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 */
114 struct scan_frame *this_prev_frame; /* this previous frame */
115 struct scan_frame *prev_frame; /* previous frame */
116 struct scan_frame *next_frame; /* next frame */
119 /* Certain characters are output as a sequence with the first being a
121 #define isBACKSLASHED_PUNCT(c) strchr("-[]\\^", c)
124 struct RExC_state_t {
125 U32 flags; /* RXf_* are we folding, multilining? */
126 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
127 char *precomp; /* uncompiled string. */
128 char *precomp_end; /* pointer to end of uncompiled string. */
129 REGEXP *rx_sv; /* The SV that is the regexp. */
130 regexp *rx; /* perl core regexp structure */
131 regexp_internal *rxi; /* internal data for regexp object
133 char *start; /* Start of input for compile */
134 char *end; /* End of input for compile */
135 char *parse; /* Input-scan pointer. */
136 char *adjusted_start; /* 'start', adjusted. See code use */
137 STRLEN precomp_adj; /* an offset beyond precomp. See code use */
138 SSize_t whilem_seen; /* number of WHILEM in this expr */
139 regnode *emit_start; /* Start of emitted-code area */
140 regnode *emit_bound; /* First regnode outside of the
142 regnode *emit; /* Code-emit pointer; if = &emit_dummy,
143 implies compiling, so don't emit */
144 regnode_ssc emit_dummy; /* placeholder for emit to point to;
145 large enough for the largest
146 non-EXACTish node, so can use it as
148 I32 naughty; /* How bad is this pattern? */
149 I32 sawback; /* Did we see \1, ...? */
151 SSize_t size; /* Code size. */
152 I32 npar; /* Capture buffer count, (OPEN) plus
153 one. ("par" 0 is the whole
155 I32 nestroot; /* root parens we are in - used by
159 regnode **open_parens; /* pointers to open parens */
160 regnode **close_parens; /* pointers to close parens */
161 regnode *end_op; /* END node in program */
162 I32 utf8; /* whether the pattern is utf8 or not */
163 I32 orig_utf8; /* whether the pattern was originally in utf8 */
164 /* XXX use this for future optimisation of case
165 * where pattern must be upgraded to utf8. */
166 I32 uni_semantics; /* If a d charset modifier should use unicode
167 rules, even if the pattern is not in
169 HV *paren_names; /* Paren names */
171 regnode **recurse; /* Recurse regops */
172 I32 recurse_count; /* Number of recurse regops we have generated */
173 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
175 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
178 I32 override_recoding;
180 I32 recode_x_to_native;
182 I32 in_multi_char_class;
183 struct reg_code_blocks *code_blocks;/* positions of literal (?{})
185 int code_index; /* next code_blocks[] slot */
186 SSize_t maxlen; /* mininum possible number of chars in string to match */
187 scan_frame *frame_head;
188 scan_frame *frame_last;
191 #ifdef ADD_TO_REGEXEC
192 char *starttry; /* -Dr: where regtry was called. */
193 #define RExC_starttry (pRExC_state->starttry)
195 SV *runtime_code_qr; /* qr with the runtime code blocks */
197 const char *lastparse;
199 AV *paren_name_list; /* idx -> name */
200 U32 study_chunk_recursed_count;
203 #define RExC_lastparse (pRExC_state->lastparse)
204 #define RExC_lastnum (pRExC_state->lastnum)
205 #define RExC_paren_name_list (pRExC_state->paren_name_list)
206 #define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
207 #define RExC_mysv (pRExC_state->mysv1)
208 #define RExC_mysv1 (pRExC_state->mysv1)
209 #define RExC_mysv2 (pRExC_state->mysv2)
212 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)
282 #define RExC_in_script_run (pRExC_state->in_script_run)
284 /* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
285 * a flag to disable back-off on the fixed/floating substrings - if it's
286 * a high complexity pattern we assume the benefit of avoiding a full match
287 * is worth the cost of checking for the substrings even if they rarely help.
289 #define RExC_naughty (pRExC_state->naughty)
290 #define TOO_NAUGHTY (10)
291 #define MARK_NAUGHTY(add) \
292 if (RExC_naughty < TOO_NAUGHTY) \
293 RExC_naughty += (add)
294 #define MARK_NAUGHTY_EXP(exp, add) \
295 if (RExC_naughty < TOO_NAUGHTY) \
296 RExC_naughty += RExC_naughty / (exp) + (add)
298 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
299 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
300 ((*s) == '{' && regcurly(s)))
303 * Flags to be passed up and down.
305 #define WORST 0 /* Worst case. */
306 #define HASWIDTH 0x01 /* Known to match non-null strings. */
308 /* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
309 * character. (There needs to be a case: in the switch statement in regexec.c
310 * for any node marked SIMPLE.) Note that this is not the same thing as
313 #define SPSTART 0x04 /* Starts with * or + */
314 #define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
315 #define TRYAGAIN 0x10 /* Weeded out a declaration. */
316 #define RESTART_PASS1 0x20 /* Need to restart sizing pass */
317 #define NEED_UTF8 0x40 /* In conjunction with RESTART_PASS1, need to
318 calcuate sizes as UTF-8 */
320 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
322 /* whether trie related optimizations are enabled */
323 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
324 #define TRIE_STUDY_OPT
325 #define FULL_TRIE_STUDY
331 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
332 #define PBITVAL(paren) (1 << ((paren) & 7))
333 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
334 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
335 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
337 #define REQUIRE_UTF8(flagp) STMT_START { \
340 *flagp = RESTART_PASS1|NEED_UTF8; \
345 /* Change from /d into /u rules, and restart the parse if we've already seen
346 * something whose size would increase as a result, by setting *flagp and
347 * returning 'restart_retval'. RExC_uni_semantics is a flag that indicates
348 * we've changed to /u during the parse. */
349 #define REQUIRE_UNI_RULES(flagp, restart_retval) \
351 if (DEPENDS_SEMANTICS) { \
353 set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); \
354 RExC_uni_semantics = 1; \
355 if (RExC_seen_unfolded_sharp_s) { \
356 *flagp |= RESTART_PASS1; \
357 return restart_retval; \
362 /* Executes a return statement with the value 'X', if 'flags' contains any of
363 * 'RESTART_PASS1', 'NEED_UTF8', or 'extra'. If so, *flagp is set to those
365 #define RETURN_X_ON_RESTART_OR_FLAGS(X, flags, flagp, extra) \
367 if ((flags) & (RESTART_PASS1|NEED_UTF8|(extra))) { \
368 *(flagp) = (flags) & (RESTART_PASS1|NEED_UTF8|(extra)); \
373 #define RETURN_NULL_ON_RESTART_OR_FLAGS(flags,flagp,extra) \
374 RETURN_X_ON_RESTART_OR_FLAGS(NULL,flags,flagp,extra)
376 #define RETURN_X_ON_RESTART(X, flags,flagp) \
377 RETURN_X_ON_RESTART_OR_FLAGS( X, flags, flagp, 0)
380 #define RETURN_NULL_ON_RESTART_FLAGP_OR_FLAGS(flagp,extra) \
381 if (*(flagp) & (RESTART_PASS1|(extra))) return NULL
383 #define MUST_RESTART(flags) ((flags) & (RESTART_PASS1))
385 #define RETURN_NULL_ON_RESTART(flags,flagp) \
386 RETURN_X_ON_RESTART(NULL, flags,flagp)
387 #define RETURN_NULL_ON_RESTART_FLAGP(flagp) \
388 RETURN_NULL_ON_RESTART_FLAGP_OR_FLAGS(flagp,0)
390 /* This converts the named class defined in regcomp.h to its equivalent class
391 * number defined in handy.h. */
392 #define namedclass_to_classnum(class) ((int) ((class) / 2))
393 #define classnum_to_namedclass(classnum) ((classnum) * 2)
395 #define _invlist_union_complement_2nd(a, b, output) \
396 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
397 #define _invlist_intersection_complement_2nd(a, b, output) \
398 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
400 /* About scan_data_t.
402 During optimisation we recurse through the regexp program performing
403 various inplace (keyhole style) optimisations. In addition study_chunk
404 and scan_commit populate this data structure with information about
405 what strings MUST appear in the pattern. We look for the longest
406 string that must appear at a fixed location, and we look for the
407 longest string that may appear at a floating location. So for instance
412 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
413 strings (because they follow a .* construct). study_chunk will identify
414 both FOO and BAR as being the longest fixed and floating strings respectively.
416 The strings can be composites, for instance
420 will result in a composite fixed substring 'foo'.
422 For each string some basic information is maintained:
425 This is the position the string must appear at, or not before.
426 It also implicitly (when combined with minlenp) tells us how many
427 characters must match before the string we are searching for.
428 Likewise when combined with minlenp and the length of the string it
429 tells us how many characters must appear after the string we have
433 Only used for floating strings. This is the rightmost point that
434 the string can appear at. If set to SSize_t_MAX it indicates that the
435 string can occur infinitely far to the right.
436 For fixed strings, it is equal to min_offset.
439 A pointer to the minimum number of characters of the pattern that the
440 string was found inside. This is important as in the case of positive
441 lookahead or positive lookbehind we can have multiple patterns
446 The minimum length of the pattern overall is 3, the minimum length
447 of the lookahead part is 3, but the minimum length of the part that
448 will actually match is 1. So 'FOO's minimum length is 3, but the
449 minimum length for the F is 1. This is important as the minimum length
450 is used to determine offsets in front of and behind the string being
451 looked for. Since strings can be composites this is the length of the
452 pattern at the time it was committed with a scan_commit. Note that
453 the length is calculated by study_chunk, so that the minimum lengths
454 are not known until the full pattern has been compiled, thus the
455 pointer to the value.
459 In the case of lookbehind the string being searched for can be
460 offset past the start point of the final matching string.
461 If this value was just blithely removed from the min_offset it would
462 invalidate some of the calculations for how many chars must match
463 before or after (as they are derived from min_offset and minlen and
464 the length of the string being searched for).
465 When the final pattern is compiled and the data is moved from the
466 scan_data_t structure into the regexp structure the information
467 about lookbehind is factored in, with the information that would
468 have been lost precalculated in the end_shift field for the
471 The fields pos_min and pos_delta are used to store the minimum offset
472 and the delta to the maximum offset at the current point in the pattern.
476 struct scan_data_substrs {
477 SV *str; /* longest substring found in pattern */
478 SSize_t min_offset; /* earliest point in string it can appear */
479 SSize_t max_offset; /* latest point in string it can appear */
480 SSize_t *minlenp; /* pointer to the minlen relevant to the string */
481 SSize_t lookbehind; /* is the pos of the string modified by LB */
482 I32 flags; /* per substring SF_* and SCF_* flags */
485 typedef struct scan_data_t {
486 /*I32 len_min; unused */
487 /*I32 len_delta; unused */
491 SSize_t last_end; /* min value, <0 unless valid. */
492 SSize_t last_start_min;
493 SSize_t last_start_max;
494 U8 cur_is_floating; /* whether the last_* values should be set as
495 * the next fixed (0) or floating (1)
498 /* [0] is longest fixed substring so far, [1] is longest float so far */
499 struct scan_data_substrs substrs[2];
501 I32 flags; /* common SF_* and SCF_* flags */
503 SSize_t *last_closep;
504 regnode_ssc *start_class;
508 * Forward declarations for pregcomp()'s friends.
511 static const scan_data_t zero_scan_data = {
512 0, 0, NULL, 0, 0, 0, 0,
514 { NULL, 0, 0, 0, 0, 0 },
515 { NULL, 0, 0, 0, 0, 0 },
522 #define SF_BEFORE_SEOL 0x0001
523 #define SF_BEFORE_MEOL 0x0002
524 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
526 #define SF_IS_INF 0x0040
527 #define SF_HAS_PAR 0x0080
528 #define SF_IN_PAR 0x0100
529 #define SF_HAS_EVAL 0x0200
532 /* SCF_DO_SUBSTR is the flag that tells the regexp analyzer to track the
533 * longest substring in the pattern. When it is not set the optimiser keeps
534 * track of position, but does not keep track of the actual strings seen,
536 * So for instance /foo/ will be parsed with SCF_DO_SUBSTR being true, but
539 * Similarly, /foo.*(blah|erm|huh).*fnorble/ will have "foo" and "fnorble"
540 * parsed with SCF_DO_SUBSTR on, but while processing the (...) it will be
541 * turned off because of the alternation (BRANCH). */
542 #define SCF_DO_SUBSTR 0x0400
544 #define SCF_DO_STCLASS_AND 0x0800
545 #define SCF_DO_STCLASS_OR 0x1000
546 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
547 #define SCF_WHILEM_VISITED_POS 0x2000
549 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
550 #define SCF_SEEN_ACCEPT 0x8000
551 #define SCF_TRIE_DOING_RESTUDY 0x10000
552 #define SCF_IN_DEFINE 0x20000
557 #define UTF cBOOL(RExC_utf8)
559 /* The enums for all these are ordered so things work out correctly */
560 #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
561 #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
562 == REGEX_DEPENDS_CHARSET)
563 #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
564 #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
565 >= REGEX_UNICODE_CHARSET)
566 #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
567 == REGEX_ASCII_RESTRICTED_CHARSET)
568 #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
569 >= REGEX_ASCII_RESTRICTED_CHARSET)
570 #define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
571 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
573 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
575 /* For programs that want to be strictly Unicode compatible by dying if any
576 * attempt is made to match a non-Unicode code point against a Unicode
578 #define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
580 #define OOB_NAMEDCLASS -1
582 /* There is no code point that is out-of-bounds, so this is problematic. But
583 * its only current use is to initialize a variable that is always set before
585 #define OOB_UNICODE 0xDEADBEEF
587 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
590 /* length of regex to show in messages that don't mark a position within */
591 #define RegexLengthToShowInErrorMessages 127
594 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
595 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
596 * op/pragma/warn/regcomp.
598 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
599 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
601 #define REPORT_LOCATION " in regex; marked by " MARKER1 \
602 " in m/%" UTF8f MARKER2 "%" UTF8f "/"
604 /* The code in this file in places uses one level of recursion with parsing
605 * rebased to an alternate string constructed by us in memory. This can take
606 * the form of something that is completely different from the input, or
607 * something that uses the input as part of the alternate. In the first case,
608 * there should be no possibility of an error, as we are in complete control of
609 * the alternate string. But in the second case we don't control the input
610 * portion, so there may be errors in that. Here's an example:
612 * is handled specially because \x{df} folds to a sequence of more than one
613 * character, 'ss'. What is done is to create and parse an alternate string,
614 * which looks like this:
615 * /(?:\x{DF}|[abc\x{DF}def])/ui
616 * where it uses the input unchanged in the middle of something it constructs,
617 * which is a branch for the DF outside the character class, and clustering
618 * parens around the whole thing. (It knows enough to skip the DF inside the
619 * class while in this substitute parse.) 'abc' and 'def' may have errors that
620 * need to be reported. The general situation looks like this:
623 * Input: ----------------------------------------------------
624 * Constructed: ---------------------------------------------------
627 * The input string sI..eI is the input pattern. The string sC..EC is the
628 * constructed substitute parse string. The portions sC..tC and eC..EC are
629 * constructed by us. The portion tC..eC is an exact duplicate of the input
630 * pattern tI..eI. In the diagram, these are vertically aligned. Suppose that
631 * while parsing, we find an error at xC. We want to display a message showing
632 * the real input string. Thus we need to find the point xI in it which
633 * corresponds to xC. xC >= tC, since the portion of the string sC..tC has
634 * been constructed by us, and so shouldn't have errors. We get:
636 * xI = sI + (tI - sI) + (xC - tC)
638 * and, the offset into sI is:
640 * (xI - sI) = (tI - sI) + (xC - tC)
642 * When the substitute is constructed, we save (tI -sI) as RExC_precomp_adj,
643 * and we save tC as RExC_adjusted_start.
645 * During normal processing of the input pattern, everything points to that,
646 * with RExC_precomp_adj set to 0, and RExC_adjusted_start set to sI.
649 #define tI_sI RExC_precomp_adj
650 #define tC RExC_adjusted_start
651 #define sC RExC_precomp
652 #define xI_offset(xC) ((IV) (tI_sI + (xC - tC)))
653 #define xI(xC) (sC + xI_offset(xC))
654 #define eC RExC_precomp_end
656 #define REPORT_LOCATION_ARGS(xC) \
658 (xI(xC) > eC) /* Don't run off end */ \
659 ? eC - sC /* Length before the <--HERE */ \
660 : ( __ASSERT_(xI_offset(xC) >= 0) xI_offset(xC) ), \
661 sC), /* The input pattern printed up to the <--HERE */ \
663 (xI(xC) > eC) ? 0 : eC - xI(xC), /* Length after <--HERE */ \
664 (xI(xC) > eC) ? eC : xI(xC)) /* pattern after <--HERE */
666 /* Used to point after bad bytes for an error message, but avoid skipping
667 * past a nul byte. */
668 #define SKIP_IF_CHAR(s) (!*(s) ? 0 : UTF ? UTF8SKIP(s) : 1)
671 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
672 * arg. Show regex, up to a maximum length. If it's too long, chop and add
675 #define _FAIL(code) STMT_START { \
676 const char *ellipses = ""; \
677 IV len = RExC_precomp_end - RExC_precomp; \
680 SAVEFREESV(RExC_rx_sv); \
681 if (len > RegexLengthToShowInErrorMessages) { \
682 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
683 len = RegexLengthToShowInErrorMessages - 10; \
689 #define FAIL(msg) _FAIL( \
690 Perl_croak(aTHX_ "%s in regex m/%" UTF8f "%s/", \
691 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
693 #define FAIL2(msg,arg) _FAIL( \
694 Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \
695 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
698 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
700 #define Simple_vFAIL(m) STMT_START { \
701 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
702 m, REPORT_LOCATION_ARGS(RExC_parse)); \
706 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
708 #define vFAIL(m) STMT_START { \
710 SAVEFREESV(RExC_rx_sv); \
715 * Like Simple_vFAIL(), but accepts two arguments.
717 #define Simple_vFAIL2(m,a1) STMT_START { \
718 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
719 REPORT_LOCATION_ARGS(RExC_parse)); \
723 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
725 #define vFAIL2(m,a1) STMT_START { \
727 SAVEFREESV(RExC_rx_sv); \
728 Simple_vFAIL2(m, a1); \
733 * Like Simple_vFAIL(), but accepts three arguments.
735 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
736 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
737 REPORT_LOCATION_ARGS(RExC_parse)); \
741 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
743 #define vFAIL3(m,a1,a2) STMT_START { \
745 SAVEFREESV(RExC_rx_sv); \
746 Simple_vFAIL3(m, a1, a2); \
750 * Like Simple_vFAIL(), but accepts four arguments.
752 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
753 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
754 REPORT_LOCATION_ARGS(RExC_parse)); \
757 #define vFAIL4(m,a1,a2,a3) STMT_START { \
759 SAVEFREESV(RExC_rx_sv); \
760 Simple_vFAIL4(m, a1, a2, a3); \
763 /* A specialized version of vFAIL2 that works with UTF8f */
764 #define vFAIL2utf8f(m, a1) STMT_START { \
766 SAVEFREESV(RExC_rx_sv); \
767 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
768 REPORT_LOCATION_ARGS(RExC_parse)); \
771 #define vFAIL3utf8f(m, a1, a2) STMT_START { \
773 SAVEFREESV(RExC_rx_sv); \
774 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
775 REPORT_LOCATION_ARGS(RExC_parse)); \
778 /* These have asserts in them because of [perl #122671] Many warnings in
779 * regcomp.c can occur twice. If they get output in pass1 and later in that
780 * pass, the pattern has to be converted to UTF-8 and the pass restarted, they
781 * would get output again. So they should be output in pass2, and these
782 * asserts make sure new warnings follow that paradigm. */
784 /* m is not necessarily a "literal string", in this macro */
785 #define reg_warn_non_literal_string(loc, m) STMT_START { \
786 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
787 "%s" REPORT_LOCATION, \
788 m, REPORT_LOCATION_ARGS(loc)); \
791 #define ckWARNreg(loc,m) STMT_START { \
792 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
794 REPORT_LOCATION_ARGS(loc)); \
797 #define vWARN(loc, m) STMT_START { \
798 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
800 REPORT_LOCATION_ARGS(loc)); \
803 #define vWARN_dep(loc, m) STMT_START { \
804 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), \
806 REPORT_LOCATION_ARGS(loc)); \
809 #define ckWARNdep(loc,m) STMT_START { \
810 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
812 REPORT_LOCATION_ARGS(loc)); \
815 #define ckWARNregdep(loc,m) STMT_START { \
816 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, \
819 REPORT_LOCATION_ARGS(loc)); \
822 #define ckWARN2reg_d(loc,m, a1) STMT_START { \
823 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
825 a1, REPORT_LOCATION_ARGS(loc)); \
828 #define ckWARN2reg(loc, m, a1) STMT_START { \
829 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
831 a1, REPORT_LOCATION_ARGS(loc)); \
834 #define vWARN3(loc, m, a1, a2) STMT_START { \
835 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
837 a1, a2, REPORT_LOCATION_ARGS(loc)); \
840 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
841 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
844 REPORT_LOCATION_ARGS(loc)); \
847 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
848 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
851 REPORT_LOCATION_ARGS(loc)); \
854 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
855 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
858 REPORT_LOCATION_ARGS(loc)); \
861 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
862 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
865 REPORT_LOCATION_ARGS(loc)); \
868 /* Macros for recording node offsets. 20001227 mjd@plover.com
869 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
870 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
871 * Element 0 holds the number n.
872 * Position is 1 indexed.
874 #ifndef RE_TRACK_PATTERN_OFFSETS
875 #define Set_Node_Offset_To_R(node,byte)
876 #define Set_Node_Offset(node,byte)
877 #define Set_Cur_Node_Offset
878 #define Set_Node_Length_To_R(node,len)
879 #define Set_Node_Length(node,len)
880 #define Set_Node_Cur_Length(node,start)
881 #define Node_Offset(n)
882 #define Node_Length(n)
883 #define Set_Node_Offset_Length(node,offset,len)
884 #define ProgLen(ri) ri->u.proglen
885 #define SetProgLen(ri,x) ri->u.proglen = x
887 #define ProgLen(ri) ri->u.offsets[0]
888 #define SetProgLen(ri,x) ri->u.offsets[0] = x
889 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
891 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
892 __LINE__, (int)(node), (int)(byte))); \
894 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
897 RExC_offsets[2*(node)-1] = (byte); \
902 #define Set_Node_Offset(node,byte) \
903 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
904 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
906 #define Set_Node_Length_To_R(node,len) STMT_START { \
908 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
909 __LINE__, (int)(node), (int)(len))); \
911 Perl_croak(aTHX_ "value of node is %d in Length macro", \
914 RExC_offsets[2*(node)] = (len); \
919 #define Set_Node_Length(node,len) \
920 Set_Node_Length_To_R((node)-RExC_emit_start, len)
921 #define Set_Node_Cur_Length(node, start) \
922 Set_Node_Length(node, RExC_parse - start)
924 /* Get offsets and lengths */
925 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
926 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
928 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
929 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
930 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
934 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
935 #define EXPERIMENTAL_INPLACESCAN
936 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
940 Perl_re_printf(pTHX_ const char *fmt, ...)
944 PerlIO *f= Perl_debug_log;
945 PERL_ARGS_ASSERT_RE_PRINTF;
947 result = PerlIO_vprintf(f, fmt, ap);
953 Perl_re_indentf(pTHX_ const char *fmt, U32 depth, ...)
957 PerlIO *f= Perl_debug_log;
958 PERL_ARGS_ASSERT_RE_INDENTF;
960 PerlIO_printf(f, "%*s", ( (int)depth % 20 ) * 2, "");
961 result = PerlIO_vprintf(f, fmt, ap);
965 #endif /* DEBUGGING */
967 #define DEBUG_RExC_seen() \
968 DEBUG_OPTIMISE_MORE_r({ \
969 Perl_re_printf( aTHX_ "RExC_seen: "); \
971 if (RExC_seen & REG_ZERO_LEN_SEEN) \
972 Perl_re_printf( aTHX_ "REG_ZERO_LEN_SEEN "); \
974 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
975 Perl_re_printf( aTHX_ "REG_LOOKBEHIND_SEEN "); \
977 if (RExC_seen & REG_GPOS_SEEN) \
978 Perl_re_printf( aTHX_ "REG_GPOS_SEEN "); \
980 if (RExC_seen & REG_RECURSE_SEEN) \
981 Perl_re_printf( aTHX_ "REG_RECURSE_SEEN "); \
983 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
984 Perl_re_printf( aTHX_ "REG_TOP_LEVEL_BRANCHES_SEEN "); \
986 if (RExC_seen & REG_VERBARG_SEEN) \
987 Perl_re_printf( aTHX_ "REG_VERBARG_SEEN "); \
989 if (RExC_seen & REG_CUTGROUP_SEEN) \
990 Perl_re_printf( aTHX_ "REG_CUTGROUP_SEEN "); \
992 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
993 Perl_re_printf( aTHX_ "REG_RUN_ON_COMMENT_SEEN "); \
995 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
996 Perl_re_printf( aTHX_ "REG_UNFOLDED_MULTI_SEEN "); \
998 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
999 Perl_re_printf( aTHX_ "REG_UNBOUNDED_QUANTIFIER_SEEN "); \
1001 Perl_re_printf( aTHX_ "\n"); \
1004 #define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
1005 if ((flags) & flag) Perl_re_printf( aTHX_ "%s ", #flag)
1010 S_debug_show_study_flags(pTHX_ U32 flags, const char *open_str,
1011 const char *close_str)
1016 Perl_re_printf( aTHX_ "%s", open_str);
1017 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_SEOL);
1018 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_MEOL);
1019 DEBUG_SHOW_STUDY_FLAG(flags, SF_IS_INF);
1020 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_PAR);
1021 DEBUG_SHOW_STUDY_FLAG(flags, SF_IN_PAR);
1022 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_EVAL);
1023 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_SUBSTR);
1024 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_AND);
1025 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_OR);
1026 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS);
1027 DEBUG_SHOW_STUDY_FLAG(flags, SCF_WHILEM_VISITED_POS);
1028 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_RESTUDY);
1029 DEBUG_SHOW_STUDY_FLAG(flags, SCF_SEEN_ACCEPT);
1030 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_DOING_RESTUDY);
1031 DEBUG_SHOW_STUDY_FLAG(flags, SCF_IN_DEFINE);
1032 Perl_re_printf( aTHX_ "%s", close_str);
1037 S_debug_studydata(pTHX_ const char *where, scan_data_t *data,
1038 U32 depth, int is_inf)
1040 GET_RE_DEBUG_FLAGS_DECL;
1042 DEBUG_OPTIMISE_MORE_r({
1045 Perl_re_indentf(aTHX_ "%s: Pos:%" IVdf "/%" IVdf " Flags: 0x%" UVXf,
1049 (IV)data->pos_delta,
1053 S_debug_show_study_flags(aTHX_ data->flags," [","]");
1055 Perl_re_printf( aTHX_
1056 " Whilem_c: %" IVdf " Lcp: %" IVdf " %s",
1058 (IV)(data->last_closep ? *((data)->last_closep) : -1),
1059 is_inf ? "INF " : ""
1062 if (data->last_found) {
1064 Perl_re_printf(aTHX_
1065 "Last:'%s' %" IVdf ":%" IVdf "/%" IVdf,
1066 SvPVX_const(data->last_found),
1068 (IV)data->last_start_min,
1069 (IV)data->last_start_max
1072 for (i = 0; i < 2; i++) {
1073 Perl_re_printf(aTHX_
1074 " %s%s: '%s' @ %" IVdf "/%" IVdf,
1075 data->cur_is_floating == i ? "*" : "",
1076 i ? "Float" : "Fixed",
1077 SvPVX_const(data->substrs[i].str),
1078 (IV)data->substrs[i].min_offset,
1079 (IV)data->substrs[i].max_offset
1081 S_debug_show_study_flags(aTHX_ data->substrs[i].flags," [","]");
1085 Perl_re_printf( aTHX_ "\n");
1091 S_debug_peep(pTHX_ const char *str, const RExC_state_t *pRExC_state,
1092 regnode *scan, U32 depth, U32 flags)
1094 GET_RE_DEBUG_FLAGS_DECL;
1101 Next = regnext(scan);
1102 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
1103 Perl_re_indentf( aTHX_ "%s>%3d: %s (%d)",
1106 REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),
1107 Next ? (REG_NODE_NUM(Next)) : 0 );
1108 S_debug_show_study_flags(aTHX_ flags," [ ","]");
1109 Perl_re_printf( aTHX_ "\n");
1114 # define DEBUG_STUDYDATA(where, data, depth, is_inf) \
1115 S_debug_studydata(aTHX_ where, data, depth, is_inf)
1117 # define DEBUG_PEEP(str, scan, depth, flags) \
1118 S_debug_peep(aTHX_ str, pRExC_state, scan, depth, flags)
1121 # define DEBUG_STUDYDATA(where, data, depth, is_inf) NOOP
1122 # define DEBUG_PEEP(str, scan, depth, flags) NOOP
1126 /* =========================================================
1127 * BEGIN edit_distance stuff.
1129 * This calculates how many single character changes of any type are needed to
1130 * transform a string into another one. It is taken from version 3.1 of
1132 * https://metacpan.org/pod/Text::Levenshtein::Damerau::XS
1135 /* Our unsorted dictionary linked list. */
1136 /* Note we use UVs, not chars. */
1141 struct dictionary* next;
1143 typedef struct dictionary item;
1146 PERL_STATIC_INLINE item*
1147 push(UV key,item* curr)
1150 Newx(head, 1, item);
1158 PERL_STATIC_INLINE item*
1159 find(item* head, UV key)
1161 item* iterator = head;
1163 if (iterator->key == key){
1166 iterator = iterator->next;
1172 PERL_STATIC_INLINE item*
1173 uniquePush(item* head,UV key)
1175 item* iterator = head;
1178 if (iterator->key == key) {
1181 iterator = iterator->next;
1184 return push(key,head);
1187 PERL_STATIC_INLINE void
1188 dict_free(item* head)
1190 item* iterator = head;
1193 item* temp = iterator;
1194 iterator = iterator->next;
1201 /* End of Dictionary Stuff */
1203 /* All calculations/work are done here */
1205 S_edit_distance(const UV* src,
1207 const STRLEN x, /* length of src[] */
1208 const STRLEN y, /* length of tgt[] */
1209 const SSize_t maxDistance
1213 UV swapCount,swapScore,targetCharCount,i,j;
1215 UV score_ceil = x + y;
1217 PERL_ARGS_ASSERT_EDIT_DISTANCE;
1219 /* intialize matrix start values */
1220 Newx(scores, ( (x + 2) * (y + 2)), UV);
1221 scores[0] = score_ceil;
1222 scores[1 * (y + 2) + 0] = score_ceil;
1223 scores[0 * (y + 2) + 1] = score_ceil;
1224 scores[1 * (y + 2) + 1] = 0;
1225 head = uniquePush(uniquePush(head,src[0]),tgt[0]);
1230 for (i=1;i<=x;i++) {
1232 head = uniquePush(head,src[i]);
1233 scores[(i+1) * (y + 2) + 1] = i;
1234 scores[(i+1) * (y + 2) + 0] = score_ceil;
1237 for (j=1;j<=y;j++) {
1240 head = uniquePush(head,tgt[j]);
1241 scores[1 * (y + 2) + (j + 1)] = j;
1242 scores[0 * (y + 2) + (j + 1)] = score_ceil;
1245 targetCharCount = find(head,tgt[j-1])->value;
1246 swapScore = scores[targetCharCount * (y + 2) + swapCount] + i - targetCharCount - 1 + j - swapCount;
1248 if (src[i-1] != tgt[j-1]){
1249 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));
1253 scores[(i+1) * (y + 2) + (j + 1)] = MIN(scores[i * (y + 2) + j], swapScore);
1257 find(head,src[i-1])->value = i;
1261 IV score = scores[(x+1) * (y + 2) + (y + 1)];
1264 return (maxDistance != 0 && maxDistance < score)?(-1):score;
1268 /* END of edit_distance() stuff
1269 * ========================================================= */
1271 /* is c a control character for which we have a mnemonic? */
1272 #define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
1275 S_cntrl_to_mnemonic(const U8 c)
1277 /* Returns the mnemonic string that represents character 'c', if one
1278 * exists; NULL otherwise. The only ones that exist for the purposes of
1279 * this routine are a few control characters */
1282 case '\a': return "\\a";
1283 case '\b': return "\\b";
1284 case ESC_NATIVE: return "\\e";
1285 case '\f': return "\\f";
1286 case '\n': return "\\n";
1287 case '\r': return "\\r";
1288 case '\t': return "\\t";
1294 /* Mark that we cannot extend a found fixed substring at this point.
1295 Update the longest found anchored substring or the longest found
1296 floating substrings if needed. */
1299 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
1300 SSize_t *minlenp, int is_inf)
1302 const STRLEN l = CHR_SVLEN(data->last_found);
1303 SV * const longest_sv = data->substrs[data->cur_is_floating].str;
1304 const STRLEN old_l = CHR_SVLEN(longest_sv);
1305 GET_RE_DEBUG_FLAGS_DECL;
1307 PERL_ARGS_ASSERT_SCAN_COMMIT;
1309 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
1310 const U8 i = data->cur_is_floating;
1311 SvSetMagicSV(longest_sv, data->last_found);
1312 data->substrs[i].min_offset = l ? data->last_start_min : data->pos_min;
1315 data->substrs[0].max_offset = data->substrs[0].min_offset;
1317 data->substrs[1].max_offset = (l
1318 ? data->last_start_max
1319 : (data->pos_delta > SSize_t_MAX - data->pos_min
1321 : data->pos_min + data->pos_delta));
1323 || (STRLEN)data->substrs[1].max_offset > (STRLEN)SSize_t_MAX)
1324 data->substrs[1].max_offset = SSize_t_MAX;
1327 if (data->flags & SF_BEFORE_EOL)
1328 data->substrs[i].flags |= (data->flags & SF_BEFORE_EOL);
1330 data->substrs[i].flags &= ~SF_BEFORE_EOL;
1331 data->substrs[i].minlenp = minlenp;
1332 data->substrs[i].lookbehind = 0;
1335 SvCUR_set(data->last_found, 0);
1337 SV * const sv = data->last_found;
1338 if (SvUTF8(sv) && SvMAGICAL(sv)) {
1339 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
1344 data->last_end = -1;
1345 data->flags &= ~SF_BEFORE_EOL;
1346 DEBUG_STUDYDATA("commit", data, 0, is_inf);
1349 /* An SSC is just a regnode_charclass_posix with an extra field: the inversion
1350 * list that describes which code points it matches */
1353 S_ssc_anything(pTHX_ regnode_ssc *ssc)
1355 /* Set the SSC 'ssc' to match an empty string or any code point */
1357 PERL_ARGS_ASSERT_SSC_ANYTHING;
1359 assert(is_ANYOF_SYNTHETIC(ssc));
1361 /* mortalize so won't leak */
1362 ssc->invlist = sv_2mortal(_add_range_to_invlist(NULL, 0, UV_MAX));
1363 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
1367 S_ssc_is_anything(const regnode_ssc *ssc)
1369 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
1370 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
1371 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1372 * in any way, so there's no point in using it */
1377 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
1379 assert(is_ANYOF_SYNTHETIC(ssc));
1381 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
1385 /* See if the list consists solely of the range 0 - Infinity */
1386 invlist_iterinit(ssc->invlist);
1387 ret = invlist_iternext(ssc->invlist, &start, &end)
1391 invlist_iterfinish(ssc->invlist);
1397 /* If e.g., both \w and \W are set, matches everything */
1398 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1400 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1401 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1411 S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
1413 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1414 * string, any code point, or any posix class under locale */
1416 PERL_ARGS_ASSERT_SSC_INIT;
1418 Zero(ssc, 1, regnode_ssc);
1419 set_ANYOF_SYNTHETIC(ssc);
1420 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
1423 /* If any portion of the regex is to operate under locale rules that aren't
1424 * fully known at compile time, initialization includes it. The reason
1425 * this isn't done for all regexes is that the optimizer was written under
1426 * the assumption that locale was all-or-nothing. Given the complexity and
1427 * lack of documentation in the optimizer, and that there are inadequate
1428 * test cases for locale, many parts of it may not work properly, it is
1429 * safest to avoid locale unless necessary. */
1430 if (RExC_contains_locale) {
1431 ANYOF_POSIXL_SETALL(ssc);
1434 ANYOF_POSIXL_ZERO(ssc);
1439 S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1440 const regnode_ssc *ssc)
1442 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1443 * to the list of code points matched, and locale posix classes; hence does
1444 * not check its flags) */
1449 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1451 assert(is_ANYOF_SYNTHETIC(ssc));
1453 invlist_iterinit(ssc->invlist);
1454 ret = invlist_iternext(ssc->invlist, &start, &end)
1458 invlist_iterfinish(ssc->invlist);
1464 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
1472 S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
1473 const regnode_charclass* const node)
1475 /* Returns a mortal inversion list defining which code points are matched
1476 * by 'node', which is of type ANYOF. Handles complementing the result if
1477 * appropriate. If some code points aren't knowable at this time, the
1478 * returned list must, and will, contain every code point that is a
1482 SV* only_utf8_locale_invlist = NULL;
1484 const U32 n = ARG(node);
1485 bool new_node_has_latin1 = FALSE;
1487 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1489 /* Look at the data structure created by S_set_ANYOF_arg() */
1490 if (n != ANYOF_ONLY_HAS_BITMAP) {
1491 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1492 AV * const av = MUTABLE_AV(SvRV(rv));
1493 SV **const ary = AvARRAY(av);
1494 assert(RExC_rxi->data->what[n] == 's');
1496 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
1497 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1])));
1499 else if (ary[0] && ary[0] != &PL_sv_undef) {
1501 /* Here, no compile-time swash, and there are things that won't be
1502 * known until runtime -- we have to assume it could be anything */
1503 invlist = sv_2mortal(_new_invlist(1));
1504 return _add_range_to_invlist(invlist, 0, UV_MAX);
1506 else if (ary[3] && ary[3] != &PL_sv_undef) {
1508 /* Here no compile-time swash, and no run-time only data. Use the
1509 * node's inversion list */
1510 invlist = sv_2mortal(invlist_clone(ary[3]));
1513 /* Get the code points valid only under UTF-8 locales */
1514 if ((ANYOF_FLAGS(node) & ANYOFL_FOLD)
1515 && ary[2] && ary[2] != &PL_sv_undef)
1517 only_utf8_locale_invlist = ary[2];
1522 invlist = sv_2mortal(_new_invlist(0));
1525 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1526 * code points, and an inversion list for the others, but if there are code
1527 * points that should match only conditionally on the target string being
1528 * UTF-8, those are placed in the inversion list, and not the bitmap.
1529 * Since there are circumstances under which they could match, they are
1530 * included in the SSC. But if the ANYOF node is to be inverted, we have
1531 * to exclude them here, so that when we invert below, the end result
1532 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1533 * have to do this here before we add the unconditionally matched code
1535 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1536 _invlist_intersection_complement_2nd(invlist,
1541 /* Add in the points from the bit map */
1542 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
1543 if (ANYOF_BITMAP_TEST(node, i)) {
1544 unsigned int start = i++;
1546 for (; i < NUM_ANYOF_CODE_POINTS && ANYOF_BITMAP_TEST(node, i); ++i) {
1549 invlist = _add_range_to_invlist(invlist, start, i-1);
1550 new_node_has_latin1 = TRUE;
1554 /* If this can match all upper Latin1 code points, have to add them
1555 * as well. But don't add them if inverting, as when that gets done below,
1556 * it would exclude all these characters, including the ones it shouldn't
1557 * that were added just above */
1558 if (! (ANYOF_FLAGS(node) & ANYOF_INVERT) && OP(node) == ANYOFD
1559 && (ANYOF_FLAGS(node) & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER))
1561 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1564 /* Similarly for these */
1565 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
1566 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
1569 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1570 _invlist_invert(invlist);
1572 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOFL_FOLD) {
1574 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1575 * locale. We can skip this if there are no 0-255 at all. */
1576 _invlist_union(invlist, PL_Latin1, &invlist);
1579 /* Similarly add the UTF-8 locale possible matches. These have to be
1580 * deferred until after the non-UTF-8 locale ones are taken care of just
1581 * above, or it leads to wrong results under ANYOF_INVERT */
1582 if (only_utf8_locale_invlist) {
1583 _invlist_union_maybe_complement_2nd(invlist,
1584 only_utf8_locale_invlist,
1585 ANYOF_FLAGS(node) & ANYOF_INVERT,
1592 /* These two functions currently do the exact same thing */
1593 #define ssc_init_zero ssc_init
1595 #define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1596 #define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1598 /* 'AND' a given class with another one. Can create false positives. 'ssc'
1599 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1600 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
1603 S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1604 const regnode_charclass *and_with)
1606 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1607 * another SSC or a regular ANYOF class. Can create false positives. */
1612 PERL_ARGS_ASSERT_SSC_AND;
1614 assert(is_ANYOF_SYNTHETIC(ssc));
1616 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1617 * the code point inversion list and just the relevant flags */
1618 if (is_ANYOF_SYNTHETIC(and_with)) {
1619 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
1620 anded_flags = ANYOF_FLAGS(and_with);
1622 /* XXX This is a kludge around what appears to be deficiencies in the
1623 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1624 * there are paths through the optimizer where it doesn't get weeded
1625 * out when it should. And if we don't make some extra provision for
1626 * it like the code just below, it doesn't get added when it should.
1627 * This solution is to add it only when AND'ing, which is here, and
1628 * only when what is being AND'ed is the pristine, original node
1629 * matching anything. Thus it is like adding it to ssc_anything() but
1630 * only when the result is to be AND'ed. Probably the same solution
1631 * could be adopted for the same problem we have with /l matching,
1632 * which is solved differently in S_ssc_init(), and that would lead to
1633 * fewer false positives than that solution has. But if this solution
1634 * creates bugs, the consequences are only that a warning isn't raised
1635 * that should be; while the consequences for having /l bugs is
1636 * incorrect matches */
1637 if (ssc_is_anything((regnode_ssc *)and_with)) {
1638 anded_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
1642 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
1643 if (OP(and_with) == ANYOFD) {
1644 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1647 anded_flags = ANYOF_FLAGS(and_with)
1648 &( ANYOF_COMMON_FLAGS
1649 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1650 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
1651 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(and_with))) {
1653 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1658 ANYOF_FLAGS(ssc) &= anded_flags;
1660 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1661 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1662 * 'and_with' may be inverted. When not inverted, we have the situation of
1664 * (C1 | P1) & (C2 | P2)
1665 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1666 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1667 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1668 * <= ((C1 & C2) | P1 | P2)
1669 * Alternatively, the last few steps could be:
1670 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1671 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1672 * <= (C1 | C2 | (P1 & P2))
1673 * We favor the second approach if either P1 or P2 is non-empty. This is
1674 * because these components are a barrier to doing optimizations, as what
1675 * they match cannot be known until the moment of matching as they are
1676 * dependent on the current locale, 'AND"ing them likely will reduce or
1678 * But we can do better if we know that C1,P1 are in their initial state (a
1679 * frequent occurrence), each matching everything:
1680 * (<everything>) & (C2 | P2) = C2 | P2
1681 * Similarly, if C2,P2 are in their initial state (again a frequent
1682 * occurrence), the result is a no-op
1683 * (C1 | P1) & (<everything>) = C1 | P1
1686 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1687 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1688 * <= (C1 & ~C2) | (P1 & ~P2)
1691 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
1692 && ! is_ANYOF_SYNTHETIC(and_with))
1696 ssc_intersection(ssc,
1698 FALSE /* Has already been inverted */
1701 /* If either P1 or P2 is empty, the intersection will be also; can skip
1703 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
1704 ANYOF_POSIXL_ZERO(ssc);
1706 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1708 /* Note that the Posix class component P from 'and_with' actually
1710 * P = Pa | Pb | ... | Pn
1711 * where each component is one posix class, such as in [\w\s].
1713 * ~P = ~(Pa | Pb | ... | Pn)
1714 * = ~Pa & ~Pb & ... & ~Pn
1715 * <= ~Pa | ~Pb | ... | ~Pn
1716 * The last is something we can easily calculate, but unfortunately
1717 * is likely to have many false positives. We could do better
1718 * in some (but certainly not all) instances if two classes in
1719 * P have known relationships. For example
1720 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1722 * :lower: & :print: = :lower:
1723 * And similarly for classes that must be disjoint. For example,
1724 * since \s and \w can have no elements in common based on rules in
1725 * the POSIX standard,
1726 * \w & ^\S = nothing
1727 * Unfortunately, some vendor locales do not meet the Posix
1728 * standard, in particular almost everything by Microsoft.
1729 * The loop below just changes e.g., \w into \W and vice versa */
1731 regnode_charclass_posixl temp;
1732 int add = 1; /* To calculate the index of the complement */
1734 Zero(&temp, 1, regnode_charclass_posixl);
1735 ANYOF_POSIXL_ZERO(&temp);
1736 for (i = 0; i < ANYOF_MAX; i++) {
1738 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1739 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
1741 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
1742 ANYOF_POSIXL_SET(&temp, i + add);
1744 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1746 ANYOF_POSIXL_AND(&temp, ssc);
1748 } /* else ssc already has no posixes */
1749 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1750 in its initial state */
1751 else if (! is_ANYOF_SYNTHETIC(and_with)
1752 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
1754 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1755 * copy it over 'ssc' */
1756 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
1757 if (is_ANYOF_SYNTHETIC(and_with)) {
1758 StructCopy(and_with, ssc, regnode_ssc);
1761 ssc->invlist = anded_cp_list;
1762 ANYOF_POSIXL_ZERO(ssc);
1763 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1764 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
1768 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
1769 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
1771 /* One or the other of P1, P2 is non-empty. */
1772 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1773 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1775 ssc_union(ssc, anded_cp_list, FALSE);
1777 else { /* P1 = P2 = empty */
1778 ssc_intersection(ssc, anded_cp_list, FALSE);
1784 S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
1785 const regnode_charclass *or_with)
1787 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1788 * another SSC or a regular ANYOF class. Can create false positives if
1789 * 'or_with' is to be inverted. */
1794 PERL_ARGS_ASSERT_SSC_OR;
1796 assert(is_ANYOF_SYNTHETIC(ssc));
1798 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1799 * the code point inversion list and just the relevant flags */
1800 if (is_ANYOF_SYNTHETIC(or_with)) {
1801 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
1802 ored_flags = ANYOF_FLAGS(or_with);
1805 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
1806 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
1807 if (OP(or_with) != ANYOFD) {
1809 |= ANYOF_FLAGS(or_with)
1810 & ( ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1811 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
1812 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(or_with))) {
1814 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1819 ANYOF_FLAGS(ssc) |= ored_flags;
1821 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1822 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1823 * 'or_with' may be inverted. When not inverted, we have the simple
1824 * situation of computing:
1825 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1826 * If P1|P2 yields a situation with both a class and its complement are
1827 * set, like having both \w and \W, this matches all code points, and we
1828 * can delete these from the P component of the ssc going forward. XXX We
1829 * might be able to delete all the P components, but I (khw) am not certain
1830 * about this, and it is better to be safe.
1833 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1834 * <= (C1 | P1) | ~C2
1835 * <= (C1 | ~C2) | P1
1836 * (which results in actually simpler code than the non-inverted case)
1839 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
1840 && ! is_ANYOF_SYNTHETIC(or_with))
1842 /* We ignore P2, leaving P1 going forward */
1843 } /* else Not inverted */
1844 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
1845 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
1846 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
1848 for (i = 0; i < ANYOF_MAX; i += 2) {
1849 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1851 ssc_match_all_cp(ssc);
1852 ANYOF_POSIXL_CLEAR(ssc, i);
1853 ANYOF_POSIXL_CLEAR(ssc, i+1);
1861 FALSE /* Already has been inverted */
1865 PERL_STATIC_INLINE void
1866 S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1868 PERL_ARGS_ASSERT_SSC_UNION;
1870 assert(is_ANYOF_SYNTHETIC(ssc));
1872 _invlist_union_maybe_complement_2nd(ssc->invlist,
1878 PERL_STATIC_INLINE void
1879 S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1881 const bool invert2nd)
1883 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1885 assert(is_ANYOF_SYNTHETIC(ssc));
1887 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1893 PERL_STATIC_INLINE void
1894 S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1896 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1898 assert(is_ANYOF_SYNTHETIC(ssc));
1900 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1903 PERL_STATIC_INLINE void
1904 S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1906 /* AND just the single code point 'cp' into the SSC 'ssc' */
1908 SV* cp_list = _new_invlist(2);
1910 PERL_ARGS_ASSERT_SSC_CP_AND;
1912 assert(is_ANYOF_SYNTHETIC(ssc));
1914 cp_list = add_cp_to_invlist(cp_list, cp);
1915 ssc_intersection(ssc, cp_list,
1916 FALSE /* Not inverted */
1918 SvREFCNT_dec_NN(cp_list);
1921 PERL_STATIC_INLINE void
1922 S_ssc_clear_locale(regnode_ssc *ssc)
1924 /* Set the SSC 'ssc' to not match any locale things */
1925 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
1927 assert(is_ANYOF_SYNTHETIC(ssc));
1929 ANYOF_POSIXL_ZERO(ssc);
1930 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
1933 #define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
1936 S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
1938 /* The synthetic start class is used to hopefully quickly winnow down
1939 * places where a pattern could start a match in the target string. If it
1940 * doesn't really narrow things down that much, there isn't much point to
1941 * having the overhead of using it. This function uses some very crude
1942 * heuristics to decide if to use the ssc or not.
1944 * It returns TRUE if 'ssc' rules out more than half what it considers to
1945 * be the "likely" possible matches, but of course it doesn't know what the
1946 * actual things being matched are going to be; these are only guesses
1948 * For /l matches, it assumes that the only likely matches are going to be
1949 * in the 0-255 range, uniformly distributed, so half of that is 127
1950 * For /a and /d matches, it assumes that the likely matches will be just
1951 * the ASCII range, so half of that is 63
1952 * For /u and there isn't anything matching above the Latin1 range, it
1953 * assumes that that is the only range likely to be matched, and uses
1954 * half that as the cut-off: 127. If anything matches above Latin1,
1955 * it assumes that all of Unicode could match (uniformly), except for
1956 * non-Unicode code points and things in the General Category "Other"
1957 * (unassigned, private use, surrogates, controls and formats). This
1958 * is a much large number. */
1960 U32 count = 0; /* Running total of number of code points matched by
1962 UV start, end; /* Start and end points of current range in inversion
1964 const U32 max_code_points = (LOC)
1966 : (( ! UNI_SEMANTICS
1967 || invlist_highest(ssc->invlist) < 256)
1970 const U32 max_match = max_code_points / 2;
1972 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
1974 invlist_iterinit(ssc->invlist);
1975 while (invlist_iternext(ssc->invlist, &start, &end)) {
1976 if (start >= max_code_points) {
1979 end = MIN(end, max_code_points - 1);
1980 count += end - start + 1;
1981 if (count >= max_match) {
1982 invlist_iterfinish(ssc->invlist);
1992 S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
1994 /* The inversion list in the SSC is marked mortal; now we need a more
1995 * permanent copy, which is stored the same way that is done in a regular
1996 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
1999 SV* invlist = invlist_clone(ssc->invlist);
2001 PERL_ARGS_ASSERT_SSC_FINALIZE;
2003 assert(is_ANYOF_SYNTHETIC(ssc));
2005 /* The code in this file assumes that all but these flags aren't relevant
2006 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
2007 * by the time we reach here */
2008 assert(! (ANYOF_FLAGS(ssc)
2009 & ~( ANYOF_COMMON_FLAGS
2010 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
2011 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)));
2013 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
2015 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
2016 NULL, NULL, NULL, FALSE);
2018 /* Make sure is clone-safe */
2019 ssc->invlist = NULL;
2021 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
2022 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
2025 if (RExC_contains_locale) {
2029 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
2032 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
2033 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
2034 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
2035 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
2036 ? (TRIE_LIST_CUR( idx ) - 1) \
2042 dump_trie(trie,widecharmap,revcharmap)
2043 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
2044 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
2046 These routines dump out a trie in a somewhat readable format.
2047 The _interim_ variants are used for debugging the interim
2048 tables that are used to generate the final compressed
2049 representation which is what dump_trie expects.
2051 Part of the reason for their existence is to provide a form
2052 of documentation as to how the different representations function.
2057 Dumps the final compressed table form of the trie to Perl_debug_log.
2058 Used for debugging make_trie().
2062 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
2063 AV *revcharmap, U32 depth)
2066 SV *sv=sv_newmortal();
2067 int colwidth= widecharmap ? 6 : 4;
2069 GET_RE_DEBUG_FLAGS_DECL;
2071 PERL_ARGS_ASSERT_DUMP_TRIE;
2073 Perl_re_indentf( aTHX_ "Char : %-6s%-6s%-4s ",
2074 depth+1, "Match","Base","Ofs" );
2076 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2077 SV ** const tmp = av_fetch( revcharmap, state, 0);
2079 Perl_re_printf( aTHX_ "%*s",
2081 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
2082 PL_colors[0], PL_colors[1],
2083 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2084 PERL_PV_ESCAPE_FIRSTCHAR
2089 Perl_re_printf( aTHX_ "\n");
2090 Perl_re_indentf( aTHX_ "State|-----------------------", depth+1);
2092 for( state = 0 ; state < trie->uniquecharcount ; state++ )
2093 Perl_re_printf( aTHX_ "%.*s", colwidth, "--------");
2094 Perl_re_printf( aTHX_ "\n");
2096 for( state = 1 ; state < trie->statecount ; state++ ) {
2097 const U32 base = trie->states[ state ].trans.base;
2099 Perl_re_indentf( aTHX_ "#%4" UVXf "|", depth+1, (UV)state);
2101 if ( trie->states[ state ].wordnum ) {
2102 Perl_re_printf( aTHX_ " W%4X", trie->states[ state ].wordnum );
2104 Perl_re_printf( aTHX_ "%6s", "" );
2107 Perl_re_printf( aTHX_ " @%4" UVXf " ", (UV)base );
2112 while( ( base + ofs < trie->uniquecharcount ) ||
2113 ( base + ofs - trie->uniquecharcount < trie->lasttrans
2114 && trie->trans[ base + ofs - trie->uniquecharcount ].check
2118 Perl_re_printf( aTHX_ "+%2" UVXf "[ ", (UV)ofs);
2120 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2121 if ( ( base + ofs >= trie->uniquecharcount )
2122 && ( base + ofs - trie->uniquecharcount
2124 && trie->trans[ base + ofs
2125 - trie->uniquecharcount ].check == state )
2127 Perl_re_printf( aTHX_ "%*" UVXf, colwidth,
2128 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next
2131 Perl_re_printf( aTHX_ "%*s",colwidth," ." );
2135 Perl_re_printf( aTHX_ "]");
2138 Perl_re_printf( aTHX_ "\n" );
2140 Perl_re_indentf( aTHX_ "word_info N:(prev,len)=",
2142 for (word=1; word <= trie->wordcount; word++) {
2143 Perl_re_printf( aTHX_ " %d:(%d,%d)",
2144 (int)word, (int)(trie->wordinfo[word].prev),
2145 (int)(trie->wordinfo[word].len));
2147 Perl_re_printf( aTHX_ "\n" );
2150 Dumps a fully constructed but uncompressed trie in list form.
2151 List tries normally only are used for construction when the number of
2152 possible chars (trie->uniquecharcount) is very high.
2153 Used for debugging make_trie().
2156 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2157 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2161 SV *sv=sv_newmortal();
2162 int colwidth= widecharmap ? 6 : 4;
2163 GET_RE_DEBUG_FLAGS_DECL;
2165 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
2167 /* print out the table precompression. */
2168 Perl_re_indentf( aTHX_ "State :Word | Transition Data\n",
2170 Perl_re_indentf( aTHX_ "%s",
2171 depth+1, "------:-----+-----------------\n" );
2173 for( state=1 ; state < next_alloc ; state ++ ) {
2176 Perl_re_indentf( aTHX_ " %4" UVXf " :",
2177 depth+1, (UV)state );
2178 if ( ! trie->states[ state ].wordnum ) {
2179 Perl_re_printf( aTHX_ "%5s| ","");
2181 Perl_re_printf( aTHX_ "W%4x| ",
2182 trie->states[ state ].wordnum
2185 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
2186 SV ** const tmp = av_fetch( revcharmap,
2187 TRIE_LIST_ITEM(state,charid).forid, 0);
2189 Perl_re_printf( aTHX_ "%*s:%3X=%4" UVXf " | ",
2191 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
2193 PL_colors[0], PL_colors[1],
2194 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2195 | PERL_PV_ESCAPE_FIRSTCHAR
2197 TRIE_LIST_ITEM(state,charid).forid,
2198 (UV)TRIE_LIST_ITEM(state,charid).newstate
2201 Perl_re_printf( aTHX_ "\n%*s| ",
2202 (int)((depth * 2) + 14), "");
2205 Perl_re_printf( aTHX_ "\n");
2210 Dumps a fully constructed but uncompressed trie in table form.
2211 This is the normal DFA style state transition table, with a few
2212 twists to facilitate compression later.
2213 Used for debugging make_trie().
2216 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2217 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2222 SV *sv=sv_newmortal();
2223 int colwidth= widecharmap ? 6 : 4;
2224 GET_RE_DEBUG_FLAGS_DECL;
2226 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
2229 print out the table precompression so that we can do a visual check
2230 that they are identical.
2233 Perl_re_indentf( aTHX_ "Char : ", depth+1 );
2235 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2236 SV ** const tmp = av_fetch( revcharmap, charid, 0);
2238 Perl_re_printf( aTHX_ "%*s",
2240 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
2241 PL_colors[0], PL_colors[1],
2242 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2243 PERL_PV_ESCAPE_FIRSTCHAR
2249 Perl_re_printf( aTHX_ "\n");
2250 Perl_re_indentf( aTHX_ "State+-", depth+1 );
2252 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
2253 Perl_re_printf( aTHX_ "%.*s", colwidth,"--------");
2256 Perl_re_printf( aTHX_ "\n" );
2258 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
2260 Perl_re_indentf( aTHX_ "%4" UVXf " : ",
2262 (UV)TRIE_NODENUM( state ) );
2264 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2265 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
2267 Perl_re_printf( aTHX_ "%*" UVXf, colwidth, v );
2269 Perl_re_printf( aTHX_ "%*s", colwidth, "." );
2271 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
2272 Perl_re_printf( aTHX_ " (%4" UVXf ")\n",
2273 (UV)trie->trans[ state ].check );
2275 Perl_re_printf( aTHX_ " (%4" UVXf ") W%4X\n",
2276 (UV)trie->trans[ state ].check,
2277 trie->states[ TRIE_NODENUM( state ) ].wordnum );
2285 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
2286 startbranch: the first branch in the whole branch sequence
2287 first : start branch of sequence of branch-exact nodes.
2288 May be the same as startbranch
2289 last : Thing following the last branch.
2290 May be the same as tail.
2291 tail : item following the branch sequence
2292 count : words in the sequence
2293 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
2294 depth : indent depth
2296 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
2298 A trie is an N'ary tree where the branches are determined by digital
2299 decomposition of the key. IE, at the root node you look up the 1st character and
2300 follow that branch repeat until you find the end of the branches. Nodes can be
2301 marked as "accepting" meaning they represent a complete word. Eg:
2305 would convert into the following structure. Numbers represent states, letters
2306 following numbers represent valid transitions on the letter from that state, if
2307 the number is in square brackets it represents an accepting state, otherwise it
2308 will be in parenthesis.
2310 +-h->+-e->[3]-+-r->(8)-+-s->[9]
2314 (1) +-i->(6)-+-s->[7]
2316 +-s->(3)-+-h->(4)-+-e->[5]
2318 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
2320 This shows that when matching against the string 'hers' we will begin at state 1
2321 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
2322 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
2323 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
2324 single traverse. We store a mapping from accepting to state to which word was
2325 matched, and then when we have multiple possibilities we try to complete the
2326 rest of the regex in the order in which they occurred in the alternation.
2328 The only prior NFA like behaviour that would be changed by the TRIE support is
2329 the silent ignoring of duplicate alternations which are of the form:
2331 / (DUPE|DUPE) X? (?{ ... }) Y /x
2333 Thus EVAL blocks following a trie may be called a different number of times with
2334 and without the optimisation. With the optimisations dupes will be silently
2335 ignored. This inconsistent behaviour of EVAL type nodes is well established as
2336 the following demonstrates:
2338 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
2340 which prints out 'word' three times, but
2342 'words'=~/(word|word|word)(?{ print $1 })S/
2344 which doesnt print it out at all. This is due to other optimisations kicking in.
2346 Example of what happens on a structural level:
2348 The regexp /(ac|ad|ab)+/ will produce the following debug output:
2350 1: CURLYM[1] {1,32767}(18)
2361 This would be optimizable with startbranch=5, first=5, last=16, tail=16
2362 and should turn into:
2364 1: CURLYM[1] {1,32767}(18)
2366 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
2374 Cases where tail != last would be like /(?foo|bar)baz/:
2384 which would be optimizable with startbranch=1, first=1, last=7, tail=8
2385 and would end up looking like:
2388 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
2395 d = uvchr_to_utf8_flags(d, uv, 0);
2397 is the recommended Unicode-aware way of saying
2402 #define TRIE_STORE_REVCHAR(val) \
2405 SV *zlopp = newSV(UTF8_MAXBYTES); \
2406 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
2407 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
2408 SvCUR_set(zlopp, kapow - flrbbbbb); \
2411 av_push(revcharmap, zlopp); \
2413 char ooooff = (char)val; \
2414 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2418 /* This gets the next character from the input, folding it if not already
2420 #define TRIE_READ_CHAR STMT_START { \
2423 /* if it is UTF then it is either already folded, or does not need \
2425 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
2427 else if (folder == PL_fold_latin1) { \
2428 /* This folder implies Unicode rules, which in the range expressible \
2429 * by not UTF is the lower case, with the two exceptions, one of \
2430 * which should have been taken care of before calling this */ \
2431 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2432 uvc = toLOWER_L1(*uc); \
2433 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2436 /* raw data, will be folded later if needed */ \
2444 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2445 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
2446 U32 ging = TRIE_LIST_LEN( state ) * 2; \
2447 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
2448 TRIE_LIST_LEN( state ) = ging; \
2450 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2451 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2452 TRIE_LIST_CUR( state )++; \
2455 #define TRIE_LIST_NEW(state) STMT_START { \
2456 Newx( trie->states[ state ].trans.list, \
2457 4, reg_trie_trans_le ); \
2458 TRIE_LIST_CUR( state ) = 1; \
2459 TRIE_LIST_LEN( state ) = 4; \
2462 #define TRIE_HANDLE_WORD(state) STMT_START { \
2463 U16 dupe= trie->states[ state ].wordnum; \
2464 regnode * const noper_next = regnext( noper ); \
2467 /* store the word for dumping */ \
2469 if (OP(noper) != NOTHING) \
2470 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
2472 tmp = newSVpvn_utf8( "", 0, UTF ); \
2473 av_push( trie_words, tmp ); \
2477 trie->wordinfo[curword].prev = 0; \
2478 trie->wordinfo[curword].len = wordlen; \
2479 trie->wordinfo[curword].accept = state; \
2481 if ( noper_next < tail ) { \
2483 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2485 trie->jump[curword] = (U16)(noper_next - convert); \
2487 jumper = noper_next; \
2489 nextbranch= regnext(cur); \
2493 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2494 /* chain, so that when the bits of chain are later */\
2495 /* linked together, the dups appear in the chain */\
2496 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2497 trie->wordinfo[dupe].prev = curword; \
2499 /* we haven't inserted this word yet. */ \
2500 trie->states[ state ].wordnum = curword; \
2505 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2506 ( ( base + charid >= ucharcount \
2507 && base + charid < ubound \
2508 && state == trie->trans[ base - ucharcount + charid ].check \
2509 && trie->trans[ base - ucharcount + charid ].next ) \
2510 ? trie->trans[ base - ucharcount + charid ].next \
2511 : ( state==1 ? special : 0 ) \
2514 #define TRIE_BITMAP_SET_FOLDED(trie, uvc, folder) \
2516 TRIE_BITMAP_SET(trie, uvc); \
2517 /* store the folded codepoint */ \
2519 TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); \
2522 /* store first byte of utf8 representation of */ \
2523 /* variant codepoints */ \
2524 if (! UVCHR_IS_INVARIANT(uvc)) { \
2525 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); \
2530 #define MADE_JUMP_TRIE 2
2531 #define MADE_EXACT_TRIE 4
2534 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2535 regnode *first, regnode *last, regnode *tail,
2536 U32 word_count, U32 flags, U32 depth)
2538 /* first pass, loop through and scan words */
2539 reg_trie_data *trie;
2540 HV *widecharmap = NULL;
2541 AV *revcharmap = newAV();
2547 regnode *jumper = NULL;
2548 regnode *nextbranch = NULL;
2549 regnode *convert = NULL;
2550 U32 *prev_states; /* temp array mapping each state to previous one */
2551 /* we just use folder as a flag in utf8 */
2552 const U8 * folder = NULL;
2554 /* in the below add_data call we are storing either 'tu' or 'tuaa'
2555 * which stands for one trie structure, one hash, optionally followed
2558 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuaa"));
2559 AV *trie_words = NULL;
2560 /* along with revcharmap, this only used during construction but both are
2561 * useful during debugging so we store them in the struct when debugging.
2564 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
2565 STRLEN trie_charcount=0;
2567 SV *re_trie_maxbuff;
2568 GET_RE_DEBUG_FLAGS_DECL;
2570 PERL_ARGS_ASSERT_MAKE_TRIE;
2572 PERL_UNUSED_ARG(depth);
2576 case EXACT: case EXACTL: break;
2580 case EXACTFLU8: folder = PL_fold_latin1; break;
2581 case EXACTF: folder = PL_fold; break;
2582 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
2585 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
2587 trie->startstate = 1;
2588 trie->wordcount = word_count;
2589 RExC_rxi->data->data[ data_slot ] = (void*)trie;
2590 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
2591 if (flags == EXACT || flags == EXACTL)
2592 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2593 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2594 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2597 trie_words = newAV();
2600 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2601 assert(re_trie_maxbuff);
2602 if (!SvIOK(re_trie_maxbuff)) {
2603 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2605 DEBUG_TRIE_COMPILE_r({
2606 Perl_re_indentf( aTHX_
2607 "make_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2609 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
2610 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
2613 /* Find the node we are going to overwrite */
2614 if ( first == startbranch && OP( last ) != BRANCH ) {
2615 /* whole branch chain */
2618 /* branch sub-chain */
2619 convert = NEXTOPER( first );
2622 /* -- First loop and Setup --
2624 We first traverse the branches and scan each word to determine if it
2625 contains widechars, and how many unique chars there are, this is
2626 important as we have to build a table with at least as many columns as we
2629 We use an array of integers to represent the character codes 0..255
2630 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2631 the native representation of the character value as the key and IV's for
2634 *TODO* If we keep track of how many times each character is used we can
2635 remap the columns so that the table compression later on is more
2636 efficient in terms of memory by ensuring the most common value is in the
2637 middle and the least common are on the outside. IMO this would be better
2638 than a most to least common mapping as theres a decent chance the most
2639 common letter will share a node with the least common, meaning the node
2640 will not be compressible. With a middle is most common approach the worst
2641 case is when we have the least common nodes twice.
2645 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2646 regnode *noper = NEXTOPER( cur );
2650 U32 wordlen = 0; /* required init */
2651 STRLEN minchars = 0;
2652 STRLEN maxchars = 0;
2653 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2656 if (OP(noper) == NOTHING) {
2657 /* skip past a NOTHING at the start of an alternation
2658 * eg, /(?:)a|(?:b)/ should be the same as /a|b/
2660 regnode *noper_next= regnext(noper);
2661 if (noper_next < tail)
2665 if ( noper < tail &&
2667 OP(noper) == flags ||
2670 OP(noper) == EXACTFU_SS
2674 uc= (U8*)STRING(noper);
2675 e= uc + STR_LEN(noper);
2682 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
2683 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2684 regardless of encoding */
2685 if (OP( noper ) == EXACTFU_SS) {
2686 /* false positives are ok, so just set this */
2687 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
2691 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2693 TRIE_CHARCOUNT(trie)++;
2696 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2697 * is in effect. Under /i, this character can match itself, or
2698 * anything that folds to it. If not under /i, it can match just
2699 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2700 * all fold to k, and all are single characters. But some folds
2701 * expand to more than one character, so for example LATIN SMALL
2702 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2703 * the string beginning at 'uc' is 'ffi', it could be matched by
2704 * three characters, or just by the one ligature character. (It
2705 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2706 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2707 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2708 * match.) The trie needs to know the minimum and maximum number
2709 * of characters that could match so that it can use size alone to
2710 * quickly reject many match attempts. The max is simple: it is
2711 * the number of folded characters in this branch (since a fold is
2712 * never shorter than what folds to it. */
2716 /* And the min is equal to the max if not under /i (indicated by
2717 * 'folder' being NULL), or there are no multi-character folds. If
2718 * there is a multi-character fold, the min is incremented just
2719 * once, for the character that folds to the sequence. Each
2720 * character in the sequence needs to be added to the list below of
2721 * characters in the trie, but we count only the first towards the
2722 * min number of characters needed. This is done through the
2723 * variable 'foldlen', which is returned by the macros that look
2724 * for these sequences as the number of bytes the sequence
2725 * occupies. Each time through the loop, we decrement 'foldlen' by
2726 * how many bytes the current char occupies. Only when it reaches
2727 * 0 do we increment 'minchars' or look for another multi-character
2729 if (folder == NULL) {
2732 else if (foldlen > 0) {
2733 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
2738 /* See if *uc is the beginning of a multi-character fold. If
2739 * so, we decrement the length remaining to look at, to account
2740 * for the current character this iteration. (We can use 'uc'
2741 * instead of the fold returned by TRIE_READ_CHAR because for
2742 * non-UTF, the latin1_safe macro is smart enough to account
2743 * for all the unfolded characters, and because for UTF, the
2744 * string will already have been folded earlier in the
2745 * compilation process */
2747 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2748 foldlen -= UTF8SKIP(uc);
2751 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2756 /* The current character (and any potential folds) should be added
2757 * to the possible matching characters for this position in this
2761 U8 folded= folder[ (U8) uvc ];
2762 if ( !trie->charmap[ folded ] ) {
2763 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2764 TRIE_STORE_REVCHAR( folded );
2767 if ( !trie->charmap[ uvc ] ) {
2768 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
2769 TRIE_STORE_REVCHAR( uvc );
2772 /* store the codepoint in the bitmap, and its folded
2774 TRIE_BITMAP_SET_FOLDED(trie, uvc, folder);
2775 set_bit = 0; /* We've done our bit :-) */
2779 /* XXX We could come up with the list of code points that fold
2780 * to this using PL_utf8_foldclosures, except not for
2781 * multi-char folds, as there may be multiple combinations
2782 * there that could work, which needs to wait until runtime to
2783 * resolve (The comment about LIGATURE FFI above is such an
2788 widecharmap = newHV();
2790 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
2793 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%" UVXf, uvc );
2795 if ( !SvTRUE( *svpp ) ) {
2796 sv_setiv( *svpp, ++trie->uniquecharcount );
2797 TRIE_STORE_REVCHAR(uvc);
2800 } /* end loop through characters in this branch of the trie */
2802 /* We take the min and max for this branch and combine to find the min
2803 * and max for all branches processed so far */
2804 if( cur == first ) {
2805 trie->minlen = minchars;
2806 trie->maxlen = maxchars;
2807 } else if (minchars < trie->minlen) {
2808 trie->minlen = minchars;
2809 } else if (maxchars > trie->maxlen) {
2810 trie->maxlen = maxchars;
2812 } /* end first pass */
2813 DEBUG_TRIE_COMPILE_r(
2814 Perl_re_indentf( aTHX_
2815 "TRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2817 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
2818 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2819 (int)trie->minlen, (int)trie->maxlen )
2823 We now know what we are dealing with in terms of unique chars and
2824 string sizes so we can calculate how much memory a naive
2825 representation using a flat table will take. If it's over a reasonable
2826 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
2827 conservative but potentially much slower representation using an array
2830 At the end we convert both representations into the same compressed
2831 form that will be used in regexec.c for matching with. The latter
2832 is a form that cannot be used to construct with but has memory
2833 properties similar to the list form and access properties similar
2834 to the table form making it both suitable for fast searches and
2835 small enough that its feasable to store for the duration of a program.
2837 See the comment in the code where the compressed table is produced
2838 inplace from the flat tabe representation for an explanation of how
2839 the compression works.
2844 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2847 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2848 > SvIV(re_trie_maxbuff) )
2851 Second Pass -- Array Of Lists Representation
2853 Each state will be represented by a list of charid:state records
2854 (reg_trie_trans_le) the first such element holds the CUR and LEN
2855 points of the allocated array. (See defines above).
2857 We build the initial structure using the lists, and then convert
2858 it into the compressed table form which allows faster lookups
2859 (but cant be modified once converted).
2862 STRLEN transcount = 1;
2864 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using list compiler\n",
2867 trie->states = (reg_trie_state *)
2868 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2869 sizeof(reg_trie_state) );
2873 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2875 regnode *noper = NEXTOPER( cur );
2876 U32 state = 1; /* required init */
2877 U16 charid = 0; /* sanity init */
2878 U32 wordlen = 0; /* required init */
2880 if (OP(noper) == NOTHING) {
2881 regnode *noper_next= regnext(noper);
2882 if (noper_next < tail)
2886 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
2887 const U8 *uc= (U8*)STRING(noper);
2888 const U8 *e= uc + STR_LEN(noper);
2890 for ( ; uc < e ; uc += len ) {
2895 charid = trie->charmap[ uvc ];
2897 SV** const svpp = hv_fetch( widecharmap,
2904 charid=(U16)SvIV( *svpp );
2907 /* charid is now 0 if we dont know the char read, or
2908 * nonzero if we do */
2915 if ( !trie->states[ state ].trans.list ) {
2916 TRIE_LIST_NEW( state );
2919 check <= TRIE_LIST_USED( state );
2922 if ( TRIE_LIST_ITEM( state, check ).forid
2925 newstate = TRIE_LIST_ITEM( state, check ).newstate;
2930 newstate = next_alloc++;
2931 prev_states[newstate] = state;
2932 TRIE_LIST_PUSH( state, charid, newstate );
2937 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
2941 TRIE_HANDLE_WORD(state);
2943 } /* end second pass */
2945 /* next alloc is the NEXT state to be allocated */
2946 trie->statecount = next_alloc;
2947 trie->states = (reg_trie_state *)
2948 PerlMemShared_realloc( trie->states,
2950 * sizeof(reg_trie_state) );
2952 /* and now dump it out before we compress it */
2953 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
2954 revcharmap, next_alloc,
2958 trie->trans = (reg_trie_trans *)
2959 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
2966 for( state=1 ; state < next_alloc ; state ++ ) {
2970 DEBUG_TRIE_COMPILE_MORE_r(
2971 Perl_re_printf( aTHX_ "tp: %d zp: %d ",tp,zp)
2975 if (trie->states[state].trans.list) {
2976 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
2980 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
2981 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
2982 if ( forid < minid ) {
2984 } else if ( forid > maxid ) {
2988 if ( transcount < tp + maxid - minid + 1) {
2990 trie->trans = (reg_trie_trans *)
2991 PerlMemShared_realloc( trie->trans,
2993 * sizeof(reg_trie_trans) );
2994 Zero( trie->trans + (transcount / 2),
2998 base = trie->uniquecharcount + tp - minid;
2999 if ( maxid == minid ) {
3001 for ( ; zp < tp ; zp++ ) {
3002 if ( ! trie->trans[ zp ].next ) {
3003 base = trie->uniquecharcount + zp - minid;
3004 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
3006 trie->trans[ zp ].check = state;
3012 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
3014 trie->trans[ tp ].check = state;
3019 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
3020 const U32 tid = base
3021 - trie->uniquecharcount
3022 + TRIE_LIST_ITEM( state, idx ).forid;
3023 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
3025 trie->trans[ tid ].check = state;
3027 tp += ( maxid - minid + 1 );
3029 Safefree(trie->states[ state ].trans.list);
3032 DEBUG_TRIE_COMPILE_MORE_r(
3033 Perl_re_printf( aTHX_ " base: %d\n",base);
3036 trie->states[ state ].trans.base=base;
3038 trie->lasttrans = tp + 1;
3042 Second Pass -- Flat Table Representation.
3044 we dont use the 0 slot of either trans[] or states[] so we add 1 to
3045 each. We know that we will need Charcount+1 trans at most to store
3046 the data (one row per char at worst case) So we preallocate both
3047 structures assuming worst case.
3049 We then construct the trie using only the .next slots of the entry
3052 We use the .check field of the first entry of the node temporarily
3053 to make compression both faster and easier by keeping track of how
3054 many non zero fields are in the node.
3056 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
3059 There are two terms at use here: state as a TRIE_NODEIDX() which is
3060 a number representing the first entry of the node, and state as a
3061 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
3062 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
3063 if there are 2 entrys per node. eg:
3071 The table is internally in the right hand, idx form. However as we
3072 also have to deal with the states array which is indexed by nodenum
3073 we have to use TRIE_NODENUM() to convert.
3076 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using table compiler\n",
3079 trie->trans = (reg_trie_trans *)
3080 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
3081 * trie->uniquecharcount + 1,
3082 sizeof(reg_trie_trans) );
3083 trie->states = (reg_trie_state *)
3084 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
3085 sizeof(reg_trie_state) );
3086 next_alloc = trie->uniquecharcount + 1;
3089 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
3091 regnode *noper = NEXTOPER( cur );
3093 U32 state = 1; /* required init */
3095 U16 charid = 0; /* sanity init */
3096 U32 accept_state = 0; /* sanity init */
3098 U32 wordlen = 0; /* required init */
3100 if (OP(noper) == NOTHING) {
3101 regnode *noper_next= regnext(noper);
3102 if (noper_next < tail)
3106 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
3107 const U8 *uc= (U8*)STRING(noper);
3108 const U8 *e= uc + STR_LEN(noper);
3110 for ( ; uc < e ; uc += len ) {
3115 charid = trie->charmap[ uvc ];
3117 SV* const * const svpp = hv_fetch( widecharmap,
3121 charid = svpp ? (U16)SvIV(*svpp) : 0;
3125 if ( !trie->trans[ state + charid ].next ) {
3126 trie->trans[ state + charid ].next = next_alloc;
3127 trie->trans[ state ].check++;
3128 prev_states[TRIE_NODENUM(next_alloc)]
3129 = TRIE_NODENUM(state);
3130 next_alloc += trie->uniquecharcount;
3132 state = trie->trans[ state + charid ].next;
3134 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
3136 /* charid is now 0 if we dont know the char read, or
3137 * nonzero if we do */
3140 accept_state = TRIE_NODENUM( state );
3141 TRIE_HANDLE_WORD(accept_state);
3143 } /* end second pass */
3145 /* and now dump it out before we compress it */
3146 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
3148 next_alloc, depth+1));
3152 * Inplace compress the table.*
3154 For sparse data sets the table constructed by the trie algorithm will
3155 be mostly 0/FAIL transitions or to put it another way mostly empty.
3156 (Note that leaf nodes will not contain any transitions.)
3158 This algorithm compresses the tables by eliminating most such
3159 transitions, at the cost of a modest bit of extra work during lookup:
3161 - Each states[] entry contains a .base field which indicates the
3162 index in the state[] array wheres its transition data is stored.
3164 - If .base is 0 there are no valid transitions from that node.
3166 - If .base is nonzero then charid is added to it to find an entry in
3169 -If trans[states[state].base+charid].check!=state then the
3170 transition is taken to be a 0/Fail transition. Thus if there are fail
3171 transitions at the front of the node then the .base offset will point
3172 somewhere inside the previous nodes data (or maybe even into a node
3173 even earlier), but the .check field determines if the transition is
3177 The following process inplace converts the table to the compressed
3178 table: We first do not compress the root node 1,and mark all its
3179 .check pointers as 1 and set its .base pointer as 1 as well. This
3180 allows us to do a DFA construction from the compressed table later,
3181 and ensures that any .base pointers we calculate later are greater
3184 - We set 'pos' to indicate the first entry of the second node.
3186 - We then iterate over the columns of the node, finding the first and
3187 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
3188 and set the .check pointers accordingly, and advance pos
3189 appropriately and repreat for the next node. Note that when we copy
3190 the next pointers we have to convert them from the original
3191 NODEIDX form to NODENUM form as the former is not valid post
3194 - If a node has no transitions used we mark its base as 0 and do not
3195 advance the pos pointer.
3197 - If a node only has one transition we use a second pointer into the
3198 structure to fill in allocated fail transitions from other states.
3199 This pointer is independent of the main pointer and scans forward
3200 looking for null transitions that are allocated to a state. When it
3201 finds one it writes the single transition into the "hole". If the
3202 pointer doesnt find one the single transition is appended as normal.
3204 - Once compressed we can Renew/realloc the structures to release the
3207 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
3208 specifically Fig 3.47 and the associated pseudocode.
3212 const U32 laststate = TRIE_NODENUM( next_alloc );
3215 trie->statecount = laststate;
3217 for ( state = 1 ; state < laststate ; state++ ) {
3219 const U32 stateidx = TRIE_NODEIDX( state );
3220 const U32 o_used = trie->trans[ stateidx ].check;
3221 U32 used = trie->trans[ stateidx ].check;
3222 trie->trans[ stateidx ].check = 0;
3225 used && charid < trie->uniquecharcount;
3228 if ( flag || trie->trans[ stateidx + charid ].next ) {
3229 if ( trie->trans[ stateidx + charid ].next ) {
3231 for ( ; zp < pos ; zp++ ) {
3232 if ( ! trie->trans[ zp ].next ) {
3236 trie->states[ state ].trans.base
3238 + trie->uniquecharcount
3240 trie->trans[ zp ].next
3241 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
3243 trie->trans[ zp ].check = state;
3244 if ( ++zp > pos ) pos = zp;
3251 trie->states[ state ].trans.base
3252 = pos + trie->uniquecharcount - charid ;
3254 trie->trans[ pos ].next
3255 = SAFE_TRIE_NODENUM(
3256 trie->trans[ stateidx + charid ].next );
3257 trie->trans[ pos ].check = state;
3262 trie->lasttrans = pos + 1;
3263 trie->states = (reg_trie_state *)
3264 PerlMemShared_realloc( trie->states, laststate
3265 * sizeof(reg_trie_state) );
3266 DEBUG_TRIE_COMPILE_MORE_r(
3267 Perl_re_indentf( aTHX_ "Alloc: %d Orig: %" IVdf " elements, Final:%" IVdf ". Savings of %%%5.2f\n",
3269 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
3273 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
3276 } /* end table compress */
3278 DEBUG_TRIE_COMPILE_MORE_r(
3279 Perl_re_indentf( aTHX_ "Statecount:%" UVxf " Lasttrans:%" UVxf "\n",
3281 (UV)trie->statecount,
3282 (UV)trie->lasttrans)
3284 /* resize the trans array to remove unused space */
3285 trie->trans = (reg_trie_trans *)
3286 PerlMemShared_realloc( trie->trans, trie->lasttrans
3287 * sizeof(reg_trie_trans) );
3289 { /* Modify the program and insert the new TRIE node */
3290 U8 nodetype =(U8)(flags & 0xFF);
3294 regnode *optimize = NULL;
3295 #ifdef RE_TRACK_PATTERN_OFFSETS
3298 U32 mjd_nodelen = 0;
3299 #endif /* RE_TRACK_PATTERN_OFFSETS */
3300 #endif /* DEBUGGING */
3302 This means we convert either the first branch or the first Exact,
3303 depending on whether the thing following (in 'last') is a branch
3304 or not and whther first is the startbranch (ie is it a sub part of
3305 the alternation or is it the whole thing.)
3306 Assuming its a sub part we convert the EXACT otherwise we convert
3307 the whole branch sequence, including the first.
3309 /* Find the node we are going to overwrite */
3310 if ( first != startbranch || OP( last ) == BRANCH ) {
3311 /* branch sub-chain */
3312 NEXT_OFF( first ) = (U16)(last - first);
3313 #ifdef RE_TRACK_PATTERN_OFFSETS
3315 mjd_offset= Node_Offset((convert));
3316 mjd_nodelen= Node_Length((convert));
3319 /* whole branch chain */
3321 #ifdef RE_TRACK_PATTERN_OFFSETS
3324 const regnode *nop = NEXTOPER( convert );
3325 mjd_offset= Node_Offset((nop));
3326 mjd_nodelen= Node_Length((nop));
3330 Perl_re_indentf( aTHX_ "MJD offset:%" UVuf " MJD length:%" UVuf "\n",
3332 (UV)mjd_offset, (UV)mjd_nodelen)
3335 /* But first we check to see if there is a common prefix we can
3336 split out as an EXACT and put in front of the TRIE node. */
3337 trie->startstate= 1;
3338 if ( trie->bitmap && !widecharmap && !trie->jump ) {
3339 /* we want to find the first state that has more than
3340 * one transition, if that state is not the first state
3341 * then we have a common prefix which we can remove.
3344 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
3346 I32 first_ofs = -1; /* keeps track of the ofs of the first
3347 transition, -1 means none */
3349 const U32 base = trie->states[ state ].trans.base;
3351 /* does this state terminate an alternation? */
3352 if ( trie->states[state].wordnum )
3355 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
3356 if ( ( base + ofs >= trie->uniquecharcount ) &&
3357 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
3358 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
3360 if ( ++count > 1 ) {
3361 /* we have more than one transition */
3364 /* if this is the first state there is no common prefix
3365 * to extract, so we can exit */
3366 if ( state == 1 ) break;
3367 tmp = av_fetch( revcharmap, ofs, 0);
3368 ch = (U8*)SvPV_nolen_const( *tmp );
3370 /* if we are on count 2 then we need to initialize the
3371 * bitmap, and store the previous char if there was one
3374 /* clear the bitmap */
3375 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
3377 Perl_re_indentf( aTHX_ "New Start State=%" UVuf " Class: [",
3380 if (first_ofs >= 0) {
3381 SV ** const tmp = av_fetch( revcharmap, first_ofs, 0);
3382 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
3384 TRIE_BITMAP_SET_FOLDED(trie,*ch,folder);
3386 Perl_re_printf( aTHX_ "%s", (char*)ch)
3390 /* store the current firstchar in the bitmap */
3391 TRIE_BITMAP_SET_FOLDED(trie,*ch,folder);
3392 DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "%s", ch));
3398 /* This state has only one transition, its transition is part
3399 * of a common prefix - we need to concatenate the char it
3400 * represents to what we have so far. */
3401 SV **tmp = av_fetch( revcharmap, first_ofs, 0);
3403 char *ch = SvPV( *tmp, len );
3405 SV *sv=sv_newmortal();
3406 Perl_re_indentf( aTHX_ "Prefix State: %" UVuf " Ofs:%" UVuf " Char='%s'\n",
3408 (UV)state, (UV)first_ofs,
3409 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
3410 PL_colors[0], PL_colors[1],
3411 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
3412 PERL_PV_ESCAPE_FIRSTCHAR
3417 OP( convert ) = nodetype;
3418 str=STRING(convert);
3421 STR_LEN(convert) += len;
3427 DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "]\n"));
3432 trie->prefixlen = (state-1);
3434 regnode *n = convert+NODE_SZ_STR(convert);
3435 NEXT_OFF(convert) = NODE_SZ_STR(convert);
3436 trie->startstate = state;
3437 trie->minlen -= (state - 1);
3438 trie->maxlen -= (state - 1);
3440 /* At least the UNICOS C compiler choked on this
3441 * being argument to DEBUG_r(), so let's just have
3444 #ifdef PERL_EXT_RE_BUILD
3450 regnode *fix = convert;
3451 U32 word = trie->wordcount;
3453 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
3454 while( ++fix < n ) {
3455 Set_Node_Offset_Length(fix, 0, 0);
3458 SV ** const tmp = av_fetch( trie_words, word, 0 );
3460 if ( STR_LEN(convert) <= SvCUR(*tmp) )
3461 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
3463 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
3471 NEXT_OFF(convert) = (U16)(tail - convert);
3472 DEBUG_r(optimize= n);
3478 if ( trie->maxlen ) {
3479 NEXT_OFF( convert ) = (U16)(tail - convert);
3480 ARG_SET( convert, data_slot );
3481 /* Store the offset to the first unabsorbed branch in
3482 jump[0], which is otherwise unused by the jump logic.
3483 We use this when dumping a trie and during optimisation. */
3485 trie->jump[0] = (U16)(nextbranch - convert);
3487 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
3488 * and there is a bitmap
3489 * and the first "jump target" node we found leaves enough room
3490 * then convert the TRIE node into a TRIEC node, with the bitmap
3491 * embedded inline in the opcode - this is hypothetically faster.
3493 if ( !trie->states[trie->startstate].wordnum
3495 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
3497 OP( convert ) = TRIEC;
3498 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
3499 PerlMemShared_free(trie->bitmap);
3502 OP( convert ) = TRIE;
3504 /* store the type in the flags */
3505 convert->flags = nodetype;
3509 + regarglen[ OP( convert ) ];
3511 /* XXX We really should free up the resource in trie now,
3512 as we won't use them - (which resources?) dmq */
3514 /* needed for dumping*/
3515 DEBUG_r(if (optimize) {
3516 regnode *opt = convert;
3518 while ( ++opt < optimize) {
3519 Set_Node_Offset_Length(opt,0,0);
3522 Try to clean up some of the debris left after the
3525 while( optimize < jumper ) {
3526 mjd_nodelen += Node_Length((optimize));
3527 OP( optimize ) = OPTIMIZED;
3528 Set_Node_Offset_Length(optimize,0,0);
3531 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
3533 } /* end node insert */
3535 /* Finish populating the prev field of the wordinfo array. Walk back
3536 * from each accept state until we find another accept state, and if
3537 * so, point the first word's .prev field at the second word. If the
3538 * second already has a .prev field set, stop now. This will be the
3539 * case either if we've already processed that word's accept state,
3540 * or that state had multiple words, and the overspill words were
3541 * already linked up earlier.
3548 for (word=1; word <= trie->wordcount; word++) {
3550 if (trie->wordinfo[word].prev)
3552 state = trie->wordinfo[word].accept;
3554 state = prev_states[state];
3557 prev = trie->states[state].wordnum;
3561 trie->wordinfo[word].prev = prev;
3563 Safefree(prev_states);
3567 /* and now dump out the compressed format */
3568 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
3570 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
3572 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
3573 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
3575 SvREFCNT_dec_NN(revcharmap);
3579 : trie->startstate>1
3585 S_construct_ahocorasick_from_trie(pTHX_ RExC_state_t *pRExC_state, regnode *source, U32 depth)
3587 /* The Trie is constructed and compressed now so we can build a fail array if
3590 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and
3592 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi,
3596 We find the fail state for each state in the trie, this state is the longest
3597 proper suffix of the current state's 'word' that is also a proper prefix of
3598 another word in our trie. State 1 represents the word '' and is thus the
3599 default fail state. This allows the DFA not to have to restart after its
3600 tried and failed a word at a given point, it simply continues as though it
3601 had been matching the other word in the first place.
3603 'abcdgu'=~/abcdefg|cdgu/
3604 When we get to 'd' we are still matching the first word, we would encounter
3605 'g' which would fail, which would bring us to the state representing 'd' in
3606 the second word where we would try 'g' and succeed, proceeding to match
3609 /* add a fail transition */
3610 const U32 trie_offset = ARG(source);
3611 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
3613 const U32 ucharcount = trie->uniquecharcount;
3614 const U32 numstates = trie->statecount;
3615 const U32 ubound = trie->lasttrans + ucharcount;
3619 U32 base = trie->states[ 1 ].trans.base;
3622 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("T"));
3624 GET_RE_DEBUG_FLAGS_DECL;
3626 PERL_ARGS_ASSERT_CONSTRUCT_AHOCORASICK_FROM_TRIE;
3627 PERL_UNUSED_CONTEXT;
3629 PERL_UNUSED_ARG(depth);
3632 if ( OP(source) == TRIE ) {
3633 struct regnode_1 *op = (struct regnode_1 *)
3634 PerlMemShared_calloc(1, sizeof(struct regnode_1));
3635 StructCopy(source,op,struct regnode_1);
3636 stclass = (regnode *)op;
3638 struct regnode_charclass *op = (struct regnode_charclass *)
3639 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
3640 StructCopy(source,op,struct regnode_charclass);
3641 stclass = (regnode *)op;
3643 OP(stclass)+=2; /* convert the TRIE type to its AHO-CORASICK equivalent */
3645 ARG_SET( stclass, data_slot );
3646 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
3647 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3648 aho->trie=trie_offset;
3649 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
3650 Copy( trie->states, aho->states, numstates, reg_trie_state );
3651 Newx( q, numstates, U32);
3652 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
3655 /* initialize fail[0..1] to be 1 so that we always have
3656 a valid final fail state */
3657 fail[ 0 ] = fail[ 1 ] = 1;
3659 for ( charid = 0; charid < ucharcount ; charid++ ) {
3660 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
3662 q[ q_write ] = newstate;
3663 /* set to point at the root */
3664 fail[ q[ q_write++ ] ]=1;
3667 while ( q_read < q_write) {
3668 const U32 cur = q[ q_read++ % numstates ];
3669 base = trie->states[ cur ].trans.base;
3671 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
3672 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
3674 U32 fail_state = cur;
3677 fail_state = fail[ fail_state ];
3678 fail_base = aho->states[ fail_state ].trans.base;
3679 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
3681 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
3682 fail[ ch_state ] = fail_state;
3683 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
3685 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
3687 q[ q_write++ % numstates] = ch_state;
3691 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
3692 when we fail in state 1, this allows us to use the
3693 charclass scan to find a valid start char. This is based on the principle
3694 that theres a good chance the string being searched contains lots of stuff
3695 that cant be a start char.
3697 fail[ 0 ] = fail[ 1 ] = 0;
3698 DEBUG_TRIE_COMPILE_r({
3699 Perl_re_indentf( aTHX_ "Stclass Failtable (%" UVuf " states): 0",
3700 depth, (UV)numstates
3702 for( q_read=1; q_read<numstates; q_read++ ) {
3703 Perl_re_printf( aTHX_ ", %" UVuf, (UV)fail[q_read]);
3705 Perl_re_printf( aTHX_ "\n");
3708 /*RExC_seen |= REG_TRIEDFA_SEEN;*/
3713 /* The below joins as many adjacent EXACTish nodes as possible into a single
3714 * one. The regop may be changed if the node(s) contain certain sequences that
3715 * require special handling. The joining is only done if:
3716 * 1) there is room in the current conglomerated node to entirely contain the
3718 * 2) they are the exact same node type
3720 * The adjacent nodes actually may be separated by NOTHING-kind nodes, and
3721 * these get optimized out
3723 * XXX khw thinks this should be enhanced to fill EXACT (at least) nodes as full
3724 * as possible, even if that means splitting an existing node so that its first
3725 * part is moved to the preceeding node. This would maximise the efficiency of
3726 * memEQ during matching.
3728 * If a node is to match under /i (folded), the number of characters it matches
3729 * can be different than its character length if it contains a multi-character
3730 * fold. *min_subtract is set to the total delta number of characters of the
3733 * And *unfolded_multi_char is set to indicate whether or not the node contains
3734 * an unfolded multi-char fold. This happens when it won't be known until
3735 * runtime whether the fold is valid or not; namely
3736 * 1) for EXACTF nodes that contain LATIN SMALL LETTER SHARP S, as only if the
3737 * target string being matched against turns out to be UTF-8 is that fold
3739 * 2) for EXACTFL nodes whose folding rules depend on the locale in force at
3741 * (Multi-char folds whose components are all above the Latin1 range are not
3742 * run-time locale dependent, and have already been folded by the time this
3743 * function is called.)
3745 * This is as good a place as any to discuss the design of handling these
3746 * multi-character fold sequences. It's been wrong in Perl for a very long
3747 * time. There are three code points in Unicode whose multi-character folds
3748 * were long ago discovered to mess things up. The previous designs for
3749 * dealing with these involved assigning a special node for them. This
3750 * approach doesn't always work, as evidenced by this example:
3751 * "\xDFs" =~ /s\xDF/ui # Used to fail before these patches
3752 * Both sides fold to "sss", but if the pattern is parsed to create a node that
3753 * would match just the \xDF, it won't be able to handle the case where a
3754 * successful match would have to cross the node's boundary. The new approach
3755 * that hopefully generally solves the problem generates an EXACTFU_SS node
3756 * that is "sss" in this case.
3758 * It turns out that there are problems with all multi-character folds, and not
3759 * just these three. Now the code is general, for all such cases. The
3760 * approach taken is:
3761 * 1) This routine examines each EXACTFish node that could contain multi-
3762 * character folded sequences. Since a single character can fold into
3763 * such a sequence, the minimum match length for this node is less than
3764 * the number of characters in the node. This routine returns in
3765 * *min_subtract how many characters to subtract from the the actual
3766 * length of the string to get a real minimum match length; it is 0 if
3767 * there are no multi-char foldeds. This delta is used by the caller to
3768 * adjust the min length of the match, and the delta between min and max,
3769 * so that the optimizer doesn't reject these possibilities based on size
3771 * 2) For the sequence involving the Sharp s (\xDF), the node type EXACTFU_SS
3772 * is used for an EXACTFU node that contains at least one "ss" sequence in
3773 * it. For non-UTF-8 patterns and strings, this is the only case where
3774 * there is a possible fold length change. That means that a regular
3775 * EXACTFU node without UTF-8 involvement doesn't have to concern itself
3776 * with length changes, and so can be processed faster. regexec.c takes
3777 * advantage of this. Generally, an EXACTFish node that is in UTF-8 is
3778 * pre-folded by regcomp.c (except EXACTFL, some of whose folds aren't
3779 * known until runtime). This saves effort in regex matching. However,
3780 * the pre-folding isn't done for non-UTF8 patterns because the fold of
3781 * the MICRO SIGN requires UTF-8, and we don't want to slow things down by
3782 * forcing the pattern into UTF8 unless necessary. Also what EXACTF (and,
3783 * again, EXACTFL) nodes fold to isn't known until runtime. The fold
3784 * possibilities for the non-UTF8 patterns are quite simple, except for
3785 * the sharp s. All the ones that don't involve a UTF-8 target string are
3786 * members of a fold-pair, and arrays are set up for all of them so that
3787 * the other member of the pair can be found quickly. Code elsewhere in
3788 * this file makes sure that in EXACTFU nodes, the sharp s gets folded to
3789 * 'ss', even if the pattern isn't UTF-8. This avoids the issues
3790 * described in the next item.
3791 * 3) A problem remains for unfolded multi-char folds. (These occur when the
3792 * validity of the fold won't be known until runtime, and so must remain
3793 * unfolded for now. This happens for the sharp s in EXACTF and EXACTFAA
3794 * nodes when the pattern isn't in UTF-8. (Note, BTW, that there cannot
3795 * be an EXACTF node with a UTF-8 pattern.) They also occur for various
3796 * folds in EXACTFL nodes, regardless of the UTF-ness of the pattern.)
3797 * The reason this is a problem is that the optimizer part of regexec.c
3798 * (probably unwittingly, in Perl_regexec_flags()) makes an assumption
3799 * that a character in the pattern corresponds to at most a single
3800 * character in the target string. (And I do mean character, and not byte
3801 * here, unlike other parts of the documentation that have never been
3802 * updated to account for multibyte Unicode.) sharp s in EXACTF and
3803 * EXACTFL nodes can match the two character string 'ss'; in EXACTFAA
3804 * nodes it can match "\x{17F}\x{17F}". These, along with other ones in
3805 * EXACTFL nodes, violate the assumption, and they are the only instances
3806 * where it is violated. I'm reluctant to try to change the assumption,
3807 * as the code involved is impenetrable to me (khw), so instead the code
3808 * here punts. This routine examines EXACTFL nodes, and (when the pattern
3809 * isn't UTF-8) EXACTF and EXACTFAA for such unfolded folds, and returns a
3810 * boolean indicating whether or not the node contains such a fold. When
3811 * it is true, the caller sets a flag that later causes the optimizer in
3812 * this file to not set values for the floating and fixed string lengths,
3813 * and thus avoids the optimizer code in regexec.c that makes the invalid
3814 * assumption. Thus, there is no optimization based on string lengths for
3815 * EXACTFL nodes that contain these few folds, nor for non-UTF8-pattern
3816 * EXACTF and EXACTFAA nodes that contain the sharp s. (The reason the
3817 * assumption is wrong only in these cases is that all other non-UTF-8
3818 * folds are 1-1; and, for UTF-8 patterns, we pre-fold all other folds to
3819 * their expanded versions. (Again, we can't prefold sharp s to 'ss' in
3820 * EXACTF nodes because we don't know at compile time if it actually
3821 * matches 'ss' or not. For EXACTF nodes it will match iff the target
3822 * string is in UTF-8. This is in contrast to EXACTFU nodes, where it
3823 * always matches; and EXACTFAA where it never does. In an EXACTFAA node
3824 * in a UTF-8 pattern, sharp s is folded to "\x{17F}\x{17F}, avoiding the
3825 * problem; but in a non-UTF8 pattern, folding it to that above-Latin1
3826 * string would require the pattern to be forced into UTF-8, the overhead
3827 * of which we want to avoid. Similarly the unfolded multi-char folds in
3828 * EXACTFL nodes will match iff the locale at the time of match is a UTF-8
3831 * Similarly, the code that generates tries doesn't currently handle
3832 * not-already-folded multi-char folds, and it looks like a pain to change
3833 * that. Therefore, trie generation of EXACTFAA nodes with the sharp s
3834 * doesn't work. Instead, such an EXACTFAA is turned into a new regnode,
3835 * EXACTFAA_NO_TRIE, which the trie code knows not to handle. Most people
3836 * using /iaa matching will be doing so almost entirely with ASCII
3837 * strings, so this should rarely be encountered in practice */
3839 #define JOIN_EXACT(scan,min_subtract,unfolded_multi_char, flags) \
3840 if (PL_regkind[OP(scan)] == EXACT) \
3841 join_exact(pRExC_state,(scan),(min_subtract),unfolded_multi_char, (flags),NULL,depth+1)
3844 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan,
3845 UV *min_subtract, bool *unfolded_multi_char,
3846 U32 flags,regnode *val, U32 depth)
3848 /* Merge several consecutive EXACTish nodes into one. */
3849 regnode *n = regnext(scan);
3851 regnode *next = scan + NODE_SZ_STR(scan);
3855 regnode *stop = scan;
3856 GET_RE_DEBUG_FLAGS_DECL;
3858 PERL_UNUSED_ARG(depth);
3861 PERL_ARGS_ASSERT_JOIN_EXACT;
3862 #ifndef EXPERIMENTAL_INPLACESCAN
3863 PERL_UNUSED_ARG(flags);
3864 PERL_UNUSED_ARG(val);
3866 DEBUG_PEEP("join", scan, depth, 0);
3868 /* Look through the subsequent nodes in the chain. Skip NOTHING, merge
3869 * EXACT ones that are mergeable to the current one. */
3871 && (PL_regkind[OP(n)] == NOTHING
3872 || (stringok && OP(n) == OP(scan)))
3874 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX)
3877 if (OP(n) == TAIL || n > next)
3879 if (PL_regkind[OP(n)] == NOTHING) {
3880 DEBUG_PEEP("skip:", n, depth, 0);
3881 NEXT_OFF(scan) += NEXT_OFF(n);
3882 next = n + NODE_STEP_REGNODE;
3889 else if (stringok) {
3890 const unsigned int oldl = STR_LEN(scan);
3891 regnode * const nnext = regnext(n);
3893 /* XXX I (khw) kind of doubt that this works on platforms (should
3894 * Perl ever run on one) where U8_MAX is above 255 because of lots
3895 * of other assumptions */
3896 /* Don't join if the sum can't fit into a single node */
3897 if (oldl + STR_LEN(n) > U8_MAX)
3900 DEBUG_PEEP("merg", n, depth, 0);
3903 NEXT_OFF(scan) += NEXT_OFF(n);
3904 STR_LEN(scan) += STR_LEN(n);
3905 next = n + NODE_SZ_STR(n);
3906 /* Now we can overwrite *n : */
3907 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
3915 #ifdef EXPERIMENTAL_INPLACESCAN
3916 if (flags && !NEXT_OFF(n)) {
3917 DEBUG_PEEP("atch", val, depth, 0);
3918 if (reg_off_by_arg[OP(n)]) {
3919 ARG_SET(n, val - n);
3922 NEXT_OFF(n) = val - n;
3930 *unfolded_multi_char = FALSE;
3932 /* Here, all the adjacent mergeable EXACTish nodes have been merged. We
3933 * can now analyze for sequences of problematic code points. (Prior to
3934 * this final joining, sequences could have been split over boundaries, and
3935 * hence missed). The sequences only happen in folding, hence for any
3936 * non-EXACT EXACTish node */
3937 if (OP(scan) != EXACT && OP(scan) != EXACTL) {
3938 U8* s0 = (U8*) STRING(scan);
3940 U8* s_end = s0 + STR_LEN(scan);
3942 int total_count_delta = 0; /* Total delta number of characters that
3943 multi-char folds expand to */
3945 /* One pass is made over the node's string looking for all the
3946 * possibilities. To avoid some tests in the loop, there are two main
3947 * cases, for UTF-8 patterns (which can't have EXACTF nodes) and
3952 if (OP(scan) == EXACTFL) {
3955 /* An EXACTFL node would already have been changed to another
3956 * node type unless there is at least one character in it that
3957 * is problematic; likely a character whose fold definition
3958 * won't be known until runtime, and so has yet to be folded.
3959 * For all but the UTF-8 locale, folds are 1-1 in length, but
3960 * to handle the UTF-8 case, we need to create a temporary
3961 * folded copy using UTF-8 locale rules in order to analyze it.
3962 * This is because our macros that look to see if a sequence is
3963 * a multi-char fold assume everything is folded (otherwise the
3964 * tests in those macros would be too complicated and slow).
3965 * Note that here, the non-problematic folds will have already
3966 * been done, so we can just copy such characters. We actually
3967 * don't completely fold the EXACTFL string. We skip the
3968 * unfolded multi-char folds, as that would just create work
3969 * below to figure out the size they already are */
3971 Newx(folded, UTF8_MAX_FOLD_CHAR_EXPAND * STR_LEN(scan) + 1, U8);
3974 STRLEN s_len = UTF8SKIP(s);
3975 if (! is_PROBLEMATIC_LOCALE_FOLD_utf8(s)) {
3976 Copy(s, d, s_len, U8);
3979 else if (is_FOLDS_TO_MULTI_utf8(s)) {
3980 *unfolded_multi_char = TRUE;
3981 Copy(s, d, s_len, U8);
3984 else if (isASCII(*s)) {
3985 *(d++) = toFOLD(*s);
3989 _toFOLD_utf8_flags(s, s_end, d, &len, FOLD_FLAGS_FULL);
3995 /* Point the remainder of the routine to look at our temporary
3999 } /* End of creating folded copy of EXACTFL string */
4001 /* Examine the string for a multi-character fold sequence. UTF-8
4002 * patterns have all characters pre-folded by the time this code is
4004 while (s < s_end - 1) /* Can stop 1 before the end, as minimum
4005 length sequence we are looking for is 2 */
4007 int count = 0; /* How many characters in a multi-char fold */
4008 int len = is_MULTI_CHAR_FOLD_utf8_safe(s, s_end);
4009 if (! len) { /* Not a multi-char fold: get next char */
4014 /* Nodes with 'ss' require special handling, except for
4015 * EXACTFAA-ish for which there is no multi-char fold to this */
4016 if (len == 2 && *s == 's' && *(s+1) == 's'
4017 && OP(scan) != EXACTFAA
4018 && OP(scan) != EXACTFAA_NO_TRIE)
4021 if (OP(scan) != EXACTFL) {
4022 OP(scan) = EXACTFU_SS;
4026 else { /* Here is a generic multi-char fold. */
4027 U8* multi_end = s + len;
4029 /* Count how many characters are in it. In the case of
4030 * /aa, no folds which contain ASCII code points are
4031 * allowed, so check for those, and skip if found. */
4032 if (OP(scan) != EXACTFAA && OP(scan) != EXACTFAA_NO_TRIE) {
4033 count = utf8_length(s, multi_end);
4037 while (s < multi_end) {
4040 goto next_iteration;
4050 /* The delta is how long the sequence is minus 1 (1 is how long
4051 * the character that folds to the sequence is) */
4052 total_count_delta += count - 1;
4056 /* We created a temporary folded copy of the string in EXACTFL
4057 * nodes. Therefore we need to be sure it doesn't go below zero,
4058 * as the real string could be shorter */
4059 if (OP(scan) == EXACTFL) {
4060 int total_chars = utf8_length((U8*) STRING(scan),
4061 (U8*) STRING(scan) + STR_LEN(scan));
4062 if (total_count_delta > total_chars) {
4063 total_count_delta = total_chars;
4067 *min_subtract += total_count_delta;
4070 else if (OP(scan) == EXACTFAA) {
4072 /* Non-UTF-8 pattern, EXACTFAA node. There can't be a multi-char
4073 * fold to the ASCII range (and there are no existing ones in the
4074 * upper latin1 range). But, as outlined in the comments preceding
4075 * this function, we need to flag any occurrences of the sharp s.
4076 * This character forbids trie formation (because of added
4078 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
4079 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
4080 || UNICODE_DOT_DOT_VERSION > 0)
4082 if (*s == LATIN_SMALL_LETTER_SHARP_S) {
4083 OP(scan) = EXACTFAA_NO_TRIE;
4084 *unfolded_multi_char = TRUE;
4092 /* Non-UTF-8 pattern, not EXACTFAA node. Look for the multi-char
4093 * folds that are all Latin1. As explained in the comments
4094 * preceding this function, we look also for the sharp s in EXACTF
4095 * and EXACTFL nodes; it can be in the final position. Otherwise
4096 * we can stop looking 1 byte earlier because have to find at least
4097 * two characters for a multi-fold */
4098 const U8* upper = (OP(scan) == EXACTF || OP(scan) == EXACTFL)
4103 int len = is_MULTI_CHAR_FOLD_latin1_safe(s, s_end);
4104 if (! len) { /* Not a multi-char fold. */
4105 if (*s == LATIN_SMALL_LETTER_SHARP_S
4106 && (OP(scan) == EXACTF || OP(scan) == EXACTFL))
4108 *unfolded_multi_char = TRUE;
4115 && isALPHA_FOLD_EQ(*s, 's')
4116 && isALPHA_FOLD_EQ(*(s+1), 's'))
4119 /* EXACTF nodes need to know that the minimum length
4120 * changed so that a sharp s in the string can match this
4121 * ss in the pattern, but they remain EXACTF nodes, as they
4122 * won't match this unless the target string is is UTF-8,
4123 * which we don't know until runtime. EXACTFL nodes can't
4124 * transform into EXACTFU nodes */
4125 if (OP(scan) != EXACTF && OP(scan) != EXACTFL) {
4126 OP(scan) = EXACTFU_SS;
4130 *min_subtract += len - 1;
4138 /* Allow dumping but overwriting the collection of skipped
4139 * ops and/or strings with fake optimized ops */
4140 n = scan + NODE_SZ_STR(scan);
4148 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl", scan, depth, 0);});
4152 /* REx optimizer. Converts nodes into quicker variants "in place".
4153 Finds fixed substrings. */
4155 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
4156 to the position after last scanned or to NULL. */
4158 #define INIT_AND_WITHP \
4159 assert(!and_withp); \
4160 Newx(and_withp,1, regnode_ssc); \
4161 SAVEFREEPV(and_withp)
4165 S_unwind_scan_frames(pTHX_ const void *p)
4167 scan_frame *f= (scan_frame *)p;
4169 scan_frame *n= f->next_frame;
4175 /* the return from this sub is the minimum length that could possibly match */
4177 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
4178 SSize_t *minlenp, SSize_t *deltap,
4183 regnode_ssc *and_withp,
4184 U32 flags, U32 depth)
4185 /* scanp: Start here (read-write). */
4186 /* deltap: Write maxlen-minlen here. */
4187 /* last: Stop before this one. */
4188 /* data: string data about the pattern */
4189 /* stopparen: treat close N as END */
4190 /* recursed: which subroutines have we recursed into */
4191 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
4193 /* There must be at least this number of characters to match */
4196 regnode *scan = *scanp, *next;
4198 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
4199 int is_inf_internal = 0; /* The studied chunk is infinite */
4200 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
4201 scan_data_t data_fake;
4202 SV *re_trie_maxbuff = NULL;
4203 regnode *first_non_open = scan;
4204 SSize_t stopmin = SSize_t_MAX;
4205 scan_frame *frame = NULL;
4206 GET_RE_DEBUG_FLAGS_DECL;
4208 PERL_ARGS_ASSERT_STUDY_CHUNK;
4209 RExC_study_started= 1;
4211 Zero(&data_fake, 1, scan_data_t);
4214 while (first_non_open && OP(first_non_open) == OPEN)
4215 first_non_open=regnext(first_non_open);
4221 RExC_study_chunk_recursed_count++;
4223 DEBUG_OPTIMISE_MORE_r(
4225 Perl_re_indentf( aTHX_ "study_chunk stopparen=%ld recursed_count=%lu depth=%lu recursed_depth=%lu scan=%p last=%p",
4226 depth, (long)stopparen,
4227 (unsigned long)RExC_study_chunk_recursed_count,
4228 (unsigned long)depth, (unsigned long)recursed_depth,
4231 if (recursed_depth) {
4234 for ( j = 0 ; j < recursed_depth ; j++ ) {
4235 for ( i = 0 ; i < (U32)RExC_npar ; i++ ) {
4237 PAREN_TEST(RExC_study_chunk_recursed +
4238 ( j * RExC_study_chunk_recursed_bytes), i )
4241 !PAREN_TEST(RExC_study_chunk_recursed +
4242 (( j - 1 ) * RExC_study_chunk_recursed_bytes), i)
4245 Perl_re_printf( aTHX_ " %d",(int)i);
4249 if ( j + 1 < recursed_depth ) {
4250 Perl_re_printf( aTHX_ ",");
4254 Perl_re_printf( aTHX_ "\n");
4257 while ( scan && OP(scan) != END && scan < last ){
4258 UV min_subtract = 0; /* How mmany chars to subtract from the minimum
4259 node length to get a real minimum (because
4260 the folded version may be shorter) */
4261 bool unfolded_multi_char = FALSE;
4262 /* Peephole optimizer: */
4263 DEBUG_STUDYDATA("Peep", data, depth, is_inf);
4264 DEBUG_PEEP("Peep", scan, depth, flags);
4267 /* The reason we do this here is that we need to deal with things like
4268 * /(?:f)(?:o)(?:o)/ which cant be dealt with by the normal EXACT
4269 * parsing code, as each (?:..) is handled by a different invocation of
4272 JOIN_EXACT(scan,&min_subtract, &unfolded_multi_char, 0);
4274 /* Follow the next-chain of the current node and optimize
4275 away all the NOTHINGs from it. */
4276 if (OP(scan) != CURLYX) {
4277 const int max = (reg_off_by_arg[OP(scan)]
4279 /* I32 may be smaller than U16 on CRAYs! */
4280 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
4281 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
4285 /* Skip NOTHING and LONGJMP. */
4286 while ((n = regnext(n))
4287 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
4288 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
4289 && off + noff < max)
4291 if (reg_off_by_arg[OP(scan)])
4294 NEXT_OFF(scan) = off;
4297 /* The principal pseudo-switch. Cannot be a switch, since we
4298 look into several different things. */
4299 if ( OP(scan) == DEFINEP ) {
4301 SSize_t deltanext = 0;
4302 SSize_t fake_last_close = 0;
4303 I32 f = SCF_IN_DEFINE;
4305 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
4306 scan = regnext(scan);
4307 assert( OP(scan) == IFTHEN );
4308 DEBUG_PEEP("expect IFTHEN", scan, depth, flags);
4310 data_fake.last_closep= &fake_last_close;
4312 next = regnext(scan);
4313 scan = NEXTOPER(NEXTOPER(scan));
4314 DEBUG_PEEP("scan", scan, depth, flags);
4315 DEBUG_PEEP("next", next, depth, flags);
4317 /* we suppose the run is continuous, last=next...
4318 * NOTE we dont use the return here! */
4319 /* DEFINEP study_chunk() recursion */
4320 (void)study_chunk(pRExC_state, &scan, &minlen,
4321 &deltanext, next, &data_fake, stopparen,
4322 recursed_depth, NULL, f, depth+1);
4327 OP(scan) == BRANCH ||
4328 OP(scan) == BRANCHJ ||
4331 next = regnext(scan);
4334 /* The op(next)==code check below is to see if we
4335 * have "BRANCH-BRANCH", "BRANCHJ-BRANCHJ", "IFTHEN-IFTHEN"
4336 * IFTHEN is special as it might not appear in pairs.
4337 * Not sure whether BRANCH-BRANCHJ is possible, regardless
4338 * we dont handle it cleanly. */
4339 if (OP(next) == code || code == IFTHEN) {
4340 /* NOTE - There is similar code to this block below for
4341 * handling TRIE nodes on a re-study. If you change stuff here
4342 * check there too. */
4343 SSize_t max1 = 0, min1 = SSize_t_MAX, num = 0;
4345 regnode * const startbranch=scan;
4347 if (flags & SCF_DO_SUBSTR) {
4348 /* Cannot merge strings after this. */
4349 scan_commit(pRExC_state, data, minlenp, is_inf);
4352 if (flags & SCF_DO_STCLASS)
4353 ssc_init_zero(pRExC_state, &accum);
4355 while (OP(scan) == code) {
4356 SSize_t deltanext, minnext, fake;
4358 regnode_ssc this_class;
4360 DEBUG_PEEP("Branch", scan, depth, flags);
4363 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
4365 data_fake.whilem_c = data->whilem_c;
4366 data_fake.last_closep = data->last_closep;
4369 data_fake.last_closep = &fake;
4371 data_fake.pos_delta = delta;
4372 next = regnext(scan);
4374 scan = NEXTOPER(scan); /* everything */
4375 if (code != BRANCH) /* everything but BRANCH */
4376 scan = NEXTOPER(scan);
4378 if (flags & SCF_DO_STCLASS) {
4379 ssc_init(pRExC_state, &this_class);
4380 data_fake.start_class = &this_class;
4381 f = SCF_DO_STCLASS_AND;
4383 if (flags & SCF_WHILEM_VISITED_POS)
4384 f |= SCF_WHILEM_VISITED_POS;
4386 /* we suppose the run is continuous, last=next...*/
4387 /* recurse study_chunk() for each BRANCH in an alternation */
4388 minnext = study_chunk(pRExC_state, &scan, minlenp,
4389 &deltanext, next, &data_fake, stopparen,
4390 recursed_depth, NULL, f,depth+1);
4394 if (deltanext == SSize_t_MAX) {
4395 is_inf = is_inf_internal = 1;
4397 } else if (max1 < minnext + deltanext)
4398 max1 = minnext + deltanext;
4400 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4402 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4403 if ( stopmin > minnext)
4404 stopmin = min + min1;
4405 flags &= ~SCF_DO_SUBSTR;
4407 data->flags |= SCF_SEEN_ACCEPT;
4410 if (data_fake.flags & SF_HAS_EVAL)
4411 data->flags |= SF_HAS_EVAL;
4412 data->whilem_c = data_fake.whilem_c;
4414 if (flags & SCF_DO_STCLASS)
4415 ssc_or(pRExC_state, &accum, (regnode_charclass*)&this_class);
4417 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
4419 if (flags & SCF_DO_SUBSTR) {
4420 data->pos_min += min1;
4421 if (data->pos_delta >= SSize_t_MAX - (max1 - min1))
4422 data->pos_delta = SSize_t_MAX;
4424 data->pos_delta += max1 - min1;
4425 if (max1 != min1 || is_inf)
4426 data->cur_is_floating = 1;
4429 if (delta == SSize_t_MAX
4430 || SSize_t_MAX - delta - (max1 - min1) < 0)
4431 delta = SSize_t_MAX;
4433 delta += max1 - min1;
4434 if (flags & SCF_DO_STCLASS_OR) {
4435 ssc_or(pRExC_state, data->start_class, (regnode_charclass*) &accum);
4437 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp);
4438 flags &= ~SCF_DO_STCLASS;
4441 else if (flags & SCF_DO_STCLASS_AND) {
4443 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &accum);
4444 flags &= ~SCF_DO_STCLASS;
4447 /* Switch to OR mode: cache the old value of
4448 * data->start_class */
4450 StructCopy(data->start_class, and_withp, regnode_ssc);
4451 flags &= ~SCF_DO_STCLASS_AND;
4452 StructCopy(&accum, data->start_class, regnode_ssc);
4453 flags |= SCF_DO_STCLASS_OR;
4457 if (PERL_ENABLE_TRIE_OPTIMISATION &&
4458 OP( startbranch ) == BRANCH )
4462 Assuming this was/is a branch we are dealing with: 'scan'
4463 now points at the item that follows the branch sequence,
4464 whatever it is. We now start at the beginning of the
4465 sequence and look for subsequences of
4471 which would be constructed from a pattern like
4474 If we can find such a subsequence we need to turn the first
4475 element into a trie and then add the subsequent branch exact
4476 strings to the trie.
4480 1. patterns where the whole set of branches can be
4483 2. patterns where only a subset can be converted.
4485 In case 1 we can replace the whole set with a single regop
4486 for the trie. In case 2 we need to keep the start and end
4489 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
4490 becomes BRANCH TRIE; BRANCH X;
4492 There is an additional case, that being where there is a
4493 common prefix, which gets split out into an EXACT like node
4494 preceding the TRIE node.
4496 If x(1..n)==tail then we can do a simple trie, if not we make
4497 a "jump" trie, such that when we match the appropriate word
4498 we "jump" to the appropriate tail node. Essentially we turn
4499 a nested if into a case structure of sorts.
4504 if (!re_trie_maxbuff) {
4505 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
4506 if (!SvIOK(re_trie_maxbuff))
4507 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
4509 if ( SvIV(re_trie_maxbuff)>=0 ) {
4511 regnode *first = (regnode *)NULL;
4512 regnode *last = (regnode *)NULL;
4513 regnode *tail = scan;
4517 /* var tail is used because there may be a TAIL
4518 regop in the way. Ie, the exacts will point to the
4519 thing following the TAIL, but the last branch will
4520 point at the TAIL. So we advance tail. If we
4521 have nested (?:) we may have to move through several
4525 while ( OP( tail ) == TAIL ) {
4526 /* this is the TAIL generated by (?:) */
4527 tail = regnext( tail );
4531 DEBUG_TRIE_COMPILE_r({
4532 regprop(RExC_rx, RExC_mysv, tail, NULL, pRExC_state);
4533 Perl_re_indentf( aTHX_ "%s %" UVuf ":%s\n",
4535 "Looking for TRIE'able sequences. Tail node is ",
4536 (UV)(tail - RExC_emit_start),
4537 SvPV_nolen_const( RExC_mysv )
4543 Step through the branches
4544 cur represents each branch,
4545 noper is the first thing to be matched as part
4547 noper_next is the regnext() of that node.
4549 We normally handle a case like this
4550 /FOO[xyz]|BAR[pqr]/ via a "jump trie" but we also
4551 support building with NOJUMPTRIE, which restricts
4552 the trie logic to structures like /FOO|BAR/.
4554 If noper is a trieable nodetype then the branch is
4555 a possible optimization target. If we are building
4556 under NOJUMPTRIE then we require that noper_next is
4557 the same as scan (our current position in the regex
4560 Once we have two or more consecutive such branches
4561 we can create a trie of the EXACT's contents and
4562 stitch it in place into the program.
4564 If the sequence represents all of the branches in
4565 the alternation we replace the entire thing with a
4568 Otherwise when it is a subsequence we need to
4569 stitch it in place and replace only the relevant
4570 branches. This means the first branch has to remain
4571 as it is used by the alternation logic, and its
4572 next pointer, and needs to be repointed at the item
4573 on the branch chain following the last branch we
4574 have optimized away.
4576 This could be either a BRANCH, in which case the
4577 subsequence is internal, or it could be the item
4578 following the branch sequence in which case the
4579 subsequence is at the end (which does not
4580 necessarily mean the first node is the start of the
4583 TRIE_TYPE(X) is a define which maps the optype to a
4587 ----------------+-----------
4591 EXACTFU_SS | EXACTFU
4594 EXACTFLU8 | EXACTFLU8
4598 #define TRIE_TYPE(X) ( ( NOTHING == (X) ) \
4600 : ( EXACT == (X) ) \
4602 : ( EXACTFU == (X) || EXACTFU_SS == (X) ) \
4604 : ( EXACTFAA == (X) ) \
4606 : ( EXACTL == (X) ) \
4608 : ( EXACTFLU8 == (X) ) \
4612 /* dont use tail as the end marker for this traverse */
4613 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
4614 regnode * const noper = NEXTOPER( cur );
4615 U8 noper_type = OP( noper );
4616 U8 noper_trietype = TRIE_TYPE( noper_type );
4617 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
4618 regnode * const noper_next = regnext( noper );
4619 U8 noper_next_type = (noper_next && noper_next < tail) ? OP(noper_next) : 0;
4620 U8 noper_next_trietype = (noper_next && noper_next < tail) ? TRIE_TYPE( noper_next_type ) :0;
4623 DEBUG_TRIE_COMPILE_r({
4624 regprop(RExC_rx, RExC_mysv, cur, NULL, pRExC_state);
4625 Perl_re_indentf( aTHX_ "- %d:%s (%d)",
4627 REG_NODE_NUM(cur), SvPV_nolen_const( RExC_mysv ), REG_NODE_NUM(cur) );
4629 regprop(RExC_rx, RExC_mysv, noper, NULL, pRExC_state);
4630 Perl_re_printf( aTHX_ " -> %d:%s",
4631 REG_NODE_NUM(noper), SvPV_nolen_const(RExC_mysv));
4634 regprop(RExC_rx, RExC_mysv, noper_next, NULL, pRExC_state);
4635 Perl_re_printf( aTHX_ "\t=> %d:%s\t",
4636 REG_NODE_NUM(noper_next), SvPV_nolen_const(RExC_mysv));
4638 Perl_re_printf( aTHX_ "(First==%d,Last==%d,Cur==%d,tt==%s,ntt==%s,nntt==%s)\n",
4639 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur),
4640 PL_reg_name[trietype], PL_reg_name[noper_trietype], PL_reg_name[noper_next_trietype]
4644 /* Is noper a trieable nodetype that can be merged
4645 * with the current trie (if there is one)? */
4649 ( noper_trietype == NOTHING )
4650 || ( trietype == NOTHING )
4651 || ( trietype == noper_trietype )
4654 && noper_next >= tail
4658 /* Handle mergable triable node Either we are
4659 * the first node in a new trieable sequence,
4660 * in which case we do some bookkeeping,
4661 * otherwise we update the end pointer. */
4664 if ( noper_trietype == NOTHING ) {
4665 #if !defined(DEBUGGING) && !defined(NOJUMPTRIE)
4666 regnode * const noper_next = regnext( noper );
4667 U8 noper_next_type = (noper_next && noper_next < tail) ? OP(noper_next) : 0;
4668 U8 noper_next_trietype = noper_next_type ? TRIE_TYPE( noper_next_type ) :0;
4671 if ( noper_next_trietype ) {
4672 trietype = noper_next_trietype;
4673 } else if (noper_next_type) {
4674 /* a NOTHING regop is 1 regop wide.
4675 * We need at least two for a trie
4676 * so we can't merge this in */
4680 trietype = noper_trietype;
4683 if ( trietype == NOTHING )
4684 trietype = noper_trietype;
4689 } /* end handle mergable triable node */
4691 /* handle unmergable node -
4692 * noper may either be a triable node which can
4693 * not be tried together with the current trie,
4694 * or a non triable node */
4696 /* If last is set and trietype is not
4697 * NOTHING then we have found at least two
4698 * triable branch sequences in a row of a
4699 * similar trietype so we can turn them
4700 * into a trie. If/when we allow NOTHING to
4701 * start a trie sequence this condition
4702 * will be required, and it isn't expensive
4703 * so we leave it in for now. */
4704 if ( trietype && trietype != NOTHING )
4705 make_trie( pRExC_state,
4706 startbranch, first, cur, tail,
4707 count, trietype, depth+1 );
4708 last = NULL; /* note: we clear/update
4709 first, trietype etc below,
4710 so we dont do it here */
4714 && noper_next >= tail
4717 /* noper is triable, so we can start a new
4721 trietype = noper_trietype;
4723 /* if we already saw a first but the
4724 * current node is not triable then we have
4725 * to reset the first information. */
4730 } /* end handle unmergable node */
4731 } /* loop over branches */
4732 DEBUG_TRIE_COMPILE_r({
4733 regprop(RExC_rx, RExC_mysv, cur, NULL, pRExC_state);
4734 Perl_re_indentf( aTHX_ "- %s (%d) <SCAN FINISHED> ",
4735 depth+1, SvPV_nolen_const( RExC_mysv ),REG_NODE_NUM(cur));
4736 Perl_re_printf( aTHX_ "(First==%d, Last==%d, Cur==%d, tt==%s)\n",
4737 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur),
4738 PL_reg_name[trietype]
4742 if ( last && trietype ) {
4743 if ( trietype != NOTHING ) {
4744 /* the last branch of the sequence was part of
4745 * a trie, so we have to construct it here
4746 * outside of the loop */
4747 made= make_trie( pRExC_state, startbranch,
4748 first, scan, tail, count,
4749 trietype, depth+1 );
4750 #ifdef TRIE_STUDY_OPT
4751 if ( ((made == MADE_EXACT_TRIE &&
4752 startbranch == first)
4753 || ( first_non_open == first )) &&
4755 flags |= SCF_TRIE_RESTUDY;
4756 if ( startbranch == first
4759 RExC_seen &=~REG_TOP_LEVEL_BRANCHES_SEEN;
4764 /* at this point we know whatever we have is a
4765 * NOTHING sequence/branch AND if 'startbranch'
4766 * is 'first' then we can turn the whole thing
4769 if ( startbranch == first ) {
4771 /* the entire thing is a NOTHING sequence,
4772 * something like this: (?:|) So we can
4773 * turn it into a plain NOTHING op. */
4774 DEBUG_TRIE_COMPILE_r({
4775 regprop(RExC_rx, RExC_mysv, cur, NULL, pRExC_state);
4776 Perl_re_indentf( aTHX_ "- %s (%d) <NOTHING BRANCH SEQUENCE>\n",
4778 SvPV_nolen_const( RExC_mysv ),REG_NODE_NUM(cur));
4781 OP(startbranch)= NOTHING;
4782 NEXT_OFF(startbranch)= tail - startbranch;
4783 for ( opt= startbranch + 1; opt < tail ; opt++ )
4787 } /* end if ( last) */
4788 } /* TRIE_MAXBUF is non zero */
4793 else if ( code == BRANCHJ ) { /* single branch is optimized. */
4794 scan = NEXTOPER(NEXTOPER(scan));
4795 } else /* single branch is optimized. */
4796 scan = NEXTOPER(scan);
4798 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB) {
4800 regnode *start = NULL;
4801 regnode *end = NULL;
4802 U32 my_recursed_depth= recursed_depth;
4804 if (OP(scan) != SUSPEND) { /* GOSUB */
4805 /* Do setup, note this code has side effects beyond
4806 * the rest of this block. Specifically setting
4807 * RExC_recurse[] must happen at least once during
4810 RExC_recurse[ARG2L(scan)] = scan;
4811 start = RExC_open_parens[paren];
4812 end = RExC_close_parens[paren];
4814 /* NOTE we MUST always execute the above code, even
4815 * if we do nothing with a GOSUB */
4817 ( flags & SCF_IN_DEFINE )
4820 (is_inf_internal || is_inf || (data && data->flags & SF_IS_INF))
4822 ( (flags & (SCF_DO_STCLASS | SCF_DO_SUBSTR)) == 0 )
4825 /* no need to do anything here if we are in a define. */
4826 /* or we are after some kind of infinite construct
4827 * so we can skip recursing into this item.
4828 * Since it is infinite we will not change the maxlen
4829 * or delta, and if we miss something that might raise
4830 * the minlen it will merely pessimise a little.
4832 * Iow /(?(DEFINE)(?<foo>foo|food))a+(?&foo)/
4833 * might result in a minlen of 1 and not of 4,
4834 * but this doesn't make us mismatch, just try a bit
4835 * harder than we should.
4837 scan= regnext(scan);
4844 !PAREN_TEST(RExC_study_chunk_recursed + ((recursed_depth-1) * RExC_study_chunk_recursed_bytes), paren)
4846 /* it is quite possible that there are more efficient ways
4847 * to do this. We maintain a bitmap per level of recursion
4848 * of which patterns we have entered so we can detect if a
4849 * pattern creates a possible infinite loop. When we
4850 * recurse down a level we copy the previous levels bitmap
4851 * down. When we are at recursion level 0 we zero the top
4852 * level bitmap. It would be nice to implement a different
4853 * more efficient way of doing this. In particular the top
4854 * level bitmap may be unnecessary.
4856 if (!recursed_depth) {
4857 Zero(RExC_study_chunk_recursed, RExC_study_chunk_recursed_bytes, U8);
4859 Copy(RExC_study_chunk_recursed + ((recursed_depth-1) * RExC_study_chunk_recursed_bytes),
4860 RExC_study_chunk_recursed + (recursed_depth * RExC_study_chunk_recursed_bytes),
4861 RExC_study_chunk_recursed_bytes, U8);
4863 /* we havent recursed into this paren yet, so recurse into it */
4864 DEBUG_STUDYDATA("gosub-set", data, depth, is_inf);
4865 PAREN_SET(RExC_study_chunk_recursed + (recursed_depth * RExC_study_chunk_recursed_bytes), paren);
4866 my_recursed_depth= recursed_depth + 1;
4868 DEBUG_STUDYDATA("gosub-inf", data, depth, is_inf);
4869 /* some form of infinite recursion, assume infinite length
4871 if (flags & SCF_DO_SUBSTR) {
4872 scan_commit(pRExC_state, data, minlenp, is_inf);
4873 data->cur_is_floating = 1;
4875 is_inf = is_inf_internal = 1;
4876 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
4877 ssc_anything(data->start_class);
4878 flags &= ~SCF_DO_STCLASS;
4880 start= NULL; /* reset start so we dont recurse later on. */
4885 end = regnext(scan);
4888 scan_frame *newframe;
4890 if (!RExC_frame_last) {
4891 Newxz(newframe, 1, scan_frame);
4892 SAVEDESTRUCTOR_X(S_unwind_scan_frames, newframe);
4893 RExC_frame_head= newframe;
4895 } else if (!RExC_frame_last->next_frame) {
4896 Newxz(newframe,1,scan_frame);
4897 RExC_frame_last->next_frame= newframe;
4898 newframe->prev_frame= RExC_frame_last;
4901 newframe= RExC_frame_last->next_frame;
4903 RExC_frame_last= newframe;
4905 newframe->next_regnode = regnext(scan);
4906 newframe->last_regnode = last;
4907 newframe->stopparen = stopparen;
4908 newframe->prev_recursed_depth = recursed_depth;
4909 newframe->this_prev_frame= frame;
4911 DEBUG_STUDYDATA("frame-new", data, depth, is_inf);
4912 DEBUG_PEEP("fnew", scan, depth, flags);
4919 recursed_depth= my_recursed_depth;
4924 else if (OP(scan) == EXACT || OP(scan) == EXACTL) {
4925 SSize_t l = STR_LEN(scan);
4929 const U8 * const s = (U8*)STRING(scan);
4930 uc = utf8_to_uvchr_buf(s, s + l, NULL);
4931 l = utf8_length(s, s + l);
4933 uc = *((U8*)STRING(scan));
4936 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
4937 /* The code below prefers earlier match for fixed
4938 offset, later match for variable offset. */
4939 if (data->last_end == -1) { /* Update the start info. */
4940 data->last_start_min = data->pos_min;
4941 data->last_start_max = is_inf
4942 ? SSize_t_MAX : data->pos_min + data->pos_delta;
4944 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
4946 SvUTF8_on(data->last_found);
4948 SV * const sv = data->last_found;
4949 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
4950 mg_find(sv, PERL_MAGIC_utf8) : NULL;
4951 if (mg && mg->mg_len >= 0)
4952 mg->mg_len += utf8_length((U8*)STRING(scan),
4953 (U8*)STRING(scan)+STR_LEN(scan));
4955 data->last_end = data->pos_min + l;
4956 data->pos_min += l; /* As in the first entry. */
4957 data->flags &= ~SF_BEFORE_EOL;
4960 /* ANDing the code point leaves at most it, and not in locale, and
4961 * can't match null string */
4962 if (flags & SCF_DO_STCLASS_AND) {
4963 ssc_cp_and(data->start_class, uc);
4964 ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING;
4965 ssc_clear_locale(data->start_class);
4967 else if (flags & SCF_DO_STCLASS_OR) {
4968 ssc_add_cp(data->start_class, uc);
4969 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp);
4971 /* See commit msg 749e076fceedeb708a624933726e7989f2302f6a */
4972 ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING;
4974 flags &= ~SCF_DO_STCLASS;
4976 else if (PL_regkind[OP(scan)] == EXACT) {
4977 /* But OP != EXACT!, so is EXACTFish */
4978 SSize_t l = STR_LEN(scan);
4979 const U8 * s = (U8*)STRING(scan);
4981 /* Search for fixed substrings supports EXACT only. */
4982 if (flags & SCF_DO_SUBSTR) {
4984 scan_commit(pRExC_state, data, minlenp, is_inf);
4987 l = utf8_length(s, s + l);
4989 if (unfolded_multi_char) {
4990 RExC_seen |= REG_UNFOLDED_MULTI_SEEN;
4992 min += l - min_subtract;
4994 delta += min_subtract;
4995 if (flags & SCF_DO_SUBSTR) {
4996 data->pos_min += l - min_subtract;
4997 if (data->pos_min < 0) {
5000 data->pos_delta += min_subtract;
5002 data->cur_is_floating = 1; /* float */
5006 if (flags & SCF_DO_STCLASS) {
5007 SV* EXACTF_invlist = _make_exactf_invlist(pRExC_state, scan);
5009 assert(EXACTF_invlist);
5010 if (flags & SCF_DO_STCLASS_AND) {
5011 if (OP(scan) != EXACTFL)
5012 ssc_clear_locale(data->start_class);
5013 ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING;
5014 ANYOF_POSIXL_ZERO(data->start_class);
5015 ssc_intersection(data->start_class, EXACTF_invlist, FALSE);
5017 else { /* SCF_DO_STCLASS_OR */
5018 ssc_union(data->start_class, EXACTF_invlist, FALSE);
5019 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp);
5021 /* See commit msg 749e076fceedeb708a624933726e7989f2302f6a */
5022 ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING;
5024 flags &= ~SCF_DO_STCLASS;
5025 SvREFCNT_dec(EXACTF_invlist);
5028 else if (REGNODE_VARIES(OP(scan))) {
5029 SSize_t mincount, maxcount, minnext, deltanext, pos_before = 0;
5030 I32 fl = 0, f = flags;
5031 regnode * const oscan = scan;
5032 regnode_ssc this_class;
5033 regnode_ssc *oclass = NULL;
5034 I32 next_is_eval = 0;
5036 switch (PL_regkind[OP(scan)]) {
5037 case WHILEM: /* End of (?:...)* . */
5038 scan = NEXTOPER(scan);
5041 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
5042 next = NEXTOPER(scan);
5043 if (OP(next) == EXACT
5044 || OP(next) == EXACTL
5045 || (flags & SCF_DO_STCLASS))
5048 maxcount = REG_INFTY;
5049 next = regnext(scan);
5050 scan = NEXTOPER(scan);
5054 if (flags & SCF_DO_SUBSTR)
5059 if (flags & SCF_DO_STCLASS) {
5061 maxcount = REG_INFTY;
5062 next = regnext(scan);
5063 scan = NEXTOPER(scan);
5066 if (flags & SCF_DO_SUBSTR) {
5067 scan_commit(pRExC_state, data, minlenp, is_inf);
5068 /* Cannot extend fixed substrings */
5069 data->cur_is_floating = 1; /* float */
5071 is_inf = is_inf_internal = 1;
5072 scan = regnext(scan);
5073 goto optimize_curly_tail;
5075 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
5076 && (scan->flags == stopparen))
5081 mincount = ARG1(scan);
5082 maxcount = ARG2(scan);
5084 next = regnext(scan);
5085 if (OP(scan) == CURLYX) {
5086 I32 lp = (data ? *(data->last_closep) : 0);
5087 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
5089 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
5090 next_is_eval = (OP(scan) == EVAL);
5092 if (flags & SCF_DO_SUBSTR) {
5094 scan_commit(pRExC_state, data, minlenp, is_inf);
5095 /* Cannot extend fixed substrings */
5096 pos_before = data->pos_min;
5100 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
5102 data->flags |= SF_IS_INF;
5104 if (flags & SCF_DO_STCLASS) {
5105 ssc_init(pRExC_state, &this_class);
5106 oclass = data->start_class;
5107 data->start_class = &this_class;
5108 f |= SCF_DO_STCLASS_AND;
5109 f &= ~SCF_DO_STCLASS_OR;
5111 /* Exclude from super-linear cache processing any {n,m}
5112 regops for which the combination of input pos and regex
5113 pos is not enough information to determine if a match
5116 For example, in the regex /foo(bar\s*){4,8}baz/ with the
5117 regex pos at the \s*, the prospects for a match depend not
5118 only on the input position but also on how many (bar\s*)
5119 repeats into the {4,8} we are. */
5120 if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY))
5121 f &= ~SCF_WHILEM_VISITED_POS;
5123 /* This will finish on WHILEM, setting scan, or on NULL: */
5124 /* recurse study_chunk() on loop bodies */
5125 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
5126 last, data, stopparen, recursed_depth, NULL,
5128 ? (f & ~SCF_DO_SUBSTR)
5132 if (flags & SCF_DO_STCLASS)
5133 data->start_class = oclass;
5134 if (mincount == 0 || minnext == 0) {
5135 if (flags & SCF_DO_STCLASS_OR) {
5136 ssc_or(pRExC_state, data->start_class, (regnode_charclass *) &this_class);
5138 else if (flags & SCF_DO_STCLASS_AND) {
5139 /* Switch to OR mode: cache the old value of
5140 * data->start_class */
5142 StructCopy(data->start_class, and_withp, regnode_ssc);
5143 flags &= ~SCF_DO_STCLASS_AND;
5144 StructCopy(&this_class, data->start_class, regnode_ssc);
5145 flags |= SCF_DO_STCLASS_OR;
5146 ANYOF_FLAGS(data->start_class)
5147 |= SSC_MATCHES_EMPTY_STRING;
5149 } else { /* Non-zero len */
5150 if (flags & SCF_DO_STCLASS_OR) {
5151 ssc_or(pRExC_state, data->start_class, (regnode_charclass *) &this_class);
5152 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp);
5154 else if (flags & SCF_DO_STCLASS_AND)
5155 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &this_class);
5156 flags &= ~SCF_DO_STCLASS;
5158 if (!scan) /* It was not CURLYX, but CURLY. */
5160 if (((flags & (SCF_TRIE_DOING_RESTUDY|SCF_DO_SUBSTR))==SCF_DO_SUBSTR)
5161 /* ? quantifier ok, except for (?{ ... }) */
5162 && (next_is_eval || !(mincount == 0 && maxcount == 1))
5163 && (minnext == 0) && (deltanext == 0)
5164 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
5165 && maxcount <= REG_INFTY/3) /* Complement check for big
5168 /* Fatal warnings may leak the regexp without this: */
5169 SAVEFREESV(RExC_rx_sv);
5170 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP),
5171 "Quantifier unexpected on zero-length expression "
5172 "in regex m/%" UTF8f "/",
5173 UTF8fARG(UTF, RExC_precomp_end - RExC_precomp,
5175 (void)ReREFCNT_inc(RExC_rx_sv);
5178 min += minnext * mincount;
5179 is_inf_internal |= deltanext == SSize_t_MAX
5180 || (maxcount == REG_INFTY && minnext + deltanext > 0);
5181 is_inf |= is_inf_internal;
5183 delta = SSize_t_MAX;
5185 delta += (minnext + deltanext) * maxcount
5186 - minnext * mincount;
5188 /* Try powerful optimization CURLYX => CURLYN. */
5189 if ( OP(oscan) == CURLYX && data
5190 && data->flags & SF_IN_PAR
5191 && !(data->flags & SF_HAS_EVAL)
5192 && !deltanext && minnext == 1 ) {
5193 /* Try to optimize to CURLYN. */
5194 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
5195 regnode * const nxt1 = nxt;
5202 if (!REGNODE_SIMPLE(OP(nxt))
5203 && !(PL_regkind[OP(nxt)] == EXACT
5204 && STR_LEN(nxt) == 1))
5210 if (OP(nxt) != CLOSE)
5212 if (RExC_open_parens) {
5213 RExC_open_parens[ARG(nxt1)]=oscan; /*open->CURLYM*/
5214 RExC_close_parens[ARG(nxt1)]=nxt+2; /*close->while*/
5216 /* Now we know that nxt2 is the only contents: */
5217 oscan->flags = (U8)ARG(nxt);
5219 OP(nxt1) = NOTHING; /* was OPEN. */
5222 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
5223 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
5224 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
5225 OP(nxt) = OPTIMIZED; /* was CLOSE. */
5226 OP(nxt + 1) = OPTIMIZED; /* was count. */
5227 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
5232 /* Try optimization CURLYX => CURLYM. */
5233 if ( OP(oscan) == CURLYX && data
5234 && !(data->flags & SF_HAS_PAR)
5235 && !(data->flags & SF_HAS_EVAL)
5236 && !deltanext /* atom is fixed width */
5237 && minnext != 0 /* CURLYM can't handle zero width */
5239 /* Nor characters whose fold at run-time may be
5240 * multi-character */
5241 && ! (RExC_seen & REG_UNFOLDED_MULTI_SEEN)
5243 /* XXXX How to optimize if data == 0? */
5244 /* Optimize to a simpler form. */
5245 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
5249 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
5250 && (OP(nxt2) != WHILEM))
5252 OP(nxt2) = SUCCEED; /* Whas WHILEM */
5253 /* Need to optimize away parenths. */
5254 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
5255 /* Set the parenth number. */
5256 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
5258 oscan->flags = (U8)ARG(nxt);
5259 if (RExC_open_parens) {
5260 RExC_open_parens[ARG(nxt1)]=oscan; /*open->CURLYM*/
5261 RExC_close_parens[ARG(nxt1)]=nxt2+1; /*close->NOTHING*/
5263 OP(nxt1) = OPTIMIZED; /* was OPEN. */
5264 OP(nxt) = OPTIMIZED; /* was CLOSE. */
5267 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
5268 OP(nxt + 1) = OPTIMIZED; /* was count. */
5269 NEXT_OFF(nxt1 + 1) = 0; /* just for consistency. */
5270 NEXT_OFF(nxt + 1) = 0; /* just for consistency. */
5273 while ( nxt1 && (OP(nxt1) != WHILEM)) {
5274 regnode *nnxt = regnext(nxt1);
5276 if (reg_off_by_arg[OP(nxt1)])
5277 ARG_SET(nxt1, nxt2 - nxt1);
5278 else if (nxt2 - nxt1 < U16_MAX)
5279 NEXT_OFF(nxt1) = nxt2 - nxt1;
5281 OP(nxt) = NOTHING; /* Cannot beautify */
5286 /* Optimize again: */
5287 /* recurse study_chunk() on optimised CURLYX => CURLYM */
5288 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
5289 NULL, stopparen, recursed_depth, NULL, 0,depth+1);
5294 else if ((OP(oscan) == CURLYX)
5295 && (flags & SCF_WHILEM_VISITED_POS)
5296 /* See the comment on a similar expression above.
5297 However, this time it's not a subexpression
5298 we care about, but the expression itself. */
5299 && (maxcount == REG_INFTY)
5301 /* This stays as CURLYX, we can put the count/of pair. */
5302 /* Find WHILEM (as in regexec.c) */
5303 regnode *nxt = oscan + NEXT_OFF(oscan);
5305 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
5307 nxt = PREVOPER(nxt);
5308 if (nxt->flags & 0xf) {
5309 /* we've already set whilem count on this node */
5310 } else if (++data->whilem_c < 16) {
5311 assert(data->whilem_c <= RExC_whilem_seen);
5312 nxt->flags = (U8)(data->whilem_c
5313 | (RExC_whilem_seen << 4)); /* On WHILEM */
5316 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
5318 if (flags & SCF_DO_SUBSTR) {
5319 SV *last_str = NULL;
5320 STRLEN last_chrs = 0;
5321 int counted = mincount != 0;
5323 if (data->last_end > 0 && mincount != 0) { /* Ends with a
5325 SSize_t b = pos_before >= data->last_start_min
5326 ? pos_before : data->last_start_min;
5328 const char * const s = SvPV_const(data->last_found, l);
5329 SSize_t old = b - data->last_start_min;
5332 old = utf8_hop((U8*)s, old) - (U8*)s;
5334 /* Get the added string: */
5335 last_str = newSVpvn_utf8(s + old, l, UTF);
5336 last_chrs = UTF ? utf8_length((U8*)(s + old),
5337 (U8*)(s + old + l)) : l;
5338 if (deltanext == 0 && pos_before == b) {
5339 /* What was added is a constant string */
5342 SvGROW(last_str, (mincount * l) + 1);
5343 repeatcpy(SvPVX(last_str) + l,
5344 SvPVX_const(last_str), l,
5346 SvCUR_set(last_str, SvCUR(last_str) * mincount);
5347 /* Add additional parts. */
5348 SvCUR_set(data->last_found,
5349 SvCUR(data->last_found) - l);
5350 sv_catsv(data->last_found, last_str);
5352 SV * sv = data->last_found;
5354 SvUTF8(sv) && SvMAGICAL(sv) ?
5355 mg_find(sv, PERL_MAGIC_utf8) : NULL;
5356 if (mg && mg->mg_len >= 0)
5357 mg->mg_len += last_chrs * (mincount-1);
5359 last_chrs *= mincount;
5360 data->last_end += l * (mincount - 1);
5363 /* start offset must point into the last copy */
5364 data->last_start_min += minnext * (mincount - 1);
5365 data->last_start_max =
5368 : data->last_start_max +
5369 (maxcount - 1) * (minnext + data->pos_delta);
5372 /* It is counted once already... */
5373 data->pos_min += minnext * (mincount - counted);
5375 Perl_re_printf( aTHX_ "counted=%" UVuf " deltanext=%" UVuf
5376 " SSize_t_MAX=%" UVuf " minnext=%" UVuf
5377 " maxcount=%" UVuf " mincount=%" UVuf "\n",
5378 (UV)counted, (UV)deltanext, (UV)SSize_t_MAX, (UV)minnext, (UV)maxcount,
5380 if (deltanext != SSize_t_MAX)
5381 Perl_re_printf( aTHX_ "LHS=%" UVuf " RHS=%" UVuf "\n",
5382 (UV)(-counted * deltanext + (minnext + deltanext) * maxcount
5383 - minnext * mincount), (UV)(SSize_t_MAX - data->pos_delta));
5385 if (deltanext == SSize_t_MAX
5386 || -counted * deltanext + (minnext + deltanext) * maxcount - minnext * mincount >= SSize_t_MAX - data->pos_delta)
5387 data->pos_delta = SSize_t_MAX;
5389 data->pos_delta += - counted * deltanext +
5390 (minnext + deltanext) * maxcount - minnext * mincount;
5391 if (mincount != maxcount) {
5392 /* Cannot extend fixed substrings found inside
5394 scan_commit(pRExC_state, data, minlenp, is_inf);
5395 if (mincount && last_str) {
5396 SV * const sv = data->last_found;
5397 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
5398 mg_find(sv, PERL_MAGIC_utf8) : NULL;
5402 sv_setsv(sv, last_str);
5403 data->last_end = data->pos_min;
5404 data->last_start_min = data->pos_min - last_chrs;
5405 data->last_start_max = is_inf
5407 : data->pos_min + data->pos_delta - last_chrs;
5409 data->cur_is_floating = 1; /* float */
5411 SvREFCNT_dec(last_str);
5413 if (data && (fl & SF_HAS_EVAL))
5414 data->flags |= SF_HAS_EVAL;
5415 optimize_curly_tail:
5416 if (OP(oscan) != CURLYX) {
5417 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
5419 NEXT_OFF(oscan) += NEXT_OFF(next);
5425 Perl_croak(aTHX_ "panic: unexpected varying REx opcode %d",
5430 if (flags & SCF_DO_SUBSTR) {
5431 /* Cannot expect anything... */
5432 scan_commit(pRExC_state, data, minlenp, is_inf);
5433 data->cur_is_floating = 1; /* float */
5435 is_inf = is_inf_internal = 1;
5436 if (flags & SCF_DO_STCLASS_OR) {
5437 if (OP(scan) == CLUMP) {
5438 /* Actually is any start char, but very few code points
5439 * aren't start characters */
5440 ssc_match_all_cp(data->start_class);
5443 ssc_anything(data->start_class);
5446 flags &= ~SCF_DO_STCLASS;
5450 else if (OP(scan) == LNBREAK) {
5451 if (flags & SCF_DO_STCLASS) {
5452 if (flags & SCF_DO_STCLASS_AND) {
5453 ssc_intersection(data->start_class,
5454 PL_XPosix_ptrs[_CC_VERTSPACE], FALSE);
5455 ssc_clear_locale(data->start_class);
5456 ANYOF_FLAGS(data->start_class)
5457 &= ~SSC_MATCHES_EMPTY_STRING;
5459 else if (flags & SCF_DO_STCLASS_OR) {
5460 ssc_union(data->start_class,
5461 PL_XPosix_ptrs[_CC_VERTSPACE],
5463 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp);
5465 /* See commit msg for
5466 * 749e076fceedeb708a624933726e7989f2302f6a */
5467 ANYOF_FLAGS(data->start_class)
5468 &= ~SSC_MATCHES_EMPTY_STRING;
5470 flags &= ~SCF_DO_STCLASS;
5473 if (delta != SSize_t_MAX)
5474 delta++; /* Because of the 2 char string cr-lf */
5475 if (flags & SCF_DO_SUBSTR) {
5476 /* Cannot expect anything... */
5477 scan_commit(pRExC_state, data, minlenp, is_inf);
5479 data->pos_delta += 1;
5480 data->cur_is_floating = 1; /* float */
5483 else if (REGNODE_SIMPLE(OP(scan))) {
5485 if (flags & SCF_DO_SUBSTR) {
5486 scan_commit(pRExC_state, data, minlenp, is_inf);
5490 if (flags & SCF_DO_STCLASS) {
5492 SV* my_invlist = NULL;
5495 /* See commit msg 749e076fceedeb708a624933726e7989f2302f6a */
5496 ANYOF_FLAGS(data->start_class) &= ~SSC_MATCHES_EMPTY_STRING;
5498 /* Some of the logic below assumes that switching
5499 locale on will only add false positives. */
5504 Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d",
5508 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
5509 ssc_match_all_cp(data->start_class);
5514 SV* REG_ANY_invlist = _new_invlist(2);
5515 REG_ANY_invlist = add_cp_to_invlist(REG_ANY_invlist,
5517 if (flags & SCF_DO_STCLASS_OR) {
5518 ssc_union(data->start_class,
5520 TRUE /* TRUE => invert, hence all but \n
5524 else if (flags & SCF_DO_STCLASS_AND) {
5525 ssc_intersection(data->start_class,
5527 TRUE /* TRUE => invert */
5529 ssc_clear_locale(data->start_class);
5531 SvREFCNT_dec_NN(REG_ANY_invlist);
5538 if (flags & SCF_DO_STCLASS_AND)
5539 ssc_and(pRExC_state, data->start_class,
5540 (regnode_charclass *) scan);
5542 ssc_or(pRExC_state, data->start_class,
5543 (regnode_charclass *) scan);
5548 SV* cp_list = get_ANYOFM_contents(scan);
5550 if (flags & SCF_DO_STCLASS_OR) {
5551 ssc_union(data->start_class,
5553 FALSE /* don't invert */
5556 else if (flags & SCF_DO_STCLASS_AND) {
5557 ssc_intersection(data->start_class,
5559 FALSE /* don't invert */
5563 SvREFCNT_dec_NN(cp_list);
5572 namedclass = classnum_to_namedclass(FLAGS(scan)) + invert;
5573 if (flags & SCF_DO_STCLASS_AND) {
5574 bool was_there = cBOOL(
5575 ANYOF_POSIXL_TEST(data->start_class,
5577 ANYOF_POSIXL_ZERO(data->start_class);
5578 if (was_there) { /* Do an AND */
5579 ANYOF_POSIXL_SET(data->start_class, namedclass);
5581 /* No individual code points can now match */
5582 data->start_class->invlist
5583 = sv_2mortal(_new_invlist(0));
5586 int complement = namedclass + ((invert) ? -1 : 1);
5588 assert(flags & SCF_DO_STCLASS_OR);
5590 /* If the complement of this class was already there,
5591 * the result is that they match all code points,
5592 * (\d + \D == everything). Remove the classes from
5593 * future consideration. Locale is not relevant in
5595 if (ANYOF_POSIXL_TEST(data->start_class, complement)) {
5596 ssc_match_all_cp(data->start_class);
5597 ANYOF_POSIXL_CLEAR(data->start_class, namedclass);
5598 ANYOF_POSIXL_CLEAR(data->start_class, complement);
5600 else { /* The usual case; just add this class to the
5602 ANYOF_POSIXL_SET(data->start_class, namedclass);
5611 my_invlist = invlist_clone(PL_XPosix_ptrs[_CC_ASCII]);
5613 /* This can be handled as a Posix class */
5614 goto join_posix_and_ascii;
5616 case NPOSIXA: /* For these, we always know the exact set of
5621 assert(FLAGS(scan) != _CC_ASCII);
5622 _invlist_intersection(PL_XPosix_ptrs[FLAGS(scan)],
5623 PL_XPosix_ptrs[_CC_ASCII],
5625 goto join_posix_and_ascii;
5633 my_invlist = invlist_clone(PL_XPosix_ptrs[FLAGS(scan)]);
5635 /* NPOSIXD matches all upper Latin1 code points unless the
5636 * target string being matched is UTF-8, which is
5637 * unknowable until match time. Since we are going to
5638 * invert, we want to get rid of all of them so that the
5639 * inversion will match all */
5640 if (OP(scan) == NPOSIXD) {
5641 _invlist_subtract(my_invlist, PL_UpperLatin1,
5645 join_posix_and_ascii:
5647 if (flags & SCF_DO_STCLASS_AND) {
5648 ssc_intersection(data->start_class, my_invlist, invert);
5649 ssc_clear_locale(data->start_class);
5652 assert(flags & SCF_DO_STCLASS_OR);
5653 ssc_union(data->start_class, my_invlist, invert);
5655 SvREFCNT_dec(my_invlist);
5657 if (flags & SCF_DO_STCLASS_OR)
5658 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp);
5659 flags &= ~SCF_DO_STCLASS;
5662 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
5663 data->flags |= (OP(scan) == MEOL
5666 scan_commit(pRExC_state, data, minlenp, is_inf);
5669 else if ( PL_regkind[OP(scan)] == BRANCHJ
5670 /* Lookbehind, or need to calculate parens/evals/stclass: */
5671 && (scan->flags || data || (flags & SCF_DO_STCLASS))
5672 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM))
5674 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
5675 || OP(scan) == UNLESSM )
5677 /* Negative Lookahead/lookbehind
5678 In this case we can't do fixed string optimisation.
5681 SSize_t deltanext, minnext, fake = 0;
5686 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
5688 data_fake.whilem_c = data->whilem_c;
5689 data_fake.last_closep = data->last_closep;
5692 data_fake.last_closep = &fake;
5693 data_fake.pos_delta = delta;
5694 if ( flags & SCF_DO_STCLASS && !scan->flags
5695 && OP(scan) == IFMATCH ) { /* Lookahead */
5696 ssc_init(pRExC_state, &intrnl);
5697 data_fake.start_class = &intrnl;
5698 f |= SCF_DO_STCLASS_AND;
5700 if (flags & SCF_WHILEM_VISITED_POS)
5701 f |= SCF_WHILEM_VISITED_POS;
5702 next = regnext(scan);
5703 nscan = NEXTOPER(NEXTOPER(scan));
5705 /* recurse study_chunk() for lookahead body */
5706 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
5707 last, &data_fake, stopparen,
5708 recursed_depth, NULL, f, depth+1);
5711 FAIL("Variable length lookbehind not implemented");
5713 else if (minnext > (I32)U8_MAX) {
5714 FAIL2("Lookbehind longer than %" UVuf " not implemented",
5717 scan->flags = (U8)minnext;
5720 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
5722 if (data_fake.flags & SF_HAS_EVAL)
5723 data->flags |= SF_HAS_EVAL;
5724 data->whilem_c = data_fake.whilem_c;
5726 if (f & SCF_DO_STCLASS_AND) {
5727 if (flags & SCF_DO_STCLASS_OR) {
5728 /* OR before, AND after: ideally we would recurse with
5729 * data_fake to get the AND applied by study of the
5730 * remainder of the pattern, and then derecurse;
5731 * *** HACK *** for now just treat as "no information".
5732 * See [perl #56690].
5734 ssc_init(pRExC_state, data->start_class);
5736 /* AND before and after: combine and continue. These
5737 * assertions are zero-length, so can match an EMPTY
5739 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &intrnl);
5740 ANYOF_FLAGS(data->start_class)
5741 |= SSC_MATCHES_EMPTY_STRING;
5745 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
5747 /* Positive Lookahead/lookbehind
5748 In this case we can do fixed string optimisation,
5749 but we must be careful about it. Note in the case of
5750 lookbehind the positions will be offset by the minimum
5751 length of the pattern, something we won't know about
5752 until after the recurse.
5754 SSize_t deltanext, fake = 0;
5758 /* We use SAVEFREEPV so that when the full compile
5759 is finished perl will clean up the allocated
5760 minlens when it's all done. This way we don't
5761 have to worry about freeing them when we know
5762 they wont be used, which would be a pain.
5765 Newx( minnextp, 1, SSize_t );
5766 SAVEFREEPV(minnextp);
5769 StructCopy(data, &data_fake, scan_data_t);
5770 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
5773 scan_commit(pRExC_state, &data_fake, minlenp, is_inf);
5774 data_fake.last_found=newSVsv(data->last_found);
5778 data_fake.last_closep = &fake;
5779 data_fake.flags = 0;
5780 data_fake.substrs[0].flags = 0;
5781 data_fake.substrs[1].flags = 0;
5782 data_fake.pos_delta = delta;
5784 data_fake.flags |= SF_IS_INF;
5785 if ( flags & SCF_DO_STCLASS && !scan->flags
5786 && OP(scan) == IFMATCH ) { /* Lookahead */
5787 ssc_init(pRExC_state, &intrnl);
5788 data_fake.start_class = &intrnl;
5789 f |= SCF_DO_STCLASS_AND;
5791 if (flags & SCF_WHILEM_VISITED_POS)
5792 f |= SCF_WHILEM_VISITED_POS;
5793 next = regnext(scan);
5794 nscan = NEXTOPER(NEXTOPER(scan));
5796 /* positive lookahead study_chunk() recursion */
5797 *minnextp = study_chunk(pRExC_state, &nscan, minnextp,
5798 &deltanext, last, &data_fake,
5799 stopparen, recursed_depth, NULL,
5803 FAIL("Variable length lookbehind not implemented");
5805 else if (*minnextp > (I32)U8_MAX) {
5806 FAIL2("Lookbehind longer than %" UVuf " not implemented",
5809 scan->flags = (U8)*minnextp;
5814 if (f & SCF_DO_STCLASS_AND) {
5815 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &intrnl);
5816 ANYOF_FLAGS(data->start_class) |= SSC_MATCHES_EMPTY_STRING;
5819 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
5821 if (data_fake.flags & SF_HAS_EVAL)
5822 data->flags |= SF_HAS_EVAL;
5823 data->whilem_c = data_fake.whilem_c;
5824 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
5826 if (RExC_rx->minlen<*minnextp)
5827 RExC_rx->minlen=*minnextp;
5828 scan_commit(pRExC_state, &data_fake, minnextp, is_inf);
5829 SvREFCNT_dec_NN(data_fake.last_found);
5831 for (i = 0; i < 2; i++) {
5832 if (data_fake.substrs[i].minlenp != minlenp) {
5833 data->substrs[i].min_offset =
5834 data_fake.substrs[i].min_offset;
5835 data->substrs[i].max_offset =
5836 data_fake.substrs[i].max_offset;
5837 data->substrs[i].minlenp =
5838 data_fake.substrs[i].minlenp;
5839 data->substrs[i].lookbehind += scan->flags;
5848 else if (OP(scan) == OPEN) {
5849 if (stopparen != (I32)ARG(scan))
5852 else if (OP(scan) == CLOSE) {
5853 if (stopparen == (I32)ARG(scan)) {
5856 if ((I32)ARG(scan) == is_par) {
5857 next = regnext(scan);
5859 if ( next && (OP(next) != WHILEM) && next < last)
5860 is_par = 0; /* Disable optimization */
5863 *(data->last_closep) = ARG(scan);
5865 else if (OP(scan) == EVAL) {
5867 data->flags |= SF_HAS_EVAL;
5869 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
5870 if (flags & SCF_DO_SUBSTR) {
5871 scan_commit(pRExC_state, data, minlenp, is_inf);
5872 flags &= ~SCF_DO_SUBSTR;
5874 if (data && OP(scan)==ACCEPT) {
5875 data->flags |= SCF_SEEN_ACCEPT;
5880 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
5882 if (flags & SCF_DO_SUBSTR) {
5883 scan_commit(pRExC_state, data, minlenp, is_inf);
5884 data->cur_is_floating = 1; /* float */
5886 is_inf = is_inf_internal = 1;
5887 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
5888 ssc_anything(data->start_class);
5889 flags &= ~SCF_DO_STCLASS;
5891 else if (OP(scan) == GPOS) {
5892 if (!(RExC_rx->intflags & PREGf_GPOS_FLOAT) &&
5893 !(delta || is_inf || (data && data->pos_delta)))
5895 if (!(RExC_rx->intflags & PREGf_ANCH) && (flags & SCF_DO_SUBSTR))
5896 RExC_rx->intflags |= PREGf_ANCH_GPOS;
5897 if (RExC_rx->gofs < (STRLEN)min)
5898 RExC_rx->gofs = min;
5900 RExC_rx->intflags |= PREGf_GPOS_FLOAT;
5904 #ifdef TRIE_STUDY_OPT
5905 #ifdef FULL_TRIE_STUDY
5906 else if (PL_regkind[OP(scan)] == TRIE) {
5907 /* NOTE - There is similar code to this block above for handling
5908 BRANCH nodes on the initial study. If you change stuff here
5910 regnode *trie_node= scan;
5911 regnode *tail= regnext(scan);
5912 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
5913 SSize_t max1 = 0, min1 = SSize_t_MAX;
5916 if (flags & SCF_DO_SUBSTR) { /* XXXX Add !SUSPEND? */
5917 /* Cannot merge strings after this. */
5918 scan_commit(pRExC_state, data, minlenp, is_inf);
5920 if (flags & SCF_DO_STCLASS)
5921 ssc_init_zero(pRExC_state, &accum);
5927 const regnode *nextbranch= NULL;
5930 for ( word=1 ; word <= trie->wordcount ; word++)
5932 SSize_t deltanext=0, minnext=0, f = 0, fake;
5933 regnode_ssc this_class;
5935 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
5937 data_fake.whilem_c = data->whilem_c;
5938 data_fake.last_closep = data->last_closep;
5941 data_fake.last_closep = &fake;
5942 data_fake.pos_delta = delta;
5943 if (flags & SCF_DO_STCLASS) {
5944 ssc_init(pRExC_state, &this_class);
5945 data_fake.start_class = &this_class;
5946 f = SCF_DO_STCLASS_AND;
5948 if (flags & SCF_WHILEM_VISITED_POS)
5949 f |= SCF_WHILEM_VISITED_POS;
5951 if (trie->jump[word]) {
5953 nextbranch = trie_node + trie->jump[0];
5954 scan= trie_node + trie->jump[word];
5955 /* We go from the jump point to the branch that follows
5956 it. Note this means we need the vestigal unused
5957 branches even though they arent otherwise used. */
5958 /* optimise study_chunk() for TRIE */
5959 minnext = study_chunk(pRExC_state, &scan, minlenp,
5960 &deltanext, (regnode *)nextbranch, &data_fake,
5961 stopparen, recursed_depth, NULL, f,depth+1);
5963 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
5964 nextbranch= regnext((regnode*)nextbranch);
5966 if (min1 > (SSize_t)(minnext + trie->minlen))
5967 min1 = minnext + trie->minlen;
5968 if (deltanext == SSize_t_MAX) {
5969 is_inf = is_inf_internal = 1;
5971 } else if (max1 < (SSize_t)(minnext + deltanext + trie->maxlen))
5972 max1 = minnext + deltanext + trie->maxlen;
5974 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
5976 if (data_fake.flags & SCF_SEEN_ACCEPT) {
5977 if ( stopmin > min + min1)
5978 stopmin = min + min1;
5979 flags &= ~SCF_DO_SUBSTR;
5981 data->flags |= SCF_SEEN_ACCEPT;
5984 if (data_fake.flags & SF_HAS_EVAL)
5985 data->flags |= SF_HAS_EVAL;
5986 data->whilem_c = data_fake.whilem_c;
5988 if (flags & SCF_DO_STCLASS)
5989 ssc_or(pRExC_state, &accum, (regnode_charclass *) &this_class);
5992 if (flags & SCF_DO_SUBSTR) {
5993 data->pos_min += min1;
5994 data->pos_delta += max1 - min1;
5995 if (max1 != min1 || is_inf)
5996 data->cur_is_floating = 1; /* float */
5999 if (delta != SSize_t_MAX) {
6000 if (SSize_t_MAX - (max1 - min1) >= delta)
6001 delta += max1 - min1;
6003 delta = SSize_t_MAX;
6005 if (flags & SCF_DO_STCLASS_OR) {
6006 ssc_or(pRExC_state, data->start_class, (regnode_charclass *) &accum);
6008 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp);
6009 flags &= ~SCF_DO_STCLASS;
6012 else if (flags & SCF_DO_STCLASS_AND) {
6014 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) &accum);
6015 flags &= ~SCF_DO_STCLASS;
6018 /* Switch to OR mode: cache the old value of
6019 * data->start_class */
6021 StructCopy(data->start_class, and_withp, regnode_ssc);
6022 flags &= ~SCF_DO_STCLASS_AND;
6023 StructCopy(&accum, data->start_class, regnode_ssc);
6024 flags |= SCF_DO_STCLASS_OR;
6031 else if (PL_regkind[OP(scan)] == TRIE) {
6032 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
6035 min += trie->minlen;
6036 delta += (trie->maxlen - trie->minlen);
6037 flags &= ~SCF_DO_STCLASS; /* xxx */
6038 if (flags & SCF_DO_SUBSTR) {
6039 /* Cannot expect anything... */
6040 scan_commit(pRExC_state, data, minlenp, is_inf);
6041 data->pos_min += trie->minlen;
6042 data->pos_delta += (trie->maxlen - trie->minlen);
6043 if (trie->maxlen != trie->minlen)
6044 data->cur_is_floating = 1; /* float */
6046 if (trie->jump) /* no more substrings -- for now /grr*/
6047 flags &= ~SCF_DO_SUBSTR;
6049 #endif /* old or new */
6050 #endif /* TRIE_STUDY_OPT */
6052 /* Else: zero-length, ignore. */
6053 scan = regnext(scan);
6058 /* we need to unwind recursion. */
6061 DEBUG_STUDYDATA("frame-end", data, depth, is_inf);
6062 DEBUG_PEEP("fend", scan, depth, flags);
6064 /* restore previous context */
6065 last = frame->last_regnode;
6066 scan = frame->next_regnode;
6067 stopparen = frame->stopparen;
6068 recursed_depth = frame->prev_recursed_depth;
6070 RExC_frame_last = frame->prev_frame;
6071 frame = frame->this_prev_frame;
6072 goto fake_study_recurse;
6076 DEBUG_STUDYDATA("pre-fin", data, depth, is_inf);
6079 *deltap = is_inf_internal ? SSize_t_MAX : delta;
6081 if (flags & SCF_DO_SUBSTR && is_inf)
6082 data->pos_delta = SSize_t_MAX - data->pos_min;
6083 if (is_par > (I32)U8_MAX)
6085 if (is_par && pars==1 && data) {
6086 data->flags |= SF_IN_PAR;
6087 data->flags &= ~SF_HAS_PAR;
6089 else if (pars && data) {
6090 data->flags |= SF_HAS_PAR;
6091 data->flags &= ~SF_IN_PAR;
6093 if (flags & SCF_DO_STCLASS_OR)
6094 ssc_and(pRExC_state, data->start_class, (regnode_charclass *) and_withp);
6095 if (flags & SCF_TRIE_RESTUDY)
6096 data->flags |= SCF_TRIE_RESTUDY;
6098 DEBUG_STUDYDATA("post-fin", data, depth, is_inf);
6101 SSize_t final_minlen= min < stopmin ? min : stopmin;
6103 if (!(RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN)) {
6104 if (final_minlen > SSize_t_MAX - delta)
6105 RExC_maxlen = SSize_t_MAX;
6106 else if (RExC_maxlen < final_minlen + delta)
6107 RExC_maxlen = final_minlen + delta;
6109 return final_minlen;
6111 NOT_REACHED; /* NOTREACHED */
6115 S_add_data(RExC_state_t* const pRExC_state, const char* const s, const U32 n)
6117 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
6119 PERL_ARGS_ASSERT_ADD_DATA;
6121 Renewc(RExC_rxi->data,
6122 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
6123 char, struct reg_data);
6125 Renew(RExC_rxi->data->what, count + n, U8);
6127 Newx(RExC_rxi->data->what, n, U8);
6128 RExC_rxi->data->count = count + n;
6129 Copy(s, RExC_rxi->data->what + count, n, U8);
6133 /*XXX: todo make this not included in a non debugging perl, but appears to be
6134 * used anyway there, in 'use re' */
6135 #ifndef PERL_IN_XSUB_RE
6137 Perl_reginitcolors(pTHX)
6139 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
6141 char *t = savepv(s);
6145 t = strchr(t, '\t');
6151 PL_colors[i] = t = (char *)"";
6156 PL_colors[i++] = (char *)"";
6163 #ifdef TRIE_STUDY_OPT
6164 #define CHECK_RESTUDY_GOTO_butfirst(dOsomething) \
6167 (data.flags & SCF_TRIE_RESTUDY) \
6175 #define CHECK_RESTUDY_GOTO_butfirst
6179 * pregcomp - compile a regular expression into internal code
6181 * Decides which engine's compiler to call based on the hint currently in
6185 #ifndef PERL_IN_XSUB_RE
6187 /* return the currently in-scope regex engine (or the default if none) */
6189 regexp_engine const *
6190 Perl_current_re_engine(pTHX)
6192 if (IN_PERL_COMPILETIME) {
6193 HV * const table = GvHV(PL_hintgv);
6196 if (!table || !(PL_hints & HINT_LOCALIZE_HH))
6197 return &PL_core_reg_engine;
6198 ptr = hv_fetchs(table, "regcomp", FALSE);
6199 if ( !(ptr && SvIOK(*ptr) && SvIV(*ptr)))
6200 return &PL_core_reg_engine;
6201 return INT2PTR(regexp_engine*,SvIV(*ptr));
6205 if (!PL_curcop->cop_hints_hash)
6206 return &PL_core_reg_engine;
6207 ptr = cop_hints_fetch_pvs(PL_curcop, "regcomp", 0);
6208 if ( !(ptr && SvIOK(ptr) && SvIV(ptr)))
6209 return &PL_core_reg_engine;
6210 return INT2PTR(regexp_engine*,SvIV(ptr));
6216 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
6218 regexp_engine const *eng = current_re_engine();
6219 GET_RE_DEBUG_FLAGS_DECL;
6221 PERL_ARGS_ASSERT_PREGCOMP;
6223 /* Dispatch a request to compile a regexp to correct regexp engine. */
6225 Perl_re_printf( aTHX_ "Using engine %" UVxf "\n",
6228 return CALLREGCOMP_ENG(eng, pattern, flags);
6232 /* public(ish) entry point for the perl core's own regex compiling code.
6233 * It's actually a wrapper for Perl_re_op_compile that only takes an SV
6234 * pattern rather than a list of OPs, and uses the internal engine rather
6235 * than the current one */
6238 Perl_re_compile(pTHX_ SV * const pattern, U32 rx_flags)
6240 SV *pat = pattern; /* defeat constness! */
6241 PERL_ARGS_ASSERT_RE_COMPILE;
6242 return Perl_re_op_compile(aTHX_ &pat, 1, NULL,
6243 #ifdef PERL_IN_XSUB_RE
6246 &PL_core_reg_engine,
6248 NULL, NULL, rx_flags, 0);
6253 S_free_codeblocks(pTHX_ struct reg_code_blocks *cbs)
6257 if (--cbs->refcnt > 0)
6259 for (n = 0; n < cbs->count; n++) {
6260 REGEXP *rx = cbs->cb[n].src_regex;
6261 cbs->cb[n].src_regex = NULL;
6269 static struct reg_code_blocks *
6270 S_alloc_code_blocks(pTHX_ int ncode)
6272 struct reg_code_blocks *cbs;
6273 Newx(cbs, 1, struct reg_code_blocks);
6276 SAVEDESTRUCTOR_X(S_free_codeblocks, cbs);
6278 Newx(cbs->cb, ncode, struct reg_code_block);
6285 /* upgrade pattern pat_p of length plen_p to UTF8, and if there are code
6286 * blocks, recalculate the indices. Update pat_p and plen_p in-place to
6287 * point to the realloced string and length.
6289 * This is essentially a copy of Perl_bytes_to_utf8() with the code index
6293 S_pat_upgrade_to_utf8(pTHX_ RExC_state_t * const pRExC_state,
6294 char **pat_p, STRLEN *plen_p, int num_code_blocks)
6296 U8 *const src = (U8*)*pat_p;
6301 GET_RE_DEBUG_FLAGS_DECL;
6303 DEBUG_PARSE_r(Perl_re_printf( aTHX_
6304 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
6306 Newx(dst, *plen_p * 2 + 1, U8);
6309 while (s < *plen_p) {
6310 append_utf8_from_native_byte(src[s], &d);
6312 if (n < num_code_blocks) {
6313 assert(pRExC_state->code_blocks);
6314 if (!do_end && pRExC_state->code_blocks->cb[n].start == s) {
6315 pRExC_state->code_blocks->cb[n].start = d - dst - 1;
6316 assert(*(d - 1) == '(');
6319 else if (do_end && pRExC_state->code_blocks->cb[n].end == s) {
6320 pRExC_state->code_blocks->cb[n].end = d - dst - 1;
6321 assert(*(d - 1) == ')');
6330 *pat_p = (char*) dst;
6332 RExC_orig_utf8 = RExC_utf8 = 1;
6337 /* S_concat_pat(): concatenate a list of args to the pattern string pat,
6338 * while recording any code block indices, and handling overloading,
6339 * nested qr// objects etc. If pat is null, it will allocate a new
6340 * string, or just return the first arg, if there's only one.
6342 * Returns the malloced/updated pat.
6343 * patternp and pat_count is the array of SVs to be concatted;
6344 * oplist is the optional list of ops that generated the SVs;
6345 * recompile_p is a pointer to a boolean that will be set if
6346 * the regex will need to be recompiled.
6347 * delim, if non-null is an SV that will be inserted between each element
6351 S_concat_pat(pTHX_ RExC_state_t * const pRExC_state,
6352 SV *pat, SV ** const patternp, int pat_count,
6353 OP *oplist, bool *recompile_p, SV *delim)
6357 bool use_delim = FALSE;
6358 bool alloced = FALSE;
6360 /* if we know we have at least two args, create an empty string,
6361 * then concatenate args to that. For no args, return an empty string */
6362 if (!pat && pat_count != 1) {
6368 for (svp = patternp; svp < patternp + pat_count; svp++) {
6371 STRLEN orig_patlen = 0;
6373 SV *msv = use_delim ? delim : *svp;
6374 if (!msv) msv = &PL_sv_undef;
6376 /* if we've got a delimiter, we go round the loop twice for each
6377 * svp slot (except the last), using the delimiter the second
6386 if (SvTYPE(msv) == SVt_PVAV) {
6387 /* we've encountered an interpolated array within
6388 * the pattern, e.g. /...@a..../. Expand the list of elements,
6389 * then recursively append elements.
6390 * The code in this block is based on S_pushav() */
6392 AV *const av = (AV*)msv;
6393 const SSize_t maxarg = AvFILL(av) + 1;
6397 assert(oplist->op_type == OP_PADAV
6398 || oplist->op_type == OP_RV2AV);
6399 oplist = OpSIBLING(oplist);
6402 if (SvRMAGICAL(av)) {
6405 Newx(array, maxarg, SV*);
6407 for (i=0; i < maxarg; i++) {
6408 SV ** const svp = av_fetch(av, i, FALSE);
6409 array[i] = svp ? *svp : &PL_sv_undef;
6413 array = AvARRAY(av);
6415 pat = S_concat_pat(aTHX_ pRExC_state, pat,
6416 array, maxarg, NULL, recompile_p,
6418 GvSV((gv_fetchpvs("\"", GV_ADDMULTI, SVt_PV))));
6424 /* we make the assumption here that each op in the list of
6425 * op_siblings maps to one SV pushed onto the stack,
6426 * except for code blocks, with have both an OP_NULL and
6428 * This allows us to match up the list of SVs against the
6429 * list of OPs to find the next code block.
6431 * Note that PUSHMARK PADSV PADSV ..
6433 * PADRANGE PADSV PADSV ..
6434 * so the alignment still works. */
6437 if (oplist->op_type == OP_NULL
6438 && (oplist->op_flags & OPf_SPECIAL))
6440 assert(n < pRExC_state->code_blocks->count);
6441 pRExC_state->code_blocks->cb[n].start = pat ? SvCUR(pat) : 0;
6442 pRExC_state->code_blocks->cb[n].block = oplist;
6443 pRExC_state->code_blocks->cb[n].src_regex = NULL;
6446 oplist = OpSIBLING(oplist); /* skip CONST */
6449 oplist = OpSIBLING(oplist);;
6452 /* apply magic and QR overloading to arg */
6455 if (SvROK(msv) && SvAMAGIC(msv)) {
6456 SV *sv = AMG_CALLunary(msv, regexp_amg);
6460 if (SvTYPE(sv) != SVt_REGEXP)
6461 Perl_croak(aTHX_ "Overloaded qr did not return a REGEXP");
6466 /* try concatenation overload ... */
6467 if (pat && (SvAMAGIC(pat) || SvAMAGIC(msv)) &&
6468 (sv = amagic_call(pat, msv, concat_amg, AMGf_assign)))
6471 /* overloading involved: all bets are off over literal
6472 * code. Pretend we haven't seen it */
6474 pRExC_state->code_blocks->count -= n;
6478 /* ... or failing that, try "" overload */
6479 while (SvAMAGIC(msv)
6480 && (sv = AMG_CALLunary(msv, string_amg))
6484 && SvRV(msv) == SvRV(sv))
6489 if (SvROK(msv) && SvTYPE(SvRV(msv)) == SVt_REGEXP)
6493 /* this is a partially unrolled
6494 * sv_catsv_nomg(pat, msv);
6495 * that allows us to adjust code block indices if
6498 char *dst = SvPV_force_nomg(pat, dlen);
6500 if (SvUTF8(msv) && !SvUTF8(pat)) {
6501 S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &dst, &dlen, n);
6502 sv_setpvn(pat, dst, dlen);
6505 sv_catsv_nomg(pat, msv);
6509 /* We have only one SV to process, but we need to verify
6510 * it is properly null terminated or we will fail asserts
6511 * later. In theory we probably shouldn't get such SV's,
6512 * but if we do we should handle it gracefully. */
6513 if ( SvTYPE(msv) != SVt_PV || (SvLEN(msv) > SvCUR(msv) && *(SvEND(msv)) == 0) || SvIsCOW_shared_hash(msv) ) {
6514 /* not a string, or a string with a trailing null */
6517 /* a string with no trailing null, we need to copy it
6518 * so it has a trailing null */
6519 pat = sv_2mortal(newSVsv(msv));
6524 pRExC_state->code_blocks->cb[n-1].end = SvCUR(pat)-1;
6527 /* extract any code blocks within any embedded qr//'s */
6528 if (rx && SvTYPE(rx) == SVt_REGEXP
6529 && RX_ENGINE((REGEXP*)rx)->op_comp)
6532 RXi_GET_DECL(ReANY((REGEXP *)rx), ri);
6533 if (ri->code_blocks && ri->code_blocks->count) {
6535 /* the presence of an embedded qr// with code means
6536 * we should always recompile: the text of the
6537 * qr// may not have changed, but it may be a
6538 * different closure than last time */
6540 if (pRExC_state->code_blocks) {
6541 int new_count = pRExC_state->code_blocks->count
6542 + ri->code_blocks->count;
6543 Renew(pRExC_state->code_blocks->cb,
6544 new_count, struct reg_code_block);
6545 pRExC_state->code_blocks->count = new_count;
6548 pRExC_state->code_blocks = S_alloc_code_blocks(aTHX_
6549 ri->code_blocks->count);
6551 for (i=0; i < ri->code_blocks->count; i++) {
6552 struct reg_code_block *src, *dst;
6553 STRLEN offset = orig_patlen
6554 + ReANY((REGEXP *)rx)->pre_prefix;
6555 assert(n < pRExC_state->code_blocks->count);
6556 src = &ri->code_blocks->cb[i];
6557 dst = &pRExC_state->code_blocks->cb[n];
6558 dst->start = src->start + offset;
6559 dst->end = src->end + offset;
6560 dst->block = src->block;
6561 dst->src_regex = (REGEXP*) SvREFCNT_inc( (SV*)
6570 /* avoid calling magic multiple times on a single element e.g. =~ $qr */
6579 /* see if there are any run-time code blocks in the pattern.
6580 * False positives are allowed */
6583 S_has_runtime_code(pTHX_ RExC_state_t * const pRExC_state,
6584 char *pat, STRLEN plen)
6589 PERL_UNUSED_CONTEXT;
6591 for (s = 0; s < plen; s++) {
6592 if ( pRExC_state->code_blocks
6593 && n < pRExC_state->code_blocks->count
6594 && s == pRExC_state->code_blocks->cb[n].start)
6596 s = pRExC_state->code_blocks->cb[n].end;
6600 /* TODO ideally should handle [..], (#..), /#.../x to reduce false
6602 if (pat[s] == '(' && s+2 <= plen && pat[s+1] == '?' &&
6604 || (s + 2 <= plen && pat[s+2] == '?' && pat[s+3] == '{'))
6611 /* Handle run-time code blocks. We will already have compiled any direct
6612 * or indirect literal code blocks. Now, take the pattern 'pat' and make a
6613 * copy of it, but with any literal code blocks blanked out and
6614 * appropriate chars escaped; then feed it into
6616 * eval "qr'modified_pattern'"
6620 * a\bc(?{"this was literal"})def'ghi\\jkl(?{"this is runtime"})mno
6624 * qr'a\\bc_______________________def\'ghi\\\\jkl(?{"this is runtime"})mno'
6626 * After eval_sv()-ing that, grab any new code blocks from the returned qr
6627 * and merge them with any code blocks of the original regexp.
6629 * If the pat is non-UTF8, while the evalled qr is UTF8, don't merge;
6630 * instead, just save the qr and return FALSE; this tells our caller that
6631 * the original pattern needs upgrading to utf8.
6635 S_compile_runtime_code(pTHX_ RExC_state_t * const pRExC_state,
6636 char *pat, STRLEN plen)
6640 GET_RE_DEBUG_FLAGS_DECL;
6642 if (pRExC_state->runtime_code_qr) {
6643 /* this is the second time we've been called; this should
6644 * only happen if the main pattern got upgraded to utf8
6645 * during compilation; re-use the qr we compiled first time
6646 * round (which should be utf8 too)
6648 qr = pRExC_state->runtime_code_qr;
6649 pRExC_state->runtime_code_qr = NULL;
6650 assert(RExC_utf8 && SvUTF8(qr));
6656 int newlen = plen + 7; /* allow for "qr''xx\0" extra chars */
6660 /* determine how many extra chars we need for ' and \ escaping */
6661 for (s = 0; s < plen; s++) {
6662 if (pat[s] == '\'' || pat[s] == '\\')
6666 Newx(newpat, newlen, char);
6668 *p++ = 'q'; *p++ = 'r'; *p++ = '\'';
6670 for (s = 0; s < plen; s++) {
6671 if ( pRExC_state->code_blocks
6672 && n < pRExC_state->code_blocks->count
6673 && s == pRExC_state->code_blocks->cb[n].start)
6675 /* blank out literal code block */
6676 assert(pat[s] == '(');
6677 while (s <= pRExC_state->code_blocks->cb[n].end) {
6685 if (pat[s] == '\'' || pat[s] == '\\')
6690 if (pRExC_state->pm_flags & RXf_PMf_EXTENDED) {
6692 if (pRExC_state->pm_flags & RXf_PMf_EXTENDED_MORE) {
6698 Perl_re_printf( aTHX_
6699 "%sre-parsing pattern for runtime code:%s %s\n",
6700 PL_colors[4],PL_colors[5],newpat);
6703 sv = newSVpvn_flags(newpat, p-newpat-1, RExC_utf8 ? SVf_UTF8 : 0);
6709 PUSHSTACKi(PERLSI_REQUIRE);
6710 /* G_RE_REPARSING causes the toker to collapse \\ into \ when
6711 * parsing qr''; normally only q'' does this. It also alters
6713 eval_sv(sv, G_SCALAR|G_RE_REPARSING);
6714 SvREFCNT_dec_NN(sv);
6719 SV * const errsv = ERRSV;
6720 if (SvTRUE_NN(errsv))
6721 /* use croak_sv ? */
6722 Perl_croak_nocontext("%" SVf, SVfARG(errsv));
6724 assert(SvROK(qr_ref));
6726 assert(SvTYPE(qr) == SVt_REGEXP && RX_ENGINE((REGEXP*)qr)->op_comp);
6727 /* the leaving below frees the tmp qr_ref.
6728 * Give qr a life of its own */
6736 if (!RExC_utf8 && SvUTF8(qr)) {
6737 /* first time through; the pattern got upgraded; save the
6738 * qr for the next time through */
6739 assert(!pRExC_state->runtime_code_qr);
6740 pRExC_state->runtime_code_qr = qr;
6745 /* extract any code blocks within the returned qr// */
6748 /* merge the main (r1) and run-time (r2) code blocks into one */
6750 RXi_GET_DECL(ReANY((REGEXP *)qr), r2);
6751 struct reg_code_block *new_block, *dst;
6752 RExC_state_t * const r1 = pRExC_state; /* convenient alias */
6756 if (!r2->code_blocks || !r2->code_blocks->count) /* we guessed wrong */
6758 SvREFCNT_dec_NN(qr);
6762 if (!r1->code_blocks)
6763 r1->code_blocks = S_alloc_code_blocks(aTHX_ 0);
6765 r1c = r1->code_blocks->count;
6766 r2c = r2->code_blocks->count;
6768 Newx(new_block, r1c + r2c, struct reg_code_block);
6772 while (i1 < r1c || i2 < r2c) {
6773 struct reg_code_block *src;
6777 src = &r2->code_blocks->cb[i2++];
6781 src = &r1->code_blocks->cb[i1++];
6782 else if ( r1->code_blocks->cb[i1].start
6783 < r2->code_blocks->cb[i2].start)
6785 src = &r1->code_blocks->cb[i1++];
6786 assert(src->end < r2->code_blocks->cb[i2].start);
6789 assert( r1->code_blocks->cb[i1].start
6790 > r2->code_blocks->cb[i2].start);
6791 src = &r2->code_blocks->cb[i2++];
6793 assert(src->end < r1->code_blocks->cb[i1].start);
6796 assert(pat[src->start] == '(');
6797 assert(pat[src->end] == ')');
6798 dst->start = src->start;
6799 dst->end = src->end;
6800 dst->block = src->block;
6801 dst->src_regex = is_qr ? (REGEXP*) SvREFCNT_inc( (SV*) qr)
6805 r1->code_blocks->count += r2c;
6806 Safefree(r1->code_blocks->cb);
6807 r1->code_blocks->cb = new_block;
6810 SvREFCNT_dec_NN(qr);
6816 S_setup_longest(pTHX_ RExC_state_t *pRExC_state,
6817 struct reg_substr_datum *rsd,
6818 struct scan_data_substrs *sub,
6819 STRLEN longest_length)
6821 /* This is the common code for setting up the floating and fixed length
6822 * string data extracted from Perl_re_op_compile() below. Returns a boolean
6823 * as to whether succeeded or not */
6827 bool eol = cBOOL(sub->flags & SF_BEFORE_EOL);
6828 bool meol = cBOOL(sub->flags & SF_BEFORE_MEOL);
6830 if (! (longest_length
6831 || (eol /* Can't have SEOL and MULTI */
6832 && (! meol || (RExC_flags & RXf_PMf_MULTILINE)))
6834 /* See comments for join_exact for why REG_UNFOLDED_MULTI_SEEN */
6835 || (RExC_seen & REG_UNFOLDED_MULTI_SEEN))
6840 /* copy the information about the longest from the reg_scan_data
6841 over to the program. */
6842 if (SvUTF8(sub->str)) {
6844 rsd->utf8_substr = sub->str;
6846 rsd->substr = sub->str;
6847 rsd->utf8_substr = NULL;
6849 /* end_shift is how many chars that must be matched that
6850 follow this item. We calculate it ahead of time as once the
6851 lookbehind offset is added in we lose the ability to correctly
6853 ml = sub->minlenp ? *(sub->minlenp) : (SSize_t)longest_length;
6854 rsd->end_shift = ml - sub->min_offset
6856 /* XXX SvTAIL is always false here - did you mean FBMcf_TAIL
6858 + (SvTAIL(sub->str) != 0)
6862 t = (eol/* Can't have SEOL and MULTI */
6863 && (! meol || (RExC_flags & RXf_PMf_MULTILINE)));
6864 fbm_compile(sub->str, t ? FBMcf_TAIL : 0);
6870 * Perl_re_op_compile - the perl internal RE engine's function to compile a
6871 * regular expression into internal code.
6872 * The pattern may be passed either as:
6873 * a list of SVs (patternp plus pat_count)
6874 * a list of OPs (expr)
6875 * If both are passed, the SV list is used, but the OP list indicates
6876 * which SVs are actually pre-compiled code blocks
6878 * The SVs in the list have magic and qr overloading applied to them (and
6879 * the list may be modified in-place with replacement SVs in the latter
6882 * If the pattern hasn't changed from old_re, then old_re will be
6885 * eng is the current engine. If that engine has an op_comp method, then
6886 * handle directly (i.e. we assume that op_comp was us); otherwise, just
6887 * do the initial concatenation of arguments and pass on to the external
6890 * If is_bare_re is not null, set it to a boolean indicating whether the
6891 * arg list reduced (after overloading) to a single bare regex which has
6892 * been returned (i.e. /$qr/).
6894 * orig_rx_flags contains RXf_* flags. See perlreapi.pod for more details.
6896 * pm_flags contains the PMf_* flags, typically based on those from the
6897 * pm_flags field of the related PMOP. Currently we're only interested in
6898 * PMf_HAS_CV, PMf_IS_QR, PMf_USE_RE_EVAL.
6900 * We can't allocate space until we know how big the compiled form will be,
6901 * but we can't compile it (and thus know how big it is) until we've got a
6902 * place to put the code. So we cheat: we compile it twice, once with code
6903 * generation turned off and size counting turned on, and once "for real".
6904 * This also means that we don't allocate space until we are sure that the
6905 * thing really will compile successfully, and we never have to move the
6906 * code and thus invalidate pointers into it. (Note that it has to be in
6907 * one piece because free() must be able to free it all.) [NB: not true in perl]
6909 * Beware that the optimization-preparation code in here knows about some
6910 * of the structure of the compiled regexp. [I'll say.]
6914 Perl_re_op_compile(pTHX_ SV ** const patternp, int pat_count,
6915 OP *expr, const regexp_engine* eng, REGEXP *old_re,
6916 bool *is_bare_re, U32 orig_rx_flags, U32 pm_flags)
6920 regexp_internal *ri;
6928 SV** new_patternp = patternp;
6930 /* these are all flags - maybe they should be turned
6931 * into a single int with different bit masks */
6932 I32 sawlookahead = 0;
6937 regex_charset initial_charset = get_regex_charset(orig_rx_flags);
6939 bool runtime_code = 0;
6941 RExC_state_t RExC_state;
6942 RExC_state_t * const pRExC_state = &RExC_state;
6943 #ifdef TRIE_STUDY_OPT
6945 RExC_state_t copyRExC_state;
6947 GET_RE_DEBUG_FLAGS_DECL;
6949 PERL_ARGS_ASSERT_RE_OP_COMPILE;
6951 DEBUG_r(if (!PL_colorset) reginitcolors());
6953 /* Initialize these here instead of as-needed, as is quick and avoids
6954 * having to test them each time otherwise */
6955 if (! PL_AboveLatin1) {
6957 char * dump_len_string;
6960 PL_AboveLatin1 = _new_invlist_C_array(AboveLatin1_invlist);
6961 PL_Latin1 = _new_invlist_C_array(Latin1_invlist);
6962 PL_UpperLatin1 = _new_invlist_C_array(UpperLatin1_invlist);
6963 PL_utf8_foldable = _new_invlist_C_array(_Perl_Any_Folds_invlist);
6964 PL_HasMultiCharFold =
6965 _new_invlist_C_array(_Perl_Folds_To_Multi_Char_invlist);
6967 /* This is calculated here, because the Perl program that generates the
6968 * static global ones doesn't currently have access to
6969 * NUM_ANYOF_CODE_POINTS */
6970 PL_InBitmap = _new_invlist(2);
6971 PL_InBitmap = _add_range_to_invlist(PL_InBitmap, 0,
6972 NUM_ANYOF_CODE_POINTS - 1);
6974 dump_len_string = PerlEnv_getenv("PERL_DUMP_RE_MAX_LEN");
6975 if ( ! dump_len_string
6976 || ! grok_atoUV(dump_len_string, (UV *)&PL_dump_re_max_len, NULL))
6978 PL_dump_re_max_len = 60; /* A reasonable default */
6983 pRExC_state->warn_text = NULL;
6984 pRExC_state->code_blocks = NULL;
6987 *is_bare_re = FALSE;
6989 if (expr && (expr->op_type == OP_LIST ||
6990 (expr->op_type == OP_NULL && expr->op_targ == OP_LIST))) {
6991 /* allocate code_blocks if needed */
6995 for (o = cLISTOPx(expr)->op_first; o; o = OpSIBLING(o))
6996 if (o->op_type == OP_NULL && (o->op_flags & OPf_SPECIAL))
6997 ncode++; /* count of DO blocks */
7000 pRExC_state->code_blocks = S_alloc_code_blocks(aTHX_ ncode);
7004 /* compile-time pattern with just OP_CONSTs and DO blocks */
7009 /* find how many CONSTs there are */
7012 if (expr->op_type == OP_CONST)
7015 for (o = cLISTOPx(expr)->op_first; o; o = OpSIBLING(o)) {
7016 if (o->op_type == OP_CONST)
7020 /* fake up an SV array */
7022 assert(!new_patternp);
7023 Newx(new_patternp, n, SV*);
7024 SAVEFREEPV(new_patternp);
7028 if (expr->op_type == OP_CONST)
7029 new_patternp[n] = cSVOPx_sv(expr);
7031 for (o = cLISTOPx(expr)->op_first; o; o = OpSIBLING(o)) {
7032 if (o->op_type == OP_CONST)
7033 new_patternp[n++] = cSVOPo_sv;
7038 DEBUG_PARSE_r(Perl_re_printf( aTHX_
7039 "Assembling pattern from %d elements%s\n", pat_count,
7040 orig_rx_flags & RXf_SPLIT ? " for split" : ""));
7042 /* set expr to the first arg op */
7044 if (pRExC_state->code_blocks && pRExC_state->code_blocks->count
7045 && expr->op_type != OP_CONST)
7047 expr = cLISTOPx(expr)->op_first;
7048 assert( expr->op_type == OP_PUSHMARK
7049 || (expr->op_type == OP_NULL && expr->op_targ == OP_PUSHMARK)
7050 || expr->op_type == OP_PADRANGE);
7051 expr = OpSIBLING(expr);
7054 pat = S_concat_pat(aTHX_ pRExC_state, NULL, new_patternp, pat_count,
7055 expr, &recompile, NULL);
7057 /* handle bare (possibly after overloading) regex: foo =~ $re */
7062 if (SvTYPE(re) == SVt_REGEXP) {
7066 DEBUG_PARSE_r(Perl_re_printf( aTHX_
7067 "Precompiled pattern%s\n",
7068 orig_rx_flags & RXf_SPLIT ? " for split" : ""));
7074 exp = SvPV_nomg(pat, plen);
7076 if (!eng->op_comp) {
7077 if ((SvUTF8(pat) && IN_BYTES)
7078 || SvGMAGICAL(pat) || SvAMAGIC(pat))
7080 /* make a temporary copy; either to convert to bytes,
7081 * or to avoid repeating get-magic / overloaded stringify */
7082 pat = newSVpvn_flags(exp, plen, SVs_TEMP |
7083 (IN_BYTES ? 0 : SvUTF8(pat)));
7085 return CALLREGCOMP_ENG(eng, pat, orig_rx_flags);
7088 /* ignore the utf8ness if the pattern is 0 length */
7089 RExC_utf8 = RExC_orig_utf8 = (plen == 0 || IN_BYTES) ? 0 : SvUTF8(pat);
7091 RExC_uni_semantics = 0;
7092 RExC_seen_unfolded_sharp_s = 0;
7093 RExC_contains_locale = 0;
7094 RExC_strict = cBOOL(pm_flags & RXf_PMf_STRICT);
7095 RExC_in_script_run = 0;
7096 RExC_study_started = 0;
7097 pRExC_state->runtime_code_qr = NULL;
7098 RExC_frame_head= NULL;
7099 RExC_frame_last= NULL;
7100 RExC_frame_count= 0;
7103 RExC_mysv1= sv_newmortal();
7104 RExC_mysv2= sv_newmortal();
7107 SV *dsv= sv_newmortal();
7108 RE_PV_QUOTED_DECL(s, RExC_utf8, dsv, exp, plen, PL_dump_re_max_len);
7109 Perl_re_printf( aTHX_ "%sCompiling REx%s %s\n",
7110 PL_colors[4],PL_colors[5],s);
7114 /* we jump here if we have to recompile, e.g., from upgrading the pattern
7117 if ((pm_flags & PMf_USE_RE_EVAL)
7118 /* this second condition covers the non-regex literal case,
7119 * i.e. $foo =~ '(?{})'. */
7120 || (IN_PERL_COMPILETIME && (PL_hints & HINT_RE_EVAL))
7122 runtime_code = S_has_runtime_code(aTHX_ pRExC_state, exp, plen);
7124 /* return old regex if pattern hasn't changed */
7125 /* XXX: note in the below we have to check the flags as well as the
7128 * Things get a touch tricky as we have to compare the utf8 flag
7129 * independently from the compile flags. */
7133 && !!RX_UTF8(old_re) == !!RExC_utf8
7134 && ( RX_COMPFLAGS(old_re) == ( orig_rx_flags & RXf_PMf_FLAGCOPYMASK ) )
7135 && RX_PRECOMP(old_re)
7136 && RX_PRELEN(old_re) == plen
7137 && memEQ(RX_PRECOMP(old_re), exp, plen)
7138 && !runtime_code /* with runtime code, always recompile */ )
7143 rx_flags = orig_rx_flags;
7145 if ( initial_charset == REGEX_DEPENDS_CHARSET
7146 && (RExC_utf8 ||RExC_uni_semantics))
7149 /* Set to use unicode semantics if the pattern is in utf8 and has the
7150 * 'depends' charset specified, as it means unicode when utf8 */
7151 set_regex_charset(&rx_flags, REGEX_UNICODE_CHARSET);
7155 RExC_precomp_adj = 0;
7156 RExC_flags = rx_flags;
7157 RExC_pm_flags = pm_flags;
7160 assert(TAINTING_get || !TAINT_get);
7162 Perl_croak(aTHX_ "Eval-group in insecure regular expression");
7164 if (!S_compile_runtime_code(aTHX_ pRExC_state, exp, plen)) {
7165 /* whoops, we have a non-utf8 pattern, whilst run-time code
7166 * got compiled as utf8. Try again with a utf8 pattern */
7167 S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &exp, &plen,
7168 pRExC_state->code_blocks ? pRExC_state->code_blocks->count : 0);
7169 goto redo_first_pass;
7172 assert(!pRExC_state->runtime_code_qr);
7178 RExC_in_lookbehind = 0;
7179 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
7182 RExC_recode_x_to_native = 0;
7184 RExC_in_multi_char_class = 0;
7186 /* First pass: determine size, legality. */
7188 RExC_start = RExC_adjusted_start = exp;
7189 RExC_end = exp + plen;
7190 RExC_precomp_end = RExC_end;
7195 RExC_emit = (regnode *) &RExC_emit_dummy;
7196 RExC_whilem_seen = 0;
7197 RExC_open_parens = NULL;
7198 RExC_close_parens = NULL;
7200 RExC_paren_names = NULL;
7202 RExC_paren_name_list = NULL;
7204 RExC_recurse = NULL;
7205 RExC_study_chunk_recursed = NULL;
7206 RExC_study_chunk_recursed_bytes= 0;
7207 RExC_recurse_count = 0;
7208 pRExC_state->code_index = 0;
7210 /* This NUL is guaranteed because the pattern comes from an SV*, and the sv
7211 * code makes sure the final byte is an uncounted NUL. But should this
7212 * ever not be the case, lots of things could read beyond the end of the
7213 * buffer: loops like
7214 * while(isFOO(*RExC_parse)) RExC_parse++;
7215 * strchr(RExC_parse, "foo");
7216 * etc. So it is worth noting. */
7217 assert(*RExC_end == '\0');
7220 Perl_re_printf( aTHX_ "Starting first pass (sizing)\n");
7222 RExC_lastparse=NULL;
7225 if (reg(pRExC_state, 0, &flags,1) == NULL) {
7226 /* It's possible to write a regexp in ascii that represents Unicode
7227 codepoints outside of the byte range, such as via \x{100}. If we
7228 detect such a sequence we have to convert the entire pattern to utf8
7229 and then recompile, as our sizing calculation will have been based
7230 on 1 byte == 1 character, but we will need to use utf8 to encode
7231 at least some part of the pattern, and therefore must convert the whole
7234 if (MUST_RESTART(flags)) {
7235 if (flags & NEED_UTF8) {
7236 S_pat_upgrade_to_utf8(aTHX_ pRExC_state, &exp, &plen,
7237 pRExC_state->code_blocks ? pRExC_state->code_blocks->count : 0);
7238 DEBUG_PARSE_r(Perl_re_printf( aTHX_ "Need to redo pass 1 after upgrade\n"));
7241 DEBUG_PARSE_r(Perl_re_printf( aTHX_ "Need to redo pass 1\n"));
7244 goto redo_first_pass;
7246 Perl_croak(aTHX_ "panic: reg returned NULL to re_op_compile for sizing pass, flags=%#" UVxf, (UV) flags);
7250 Perl_re_printf( aTHX_
7251 "Required size %" IVdf " nodes\n"
7252 "Starting second pass (creation)\n",
7255 RExC_lastparse=NULL;
7258 /* The first pass could have found things that force Unicode semantics */
7259 if ((RExC_utf8 || RExC_uni_semantics)
7260 && get_regex_charset(rx_flags) == REGEX_DEPENDS_CHARSET)
7262 set_regex_charset(&rx_flags, REGEX_UNICODE_CHARSET);
7265 /* Small enough for pointer-storage convention?
7266 If extralen==0, this means that we will not need long jumps. */
7267 if (RExC_size >= 0x10000L && RExC_extralen)
7268 RExC_size += RExC_extralen;
7271 if (RExC_whilem_seen > 15)
7272 RExC_whilem_seen = 15;
7274 /* Allocate space and zero-initialize. Note, the two step process
7275 of zeroing when in debug mode, thus anything assigned has to
7276 happen after that */
7277 rx = (REGEXP*) newSV_type(SVt_REGEXP);
7279 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
7280 char, regexp_internal);
7281 if ( r == NULL || ri == NULL )
7282 FAIL("Regexp out of space");
7284 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
7285 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
7288 /* bulk initialize base fields with 0. */
7289 Zero(ri, sizeof(regexp_internal), char);
7292 /* non-zero initialization begins here */
7295 r->extflags = rx_flags;
7296 RXp_COMPFLAGS(r) = orig_rx_flags & RXf_PMf_FLAGCOPYMASK;
7298 if (pm_flags & PMf_IS_QR) {
7299 ri->code_blocks = pRExC_state->code_blocks;
7300 if (ri->code_blocks)
7301 ri->code_blocks->refcnt++;
7305 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
7306 bool has_charset = (get_regex_charset(r->extflags)
7307 != REGEX_DEPENDS_CHARSET);
7309 /* The caret is output if there are any defaults: if not all the STD
7310 * flags are set, or if no character set specifier is needed */
7312 (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD)
7314 bool has_runon = ((RExC_seen & REG_RUN_ON_COMMENT_SEEN)
7315 == REG_RUN_ON_COMMENT_SEEN);
7316 U8 reganch = (U8)((r->extflags & RXf_PMf_STD_PMMOD)
7317 >> RXf_PMf_STD_PMMOD_SHIFT);
7318 const char *fptr = STD_PAT_MODS; /*"msixxn"*/
7321 /* We output all the necessary flags; we never output a minus, as all
7322 * those are defaults, so are
7323 * covered by the caret */
7324 const STRLEN wraplen = plen + has_p + has_runon
7325 + has_default /* If needs a caret */
7326 + PL_bitcount[reganch] /* 1 char for each set standard flag */
7328 /* If needs a character set specifier */
7329 + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0)
7330 + (sizeof("(?:)") - 1);
7332 /* make sure PL_bitcount bounds not exceeded */
7333 assert(sizeof(STD_PAT_MODS) <= 8);
7335 p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */
7338 SvFLAGS(rx) |= SVf_UTF8;
7341 /* If a default, cover it using the caret */
7343 *p++= DEFAULT_PAT_MOD;
7347 const char* const name = get_regex_charset_name(r->extflags, &len);
7348 Copy(name, p, len, char);
7352 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
7355 while((ch = *fptr++)) {
7363 Copy(RExC_precomp, p, plen, char);
7364 assert ((RX_WRAPPED(rx) - p) < 16);
7365 r->pre_prefix = p - RX_WRAPPED(rx);
7371 SvCUR_set(rx, p - RX_WRAPPED(rx));
7375 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
7377 /* Useful during FAIL. */
7378 #ifdef RE_TRACK_PATTERN_OFFSETS
7379 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
7380 DEBUG_OFFSETS_r(Perl_re_printf( aTHX_
7381 "%s %" UVuf " bytes for offset annotations.\n",
7382 ri->u.offsets ? "Got" : "Couldn't get",
7383 (UV)((2*RExC_size+1) * sizeof(U32))));
7385 SetProgLen(ri,RExC_size);
7390 /* Second pass: emit code. */
7391 RExC_flags = rx_flags; /* don't let top level (?i) bleed */
7392 RExC_pm_flags = pm_flags;
7394 RExC_end = exp + plen;
7396 RExC_emit_start = ri->program;
7397 RExC_emit = ri->program;
7398 RExC_emit_bound = ri->program + RExC_size + 1;
7399 pRExC_state->code_index = 0;
7401 *((char*) RExC_emit++) = (char) REG_MAGIC;
7402 /* setup various meta data about recursion, this all requires
7403 * RExC_npar to be correctly set, and a bit later on we clear it */
7404 if (RExC_seen & REG_RECURSE_SEEN) {
7405 DEBUG_OPTIMISE_MORE_r(Perl_re_printf( aTHX_
7406 "%*s%*s Setting up open/close parens\n",
7407 22, "| |", (int)(0 * 2 + 1), ""));
7409 /* setup RExC_open_parens, which holds the address of each
7410 * OPEN tag, and to make things simpler for the 0 index
7411 * the start of the program - this is used later for offsets */
7412 Newxz(RExC_open_parens, RExC_npar,regnode *);
7413 SAVEFREEPV(RExC_open_parens);
7414 RExC_open_parens[0] = RExC_emit;
7416 /* setup RExC_close_parens, which holds the address of each
7417 * CLOSE tag, and to make things simpler for the 0 index
7418 * the end of the program - this is used later for offsets */
7419 Newxz(RExC_close_parens, RExC_npar,regnode *);
7420 SAVEFREEPV(RExC_close_parens);
7421 /* we dont know where end op starts yet, so we dont
7422 * need to set RExC_close_parens[0] like we do RExC_open_parens[0] above */
7424 /* Note, RExC_npar is 1 + the number of parens in a pattern.
7425 * So its 1 if there are no parens. */
7426 RExC_study_chunk_recursed_bytes= (RExC_npar >> 3) +
7427 ((RExC_npar & 0x07) != 0);
7428 Newx(RExC_study_chunk_recursed,
7429 RExC_study_chunk_recursed_bytes * RExC_npar, U8);
7430 SAVEFREEPV(RExC_study_chunk_recursed);
7433 if (reg(pRExC_state, 0, &flags,1) == NULL) {
7435 Perl_croak(aTHX_ "panic: reg returned NULL to re_op_compile for generation pass, flags=%#" UVxf, (UV) flags);
7438 Perl_re_printf( aTHX_ "Starting post parse optimization\n");
7441 /* XXXX To minimize changes to RE engine we always allocate
7442 3-units-long substrs field. */
7443 Newx(r->substrs, 1, struct reg_substr_data);
7444 if (RExC_recurse_count) {
7445 Newx(RExC_recurse,RExC_recurse_count,regnode *);
7446 SAVEFREEPV(RExC_recurse);
7450 r->minlen = minlen = sawlookahead = sawplus = sawopen = sawminmod = 0;
7452 RExC_study_chunk_recursed_count= 0;
7454 Zero(r->substrs, 1, struct reg_substr_data);
7455 if (RExC_study_chunk_recursed) {
7456 Zero(RExC_study_chunk_recursed,
7457 RExC_study_chunk_recursed_bytes * RExC_npar, U8);
7461 #ifdef TRIE_STUDY_OPT
7463 StructCopy(&zero_scan_data, &data, scan_data_t);
7464 copyRExC_state = RExC_state;
7467 DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "Restudying\n"));
7469 RExC_state = copyRExC_state;
7470 if (seen & REG_TOP_LEVEL_BRANCHES_SEEN)
7471 RExC_seen |= REG_TOP_LEVEL_BRANCHES_SEEN;
7473 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES_SEEN;
7474 StructCopy(&zero_scan_data, &data, scan_data_t);
7477 StructCopy(&zero_scan_data, &data, scan_data_t);
7480 /* Dig out information for optimizations. */
7481 r->extflags = RExC_flags; /* was pm_op */
7482 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
7485 SvUTF8_on(rx); /* Unicode in it? */
7486 ri->regstclass = NULL;
7487 if (RExC_naughty >= TOO_NAUGHTY) /* Probably an expensive pattern. */
7488 r->intflags |= PREGf_NAUGHTY;
7489 scan = ri->program + 1; /* First BRANCH. */
7491 /* testing for BRANCH here tells us whether there is "must appear"
7492 data in the pattern. If there is then we can use it for optimisations */
7493 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN)) { /* Only one top-level choice.
7496 STRLEN longest_length[2];
7497 regnode_ssc ch_class; /* pointed to by data */
7499 SSize_t last_close = 0; /* pointed to by data */
7500 regnode *first= scan;
7501 regnode *first_next= regnext(first);
7505 * Skip introductions and multiplicators >= 1
7506 * so that we can extract the 'meat' of the pattern that must
7507 * match in the large if() sequence following.
7508 * NOTE that EXACT is NOT covered here, as it is normally
7509 * picked up by the optimiser separately.
7511 * This is unfortunate as the optimiser isnt handling lookahead
7512 * properly currently.
7515 while ((OP(first) == OPEN && (sawopen = 1)) ||
7516 /* An OR of *one* alternative - should not happen now. */
7517 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
7518 /* for now we can't handle lookbehind IFMATCH*/
7519 (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) ||
7520 (OP(first) == PLUS) ||
7521 (OP(first) == MINMOD) ||
7522 /* An {n,m} with n>0 */
7523 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
7524 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
7527 * the only op that could be a regnode is PLUS, all the rest
7528 * will be regnode_1 or regnode_2.
7530 * (yves doesn't think this is true)
7532 if (OP(first) == PLUS)
7535 if (OP(first) == MINMOD)
7537 first += regarglen[OP(first)];
7539 first = NEXTOPER(first);
7540 first_next= regnext(first);
7543 /* Starting-point info. */
7545 DEBUG_PEEP("first:", first, 0, 0);
7546 /* Ignore EXACT as we deal with it later. */
7547 if (PL_regkind[OP(first)] == EXACT) {
7548 if (OP(first) == EXACT || OP(first) == EXACTL)
7549 NOOP; /* Empty, get anchored substr later. */
7551 ri->regstclass = first;
7554 else if (PL_regkind[OP(first)] == TRIE &&
7555 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
7557 /* this can happen only on restudy */
7558 ri->regstclass = construct_ahocorasick_from_trie(pRExC_state, (regnode *)first, 0);
7561 else if (REGNODE_SIMPLE(OP(first)))
7562 ri->regstclass = first;
7563 else if (PL_regkind[OP(first)] == BOUND ||
7564 PL_regkind[OP(first)] == NBOUND)
7565 ri->regstclass = first;
7566 else if (PL_regkind[OP(first)] == BOL) {
7567 r->intflags |= (OP(first) == MBOL
7570 first = NEXTOPER(first);
7573 else if (OP(first) == GPOS) {
7574 r->intflags |= PREGf_ANCH_GPOS;
7575 first = NEXTOPER(first);
7578 else if ((!sawopen || !RExC_sawback) &&
7580 (OP(first) == STAR &&
7581 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
7582 !(r->intflags & PREGf_ANCH) && !pRExC_state->code_blocks)
7584 /* turn .* into ^.* with an implied $*=1 */
7586 (OP(NEXTOPER(first)) == REG_ANY)
7589 r->intflags |= (type | PREGf_IMPLICIT);
7590 first = NEXTOPER(first);
7593 if (sawplus && !sawminmod && !sawlookahead
7594 && (!sawopen || !RExC_sawback)
7595 && !pRExC_state->code_blocks) /* May examine pos and $& */
7596 /* x+ must match at the 1st pos of run of x's */
7597 r->intflags |= PREGf_SKIP;
7599 /* Scan is after the zeroth branch, first is atomic matcher. */
7600 #ifdef TRIE_STUDY_OPT
7603 Perl_re_printf( aTHX_ "first at %" IVdf "\n",
7604 (IV)(first - scan + 1))
7608 Perl_re_printf( aTHX_ "first at %" IVdf "\n",
7609 (IV)(first - scan + 1))
7615 * If there's something expensive in the r.e., find the
7616 * longest literal string that must appear and make it the
7617 * regmust. Resolve ties in favor of later strings, since
7618 * the regstart check works with the beginning of the r.e.
7619 * and avoiding duplication strengthens checking. Not a
7620 * strong reason, but sufficient in the absence of others.
7621 * [Now we resolve ties in favor of the earlier string if
7622 * it happens that c_offset_min has been invalidated, since the
7623 * earlier string may buy us something the later one won't.]
7626 data.substrs[0].str = newSVpvs("");
7627 data.substrs[1].str = newSVpvs("");
7628 data.last_found = newSVpvs("");
7629 data.cur_is_floating = 0; /* initially any found substring is fixed */
7630 ENTER_with_name("study_chunk");
7631 SAVEFREESV(data.substrs[0].str);
7632 SAVEFREESV(data.substrs[1].str);
7633 SAVEFREESV(data.last_found);
7635 if (!ri->regstclass) {
7636 ssc_init(pRExC_state, &ch_class);
7637 data.start_class = &ch_class;
7638 stclass_flag = SCF_DO_STCLASS_AND;
7639 } else /* XXXX Check for BOUND? */
7641 data.last_closep = &last_close;
7645 * MAIN ENTRY FOR study_chunk() FOR m/PATTERN/
7646 * (NO top level branches)
7648 minlen = study_chunk(pRExC_state, &first, &minlen, &fake,
7649 scan + RExC_size, /* Up to end */
7651 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag
7652 | (restudied ? SCF_TRIE_DOING_RESTUDY : 0),
7656 CHECK_RESTUDY_GOTO_butfirst(LEAVE_with_name("study_chunk"));
7659 if ( RExC_npar == 1 && !data.cur_is_floating
7660 && data.last_start_min == 0 && data.last_end > 0
7661 && !RExC_seen_zerolen
7662 && !(RExC_seen & REG_VERBARG_SEEN)
7663 && !(RExC_seen & REG_GPOS_SEEN)
7665 r->extflags |= RXf_CHECK_ALL;
7667 scan_commit(pRExC_state, &data,&minlen,0);
7670 /* XXX this is done in reverse order because that's the way the
7671 * code was before it was parameterised. Don't know whether it
7672 * actually needs doing in reverse order. DAPM */
7673 for (i = 1; i >= 0; i--) {
7674 longest_length[i] = CHR_SVLEN(data.substrs[i].str);
7677 && SvCUR(data.substrs[0].str) /* ok to leave SvCUR */
7678 && data.substrs[0].min_offset
7679 == data.substrs[1].min_offset
7680 && SvCUR(data.substrs[0].str)
7681 == SvCUR(data.substrs[1].str)
7683 && S_setup_longest (aTHX_ pRExC_state,
7684 &(r->substrs->data[i]),
7688 r->substrs->data[i].min_offset =
7689 data.substrs[i].min_offset - data.substrs[i].lookbehind;
7691 r->substrs->data[i].max_offset = data.substrs[i].max_offset;
7692 /* Don't offset infinity */
7693 if (data.substrs[i].max_offset < SSize_t_MAX)
7694 r->substrs->data[i].max_offset -= data.substrs[i].lookbehind;
7695 SvREFCNT_inc_simple_void_NN(data.substrs[i].str);
7698 r->substrs->data[i].substr = NULL;
7699 r->substrs->data[i].utf8_substr = NULL;
7700 longest_length[i] = 0;
7704 LEAVE_with_name("study_chunk");
7707 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
7708 ri->regstclass = NULL;
7710 if ((!(r->substrs->data[0].substr || r->substrs->data[0].utf8_substr)
7711 || r->substrs->data[0].min_offset)
7713 && ! (ANYOF_FLAGS(data.start_class) & SSC_MATCHES_EMPTY_STRING)
7714 && is_ssc_worth_it(pRExC_state, data.start_class))
7716 const U32 n = add_data(pRExC_state, STR_WITH_LEN("f"));
7718 ssc_finalize(pRExC_state, data.start_class);
7720 Newx(RExC_rxi->data->data[n], 1, regnode_ssc);
7721 StructCopy(data.start_class,
7722 (regnode_ssc*)RExC_rxi->data->data[n],
7724 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
7725 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
7726 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
7727 regprop(r, sv, (regnode*)data.start_class, NULL, pRExC_state);
7728 Perl_re_printf( aTHX_
7729 "synthetic stclass \"%s\".\n",
7730 SvPVX_const(sv));});
7731 data.start_class = NULL;
7734 /* A temporary algorithm prefers floated substr to fixed one of
7735 * same length to dig more info. */
7736 i = (longest_length[0] <= longest_length[1]);
7737 r->substrs->check_ix = i;
7738 r->check_end_shift = r->substrs->data[i].end_shift;
7739 r->check_substr = r->substrs->data[i].substr;
7740 r->check_utf8 = r->substrs->data[i].utf8_substr;
7741 r->check_offset_min = r->substrs->data[i].min_offset;
7742 r->check_offset_max = r->substrs->data[i].max_offset;
7743 if (!i && (r->intflags & (PREGf_ANCH_SBOL|PREGf_ANCH_GPOS)))
7744 r->intflags |= PREGf_NOSCAN;
7746 if ((r->check_substr || r->check_utf8) ) {
7747 r->extflags |= RXf_USE_INTUIT;
7748 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
7749 r->extflags |= RXf_INTUIT_TAIL;
7752 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
7753 if ( (STRLEN)minlen < longest_length[1] )
7754 minlen= longest_length[1];
7755 if ( (STRLEN)minlen < longest_length[0] )
7756 minlen= longest_length[0];
7760 /* Several toplevels. Best we can is to set minlen. */
7762 regnode_ssc ch_class;
7763 SSize_t last_close = 0;
7765 DEBUG_PARSE_r(Perl_re_printf( aTHX_ "\nMulti Top Level\n"));
7767 scan = ri->program + 1;
7768 ssc_init(pRExC_state, &ch_class);
7769 data.start_class = &ch_class;
7770 data.last_closep = &last_close;
7774 * MAIN ENTRY FOR study_chunk() FOR m/P1|P2|.../
7775 * (patterns WITH top level branches)
7777 minlen = study_chunk(pRExC_state,
7778 &scan, &minlen, &fake, scan + RExC_size, &data, -1, 0, NULL,
7779 SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS|(restudied
7780 ? SCF_TRIE_DOING_RESTUDY
7784 CHECK_RESTUDY_GOTO_butfirst(NOOP);
7786 r->check_substr = NULL;
7787 r->check_utf8 = NULL;
7788 r->substrs->data[0].substr = NULL;
7789 r->substrs->data[0].utf8_substr = NULL;
7790 r->substrs->data[1].substr = NULL;
7791 r->substrs->data[1].utf8_substr = NULL;
7793 if (! (ANYOF_FLAGS(data.start_class) & SSC_MATCHES_EMPTY_STRING)
7794 && is_ssc_worth_it(pRExC_state, data.start_class))
7796 const U32 n = add_data(pRExC_state, STR_WITH_LEN("f"));
7798 ssc_finalize(pRExC_state, data.start_class);
7800 Newx(RExC_rxi->data->data[n], 1, regnode_ssc);
7801 StructCopy(data.start_class,
7802 (regnode_ssc*)RExC_rxi->data->data[n],
7804 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
7805 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
7806 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
7807 regprop(r, sv, (regnode*)data.start_class, NULL, pRExC_state);
7808 Perl_re_printf( aTHX_
7809 "synthetic stclass \"%s\".\n",
7810 SvPVX_const(sv));});
7811 data.start_class = NULL;
7815 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) {
7816 r->extflags |= RXf_UNBOUNDED_QUANTIFIER_SEEN;
7817 r->maxlen = REG_INFTY;
7820 r->maxlen = RExC_maxlen;
7823 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
7824 the "real" pattern. */
7826 Perl_re_printf( aTHX_ "minlen: %" IVdf " r->minlen:%" IVdf " maxlen:%" IVdf "\n",
7827 (IV)minlen, (IV)r->minlen, (IV)RExC_maxlen);
7829 r->minlenret = minlen;
7830 if (r->minlen < minlen)
7833 if (RExC_seen & REG_RECURSE_SEEN ) {
7834 r->intflags |= PREGf_RECURSE_SEEN;
7835 Newx(r->recurse_locinput, r->nparens + 1, char *);
7837 if (RExC_seen & REG_GPOS_SEEN)
7838 r->intflags |= PREGf_GPOS_SEEN;
7839 if (RExC_seen & REG_LOOKBEHIND_SEEN)
7840 r->extflags |= RXf_NO_INPLACE_SUBST; /* inplace might break the
7842 if (pRExC_state->code_blocks)
7843 r->extflags |= RXf_EVAL_SEEN;
7844 if (RExC_seen & REG_VERBARG_SEEN)
7846 r->intflags |= PREGf_VERBARG_SEEN;
7847 r->extflags |= RXf_NO_INPLACE_SUBST; /* don't understand this! Yves */
7849 if (RExC_seen & REG_CUTGROUP_SEEN)
7850 r->intflags |= PREGf_CUTGROUP_SEEN;
7851 if (pm_flags & PMf_USE_RE_EVAL)
7852 r->intflags |= PREGf_USE_RE_EVAL;
7853 if (RExC_paren_names)
7854 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
7856 RXp_PAREN_NAMES(r) = NULL;
7858 /* If we have seen an anchor in our pattern then we set the extflag RXf_IS_ANCHORED
7859 * so it can be used in pp.c */
7860 if (r->intflags & PREGf_ANCH)
7861 r->extflags |= RXf_IS_ANCHORED;
7865 /* this is used to identify "special" patterns that might result
7866 * in Perl NOT calling the regex engine and instead doing the match "itself",
7867 * particularly special cases in split//. By having the regex compiler
7868 * do this pattern matching at a regop level (instead of by inspecting the pattern)
7869 * we avoid weird issues with equivalent patterns resulting in different behavior,
7870 * AND we allow non Perl engines to get the same optimizations by the setting the
7871 * flags appropriately - Yves */
7872 regnode *first = ri->program + 1;
7874 regnode *next = regnext(first);
7877 if (PL_regkind[fop] == NOTHING && nop == END)
7878 r->extflags |= RXf_NULL;
7879 else if ((fop == MBOL || (fop == SBOL && !first->flags)) && nop == END)
7880 /* when fop is SBOL first->flags will be true only when it was
7881 * produced by parsing /\A/, and not when parsing /^/. This is
7882 * very important for the split code as there we want to
7883 * treat /^/ as /^/m, but we do not want to treat /\A/ as /^/m.
7884 * See rt #122761 for more details. -- Yves */
7885 r->extflags |= RXf_START_ONLY;
7886 else if (fop == PLUS
7887 && PL_regkind[nop] == POSIXD && FLAGS(next) == _CC_SPACE
7889 r->extflags |= RXf_WHITE;
7890 else if ( r->extflags & RXf_SPLIT
7891 && (fop == EXACT || fop == EXACTL)
7892 && STR_LEN(first) == 1
7893 && *(STRING(first)) == ' '
7895 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
7899 if (RExC_contains_locale) {
7900 RXp_EXTFLAGS(r) |= RXf_TAINTED;
7904 if (RExC_paren_names) {
7905 ri->name_list_idx = add_data( pRExC_state, STR_WITH_LEN("a"));
7906 ri->data->data[ri->name_list_idx]
7907 = (void*)SvREFCNT_inc(RExC_paren_name_list);
7910 ri->name_list_idx = 0;
7912 while ( RExC_recurse_count > 0 ) {
7913 const regnode *scan = RExC_recurse[ --RExC_recurse_count ];
7915 * This data structure is set up in study_chunk() and is used
7916 * to calculate the distance between a GOSUB regopcode and
7917 * the OPEN/CURLYM (CURLYM's are special and can act like OPEN's)
7920 * If for some reason someone writes code that optimises
7921 * away a GOSUB opcode then the assert should be changed to
7922 * an if(scan) to guard the ARG2L_SET() - Yves
7925 assert(scan && OP(scan) == GOSUB);
7926 ARG2L_SET( scan, RExC_open_parens[ARG(scan)] - scan );
7929 Newxz(r->offs, RExC_npar, regexp_paren_pair);
7930 /* assume we don't need to swap parens around before we match */
7932 Perl_re_printf( aTHX_ "study_chunk_recursed_count: %lu\n",
7933 (unsigned long)RExC_study_chunk_recursed_count);
7937 Perl_re_printf( aTHX_ "Final program:\n");
7940 #ifdef RE_TRACK_PATTERN_OFFSETS
7941 DEBUG_OFFSETS_r(if (ri->u.offsets) {
7942 const STRLEN len = ri->u.offsets[0];
7944 GET_RE_DEBUG_FLAGS_DECL;
7945 Perl_re_printf( aTHX_
7946 "Offsets: [%" UVuf "]\n\t", (UV)ri->u.offsets[0]);
7947 for (i = 1; i <= len; i++) {
7948 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
7949 Perl_re_printf( aTHX_ "%" UVuf ":%" UVuf "[%" UVuf "] ",
7950 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
7952 Perl_re_printf( aTHX_ "\n");
7957 /* under ithreads the ?pat? PMf_USED flag on the pmop is simulated
7958 * by setting the regexp SV to readonly-only instead. If the
7959 * pattern's been recompiled, the USEDness should remain. */
7960 if (old_re && SvREADONLY(old_re))
7968 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
7971 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
7973 PERL_UNUSED_ARG(value);
7975 if (flags & RXapif_FETCH) {
7976 return reg_named_buff_fetch(rx, key, flags);
7977 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
7978 Perl_croak_no_modify();
7980 } else if (flags & RXapif_EXISTS) {
7981 return reg_named_buff_exists(rx, key, flags)
7984 } else if (flags & RXapif_REGNAMES) {
7985 return reg_named_buff_all(rx, flags);
7986 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
7987 return reg_named_buff_scalar(rx, flags);
7989 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
7995 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
7998 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
7999 PERL_UNUSED_ARG(lastkey);
8001 if (flags & RXapif_FIRSTKEY)
8002 return reg_named_buff_firstkey(rx, flags);
8003 else if (flags & RXapif_NEXTKEY)
8004 return reg_named_buff_nextkey(rx, flags);
8006 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter",
8013 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
8017 struct regexp *const rx = ReANY(r);
8019 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
8021 if (rx && RXp_PAREN_NAMES(rx)) {
8022 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
8025 SV* sv_dat=HeVAL(he_str);
8026 I32 *nums=(I32*)SvPVX(sv_dat);
8027 AV * const retarray = (flags & RXapif_ALL) ? newAV() : NULL;
8028 for ( i=0; i<SvIVX(sv_dat); i++ ) {
8029 if ((I32)(rx->nparens) >= nums[i]
8030 && rx->offs[nums[i]].start != -1
8031 && rx->offs[nums[i]].end != -1)
8034 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
8039 ret = newSVsv(&PL_sv_undef);
8042 av_push(retarray, ret);
8045 return newRV_noinc(MUTABLE_SV(retarray));
8052 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
8055 struct regexp *const rx = ReANY(r);
8057 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
8059 if (rx && RXp_PAREN_NAMES(rx)) {
8060 if (flags & RXapif_ALL) {
8061 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
8063 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
8065 SvREFCNT_dec_NN(sv);
8077 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
8079 struct regexp *const rx = ReANY(r);
8081 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
8083 if ( rx && RXp_PAREN_NAMES(rx) ) {
8084 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
8086 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
8093 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
8095 struct regexp *const rx = ReANY(r);
8096 GET_RE_DEBUG_FLAGS_DECL;
8098 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
8100 if (rx && RXp_PAREN_NAMES(rx)) {
8101 HV *hv = RXp_PAREN_NAMES(rx);
8103 while ( (temphe = hv_iternext_flags(hv,0)) ) {
8106 SV* sv_dat = HeVAL(temphe);
8107 I32 *nums = (I32*)SvPVX(sv_dat);
8108 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
8109 if ((I32)(rx->lastparen) >= nums[i] &&
8110 rx->offs[nums[i]].start != -1 &&
8111 rx->offs[nums[i]].end != -1)
8117 if (parno || flags & RXapif_ALL) {
8118 return newSVhek(HeKEY_hek(temphe));
8126 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
8131 struct regexp *const rx = ReANY(r);
8133 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
8135 if (rx && RXp_PAREN_NAMES(rx)) {
8136 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
8137 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
8138 } else if (flags & RXapif_ONE) {
8139 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
8140 av = MUTABLE_AV(SvRV(ret));
8141 length = av_tindex(av);
8142 SvREFCNT_dec_NN(ret);
8143 return newSViv(length + 1);
8145 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar",
8150 return &PL_sv_undef;
8154 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
8156 struct regexp *const rx = ReANY(r);
8159 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
8161 if (rx && RXp_PAREN_NAMES(rx)) {
8162 HV *hv= RXp_PAREN_NAMES(rx);
8164 (void)hv_iterinit(hv);
8165 while ( (temphe = hv_iternext_flags(hv,0)) ) {
8168 SV* sv_dat = HeVAL(temphe);
8169 I32 *nums = (I32*)SvPVX(sv_dat);
8170 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
8171 if ((I32)(rx->lastparen) >= nums[i] &&
8172 rx->offs[nums[i]].start != -1 &&
8173 rx->offs[nums[i]].end != -1)
8179 if (parno || flags & RXapif_ALL) {
8180 av_push(av, newSVhek(HeKEY_hek(temphe)));
8185 return newRV_noinc(MUTABLE_SV(av));
8189 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
8192 struct regexp *const rx = ReANY(r);
8198 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
8200 if ( n == RX_BUFF_IDX_CARET_PREMATCH
8201 || n == RX_BUFF_IDX_CARET_FULLMATCH
8202 || n == RX_BUFF_IDX_CARET_POSTMATCH
8205 bool keepcopy = cBOOL(rx->extflags & RXf_PMf_KEEPCOPY);
8207 /* on something like
8210 * the KEEPCOPY is set on the PMOP rather than the regex */
8211 if (PL_curpm && r == PM_GETRE(PL_curpm))
8212 keepcopy = cBOOL(PL_curpm->op_pmflags & PMf_KEEPCOPY);
8221 if (n == RX_BUFF_IDX_CARET_FULLMATCH)
8222 /* no need to distinguish between them any more */
8223 n = RX_BUFF_IDX_FULLMATCH;
8225 if ((n == RX_BUFF_IDX_PREMATCH || n == RX_BUFF_IDX_CARET_PREMATCH)
8226 && rx->offs[0].start != -1)
8228 /* $`, ${^PREMATCH} */
8229 i = rx->offs[0].start;
8233 if ((n == RX_BUFF_IDX_POSTMATCH || n == RX_BUFF_IDX_CARET_POSTMATCH)
8234 && rx->offs[0].end != -1)
8236 /* $', ${^POSTMATCH} */
8237 s = rx->subbeg - rx->suboffset + rx->offs[0].end;
8238 i = rx->sublen + rx->suboffset - rx->offs[0].end;
8241 if ( 0 <= n && n <= (I32)rx->nparens &&
8242 (s1 = rx->offs[n].start) != -1 &&
8243 (t1 = rx->offs[n].end) != -1)
8245 /* $&, ${^MATCH}, $1 ... */
8247 s = rx->subbeg + s1 - rx->suboffset;
8252 assert(s >= rx->subbeg);
8253 assert((STRLEN)rx->sublen >= (STRLEN)((s - rx->subbeg) + i) );
8255 #ifdef NO_TAINT_SUPPORT
8256 sv_setpvn(sv, s, i);
8258 const int oldtainted = TAINT_get;
8260 sv_setpvn(sv, s, i);
8261 TAINT_set(oldtainted);
8263 if (RXp_MATCH_UTF8(rx))
8268 if (RXp_MATCH_TAINTED(rx)) {
8269 if (SvTYPE(sv) >= SVt_PVMG) {
8270 MAGIC* const mg = SvMAGIC(sv);
8273 SvMAGIC_set(sv, mg->mg_moremagic);
8275 if ((mgt = SvMAGIC(sv))) {
8276 mg->mg_moremagic = mgt;
8277 SvMAGIC_set(sv, mg);
8294 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
8295 SV const * const value)
8297 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
8299 PERL_UNUSED_ARG(rx);
8300 PERL_UNUSED_ARG(paren);
8301 PERL_UNUSED_ARG(value);
8304 Perl_croak_no_modify();
8308 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
8311 struct regexp *const rx = ReANY(r);
8315 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
8317 if ( paren == RX_BUFF_IDX_CARET_PREMATCH
8318 || paren == RX_BUFF_IDX_CARET_FULLMATCH
8319 || paren == RX_BUFF_IDX_CARET_POSTMATCH
8322 bool keepcopy = cBOOL(rx->extflags & RXf_PMf_KEEPCOPY);
8324 /* on something like
8327 * the KEEPCOPY is set on the PMOP rather than the regex */
8328 if (PL_curpm && r == PM_GETRE(PL_curpm))
8329 keepcopy = cBOOL(PL_curpm->op_pmflags & PMf_KEEPCOPY);
8335 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
8337 case RX_BUFF_IDX_CARET_PREMATCH: /* ${^PREMATCH} */
8338 case RX_BUFF_IDX_PREMATCH: /* $` */
8339 if (rx->offs[0].start != -1) {
8340 i = rx->offs[0].start;
8349 case RX_BUFF_IDX_CARET_POSTMATCH: /* ${^POSTMATCH} */
8350 case RX_BUFF_IDX_POSTMATCH: /* $' */
8351 if (rx->offs[0].end != -1) {
8352 i = rx->sublen - rx->offs[0].end;
8354 s1 = rx->offs[0].end;
8361 default: /* $& / ${^MATCH}, $1, $2, ... */
8362 if (paren <= (I32)rx->nparens &&
8363 (s1 = rx->offs[paren].start) != -1 &&
8364 (t1 = rx->offs[paren].end) != -1)
8370 if (ckWARN(WARN_UNINITIALIZED))
8371 report_uninit((const SV *)sv);
8376 if (i > 0 && RXp_MATCH_UTF8(rx)) {
8377 const char * const s = rx->subbeg - rx->suboffset + s1;
8382 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
8389 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
8391 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
8392 PERL_UNUSED_ARG(rx);
8396 return newSVpvs("Regexp");
8399 /* Scans the name of a named buffer from the pattern.
8400 * If flags is REG_RSN_RETURN_NULL returns null.
8401 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
8402 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
8403 * to the parsed name as looked up in the RExC_paren_names hash.
8404 * If there is an error throws a vFAIL().. type exception.
8407 #define REG_RSN_RETURN_NULL 0
8408 #define REG_RSN_RETURN_NAME 1
8409 #define REG_RSN_RETURN_DATA 2
8412 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
8414 char *name_start = RExC_parse;
8416 PERL_ARGS_ASSERT_REG_SCAN_NAME;
8418 assert (RExC_parse <= RExC_end);
8419 if (RExC_parse == RExC_end) NOOP;
8420 else if (isIDFIRST_lazy_if_safe(RExC_parse, RExC_end, UTF)) {
8421 /* Note that the code here assumes well-formed UTF-8. Skip IDFIRST by
8422 * using do...while */
8425 RExC_parse += UTF8SKIP(RExC_parse);
8426 } while ( RExC_parse < RExC_end
8427 && isWORDCHAR_utf8_safe((U8*)RExC_parse, (U8*) RExC_end));
8431 } while (RExC_parse < RExC_end && isWORDCHAR(*RExC_parse));
8433 RExC_parse++; /* so the <- from the vFAIL is after the offending
8435 vFAIL("Group name must start with a non-digit word character");
8439 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
8440 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
8441 if ( flags == REG_RSN_RETURN_NAME)
8443 else if (flags==REG_RSN_RETURN_DATA) {
8446 if ( ! sv_name ) /* should not happen*/
8447 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
8448 if (RExC_paren_names)
8449 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
8451 sv_dat = HeVAL(he_str);
8453 vFAIL("Reference to nonexistent named group");
8457 Perl_croak(aTHX_ "panic: bad flag %lx in reg_scan_name",
8458 (unsigned long) flags);
8460 NOT_REACHED; /* NOTREACHED */
8465 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
8467 if (RExC_lastparse!=RExC_parse) { \
8468 Perl_re_printf( aTHX_ "%s", \
8469 Perl_pv_pretty(aTHX_ RExC_mysv1, RExC_parse, \
8470 RExC_end - RExC_parse, 16, \
8472 PERL_PV_ESCAPE_UNI_DETECT | \
8473 PERL_PV_PRETTY_ELLIPSES | \
8474 PERL_PV_PRETTY_LTGT | \
8475 PERL_PV_ESCAPE_RE | \
8476 PERL_PV_PRETTY_EXACTSIZE \
8480 Perl_re_printf( aTHX_ "%16s",""); \
8483 num = RExC_size + 1; \
8485 num=REG_NODE_NUM(RExC_emit); \
8486 if (RExC_lastnum!=num) \
8487 Perl_re_printf( aTHX_ "|%4d",num); \
8489 Perl_re_printf( aTHX_ "|%4s",""); \
8490 Perl_re_printf( aTHX_ "|%*s%-4s", \
8491 (int)((depth*2)), "", \
8495 RExC_lastparse=RExC_parse; \
8500 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
8501 DEBUG_PARSE_MSG((funcname)); \
8502 Perl_re_printf( aTHX_ "%4s","\n"); \
8504 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({\
8505 DEBUG_PARSE_MSG((funcname)); \
8506 Perl_re_printf( aTHX_ fmt "\n",args); \
8509 /* This section of code defines the inversion list object and its methods. The
8510 * interfaces are highly subject to change, so as much as possible is static to
8511 * this file. An inversion list is here implemented as a malloc'd C UV array
8512 * as an SVt_INVLIST scalar.
8514 * An inversion list for Unicode is an array of code points, sorted by ordinal
8515 * number. Each element gives the code point that begins a range that extends
8516 * up-to but not including the code point given by the next element. The final
8517 * element gives the first code point of a range that extends to the platform's
8518 * infinity. The even-numbered elements (invlist[0], invlist[2], invlist[4],
8519 * ...) give ranges whose code points are all in the inversion list. We say
8520 * that those ranges are in the set. The odd-numbered elements give ranges
8521 * whose code points are not in the inversion list, and hence not in the set.
8522 * Thus, element [0] is the first code point in the list. Element [1]
8523 * is the first code point beyond that not in the list; and element [2] is the
8524 * first code point beyond that that is in the list. In other words, the first
8525 * range is invlist[0]..(invlist[1]-1), and all code points in that range are
8526 * in the inversion list. The second range is invlist[1]..(invlist[2]-1), and
8527 * all code points in that range are not in the inversion list. The third
8528 * range invlist[2]..(invlist[3]-1) gives code points that are in the inversion
8529 * list, and so forth. Thus every element whose index is divisible by two
8530 * gives the beginning of a range that is in the list, and every element whose
8531 * index is not divisible by two gives the beginning of a range not in the
8532 * list. If the final element's index is divisible by two, the inversion list
8533 * extends to the platform's infinity; otherwise the highest code point in the
8534 * inversion list is the contents of that element minus 1.
8536 * A range that contains just a single code point N will look like
8538 * invlist[i+1] == N+1
8540 * If N is UV_MAX (the highest representable code point on the machine), N+1 is
8541 * impossible to represent, so element [i+1] is omitted. The single element
8543 * invlist[0] == UV_MAX
8544 * contains just UV_MAX, but is interpreted as matching to infinity.
8546 * Taking the complement (inverting) an inversion list is quite simple, if the
8547 * first element is 0, remove it; otherwise add a 0 element at the beginning.
8548 * This implementation reserves an element at the beginning of each inversion
8549 * list to always contain 0; there is an additional flag in the header which
8550 * indicates if the list begins at the 0, or is offset to begin at the next
8551 * element. This means that the inversion list can be inverted without any
8552 * copying; just flip the flag.
8554 * More about inversion lists can be found in "Unicode Demystified"
8555 * Chapter 13 by Richard Gillam, published by Addison-Wesley.
8557 * The inversion list data structure is currently implemented as an SV pointing
8558 * to an array of UVs that the SV thinks are bytes. This allows us to have an
8559 * array of UV whose memory management is automatically handled by the existing
8560 * facilities for SV's.
8562 * Some of the methods should always be private to the implementation, and some
8563 * should eventually be made public */
8565 /* The header definitions are in F<invlist_inline.h> */
8567 #ifndef PERL_IN_XSUB_RE
8569 PERL_STATIC_INLINE UV*
8570 S__invlist_array_init(SV* const invlist, const bool will_have_0)
8572 /* Returns a pointer to the first element in the inversion list's array.
8573 * This is called upon initialization of an inversion list. Where the
8574 * array begins depends on whether the list has the code point U+0000 in it
8575 * or not. The other parameter tells it whether the code that follows this
8576 * call is about to put a 0 in the inversion list or not. The first
8577 * element is either the element reserved for 0, if TRUE, or the element
8578 * after it, if FALSE */
8580 bool* offset = get_invlist_offset_addr(invlist);
8581 UV* zero_addr = (UV *) SvPVX(invlist);
8583 PERL_ARGS_ASSERT__INVLIST_ARRAY_INIT;
8586 assert(! _invlist_len(invlist));
8590 /* 1^1 = 0; 1^0 = 1 */
8591 *offset = 1 ^ will_have_0;
8592 return zero_addr + *offset;
8597 PERL_STATIC_INLINE void
8598 S_invlist_set_len(pTHX_ SV* const invlist, const UV len, const bool offset)
8600 /* Sets the current number of elements stored in the inversion list.
8601 * Updates SvCUR correspondingly */
8602 PERL_UNUSED_CONTEXT;
8603 PERL_ARGS_ASSERT_INVLIST_SET_LEN;
8605 assert(SvTYPE(invlist) == SVt_INVLIST);
8610 : TO_INTERNAL_SIZE(len + offset));
8611 assert(SvLEN(invlist) == 0 || SvCUR(invlist) <= SvLEN(invlist));
8614 #ifndef PERL_IN_XSUB_RE
8617 S_invlist_replace_list_destroys_src(pTHX_ SV * dest, SV * src)
8619 /* Replaces the inversion list in 'dest' with the one from 'src'. It
8620 * steals the list from 'src', so 'src' is made to have a NULL list. This
8621 * is similar to what SvSetMagicSV() would do, if it were implemented on
8622 * inversion lists, though this routine avoids a copy */
8624 const UV src_len = _invlist_len(src);
8625 const bool src_offset = *get_invlist_offset_addr(src);
8626 const STRLEN src_byte_len = SvLEN(src);
8627 char * array = SvPVX(src);
8629 const int oldtainted = TAINT_get;
8631 PERL_ARGS_ASSERT_INVLIST_REPLACE_LIST_DESTROYS_SRC;
8633 assert(SvTYPE(src) == SVt_INVLIST);
8634 assert(SvTYPE(dest) == SVt_INVLIST);
8635 assert(! invlist_is_iterating(src));
8636 assert(SvCUR(src) == 0 || SvCUR(src) < SvLEN(src));
8638 /* Make sure it ends in the right place with a NUL, as our inversion list
8639 * manipulations aren't careful to keep this true, but sv_usepvn_flags()
8641 array[src_byte_len - 1] = '\0';
8643 TAINT_NOT; /* Otherwise it breaks */
8644 sv_usepvn_flags(dest,
8648 /* This flag is documented to cause a copy to be avoided */
8649 SV_HAS_TRAILING_NUL);
8650 TAINT_set(oldtainted);
8655 /* Finish up copying over the other fields in an inversion list */
8656 *get_invlist_offset_addr(dest) = src_offset;
8657 invlist_set_len(dest, src_len, src_offset);
8658 *get_invlist_previous_index_addr(dest) = 0;
8659 invlist_iterfinish(dest);
8662 PERL_STATIC_INLINE IV*
8663 S_get_invlist_previous_index_addr(SV* invlist)
8665 /* Return the address of the IV that is reserved to hold the cached index
8667 PERL_ARGS_ASSERT_GET_INVLIST_PREVIOUS_INDEX_ADDR;
8669 assert(SvTYPE(invlist) == SVt_INVLIST);
8671 return &(((XINVLIST*) SvANY(invlist))->prev_index);
8674 PERL_STATIC_INLINE IV
8675 S_invlist_previous_index(SV* const invlist)
8677 /* Returns cached index of previous search */
8679 PERL_ARGS_ASSERT_INVLIST_PREVIOUS_INDEX;
8681 return *get_invlist_previous_index_addr(invlist);
8684 PERL_STATIC_INLINE void
8685 S_invlist_set_previous_index(SV* const invlist, const IV index)
8687 /* Caches <index> for later retrieval */
8689 PERL_ARGS_ASSERT_INVLIST_SET_PREVIOUS_INDEX;
8691 assert(index == 0 || index < (int) _invlist_len(invlist));
8693 *get_invlist_previous_index_addr(invlist) = index;
8696 PERL_STATIC_INLINE void
8697 S_invlist_trim(SV* invlist)
8699 /* Free the not currently-being-used space in an inversion list */
8701 /* But don't free up the space needed for the 0 UV that is always at the
8702 * beginning of the list, nor the trailing NUL */
8703 const UV min_size = TO_INTERNAL_SIZE(1) + 1;
8705 PERL_ARGS_ASSERT_INVLIST_TRIM;
8707 assert(SvTYPE(invlist) == SVt_INVLIST);
8709 SvPV_renew(invlist, MAX(min_size, SvCUR(invlist) + 1));
8712 PERL_STATIC_INLINE void
8713 S_invlist_clear(pTHX_ SV* invlist) /* Empty the inversion list */
8715 PERL_ARGS_ASSERT_INVLIST_CLEAR;
8717 assert(SvTYPE(invlist) == SVt_INVLIST);
8719 invlist_set_len(invlist, 0, 0);
8720 invlist_trim(invlist);
8723 #endif /* ifndef PERL_IN_XSUB_RE */
8725 PERL_STATIC_INLINE bool
8726 S_invlist_is_iterating(SV* const invlist)
8728 PERL_ARGS_ASSERT_INVLIST_IS_ITERATING;
8730 return *(get_invlist_iter_addr(invlist)) < (STRLEN) UV_MAX;
8733 #ifndef PERL_IN_XSUB_RE
8735 PERL_STATIC_INLINE UV
8736 S_invlist_max(SV* const invlist)
8738 /* Returns the maximum number of elements storable in the inversion list's
8739 * array, without having to realloc() */
8741 PERL_ARGS_ASSERT_INVLIST_MAX;
8743 assert(SvTYPE(invlist) == SVt_INVLIST);
8745 /* Assumes worst case, in which the 0 element is not counted in the
8746 * inversion list, so subtracts 1 for that */
8747 return SvLEN(invlist) == 0 /* This happens under _new_invlist_C_array */
8748 ? FROM_INTERNAL_SIZE(SvCUR(invlist)) - 1
8749 : FROM_INTERNAL_SIZE(SvLEN(invlist)) - 1;
8752 Perl__new_invlist(pTHX_ IV initial_size)
8755 /* Return a pointer to a newly constructed inversion list, with enough
8756 * space to store 'initial_size' elements. If that number is negative, a
8757 * system default is used instead */
8761 if (initial_size < 0) {
8765 /* Allocate the initial space */
8766 new_list = newSV_type(SVt_INVLIST);
8768 /* First 1 is in case the zero element isn't in the list; second 1 is for
8770 SvGROW(new_list, TO_INTERNAL_SIZE(initial_size + 1) + 1);
8771 invlist_set_len(new_list, 0, 0);
8773 /* Force iterinit() to be used to get iteration to work */
8774 *get_invlist_iter_addr(new_list) = (STRLEN) UV_MAX;
8776 *get_invlist_previous_index_addr(new_list) = 0;
8782 Perl__new_invlist_C_array(pTHX_ const UV* const list)
8784 /* Return a pointer to a newly constructed inversion list, initialized to
8785 * point to <list>, which has to be in the exact correct inversion list
8786 * form, including internal fields. Thus this is a dangerous routine that
8787 * should not be used in the wrong hands. The passed in 'list' contains
8788 * several header fields at the beginning that are not part of the
8789 * inversion list body proper */
8791 const STRLEN length = (STRLEN) list[0];
8792 const UV version_id = list[1];
8793 const bool offset = cBOOL(list[2]);
8794 #define HEADER_LENGTH 3
8795 /* If any of the above changes in any way, you must change HEADER_LENGTH
8796 * (if appropriate) and regenerate INVLIST_VERSION_ID by running
8797 * perl -E 'say int(rand 2**31-1)'
8799 #define INVLIST_VERSION_ID 148565664 /* This is a combination of a version and
8800 data structure type, so that one being
8801 passed in can be validated to be an
8802 inversion list of the correct vintage.
8805 SV* invlist = newSV_type(SVt_INVLIST);
8807 PERL_ARGS_ASSERT__NEW_INVLIST_C_ARRAY;
8809 if (version_id != INVLIST_VERSION_ID) {
8810 Perl_croak(aTHX_ "panic: Incorrect version for previously generated inversion list");
8813 /* The generated array passed in includes header elements that aren't part
8814 * of the list proper, so start it just after them */
8815 SvPV_set(invlist, (char *) (list + HEADER_LENGTH));
8817 SvLEN_set(invlist, 0); /* Means we own the contents, and the system
8818 shouldn't touch it */
8820 *(get_invlist_offset_addr(invlist)) = offset;
8822 /* The 'length' passed to us is the physical number of elements in the
8823 * inversion list. But if there is an offset the logical number is one
8825 invlist_set_len(invlist, length - offset, offset);
8827 invlist_set_previous_index(invlist, 0);
8829 /* Initialize the iteration pointer. */
8830 invlist_iterfinish(invlist);
8832 SvREADONLY_on(invlist);
8838 S_invlist_extend(pTHX_ SV* const invlist, const UV new_max)
8840 /* Grow the maximum size of an inversion list */
8842 PERL_ARGS_ASSERT_INVLIST_EXTEND;
8844 assert(SvTYPE(invlist) == SVt_INVLIST);
8846 /* Add one to account for the zero element at the beginning which may not
8847 * be counted by the calling parameters */
8848 SvGROW((SV *)invlist, TO_INTERNAL_SIZE(new_max + 1));
8852 S__append_range_to_invlist(pTHX_ SV* const invlist,
8853 const UV start, const UV end)
8855 /* Subject to change or removal. Append the range from 'start' to 'end' at
8856 * the end of the inversion list. The range must be above any existing
8860 UV max = invlist_max(invlist);
8861 UV len = _invlist_len(invlist);
8864 PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST;
8866 if (len == 0) { /* Empty lists must be initialized */
8867 offset = start != 0;
8868 array = _invlist_array_init(invlist, ! offset);
8871 /* Here, the existing list is non-empty. The current max entry in the
8872 * list is generally the first value not in the set, except when the
8873 * set extends to the end of permissible values, in which case it is
8874 * the first entry in that final set, and so this call is an attempt to
8875 * append out-of-order */
8877 UV final_element = len - 1;
8878 array = invlist_array(invlist);
8879 if ( array[final_element] > start
8880 || ELEMENT_RANGE_MATCHES_INVLIST(final_element))
8882 Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list, final=%" UVuf ", start=%" UVuf ", match=%c",
8883 array[final_element], start,
8884 ELEMENT_RANGE_MATCHES_INVLIST(final_element) ? 't' : 'f');
8887 /* Here, it is a legal append. If the new range begins 1 above the end
8888 * of the range below it, it is extending the range below it, so the
8889 * new first value not in the set is one greater than the newly
8890 * extended range. */
8891 offset = *get_invlist_offset_addr(invlist);
8892 if (array[final_element] == start) {
8893 if (end != UV_MAX) {
8894 array[final_element] = end + 1;
8897 /* But if the end is the maximum representable on the machine,
8898 * assume that infinity was actually what was meant. Just let
8899 * the range that this would extend to have no end */
8900 invlist_set_len(invlist, len - 1, offset);
8906 /* Here the new range doesn't extend any existing set. Add it */
8908 len += 2; /* Includes an element each for the start and end of range */
8910 /* If wll overflow the existing space, extend, which may cause the array to
8913 invlist_extend(invlist, len);
8915 /* Have to set len here to avoid assert failure in invlist_array() */
8916 invlist_set_len(invlist, len, offset);
8918 array = invlist_array(invlist);
8921 invlist_set_len(invlist, len, offset);
8924 /* The next item on the list starts the range, the one after that is
8925 * one past the new range. */
8926 array[len - 2] = start;
8927 if (end != UV_MAX) {
8928 array[len - 1] = end + 1;
8931 /* But if the end is the maximum representable on the machine, just let
8932 * the range have no end */
8933 invlist_set_len(invlist, len - 1, offset);
8938 Perl__invlist_search(SV* const invlist, const UV cp)
8940 /* Searches the inversion list for the entry that contains the input code
8941 * point <cp>. If <cp> is not in the list, -1 is returned. Otherwise, the
8942 * return value is the index into the list's array of the range that
8943 * contains <cp>, that is, 'i' such that
8944 * array[i] <= cp < array[i+1]
8949 IV high = _invlist_len(invlist);
8950 const IV highest_element = high - 1;
8953 PERL_ARGS_ASSERT__INVLIST_SEARCH;
8955 /* If list is empty, return failure. */
8960 /* (We can't get the array unless we know the list is non-empty) */
8961 array = invlist_array(invlist);
8963 mid = invlist_previous_index(invlist);
8965 if (mid > highest_element) {
8966 mid = highest_element;
8969 /* <mid> contains the cache of the result of the previous call to this
8970 * function (0 the first time). See if this call is for the same result,
8971 * or if it is for mid-1. This is under the theory that calls to this
8972 * function will often be for related code points that are near each other.
8973 * And benchmarks show that caching gives better results. We also test
8974 * here if the code point is within the bounds of the list. These tests
8975 * replace others that would have had to be made anyway to make sure that
8976 * the array bounds were not exceeded, and these give us extra information
8977 * at the same time */
8978 if (cp >= array[mid]) {
8979 if (cp >= array[highest_element]) {
8980 return highest_element;
8983 /* Here, array[mid] <= cp < array[highest_element]. This means that
8984 * the final element is not the answer, so can exclude it; it also
8985 * means that <mid> is not the final element, so can refer to 'mid + 1'
8987 if (cp < array[mid + 1]) {
8993 else { /* cp < aray[mid] */
8994 if (cp < array[0]) { /* Fail if outside the array */
8998 if (cp >= array[mid - 1]) {
9003 /* Binary search. What we are looking for is <i> such that
9004 * array[i] <= cp < array[i+1]
9005 * The loop below converges on the i+1. Note that there may not be an
9006 * (i+1)th element in the array, and things work nonetheless */
9007 while (low < high) {
9008 mid = (low + high) / 2;
9009 assert(mid <= highest_element);
9010 if (array[mid] <= cp) { /* cp >= array[mid] */
9013 /* We could do this extra test to exit the loop early.
9014 if (cp < array[low]) {
9019 else { /* cp < array[mid] */
9026 invlist_set_previous_index(invlist, high);
9031 Perl__invlist_populate_swatch(SV* const invlist,
9032 const UV start, const UV end, U8* swatch)
9034 /* populates a swatch of a swash the same way swatch_get() does in utf8.c,
9035 * but is used when the swash has an inversion list. This makes this much
9036 * faster, as it uses a binary search instead of a linear one. This is
9037 * intimately tied to that function, and perhaps should be in utf8.c,
9038 * except it is intimately tied to inversion lists as well. It assumes
9039 * that <swatch> is all 0's on input */
9042 const IV len = _invlist_len(invlist);
9046 PERL_ARGS_ASSERT__INVLIST_POPULATE_SWATCH;
9048 if (len == 0) { /* Empty inversion list */
9052 array = invlist_array(invlist);
9054 /* Find which element it is */
9055 i = _invlist_search(invlist, start);
9057 /* We populate from <start> to <end> */
9058 while (current < end) {
9061 /* The inversion list gives the results for every possible code point
9062 * after the first one in the list. Only those ranges whose index is
9063 * even are ones that the inversion list matches. For the odd ones,
9064 * and if the initial code point is not in the list, we have to skip
9065 * forward to the next element */
9066 if (i == -1 || ! ELEMENT_RANGE_MATCHES_INVLIST(i)) {
9068 if (i >= len) { /* Finished if beyond the end of the array */
9072 if (current >= end) { /* Finished if beyond the end of what we
9074 if (LIKELY(end < UV_MAX)) {
9078 /* We get here when the upper bound is the maximum
9079 * representable on the machine, and we are looking for just
9080 * that code point. Have to special case it */
9082 goto join_end_of_list;
9085 assert(current >= start);
9087 /* The current range ends one below the next one, except don't go past
9090 upper = (i < len && array[i] < end) ? array[i] : end;
9092 /* Here we are in a range that matches. Populate a bit in the 3-bit U8
9093 * for each code point in it */
9094 for (; current < upper; current++) {
9095 const STRLEN offset = (STRLEN)(current - start);
9096 swatch[offset >> 3] |= 1 << (offset & 7);
9101 /* Quit if at the end of the list */
9104 /* But first, have to deal with the highest possible code point on
9105 * the platform. The previous code assumes that <end> is one
9106 * beyond where we want to populate, but that is impossible at the
9107 * platform's infinity, so have to handle it specially */
9108 if (UNLIKELY(end == UV_MAX && ELEMENT_RANGE_MATCHES_INVLIST(len-1)))
9110 const STRLEN offset = (STRLEN)(end - start);
9111 swatch[offset >> 3] |= 1 << (offset & 7);
9116 /* Advance to the next range, which will be for code points not in the
9125 Perl__invlist_union_maybe_complement_2nd(pTHX_ SV* const a, SV* const b,
9126 const bool complement_b, SV** output)
9128 /* Take the union of two inversion lists and point '*output' to it. On
9129 * input, '*output' MUST POINT TO NULL OR TO AN SV* INVERSION LIST (possibly
9130 * even 'a' or 'b'). If to an inversion list, the contents of the original
9131 * list will be replaced by the union. The first list, 'a', may be
9132 * NULL, in which case a copy of the second list is placed in '*output'.
9133 * If 'complement_b' is TRUE, the union is taken of the complement
9134 * (inversion) of 'b' instead of b itself.
9136 * The basis for this comes from "Unicode Demystified" Chapter 13 by
9137 * Richard Gillam, published by Addison-Wesley, and explained at some
9138 * length there. The preface says to incorporate its examples into your
9139 * code at your own risk.
9141 * The algorithm is like a merge sort. */
9143 const UV* array_a; /* a's array */
9145 UV len_a; /* length of a's array */
9148 SV* u; /* the resulting union */
9152 UV i_a = 0; /* current index into a's array */
9156 /* running count, as explained in the algorithm source book; items are
9157 * stopped accumulating and are output when the count changes to/from 0.
9158 * The count is incremented when we start a range that's in an input's set,
9159 * and decremented when we start a range that's not in a set. So this
9160 * variable can be 0, 1, or 2. When it is 0 neither input is in their set,
9161 * and hence nothing goes into the union; 1, just one of the inputs is in
9162 * its set (and its current range gets added to the union); and 2 when both
9163 * inputs are in their sets. */
9166 PERL_ARGS_ASSERT__INVLIST_UNION_MAYBE_COMPLEMENT_2ND;
9168 assert(*output == NULL || SvTYPE(*output) == SVt_INVLIST);
9170 len_b = _invlist_len(b);
9173 /* Here, 'b' is empty, hence it's complement is all possible code
9174 * points. So if the union includes the complement of 'b', it includes
9175 * everything, and we need not even look at 'a'. It's easiest to
9176 * create a new inversion list that matches everything. */
9178 SV* everything = _add_range_to_invlist(NULL, 0, UV_MAX);
9180 if (*output == NULL) { /* If the output didn't exist, just point it
9182 *output = everything;
9184 else { /* Otherwise, replace its contents with the new list */
9185 invlist_replace_list_destroys_src(*output, everything);
9186 SvREFCNT_dec_NN(everything);
9192 /* Here, we don't want the complement of 'b', and since 'b' is empty,
9193 * the union will come entirely from 'a'. If 'a' is NULL or empty, the
9194 * output will be empty */
9196 if (a == NULL || _invlist_len(a) == 0) {
9197 if (*output == NULL) {
9198 *output = _new_invlist(0);
9201 invlist_clear(*output);
9206 /* Here, 'a' is not empty, but 'b' is, so 'a' entirely determines the
9207 * union. We can just return a copy of 'a' if '*output' doesn't point
9208 * to an existing list */
9209 if (*output == NULL) {
9210 *output = invlist_clone(a);
9214 /* If the output is to overwrite 'a', we have a no-op, as it's
9220 /* Here, '*output' is to be overwritten by 'a' */
9221 u = invlist_clone(a);
9222 invlist_replace_list_destroys_src(*output, u);
9228 /* Here 'b' is not empty. See about 'a' */
9230 if (a == NULL || ((len_a = _invlist_len(a)) == 0)) {
9232 /* Here, 'a' is empty (and b is not). That means the union will come
9233 * entirely from 'b'. If '*output' is NULL, we can directly return a
9234 * clone of 'b'. Otherwise, we replace the contents of '*output' with
9237 SV ** dest = (*output == NULL) ? output : &u;
9238 *dest = invlist_clone(b);
9240 _invlist_invert(*dest);
9244 invlist_replace_list_destroys_src(*output, u);
9251 /* Here both lists exist and are non-empty */
9252 array_a = invlist_array(a);
9253 array_b = invlist_array(b);
9255 /* If are to take the union of 'a' with the complement of b, set it
9256 * up so are looking at b's complement. */
9259 /* To complement, we invert: if the first element is 0, remove it. To
9260 * do this, we just pretend the array starts one later */
9261 if (array_b[0] == 0) {
9267 /* But if the first element is not zero, we pretend the list starts
9268 * at the 0 that is always stored immediately before the array. */
9274 /* Size the union for the worst case: that the sets are completely
9276 u = _new_invlist(len_a + len_b);
9278 /* Will contain U+0000 if either component does */
9279 array_u = _invlist_array_init(u, ( len_a > 0 && array_a[0] == 0)
9280 || (len_b > 0 && array_b[0] == 0));
9282 /* Go through each input list item by item, stopping when have exhausted
9284 while (i_a < len_a && i_b < len_b) {
9285 UV cp; /* The element to potentially add to the union's array */
9286 bool cp_in_set; /* is it in the the input list's set or not */
9288 /* We need to take one or the other of the two inputs for the union.
9289 * Since we are merging two sorted lists, we take the smaller of the
9290 * next items. In case of a tie, we take first the one that is in its
9291 * set. If we first took the one not in its set, it would decrement
9292 * the count, possibly to 0 which would cause it to be output as ending
9293 * the range, and the next time through we would take the same number,
9294 * and output it again as beginning the next range. By doing it the
9295 * opposite way, there is no possibility that the count will be
9296 * momentarily decremented to 0, and thus the two adjoining ranges will
9297 * be seamlessly merged. (In a tie and both are in the set or both not
9298 * in the set, it doesn't matter which we take first.) */
9299 if ( array_a[i_a] < array_b[i_b]
9300 || ( array_a[i_a] == array_b[i_b]
9301 && ELEMENT_RANGE_MATCHES_INVLIST(i_a)))
9303 cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_a);
9304 cp = array_a[i_a++];
9307 cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_b);
9308 cp = array_b[i_b++];
9311 /* Here, have chosen which of the two inputs to look at. Only output
9312 * if the running count changes to/from 0, which marks the
9313 * beginning/end of a range that's in the set */
9316 array_u[i_u++] = cp;
9323 array_u[i_u++] = cp;
9329 /* The loop above increments the index into exactly one of the input lists
9330 * each iteration, and ends when either index gets to its list end. That
9331 * means the other index is lower than its end, and so something is
9332 * remaining in that one. We decrement 'count', as explained below, if
9333 * that list is in its set. (i_a and i_b each currently index the element
9334 * beyond the one we care about.) */
9335 if ( (i_a != len_a && PREV_RANGE_MATCHES_INVLIST(i_a))
9336 || (i_b != len_b && PREV_RANGE_MATCHES_INVLIST(i_b)))
9341 /* Above we decremented 'count' if the list that had unexamined elements in
9342 * it was in its set. This has made it so that 'count' being non-zero
9343 * means there isn't anything left to output; and 'count' equal to 0 means
9344 * that what is left to output is precisely that which is left in the
9345 * non-exhausted input list.
9347 * To see why, note first that the exhausted input obviously has nothing
9348 * left to add to the union. If it was in its set at its end, that means
9349 * the set extends from here to the platform's infinity, and hence so does
9350 * the union and the non-exhausted set is irrelevant. The exhausted set
9351 * also contributed 1 to 'count'. If 'count' was 2, it got decremented to
9352 * 1, but if it was 1, the non-exhausted set wasn't in its set, and so
9353 * 'count' remains at 1. This is consistent with the decremented 'count'
9354 * != 0 meaning there's nothing left to add to the union.
9356 * But if the exhausted input wasn't in its set, it contributed 0 to
9357 * 'count', and the rest of the union will be whatever the other input is.
9358 * If 'count' was 0, neither list was in its set, and 'count' remains 0;
9359 * otherwise it gets decremented to 0. This is consistent with 'count'
9360 * == 0 meaning the remainder of the union is whatever is left in the
9361 * non-exhausted list. */
9366 IV copy_count = len_a - i_a;
9367 if (copy_count > 0) { /* The non-exhausted input is 'a' */
9368 Copy(array_a + i_a, array_u + i_u, copy_count, UV);
9370 else { /* The non-exhausted input is b */
9371 copy_count = len_b - i_b;
9372 Copy(array_b + i_b, array_u + i_u, copy_count, UV);
9374 len_u = i_u + copy_count;
9377 /* Set the result to the final length, which can change the pointer to
9378 * array_u, so re-find it. (Note that it is unlikely that this will
9379 * change, as we are shrinking the space, not enlarging it) */
9380 if (len_u != _invlist_len(u)) {
9381 invlist_set_len(u, len_u, *get_invlist_offset_addr(u));
9383 array_u = invlist_array(u);
9386 if (*output == NULL) { /* Simply return the new inversion list */
9390 /* Otherwise, overwrite the inversion list that was in '*output'. We
9391 * could instead free '*output', and then set it to 'u', but experience
9392 * has shown [perl #127392] that if the input is a mortal, we can get a
9393 * huge build-up of these during regex compilation before they get
9395 invlist_replace_list_destroys_src(*output, u);
9403 Perl__invlist_intersection_maybe_complement_2nd(pTHX_ SV* const a, SV* const b,
9404 const bool complement_b, SV** i)
9406 /* Take the intersection of two inversion lists and point '*i' to it. On
9407 * input, '*i' MUST POINT TO NULL OR TO AN SV* INVERSION LIST (possibly
9408 * even 'a' or 'b'). If to an inversion list, the contents of the original
9409 * list will be replaced by the intersection. The first list, 'a', may be
9410 * NULL, in which case '*i' will be an empty list. If 'complement_b' is
9411 * TRUE, the result will be the intersection of 'a' and the complement (or
9412 * inversion) of 'b' instead of 'b' directly.
9414 * The basis for this comes from "Unicode Demystified" Chapter 13 by
9415 * Richard Gillam, published by Addison-Wesley, and explained at some
9416 * length there. The preface says to incorporate its examples into your
9417 * code at your own risk. In fact, it had bugs
9419 * The algorithm is like a merge sort, and is essentially the same as the
9423 const UV* array_a; /* a's array */
9425 UV len_a; /* length of a's array */
9428 SV* r; /* the resulting intersection */
9432 UV i_a = 0; /* current index into a's array */
9436 /* running count of how many of the two inputs are postitioned at ranges
9437 * that are in their sets. As explained in the algorithm source book,
9438 * items are stopped accumulating and are output when the count changes
9439 * to/from 2. The count is incremented when we start a range that's in an
9440 * input's set, and decremented when we start a range that's not in a set.
9441 * Only when it is 2 are we in the intersection. */
9444 PERL_ARGS_ASSERT__INVLIST_INTERSECTION_MAYBE_COMPLEMENT_2ND;
9446 assert(*i == NULL || SvTYPE(*i) == SVt_INVLIST);
9448 /* Special case if either one is empty */
9449 len_a = (a == NULL) ? 0 : _invlist_len(a);
9450 if ((len_a == 0) || ((len_b = _invlist_len(b)) == 0)) {
9451 if (len_a != 0 && complement_b) {
9453 /* Here, 'a' is not empty, therefore from the enclosing 'if', 'b'
9454 * must be empty. Here, also we are using 'b's complement, which
9455 * hence must be every possible code point. Thus the intersection
9458 if (*i == a) { /* No-op */
9463 *i = invlist_clone(a);
9467 r = invlist_clone(a);
9468 invlist_replace_list_destroys_src(*i, r);
9473 /* Here, 'a' or 'b' is empty and not using the complement of 'b'. The
9474 * intersection must be empty */
9476 *i = _new_invlist(0);
9484 /* Here both lists exist and are non-empty */
9485 array_a = invlist_array(a);
9486 array_b = invlist_array(b);
9488 /* If are to take the intersection of 'a' with the complement of b, set it
9489 * up so are looking at b's complement. */
9492 /* To complement, we invert: if the first element is 0, remove it. To
9493 * do this, we just pretend the array starts one later */
9494 if (array_b[0] == 0) {
9500 /* But if the first element is not zero, we pretend the list starts
9501 * at the 0 that is always stored immediately before the array. */
9507 /* Size the intersection for the worst case: that the intersection ends up
9508 * fragmenting everything to be completely disjoint */
9509 r= _new_invlist(len_a + len_b);
9511 /* Will contain U+0000 iff both components do */
9512 array_r = _invlist_array_init(r, len_a > 0 && array_a[0] == 0
9513 && len_b > 0 && array_b[0] == 0);
9515 /* Go through each list item by item, stopping when have exhausted one of
9517 while (i_a < len_a && i_b < len_b) {
9518 UV cp; /* The element to potentially add to the intersection's
9520 bool cp_in_set; /* Is it in the input list's set or not */
9522 /* We need to take one or the other of the two inputs for the
9523 * intersection. Since we are merging two sorted lists, we take the
9524 * smaller of the next items. In case of a tie, we take first the one
9525 * that is not in its set (a difference from the union algorithm). If
9526 * we first took the one in its set, it would increment the count,
9527 * possibly to 2 which would cause it to be output as starting a range
9528 * in the intersection, and the next time through we would take that
9529 * same number, and output it again as ending the set. By doing the
9530 * opposite of this, there is no possibility that the count will be
9531 * momentarily incremented to 2. (In a tie and both are in the set or
9532 * both not in the set, it doesn't matter which we take first.) */
9533 if ( array_a[i_a] < array_b[i_b]
9534 || ( array_a[i_a] == array_b[i_b]
9535 && ! ELEMENT_RANGE_MATCHES_INVLIST(i_a)))
9537 cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_a);
9538 cp = array_a[i_a++];
9541 cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_b);
9545 /* Here, have chosen which of the two inputs to look at. Only output
9546 * if the running count changes to/from 2, which marks the
9547 * beginning/end of a range that's in the intersection */
9551 array_r[i_r++] = cp;
9556 array_r[i_r++] = cp;
9563 /* The loop above increments the index into exactly one of the input lists
9564 * each iteration, and ends when either index gets to its list end. That
9565 * means the other index is lower than its end, and so something is
9566 * remaining in that one. We increment 'count', as explained below, if the
9567 * exhausted list was in its set. (i_a and i_b each currently index the
9568 * element beyond the one we care about.) */
9569 if ( (i_a == len_a && PREV_RANGE_MATCHES_INVLIST(i_a))
9570 || (i_b == len_b && PREV_RANGE_MATCHES_INVLIST(i_b)))
9575 /* Above we incremented 'count' if the exhausted list was in its set. This
9576 * has made it so that 'count' being below 2 means there is nothing left to
9577 * output; otheriwse what's left to add to the intersection is precisely
9578 * that which is left in the non-exhausted input list.
9580 * To see why, note first that the exhausted input obviously has nothing
9581 * left to affect the intersection. If it was in its set at its end, that
9582 * means the set extends from here to the platform's infinity, and hence
9583 * anything in the non-exhausted's list will be in the intersection, and
9584 * anything not in it won't be. Hence, the rest of the intersection is
9585 * precisely what's in the non-exhausted list The exhausted set also
9586 * contributed 1 to 'count', meaning 'count' was at least 1. Incrementing
9587 * it means 'count' is now at least 2. This is consistent with the
9588 * incremented 'count' being >= 2 means to add the non-exhausted list to
9591 * But if the exhausted input wasn't in its set, it contributed 0 to
9592 * 'count', and the intersection can't include anything further; the
9593 * non-exhausted set is irrelevant. 'count' was at most 1, and doesn't get
9594 * incremented. This is consistent with 'count' being < 2 meaning nothing
9595 * further to add to the intersection. */
9596 if (count < 2) { /* Nothing left to put in the intersection. */
9599 else { /* copy the non-exhausted list, unchanged. */
9600 IV copy_count = len_a - i_a;
9601 if (copy_count > 0) { /* a is the one with stuff left */
9602 Copy(array_a + i_a, array_r + i_r, copy_count, UV);
9604 else { /* b is the one with stuff left */
9605 copy_count = len_b - i_b;
9606 Copy(array_b + i_b, array_r + i_r, copy_count, UV);
9608 len_r = i_r + copy_count;
9611 /* Set the result to the final length, which can change the pointer to
9612 * array_r, so re-find it. (Note that it is unlikely that this will
9613 * change, as we are shrinking the space, not enlarging it) */
9614 if (len_r != _invlist_len(r)) {
9615 invlist_set_len(r, len_r, *get_invlist_offset_addr(r));
9617 array_r = invlist_array(r);
9620 if (*i == NULL) { /* Simply return the calculated intersection */
9623 else { /* Otherwise, replace the existing inversion list in '*i'. We could
9624 instead free '*i', and then set it to 'r', but experience has
9625 shown [perl #127392] that if the input is a mortal, we can get a
9626 huge build-up of these during regex compilation before they get
9629 invlist_replace_list_destroys_src(*i, r);
9641 Perl__add_range_to_invlist(pTHX_ SV* invlist, UV start, UV end)
9643 /* Add the range from 'start' to 'end' inclusive to the inversion list's
9644 * set. A pointer to the inversion list is returned. This may actually be
9645 * a new list, in which case the passed in one has been destroyed. The
9646 * passed-in inversion list can be NULL, in which case a new one is created
9647 * with just the one range in it. The new list is not necessarily
9648 * NUL-terminated. Space is not freed if the inversion list shrinks as a
9649 * result of this function. The gain would not be large, and in many
9650 * cases, this is called multiple times on a single inversion list, so
9651 * anything freed may almost immediately be needed again.
9653 * This used to mostly call the 'union' routine, but that is much more
9654 * heavyweight than really needed for a single range addition */
9656 UV* array; /* The array implementing the inversion list */
9657 UV len; /* How many elements in 'array' */
9658 SSize_t i_s; /* index into the invlist array where 'start'
9660 SSize_t i_e = 0; /* And the index where 'end' should go */
9661 UV cur_highest; /* The highest code point in the inversion list
9662 upon entry to this function */
9664 /* This range becomes the whole inversion list if none already existed */
9665 if (invlist == NULL) {
9666 invlist = _new_invlist(2);
9667 _append_range_to_invlist(invlist, start, end);
9671 /* Likewise, if the inversion list is currently empty */
9672 len = _invlist_len(invlist);
9674 _append_range_to_invlist(invlist, start, end);
9678 /* Starting here, we have to know the internals of the list */
9679 array = invlist_array(invlist);
9681 /* If the new range ends higher than the current highest ... */
9682 cur_highest = invlist_highest(invlist);
9683 if (end > cur_highest) {
9685 /* If the whole range is higher, we can just append it */
9686 if (start > cur_highest) {
9687 _append_range_to_invlist(invlist, start, end);
9691 /* Otherwise, add the portion that is higher ... */
9692 _append_range_to_invlist(invlist, cur_highest + 1, end);
9694 /* ... and continue on below to handle the rest. As a result of the
9695 * above append, we know that the index of the end of the range is the
9696 * final even numbered one of the array. Recall that the final element
9697 * always starts a range that extends to infinity. If that range is in
9698 * the set (meaning the set goes from here to infinity), it will be an
9699 * even index, but if it isn't in the set, it's odd, and the final
9700 * range in the set is one less, which is even. */
9701 if (end == UV_MAX) {
9709 /* We have dealt with appending, now see about prepending. If the new
9710 * range starts lower than the current lowest ... */
9711 if (start < array[0]) {
9713 /* Adding something which has 0 in it is somewhat tricky, and uncommon.
9714 * Let the union code handle it, rather than having to know the
9715 * trickiness in two code places. */
9716 if (UNLIKELY(start == 0)) {
9719 range_invlist = _new_invlist(2);
9720 _append_range_to_invlist(range_invlist, start, end);
9722 _invlist_union(invlist, range_invlist, &invlist);
9724 SvREFCNT_dec_NN(range_invlist);
9729 /* If the whole new range comes before the first entry, and doesn't
9730 * extend it, we have to insert it as an additional range */
9731 if (end < array[0] - 1) {
9733 goto splice_in_new_range;
9736 /* Here the new range adjoins the existing first range, extending it
9740 /* And continue on below to handle the rest. We know that the index of
9741 * the beginning of the range is the first one of the array */
9744 else { /* Not prepending any part of the new range to the existing list.
9745 * Find where in the list it should go. This finds i_s, such that:
9746 * invlist[i_s] <= start < array[i_s+1]
9748 i_s = _invlist_search(invlist, start);
9751 /* At this point, any extending before the beginning of the inversion list
9752 * and/or after the end has been done. This has made it so that, in the
9753 * code below, each endpoint of the new range is either in a range that is
9754 * in the set, or is in a gap between two ranges that are. This means we
9755 * don't have to worry about exceeding the array bounds.
9757 * Find where in the list the new range ends (but we can skip this if we
9758 * have already determined what it is, or if it will be the same as i_s,
9759 * which we already have computed) */
9761 i_e = (start == end)
9763 : _invlist_search(invlist, end);
9766 /* Here generally invlist[i_e] <= end < array[i_e+1]. But if invlist[i_e]
9767 * is a range that goes to infinity there is no element at invlist[i_e+1],
9768 * so only the first relation holds. */
9770 if ( ! ELEMENT_RANGE_MATCHES_INVLIST(i_s)) {
9772 /* Here, the ranges on either side of the beginning of the new range
9773 * are in the set, and this range starts in the gap between them.
9775 * The new range extends the range above it downwards if the new range
9776 * ends at or above that range's start */
9777 const bool extends_the_range_above = ( end == UV_MAX
9778 || end + 1 >= array[i_s+1]);
9780 /* The new range extends the range below it upwards if it begins just
9781 * after where that range ends */
9782 if (start == array[i_s]) {
9784 /* If the new range fills the entire gap between the other ranges,
9785 * they will get merged together. Other ranges may also get
9786 * merged, depending on how many of them the new range spans. In
9787 * the general case, we do the merge later, just once, after we
9788 * figure out how many to merge. But in the case where the new
9789 * range exactly spans just this one gap (possibly extending into
9790 * the one above), we do the merge here, and an early exit. This
9791 * is done here to avoid having to special case later. */
9792 if (i_e - i_s <= 1) {
9794 /* If i_e - i_s == 1, it means that the new range terminates
9795 * within the range above, and hence 'extends_the_range_above'
9796 * must be true. (If the range above it extends to infinity,
9797 * 'i_s+2' will be above the array's limit, but 'len-i_s-2'
9798 * will be 0, so no harm done.) */
9799 if (extends_the_range_above) {
9800 Move(array + i_s + 2, array + i_s, len - i_s - 2, UV);
9801 invlist_set_len(invlist,
9803 *(get_invlist_offset_addr(invlist)));
9807 /* Here, i_e must == i_s. We keep them in sync, as they apply
9808 * to the same range, and below we are about to decrement i_s
9813 /* Here, the new range is adjacent to the one below. (It may also
9814 * span beyond the range above, but that will get resolved later.)
9815 * Extend the range below to include this one. */
9816 array[i_s] = (end == UV_MAX) ? UV_MAX : end + 1;
9820 else if (extends_the_range_above) {
9822 /* Here the new range only extends the range above it, but not the
9823 * one below. It merges with the one above. Again, we keep i_e
9824 * and i_s in sync if they point to the same range */
9833 /* Here, we've dealt with the new range start extending any adjoining
9836 * If the new range extends to infinity, it is now the final one,
9837 * regardless of what was there before */
9838 if (UNLIKELY(end == UV_MAX)) {
9839 invlist_set_len(invlist, i_s + 1, *(get_invlist_offset_addr(invlist)));
9843 /* If i_e started as == i_s, it has also been dealt with,
9844 * and been updated to the new i_s, which will fail the following if */
9845 if (! ELEMENT_RANGE_MATCHES_INVLIST(i_e)) {
9847 /* Here, the ranges on either side of the end of the new range are in
9848 * the set, and this range ends in the gap between them.
9850 * If this range is adjacent to (hence extends) the range above it, it
9851 * becomes part of that range; likewise if it extends the range below,
9852 * it becomes part of that range */
9853 if (end + 1 == array[i_e+1]) {
9857 else if (start <= array[i_e]) {
9858 array[i_e] = end + 1;
9865 /* If the range fits entirely in an existing range (as possibly already
9866 * extended above), it doesn't add anything new */
9867 if (ELEMENT_RANGE_MATCHES_INVLIST(i_s)) {
9871 /* Here, no part of the range is in the list. Must add it. It will
9872 * occupy 2 more slots */
9873 splice_in_new_range:
9875 invlist_extend(invlist, len + 2);
9876 array = invlist_array(invlist);
9877 /* Move the rest of the array down two slots. Don't include any
9879 Move(array + i_e + 1, array + i_e + 3, len - i_e - 1, UV);
9881 /* Do the actual splice */
9882 array[i_e+1] = start;
9883 array[i_e+2] = end + 1;
9884 invlist_set_len(invlist, len + 2, *(get_invlist_offset_addr(invlist)));
9888 /* Here the new range crossed the boundaries of a pre-existing range. The
9889 * code above has adjusted things so that both ends are in ranges that are
9890 * in the set. This means everything in between must also be in the set.
9891 * Just squash things together */
9892 Move(array + i_e + 1, array + i_s + 1, len - i_e - 1, UV);
9893 invlist_set_len(invlist,
9895 *(get_invlist_offset_addr(invlist)));
9901 Perl__setup_canned_invlist(pTHX_ const STRLEN size, const UV element0,
9902 UV** other_elements_ptr)
9904 /* Create and return an inversion list whose contents are to be populated
9905 * by the caller. The caller gives the number of elements (in 'size') and
9906 * the very first element ('element0'). This function will set
9907 * '*other_elements_ptr' to an array of UVs, where the remaining elements
9910 * Obviously there is some trust involved that the caller will properly
9911 * fill in the other elements of the array.
9913 * (The first element needs to be passed in, as the underlying code does
9914 * things differently depending on whether it is zero or non-zero) */
9916 SV* invlist = _new_invlist(size);
9919 PERL_ARGS_ASSERT__SETUP_CANNED_INVLIST;
9921 invlist = add_cp_to_invlist(invlist, element0);
9922 offset = *get_invlist_offset_addr(invlist);
9924 invlist_set_len(invlist, size, offset);
9925 *other_elements_ptr = invlist_array(invlist) + 1;
9931 PERL_STATIC_INLINE SV*
9932 S_add_cp_to_invlist(pTHX_ SV* invlist, const UV cp) {
9933 return _add_range_to_invlist(invlist, cp, cp);
9936 #ifndef PERL_IN_XSUB_RE
9938 Perl__invlist_invert(pTHX_ SV* const invlist)
9940 /* Complement the input inversion list. This adds a 0 if the list didn't
9941 * have a zero; removes it otherwise. As described above, the data
9942 * structure is set up so that this is very efficient */
9944 PERL_ARGS_ASSERT__INVLIST_INVERT;
9946 assert(! invlist_is_iterating(invlist));
9948 /* The inverse of matching nothing is matching everything */
9949 if (_invlist_len(invlist) == 0) {
9950 _append_range_to_invlist(invlist, 0, UV_MAX);
9954 *get_invlist_offset_addr(invlist) = ! *get_invlist_offset_addr(invlist);
9959 PERL_STATIC_INLINE SV*
9960 S_invlist_clone(pTHX_ SV* const invlist)
9963 /* Return a new inversion list that is a copy of the input one, which is
9964 * unchanged. The new list will not be mortal even if the old one was. */
9966 /* Need to allocate extra space to accommodate Perl's addition of a
9967 * trailing NUL to SvPV's, since it thinks they are always strings */
9968 SV* new_invlist = _new_invlist(_invlist_len(invlist) + 1);
9969 STRLEN physical_length = SvCUR(invlist);
9970 bool offset = *(get_invlist_offset_addr(invlist));
9972 PERL_ARGS_ASSERT_INVLIST_CLONE;
9974 *(get_invlist_offset_addr(new_invlist)) = offset;
9975 invlist_set_len(new_invlist, _invlist_len(invlist), offset);
9976 Copy(SvPVX(invlist), SvPVX(new_invlist), physical_length, char);
9981 PERL_STATIC_INLINE STRLEN*
9982 S_get_invlist_iter_addr(SV* invlist)
9984 /* Return the address of the UV that contains the current iteration
9987 PERL_ARGS_ASSERT_GET_INVLIST_ITER_ADDR;
9989 assert(SvTYPE(invlist) == SVt_INVLIST);
9991 return &(((XINVLIST*) SvANY(invlist))->iterator);
9994 PERL_STATIC_INLINE void
9995 S_invlist_iterinit(SV* invlist) /* Initialize iterator for invlist */
9997 PERL_ARGS_ASSERT_INVLIST_ITERINIT;
9999 *get_invlist_iter_addr(invlist) = 0;
10002 PERL_STATIC_INLINE void
10003 S_invlist_iterfinish(SV* invlist)
10005 /* Terminate iterator for invlist. This is to catch development errors.
10006 * Any iteration that is interrupted before completed should call this
10007 * function. Functions that add code points anywhere else but to the end
10008 * of an inversion list assert that they are not in the middle of an
10009 * iteration. If they were, the addition would make the iteration
10010 * problematical: if the iteration hadn't reached the place where things
10011 * were being added, it would be ok */
10013 PERL_ARGS_ASSERT_INVLIST_ITERFINISH;
10015 *get_invlist_iter_addr(invlist) = (STRLEN) UV_MAX;
10019 S_invlist_iternext(SV* invlist, UV* start, UV* end)
10021 /* An C<invlist_iterinit> call on <invlist> must be used to set this up.
10022 * This call sets in <*start> and <*end>, the next range in <invlist>.
10023 * Returns <TRUE> if successful and the next call will return the next
10024 * range; <FALSE> if was already at the end of the list. If the latter,
10025 * <*start> and <*end> are unchanged, and the next call to this function
10026 * will start over at the beginning of the list */
10028 STRLEN* pos = get_invlist_iter_addr(invlist);
10029 UV len = _invlist_len(invlist);
10032 PERL_ARGS_ASSERT_INVLIST_ITERNEXT;
10035 *pos = (STRLEN) UV_MAX; /* Force iterinit() to be required next time */
10039 array = invlist_array(invlist);
10041 *start = array[(*pos)++];
10047 *end = array[(*pos)++] - 1;
10053 PERL_STATIC_INLINE UV
10054 S_invlist_highest(SV* const invlist)
10056 /* Returns the highest code point that matches an inversion list. This API
10057 * has an ambiguity, as it returns 0 under either the highest is actually
10058 * 0, or if the list is empty. If this distinction matters to you, check
10059 * for emptiness before calling this function */
10061 UV len = _invlist_len(invlist);
10064 PERL_ARGS_ASSERT_INVLIST_HIGHEST;
10070 array = invlist_array(invlist);
10072 /* The last element in the array in the inversion list always starts a
10073 * range that goes to infinity. That range may be for code points that are
10074 * matched in the inversion list, or it may be for ones that aren't
10075 * matched. In the latter case, the highest code point in the set is one
10076 * less than the beginning of this range; otherwise it is the final element
10077 * of this range: infinity */
10078 return (ELEMENT_RANGE_MATCHES_INVLIST(len - 1))
10080 : array[len - 1] - 1;
10084 S_invlist_contents(pTHX_ SV* const invlist, const bool traditional_style)
10086 /* Get the contents of an inversion list into a string SV so that they can
10087 * be printed out. If 'traditional_style' is TRUE, it uses the format
10088 * traditionally done for debug tracing; otherwise it uses a format
10089 * suitable for just copying to the output, with blanks between ranges and
10090 * a dash between range components */
10094 const char intra_range_delimiter = (traditional_style ? '\t' : '-');
10095 const char inter_range_delimiter = (traditional_style ? '\n' : ' ');
10097 if (traditional_style) {
10098 output = newSVpvs("\n");
10101 output = newSVpvs("");
10104 PERL_ARGS_ASSERT_INVLIST_CONTENTS;
10106 assert(! invlist_is_iterating(invlist));
10108 invlist_iterinit(invlist);
10109 while (invlist_iternext(invlist, &start, &end)) {
10110 if (end == UV_MAX) {
10111 Perl_sv_catpvf(aTHX_ output, "%04" UVXf "%cINFINITY%c",
10112 start, intra_range_delimiter,
10113 inter_range_delimiter);
10115 else if (end != start) {
10116 Perl_sv_catpvf(aTHX_ output, "%04" UVXf "%c%04" UVXf "%c",
10118 intra_range_delimiter,
10119 end, inter_range_delimiter);
10122 Perl_sv_catpvf(aTHX_ output, "%04" UVXf "%c",
10123 start, inter_range_delimiter);
10127 if (SvCUR(output) && ! traditional_style) {/* Get rid of trailing blank */
10128 SvCUR_set(output, SvCUR(output) - 1);
10134 #ifndef PERL_IN_XSUB_RE
10136 Perl__invlist_dump(pTHX_ PerlIO *file, I32 level,
10137 const char * const indent, SV* const invlist)
10139 /* Designed to be called only by do_sv_dump(). Dumps out the ranges of the
10140 * inversion list 'invlist' to 'file' at 'level' Each line is prefixed by
10141 * the string 'indent'. The output looks like this:
10142 [0] 0x000A .. 0x000D
10144 [4] 0x2028 .. 0x2029
10145 [6] 0x3104 .. INFINITY
10146 * This means that the first range of code points matched by the list are
10147 * 0xA through 0xD; the second range contains only the single code point
10148 * 0x85, etc. An inversion list is an array of UVs. Two array elements
10149 * are used to define each range (except if the final range extends to
10150 * infinity, only a single element is needed). The array index of the
10151 * first element for the corresponding range is given in brackets. */
10156 PERL_ARGS_ASSERT__INVLIST_DUMP;
10158 if (invlist_is_iterating(invlist)) {
10159 Perl_dump_indent(aTHX_ level, file,
10160 "%sCan't dump inversion list because is in middle of iterating\n",
10165 invlist_iterinit(invlist);
10166 while (invlist_iternext(invlist, &start, &end)) {
10167 if (end == UV_MAX) {
10168 Perl_dump_indent(aTHX_ level, file,
10169 "%s[%" UVuf "] 0x%04" UVXf " .. INFINITY\n",
10170 indent, (UV)count, start);
10172 else if (end != start) {
10173 Perl_dump_indent(aTHX_ level, file,
10174 "%s[%" UVuf "] 0x%04" UVXf " .. 0x%04" UVXf "\n",
10175 indent, (UV)count, start, end);
10178 Perl_dump_indent(aTHX_ level, file, "%s[%" UVuf "] 0x%04" UVXf "\n",
10179 indent, (UV)count, start);
10186 Perl__load_PL_utf8_foldclosures (pTHX)
10188 assert(! PL_utf8_foldclosures);
10190 /* If the folds haven't been read in, call a fold function
10192 if (! PL_utf8_tofold) {
10193 U8 dummy[UTF8_MAXBYTES_CASE+1];
10194 const U8 hyphen[] = HYPHEN_UTF8;
10196 /* This string is just a short named one above \xff */
10197 toFOLD_utf8_safe(hyphen, hyphen + sizeof(hyphen) - 1, dummy, NULL);
10198 assert(PL_utf8_tofold); /* Verify that worked */
10200 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
10204 #if defined(PERL_ARGS_ASSERT__INVLISTEQ) && !defined(PERL_IN_XSUB_RE)
10206 Perl__invlistEQ(pTHX_ SV* const a, SV* const b, const bool complement_b)
10208 /* Return a boolean as to if the two passed in inversion lists are
10209 * identical. The final argument, if TRUE, says to take the complement of
10210 * the second inversion list before doing the comparison */
10212 const UV* array_a = invlist_array(a);
10213 const UV* array_b = invlist_array(b);
10214 UV len_a = _invlist_len(a);
10215 UV len_b = _invlist_len(b);
10217 PERL_ARGS_ASSERT__INVLISTEQ;
10219 /* If are to compare 'a' with the complement of b, set it
10220 * up so are looking at b's complement. */
10221 if (complement_b) {
10223 /* The complement of nothing is everything, so <a> would have to have
10224 * just one element, starting at zero (ending at infinity) */
10226 return (len_a == 1 && array_a[0] == 0);
10228 else if (array_b[0] == 0) {
10230 /* Otherwise, to complement, we invert. Here, the first element is
10231 * 0, just remove it. To do this, we just pretend the array starts
10239 /* But if the first element is not zero, we pretend the list starts
10240 * at the 0 that is always stored immediately before the array. */
10246 return len_a == len_b
10247 && memEQ(array_a, array_b, len_a * sizeof(array_a[0]));
10253 * As best we can, determine the characters that can match the start of
10254 * the given EXACTF-ish node.
10256 * Returns the invlist as a new SV*; it is the caller's responsibility to
10257 * call SvREFCNT_dec() when done with it.
10260 S__make_exactf_invlist(pTHX_ RExC_state_t *pRExC_state, regnode *node)
10262 const U8 * s = (U8*)STRING(node);
10263 SSize_t bytelen = STR_LEN(node);
10265 /* Start out big enough for 2 separate code points */
10266 SV* invlist = _new_invlist(4);
10268 PERL_ARGS_ASSERT__MAKE_EXACTF_INVLIST;
10273 /* We punt and assume can match anything if the node begins
10274 * with a multi-character fold. Things are complicated. For
10275 * example, /ffi/i could match any of:
10276 * "\N{LATIN SMALL LIGATURE FFI}"
10277 * "\N{LATIN SMALL LIGATURE FF}I"
10278 * "F\N{LATIN SMALL LIGATURE FI}"
10279 * plus several other things; and making sure we have all the
10280 * possibilities is hard. */
10281 if (is_MULTI_CHAR_FOLD_latin1_safe(s, s + bytelen)) {
10282 invlist = _add_range_to_invlist(invlist, 0, UV_MAX);
10285 /* Any Latin1 range character can potentially match any
10286 * other depending on the locale */
10287 if (OP(node) == EXACTFL) {
10288 _invlist_union(invlist, PL_Latin1, &invlist);
10291 /* But otherwise, it matches at least itself. We can
10292 * quickly tell if it has a distinct fold, and if so,
10293 * it matches that as well */
10294 invlist = add_cp_to_invlist(invlist, uc);
10295 if (IS_IN_SOME_FOLD_L1(uc))
10296 invlist = add_cp_to_invlist(invlist, PL_fold_latin1[uc]);
10299 /* Some characters match above-Latin1 ones under /i. This
10300 * is true of EXACTFL ones when the locale is UTF-8 */
10301 if (HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(uc)
10302 && (! isASCII(uc) || (OP(node) != EXACTFAA
10303 && OP(node) != EXACTFAA_NO_TRIE)))
10305 add_above_Latin1_folds(pRExC_state, (U8) uc, &invlist);
10309 else { /* Pattern is UTF-8 */
10310 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
10311 STRLEN foldlen = UTF8SKIP(s);
10312 const U8* e = s + bytelen;
10315 uc = utf8_to_uvchr_buf(s, s + bytelen, NULL);
10317 /* The only code points that aren't folded in a UTF EXACTFish
10318 * node are are the problematic ones in EXACTFL nodes */
10319 if (OP(node) == EXACTFL && is_PROBLEMATIC_LOCALE_FOLDEDS_START_cp(uc)) {
10320 /* We need to check for the possibility that this EXACTFL
10321 * node begins with a multi-char fold. Therefore we fold
10322 * the first few characters of it so that we can make that
10327 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < e; i++) {
10329 *(d++) = (U8) toFOLD(*s);
10334 toFOLD_utf8_safe(s, e, d, &len);
10340 /* And set up so the code below that looks in this folded
10341 * buffer instead of the node's string */
10343 foldlen = UTF8SKIP(folded);
10347 /* When we reach here 's' points to the fold of the first
10348 * character(s) of the node; and 'e' points to far enough along
10349 * the folded string to be just past any possible multi-char
10350 * fold. 'foldlen' is the length in bytes of the first
10353 * Unlike the non-UTF-8 case, the macro for determining if a
10354 * string is a multi-char fold requires all the characters to
10355 * already be folded. This is because of all the complications
10356 * if not. Note that they are folded anyway, except in EXACTFL
10357 * nodes. Like the non-UTF case above, we punt if the node
10358 * begins with a multi-char fold */
10360 if (is_MULTI_CHAR_FOLD_utf8_safe(s, e)) {
10361 invlist = _add_range_to_invlist(invlist, 0, UV_MAX);
10363 else { /* Single char fold */
10365 /* It matches all the things that fold to it, which are
10366 * found in PL_utf8_foldclosures (including itself) */
10367 invlist = add_cp_to_invlist(invlist, uc);
10368 if (! PL_utf8_foldclosures)
10369 _load_PL_utf8_foldclosures();
10370 if ((listp = hv_fetch(PL_utf8_foldclosures,
10371 (char *) s, foldlen, FALSE)))
10373 AV* list = (AV*) *listp;
10375 for (k = 0; k <= av_tindex_skip_len_mg(list); k++) {
10376 SV** c_p = av_fetch(list, k, FALSE);
10382 /* /aa doesn't allow folds between ASCII and non- */
10383 if ((OP(node) == EXACTFAA || OP(node) == EXACTFAA_NO_TRIE)
10384 && isASCII(c) != isASCII(uc))
10389 invlist = add_cp_to_invlist(invlist, c);
10398 #undef HEADER_LENGTH
10399 #undef TO_INTERNAL_SIZE
10400 #undef FROM_INTERNAL_SIZE
10401 #undef INVLIST_VERSION_ID
10403 /* End of inversion list object */
10406 S_parse_lparen_question_flags(pTHX_ RExC_state_t *pRExC_state)
10408 /* This parses the flags that are in either the '(?foo)' or '(?foo:bar)'
10409 * constructs, and updates RExC_flags with them. On input, RExC_parse
10410 * should point to the first flag; it is updated on output to point to the
10411 * final ')' or ':'. There needs to be at least one flag, or this will
10414 /* for (?g), (?gc), and (?o) warnings; warning
10415 about (?c) will warn about (?g) -- japhy */
10417 #define WASTED_O 0x01
10418 #define WASTED_G 0x02
10419 #define WASTED_C 0x04
10420 #define WASTED_GC (WASTED_G|WASTED_C)
10421 I32 wastedflags = 0x00;
10422 U32 posflags = 0, negflags = 0;
10423 U32 *flagsp = &posflags;
10424 char has_charset_modifier = '\0';
10426 bool has_use_defaults = FALSE;
10427 const char* const seqstart = RExC_parse - 1; /* Point to the '?' */
10428 int x_mod_count = 0;
10430 PERL_ARGS_ASSERT_PARSE_LPAREN_QUESTION_FLAGS;
10432 /* '^' as an initial flag sets certain defaults */
10433 if (UCHARAT(RExC_parse) == '^') {
10435 has_use_defaults = TRUE;
10436 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
10437 set_regex_charset(&RExC_flags, (RExC_utf8 || RExC_uni_semantics)
10438 ? REGEX_UNICODE_CHARSET
10439 : REGEX_DEPENDS_CHARSET);
10442 cs = get_regex_charset(RExC_flags);
10443 if (cs == REGEX_DEPENDS_CHARSET
10444 && (RExC_utf8 || RExC_uni_semantics))
10446 cs = REGEX_UNICODE_CHARSET;
10449 while (RExC_parse < RExC_end) {
10450 /* && strchr("iogcmsx", *RExC_parse) */
10451 /* (?g), (?gc) and (?o) are useless here
10452 and must be globally applied -- japhy */
10453 switch (*RExC_parse) {
10455 /* Code for the imsxn flags */
10456 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp, x_mod_count);
10458 case LOCALE_PAT_MOD:
10459 if (has_charset_modifier) {
10460 goto excess_modifier;
10462 else if (flagsp == &negflags) {
10465 cs = REGEX_LOCALE_CHARSET;
10466 has_charset_modifier = LOCALE_PAT_MOD;
10468 case UNICODE_PAT_MOD:
10469 if (has_charset_modifier) {
10470 goto excess_modifier;
10472 else if (flagsp == &negflags) {
10475 cs = REGEX_UNICODE_CHARSET;
10476 has_charset_modifier = UNICODE_PAT_MOD;
10478 case ASCII_RESTRICT_PAT_MOD:
10479 if (flagsp == &negflags) {
10482 if (has_charset_modifier) {
10483 if (cs != REGEX_ASCII_RESTRICTED_CHARSET) {
10484 goto excess_modifier;
10486 /* Doubled modifier implies more restricted */
10487 cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET;
10490 cs = REGEX_ASCII_RESTRICTED_CHARSET;
10492 has_charset_modifier = ASCII_RESTRICT_PAT_MOD;
10494 case DEPENDS_PAT_MOD:
10495 if (has_use_defaults) {
10496 goto fail_modifiers;
10498 else if (flagsp == &negflags) {
10501 else if (has_charset_modifier) {
10502 goto excess_modifier;
10505 /* The dual charset means unicode semantics if the
10506 * pattern (or target, not known until runtime) are
10507 * utf8, or something in the pattern indicates unicode
10509 cs = (RExC_utf8 || RExC_uni_semantics)
10510 ? REGEX_UNICODE_CHARSET
10511 : REGEX_DEPENDS_CHARSET;
10512 has_charset_modifier = DEPENDS_PAT_MOD;
10516 if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) {
10517 vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD);
10519 else if (has_charset_modifier == *(RExC_parse - 1)) {
10520 vFAIL2("Regexp modifier \"%c\" may not appear twice",
10521 *(RExC_parse - 1));
10524 vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1));
10526 NOT_REACHED; /*NOTREACHED*/
10529 vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"",
10530 *(RExC_parse - 1));
10531 NOT_REACHED; /*NOTREACHED*/
10532 case ONCE_PAT_MOD: /* 'o' */
10533 case GLOBAL_PAT_MOD: /* 'g' */
10534 if (PASS2 && ckWARN(WARN_REGEXP)) {
10535 const I32 wflagbit = *RExC_parse == 'o'
10538 if (! (wastedflags & wflagbit) ) {
10539 wastedflags |= wflagbit;
10540 /* diag_listed_as: Useless (?-%s) - don't use /%s modifier in regex; marked by <-- HERE in m/%s/ */
10543 "Useless (%s%c) - %suse /%c modifier",
10544 flagsp == &negflags ? "?-" : "?",
10546 flagsp == &negflags ? "don't " : "",
10553 case CONTINUE_PAT_MOD: /* 'c' */
10554 if (PASS2 && ckWARN(WARN_REGEXP)) {
10555 if (! (wastedflags & WASTED_C) ) {
10556 wastedflags |= WASTED_GC;
10557 /* diag_listed_as: Useless (?-%s) - don't use /%s modifier in regex; marked by <-- HERE in m/%s/ */
10560 "Useless (%sc) - %suse /gc modifier",
10561 flagsp == &negflags ? "?-" : "?",
10562 flagsp == &negflags ? "don't " : ""
10567 case KEEPCOPY_PAT_MOD: /* 'p' */
10568 if (flagsp == &negflags) {
10570 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
10572 *flagsp |= RXf_PMf_KEEPCOPY;
10576 /* A flag is a default iff it is following a minus, so
10577 * if there is a minus, it means will be trying to
10578 * re-specify a default which is an error */
10579 if (has_use_defaults || flagsp == &negflags) {
10580 goto fail_modifiers;
10582 flagsp = &negflags;
10583 wastedflags = 0; /* reset so (?g-c) warns twice */
10589 if ((posflags & (RXf_PMf_EXTENDED|RXf_PMf_EXTENDED_MORE)) == RXf_PMf_EXTENDED) {
10590 negflags |= RXf_PMf_EXTENDED_MORE;
10592 RExC_flags |= posflags;
10594 if (negflags & RXf_PMf_EXTENDED) {
10595 negflags |= RXf_PMf_EXTENDED_MORE;
10597 RExC_flags &= ~negflags;
10598 set_regex_charset(&RExC_flags, cs);
10603 RExC_parse += SKIP_IF_CHAR(RExC_parse);
10604 /* diag_listed_as: Sequence (?%s...) not recognized in regex; marked by <-- HERE in m/%s/ */
10605 vFAIL2utf8f("Sequence (%" UTF8f "...) not recognized",
10606 UTF8fARG(UTF, RExC_parse-seqstart, seqstart));
10607 NOT_REACHED; /*NOTREACHED*/
10610 RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
10613 vFAIL("Sequence (?... not terminated");
10617 - reg - regular expression, i.e. main body or parenthesized thing
10619 * Caller must absorb opening parenthesis.
10621 * Combining parenthesis handling with the base level of regular expression
10622 * is a trifle forced, but the need to tie the tails of the branches to what
10623 * follows makes it hard to avoid.
10625 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
10627 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
10629 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
10632 PERL_STATIC_INLINE regnode *
10633 S_handle_named_backref(pTHX_ RExC_state_t *pRExC_state,
10635 char * parse_start,
10640 char* name_start = RExC_parse;
10642 SV *sv_dat = reg_scan_name(pRExC_state, SIZE_ONLY
10643 ? REG_RSN_RETURN_NULL
10644 : REG_RSN_RETURN_DATA);
10645 GET_RE_DEBUG_FLAGS_DECL;
10647 PERL_ARGS_ASSERT_HANDLE_NAMED_BACKREF;
10649 if (RExC_parse == name_start || *RExC_parse != ch) {
10650 /* diag_listed_as: Sequence \%s... not terminated in regex; marked by <-- HERE in m/%s/ */
10651 vFAIL2("Sequence %.3s... not terminated",parse_start);
10655 num = add_data( pRExC_state, STR_WITH_LEN("S"));
10656 RExC_rxi->data->data[num]=(void*)sv_dat;
10657 SvREFCNT_inc_simple_void(sv_dat);
10660 ret = reganode(pRExC_state,
10663 : (ASCII_FOLD_RESTRICTED)
10665 : (AT_LEAST_UNI_SEMANTICS)
10671 *flagp |= HASWIDTH;
10673 Set_Node_Offset(ret, parse_start+1);
10674 Set_Node_Cur_Length(ret, parse_start);
10676 nextchar(pRExC_state);
10680 /* Returns NULL, setting *flagp to TRYAGAIN at the end of (?) that only sets
10681 flags. Returns NULL, setting *flagp to RESTART_PASS1 if the sizing scan
10682 needs to be restarted, or'd with NEED_UTF8 if the pattern needs to be
10683 upgraded to UTF-8. Otherwise would only return NULL if regbranch() returns
10684 NULL, which cannot happen. */
10686 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
10687 /* paren: Parenthesized? 0=top; 1,2=inside '(': changed to letter.
10688 * 2 is like 1, but indicates that nextchar() has been called to advance
10689 * RExC_parse beyond the '('. Things like '(?' are indivisible tokens, and
10690 * this flag alerts us to the need to check for that */
10692 regnode *ret = NULL; /* Will be the head of the group. */
10695 regnode *ender = NULL;
10698 U32 oregflags = RExC_flags;
10699 bool have_branch = 0;
10701 I32 freeze_paren = 0;
10702 I32 after_freeze = 0;
10703 I32 num; /* numeric backreferences */
10705 char * parse_start = RExC_parse; /* MJD */
10706 char * const oregcomp_parse = RExC_parse;
10708 GET_RE_DEBUG_FLAGS_DECL;
10710 PERL_ARGS_ASSERT_REG;
10711 DEBUG_PARSE("reg ");
10713 *flagp = 0; /* Tentatively. */
10715 /* Having this true makes it feasible to have a lot fewer tests for the
10716 * parse pointer being in scope. For example, we can write
10717 * while(isFOO(*RExC_parse)) RExC_parse++;
10719 * while(RExC_parse < RExC_end && isFOO(*RExC_parse)) RExC_parse++;
10721 assert(*RExC_end == '\0');
10723 /* Make an OPEN node, if parenthesized. */
10726 /* Under /x, space and comments can be gobbled up between the '(' and
10727 * here (if paren ==2). The forms '(*VERB' and '(?...' disallow such
10728 * intervening space, as the sequence is a token, and a token should be
10730 bool has_intervening_patws = (paren == 2)
10731 && *(RExC_parse - 1) != '(';
10733 if (RExC_parse >= RExC_end) {
10734 vFAIL("Unmatched (");
10737 if (paren == 'r') { /* Atomic script run */
10741 else if ( *RExC_parse == '*') { /* (*VERB:ARG), (*construct:...) */
10742 char *start_verb = RExC_parse + 1;
10744 char *start_arg = NULL;
10745 unsigned char op = 0;
10746 int arg_required = 0;
10747 int internal_argval = -1; /* if >-1 we are not allowed an argument*/
10748 bool has_upper = FALSE;
10750 if (has_intervening_patws) {
10751 RExC_parse++; /* past the '*' */
10753 /* For strict backwards compatibility, don't change the message
10754 * now that we also have lowercase operands */
10755 if (isUPPER(*RExC_parse)) {
10756 vFAIL("In '(*VERB...)', the '(' and '*' must be adjacent");
10759 vFAIL("In '(*...)', the '(' and '*' must be adjacent");
10762 while (RExC_parse < RExC_end && *RExC_parse != ')' ) {
10763 if ( *RExC_parse == ':' ) {
10764 start_arg = RExC_parse + 1;
10768 if (isUPPER(*RExC_parse)) {
10774 RExC_parse += UTF8SKIP(RExC_parse);
10777 verb_len = RExC_parse - start_verb;
10779 if (RExC_parse >= RExC_end) {
10780 goto unterminated_verb_pattern;
10783 RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
10784 while ( RExC_parse < RExC_end && *RExC_parse != ')' ) {
10785 RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
10787 if ( RExC_parse >= RExC_end || *RExC_parse != ')' ) {
10788 unterminated_verb_pattern:
10790 vFAIL("Unterminated verb pattern argument");
10793 vFAIL("Unterminated '(*...' argument");
10797 if ( RExC_parse >= RExC_end || *RExC_parse != ')' ) {
10799 vFAIL("Unterminated verb pattern");
10802 vFAIL("Unterminated '(*...' construct");
10807 /* Here, we know that RExC_parse < RExC_end */
10809 switch ( *start_verb ) {
10810 case 'A': /* (*ACCEPT) */
10811 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
10813 internal_argval = RExC_nestroot;
10816 case 'C': /* (*COMMIT) */
10817 if ( memEQs(start_verb,verb_len,"COMMIT") )
10820 case 'F': /* (*FAIL) */
10821 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
10825 case ':': /* (*:NAME) */
10826 case 'M': /* (*MARK:NAME) */
10827 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
10832 case 'P': /* (*PRUNE) */
10833 if ( memEQs(start_verb,verb_len,"PRUNE") )
10836 case 'S': /* (*SKIP) */
10837 if ( memEQs(start_verb,verb_len,"SKIP") )
10840 case 'T': /* (*THEN) */
10841 /* [19:06] <TimToady> :: is then */
10842 if ( memEQs(start_verb,verb_len,"THEN") ) {
10844 RExC_seen |= REG_CUTGROUP_SEEN;
10848 if ( memEQs(start_verb, verb_len, "asr")
10849 || memEQs(start_verb, verb_len, "atomic_script_run"))
10851 paren = 'r'; /* Mnemonic: recursed run */
10854 else if (memEQs(start_verb, verb_len, "atomic")) {
10855 paren = 't'; /* AtOMIC */
10856 goto alpha_assertions;
10860 if ( memEQs(start_verb, verb_len, "plb")
10861 || memEQs(start_verb, verb_len, "positive_lookbehind"))
10864 goto lookbehind_alpha_assertions;
10866 else if ( memEQs(start_verb, verb_len, "pla")
10867 || memEQs(start_verb, verb_len, "positive_lookahead"))
10870 goto alpha_assertions;
10874 if ( memEQs(start_verb, verb_len, "nlb")
10875 || memEQs(start_verb, verb_len, "negative_lookbehind"))
10878 goto lookbehind_alpha_assertions;
10880 else if ( memEQs(start_verb, verb_len, "nla")
10881 || memEQs(start_verb, verb_len, "negative_lookahead"))
10884 goto alpha_assertions;
10888 if ( memEQs(start_verb, verb_len, "sr")
10889 || memEQs(start_verb, verb_len, "script_run"))
10897 /* This indicates Unicode rules. */
10898 REQUIRE_UNI_RULES(flagp, NULL);
10904 RExC_parse = start_arg;
10906 if (RExC_in_script_run) {
10908 /* Nested script runs are treated as no-ops, because
10909 * if the nested one fails, the outer one must as
10910 * well. It could fail sooner, and avoid (??{} with
10911 * side effects, but that is explicitly documented as
10912 * undefined behavior. */
10916 if (paren == 's') {
10921 /* But, the atomic part of a nested atomic script run
10922 * isn't a no-op, but can be treated just like a '(?>'
10928 /* By doing this here, we avoid extra warnings for nested
10931 Perl_ck_warner_d(aTHX_
10932 packWARN(WARN_EXPERIMENTAL__SCRIPT_RUN),
10933 "The script_run feature is experimental"
10934 REPORT_LOCATION, REPORT_LOCATION_ARGS(RExC_parse));
10938 if (paren == 's') {
10939 /* Here, we're starting a new regular script run */
10940 ret = reg_node(pRExC_state, SROPEN);
10941 RExC_in_script_run = 1;
10946 /* Here, we are starting an atomic script run. This is
10947 * handled by recursing to deal with the atomic portion
10948 * separately, enclosed in SROPEN ... SRCLOSE nodes */
10950 ret = reg_node(pRExC_state, SROPEN);
10952 RExC_in_script_run = 1;
10954 atomic = reg(pRExC_state, 'r', &flags, depth);
10955 if (flags & (RESTART_PASS1|NEED_UTF8)) {
10956 *flagp = flags & (RESTART_PASS1|NEED_UTF8);
10960 REGTAIL(pRExC_state, ret, atomic);
10962 REGTAIL(pRExC_state, atomic,
10963 reg_node(pRExC_state, SRCLOSE));
10965 RExC_in_script_run = 0;
10971 lookbehind_alpha_assertions:
10972 RExC_seen |= REG_LOOKBEHIND_SEEN;
10973 RExC_in_lookbehind++;
10979 Perl_ck_warner_d(aTHX_
10980 packWARN(WARN_EXPERIMENTAL__ALPHA_ASSERTIONS),
10981 "The alpha_assertions feature is experimental"
10982 REPORT_LOCATION, REPORT_LOCATION_ARGS(RExC_parse));
10985 RExC_seen_zerolen++;
10991 /* An empty negative lookahead assertion simply is failure */
10992 if (paren == 'A' && RExC_parse == start_arg) {
10993 ret=reganode(pRExC_state, OPFAIL, 0);
10994 nextchar(pRExC_state);
10998 RExC_parse = start_arg;
11003 "'(*%" UTF8f "' requires a terminating ':'",
11004 UTF8fARG(UTF, verb_len, start_verb));
11005 NOT_REACHED; /*NOTREACHED*/
11007 } /* End of switch */
11009 RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
11010 if (has_upper || verb_len == 0) {
11012 "Unknown verb pattern '%" UTF8f "'",
11013 UTF8fARG(UTF, verb_len, start_verb));
11017 "Unknown '(*...)' construct '%" UTF8f "'",
11018 UTF8fARG(UTF, verb_len, start_verb));
11021 if ( RExC_parse == start_arg ) {
11024 if ( arg_required && !start_arg ) {
11025 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
11026 verb_len, start_verb);
11028 if (internal_argval == -1) {
11029 ret = reganode(pRExC_state, op, 0);
11031 ret = reg2Lanode(pRExC_state, op, 0, internal_argval);
11033 RExC_seen |= REG_VERBARG_SEEN;
11034 if ( ! SIZE_ONLY ) {
11036 SV *sv = newSVpvn( start_arg,
11037 RExC_parse - start_arg);
11038 ARG(ret) = add_data( pRExC_state,
11039 STR_WITH_LEN("S"));
11040 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
11045 if ( internal_argval != -1 )
11046 ARG2L_SET(ret, internal_argval);
11048 nextchar(pRExC_state);
11051 else if (*RExC_parse == '?') { /* (?...) */
11052 bool is_logical = 0;
11053 const char * const seqstart = RExC_parse;
11054 const char * endptr;
11055 if (has_intervening_patws) {
11057 vFAIL("In '(?...)', the '(' and '?' must be adjacent");
11060 RExC_parse++; /* past the '?' */
11061 paren = *RExC_parse; /* might be a trailing NUL, if not
11063 RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
11064 if (RExC_parse > RExC_end) {
11067 ret = NULL; /* For look-ahead/behind. */
11070 case 'P': /* (?P...) variants for those used to PCRE/Python */
11071 paren = *RExC_parse;
11072 if ( paren == '<') { /* (?P<...>) named capture */
11074 if (RExC_parse >= RExC_end) {
11075 vFAIL("Sequence (?P<... not terminated");
11077 goto named_capture;
11079 else if (paren == '>') { /* (?P>name) named recursion */
11081 if (RExC_parse >= RExC_end) {
11082 vFAIL("Sequence (?P>... not terminated");
11084 goto named_recursion;
11086 else if (paren == '=') { /* (?P=...) named backref */
11088 return handle_named_backref(pRExC_state, flagp,
11091 RExC_parse += SKIP_IF_CHAR(RExC_parse);
11092 /* diag_listed_as: Sequence (?%s...) not recognized in regex; marked by <-- HERE in m/%s/ */
11093 vFAIL3("Sequence (%.*s...) not recognized",
11094 RExC_parse-seqstart, seqstart);
11095 NOT_REACHED; /*NOTREACHED*/
11096 case '<': /* (?<...) */
11097 if (*RExC_parse == '!')
11099 else if (*RExC_parse != '=')
11106 case '\'': /* (?'...') */
11107 name_start = RExC_parse;
11108 svname = reg_scan_name(pRExC_state,
11109 SIZE_ONLY /* reverse test from the others */
11110 ? REG_RSN_RETURN_NAME
11111 : REG_RSN_RETURN_NULL);
11112 if ( RExC_parse == name_start
11113 || RExC_parse >= RExC_end
11114 || *RExC_parse != paren)
11116 vFAIL2("Sequence (?%c... not terminated",
11117 paren=='>' ? '<' : paren);
11122 if (!svname) /* shouldn't happen */
11124 "panic: reg_scan_name returned NULL");
11125 if (!RExC_paren_names) {
11126 RExC_paren_names= newHV();
11127 sv_2mortal(MUTABLE_SV(RExC_paren_names));
11129 RExC_paren_name_list= newAV();
11130 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
11133 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
11135 sv_dat = HeVAL(he_str);
11137 /* croak baby croak */
11139 "panic: paren_name hash element allocation failed");
11140 } else if ( SvPOK(sv_dat) ) {
11141 /* (?|...) can mean we have dupes so scan to check
11142 its already been stored. Maybe a flag indicating
11143 we are inside such a construct would be useful,
11144 but the arrays are likely to be quite small, so
11145 for now we punt -- dmq */
11146 IV count = SvIV(sv_dat);
11147 I32 *pv = (I32*)SvPVX(sv_dat);
11149 for ( i = 0 ; i < count ; i++ ) {
11150 if ( pv[i] == RExC_npar ) {
11156 pv = (I32*)SvGROW(sv_dat,
11157 SvCUR(sv_dat) + sizeof(I32)+1);
11158 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
11159 pv[count] = RExC_npar;
11160 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
11163 (void)SvUPGRADE(sv_dat,SVt_PVNV);
11164 sv_setpvn(sv_dat, (char *)&(RExC_npar),
11167 SvIV_set(sv_dat, 1);
11170 /* Yes this does cause a memory leak in debugging Perls
11172 if (!av_store(RExC_paren_name_list,
11173 RExC_npar, SvREFCNT_inc(svname)))
11174 SvREFCNT_dec_NN(svname);
11177 /*sv_dump(sv_dat);*/
11179 nextchar(pRExC_state);
11181 goto capturing_parens;
11184 RExC_seen |= REG_LOOKBEHIND_SEEN;
11185 RExC_in_lookbehind++;
11187 if (RExC_parse >= RExC_end) {
11188 vFAIL("Sequence (?... not terminated");
11192 case '=': /* (?=...) */
11193 RExC_seen_zerolen++;
11195 case '!': /* (?!...) */
11196 RExC_seen_zerolen++;
11197 /* check if we're really just a "FAIL" assertion */
11198 skip_to_be_ignored_text(pRExC_state, &RExC_parse,
11199 FALSE /* Don't force to /x */ );
11200 if (*RExC_parse == ')') {
11201 ret=reganode(pRExC_state, OPFAIL, 0);
11202 nextchar(pRExC_state);
11206 case '|': /* (?|...) */
11207 /* branch reset, behave like a (?:...) except that
11208 buffers in alternations share the same numbers */
11210 after_freeze = freeze_paren = RExC_npar;
11212 case ':': /* (?:...) */
11213 case '>': /* (?>...) */
11215 case '$': /* (?$...) */
11216 case '@': /* (?@...) */
11217 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
11219 case '0' : /* (?0) */
11220 case 'R' : /* (?R) */
11221 if (RExC_parse == RExC_end || *RExC_parse != ')')
11222 FAIL("Sequence (?R) not terminated");
11224 RExC_seen |= REG_RECURSE_SEEN;
11225 *flagp |= POSTPONED;
11226 goto gen_recurse_regop;
11228 /* named and numeric backreferences */
11229 case '&': /* (?&NAME) */
11230 parse_start = RExC_parse - 1;
11233 SV *sv_dat = reg_scan_name(pRExC_state,
11234 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
11235 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
11237 if (RExC_parse >= RExC_end || *RExC_parse != ')')
11238 vFAIL("Sequence (?&... not terminated");
11239 goto gen_recurse_regop;
11242 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
11244 vFAIL("Illegal pattern");
11246 goto parse_recursion;
11248 case '-': /* (?-1) */
11249 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
11250 RExC_parse--; /* rewind to let it be handled later */
11254 case '1': case '2': case '3': case '4': /* (?1) */
11255 case '5': case '6': case '7': case '8': case '9':
11256 RExC_parse = (char *) seqstart + 1; /* Point to the digit */
11259 bool is_neg = FALSE;
11261 parse_start = RExC_parse - 1; /* MJD */
11262 if (*RExC_parse == '-') {
11266 if (grok_atoUV(RExC_parse, &unum, &endptr)
11270 RExC_parse = (char*)endptr;
11274 /* Some limit for num? */
11278 if (*RExC_parse!=')')
11279 vFAIL("Expecting close bracket");
11282 if ( paren == '-' ) {
11284 Diagram of capture buffer numbering.
11285 Top line is the normal capture buffer numbers
11286 Bottom line is the negative indexing as from
11290 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
11294 num = RExC_npar + num;
11297 vFAIL("Reference to nonexistent group");
11299 } else if ( paren == '+' ) {
11300 num = RExC_npar + num - 1;
11302 /* We keep track how many GOSUB items we have produced.
11303 To start off the ARG2L() of the GOSUB holds its "id",
11304 which is used later in conjunction with RExC_recurse
11305 to calculate the offset we need to jump for the GOSUB,
11306 which it will store in the final representation.
11307 We have to defer the actual calculation until much later
11308 as the regop may move.
11311 ret = reg2Lanode(pRExC_state, GOSUB, num, RExC_recurse_count);
11313 if (num > (I32)RExC_rx->nparens) {
11315 vFAIL("Reference to nonexistent group");
11317 RExC_recurse_count++;
11318 DEBUG_OPTIMISE_MORE_r(Perl_re_printf( aTHX_
11319 "%*s%*s Recurse #%" UVuf " to %" IVdf "\n",
11320 22, "| |", (int)(depth * 2 + 1), "",
11321 (UV)ARG(ret), (IV)ARG2L(ret)));
11323 RExC_seen |= REG_RECURSE_SEEN;
11325 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
11326 Set_Node_Offset(ret, parse_start); /* MJD */
11328 *flagp |= POSTPONED;
11329 assert(*RExC_parse == ')');
11330 nextchar(pRExC_state);
11335 case '?': /* (??...) */
11337 if (*RExC_parse != '{') {
11338 RExC_parse += SKIP_IF_CHAR(RExC_parse);
11339 /* diag_listed_as: Sequence (?%s...) not recognized in regex; marked by <-- HERE in m/%s/ */
11341 "Sequence (%" UTF8f "...) not recognized",
11342 UTF8fARG(UTF, RExC_parse-seqstart, seqstart));
11343 NOT_REACHED; /*NOTREACHED*/
11345 *flagp |= POSTPONED;
11349 case '{': /* (?{...}) */
11352 struct reg_code_block *cb;
11354 RExC_seen_zerolen++;
11356 if ( !pRExC_state->code_blocks
11357 || pRExC_state->code_index
11358 >= pRExC_state->code_blocks->count
11359 || pRExC_state->code_blocks->cb[pRExC_state->code_index].start
11360 != (STRLEN)((RExC_parse -3 - (is_logical ? 1 : 0))
11363 if (RExC_pm_flags & PMf_USE_RE_EVAL)
11364 FAIL("panic: Sequence (?{...}): no code block found\n");
11365 FAIL("Eval-group not allowed at runtime, use re 'eval'");
11367 /* this is a pre-compiled code block (?{...}) */
11368 cb = &pRExC_state->code_blocks->cb[pRExC_state->code_index];
11369 RExC_parse = RExC_start + cb->end;
11372 if (cb->src_regex) {
11373 n = add_data(pRExC_state, STR_WITH_LEN("rl"));
11374 RExC_rxi->data->data[n] =
11375 (void*)SvREFCNT_inc((SV*)cb->src_regex);
11376 RExC_rxi->data->data[n+1] = (void*)o;
11379 n = add_data(pRExC_state,
11380 (RExC_pm_flags & PMf_HAS_CV) ? "L" : "l", 1);
11381 RExC_rxi->data->data[n] = (void*)o;
11384 pRExC_state->code_index++;
11385 nextchar(pRExC_state);
11389 ret = reg_node(pRExC_state, LOGICAL);
11391 eval = reg2Lanode(pRExC_state, EVAL,
11394 /* for later propagation into (??{})
11396 RExC_flags & RXf_PMf_COMPILETIME
11401 REGTAIL(pRExC_state, ret, eval);
11402 /* deal with the length of this later - MJD */
11405 ret = reg2Lanode(pRExC_state, EVAL, n, 0);
11406 Set_Node_Length(ret, RExC_parse - parse_start + 1);
11407 Set_Node_Offset(ret, parse_start);
11410 case '(': /* (?(?{...})...) and (?(?=...)...) */
11413 const int DEFINE_len = sizeof("DEFINE") - 1;
11414 if ( RExC_parse < RExC_end - 1
11415 && ( ( RExC_parse[0] == '?' /* (?(?...)) */
11416 && ( RExC_parse[1] == '='
11417 || RExC_parse[1] == '!'
11418 || RExC_parse[1] == '<'
11419 || RExC_parse[1] == '{'))
11420 || ( RExC_parse[0] == '*' /* (?(*...)) */
11421 && ( memBEGINs(RExC_parse + 1,
11422 (Size_t) (RExC_end - (RExC_parse + 1)),
11424 || memBEGINs(RExC_parse + 1,
11425 (Size_t) (RExC_end - (RExC_parse + 1)),
11427 || memBEGINs(RExC_parse + 1,
11428 (Size_t) (RExC_end - (RExC_parse + 1)),
11430 || memBEGINs(RExC_parse + 1,
11431 (Size_t) (RExC_end - (RExC_parse + 1)),
11433 || memBEGINs(RExC_parse + 1,
11434 (Size_t) (RExC_end - (RExC_parse + 1)),
11435 "positive_lookahead:")
11436 || memBEGINs(RExC_parse + 1,
11437 (Size_t) (RExC_end - (RExC_parse + 1)),
11438 "positive_lookbehind:")
11439 || memBEGINs(RExC_parse + 1,
11440 (Size_t) (RExC_end - (RExC_parse + 1)),
11441 "negative_lookahead:")
11442 || memBEGINs(RExC_parse + 1,
11443 (Size_t) (RExC_end - (RExC_parse + 1)),
11444 "negative_lookbehind:"))))
11445 ) { /* Lookahead or eval. */
11449 ret = reg_node(pRExC_state, LOGICAL);
11453 tail = reg(pRExC_state, 1, &flag, depth+1);
11454 RETURN_NULL_ON_RESTART(flag,flagp);
11455 REGTAIL(pRExC_state, ret, tail);
11458 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
11459 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
11461 char ch = RExC_parse[0] == '<' ? '>' : '\'';
11462 char *name_start= RExC_parse++;
11464 SV *sv_dat=reg_scan_name(pRExC_state,
11465 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
11466 if ( RExC_parse == name_start
11467 || RExC_parse >= RExC_end
11468 || *RExC_parse != ch)
11470 vFAIL2("Sequence (?(%c... not terminated",
11471 (ch == '>' ? '<' : ch));
11475 num = add_data( pRExC_state, STR_WITH_LEN("S"));
11476 RExC_rxi->data->data[num]=(void*)sv_dat;
11477 SvREFCNT_inc_simple_void(sv_dat);
11479 ret = reganode(pRExC_state,NGROUPP,num);
11480 goto insert_if_check_paren;
11482 else if (memBEGINs(RExC_parse,
11483 (STRLEN) (RExC_end - RExC_parse),
11486 ret = reganode(pRExC_state,DEFINEP,0);
11487 RExC_parse += DEFINE_len;
11489 goto insert_if_check_paren;
11491 else if (RExC_parse[0] == 'R') {
11493 /* parno == 0 => /(?(R)YES|NO)/ "in any form of recursion OR eval"
11494 * parno == 1 => /(?(R0)YES|NO)/ "in GOSUB (?0) / (?R)"
11495 * parno == 2 => /(?(R1)YES|NO)/ "in GOSUB (?1) (parno-1)"
11498 if (RExC_parse[0] == '0') {
11502 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
11504 if (grok_atoUV(RExC_parse, &uv, &endptr)
11507 parno = (I32)uv + 1;
11508 RExC_parse = (char*)endptr;
11510 /* else "Switch condition not recognized" below */
11511 } else if (RExC_parse[0] == '&') {
11514 sv_dat = reg_scan_name(pRExC_state,
11516 ? REG_RSN_RETURN_NULL
11517 : REG_RSN_RETURN_DATA);
11519 /* we should only have a false sv_dat when
11520 * SIZE_ONLY is true, and we always have false
11521 * sv_dat when SIZE_ONLY is true.
11522 * reg_scan_name() will VFAIL() if the name is
11523 * unknown when SIZE_ONLY is false, and otherwise
11524 * will return something, and when SIZE_ONLY is
11525 * true, reg_scan_name() just parses the string,
11526 * and doesnt return anything. (in theory) */
11527 assert(SIZE_ONLY ? !sv_dat : !!sv_dat);
11530 parno = 1 + *((I32 *)SvPVX(sv_dat));
11532 ret = reganode(pRExC_state,INSUBP,parno);
11533 goto insert_if_check_paren;
11535 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
11539 if (grok_atoUV(RExC_parse, &uv, &endptr)
11543 RExC_parse = (char*)endptr;
11546 vFAIL("panic: grok_atoUV returned FALSE");
11548 ret = reganode(pRExC_state, GROUPP, parno);
11550 insert_if_check_paren:
11551 if (UCHARAT(RExC_parse) != ')') {
11552 RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
11553 vFAIL("Switch condition not recognized");
11555 nextchar(pRExC_state);
11557 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
11558 br = regbranch(pRExC_state, &flags, 1,depth+1);
11560 RETURN_NULL_ON_RESTART(flags,flagp);
11561 FAIL2("panic: regbranch returned NULL, flags=%#" UVxf,
11564 REGTAIL(pRExC_state, br, reganode(pRExC_state,
11566 c = UCHARAT(RExC_parse);
11567 nextchar(pRExC_state);
11568 if (flags&HASWIDTH)
11569 *flagp |= HASWIDTH;
11572 vFAIL("(?(DEFINE)....) does not allow branches");
11574 /* Fake one for optimizer. */
11575 lastbr = reganode(pRExC_state, IFTHEN, 0);
11577 if (!regbranch(pRExC_state, &flags, 1,depth+1)) {
11578 RETURN_NULL_ON_RESTART(flags,flagp);
11579 FAIL2("panic: regbranch returned NULL, flags=%#" UVxf,
11582 REGTAIL(pRExC_state, ret, lastbr);
11583 if (flags&HASWIDTH)
11584 *flagp |= HASWIDTH;
11585 c = UCHARAT(RExC_parse);
11586 nextchar(pRExC_state);
11591 if (RExC_parse >= RExC_end)
11592 vFAIL("Switch (?(condition)... not terminated");
11594 vFAIL("Switch (?(condition)... contains too many branches");
11596 ender = reg_node(pRExC_state, TAIL);
11597 REGTAIL(pRExC_state, br, ender);
11599 REGTAIL(pRExC_state, lastbr, ender);
11600 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
11603 REGTAIL(pRExC_state, ret, ender);
11604 RExC_size++; /* XXX WHY do we need this?!!
11605 For large programs it seems to be required
11606 but I can't figure out why. -- dmq*/
11609 RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
11610 vFAIL("Unknown switch condition (?(...))");
11612 case '[': /* (?[ ... ]) */
11613 return handle_regex_sets(pRExC_state, NULL, flagp, depth+1,
11615 case 0: /* A NUL */
11616 RExC_parse--; /* for vFAIL to print correctly */
11617 vFAIL("Sequence (? incomplete");
11619 default: /* e.g., (?i) */
11620 RExC_parse = (char *) seqstart + 1;
11622 parse_lparen_question_flags(pRExC_state);
11623 if (UCHARAT(RExC_parse) != ':') {
11624 if (RExC_parse < RExC_end)
11625 nextchar(pRExC_state);
11630 nextchar(pRExC_state);
11636 if (*RExC_parse == '{' && PASS2) {
11637 ckWARNregdep(RExC_parse + 1,
11638 "Unescaped left brace in regex is "
11639 "deprecated here (and will be fatal "
11640 "in Perl 5.32), passed through");
11642 /* Not bothering to indent here, as the above 'else' is temporary
11644 if (!(RExC_flags & RXf_PMf_NOCAPTURE)) { /* (...) */
11649 ret = reganode(pRExC_state, OPEN, parno);
11651 if (!RExC_nestroot)
11652 RExC_nestroot = parno;
11653 if (RExC_open_parens && !RExC_open_parens[parno])
11655 DEBUG_OPTIMISE_MORE_r(Perl_re_printf( aTHX_
11656 "%*s%*s Setting open paren #%" IVdf " to %d\n",
11657 22, "| |", (int)(depth * 2 + 1), "",
11658 (IV)parno, REG_NODE_NUM(ret)));
11659 RExC_open_parens[parno]= ret;
11662 Set_Node_Length(ret, 1); /* MJD */
11663 Set_Node_Offset(ret, RExC_parse); /* MJD */
11666 /* with RXf_PMf_NOCAPTURE treat (...) as (?:...) */
11676 /* Pick up the branches, linking them together. */
11677 parse_start = RExC_parse; /* MJD */
11678 br = regbranch(pRExC_state, &flags, 1,depth+1);
11680 /* branch_len = (paren != 0); */
11683 RETURN_NULL_ON_RESTART(flags,flagp);
11684 FAIL2("panic: regbranch returned NULL, flags=%#" UVxf, (UV) flags);
11686 if (*RExC_parse == '|') {
11687 if (!SIZE_ONLY && RExC_extralen) {
11688 reginsert(pRExC_state, BRANCHJ, br, depth+1);
11691 reginsert(pRExC_state, BRANCH, br, depth+1);
11692 Set_Node_Length(br, paren != 0);
11693 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
11697 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
11699 else if (paren == ':') {
11700 *flagp |= flags&SIMPLE;
11702 if (is_open) { /* Starts with OPEN. */
11703 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
11705 else if (paren != '?') /* Not Conditional */
11707 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
11709 while (*RExC_parse == '|') {
11710 if (!SIZE_ONLY && RExC_extralen) {
11711 ender = reganode(pRExC_state, LONGJMP,0);
11713 /* Append to the previous. */
11714 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
11717 RExC_extralen += 2; /* Account for LONGJMP. */
11718 nextchar(pRExC_state);
11719 if (freeze_paren) {
11720 if (RExC_npar > after_freeze)
11721 after_freeze = RExC_npar;
11722 RExC_npar = freeze_paren;
11724 br = regbranch(pRExC_state, &flags, 0, depth+1);
11727 RETURN_NULL_ON_RESTART(flags,flagp);
11728 FAIL2("panic: regbranch returned NULL, flags=%#" UVxf, (UV) flags);
11730 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
11732 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
11735 if (have_branch || paren != ':') {
11736 /* Make a closing node, and hook it on the end. */
11739 ender = reg_node(pRExC_state, TAIL);
11742 ender = reganode(pRExC_state, CLOSE, parno);
11743 if ( RExC_close_parens ) {
11744 DEBUG_OPTIMISE_MORE_r(Perl_re_printf( aTHX_
11745 "%*s%*s Setting close paren #%" IVdf " to %d\n",
11746 22, "| |", (int)(depth * 2 + 1), "", (IV)parno, REG_NODE_NUM(ender)));
11747 RExC_close_parens[parno]= ender;
11748 if (RExC_nestroot == parno)
11751 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
11752 Set_Node_Length(ender,1); /* MJD */
11755 ender = reg_node(pRExC_state, SRCLOSE);
11756 RExC_in_script_run = 0;
11766 *flagp &= ~HASWIDTH;
11768 case 't': /* aTomic */
11770 ender = reg_node(pRExC_state, SUCCEED);
11773 ender = reg_node(pRExC_state, END);
11775 assert(!RExC_end_op); /* there can only be one! */
11776 RExC_end_op = ender;
11777 if (RExC_close_parens) {
11778 DEBUG_OPTIMISE_MORE_r(Perl_re_printf( aTHX_
11779 "%*s%*s Setting close paren #0 (END) to %d\n",
11780 22, "| |", (int)(depth * 2 + 1), "", REG_NODE_NUM(ender)));
11782 RExC_close_parens[0]= ender;
11787 DEBUG_PARSE_r(if (!SIZE_ONLY) {
11788 DEBUG_PARSE_MSG("lsbr");
11789 regprop(RExC_rx, RExC_mysv1, lastbr, NULL, pRExC_state);
11790 regprop(RExC_rx, RExC_mysv2, ender, NULL, pRExC_state);
11791 Perl_re_printf( aTHX_ "~ tying lastbr %s (%" IVdf ") to ender %s (%" IVdf ") offset %" IVdf "\n",
11792 SvPV_nolen_const(RExC_mysv1),
11793 (IV)REG_NODE_NUM(lastbr),
11794 SvPV_nolen_const(RExC_mysv2),
11795 (IV)REG_NODE_NUM(ender),
11796 (IV)(ender - lastbr)
11799 REGTAIL(pRExC_state, lastbr, ender);
11801 if (have_branch && !SIZE_ONLY) {
11802 char is_nothing= 1;
11804 RExC_seen |= REG_TOP_LEVEL_BRANCHES_SEEN;
11806 /* Hook the tails of the branches to the closing node. */
11807 for (br = ret; br; br = regnext(br)) {
11808 const U8 op = PL_regkind[OP(br)];
11809 if (op == BRANCH) {
11810 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
11811 if ( OP(NEXTOPER(br)) != NOTHING
11812 || regnext(NEXTOPER(br)) != ender)
11815 else if (op == BRANCHJ) {
11816 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
11817 /* for now we always disable this optimisation * /
11818 if ( OP(NEXTOPER(NEXTOPER(br))) != NOTHING
11819 || regnext(NEXTOPER(NEXTOPER(br))) != ender)
11825 br= PL_regkind[OP(ret)] != BRANCH ? regnext(ret) : ret;
11826 DEBUG_PARSE_r(if (!SIZE_ONLY) {
11827 DEBUG_PARSE_MSG("NADA");
11828 regprop(RExC_rx, RExC_mysv1, ret, NULL, pRExC_state);
11829 regprop(RExC_rx, RExC_mysv2, ender, NULL, pRExC_state);
11830 Perl_re_printf( aTHX_ "~ converting ret %s (%" IVdf ") to ender %s (%" IVdf ") offset %" IVdf "\n",
11831 SvPV_nolen_const(RExC_mysv1),
11832 (IV)REG_NODE_NUM(ret),
11833 SvPV_nolen_const(RExC_mysv2),
11834 (IV)REG_NODE_NUM(ender),
11839 if (OP(ender) == TAIL) {
11844 for ( opt= br + 1; opt < ender ; opt++ )
11845 OP(opt)= OPTIMIZED;
11846 NEXT_OFF(br)= ender - br;
11854 /* Even/odd or x=don't care: 010101x10x */
11855 static const char parens[] = "=!aA<,>Bbt";
11856 /* flag below is set to 0 up through 'A'; 1 for larger */
11858 if (paren && (p = strchr(parens, paren))) {
11859 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
11860 int flag = (p - parens) > 3;
11862 if (paren == '>' || paren == 't') {
11863 node = SUSPEND, flag = 0;
11866 reginsert(pRExC_state, node,ret, depth+1);
11867 Set_Node_Cur_Length(ret, parse_start);
11868 Set_Node_Offset(ret, parse_start + 1);
11870 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
11874 /* Check for proper termination. */
11876 /* restore original flags, but keep (?p) and, if we've changed from /d
11877 * rules to /u, keep the /u */
11878 RExC_flags = oregflags | (RExC_flags & RXf_PMf_KEEPCOPY);
11879 if (DEPENDS_SEMANTICS && RExC_uni_semantics) {
11880 set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET);
11882 if (RExC_parse >= RExC_end || UCHARAT(RExC_parse) != ')') {
11883 RExC_parse = oregcomp_parse;
11884 vFAIL("Unmatched (");
11886 nextchar(pRExC_state);
11888 else if (!paren && RExC_parse < RExC_end) {
11889 if (*RExC_parse == ')') {
11891 vFAIL("Unmatched )");
11894 FAIL("Junk on end of regexp"); /* "Can't happen". */
11895 NOT_REACHED; /* NOTREACHED */
11898 if (RExC_in_lookbehind) {
11899 RExC_in_lookbehind--;
11901 if (after_freeze > RExC_npar)
11902 RExC_npar = after_freeze;
11907 - regbranch - one alternative of an | operator
11909 * Implements the concatenation operator.
11911 * Returns NULL, setting *flagp to RESTART_PASS1 if the sizing scan needs to be
11912 * restarted, or'd with NEED_UTF8 if the pattern needs to be upgraded to UTF-8
11915 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
11918 regnode *chain = NULL;
11920 I32 flags = 0, c = 0;
11921 GET_RE_DEBUG_FLAGS_DECL;
11923 PERL_ARGS_ASSERT_REGBRANCH;
11925 DEBUG_PARSE("brnc");
11930 if (!SIZE_ONLY && RExC_extralen)
11931 ret = reganode(pRExC_state, BRANCHJ,0);
11933 ret = reg_node(pRExC_state, BRANCH);
11934 Set_Node_Length(ret, 1);
11938 if (!first && SIZE_ONLY)
11939 RExC_extralen += 1; /* BRANCHJ */
11941 *flagp = WORST; /* Tentatively. */
11943 skip_to_be_ignored_text(pRExC_state, &RExC_parse,
11944 FALSE /* Don't force to /x */ );
11945 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
11946 flags &= ~TRYAGAIN;
11947 latest = regpiece(pRExC_state, &flags,depth+1);
11948 if (latest == NULL) {
11949 if (flags & TRYAGAIN)
11951 RETURN_NULL_ON_RESTART(flags,flagp);
11952 FAIL2("panic: regpiece returned NULL, flags=%#" UVxf, (UV) flags);
11954 else if (ret == NULL)
11956 *flagp |= flags&(HASWIDTH|POSTPONED);
11957 if (chain == NULL) /* First piece. */
11958 *flagp |= flags&SPSTART;
11960 /* FIXME adding one for every branch after the first is probably
11961 * excessive now we have TRIE support. (hv) */
11963 REGTAIL(pRExC_state, chain, latest);
11968 if (chain == NULL) { /* Loop ran zero times. */
11969 chain = reg_node(pRExC_state, NOTHING);
11974 *flagp |= flags&SIMPLE;
11981 - regpiece - something followed by possible quantifier * + ? {n,m}
11983 * Note that the branching code sequences used for ? and the general cases
11984 * of * and + are somewhat optimized: they use the same NOTHING node as
11985 * both the endmarker for their branch list and the body of the last branch.
11986 * It might seem that this node could be dispensed with entirely, but the
11987 * endmarker role is not redundant.
11989 * Returns NULL, setting *flagp to TRYAGAIN if regatom() returns NULL with
11991 * Returns NULL, setting *flagp to RESTART_PASS1 if the sizing scan needs to be
11992 * restarted, or'd with NEED_UTF8 if the pattern needs to be upgraded to UTF-8
11995 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
12001 const char * const origparse = RExC_parse;
12003 I32 max = REG_INFTY;
12004 #ifdef RE_TRACK_PATTERN_OFFSETS
12007 const char *maxpos = NULL;
12010 /* Save the original in case we change the emitted regop to a FAIL. */
12011 regnode * const orig_emit = RExC_emit;
12013 GET_RE_DEBUG_FLAGS_DECL;
12015 PERL_ARGS_ASSERT_REGPIECE;
12017 DEBUG_PARSE("piec");
12019 ret = regatom(pRExC_state, &flags,depth+1);
12021 RETURN_NULL_ON_RESTART_OR_FLAGS(flags,flagp,TRYAGAIN);
12022 FAIL2("panic: regatom returned NULL, flags=%#" UVxf, (UV) flags);
12027 if (op == '{' && regcurly(RExC_parse)) {
12029 #ifdef RE_TRACK_PATTERN_OFFSETS
12030 parse_start = RExC_parse; /* MJD */
12032 next = RExC_parse + 1;
12033 while (isDIGIT(*next) || *next == ',') {
12034 if (*next == ',') {
12042 if (*next == '}') { /* got one */
12043 const char* endptr;
12047 if (isDIGIT(*RExC_parse)) {
12048 if (!grok_atoUV(RExC_parse, &uv, &endptr))
12049 vFAIL("Invalid quantifier in {,}");
12050 if (uv >= REG_INFTY)
12051 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
12056 if (*maxpos == ',')
12059 maxpos = RExC_parse;
12060 if (isDIGIT(*maxpos)) {
12061 if (!grok_atoUV(maxpos, &uv, &endptr))
12062 vFAIL("Invalid quantifier in {,}");
12063 if (uv >= REG_INFTY)
12064 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
12067 max = REG_INFTY; /* meaning "infinity" */
12070 nextchar(pRExC_state);
12071 if (max < min) { /* If can't match, warn and optimize to fail
12073 reginsert(pRExC_state, OPFAIL, orig_emit, depth+1);
12075 ckWARNreg(RExC_parse, "Quantifier {n,m} with n > m can't match");
12076 NEXT_OFF(orig_emit)= regarglen[OPFAIL] + NODE_STEP_REGNODE;
12080 else if (min == max && *RExC_parse == '?')
12083 ckWARN2reg(RExC_parse + 1,
12084 "Useless use of greediness modifier '%c'",
12090 if ((flags&SIMPLE)) {
12091 if (min == 0 && max == REG_INFTY) {
12092 reginsert(pRExC_state, STAR, ret, depth+1);
12094 RExC_seen |= REG_UNBOUNDED_QUANTIFIER_SEEN;
12097 if (min == 1 && max == REG_INFTY) {
12098 reginsert(pRExC_state, PLUS, ret, depth+1);
12100 RExC_seen |= REG_UNBOUNDED_QUANTIFIER_SEEN;
12103 MARK_NAUGHTY_EXP(2, 2);
12104 reginsert(pRExC_state, CURLY, ret, depth+1);
12105 Set_Node_Offset(ret, parse_start+1); /* MJD */
12106 Set_Node_Cur_Length(ret, parse_start);
12109 regnode * const w = reg_node(pRExC_state, WHILEM);
12112 REGTAIL(pRExC_state, ret, w);
12113 if (!SIZE_ONLY && RExC_extralen) {
12114 reginsert(pRExC_state, LONGJMP,ret, depth+1);
12115 reginsert(pRExC_state, NOTHING,ret, depth+1);
12116 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
12118 reginsert(pRExC_state, CURLYX,ret, depth+1);
12120 Set_Node_Offset(ret, parse_start+1);
12121 Set_Node_Length(ret,
12122 op == '{' ? (RExC_parse - parse_start) : 1);
12124 if (!SIZE_ONLY && RExC_extralen)
12125 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
12126 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
12128 RExC_whilem_seen++, RExC_extralen += 3;
12129 MARK_NAUGHTY_EXP(1, 4); /* compound interest */
12136 *flagp |= HASWIDTH;
12138 ARG1_SET(ret, (U16)min);
12139 ARG2_SET(ret, (U16)max);
12141 if (max == REG_INFTY)
12142 RExC_seen |= REG_UNBOUNDED_QUANTIFIER_SEEN;
12148 if (!ISMULT1(op)) {
12153 #if 0 /* Now runtime fix should be reliable. */
12155 /* if this is reinstated, don't forget to put this back into perldiag:
12157 =item Regexp *+ operand could be empty at {#} in regex m/%s/
12159 (F) The part of the regexp subject to either the * or + quantifier
12160 could match an empty string. The {#} shows in the regular
12161 expression about where the problem was discovered.
12165 if (!(flags&HASWIDTH) && op != '?')
12166 vFAIL("Regexp *+ operand could be empty");
12169 #ifdef RE_TRACK_PATTERN_OFFSETS
12170 parse_start = RExC_parse;
12172 nextchar(pRExC_state);
12174 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
12180 else if (op == '+') {
12184 else if (op == '?') {
12189 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
12190 SAVEFREESV(RExC_rx_sv); /* in case of fatal warnings */
12191 ckWARN2reg(RExC_parse,
12192 "%" UTF8f " matches null string many times",
12193 UTF8fARG(UTF, (RExC_parse >= origparse
12194 ? RExC_parse - origparse
12197 (void)ReREFCNT_inc(RExC_rx_sv);
12200 if (*RExC_parse == '?') {
12201 nextchar(pRExC_state);
12202 reginsert(pRExC_state, MINMOD, ret, depth+1);
12203 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
12205 else if (*RExC_parse == '+') {
12207 nextchar(pRExC_state);
12208 ender = reg_node(pRExC_state, SUCCEED);
12209 REGTAIL(pRExC_state, ret, ender);
12210 reginsert(pRExC_state, SUSPEND, ret, depth+1);
12211 ender = reg_node(pRExC_state, TAIL);
12212 REGTAIL(pRExC_state, ret, ender);
12215 if (ISMULT2(RExC_parse)) {
12217 vFAIL("Nested quantifiers");
12224 S_grok_bslash_N(pTHX_ RExC_state_t *pRExC_state,
12233 /* This routine teases apart the various meanings of \N and returns
12234 * accordingly. The input parameters constrain which meaning(s) is/are valid
12235 * in the current context.
12237 * Exactly one of <node_p> and <code_point_p> must be non-NULL.
12239 * If <code_point_p> is not NULL, the context is expecting the result to be a
12240 * single code point. If this \N instance turns out to a single code point,
12241 * the function returns TRUE and sets *code_point_p to that code point.
12243 * If <node_p> is not NULL, the context is expecting the result to be one of
12244 * the things representable by a regnode. If this \N instance turns out to be
12245 * one such, the function generates the regnode, returns TRUE and sets *node_p
12246 * to point to that regnode.
12248 * If this instance of \N isn't legal in any context, this function will
12249 * generate a fatal error and not return.
12251 * On input, RExC_parse should point to the first char following the \N at the
12252 * time of the call. On successful return, RExC_parse will have been updated
12253 * to point to just after the sequence identified by this routine. Also
12254 * *flagp has been updated as needed.
12256 * When there is some problem with the current context and this \N instance,
12257 * the function returns FALSE, without advancing RExC_parse, nor setting
12258 * *node_p, nor *code_point_p, nor *flagp.
12260 * If <cp_count> is not NULL, the caller wants to know the length (in code
12261 * points) that this \N sequence matches. This is set even if the function
12262 * returns FALSE, as detailed below.
12264 * There are 5 possibilities here, as detailed in the next 5 paragraphs.
12266 * Probably the most common case is for the \N to specify a single code point.
12267 * *cp_count will be set to 1, and *code_point_p will be set to that code
12270 * Another possibility is for the input to be an empty \N{}, which for
12271 * backwards compatibility we accept. *cp_count will be set to 0. *node_p
12272 * will be set to a generated NOTHING node.
12274 * Still another possibility is for the \N to mean [^\n]. *cp_count will be
12275 * set to 0. *node_p will be set to a generated REG_ANY node.
12277 * The fourth possibility is that \N resolves to a sequence of more than one
12278 * code points. *cp_count will be set to the number of code points in the
12279 * sequence. *node_p * will be set to a generated node returned by this
12280 * function calling S_reg().
12282 * The final possibility is that it is premature to be calling this function;
12283 * that pass1 needs to be restarted. This can happen when this changes from
12284 * /d to /u rules, or when the pattern needs to be upgraded to UTF-8. The
12285 * latter occurs only when the fourth possibility would otherwise be in
12286 * effect, and is because one of those code points requires the pattern to be
12287 * recompiled as UTF-8. The function returns FALSE, and sets the
12288 * RESTART_PASS1 and NEED_UTF8 flags in *flagp, as appropriate. When this
12289 * happens, the caller needs to desist from continuing parsing, and return
12290 * this information to its caller. This is not set for when there is only one
12291 * code point, as this can be called as part of an ANYOF node, and they can
12292 * store above-Latin1 code points without the pattern having to be in UTF-8.
12294 * For non-single-quoted regexes, the tokenizer has resolved character and
12295 * sequence names inside \N{...} into their Unicode values, normalizing the
12296 * result into what we should see here: '\N{U+c1.c2...}', where c1... are the
12297 * hex-represented code points in the sequence. This is done there because
12298 * the names can vary based on what charnames pragma is in scope at the time,
12299 * so we need a way to take a snapshot of what they resolve to at the time of
12300 * the original parse. [perl #56444].
12302 * That parsing is skipped for single-quoted regexes, so we may here get
12303 * '\N{NAME}'. This is a fatal error. These names have to be resolved by the
12304 * parser. But if the single-quoted regex is something like '\N{U+41}', that
12305 * is legal and handled here. The code point is Unicode, and has to be
12306 * translated into the native character set for non-ASCII platforms.
12309 char * endbrace; /* points to '}' following the name */
12310 char * endchar; /* Points to '.' or '}' ending cur char in the input
12312 char* p = RExC_parse; /* Temporary */
12314 SV * substitute_parse;
12320 GET_RE_DEBUG_FLAGS_DECL;
12322 PERL_ARGS_ASSERT_GROK_BSLASH_N;
12324 GET_RE_DEBUG_FLAGS;
12326 assert(cBOOL(node_p) ^ cBOOL(code_point_p)); /* Exactly one should be set */
12327 assert(! (node_p && cp_count)); /* At most 1 should be set */
12329 if (cp_count) { /* Initialize return for the most common case */
12333 /* The [^\n] meaning of \N ignores spaces and comments under the /x
12334 * modifier. The other meanings do not, so use a temporary until we find
12335 * out which we are being called with */
12336 skip_to_be_ignored_text(pRExC_state, &p,
12337 FALSE /* Don't force to /x */ );
12339 /* Disambiguate between \N meaning a named character versus \N meaning
12340 * [^\n]. The latter is assumed when the {...} following the \N is a legal
12341 * quantifier, or there is no '{' at all */
12342 if (*p != '{' || regcurly(p)) {
12352 *node_p = reg_node(pRExC_state, REG_ANY);
12353 *flagp |= HASWIDTH|SIMPLE;
12355 Set_Node_Length(*node_p, 1); /* MJD */
12359 /* The test above made sure that the next real character is a '{', but
12360 * under the /x modifier, it could be separated by space (or a comment and
12361 * \n) and this is not allowed (for consistency with \x{...} and the
12362 * tokenizer handling of \N{NAME}). */
12363 if (*RExC_parse != '{') {
12364 vFAIL("Missing braces on \\N{}");
12367 RExC_parse++; /* Skip past the '{' */
12369 endbrace = (char *) memchr(RExC_parse, '}', RExC_end - RExC_parse);
12370 if (! endbrace) { /* no trailing brace */
12371 vFAIL2("Missing right brace on \\%c{}", 'N');
12374 /* Here, we have decided it should be a named character or sequence */
12375 REQUIRE_UNI_RULES(flagp, FALSE); /* Unicode named chars imply Unicode
12378 if (endbrace == RExC_parse) { /* empty: \N{} */
12380 RExC_parse++; /* Position after the "}" */
12381 vFAIL("Zero length \\N{}");
12386 nextchar(pRExC_state);
12391 *node_p = reg_node(pRExC_state,NOTHING);
12395 /* If we haven't got something that begins with 'U+', then it didn't get lexed. */
12396 if ( endbrace - RExC_parse < 2
12397 || strnNE(RExC_parse, "U+", 2))
12399 RExC_parse = endbrace; /* position msg's '<--HERE' */
12400 vFAIL("\\N{NAME} must be resolved by the lexer");
12403 RExC_parse += 2; /* Skip past the 'U+' */
12405 /* Because toke.c has generated a special construct for us guaranteed
12406 * not to have NULs, we can use a str function */
12407 endchar = RExC_parse + strcspn(RExC_parse, ".}");
12409 /* Code points are separated by dots. If none, there is only one code
12410 * point, and is terminated by the brace */
12412 if (endchar >= endbrace) {
12413 STRLEN length_of_hex;
12414 I32 grok_hex_flags;
12416 /* Here, exactly one code point. If that isn't what is wanted,
12418 if (! code_point_p) {
12423 /* Convert code point from hex */
12424 length_of_hex = (STRLEN)(endchar - RExC_parse);
12425 grok_hex_flags = PERL_SCAN_ALLOW_UNDERSCORES
12426 | PERL_SCAN_DISALLOW_PREFIX
12428 /* No errors in the first pass (See [perl
12429 * #122671].) We let the code below find the
12430 * errors when there are multiple chars. */
12432 ? PERL_SCAN_SILENT_ILLDIGIT
12435 /* This routine is the one place where both single- and
12436 * double-quotish \N{U+xxxx} are evaluated. The value is a Unicode
12437 * code point which must be converted to native. */
12438 *code_point_p = UNI_TO_NATIVE(grok_hex(RExC_parse,
12443 /* The tokenizer should have guaranteed validity, but it's possible
12444 * to bypass it by using single quoting, so check. Don't do the
12445 * check here when there are multiple chars; we do it below anyway.
12447 if (length_of_hex == 0
12448 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
12450 RExC_parse += length_of_hex; /* Includes all the valid */
12451 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
12452 ? UTF8SKIP(RExC_parse)
12454 /* Guard against malformed utf8 */
12455 if (RExC_parse >= endchar) {
12456 RExC_parse = endchar;
12458 vFAIL("Invalid hexadecimal number in \\N{U+...}");
12461 RExC_parse = endbrace + 1;
12465 /* Here, is a multiple character sequence */
12467 /* Count the code points, if desired, in the sequence */
12470 while (RExC_parse < endbrace) {
12471 /* Point to the beginning of the next character in the sequence. */
12472 RExC_parse = endchar + 1;
12473 endchar = RExC_parse + strcspn(RExC_parse, ".}");
12478 /* Fail if caller doesn't want to handle a multi-code-point sequence.
12479 * But don't backup up the pointer if the caller wants to know how many
12480 * code points there are (they can then handle things) */
12488 /* What is done here is to convert this to a sub-pattern of the form
12489 * \x{char1}\x{char2}... and then call reg recursively to parse it
12490 * (enclosing in "(?: ... )" ). That way, it retains its atomicness,
12491 * while not having to worry about special handling that some code
12492 * points may have. */
12494 substitute_parse = newSVpvs("?:");
12496 while (RExC_parse < endbrace) {
12498 /* Convert to notation the rest of the code understands */
12499 sv_catpv(substitute_parse, "\\x{");
12500 sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse);
12501 sv_catpv(substitute_parse, "}");
12503 /* Point to the beginning of the next character in the sequence. */
12504 RExC_parse = endchar + 1;
12505 endchar = RExC_parse + strcspn(RExC_parse, ".}");
12508 sv_catpv(substitute_parse, ")");
12510 len = SvCUR(substitute_parse);
12512 /* Don't allow empty number */
12513 if (len < (STRLEN) 8) {
12514 RExC_parse = endbrace;
12515 vFAIL("Invalid hexadecimal number in \\N{U+...}");
12518 /* The values are Unicode, and therefore not subject to recoding, but
12519 * have to be converted to native on a non-Unicode (meaning non-ASCII)
12522 RExC_recode_x_to_native = 1;
12525 save_start = RExC_start;
12526 orig_end = RExC_end;
12528 RExC_parse = RExC_start = RExC_adjusted_start = SvPV(substitute_parse,
12530 RExC_end = RExC_parse + len;
12532 *node_p = reg(pRExC_state, 1, &flags, depth+1);
12534 /* Restore the saved values */
12535 RExC_start = RExC_adjusted_start = save_start;
12536 RExC_parse = endbrace;
12537 RExC_end = orig_end;
12539 RExC_recode_x_to_native = 0;
12542 SvREFCNT_dec_NN(substitute_parse);
12545 RETURN_X_ON_RESTART(FALSE, flags,flagp);
12546 FAIL2("panic: reg returned NULL to grok_bslash_N, flags=%#" UVxf,
12549 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
12551 nextchar(pRExC_state);
12557 PERL_STATIC_INLINE U8
12558 S_compute_EXACTish(RExC_state_t *pRExC_state)
12562 PERL_ARGS_ASSERT_COMPUTE_EXACTISH;
12570 op = get_regex_charset(RExC_flags);
12571 if (op >= REGEX_ASCII_RESTRICTED_CHARSET) {
12572 op--; /* /a is same as /u, and map /aa's offset to what /a's would have
12573 been, so there is no hole */
12576 return op + EXACTF;
12579 PERL_STATIC_INLINE void
12580 S_alloc_maybe_populate_EXACT(pTHX_ RExC_state_t *pRExC_state,
12581 regnode *node, I32* flagp, STRLEN len, UV code_point,
12584 /* This knows the details about sizing an EXACTish node, setting flags for
12585 * it (by setting <*flagp>, and potentially populating it with a single
12588 * If <len> (the length in bytes) is non-zero, this function assumes that
12589 * the node has already been populated, and just does the sizing. In this
12590 * case <code_point> should be the final code point that has already been
12591 * placed into the node. This value will be ignored except that under some
12592 * circumstances <*flagp> is set based on it.
12594 * If <len> is zero, the function assumes that the node is to contain only
12595 * the single character given by <code_point> and calculates what <len>
12596 * should be. In pass 1, it sizes the node appropriately. In pass 2, it
12597 * additionally will populate the node's STRING with <code_point> or its
12600 * In both cases <*flagp> is appropriately set
12602 * It knows that under FOLD, the Latin Sharp S and UTF characters above
12603 * 255, must be folded (the former only when the rules indicate it can
12606 * When it does the populating, it looks at the flag 'downgradable'. If
12607 * true with a node that folds, it checks if the single code point
12608 * participates in a fold, and if not downgrades the node to an EXACT.
12609 * This helps the optimizer */
12611 bool len_passed_in = cBOOL(len != 0);
12612 U8 character[UTF8_MAXBYTES_CASE+1];
12614 PERL_ARGS_ASSERT_ALLOC_MAYBE_POPULATE_EXACT;
12616 /* Don't bother to check for downgrading in PASS1, as it doesn't make any
12617 * sizing difference, and is extra work that is thrown away */
12618 if (downgradable && ! PASS2) {
12619 downgradable = FALSE;
12622 if (! len_passed_in) {
12624 if (UVCHR_IS_INVARIANT(code_point)) {
12625 if (LOC || ! FOLD) { /* /l defers folding until runtime */
12626 *character = (U8) code_point;
12628 else { /* Here is /i and not /l. (toFOLD() is defined on just
12629 ASCII, which isn't the same thing as INVARIANT on
12630 EBCDIC, but it works there, as the extra invariants
12631 fold to themselves) */
12632 *character = toFOLD((U8) code_point);
12634 /* We can downgrade to an EXACT node if this character
12635 * isn't a folding one. Note that this assumes that
12636 * nothing above Latin1 folds to some other invariant than
12637 * one of these alphabetics; otherwise we would also have
12639 * && (! HAS_NONLATIN1_FOLD_CLOSURE(code_point)
12640 * || ASCII_FOLD_RESTRICTED))
12642 if (downgradable && PL_fold[code_point] == code_point) {
12648 else if (FOLD && (! LOC
12649 || ! is_PROBLEMATIC_LOCALE_FOLD_cp(code_point)))
12650 { /* Folding, and ok to do so now */
12651 UV folded = _to_uni_fold_flags(
12655 FOLD_FLAGS_FULL | ((ASCII_FOLD_RESTRICTED)
12656 ? FOLD_FLAGS_NOMIX_ASCII
12659 && folded == code_point /* This quickly rules out many
12660 cases, avoiding the
12661 _invlist_contains_cp() overhead
12663 && ! _invlist_contains_cp(PL_utf8_foldable, code_point))
12670 else if (code_point <= MAX_UTF8_TWO_BYTE) {
12672 /* Not folding this cp, and can output it directly */
12673 *character = UTF8_TWO_BYTE_HI(code_point);
12674 *(character + 1) = UTF8_TWO_BYTE_LO(code_point);
12678 uvchr_to_utf8( character, code_point);
12679 len = UTF8SKIP(character);
12681 } /* Else pattern isn't UTF8. */
12683 *character = (U8) code_point;
12685 } /* Else is folded non-UTF8 */
12686 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
12687 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
12688 || UNICODE_DOT_DOT_VERSION > 0)
12689 else if (LIKELY(code_point != LATIN_SMALL_LETTER_SHARP_S)) {
12693 /* We don't fold any non-UTF8 except possibly the Sharp s (see
12694 * comments at join_exact()); */
12695 *character = (U8) code_point;
12698 /* Can turn into an EXACT node if we know the fold at compile time,
12699 * and it folds to itself and doesn't particpate in other folds */
12702 && PL_fold_latin1[code_point] == code_point
12703 && (! HAS_NONLATIN1_FOLD_CLOSURE(code_point)
12704 || (isASCII(code_point) && ASCII_FOLD_RESTRICTED)))
12708 } /* else is Sharp s. May need to fold it */
12709 else if (AT_LEAST_UNI_SEMANTICS && ! ASCII_FOLD_RESTRICTED) {
12711 *(character + 1) = 's';
12715 *character = LATIN_SMALL_LETTER_SHARP_S;
12721 RExC_size += STR_SZ(len);
12724 RExC_emit += STR_SZ(len);
12725 STR_LEN(node) = len;
12726 if (! len_passed_in) {
12727 Copy((char *) character, STRING(node), len, char);
12731 *flagp |= HASWIDTH;
12733 /* A single character node is SIMPLE, except for the special-cased SHARP S
12735 if ((len == 1 || (UTF && len == UVCHR_SKIP(code_point)))
12736 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
12737 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
12738 || UNICODE_DOT_DOT_VERSION > 0)
12739 && ( code_point != LATIN_SMALL_LETTER_SHARP_S
12740 || ! FOLD || ! DEPENDS_SEMANTICS)
12746 /* The OP may not be well defined in PASS1 */
12747 if (PASS2 && OP(node) == EXACTFL) {
12748 RExC_contains_locale = 1;
12753 S_new_regcurly(const char *s, const char *e)
12755 /* This is a temporary function designed to match the most lenient form of
12756 * a {m,n} quantifier we ever envision, with either number omitted, and
12757 * spaces anywhere between/before/after them.
12759 * If this function fails, then the string it matches is very unlikely to
12760 * ever be considered a valid quantifier, so we can allow the '{' that
12761 * begins it to be considered as a literal */
12763 bool has_min = FALSE;
12764 bool has_max = FALSE;
12766 PERL_ARGS_ASSERT_NEW_REGCURLY;
12768 if (s >= e || *s++ != '{')
12771 while (s < e && isSPACE(*s)) {
12774 while (s < e && isDIGIT(*s)) {
12778 while (s < e && isSPACE(*s)) {
12784 while (s < e && isSPACE(*s)) {
12787 while (s < e && isDIGIT(*s)) {
12791 while (s < e && isSPACE(*s)) {
12796 return s < e && *s == '}' && (has_min || has_max);
12799 /* Parse backref decimal value, unless it's too big to sensibly be a backref,
12800 * in which case return I32_MAX (rather than possibly 32-bit wrapping) */
12803 S_backref_value(char *p)
12805 const char* endptr;
12807 if (grok_atoUV(p, &val, &endptr) && val <= I32_MAX)
12814 - regatom - the lowest level
12816 Try to identify anything special at the start of the current parse position.
12817 If there is, then handle it as required. This may involve generating a
12818 single regop, such as for an assertion; or it may involve recursing, such as
12819 to handle a () structure.
12821 If the string doesn't start with something special then we gobble up
12822 as much literal text as we can. If we encounter a quantifier, we have to
12823 back off the final literal character, as that quantifier applies to just it
12824 and not to the whole string of literals.
12826 Once we have been able to handle whatever type of thing started the
12827 sequence, we return.
12829 Note: we have to be careful with escapes, as they can be both literal
12830 and special, and in the case of \10 and friends, context determines which.
12832 A summary of the code structure is:
12834 switch (first_byte) {
12835 cases for each special:
12836 handle this special;
12839 switch (2nd byte) {
12840 cases for each unambiguous special:
12841 handle this special;
12843 cases for each ambigous special/literal:
12845 if (special) handle here
12847 default: // unambiguously literal:
12850 default: // is a literal char
12853 create EXACTish node for literal;
12854 while (more input and node isn't full) {
12855 switch (input_byte) {
12856 cases for each special;
12857 make sure parse pointer is set so that the next call to
12858 regatom will see this special first
12859 goto loopdone; // EXACTish node terminated by prev. char
12861 append char to EXACTISH node;
12863 get next input byte;
12867 return the generated node;
12869 Specifically there are two separate switches for handling
12870 escape sequences, with the one for handling literal escapes requiring
12871 a dummy entry for all of the special escapes that are actually handled
12874 Returns NULL, setting *flagp to TRYAGAIN if reg() returns NULL with
12876 Returns NULL, setting *flagp to RESTART_PASS1 if the sizing scan needs to be
12877 restarted, or'd with NEED_UTF8 if the pattern needs to be upgraded to UTF-8
12878 Otherwise does not return NULL.
12882 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
12884 regnode *ret = NULL;
12891 GET_RE_DEBUG_FLAGS_DECL;
12893 *flagp = WORST; /* Tentatively. */
12895 DEBUG_PARSE("atom");
12897 PERL_ARGS_ASSERT_REGATOM;
12900 parse_start = RExC_parse;
12901 assert(RExC_parse < RExC_end);
12902 switch ((U8)*RExC_parse) {
12904 RExC_seen_zerolen++;
12905 nextchar(pRExC_state);
12906 if (RExC_flags & RXf_PMf_MULTILINE)
12907 ret = reg_node(pRExC_state, MBOL);
12909 ret = reg_node(pRExC_state, SBOL);
12910 Set_Node_Length(ret, 1); /* MJD */
12913 nextchar(pRExC_state);
12915 RExC_seen_zerolen++;
12916 if (RExC_flags & RXf_PMf_MULTILINE)
12917 ret = reg_node(pRExC_state, MEOL);
12919 ret = reg_node(pRExC_state, SEOL);
12920 Set_Node_Length(ret, 1); /* MJD */
12923 nextchar(pRExC_state);
12924 if (RExC_flags & RXf_PMf_SINGLELINE)
12925 ret = reg_node(pRExC_state, SANY);
12927 ret = reg_node(pRExC_state, REG_ANY);
12928 *flagp |= HASWIDTH|SIMPLE;
12930 Set_Node_Length(ret, 1); /* MJD */
12934 char * const oregcomp_parse = ++RExC_parse;
12935 ret = regclass(pRExC_state, flagp,depth+1,
12936 FALSE, /* means parse the whole char class */
12937 TRUE, /* allow multi-char folds */
12938 FALSE, /* don't silence non-portable warnings. */
12939 (bool) RExC_strict,
12940 TRUE, /* Allow an optimized regnode result */
12944 RETURN_NULL_ON_RESTART_FLAGP_OR_FLAGS(flagp,NEED_UTF8);
12945 FAIL2("panic: regclass returned NULL to regatom, flags=%#" UVxf,
12948 if (*RExC_parse != ']') {
12949 RExC_parse = oregcomp_parse;
12950 vFAIL("Unmatched [");
12952 nextchar(pRExC_state);
12953 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
12957 nextchar(pRExC_state);
12958 ret = reg(pRExC_state, 2, &flags,depth+1);
12960 if (flags & TRYAGAIN) {
12961 if (RExC_parse >= RExC_end) {
12962 /* Make parent create an empty node if needed. */
12963 *flagp |= TRYAGAIN;
12968 RETURN_NULL_ON_RESTART(flags,flagp);
12969 FAIL2("panic: reg returned NULL to regatom, flags=%#" UVxf,
12972 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
12976 if (flags & TRYAGAIN) {
12977 *flagp |= TRYAGAIN;
12980 vFAIL("Internal urp");
12981 /* Supposed to be caught earlier. */
12987 vFAIL("Quantifier follows nothing");
12992 This switch handles escape sequences that resolve to some kind
12993 of special regop and not to literal text. Escape sequnces that
12994 resolve to literal text are handled below in the switch marked
12997 Every entry in this switch *must* have a corresponding entry
12998 in the literal escape switch. However, the opposite is not
12999 required, as the default for this switch is to jump to the
13000 literal text handling code.
13003 switch ((U8)*RExC_parse) {
13004 /* Special Escapes */
13006 RExC_seen_zerolen++;
13007 ret = reg_node(pRExC_state, SBOL);
13008 /* SBOL is shared with /^/ so we set the flags so we can tell
13009 * /\A/ from /^/ in split. We check ret because first pass we
13010 * have no regop struct to set the flags on. */
13014 goto finish_meta_pat;
13016 ret = reg_node(pRExC_state, GPOS);
13017 RExC_seen |= REG_GPOS_SEEN;
13019 goto finish_meta_pat;
13021 RExC_seen_zerolen++;
13022 ret = reg_node(pRExC_state, KEEPS);
13024 /* XXX:dmq : disabling in-place substitution seems to
13025 * be necessary here to avoid cases of memory corruption, as
13026 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
13028 RExC_seen |= REG_LOOKBEHIND_SEEN;
13029 goto finish_meta_pat;
13031 ret = reg_node(pRExC_state, SEOL);
13033 RExC_seen_zerolen++; /* Do not optimize RE away */
13034 goto finish_meta_pat;
13036 ret = reg_node(pRExC_state, EOS);
13038 RExC_seen_zerolen++; /* Do not optimize RE away */
13039 goto finish_meta_pat;
13041 vFAIL("\\C no longer supported");
13043 ret = reg_node(pRExC_state, CLUMP);
13044 *flagp |= HASWIDTH;
13045 goto finish_meta_pat;
13051 arg = ANYOF_WORDCHAR;
13059 regex_charset charset = get_regex_charset(RExC_flags);
13061 RExC_seen_zerolen++;
13062 RExC_seen |= REG_LOOKBEHIND_SEEN;
13063 op = BOUND + charset;
13065 if (op == BOUNDL) {
13066 RExC_contains_locale = 1;
13069 ret = reg_node(pRExC_state, op);
13071 if (RExC_parse >= RExC_end || *(RExC_parse + 1) != '{') {
13072 FLAGS(ret) = TRADITIONAL_BOUND;
13073 if (PASS2 && op > BOUNDA) { /* /aa is same as /a */
13079 char name = *RExC_parse;
13080 char * endbrace = NULL;
13082 if (RExC_parse < RExC_end) {
13083 endbrace = (char *) memchr(RExC_parse, '}', RExC_end - RExC_parse);
13087 vFAIL2("Missing right brace on \\%c{}", name);
13089 /* XXX Need to decide whether to take spaces or not. Should be
13090 * consistent with \p{}, but that currently is SPACE, which
13091 * means vertical too, which seems wrong
13092 * while (isBLANK(*RExC_parse)) {
13095 if (endbrace == RExC_parse) {
13096 RExC_parse++; /* After the '}' */
13097 vFAIL2("Empty \\%c{}", name);
13099 length = endbrace - RExC_parse;
13100 /*while (isBLANK(*(RExC_parse + length - 1))) {
13103 switch (*RExC_parse) {
13106 && (memNEs(RExC_parse + 1, length - 1, "cb")))
13108 goto bad_bound_type;
13110 FLAGS(ret) = GCB_BOUND;
13113 if (length != 2 || *(RExC_parse + 1) != 'b') {
13114 goto bad_bound_type;
13116 FLAGS(ret) = LB_BOUND;
13119 if (length != 2 || *(RExC_parse + 1) != 'b') {
13120 goto bad_bound_type;
13122 FLAGS(ret) = SB_BOUND;
13125 if (length != 2 || *(RExC_parse + 1) != 'b') {
13126 goto bad_bound_type;
13128 FLAGS(ret) = WB_BOUND;
13132 RExC_parse = endbrace;
13134 "'%" UTF8f "' is an unknown bound type",
13135 UTF8fARG(UTF, length, endbrace - length));
13136 NOT_REACHED; /*NOTREACHED*/
13138 RExC_parse = endbrace;
13139 REQUIRE_UNI_RULES(flagp, NULL);
13141 if (PASS2 && op >= BOUNDA) { /* /aa is same as /a */
13145 /* Don't have to worry about UTF-8, in this message because
13146 * to get here the contents of the \b must be ASCII */
13147 ckWARN4reg(RExC_parse + 1, /* Include the '}' in msg */
13148 "Using /u for '%.*s' instead of /%s",
13150 endbrace - length + 1,
13151 (charset == REGEX_ASCII_RESTRICTED_CHARSET)
13152 ? ASCII_RESTRICT_PAT_MODS
13153 : ASCII_MORE_RESTRICT_PAT_MODS);
13157 if (PASS2 && invert) {
13158 OP(ret) += NBOUND - BOUND;
13160 goto finish_meta_pat;
13168 if (! DEPENDS_SEMANTICS) {
13172 /* \d doesn't have any matches in the upper Latin1 range, hence /d
13173 * is equivalent to /u. Changing to /u saves some branches at
13176 goto join_posix_op_known;
13179 ret = reg_node(pRExC_state, LNBREAK);
13180 *flagp |= HASWIDTH|SIMPLE;
13181 goto finish_meta_pat;
13189 goto join_posix_op_known;
13195 arg = ANYOF_VERTWS;
13197 goto join_posix_op_known;
13207 op = POSIXD + get_regex_charset(RExC_flags);
13208 if (op > POSIXA) { /* /aa is same as /a */
13211 else if (op == POSIXL) {
13212 RExC_contains_locale = 1;
13215 join_posix_op_known:
13218 op += NPOSIXD - POSIXD;
13221 ret = reg_node(pRExC_state, op);
13223 FLAGS(ret) = namedclass_to_classnum(arg);
13226 *flagp |= HASWIDTH|SIMPLE;
13230 if ( UCHARAT(RExC_parse + 1) == '{'
13231 && UNLIKELY(! new_regcurly(RExC_parse + 1, RExC_end)))
13234 vFAIL("Unescaped left brace in regex is illegal here");
13236 nextchar(pRExC_state);
13237 Set_Node_Length(ret, 2); /* MJD */
13243 ret = regclass(pRExC_state, flagp,depth+1,
13244 TRUE, /* means just parse this element */
13245 FALSE, /* don't allow multi-char folds */
13246 FALSE, /* don't silence non-portable warnings. It
13247 would be a bug if these returned
13249 (bool) RExC_strict,
13250 TRUE, /* Allow an optimized regnode result */
13253 RETURN_NULL_ON_RESTART_FLAGP(flagp);
13254 /* regclass() can only return RESTART_PASS1 and NEED_UTF8 if
13255 * multi-char folds are allowed. */
13257 FAIL2("panic: regclass returned NULL to regatom, flags=%#" UVxf,
13262 Set_Node_Offset(ret, parse_start);
13263 Set_Node_Cur_Length(ret, parse_start - 2);
13264 nextchar(pRExC_state);
13267 /* Handle \N, \N{} and \N{NAMED SEQUENCE} (the latter meaning the
13268 * \N{...} evaluates to a sequence of more than one code points).
13269 * The function call below returns a regnode, which is our result.
13270 * The parameters cause it to fail if the \N{} evaluates to a
13271 * single code point; we handle those like any other literal. The
13272 * reason that the multicharacter case is handled here and not as
13273 * part of the EXACtish code is because of quantifiers. In
13274 * /\N{BLAH}+/, the '+' applies to the whole thing, and doing it
13275 * this way makes that Just Happen. dmq.
13276 * join_exact() will join this up with adjacent EXACTish nodes
13277 * later on, if appropriate. */
13279 if (grok_bslash_N(pRExC_state,
13280 &ret, /* Want a regnode returned */
13281 NULL, /* Fail if evaluates to a single code
13283 NULL, /* Don't need a count of how many code
13292 RETURN_NULL_ON_RESTART_FLAGP(flagp);
13294 /* Here, evaluates to a single code point. Go get that */
13295 RExC_parse = parse_start;
13298 case 'k': /* Handle \k<NAME> and \k'NAME' */
13302 if ( RExC_parse >= RExC_end - 1
13303 || (( ch = RExC_parse[1]) != '<'
13308 /* diag_listed_as: Sequence \%s... not terminated in regex; marked by <-- HERE in m/%s/ */
13309 vFAIL2("Sequence %.2s... not terminated",parse_start);
13312 ret = handle_named_backref(pRExC_state,
13324 case '1': case '2': case '3': case '4':
13325 case '5': case '6': case '7': case '8': case '9':
13330 if (*RExC_parse == 'g') {
13334 if (*RExC_parse == '{') {
13338 if (*RExC_parse == '-') {
13342 if (hasbrace && !isDIGIT(*RExC_parse)) {
13343 if (isrel) RExC_parse--;
13345 goto parse_named_seq;
13348 if (RExC_parse >= RExC_end) {
13349 goto unterminated_g;
13351 num = S_backref_value(RExC_parse);
13353 vFAIL("Reference to invalid group 0");
13354 else if (num == I32_MAX) {
13355 if (isDIGIT(*RExC_parse))
13356 vFAIL("Reference to nonexistent group");
13359 vFAIL("Unterminated \\g... pattern");
13363 num = RExC_npar - num;
13365 vFAIL("Reference to nonexistent or unclosed group");
13369 num = S_backref_value(RExC_parse);
13370 /* bare \NNN might be backref or octal - if it is larger
13371 * than or equal RExC_npar then it is assumed to be an
13372 * octal escape. Note RExC_npar is +1 from the actual
13373 * number of parens. */
13374 /* Note we do NOT check if num == I32_MAX here, as that is
13375 * handled by the RExC_npar check */
13378 /* any numeric escape < 10 is always a backref */
13380 /* any numeric escape < RExC_npar is a backref */
13381 && num >= RExC_npar
13382 /* cannot be an octal escape if it starts with 8 */
13383 && *RExC_parse != '8'
13384 /* cannot be an octal escape it it starts with 9 */
13385 && *RExC_parse != '9'
13388 /* Probably not a backref, instead likely to be an
13389 * octal character escape, e.g. \35 or \777.
13390 * The above logic should make it obvious why using
13391 * octal escapes in patterns is problematic. - Yves */
13392 RExC_parse = parse_start;
13397 /* At this point RExC_parse points at a numeric escape like
13398 * \12 or \88 or something similar, which we should NOT treat
13399 * as an octal escape. It may or may not be a valid backref
13400 * escape. For instance \88888888 is unlikely to be a valid
13402 while (isDIGIT(*RExC_parse))
13405 if (*RExC_parse != '}')
13406 vFAIL("Unterminated \\g{...} pattern");
13410 if (num > (I32)RExC_rx->nparens)
13411 vFAIL("Reference to nonexistent group");
13414 ret = reganode(pRExC_state,
13417 : (ASCII_FOLD_RESTRICTED)
13419 : (AT_LEAST_UNI_SEMANTICS)
13425 *flagp |= HASWIDTH;
13427 /* override incorrect value set in reganode MJD */
13428 Set_Node_Offset(ret, parse_start);
13429 Set_Node_Cur_Length(ret, parse_start-1);
13430 skip_to_be_ignored_text(pRExC_state, &RExC_parse,
13431 FALSE /* Don't force to /x */ );
13435 if (RExC_parse >= RExC_end)
13436 FAIL("Trailing \\");
13439 /* Do not generate "unrecognized" warnings here, we fall
13440 back into the quick-grab loop below */
13441 RExC_parse = parse_start;
13443 } /* end of switch on a \foo sequence */
13448 /* '#' comments should have been spaced over before this function was
13450 assert((RExC_flags & RXf_PMf_EXTENDED) == 0);
13452 if (RExC_flags & RXf_PMf_EXTENDED) {
13453 RExC_parse = reg_skipcomment( pRExC_state, RExC_parse );
13454 if (RExC_parse < RExC_end)
13464 /* Here, we have determined that the next thing is probably a
13465 * literal character. RExC_parse points to the first byte of its
13466 * definition. (It still may be an escape sequence that evaluates
13467 * to a single character) */
13474 /* This allows us to fill a node with just enough spare so that if the final
13475 * character folds, its expansion is guaranteed to fit */
13476 #define MAX_NODE_STRING_SIZE (255-UTF8_MAXBYTES_CASE)
13477 char foldbuf[MAX_NODE_STRING_SIZE+UTF8_MAXBYTES_CASE+1];
13480 U8 upper_parse = MAX_NODE_STRING_SIZE;
13482 /* We start out as an EXACT node, even if under /i, until we find a
13483 * character which is in a fold. The algorithm now segregates into
13484 * separate nodes, characters that fold from those that don't under
13485 * /i. (This hopefull will create nodes that are fixed strings
13486 * even under /i, giving the optimizer something to grab onto to.)
13487 * So, if a node has something in it and the next character is in
13488 * the opposite category, that node is closed up, and the function
13489 * returns. Then regatom is called again, and a new node is
13490 * created for the new category. */
13491 U8 node_type = EXACT;
13493 bool next_is_quantifier;
13494 char * oldp = NULL;
13496 /* We can convert EXACTF nodes to EXACTFU if they contain only
13497 * characters that match identically regardless of the target
13498 * string's UTF8ness. The reason to do this is that EXACTF is not
13499 * trie-able, EXACTFU is.
13501 * Similarly, we can convert EXACTFL nodes to EXACTFLU8 if they
13502 * contain only above-Latin1 characters (hence must be in UTF8),
13503 * which don't participate in folds with Latin1-range characters,
13504 * as the latter's folds aren't known until runtime. (We don't
13505 * need to figure this out until pass 2) */
13506 bool maybe_exactfu = PASS2;
13508 /* The node_type may change below, but since the size of the node
13509 * doesn't change, it works */
13510 ret = reg_node(pRExC_state, node_type);
13512 /* In pass1, folded, we use a temporary buffer instead of the
13513 * actual node, as the node doesn't exist yet */
13514 s = (SIZE_ONLY && FOLD) ? foldbuf : STRING(ret);
13520 /* This breaks under rare circumstances. If folding, we do not
13521 * want to split a node at a character that is a non-final in a
13522 * multi-char fold, as an input string could just happen to want to
13523 * match across the node boundary. The code at the end of the loop
13524 * looks for this, and backs off until it finds not such a
13525 * character, but it is possible (though extremely, extremely
13526 * unlikely) for all characters in the node to be non-final fold
13527 * ones, in which case we just leave the node fully filled, and
13528 * hope that it doesn't match the string in just the wrong place */
13530 assert( ! UTF /* Is at the beginning of a character */
13531 || UTF8_IS_INVARIANT(UCHARAT(RExC_parse))
13532 || UTF8_IS_START(UCHARAT(RExC_parse)));
13534 /* Here, we have a literal character. Find the maximal string of
13535 * them in the input that we can fit into a single EXACTish node.
13536 * We quit at the first non-literal or when the node gets full, or
13537 * under /i the categorization of folding/non-folding character
13539 for (p = RExC_parse; len < upper_parse && p < RExC_end; ) {
13541 /* In most cases each iteration adds one byte to the output.
13542 * The exceptions override this */
13543 Size_t added_len = 1;
13547 /* White space has already been ignored */
13548 assert( (RExC_flags & RXf_PMf_EXTENDED) == 0
13549 || ! is_PATWS_safe((p), RExC_end, UTF));
13561 /* Literal Escapes Switch
13563 This switch is meant to handle escape sequences that
13564 resolve to a literal character.
13566 Every escape sequence that represents something
13567 else, like an assertion or a char class, is handled
13568 in the switch marked 'Special Escapes' above in this
13569 routine, but also has an entry here as anything that
13570 isn't explicitly mentioned here will be treated as
13571 an unescaped equivalent literal.
13574 switch ((U8)*++p) {
13575 /* These are all the special escapes. */
13576 case 'A': /* Start assertion */
13577 case 'b': case 'B': /* Word-boundary assertion*/
13578 case 'C': /* Single char !DANGEROUS! */
13579 case 'd': case 'D': /* digit class */
13580 case 'g': case 'G': /* generic-backref, pos assertion */
13581 case 'h': case 'H': /* HORIZWS */
13582 case 'k': case 'K': /* named backref, keep marker */
13583 case 'p': case 'P': /* Unicode property */
13584 case 'R': /* LNBREAK */
13585 case 's': case 'S': /* space class */
13586 case 'v': case 'V': /* VERTWS */
13587 case 'w': case 'W': /* word class */
13588 case 'X': /* eXtended Unicode "combining
13589 character sequence" */
13590 case 'z': case 'Z': /* End of line/string assertion */
13594 /* Anything after here is an escape that resolves to a
13595 literal. (Except digits, which may or may not)
13601 case 'N': /* Handle a single-code point named character. */
13602 RExC_parse = p + 1;
13603 if (! grok_bslash_N(pRExC_state,
13604 NULL, /* Fail if evaluates to
13605 anything other than a
13606 single code point */
13607 &ender, /* The returned single code
13609 NULL, /* Don't need a count of
13610 how many code points */
13615 if (*flagp & NEED_UTF8)
13616 FAIL("panic: grok_bslash_N set NEED_UTF8");
13617 RETURN_NULL_ON_RESTART_FLAGP(flagp);
13619 /* Here, it wasn't a single code point. Go close
13620 * up this EXACTish node. The switch() prior to
13621 * this switch handles the other cases */
13622 RExC_parse = p = oldp;
13626 RExC_parse = parse_start;
13627 if (ender > 0xff) {
13628 REQUIRE_UTF8(flagp);
13644 ender = ESC_NATIVE;
13654 const char* error_msg;
13656 bool valid = grok_bslash_o(&p,
13660 PASS2, /* out warnings */
13661 (bool) RExC_strict,
13662 TRUE, /* Output warnings
13667 RExC_parse = p; /* going to die anyway; point
13668 to exact spot of failure */
13672 if (ender > 0xff) {
13673 REQUIRE_UTF8(flagp);
13679 UV result = UV_MAX; /* initialize to erroneous
13681 const char* error_msg;
13683 bool valid = grok_bslash_x(&p,
13687 PASS2, /* out warnings */
13688 (bool) RExC_strict,
13689 TRUE, /* Silence warnings
13694 RExC_parse = p; /* going to die anyway; point
13695 to exact spot of failure */
13700 if (ender < 0x100) {
13702 if (RExC_recode_x_to_native) {
13703 ender = LATIN1_TO_NATIVE(ender);
13708 REQUIRE_UTF8(flagp);
13714 ender = grok_bslash_c(*p++, PASS2);
13716 case '8': case '9': /* must be a backreference */
13718 /* we have an escape like \8 which cannot be an octal escape
13719 * so we exit the loop, and let the outer loop handle this
13720 * escape which may or may not be a legitimate backref. */
13722 case '1': case '2': case '3':case '4':
13723 case '5': case '6': case '7':
13724 /* When we parse backslash escapes there is ambiguity
13725 * between backreferences and octal escapes. Any escape
13726 * from \1 - \9 is a backreference, any multi-digit
13727 * escape which does not start with 0 and which when
13728 * evaluated as decimal could refer to an already
13729 * parsed capture buffer is a back reference. Anything
13732 * Note this implies that \118 could be interpreted as
13733 * 118 OR as "\11" . "8" depending on whether there
13734 * were 118 capture buffers defined already in the
13737 /* NOTE, RExC_npar is 1 more than the actual number of
13738 * parens we have seen so far, hence the < RExC_npar below. */
13740 if ( !isDIGIT(p[1]) || S_backref_value(p) < RExC_npar)
13741 { /* Not to be treated as an octal constant, go
13749 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
13751 ender = grok_oct(p, &numlen, &flags, NULL);
13752 if (ender > 0xff) {
13753 REQUIRE_UTF8(flagp);
13756 if (PASS2 /* like \08, \178 */
13758 && isDIGIT(*p) && ckWARN(WARN_REGEXP))
13760 reg_warn_non_literal_string(
13762 form_short_octal_warning(p, numlen));
13768 FAIL("Trailing \\");
13771 if (!SIZE_ONLY&& isALPHANUMERIC(*p)) {
13772 /* Include any left brace following the alpha to emphasize
13773 * that it could be part of an escape at some point
13775 int len = (isALPHA(*p) && *(p + 1) == '{') ? 2 : 1;
13776 ckWARN3reg(p + len, "Unrecognized escape \\%.*s passed through", len, p);
13778 goto normal_default;
13779 } /* End of switch on '\' */
13782 /* Currently we allow an lbrace at the start of a construct
13783 * without raising a warning. This is because we think we
13784 * will never want such a brace to be meant to be other
13785 * than taken literally. */
13786 if (len || (p > RExC_start && isALPHA_A(*(p - 1)))) {
13788 /* But, we raise a fatal warning otherwise, as the
13789 * deprecation cycle has come and gone. Except that it
13790 * turns out that some heavily-relied on upstream
13791 * software, notably GNU Autoconf, have failed to fix
13792 * their uses. For these, don't make it fatal unless
13793 * we anticipate using the '{' for something else.
13794 * This happens after any alpha, and for a looser {m,n}
13795 * quantifier specification */
13797 || ( p > parse_start + 1
13798 && isALPHA_A(*(p - 1))
13799 && *(p - 2) == '\\')
13800 || new_regcurly(p, RExC_end))
13802 RExC_parse = p + 1;
13803 vFAIL("Unescaped left brace in regex is "
13807 ckWARNregdep(p + 1,
13808 "Unescaped left brace in regex is "
13809 "deprecated here (and will be fatal "
13810 "in Perl 5.30), passed through");
13813 goto normal_default;
13816 if (PASS2 && p > RExC_parse && RExC_strict) {
13817 ckWARN2reg(p + 1, "Unescaped literal '%c'", *p);
13820 default: /* A literal character */
13822 if (! UTF8_IS_INVARIANT(*p) && UTF) {
13824 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
13825 &numlen, UTF8_ALLOW_DEFAULT);
13831 } /* End of switch on the literal */
13833 /* Here, have looked at the literal character, and <ender>
13834 * contains its ordinal; <p> points to the character after it.
13835 * We need to check if the next non-ignored thing is a
13836 * quantifier. Move <p> to after anything that should be
13837 * ignored, which, as a side effect, positions <p> for the next
13838 * loop iteration */
13839 skip_to_be_ignored_text(pRExC_state, &p,
13840 FALSE /* Don't force to /x */ );
13842 /* If the next thing is a quantifier, it applies to this
13843 * character only, which means that this character has to be in
13844 * its own node and can't just be appended to the string in an
13845 * existing node, so if there are already other characters in
13846 * the node, close the node with just them, and set up to do
13847 * this character again next time through, when it will be the
13848 * only thing in its new node */
13850 next_is_quantifier = LIKELY(p < RExC_end)
13851 && UNLIKELY(ISMULT2(p));
13853 if (next_is_quantifier && LIKELY(len)) {
13858 /* Ready to add 'ender' to the node */
13860 if (! FOLD) { /* The simple case, just append the literal */
13862 /* In the sizing pass, we need only the size of the
13863 * character we are appending, hence we can delay getting
13864 * its representation until PASS2. */
13866 if (UTF && ! UVCHR_IS_INVARIANT(ender)) {
13867 const STRLEN unilen = UVCHR_SKIP(ender);
13869 added_len = unilen;
13874 } else { /* PASS2 */
13876 if (UTF && ! UVCHR_IS_INVARIANT(ender)) {
13877 U8 * new_s = uvchr_to_utf8((U8*)s, ender);
13878 added_len = (char *) new_s - s;
13879 s = (char *) new_s;
13882 *(s++) = (char) ender;
13886 else if (LOC && is_PROBLEMATIC_LOCALE_FOLD_cp(ender)) {
13888 /* Here are folding under /l, and the code point is
13889 * problematic. If this is the first character in the
13890 * node, change the node type to folding. Otherwise, if
13891 * this is the first problematic character, close up the
13892 * existing node, so can start a new node with this one */
13894 node_type = EXACTFL;
13896 else if (node_type == EXACT) {
13901 /* This code point means we can't simplify things */
13902 maybe_exactfu = FALSE;
13904 /* A problematic code point in this context means that its
13905 * fold isn't known until runtime, so we can't fold it now.
13906 * (The non-problematic code points are the above-Latin1
13907 * ones that fold to also all above-Latin1. Their folds
13908 * don't vary no matter what the locale is.) But here we
13909 * have characters whose fold depends on the locale.
13910 * Unlike the non-folding case above, we have to keep track
13911 * of these in the sizing pass, so that we can make sure we
13912 * don't split too-long nodes in the middle of a potential
13913 * multi-char fold. And unlike the regular fold case
13914 * handled in the else clauses below, we don't actually
13915 * fold and don't have special cases to consider. What we
13916 * do for both passes is the PASS2 code for non-folding */
13917 goto not_fold_common;
13919 else /* A regular FOLD code point */
13922 /* Here, are folding and are not UTF-8 encoded; therefore
13923 * the character must be in the range 0-255, and is not /l.
13924 * (Not /l because we already handled these under /l in
13925 * is_PROBLEMATIC_LOCALE_FOLD_cp) */
13926 if (! IS_IN_SOME_FOLD_L1(ender)) {
13928 /* Start a new node for this non-folding character if
13929 * previous ones in the node were folded */
13930 if (len && node_type != EXACT) {
13935 *(s++) = (char) ender;
13937 else { /* Here, does participate in some fold */
13939 /* if this is the first character in the node, change
13940 * its type to folding. Otherwise, if this is the
13941 * first folding character in the node, close up the
13942 * existing node, so can start a new node with this
13945 node_type = compute_EXACTish(pRExC_state);
13947 else if (node_type == EXACT) {
13952 /* See if the character's fold differs between /d and
13953 * /u. On non-ancient Unicode versions, this includes
13954 * the multi-char fold SHARP S to 'ss' */
13956 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
13957 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
13958 || UNICODE_DOT_DOT_VERSION > 0)
13960 if (UNLIKELY(ender == LATIN_SMALL_LETTER_SHARP_S)) {
13962 /* See comments for join_exact() as to why we fold
13963 * this non-UTF at compile time */
13964 if (node_type == EXACTFU) {
13967 /* Let the code below add in the extra 's' */
13972 RExC_seen_unfolded_sharp_s = 1;
13973 maybe_exactfu = FALSE;
13977 && isALPHA_FOLD_EQ(ender, 's')
13978 && isALPHA_FOLD_EQ(*(s-1), 's'))
13980 maybe_exactfu = FALSE;
13985 if (PL_fold[ender] != PL_fold_latin1[ender]) {
13986 maybe_exactfu = FALSE;
13989 /* Even when folding, we store just the input
13990 * character, as we have an array that finds its fold
13992 *(s++) = (char) ender;
13995 else { /* FOLD, and UTF */
13996 /* Unlike the non-fold case, we do actually have to
13997 * calculate the fold in pass 1. This is for two reasons,
13998 * the folded length may be longer than the unfolded, and
13999 * we have to calculate how many EXACTish nodes it will
14000 * take; and we may run out of room in a node in the middle
14001 * of a potential multi-char fold, and have to back off
14004 if (isASCII_uni(ender)) {
14006 /* As above, we close up and start a new node if the
14007 * previous characters don't match the fold/non-fold
14008 * state of this one. And if this is the first
14009 * character in the node, and it folds, we change the
14010 * node away from being EXACT */
14011 if (! IS_IN_SOME_FOLD_L1(ender)) {
14012 if (len && node_type != EXACT) {
14017 *(s)++ = (U8) ender;
14019 else { /* Is in a fold */
14022 node_type = compute_EXACTish(pRExC_state);
14024 else if (node_type == EXACT) {
14029 *(s)++ = (U8) toFOLD(ender);
14032 else { /* Not ASCII */
14035 /* As above, we close up and start a new node if the
14036 * previous characters don't match the fold/non-fold
14037 * state of this one. And if this is the first
14038 * character in the node, and it folds, we change the
14039 * node away from being EXACT */
14040 if (! _invlist_contains_cp(PL_utf8_foldable, ender)) {
14041 if (len && node_type != EXACT) {
14046 s = (char *) uvchr_to_utf8((U8 *) s, ender);
14047 added_len = UVCHR_SKIP(ender);
14052 node_type = compute_EXACTish(pRExC_state);
14054 else if (node_type == EXACT) {
14059 ender = _to_uni_fold_flags(
14063 FOLD_FLAGS_FULL | ((ASCII_FOLD_RESTRICTED)
14064 ? FOLD_FLAGS_NOMIX_ASCII
14067 added_len = foldlen;
14074 if (next_is_quantifier) {
14076 /* Here, the next input is a quantifier, and to get here,
14077 * the current character is the only one in the node. */
14081 } /* End of loop through literal characters */
14083 /* Here we have either exhausted the input or ran out of room in
14084 * the node. (If we encountered a character that can't be in the
14085 * node, transfer is made directly to <loopdone>, and so we
14086 * wouldn't have fallen off the end of the loop.) In the latter
14087 * case, we artificially have to split the node into two, because
14088 * we just don't have enough space to hold everything. This
14089 * creates a problem if the final character participates in a
14090 * multi-character fold in the non-final position, as a match that
14091 * should have occurred won't, due to the way nodes are matched,
14092 * and our artificial boundary. So back off until we find a non-
14093 * problematic character -- one that isn't at the beginning or
14094 * middle of such a fold. (Either it doesn't participate in any
14095 * folds, or appears only in the final position of all the folds it
14096 * does participate in.) A better solution with far fewer false
14097 * positives, and that would fill the nodes more completely, would
14098 * be to actually have available all the multi-character folds to
14099 * test against, and to back-off only far enough to be sure that
14100 * this node isn't ending with a partial one. <upper_parse> is set
14101 * further below (if we need to reparse the node) to include just
14102 * up through that final non-problematic character that this code
14103 * identifies, so when it is set to less than the full node, we can
14104 * skip the rest of this */
14105 if (FOLD && p < RExC_end && upper_parse == MAX_NODE_STRING_SIZE) {
14107 const STRLEN full_len = len;
14109 assert(len >= MAX_NODE_STRING_SIZE);
14111 /* Here, <s> points to the final byte of the final character.
14112 * Look backwards through the string until find a non-
14113 * problematic character */
14117 /* This has no multi-char folds to non-UTF characters */
14118 if (ASCII_FOLD_RESTRICTED) {
14122 while (--s >= s0 && IS_NON_FINAL_FOLD(*s)) { }
14126 if (! PL_NonL1NonFinalFold) {
14127 PL_NonL1NonFinalFold = _new_invlist_C_array(
14128 NonL1_Perl_Non_Final_Folds_invlist);
14131 /* Point to the first byte of the final character */
14132 s = (char *) utf8_hop((U8 *) s, -1);
14134 while (s >= s0) { /* Search backwards until find
14135 a non-problematic char */
14136 if (UTF8_IS_INVARIANT(*s)) {
14138 /* There are no ascii characters that participate
14139 * in multi-char folds under /aa. In EBCDIC, the
14140 * non-ascii invariants are all control characters,
14141 * so don't ever participate in any folds. */
14142 if (ASCII_FOLD_RESTRICTED
14143 || ! IS_NON_FINAL_FOLD(*s))
14148 else if (UTF8_IS_DOWNGRADEABLE_START(*s)) {
14149 if (! IS_NON_FINAL_FOLD(EIGHT_BIT_UTF8_TO_NATIVE(
14155 else if (! _invlist_contains_cp(
14156 PL_NonL1NonFinalFold,
14157 valid_utf8_to_uvchr((U8 *) s, NULL)))
14162 /* Here, the current character is problematic in that
14163 * it does occur in the non-final position of some
14164 * fold, so try the character before it, but have to
14165 * special case the very first byte in the string, so
14166 * we don't read outside the string */
14167 s = (s == s0) ? s -1 : (char *) utf8_hop((U8 *) s, -1);
14168 } /* End of loop backwards through the string */
14170 /* If there were only problematic characters in the string,
14171 * <s> will point to before s0, in which case the length
14172 * should be 0, otherwise include the length of the
14173 * non-problematic character just found */
14174 len = (s < s0) ? 0 : s - s0 + UTF8SKIP(s);
14177 /* Here, have found the final character, if any, that is
14178 * non-problematic as far as ending the node without splitting
14179 * it across a potential multi-char fold. <len> contains the
14180 * number of bytes in the node up-to and including that
14181 * character, or is 0 if there is no such character, meaning
14182 * the whole node contains only problematic characters. In
14183 * this case, give up and just take the node as-is. We can't
14188 /* If the node ends in an 's' we make sure it stays EXACTF,
14189 * as if it turns into an EXACTFU, it could later get
14190 * joined with another 's' that would then wrongly match
14192 if (maybe_exactfu && isALPHA_FOLD_EQ(ender, 's'))
14194 maybe_exactfu = FALSE;
14198 /* Here, the node does contain some characters that aren't
14199 * problematic. If one such is the final character in the
14200 * node, we are done */
14201 if (len == full_len) {
14204 else if (len + ((UTF) ? UTF8SKIP(s) : 1) == full_len) {
14206 /* If the final character is problematic, but the
14207 * penultimate is not, back-off that last character to
14208 * later start a new node with it */
14213 /* Here, the final non-problematic character is earlier
14214 * in the input than the penultimate character. What we do
14215 * is reparse from the beginning, going up only as far as
14216 * this final ok one, thus guaranteeing that the node ends
14217 * in an acceptable character. The reason we reparse is
14218 * that we know how far in the character is, but we don't
14219 * know how to correlate its position with the input parse.
14220 * An alternate implementation would be to build that
14221 * correlation as we go along during the original parse,
14222 * but that would entail extra work for every node, whereas
14223 * this code gets executed only when the string is too
14224 * large for the node, and the final two characters are
14225 * problematic, an infrequent occurrence. Yet another
14226 * possible strategy would be to save the tail of the
14227 * string, and the next time regatom is called, initialize
14228 * with that. The problem with this is that unless you
14229 * back off one more character, you won't be guaranteed
14230 * regatom will get called again, unless regbranch,
14231 * regpiece ... are also changed. If you do back off that
14232 * extra character, so that there is input guaranteed to
14233 * force calling regatom, you can't handle the case where
14234 * just the first character in the node is acceptable. I
14235 * (khw) decided to try this method which doesn't have that
14236 * pitfall; if performance issues are found, we can do a
14237 * combination of the current approach plus that one */
14243 } /* End of verifying node ends with an appropriate char */
14245 loopdone: /* Jumped to when encounters something that shouldn't be
14248 /* I (khw) don't know if you can get here with zero length, but the
14249 * old code handled this situation by creating a zero-length EXACT
14250 * node. Might as well be NOTHING instead */
14255 OP(ret) = node_type;
14257 /* If the node type is EXACT here, check to see if it
14258 * should be EXACTL. */
14259 if (node_type == EXACT) {
14266 /* If 'maybe_exactfu' is set, then there are no code points
14267 * that match differently depending on UTF8ness of the
14268 * target string (for /u), or depending on locale for /l */
14269 if (maybe_exactfu) {
14270 if (node_type == EXACTF) {
14273 else if (node_type == EXACTFL) {
14274 OP(ret) = EXACTFLU8;
14279 alloc_maybe_populate_EXACT(pRExC_state, ret, flagp, len, ender,
14280 FALSE /* Don't look to see if could
14281 be turned into an EXACT
14282 node, as we have already
14287 RExC_parse = p - 1;
14288 Set_Node_Cur_Length(ret, parse_start);
14291 /* len is STRLEN which is unsigned, need to copy to signed */
14294 vFAIL("Internal disaster");
14297 } /* End of label 'defchar:' */
14299 } /* End of giant switch on input character */
14301 /* Position parse to next real character */
14302 skip_to_be_ignored_text(pRExC_state, &RExC_parse,
14303 FALSE /* Don't force to /x */ );
14304 if (PASS2 && *RExC_parse == '{' && OP(ret) != SBOL && ! regcurly(RExC_parse)) {
14305 ckWARNregdep(RExC_parse + 1, "Unescaped left brace in regex is deprecated here (and will be fatal in Perl 5.30), passed through");
14313 S_populate_ANYOF_from_invlist(pTHX_ regnode *node, SV** invlist_ptr)
14315 /* Uses the inversion list '*invlist_ptr' to populate the ANYOF 'node'. It
14316 * sets up the bitmap and any flags, removing those code points from the
14317 * inversion list, setting it to NULL should it become completely empty */
14319 PERL_ARGS_ASSERT_POPULATE_ANYOF_FROM_INVLIST;
14320 assert(PL_regkind[OP(node)] == ANYOF);
14322 ANYOF_BITMAP_ZERO(node);
14323 if (*invlist_ptr) {
14325 /* This gets set if we actually need to modify things */
14326 bool change_invlist = FALSE;
14330 /* Start looking through *invlist_ptr */
14331 invlist_iterinit(*invlist_ptr);
14332 while (invlist_iternext(*invlist_ptr, &start, &end)) {
14336 if (end == UV_MAX && start <= NUM_ANYOF_CODE_POINTS) {
14337 ANYOF_FLAGS(node) |= ANYOF_MATCHES_ALL_ABOVE_BITMAP;
14340 /* Quit if are above what we should change */
14341 if (start >= NUM_ANYOF_CODE_POINTS) {
14345 change_invlist = TRUE;
14347 /* Set all the bits in the range, up to the max that we are doing */
14348 high = (end < NUM_ANYOF_CODE_POINTS - 1)
14350 : NUM_ANYOF_CODE_POINTS - 1;
14351 for (i = start; i <= (int) high; i++) {
14352 if (! ANYOF_BITMAP_TEST(node, i)) {
14353 ANYOF_BITMAP_SET(node, i);
14357 invlist_iterfinish(*invlist_ptr);
14359 /* Done with loop; remove any code points that are in the bitmap from
14360 * *invlist_ptr; similarly for code points above the bitmap if we have
14361 * a flag to match all of them anyways */
14362 if (change_invlist) {
14363 _invlist_subtract(*invlist_ptr, PL_InBitmap, invlist_ptr);
14365 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
14366 _invlist_intersection(*invlist_ptr, PL_InBitmap, invlist_ptr);
14369 /* If have completely emptied it, remove it completely */
14370 if (_invlist_len(*invlist_ptr) == 0) {
14371 SvREFCNT_dec_NN(*invlist_ptr);
14372 *invlist_ptr = NULL;
14377 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
14378 Character classes ([:foo:]) can also be negated ([:^foo:]).
14379 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
14380 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
14381 but trigger failures because they are currently unimplemented. */
14383 #define POSIXCC_DONE(c) ((c) == ':')
14384 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
14385 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
14386 #define MAYBE_POSIXCC(c) (POSIXCC(c) || (c) == '^' || (c) == ';')
14388 #define WARNING_PREFIX "Assuming NOT a POSIX class since "
14389 #define NO_BLANKS_POSIX_WARNING "no blanks are allowed in one"
14390 #define SEMI_COLON_POSIX_WARNING "a semi-colon was found instead of a colon"
14392 #define NOT_MEANT_TO_BE_A_POSIX_CLASS (OOB_NAMEDCLASS - 1)
14394 /* 'posix_warnings' and 'warn_text' are names of variables in the following
14396 #define ADD_POSIX_WARNING(p, text) STMT_START { \
14397 if (posix_warnings) { \
14398 if (! RExC_warn_text ) RExC_warn_text = (AV *) sv_2mortal((SV *) newAV()); \
14399 av_push(RExC_warn_text, Perl_newSVpvf(aTHX_ \
14403 REPORT_LOCATION_ARGS(p))); \
14406 #define CLEAR_POSIX_WARNINGS() \
14408 if (posix_warnings && RExC_warn_text) \
14409 av_clear(RExC_warn_text); \
14412 #define CLEAR_POSIX_WARNINGS_AND_RETURN(ret) \
14414 CLEAR_POSIX_WARNINGS(); \
14419 S_handle_possible_posix(pTHX_ RExC_state_t *pRExC_state,
14421 const char * const s, /* Where the putative posix class begins.
14422 Normally, this is one past the '['. This
14423 parameter exists so it can be somewhere
14424 besides RExC_parse. */
14425 char ** updated_parse_ptr, /* Where to set the updated parse pointer, or
14427 AV ** posix_warnings, /* Where to place any generated warnings, or
14429 const bool check_only /* Don't die if error */
14432 /* This parses what the caller thinks may be one of the three POSIX
14434 * 1) a character class, like [:blank:]
14435 * 2) a collating symbol, like [. .]
14436 * 3) an equivalence class, like [= =]
14437 * In the latter two cases, it croaks if it finds a syntactically legal
14438 * one, as these are not handled by Perl.
14440 * The main purpose is to look for a POSIX character class. It returns:
14441 * a) the class number
14442 * if it is a completely syntactically and semantically legal class.
14443 * 'updated_parse_ptr', if not NULL, is set to point to just after the
14444 * closing ']' of the class
14445 * b) OOB_NAMEDCLASS
14446 * if it appears that one of the three POSIX constructs was meant, but
14447 * its specification was somehow defective. 'updated_parse_ptr', if
14448 * not NULL, is set to point to the character just after the end
14449 * character of the class. See below for handling of warnings.
14450 * c) NOT_MEANT_TO_BE_A_POSIX_CLASS
14451 * if it doesn't appear that a POSIX construct was intended.
14452 * 'updated_parse_ptr' is not changed. No warnings nor errors are
14455 * In b) there may be errors or warnings generated. If 'check_only' is
14456 * TRUE, then any errors are discarded. Warnings are returned to the
14457 * caller via an AV* created into '*posix_warnings' if it is not NULL. If
14458 * instead it is NULL, warnings are suppressed. This is done in all
14459 * passes. The reason for this is that the rest of the parsing is heavily
14460 * dependent on whether this routine found a valid posix class or not. If
14461 * it did, the closing ']' is absorbed as part of the class. If no class,
14462 * or an invalid one is found, any ']' will be considered the terminator of
14463 * the outer bracketed character class, leading to very different results.
14464 * In particular, a '(?[ ])' construct will likely have a syntax error if
14465 * the class is parsed other than intended, and this will happen in pass1,
14466 * before the warnings would normally be output. This mechanism allows the
14467 * caller to output those warnings in pass1 just before dieing, giving a
14468 * much better clue as to what is wrong.
14470 * The reason for this function, and its complexity is that a bracketed
14471 * character class can contain just about anything. But it's easy to
14472 * mistype the very specific posix class syntax but yielding a valid
14473 * regular bracketed class, so it silently gets compiled into something
14474 * quite unintended.
14476 * The solution adopted here maintains backward compatibility except that
14477 * it adds a warning if it looks like a posix class was intended but
14478 * improperly specified. The warning is not raised unless what is input
14479 * very closely resembles one of the 14 legal posix classes. To do this,
14480 * it uses fuzzy parsing. It calculates how many single-character edits it
14481 * would take to transform what was input into a legal posix class. Only
14482 * if that number is quite small does it think that the intention was a
14483 * posix class. Obviously these are heuristics, and there will be cases
14484 * where it errs on one side or another, and they can be tweaked as
14485 * experience informs.
14487 * The syntax for a legal posix class is:
14489 * qr/(?xa: \[ : \^? [[:lower:]]{4,6} : \] )/
14491 * What this routine considers syntactically to be an intended posix class
14492 * is this (the comments indicate some restrictions that the pattern
14495 * qr/(?x: \[? # The left bracket, possibly
14497 * \h* # possibly followed by blanks
14498 * (?: \^ \h* )? # possibly a misplaced caret
14499 * [:;]? # The opening class character,
14500 * # possibly omitted. A typo
14501 * # semi-colon can also be used.
14503 * \^? # possibly a correctly placed
14504 * # caret, but not if there was also
14505 * # a misplaced one
14507 * .{3,15} # The class name. If there are
14508 * # deviations from the legal syntax,
14509 * # its edit distance must be close
14510 * # to a real class name in order
14511 * # for it to be considered to be
14512 * # an intended posix class.
14514 * [[:punct:]]? # The closing class character,
14515 * # possibly omitted. If not a colon
14516 * # nor semi colon, the class name
14517 * # must be even closer to a valid
14520 * \]? # The right bracket, possibly
14524 * In the above, \h must be ASCII-only.
14526 * These are heuristics, and can be tweaked as field experience dictates.
14527 * There will be cases when someone didn't intend to specify a posix class
14528 * that this warns as being so. The goal is to minimize these, while
14529 * maximizing the catching of things intended to be a posix class that
14530 * aren't parsed as such.
14534 const char * const e = RExC_end;
14535 unsigned complement = 0; /* If to complement the class */
14536 bool found_problem = FALSE; /* Assume OK until proven otherwise */
14537 bool has_opening_bracket = FALSE;
14538 bool has_opening_colon = FALSE;
14539 int class_number = OOB_NAMEDCLASS; /* Out-of-bounds until find
14541 const char * possible_end = NULL; /* used for a 2nd parse pass */
14542 const char* name_start; /* ptr to class name first char */
14544 /* If the number of single-character typos the input name is away from a
14545 * legal name is no more than this number, it is considered to have meant
14546 * the legal name */
14547 int max_distance = 2;
14549 /* to store the name. The size determines the maximum length before we
14550 * decide that no posix class was intended. Should be at least
14551 * sizeof("alphanumeric") */
14553 STATIC_ASSERT_DECL(C_ARRAY_LENGTH(input_text) >= sizeof "alphanumeric");
14555 PERL_ARGS_ASSERT_HANDLE_POSSIBLE_POSIX;
14557 CLEAR_POSIX_WARNINGS();
14560 return NOT_MEANT_TO_BE_A_POSIX_CLASS;
14563 if (*(p - 1) != '[') {
14564 ADD_POSIX_WARNING(p, "it doesn't start with a '['");
14565 found_problem = TRUE;
14568 has_opening_bracket = TRUE;
14571 /* They could be confused and think you can put spaces between the
14574 found_problem = TRUE;
14578 } while (p < e && isBLANK(*p));
14580 ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING);
14583 /* For [. .] and [= =]. These are quite different internally from [: :],
14584 * so they are handled separately. */
14585 if (POSIXCC_NOTYET(*p) && p < e - 3) /* 1 for the close, and 1 for the ']'
14586 and 1 for at least one char in it
14589 const char open_char = *p;
14590 const char * temp_ptr = p + 1;
14592 /* These two constructs are not handled by perl, and if we find a
14593 * syntactically valid one, we croak. khw, who wrote this code, finds
14594 * this explanation of them very unclear:
14595 * http://pubs.opengroup.org/onlinepubs/009696899/basedefs/xbd_chap09.html
14596 * And searching the rest of the internet wasn't very helpful either.
14597 * It looks like just about any byte can be in these constructs,
14598 * depending on the locale. But unless the pattern is being compiled
14599 * under /l, which is very rare, Perl runs under the C or POSIX locale.
14600 * In that case, it looks like [= =] isn't allowed at all, and that
14601 * [. .] could be any single code point, but for longer strings the
14602 * constituent characters would have to be the ASCII alphabetics plus
14603 * the minus-hyphen. Any sensible locale definition would limit itself
14604 * to these. And any portable one definitely should. Trying to parse
14605 * the general case is a nightmare (see [perl #127604]). So, this code
14606 * looks only for interiors of these constructs that match:
14608 * Using \w relaxes the apparent rules a little, without adding much
14609 * danger of mistaking something else for one of these constructs.
14611 * [. .] in some implementations described on the internet is usable to
14612 * escape a character that otherwise is special in bracketed character
14613 * classes. For example [.].] means a literal right bracket instead of
14614 * the ending of the class
14616 * [= =] can legitimately contain a [. .] construct, but we don't
14617 * handle this case, as that [. .] construct will later get parsed
14618 * itself and croak then. And [= =] is checked for even when not under
14619 * /l, as Perl has long done so.
14621 * The code below relies on there being a trailing NUL, so it doesn't
14622 * have to keep checking if the parse ptr < e.
14624 if (temp_ptr[1] == open_char) {
14627 else while ( temp_ptr < e
14628 && (isWORDCHAR(*temp_ptr) || *temp_ptr == '-'))
14633 if (*temp_ptr == open_char) {
14635 if (*temp_ptr == ']') {
14637 if (! found_problem && ! check_only) {
14638 RExC_parse = (char *) temp_ptr;
14639 vFAIL3("POSIX syntax [%c %c] is reserved for future "
14640 "extensions", open_char, open_char);
14643 /* Here, the syntax wasn't completely valid, or else the call
14644 * is to check-only */
14645 if (updated_parse_ptr) {
14646 *updated_parse_ptr = (char *) temp_ptr;
14649 CLEAR_POSIX_WARNINGS_AND_RETURN(OOB_NAMEDCLASS);
14653 /* If we find something that started out to look like one of these
14654 * constructs, but isn't, we continue below so that it can be checked
14655 * for being a class name with a typo of '.' or '=' instead of a colon.
14659 /* Here, we think there is a possibility that a [: :] class was meant, and
14660 * we have the first real character. It could be they think the '^' comes
14663 found_problem = TRUE;
14664 ADD_POSIX_WARNING(p + 1, "the '^' must come after the colon");
14669 found_problem = TRUE;
14673 } while (p < e && isBLANK(*p));
14675 ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING);
14679 /* But the first character should be a colon, which they could have easily
14680 * mistyped on a qwerty keyboard as a semi-colon (and which may be hard to
14681 * distinguish from a colon, so treat that as a colon). */
14684 has_opening_colon = TRUE;
14686 else if (*p == ';') {
14687 found_problem = TRUE;
14689 ADD_POSIX_WARNING(p, SEMI_COLON_POSIX_WARNING);
14690 has_opening_colon = TRUE;
14693 found_problem = TRUE;
14694 ADD_POSIX_WARNING(p, "there must be a starting ':'");
14696 /* Consider an initial punctuation (not one of the recognized ones) to
14697 * be a left terminator */
14698 if (*p != '^' && *p != ']' && isPUNCT(*p)) {
14703 /* They may think that you can put spaces between the components */
14705 found_problem = TRUE;
14709 } while (p < e && isBLANK(*p));
14711 ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING);
14716 /* We consider something like [^:^alnum:]] to not have been intended to
14717 * be a posix class, but XXX maybe we should */
14719 CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS);
14726 /* Again, they may think that you can put spaces between the components */
14728 found_problem = TRUE;
14732 } while (p < e && isBLANK(*p));
14734 ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING);
14739 /* XXX This ']' may be a typo, and something else was meant. But
14740 * treating it as such creates enough complications, that that
14741 * possibility isn't currently considered here. So we assume that the
14742 * ']' is what is intended, and if we've already found an initial '[',
14743 * this leaves this construct looking like [:] or [:^], which almost
14744 * certainly weren't intended to be posix classes */
14745 if (has_opening_bracket) {
14746 CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS);
14749 /* But this function can be called when we parse the colon for
14750 * something like qr/[alpha:]]/, so we back up to look for the
14755 found_problem = TRUE;
14756 ADD_POSIX_WARNING(p, SEMI_COLON_POSIX_WARNING);
14758 else if (*p != ':') {
14760 /* XXX We are currently very restrictive here, so this code doesn't
14761 * consider the possibility that, say, /[alpha.]]/ was intended to
14762 * be a posix class. */
14763 CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS);
14766 /* Here we have something like 'foo:]'. There was no initial colon,
14767 * and we back up over 'foo. XXX Unlike the going forward case, we
14768 * don't handle typos of non-word chars in the middle */
14769 has_opening_colon = FALSE;
14772 while (p > RExC_start && isWORDCHAR(*p)) {
14777 /* Here, we have positioned ourselves to where we think the first
14778 * character in the potential class is */
14781 /* Now the interior really starts. There are certain key characters that
14782 * can end the interior, or these could just be typos. To catch both
14783 * cases, we may have to do two passes. In the first pass, we keep on
14784 * going unless we come to a sequence that matches
14785 * qr/ [[:punct:]] [[:blank:]]* \] /xa
14786 * This means it takes a sequence to end the pass, so two typos in a row if
14787 * that wasn't what was intended. If the class is perfectly formed, just
14788 * this one pass is needed. We also stop if there are too many characters
14789 * being accumulated, but this number is deliberately set higher than any
14790 * real class. It is set high enough so that someone who thinks that
14791 * 'alphanumeric' is a correct name would get warned that it wasn't.
14792 * While doing the pass, we keep track of where the key characters were in
14793 * it. If we don't find an end to the class, and one of the key characters
14794 * was found, we redo the pass, but stop when we get to that character.
14795 * Thus the key character was considered a typo in the first pass, but a
14796 * terminator in the second. If two key characters are found, we stop at
14797 * the second one in the first pass. Again this can miss two typos, but
14798 * catches a single one
14800 * In the first pass, 'possible_end' starts as NULL, and then gets set to
14801 * point to the first key character. For the second pass, it starts as -1.
14807 bool has_blank = FALSE;
14808 bool has_upper = FALSE;
14809 bool has_terminating_colon = FALSE;
14810 bool has_terminating_bracket = FALSE;
14811 bool has_semi_colon = FALSE;
14812 unsigned int name_len = 0;
14813 int punct_count = 0;
14817 /* Squeeze out blanks when looking up the class name below */
14818 if (isBLANK(*p) ) {
14820 found_problem = TRUE;
14825 /* The name will end with a punctuation */
14827 const char * peek = p + 1;
14829 /* Treat any non-']' punctuation followed by a ']' (possibly
14830 * with intervening blanks) as trying to terminate the class.
14831 * ']]' is very likely to mean a class was intended (but
14832 * missing the colon), but the warning message that gets
14833 * generated shows the error position better if we exit the
14834 * loop at the bottom (eventually), so skip it here. */
14836 if (peek < e && isBLANK(*peek)) {
14838 found_problem = TRUE;
14841 } while (peek < e && isBLANK(*peek));
14844 if (peek < e && *peek == ']') {
14845 has_terminating_bracket = TRUE;
14847 has_terminating_colon = TRUE;
14849 else if (*p == ';') {
14850 has_semi_colon = TRUE;
14851 has_terminating_colon = TRUE;
14854 found_problem = TRUE;
14861 /* Here we have punctuation we thought didn't end the class.
14862 * Keep track of the position of the key characters that are
14863 * more likely to have been class-enders */
14864 if (*p == ']' || *p == '[' || *p == ':' || *p == ';') {
14866 /* Allow just one such possible class-ender not actually
14867 * ending the class. */
14868 if (possible_end) {
14874 /* If we have too many punctuation characters, no use in
14876 if (++punct_count > max_distance) {
14880 /* Treat the punctuation as a typo. */
14881 input_text[name_len++] = *p;
14884 else if (isUPPER(*p)) { /* Use lowercase for lookup */
14885 input_text[name_len++] = toLOWER(*p);
14887 found_problem = TRUE;
14889 } else if (! UTF || UTF8_IS_INVARIANT(*p)) {
14890 input_text[name_len++] = *p;
14894 input_text[name_len++] = utf8_to_uvchr_buf((U8 *) p, e, NULL);
14898 /* The declaration of 'input_text' is how long we allow a potential
14899 * class name to be, before saying they didn't mean a class name at
14901 if (name_len >= C_ARRAY_LENGTH(input_text)) {
14906 /* We get to here when the possible class name hasn't been properly
14907 * terminated before:
14908 * 1) we ran off the end of the pattern; or
14909 * 2) found two characters, each of which might have been intended to
14910 * be the name's terminator
14911 * 3) found so many punctuation characters in the purported name,
14912 * that the edit distance to a valid one is exceeded
14913 * 4) we decided it was more characters than anyone could have
14914 * intended to be one. */
14916 found_problem = TRUE;
14918 /* In the final two cases, we know that looking up what we've
14919 * accumulated won't lead to a match, even a fuzzy one. */
14920 if ( name_len >= C_ARRAY_LENGTH(input_text)
14921 || punct_count > max_distance)
14923 /* If there was an intermediate key character that could have been
14924 * an intended end, redo the parse, but stop there */
14925 if (possible_end && possible_end != (char *) -1) {
14926 possible_end = (char *) -1; /* Special signal value to say
14927 we've done a first pass */
14932 /* Otherwise, it can't have meant to have been a class */
14933 CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS);
14936 /* If we ran off the end, and the final character was a punctuation
14937 * one, back up one, to look at that final one just below. Later, we
14938 * will restore the parse pointer if appropriate */
14939 if (name_len && p == e && isPUNCT(*(p-1))) {
14944 if (p < e && isPUNCT(*p)) {
14946 has_terminating_bracket = TRUE;
14948 /* If this is a 2nd ']', and the first one is just below this
14949 * one, consider that to be the real terminator. This gives a
14950 * uniform and better positioning for the warning message */
14952 && possible_end != (char *) -1
14953 && *possible_end == ']'
14954 && name_len && input_text[name_len - 1] == ']')
14959 /* And this is actually equivalent to having done the 2nd
14960 * pass now, so set it to not try again */
14961 possible_end = (char *) -1;
14966 has_terminating_colon = TRUE;
14968 else if (*p == ';') {
14969 has_semi_colon = TRUE;
14970 has_terminating_colon = TRUE;
14978 /* Here, we have a class name to look up. We can short circuit the
14979 * stuff below for short names that can't possibly be meant to be a
14980 * class name. (We can do this on the first pass, as any second pass
14981 * will yield an even shorter name) */
14982 if (name_len < 3) {
14983 CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS);
14986 /* Find which class it is. Initially switch on the length of the name.
14988 switch (name_len) {
14990 if (memEQs(name_start, 4, "word")) {
14991 /* this is not POSIX, this is the Perl \w */
14992 class_number = ANYOF_WORDCHAR;
14996 /* Names all of length 5: alnum alpha ascii blank cntrl digit
14997 * graph lower print punct space upper
14998 * Offset 4 gives the best switch position. */
14999 switch (name_start[4]) {
15001 if (memBEGINs(name_start, 5, "alph")) /* alpha */
15002 class_number = ANYOF_ALPHA;
15005 if (memBEGINs(name_start, 5, "spac")) /* space */
15006 class_number = ANYOF_SPACE;
15009 if (memBEGINs(name_start, 5, "grap")) /* graph */
15010 class_number = ANYOF_GRAPH;
15013 if (memBEGINs(name_start, 5, "asci")) /* ascii */
15014 class_number = ANYOF_ASCII;
15017 if (memBEGINs(name_start, 5, "blan")) /* blank */
15018 class_number = ANYOF_BLANK;
15021 if (memBEGINs(name_start, 5, "cntr")) /* cntrl */
15022 class_number = ANYOF_CNTRL;
15025 if (memBEGINs(name_start, 5, "alnu")) /* alnum */
15026 class_number = ANYOF_ALPHANUMERIC;
15029 if (memBEGINs(name_start, 5, "lowe")) /* lower */
15030 class_number = (FOLD) ? ANYOF_CASED : ANYOF_LOWER;
15031 else if (memBEGINs(name_start, 5, "uppe")) /* upper */
15032 class_number = (FOLD) ? ANYOF_CASED : ANYOF_UPPER;
15035 if (memBEGINs(name_start, 5, "digi")) /* digit */
15036 class_number = ANYOF_DIGIT;
15037 else if (memBEGINs(name_start, 5, "prin")) /* print */
15038 class_number = ANYOF_PRINT;
15039 else if (memBEGINs(name_start, 5, "punc")) /* punct */
15040 class_number = ANYOF_PUNCT;
15045 if (memEQs(name_start, 6, "xdigit"))
15046 class_number = ANYOF_XDIGIT;
15050 /* If the name exactly matches a posix class name the class number will
15051 * here be set to it, and the input almost certainly was meant to be a
15052 * posix class, so we can skip further checking. If instead the syntax
15053 * is exactly correct, but the name isn't one of the legal ones, we
15054 * will return that as an error below. But if neither of these apply,
15055 * it could be that no posix class was intended at all, or that one
15056 * was, but there was a typo. We tease these apart by doing fuzzy
15057 * matching on the name */
15058 if (class_number == OOB_NAMEDCLASS && found_problem) {
15059 const UV posix_names[][6] = {
15060 { 'a', 'l', 'n', 'u', 'm' },
15061 { 'a', 'l', 'p', 'h', 'a' },
15062 { 'a', 's', 'c', 'i', 'i' },
15063 { 'b', 'l', 'a', 'n', 'k' },
15064 { 'c', 'n', 't', 'r', 'l' },
15065 { 'd', 'i', 'g', 'i', 't' },
15066 { 'g', 'r', 'a', 'p', 'h' },
15067 { 'l', 'o', 'w', 'e', 'r' },
15068 { 'p', 'r', 'i', 'n', 't' },
15069 { 'p', 'u', 'n', 'c', 't' },
15070 { 's', 'p', 'a', 'c', 'e' },
15071 { 'u', 'p', 'p', 'e', 'r' },
15072 { 'w', 'o', 'r', 'd' },
15073 { 'x', 'd', 'i', 'g', 'i', 't' }
15075 /* The names of the above all have added NULs to make them the same
15076 * size, so we need to also have the real lengths */
15077 const UV posix_name_lengths[] = {
15078 sizeof("alnum") - 1,
15079 sizeof("alpha") - 1,
15080 sizeof("ascii") - 1,
15081 sizeof("blank") - 1,
15082 sizeof("cntrl") - 1,
15083 sizeof("digit") - 1,
15084 sizeof("graph") - 1,
15085 sizeof("lower") - 1,
15086 sizeof("print") - 1,
15087 sizeof("punct") - 1,
15088 sizeof("space") - 1,
15089 sizeof("upper") - 1,
15090 sizeof("word") - 1,
15091 sizeof("xdigit")- 1
15094 int temp_max = max_distance; /* Use a temporary, so if we
15095 reparse, we haven't changed the
15098 /* Use a smaller max edit distance if we are missing one of the
15100 if ( has_opening_bracket + has_opening_colon < 2
15101 || has_terminating_bracket + has_terminating_colon < 2)
15106 /* See if the input name is close to a legal one */
15107 for (i = 0; i < C_ARRAY_LENGTH(posix_names); i++) {
15109 /* Short circuit call if the lengths are too far apart to be
15111 if (abs( (int) (name_len - posix_name_lengths[i]))
15117 if (edit_distance(input_text,
15120 posix_name_lengths[i],
15124 { /* If it is close, it probably was intended to be a class */
15125 goto probably_meant_to_be;
15129 /* Here the input name is not close enough to a valid class name
15130 * for us to consider it to be intended to be a posix class. If
15131 * we haven't already done so, and the parse found a character that
15132 * could have been terminators for the name, but which we absorbed
15133 * as typos during the first pass, repeat the parse, signalling it
15134 * to stop at that character */
15135 if (possible_end && possible_end != (char *) -1) {
15136 possible_end = (char *) -1;
15141 /* Here neither pass found a close-enough class name */
15142 CLEAR_POSIX_WARNINGS_AND_RETURN(NOT_MEANT_TO_BE_A_POSIX_CLASS);
15145 probably_meant_to_be:
15147 /* Here we think that a posix specification was intended. Update any
15149 if (updated_parse_ptr) {
15150 *updated_parse_ptr = (char *) p;
15153 /* If a posix class name was intended but incorrectly specified, we
15154 * output or return the warnings */
15155 if (found_problem) {
15157 /* We set flags for these issues in the parse loop above instead of
15158 * adding them to the list of warnings, because we can parse it
15159 * twice, and we only want one warning instance */
15161 ADD_POSIX_WARNING(p, "the name must be all lowercase letters");
15164 ADD_POSIX_WARNING(p, NO_BLANKS_POSIX_WARNING);
15166 if (has_semi_colon) {
15167 ADD_POSIX_WARNING(p, SEMI_COLON_POSIX_WARNING);
15169 else if (! has_terminating_colon) {
15170 ADD_POSIX_WARNING(p, "there is no terminating ':'");
15172 if (! has_terminating_bracket) {
15173 ADD_POSIX_WARNING(p, "there is no terminating ']'");
15176 if (posix_warnings && RExC_warn_text && av_top_index(RExC_warn_text) > -1) {
15177 *posix_warnings = RExC_warn_text;
15180 else if (class_number != OOB_NAMEDCLASS) {
15181 /* If it is a known class, return the class. The class number
15182 * #defines are structured so each complement is +1 to the normal
15184 CLEAR_POSIX_WARNINGS_AND_RETURN(class_number + complement);
15186 else if (! check_only) {
15188 /* Here, it is an unrecognized class. This is an error (unless the
15189 * call is to check only, which we've already handled above) */
15190 const char * const complement_string = (complement)
15193 RExC_parse = (char *) p;
15194 vFAIL3utf8f("POSIX class [:%s%" UTF8f ":] unknown",
15196 UTF8fARG(UTF, RExC_parse - name_start - 2, name_start));
15200 return OOB_NAMEDCLASS;
15202 #undef ADD_POSIX_WARNING
15204 STATIC unsigned int
15205 S_regex_set_precedence(const U8 my_operator) {
15207 /* Returns the precedence in the (?[...]) construct of the input operator,
15208 * specified by its character representation. The precedence follows
15209 * general Perl rules, but it extends this so that ')' and ']' have (low)
15210 * precedence even though they aren't really operators */
15212 switch (my_operator) {
15228 NOT_REACHED; /* NOTREACHED */
15229 return 0; /* Silence compiler warning */
15233 S_handle_regex_sets(pTHX_ RExC_state_t *pRExC_state, SV** return_invlist,
15234 I32 *flagp, U32 depth,
15235 char * const oregcomp_parse)
15237 /* Handle the (?[...]) construct to do set operations */
15239 U8 curchar; /* Current character being parsed */
15240 UV start, end; /* End points of code point ranges */
15241 SV* final = NULL; /* The end result inversion list */
15242 SV* result_string; /* 'final' stringified */
15243 AV* stack; /* stack of operators and operands not yet
15245 AV* fence_stack = NULL; /* A stack containing the positions in
15246 'stack' of where the undealt-with left
15247 parens would be if they were actually
15249 /* The 'volatile' is a workaround for an optimiser bug
15250 * in Solaris Studio 12.3. See RT #127455 */
15251 volatile IV fence = 0; /* Position of where most recent undealt-
15252 with left paren in stack is; -1 if none.
15254 STRLEN len; /* Temporary */
15255 regnode* node; /* Temporary, and final regnode returned by
15257 const bool save_fold = FOLD; /* Temporary */
15258 char *save_end, *save_parse; /* Temporaries */
15259 const bool in_locale = LOC; /* we turn off /l during processing */
15260 AV* posix_warnings = NULL;
15262 GET_RE_DEBUG_FLAGS_DECL;
15264 PERL_ARGS_ASSERT_HANDLE_REGEX_SETS;
15266 DEBUG_PARSE("xcls");
15269 set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET);
15272 REQUIRE_UNI_RULES(flagp, NULL); /* The use of this operator implies /u.
15273 This is required so that the compile
15274 time values are valid in all runtime
15277 /* This will return only an ANYOF regnode, or (unlikely) something smaller
15278 * (such as EXACT). Thus we can skip most everything if just sizing. We
15279 * call regclass to handle '[]' so as to not have to reinvent its parsing
15280 * rules here (throwing away the size it computes each time). And, we exit
15281 * upon an unescaped ']' that isn't one ending a regclass. To do both
15282 * these things, we need to realize that something preceded by a backslash
15283 * is escaped, so we have to keep track of backslashes */
15285 UV nest_depth = 0; /* how many nested (?[...]) constructs */
15287 while (RExC_parse < RExC_end) {
15288 SV* current = NULL;
15290 skip_to_be_ignored_text(pRExC_state, &RExC_parse,
15291 TRUE /* Force /x */ );
15293 switch (*RExC_parse) {
15295 if (RExC_parse[1] == '?' && RExC_parse[2] == '[')
15296 nest_depth++, RExC_parse+=2;
15301 /* Skip past this, so the next character gets skipped, after
15304 if (*RExC_parse == 'c') {
15305 /* Skip the \cX notation for control characters */
15306 RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
15312 /* See if this is a [:posix:] class. */
15313 bool is_posix_class = (OOB_NAMEDCLASS
15314 < handle_possible_posix(pRExC_state,
15318 TRUE /* checking only */));
15319 /* If it is a posix class, leave the parse pointer at the
15320 * '[' to fool regclass() into thinking it is part of a
15321 * '[[:posix:]]'. */
15322 if (! is_posix_class) {
15326 /* regclass() can only return RESTART_PASS1 and NEED_UTF8
15327 * if multi-char folds are allowed. */
15328 if (!regclass(pRExC_state, flagp,depth+1,
15329 is_posix_class, /* parse the whole char
15330 class only if not a
15332 FALSE, /* don't allow multi-char folds */
15333 TRUE, /* silence non-portable warnings. */
15335 FALSE, /* Require return to be an ANYOF */
15339 FAIL2("panic: regclass returned NULL to handle_sets, "
15340 "flags=%#" UVxf, (UV) *flagp);
15342 /* function call leaves parse pointing to the ']', except
15343 * if we faked it */
15344 if (is_posix_class) {
15348 SvREFCNT_dec(current); /* In case it returned something */
15353 if (RExC_parse[1] == ')') {
15355 if (nest_depth--) break;
15356 node = reganode(pRExC_state, ANYOF, 0);
15357 RExC_size += ANYOF_SKIP;
15358 nextchar(pRExC_state);
15359 Set_Node_Length(node,
15360 RExC_parse - oregcomp_parse + 1); /* MJD */
15362 set_regex_charset(&RExC_flags, REGEX_LOCALE_CHARSET);
15367 /* We output the messages even if warnings are off, because we'll fail
15368 * the very next thing, and these give a likely diagnosis for that */
15369 if (posix_warnings && av_tindex_skip_len_mg(posix_warnings) >= 0) {
15370 output_or_return_posix_warnings(pRExC_state, posix_warnings, NULL);
15373 vFAIL("Unexpected ']' with no following ')' in (?[...");
15376 RExC_parse += UTF ? UTF8SKIP(RExC_parse) : 1;
15379 /* We output the messages even if warnings are off, because we'll fail
15380 * the very next thing, and these give a likely diagnosis for that */
15381 if (posix_warnings && av_tindex_skip_len_mg(posix_warnings) >= 0) {
15382 output_or_return_posix_warnings(pRExC_state, posix_warnings, NULL);
15385 vFAIL("Syntax error in (?[...])");
15388 /* Pass 2 only after this. */
15389 Perl_ck_warner_d(aTHX_
15390 packWARN(WARN_EXPERIMENTAL__REGEX_SETS),
15391 "The regex_sets feature is experimental" REPORT_LOCATION,
15392 REPORT_LOCATION_ARGS(RExC_parse));
15394 /* Everything in this construct is a metacharacter. Operands begin with
15395 * either a '\' (for an escape sequence), or a '[' for a bracketed
15396 * character class. Any other character should be an operator, or
15397 * parenthesis for grouping. Both types of operands are handled by calling
15398 * regclass() to parse them. It is called with a parameter to indicate to
15399 * return the computed inversion list. The parsing here is implemented via
15400 * a stack. Each entry on the stack is a single character representing one
15401 * of the operators; or else a pointer to an operand inversion list. */
15403 #define IS_OPERATOR(a) SvIOK(a)
15404 #define IS_OPERAND(a) (! IS_OPERATOR(a))
15406 /* The stack is kept in Łukasiewicz order. (That's pronounced similar
15407 * to luke-a-shave-itch (or -itz), but people who didn't want to bother
15408 * with pronouncing it called it Reverse Polish instead, but now that YOU
15409 * know how to pronounce it you can use the correct term, thus giving due
15410 * credit to the person who invented it, and impressing your geek friends.
15411 * Wikipedia says that the pronounciation of "Ł" has been changing so that
15412 * it is now more like an English initial W (as in wonk) than an L.)
15414 * This means that, for example, 'a | b & c' is stored on the stack as
15422 * where the numbers in brackets give the stack [array] element number.
15423 * In this implementation, parentheses are not stored on the stack.
15424 * Instead a '(' creates a "fence" so that the part of the stack below the
15425 * fence is invisible except to the corresponding ')' (this allows us to
15426 * replace testing for parens, by using instead subtraction of the fence
15427 * position). As new operands are processed they are pushed onto the stack
15428 * (except as noted in the next paragraph). New operators of higher
15429 * precedence than the current final one are inserted on the stack before
15430 * the lhs operand (so that when the rhs is pushed next, everything will be
15431 * in the correct positions shown above. When an operator of equal or
15432 * lower precedence is encountered in parsing, all the stacked operations
15433 * of equal or higher precedence are evaluated, leaving the result as the
15434 * top entry on the stack. This makes higher precedence operations
15435 * evaluate before lower precedence ones, and causes operations of equal
15436 * precedence to left associate.
15438 * The only unary operator '!' is immediately pushed onto the stack when
15439 * encountered. When an operand is encountered, if the top of the stack is
15440 * a '!", the complement is immediately performed, and the '!' popped. The
15441 * resulting value is treated as a new operand, and the logic in the
15442 * previous paragraph is executed. Thus in the expression
15444 * the stack looks like
15450 * as 'b' gets parsed, the latter gets evaluated to '!b', and the stack
15457 * A ')' is treated as an operator with lower precedence than all the
15458 * aforementioned ones, which causes all operations on the stack above the
15459 * corresponding '(' to be evaluated down to a single resultant operand.
15460 * Then the fence for the '(' is removed, and the operand goes through the
15461 * algorithm above, without the fence.
15463 * A separate stack is kept of the fence positions, so that the position of
15464 * the latest so-far unbalanced '(' is at the top of it.
15466 * The ']' ending the construct is treated as the lowest operator of all,
15467 * so that everything gets evaluated down to a single operand, which is the
15470 sv_2mortal((SV *)(stack = newAV()));
15471 sv_2mortal((SV *)(fence_stack = newAV()));
15473 while (RExC_parse < RExC_end) {
15474 I32 top_index; /* Index of top-most element in 'stack' */
15475 SV** top_ptr; /* Pointer to top 'stack' element */
15476 SV* current = NULL; /* To contain the current inversion list
15478 SV* only_to_avoid_leaks;
15480 skip_to_be_ignored_text(pRExC_state, &RExC_parse,
15481 TRUE /* Force /x */ );
15482 if (RExC_parse >= RExC_end) {
15483 Perl_croak(aTHX_ "panic: Read past end of '(?[ ])'");
15486 curchar = UCHARAT(RExC_parse);
15490 #ifdef ENABLE_REGEX_SETS_DEBUGGING
15491 /* Enable with -Accflags=-DENABLE_REGEX_SETS_DEBUGGING */
15492 DEBUG_U(dump_regex_sets_structures(pRExC_state,
15493 stack, fence, fence_stack));
15496 top_index = av_tindex_skip_len_mg(stack);
15499 SV** stacked_ptr; /* Ptr to something already on 'stack' */
15500 char stacked_operator; /* The topmost operator on the 'stack'. */
15501 SV* lhs; /* Operand to the left of the operator */
15502 SV* rhs; /* Operand to the right of the operator */
15503 SV* fence_ptr; /* Pointer to top element of the fence
15508 if ( RExC_parse < RExC_end - 1
15509 && (UCHARAT(RExC_parse + 1) == '?'))
15511 /* If is a '(?', could be an embedded '(?flags:(?[...])'.
15512 * This happens when we have some thing like
15514 * my $thai_or_lao = qr/(?[ \p{Thai} + \p{Lao} ])/;
15516 * qr/(?[ \p{Digit} & $thai_or_lao ])/;
15518 * Here we would be handling the interpolated
15519 * '$thai_or_lao'. We handle this by a recursive call to
15520 * ourselves which returns the inversion list the
15521 * interpolated expression evaluates to. We use the flags
15522 * from the interpolated pattern. */
15523 U32 save_flags = RExC_flags;
15524 const char * save_parse;
15526 RExC_parse += 2; /* Skip past the '(?' */
15527 save_parse = RExC_parse;
15529 /* Parse any flags for the '(?' */
15530 parse_lparen_question_flags(pRExC_state);
15532 if (RExC_parse == save_parse /* Makes sure there was at
15533 least one flag (or else
15534 this embedding wasn't
15536 || RExC_parse >= RExC_end - 4
15537 || UCHARAT(RExC_parse) != ':'
15538 || UCHARAT(++RExC_parse) != '('
15539 || UCHARAT(++RExC_parse) != '?'
15540 || UCHARAT(++RExC_parse) != '[')
15543 /* In combination with the above, this moves the
15544 * pointer to the point just after the first erroneous
15545 * character (or if there are no flags, to where they
15546 * should have been) */
15547 if (RExC_parse >= RExC_end - 4) {
15548 RExC_parse = RExC_end;
15550 else if (RExC_parse != save_parse) {
15551 RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1;
15553 vFAIL("Expecting '(?flags:(?[...'");
15556 /* Recurse, with the meat of the embedded expression */
15558 (void) handle_regex_sets(pRExC_state, ¤t, flagp,
15559 depth+1, oregcomp_parse);
15561 /* Here, 'current' contains the embedded expression's
15562 * inversion list, and RExC_parse points to the trailing
15563 * ']'; the next character should be the ')' */
15565 if (UCHARAT(RExC_parse) != ')')
15566 vFAIL("Expecting close paren for nested extended charclass");
15568 /* Then the ')' matching the original '(' handled by this
15569 * case: statement */
15571 if (UCHARAT(RExC_parse) != ')')
15572 vFAIL("Expecting close paren for wrapper for nested extended charclass");
15575 RExC_flags = save_flags;
15576 goto handle_operand;
15579 /* A regular '('. Look behind for illegal syntax */
15580 if (top_index - fence >= 0) {
15581 /* If the top entry on the stack is an operator, it had
15582 * better be a '!', otherwise the entry below the top
15583 * operand should be an operator */
15584 if ( ! (top_ptr = av_fetch(stack, top_index, FALSE))
15585 || (IS_OPERATOR(*top_ptr) && SvUV(*top_ptr) != '!')
15586 || ( IS_OPERAND(*top_ptr)
15587 && ( top_index - fence < 1
15588 || ! (stacked_ptr = av_fetch(stack,
15591 || ! IS_OPERATOR(*stacked_ptr))))
15594 vFAIL("Unexpected '(' with no preceding operator");
15598 /* Stack the position of this undealt-with left paren */
15599 av_push(fence_stack, newSViv(fence));
15600 fence = top_index + 1;
15604 /* regclass() can only return RESTART_PASS1 and NEED_UTF8 if
15605 * multi-char folds are allowed. */
15606 if (!regclass(pRExC_state, flagp,depth+1,
15607 TRUE, /* means parse just the next thing */
15608 FALSE, /* don't allow multi-char folds */
15609 FALSE, /* don't silence non-portable warnings. */
15611 FALSE, /* Require return to be an ANYOF */
15615 FAIL2("panic: regclass returned NULL to handle_sets, "
15616 "flags=%#" UVxf, (UV) *flagp);
15619 /* regclass() will return with parsing just the \ sequence,
15620 * leaving the parse pointer at the next thing to parse */
15622 goto handle_operand;
15624 case '[': /* Is a bracketed character class */
15626 /* See if this is a [:posix:] class. */
15627 bool is_posix_class = (OOB_NAMEDCLASS
15628 < handle_possible_posix(pRExC_state,
15632 TRUE /* checking only */));
15633 /* If it is a posix class, leave the parse pointer at the '['
15634 * to fool regclass() into thinking it is part of a
15635 * '[[:posix:]]'. */
15636 if (! is_posix_class) {
15640 /* regclass() can only return RESTART_PASS1 and NEED_UTF8 if
15641 * multi-char folds are allowed. */
15642 if (!regclass(pRExC_state, flagp,depth+1,
15643 is_posix_class, /* parse the whole char
15644 class only if not a
15646 FALSE, /* don't allow multi-char folds */
15647 TRUE, /* silence non-portable warnings. */
15649 FALSE, /* Require return to be an ANYOF */
15654 FAIL2("panic: regclass returned NULL to handle_sets, "
15655 "flags=%#" UVxf, (UV) *flagp);
15658 /* function call leaves parse pointing to the ']', except if we
15660 if (is_posix_class) {
15664 goto handle_operand;
15668 if (top_index >= 1) {
15669 goto join_operators;
15672 /* Only a single operand on the stack: are done */
15676 if (av_tindex_skip_len_mg(fence_stack) < 0) {
15678 vFAIL("Unexpected ')'");
15681 /* If nothing after the fence, is missing an operand */
15682 if (top_index - fence < 0) {
15686 /* If at least two things on the stack, treat this as an
15688 if (top_index - fence >= 1) {
15689 goto join_operators;
15692 /* Here only a single thing on the fenced stack, and there is a
15693 * fence. Get rid of it */
15694 fence_ptr = av_pop(fence_stack);
15696 fence = SvIV(fence_ptr) - 1;
15697 SvREFCNT_dec_NN(fence_ptr);
15704 /* Having gotten rid of the fence, we pop the operand at the
15705 * stack top and process it as a newly encountered operand */
15706 current = av_pop(stack);
15707 if (IS_OPERAND(current)) {
15708 goto handle_operand;
15720 /* These binary operators should have a left operand already
15722 if ( top_index - fence < 0
15723 || top_index - fence == 1
15724 || ( ! (top_ptr = av_fetch(stack, top_index, FALSE)))
15725 || ! IS_OPERAND(*top_ptr))
15727 goto unexpected_binary;
15730 /* If only the one operand is on the part of the stack visible
15731 * to us, we just place this operator in the proper position */
15732 if (top_index - fence < 2) {
15734 /* Place the operator before the operand */
15736 SV* lhs = av_pop(stack);
15737 av_push(stack, newSVuv(curchar));
15738 av_push(stack, lhs);
15742 /* But if there is something else on the stack, we need to
15743 * process it before this new operator if and only if the
15744 * stacked operation has equal or higher precedence than the
15749 /* The operator on the stack is supposed to be below both its
15751 if ( ! (stacked_ptr = av_fetch(stack, top_index - 2, FALSE))
15752 || IS_OPERAND(*stacked_ptr))
15754 /* But if not, it's legal and indicates we are completely
15755 * done if and only if we're currently processing a ']',
15756 * which should be the final thing in the expression */
15757 if (curchar == ']') {
15763 vFAIL2("Unexpected binary operator '%c' with no "
15764 "preceding operand", curchar);
15766 stacked_operator = (char) SvUV(*stacked_ptr);
15768 if (regex_set_precedence(curchar)
15769 > regex_set_precedence(stacked_operator))
15771 /* Here, the new operator has higher precedence than the
15772 * stacked one. This means we need to add the new one to
15773 * the stack to await its rhs operand (and maybe more
15774 * stuff). We put it before the lhs operand, leaving
15775 * untouched the stacked operator and everything below it
15777 lhs = av_pop(stack);
15778 assert(IS_OPERAND(lhs));
15780 av_push(stack, newSVuv(curchar));
15781 av_push(stack, lhs);
15785 /* Here, the new operator has equal or lower precedence than
15786 * what's already there. This means the operation already
15787 * there should be performed now, before the new one. */
15789 rhs = av_pop(stack);
15790 if (! IS_OPERAND(rhs)) {
15792 /* This can happen when a ! is not followed by an operand,
15793 * like in /(?[\t &!])/ */
15797 lhs = av_pop(stack);
15799 if (! IS_OPERAND(lhs)) {
15801 /* This can happen when there is an empty (), like in
15802 * /(?[[0]+()+])/ */
15806 switch (stacked_operator) {
15808 _invlist_intersection(lhs, rhs, &rhs);
15813 _invlist_union(lhs, rhs, &rhs);
15817 _invlist_subtract(lhs, rhs, &rhs);
15820 case '^': /* The union minus the intersection */
15825 _invlist_union(lhs, rhs, &u);
15826 _invlist_intersection(lhs, rhs, &i);
15827 _invlist_subtract(u, i, &rhs);
15828 SvREFCNT_dec_NN(i);
15829 SvREFCNT_dec_NN(u);
15835 /* Here, the higher precedence operation has been done, and the
15836 * result is in 'rhs'. We overwrite the stacked operator with
15837 * the result. Then we redo this code to either push the new
15838 * operator onto the stack or perform any higher precedence
15839 * stacked operation */
15840 only_to_avoid_leaks = av_pop(stack);
15841 SvREFCNT_dec(only_to_avoid_leaks);
15842 av_push(stack, rhs);
15845 case '!': /* Highest priority, right associative */
15847 /* If what's already at the top of the stack is another '!",
15848 * they just cancel each other out */
15849 if ( (top_ptr = av_fetch(stack, top_index, FALSE))
15850 && (IS_OPERATOR(*top_ptr) && SvUV(*top_ptr) == '!'))
15852 only_to_avoid_leaks = av_pop(stack);
15853 SvREFCNT_dec(only_to_avoid_leaks);
15855 else { /* Otherwise, since it's right associative, just push
15857 av_push(stack, newSVuv(curchar));
15862 RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1;
15863 vFAIL("Unexpected character");
15867 /* Here 'current' is the operand. If something is already on the
15868 * stack, we have to check if it is a !. But first, the code above
15869 * may have altered the stack in the time since we earlier set
15872 top_index = av_tindex_skip_len_mg(stack);
15873 if (top_index - fence >= 0) {
15874 /* If the top entry on the stack is an operator, it had better
15875 * be a '!', otherwise the entry below the top operand should
15876 * be an operator */
15877 top_ptr = av_fetch(stack, top_index, FALSE);
15879 if (IS_OPERATOR(*top_ptr)) {
15881 /* The only permissible operator at the top of the stack is
15882 * '!', which is applied immediately to this operand. */
15883 curchar = (char) SvUV(*top_ptr);
15884 if (curchar != '!') {
15885 SvREFCNT_dec(current);
15886 vFAIL2("Unexpected binary operator '%c' with no "
15887 "preceding operand", curchar);
15890 _invlist_invert(current);
15892 only_to_avoid_leaks = av_pop(stack);
15893 SvREFCNT_dec(only_to_avoid_leaks);
15895 /* And we redo with the inverted operand. This allows
15896 * handling multiple ! in a row */
15897 goto handle_operand;
15899 /* Single operand is ok only for the non-binary ')'
15901 else if ((top_index - fence == 0 && curchar != ')')
15902 || (top_index - fence > 0
15903 && (! (stacked_ptr = av_fetch(stack,
15906 || IS_OPERAND(*stacked_ptr))))
15908 SvREFCNT_dec(current);
15909 vFAIL("Operand with no preceding operator");
15913 /* Here there was nothing on the stack or the top element was
15914 * another operand. Just add this new one */
15915 av_push(stack, current);
15917 } /* End of switch on next parse token */
15919 RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1;
15920 } /* End of loop parsing through the construct */
15923 if (av_tindex_skip_len_mg(fence_stack) >= 0) {
15924 vFAIL("Unmatched (");
15927 if (av_tindex_skip_len_mg(stack) < 0 /* Was empty */
15928 || ((final = av_pop(stack)) == NULL)
15929 || ! IS_OPERAND(final)
15930 || SvTYPE(final) != SVt_INVLIST
15931 || av_tindex_skip_len_mg(stack) >= 0) /* More left on stack */
15934 SvREFCNT_dec(final);
15935 vFAIL("Incomplete expression within '(?[ ])'");
15938 /* Here, 'final' is the resultant inversion list from evaluating the
15939 * expression. Return it if so requested */
15940 if (return_invlist) {
15941 *return_invlist = final;
15945 /* Otherwise generate a resultant node, based on 'final'. regclass() is
15946 * expecting a string of ranges and individual code points */
15947 invlist_iterinit(final);
15948 result_string = newSVpvs("");
15949 while (invlist_iternext(final, &start, &end)) {
15950 if (start == end) {
15951 Perl_sv_catpvf(aTHX_ result_string, "\\x{%" UVXf "}", start);
15954 Perl_sv_catpvf(aTHX_ result_string, "\\x{%" UVXf "}-\\x{%" UVXf "}",
15959 /* About to generate an ANYOF (or similar) node from the inversion list we
15960 * have calculated */
15961 save_parse = RExC_parse;
15962 RExC_parse = SvPV(result_string, len);
15963 save_end = RExC_end;
15964 RExC_end = RExC_parse + len;
15966 /* We turn off folding around the call, as the class we have constructed
15967 * already has all folding taken into consideration, and we don't want
15968 * regclass() to add to that */
15969 RExC_flags &= ~RXf_PMf_FOLD;
15970 /* regclass() can only return RESTART_PASS1 and NEED_UTF8 if multi-char
15971 * folds are allowed. */
15972 node = regclass(pRExC_state, flagp,depth+1,
15973 FALSE, /* means parse the whole char class */
15974 FALSE, /* don't allow multi-char folds */
15975 TRUE, /* silence non-portable warnings. The above may very
15976 well have generated non-portable code points, but
15977 they're valid on this machine */
15978 FALSE, /* similarly, no need for strict */
15979 FALSE, /* Require return to be an ANYOF */
15984 FAIL2("panic: regclass returned NULL to handle_sets, flags=%#" UVxf,
15987 /* Fix up the node type if we are in locale. (We have pretended we are
15988 * under /u for the purposes of regclass(), as this construct will only
15989 * work under UTF-8 locales. But now we change the opcode to be ANYOFL (so
15990 * as to cause any warnings about bad locales to be output in regexec.c),
15991 * and add the flag that indicates to check if not in a UTF-8 locale. The
15992 * reason we above forbid optimization into something other than an ANYOF
15993 * node is simply to minimize the number of code changes in regexec.c.
15994 * Otherwise we would have to create new EXACTish node types and deal with
15995 * them. This decision could be revisited should this construct become
15998 * (One might think we could look at the resulting ANYOF node and suppress
15999 * the flag if everything is above 255, as those would be UTF-8 only,
16000 * but this isn't true, as the components that led to that result could
16001 * have been locale-affected, and just happen to cancel each other out
16002 * under UTF-8 locales.) */
16004 set_regex_charset(&RExC_flags, REGEX_LOCALE_CHARSET);
16006 assert(OP(node) == ANYOF);
16010 |= ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
16014 RExC_flags |= RXf_PMf_FOLD;
16017 RExC_parse = save_parse + 1;
16018 RExC_end = save_end;
16019 SvREFCNT_dec_NN(final);
16020 SvREFCNT_dec_NN(result_string);
16022 nextchar(pRExC_state);
16023 Set_Node_Length(node, RExC_parse - oregcomp_parse + 1); /* MJD */
16027 #ifdef ENABLE_REGEX_SETS_DEBUGGING
16030 S_dump_regex_sets_structures(pTHX_ RExC_state_t *pRExC_state,
16031 AV * stack, const IV fence, AV * fence_stack)
16032 { /* Dumps the stacks in handle_regex_sets() */
16034 const SSize_t stack_top = av_tindex_skip_len_mg(stack);
16035 const SSize_t fence_stack_top = av_tindex_skip_len_mg(fence_stack);
16038 PERL_ARGS_ASSERT_DUMP_REGEX_SETS_STRUCTURES;
16040 PerlIO_printf(Perl_debug_log, "\nParse position is:%s\n", RExC_parse);
16042 if (stack_top < 0) {
16043 PerlIO_printf(Perl_debug_log, "Nothing on stack\n");
16046 PerlIO_printf(Perl_debug_log, "Stack: (fence=%d)\n", (int) fence);
16047 for (i = stack_top; i >= 0; i--) {
16048 SV ** element_ptr = av_fetch(stack, i, FALSE);
16049 if (! element_ptr) {
16052 if (IS_OPERATOR(*element_ptr)) {
16053 PerlIO_printf(Perl_debug_log, "[%d]: %c\n",
16054 (int) i, (int) SvIV(*element_ptr));
16057 PerlIO_printf(Perl_debug_log, "[%d] ", (int) i);
16058 sv_dump(*element_ptr);
16063 if (fence_stack_top < 0) {
16064 PerlIO_printf(Perl_debug_log, "Nothing on fence_stack\n");
16067 PerlIO_printf(Perl_debug_log, "Fence_stack: \n");
16068 for (i = fence_stack_top; i >= 0; i--) {
16069 SV ** element_ptr = av_fetch(fence_stack, i, FALSE);
16070 if (! element_ptr) {
16073 PerlIO_printf(Perl_debug_log, "[%d]: %d\n",
16074 (int) i, (int) SvIV(*element_ptr));
16085 S_add_above_Latin1_folds(pTHX_ RExC_state_t *pRExC_state, const U8 cp, SV** invlist)
16087 /* This hard-codes the Latin1/above-Latin1 folding rules, so that an
16088 * innocent-looking character class, like /[ks]/i won't have to go out to
16089 * disk to find the possible matches.
16091 * This should be called only for a Latin1-range code points, cp, which is
16092 * known to be involved in a simple fold with other code points above
16093 * Latin1. It would give false results if /aa has been specified.
16094 * Multi-char folds are outside the scope of this, and must be handled
16097 * XXX It would be better to generate these via regen, in case a new
16098 * version of the Unicode standard adds new mappings, though that is not
16099 * really likely, and may be caught by the default: case of the switch
16102 PERL_ARGS_ASSERT_ADD_ABOVE_LATIN1_FOLDS;
16104 assert(HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(cp));
16110 add_cp_to_invlist(*invlist, KELVIN_SIGN);
16114 *invlist = add_cp_to_invlist(*invlist, LATIN_SMALL_LETTER_LONG_S);
16117 *invlist = add_cp_to_invlist(*invlist, GREEK_CAPITAL_LETTER_MU);
16118 *invlist = add_cp_to_invlist(*invlist, GREEK_SMALL_LETTER_MU);
16120 case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
16121 case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
16122 *invlist = add_cp_to_invlist(*invlist, ANGSTROM_SIGN);
16124 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
16125 *invlist = add_cp_to_invlist(*invlist,
16126 LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
16129 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
16131 case LATIN_SMALL_LETTER_SHARP_S:
16132 *invlist = add_cp_to_invlist(*invlist, LATIN_CAPITAL_LETTER_SHARP_S);
16137 #if UNICODE_MAJOR_VERSION < 3 \
16138 || (UNICODE_MAJOR_VERSION == 3 && UNICODE_DOT_VERSION == 0)
16140 /* In 3.0 and earlier, U+0130 folded simply to 'i'; and in 3.0.1 so did
16145 add_cp_to_invlist(*invlist, LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
16146 # if UNICODE_DOT_DOT_VERSION == 1
16147 *invlist = add_cp_to_invlist(*invlist, LATIN_SMALL_LETTER_DOTLESS_I);
16153 /* Use deprecated warning to increase the chances of this being
16156 ckWARN2reg_d(RExC_parse, "Perl folding rules are not up-to-date for 0x%02X; please use the perlbug utility to report;", cp);
16163 S_output_or_return_posix_warnings(pTHX_ RExC_state_t *pRExC_state, AV* posix_warnings, AV** return_posix_warnings)
16165 /* If the final parameter is NULL, output the elements of the array given
16166 * by '*posix_warnings' as REGEXP warnings. Otherwise, the elements are
16167 * pushed onto it, (creating if necessary) */
16170 const bool first_is_fatal = ! return_posix_warnings
16171 && ckDEAD(packWARN(WARN_REGEXP));
16173 PERL_ARGS_ASSERT_OUTPUT_OR_RETURN_POSIX_WARNINGS;
16175 while ((msg = av_shift(posix_warnings)) != &PL_sv_undef) {
16176 if (return_posix_warnings) {
16177 if (! *return_posix_warnings) { /* mortalize to not leak if
16178 warnings are fatal */
16179 *return_posix_warnings = (AV *) sv_2mortal((SV *) newAV());
16181 av_push(*return_posix_warnings, msg);
16184 if (first_is_fatal) { /* Avoid leaking this */
16185 av_undef(posix_warnings); /* This isn't necessary if the
16186 array is mortal, but is a
16188 (void) sv_2mortal(msg);
16190 SAVEFREESV(RExC_rx_sv);
16193 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s", SvPVX(msg));
16194 SvREFCNT_dec_NN(msg);
16200 S_add_multi_match(pTHX_ AV* multi_char_matches, SV* multi_string, const STRLEN cp_count)
16202 /* This adds the string scalar <multi_string> to the array
16203 * <multi_char_matches>. <multi_string> is known to have exactly
16204 * <cp_count> code points in it. This is used when constructing a
16205 * bracketed character class and we find something that needs to match more
16206 * than a single character.
16208 * <multi_char_matches> is actually an array of arrays. Each top-level
16209 * element is an array that contains all the strings known so far that are
16210 * the same length. And that length (in number of code points) is the same
16211 * as the index of the top-level array. Hence, the [2] element is an
16212 * array, each element thereof is a string containing TWO code points;
16213 * while element [3] is for strings of THREE characters, and so on. Since
16214 * this is for multi-char strings there can never be a [0] nor [1] element.
16216 * When we rewrite the character class below, we will do so such that the
16217 * longest strings are written first, so that it prefers the longest
16218 * matching strings first. This is done even if it turns out that any
16219 * quantifier is non-greedy, out of this programmer's (khw) laziness. Tom
16220 * Christiansen has agreed that this is ok. This makes the test for the
16221 * ligature 'ffi' come before the test for 'ff', for example */
16224 AV** this_array_ptr;
16226 PERL_ARGS_ASSERT_ADD_MULTI_MATCH;
16228 if (! multi_char_matches) {
16229 multi_char_matches = newAV();
16232 if (av_exists(multi_char_matches, cp_count)) {
16233 this_array_ptr = (AV**) av_fetch(multi_char_matches, cp_count, FALSE);
16234 this_array = *this_array_ptr;
16237 this_array = newAV();
16238 av_store(multi_char_matches, cp_count,
16241 av_push(this_array, multi_string);
16243 return multi_char_matches;
16246 /* The names of properties whose definitions are not known at compile time are
16247 * stored in this SV, after a constant heading. So if the length has been
16248 * changed since initialization, then there is a run-time definition. */
16249 #define HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION \
16250 (SvCUR(listsv) != initial_listsv_len)
16252 /* There is a restricted set of white space characters that are legal when
16253 * ignoring white space in a bracketed character class. This generates the
16254 * code to skip them.
16256 * There is a line below that uses the same white space criteria but is outside
16257 * this macro. Both here and there must use the same definition */
16258 #define SKIP_BRACKETED_WHITE_SPACE(do_skip, p) \
16261 while (isBLANK_A(UCHARAT(p))) \
16269 S_regclass(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth,
16270 const bool stop_at_1, /* Just parse the next thing, don't
16271 look for a full character class */
16272 bool allow_multi_folds,
16273 const bool silence_non_portable, /* Don't output warnings
16277 bool optimizable, /* ? Allow a non-ANYOF return
16279 SV** ret_invlist, /* Return an inversion list, not a node */
16280 AV** return_posix_warnings
16283 /* parse a bracketed class specification. Most of these will produce an
16284 * ANYOF node; but something like [a] will produce an EXACT node; [aA], an
16285 * EXACTFish node; [[:ascii:]], a POSIXA node; etc. It is more complex
16286 * under /i with multi-character folds: it will be rewritten following the
16287 * paradigm of this example, where the <multi-fold>s are characters which
16288 * fold to multiple character sequences:
16289 * /[abc\x{multi-fold1}def\x{multi-fold2}ghi]/i
16290 * gets effectively rewritten as:
16291 * /(?:\x{multi-fold1}|\x{multi-fold2}|[abcdefghi]/i
16292 * reg() gets called (recursively) on the rewritten version, and this
16293 * function will return what it constructs. (Actually the <multi-fold>s
16294 * aren't physically removed from the [abcdefghi], it's just that they are
16295 * ignored in the recursion by means of a flag:
16296 * <RExC_in_multi_char_class>.)
16298 * ANYOF nodes contain a bit map for the first NUM_ANYOF_CODE_POINTS
16299 * characters, with the corresponding bit set if that character is in the
16300 * list. For characters above this, a range list or swash is used. There
16301 * are extra bits for \w, etc. in locale ANYOFs, as what these match is not
16302 * determinable at compile time
16304 * Returns NULL, setting *flagp to RESTART_PASS1 if the sizing scan needs
16305 * to be restarted, or'd with NEED_UTF8 if the pattern needs to be upgraded
16306 * to UTF-8. This can only happen if ret_invlist is non-NULL.
16309 UV prevvalue = OOB_UNICODE, save_prevvalue = OOB_UNICODE;
16311 UV value = OOB_UNICODE, save_value = OOB_UNICODE;
16314 int namedclass = OOB_NAMEDCLASS;
16315 char *rangebegin = NULL;
16316 bool need_class = 0;
16318 STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more
16319 than just initialized. */
16320 SV* properties = NULL; /* Code points that match \p{} \P{} */
16321 SV* posixes = NULL; /* Code points that match classes like [:word:],
16322 extended beyond the Latin1 range. These have to
16323 be kept separate from other code points for much
16324 of this function because their handling is
16325 different under /i, and for most classes under
16327 SV* nposixes = NULL; /* Similarly for [:^word:]. These are kept
16328 separate for a while from the non-complemented
16329 versions because of complications with /d
16331 SV* simple_posixes = NULL; /* But under some conditions, the classes can be
16332 treated more simply than the general case,
16333 leading to less compilation and execution
16335 UV element_count = 0; /* Number of distinct elements in the class.
16336 Optimizations may be possible if this is tiny */
16337 AV * multi_char_matches = NULL; /* Code points that fold to more than one
16338 character; used under /i */
16340 char * stop_ptr = RExC_end; /* where to stop parsing */
16342 /* ignore unescaped whitespace? */
16343 const bool skip_white = cBOOL( ret_invlist
16344 || (RExC_flags & RXf_PMf_EXTENDED_MORE));
16346 /* Unicode properties are stored in a swash; this holds the current one
16347 * being parsed. If this swash is the only above-latin1 component of the
16348 * character class, an optimization is to pass it directly on to the
16349 * execution engine. Otherwise, it is set to NULL to indicate that there
16350 * are other things in the class that have to be dealt with at execution
16352 SV* swash = NULL; /* Code points that match \p{} \P{} */
16354 /* Set if a component of this character class is user-defined; just passed
16355 * on to the engine */
16356 bool has_user_defined_property = FALSE;
16358 /* inversion list of code points this node matches only when the target
16359 * string is in UTF-8. These are all non-ASCII, < 256. (Because is under
16361 SV* has_upper_latin1_only_utf8_matches = NULL;
16363 /* Inversion list of code points this node matches regardless of things
16364 * like locale, folding, utf8ness of the target string */
16365 SV* cp_list = NULL;
16367 /* Like cp_list, but code points on this list need to be checked for things
16368 * that fold to/from them under /i */
16369 SV* cp_foldable_list = NULL;
16371 /* Like cp_list, but code points on this list are valid only when the
16372 * runtime locale is UTF-8 */
16373 SV* only_utf8_locale_list = NULL;
16375 /* In a range, if one of the endpoints is non-character-set portable,
16376 * meaning that it hard-codes a code point that may mean a different
16377 * charactger in ASCII vs. EBCDIC, as opposed to, say, a literal 'A' or a
16378 * mnemonic '\t' which each mean the same character no matter which
16379 * character set the platform is on. */
16380 unsigned int non_portable_endpoint = 0;
16382 /* Is the range unicode? which means on a platform that isn't 1-1 native
16383 * to Unicode (i.e. non-ASCII), each code point in it should be considered
16384 * to be a Unicode value. */
16385 bool unicode_range = FALSE;
16386 bool invert = FALSE; /* Is this class to be complemented */
16388 bool warn_super = ALWAYS_WARN_SUPER;
16390 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
16391 case we need to change the emitted regop to an EXACT. */
16392 const char * orig_parse = RExC_parse;
16393 const SSize_t orig_size = RExC_size;
16394 bool posixl_matches_all = FALSE; /* Does /l class have both e.g. \W,\w ? */
16396 /* This variable is used to mark where the end in the input is of something
16397 * that looks like a POSIX construct but isn't. During the parse, when
16398 * something looks like it could be such a construct is encountered, it is
16399 * checked for being one, but not if we've already checked this area of the
16400 * input. Only after this position is reached do we check again */
16401 char *not_posix_region_end = RExC_parse - 1;
16403 AV* posix_warnings = NULL;
16404 const bool do_posix_warnings = return_posix_warnings
16405 || (PASS2 && ckWARN(WARN_REGEXP));
16407 GET_RE_DEBUG_FLAGS_DECL;
16409 PERL_ARGS_ASSERT_REGCLASS;
16411 PERL_UNUSED_ARG(depth);
16414 DEBUG_PARSE("clas");
16416 #if UNICODE_MAJOR_VERSION < 3 /* no multifolds in early Unicode */ \
16417 || (UNICODE_MAJOR_VERSION == 3 && UNICODE_DOT_VERSION == 0 \
16418 && UNICODE_DOT_DOT_VERSION == 0)
16419 allow_multi_folds = FALSE;
16422 /* Assume we are going to generate an ANYOF node. */
16423 ret = reganode(pRExC_state,
16430 RExC_size += ANYOF_SKIP;
16431 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
16434 ANYOF_FLAGS(ret) = 0;
16436 RExC_emit += ANYOF_SKIP;
16437 listsv = newSVpvs_flags("# comment\n", SVs_TEMP);
16438 initial_listsv_len = SvCUR(listsv);
16439 SvTEMP_off(listsv); /* Grr, TEMPs and mortals are conflated. */
16442 SKIP_BRACKETED_WHITE_SPACE(skip_white, RExC_parse);
16444 assert(RExC_parse <= RExC_end);
16446 if (UCHARAT(RExC_parse) == '^') { /* Complement the class */
16449 allow_multi_folds = FALSE;
16451 SKIP_BRACKETED_WHITE_SPACE(skip_white, RExC_parse);
16454 /* Check that they didn't say [:posix:] instead of [[:posix:]] */
16455 if (! ret_invlist && MAYBE_POSIXCC(UCHARAT(RExC_parse))) {
16456 int maybe_class = handle_possible_posix(pRExC_state,
16458 ¬_posix_region_end,
16460 TRUE /* checking only */);
16461 if (PASS2 && maybe_class >= OOB_NAMEDCLASS && do_posix_warnings) {
16462 SAVEFREESV(RExC_rx_sv);
16463 ckWARN4reg(not_posix_region_end,
16464 "POSIX syntax [%c %c] belongs inside character classes%s",
16465 *RExC_parse, *RExC_parse,
16466 (maybe_class == OOB_NAMEDCLASS)
16467 ? ((POSIXCC_NOTYET(*RExC_parse))
16468 ? " (but this one isn't implemented)"
16469 : " (but this one isn't fully valid)")
16472 (void)ReREFCNT_inc(RExC_rx_sv);
16476 /* If the caller wants us to just parse a single element, accomplish this
16477 * by faking the loop ending condition */
16478 if (stop_at_1 && RExC_end > RExC_parse) {
16479 stop_ptr = RExC_parse + 1;
16482 /* allow 1st char to be ']' (allowing it to be '-' is dealt with later) */
16483 if (UCHARAT(RExC_parse) == ']')
16484 goto charclassloop;
16488 if ( posix_warnings
16489 && av_tindex_skip_len_mg(posix_warnings) >= 0
16490 && RExC_parse > not_posix_region_end)
16492 /* Warnings about posix class issues are considered tentative until
16493 * we are far enough along in the parse that we can no longer
16494 * change our mind, at which point we either output them or add
16495 * them, if it has so specified, to what gets returned to the
16496 * caller. This is done each time through the loop so that a later
16497 * class won't zap them before they have been dealt with. */
16498 output_or_return_posix_warnings(pRExC_state, posix_warnings,
16499 return_posix_warnings);
16502 if (RExC_parse >= stop_ptr) {
16506 SKIP_BRACKETED_WHITE_SPACE(skip_white, RExC_parse);
16508 if (UCHARAT(RExC_parse) == ']') {
16514 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
16515 save_value = value;
16516 save_prevvalue = prevvalue;
16519 rangebegin = RExC_parse;
16521 non_portable_endpoint = 0;
16523 if (UTF && ! UTF8_IS_INVARIANT(* RExC_parse)) {
16524 value = utf8n_to_uvchr((U8*)RExC_parse,
16525 RExC_end - RExC_parse,
16526 &numlen, UTF8_ALLOW_DEFAULT);
16527 RExC_parse += numlen;
16530 value = UCHARAT(RExC_parse++);
16532 if (value == '[') {
16533 char * posix_class_end;
16534 namedclass = handle_possible_posix(pRExC_state,
16537 do_posix_warnings ? &posix_warnings : NULL,
16538 FALSE /* die if error */);
16539 if (namedclass > OOB_NAMEDCLASS) {
16541 /* If there was an earlier attempt to parse this particular
16542 * posix class, and it failed, it was a false alarm, as this
16543 * successful one proves */
16544 if ( posix_warnings
16545 && av_tindex_skip_len_mg(posix_warnings) >= 0
16546 && not_posix_region_end >= RExC_parse
16547 && not_posix_region_end <= posix_class_end)
16549 av_undef(posix_warnings);
16552 RExC_parse = posix_class_end;
16554 else if (namedclass == OOB_NAMEDCLASS) {
16555 not_posix_region_end = posix_class_end;
16558 namedclass = OOB_NAMEDCLASS;
16561 else if ( RExC_parse - 1 > not_posix_region_end
16562 && MAYBE_POSIXCC(value))
16564 (void) handle_possible_posix(
16566 RExC_parse - 1, /* -1 because parse has already been
16568 ¬_posix_region_end,
16569 do_posix_warnings ? &posix_warnings : NULL,
16570 TRUE /* checking only */);
16572 else if ( strict && ! skip_white
16573 && ( _generic_isCC(value, _CC_VERTSPACE)
16574 || is_VERTWS_cp_high(value)))
16576 vFAIL("Literal vertical space in [] is illegal except under /x");
16578 else if (value == '\\') {
16579 /* Is a backslash; get the code point of the char after it */
16581 if (RExC_parse >= RExC_end) {
16582 vFAIL("Unmatched [");
16585 if (UTF && ! UTF8_IS_INVARIANT(UCHARAT(RExC_parse))) {
16586 value = utf8n_to_uvchr((U8*)RExC_parse,
16587 RExC_end - RExC_parse,
16588 &numlen, UTF8_ALLOW_DEFAULT);
16589 RExC_parse += numlen;
16592 value = UCHARAT(RExC_parse++);
16594 /* Some compilers cannot handle switching on 64-bit integer
16595 * values, therefore value cannot be an UV. Yes, this will
16596 * be a problem later if we want switch on Unicode.
16597 * A similar issue a little bit later when switching on
16598 * namedclass. --jhi */
16600 /* If the \ is escaping white space when white space is being
16601 * skipped, it means that that white space is wanted literally, and
16602 * is already in 'value'. Otherwise, need to translate the escape
16603 * into what it signifies. */
16604 if (! skip_white || ! isBLANK_A(value)) switch ((I32)value) {
16606 case 'w': namedclass = ANYOF_WORDCHAR; break;
16607 case 'W': namedclass = ANYOF_NWORDCHAR; break;
16608 case 's': namedclass = ANYOF_SPACE; break;
16609 case 'S': namedclass = ANYOF_NSPACE; break;
16610 case 'd': namedclass = ANYOF_DIGIT; break;
16611 case 'D': namedclass = ANYOF_NDIGIT; break;
16612 case 'v': namedclass = ANYOF_VERTWS; break;
16613 case 'V': namedclass = ANYOF_NVERTWS; break;
16614 case 'h': namedclass = ANYOF_HORIZWS; break;
16615 case 'H': namedclass = ANYOF_NHORIZWS; break;
16616 case 'N': /* Handle \N{NAME} in class */
16618 const char * const backslash_N_beg = RExC_parse - 2;
16621 if (! grok_bslash_N(pRExC_state,
16622 NULL, /* No regnode */
16623 &value, /* Yes single value */
16624 &cp_count, /* Multiple code pt count */
16630 if (*flagp & NEED_UTF8)
16631 FAIL("panic: grok_bslash_N set NEED_UTF8");
16633 RETURN_NULL_ON_RESTART_FLAGP(flagp);
16635 if (cp_count < 0) {
16636 vFAIL("\\N in a character class must be a named character: \\N{...}");
16638 else if (cp_count == 0) {
16640 ckWARNreg(RExC_parse,
16641 "Ignoring zero length \\N{} in character class");
16644 else { /* cp_count > 1 */
16645 if (! RExC_in_multi_char_class) {
16646 if (invert || range || *RExC_parse == '-') {
16649 vFAIL("\\N{} in inverted character class or as a range end-point is restricted to one character");
16652 ckWARNreg(RExC_parse, "Using just the first character returned by \\N{} in character class");
16654 break; /* <value> contains the first code
16655 point. Drop out of the switch to
16659 SV * multi_char_N = newSVpvn(backslash_N_beg,
16660 RExC_parse - backslash_N_beg);
16662 = add_multi_match(multi_char_matches,
16667 } /* End of cp_count != 1 */
16669 /* This element should not be processed further in this
16672 value = save_value;
16673 prevvalue = save_prevvalue;
16674 continue; /* Back to top of loop to get next char */
16677 /* Here, is a single code point, and <value> contains it */
16678 unicode_range = TRUE; /* \N{} are Unicode */
16686 /* We will handle any undefined properties ourselves */
16687 U8 swash_init_flags = _CORE_SWASH_INIT_RETURN_IF_UNDEF
16688 /* And we actually would prefer to get
16689 * the straight inversion list of the
16690 * swash, since we will be accessing it
16691 * anyway, to save a little time */
16692 |_CORE_SWASH_INIT_ACCEPT_INVLIST;
16694 if (RExC_parse >= RExC_end)
16695 vFAIL2("Empty \\%c", (U8)value);
16696 if (*RExC_parse == '{') {
16697 const U8 c = (U8)value;
16698 e = (char *) memchr(RExC_parse, '}', RExC_end - RExC_parse);
16701 vFAIL2("Missing right brace on \\%c{}", c);
16705 while (isSPACE(*RExC_parse)) {
16709 if (UCHARAT(RExC_parse) == '^') {
16711 /* toggle. (The rhs xor gets the single bit that
16712 * differs between P and p; the other xor inverts just
16714 value ^= 'P' ^ 'p';
16717 while (isSPACE(*RExC_parse)) {
16722 if (e == RExC_parse)
16723 vFAIL2("Empty \\%c{}", c);
16725 n = e - RExC_parse;
16726 while (isSPACE(*(RExC_parse + n - 1)))
16728 } /* The \p isn't immediately followed by a '{' */
16729 else if (! isALPHA(*RExC_parse)) {
16730 RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1;
16731 vFAIL2("Character following \\%c must be '{' or a "
16732 "single-character Unicode property name",
16742 char* base_name; /* name after any packages are stripped */
16743 char* lookup_name = NULL;
16744 const char * const colon_colon = "::";
16746 /* Try to get the definition of the property into
16747 * <invlist>. If /i is in effect, the effective property
16748 * will have its name be <__NAME_i>. The design is
16749 * discussed in commit
16750 * 2f833f5208e26b208886e51e09e2c072b5eabb46 */
16751 name = savepv(Perl_form(aTHX_ "%.*s", (int)n, RExC_parse));
16754 lookup_name = savepv(Perl_form(aTHX_ "__%s_i", name));
16756 /* The function call just below that uses this can fail
16757 * to return, leaking memory if we don't do this */
16758 SAVEFREEPV(lookup_name);
16761 /* Look up the property name, and get its swash and
16762 * inversion list, if the property is found */
16763 SvREFCNT_dec(swash); /* Free any left-overs */
16764 swash = _core_swash_init("utf8",
16771 NULL, /* No inversion list */
16774 if (! swash || ! (invlist = _get_swash_invlist(swash))) {
16775 HV* curpkg = (IN_PERL_COMPILETIME)
16777 : CopSTASH(PL_curcop);
16781 if (swash) { /* Got a swash but no inversion list.
16782 Something is likely wrong that will
16783 be sorted-out later */
16784 SvREFCNT_dec_NN(swash);
16788 /* Here didn't find it. It could be a an error (like a
16789 * typo) in specifying a Unicode property, or it could
16790 * be a user-defined property that will be available at
16791 * run-time. The names of these must begin with 'In'
16792 * or 'Is' (after any packages are stripped off). So
16793 * if not one of those, or if we accept only
16794 * compile-time properties, is an error; otherwise add
16795 * it to the list for run-time look up. */
16796 if ((base_name = rninstr(name, name + n,
16797 colon_colon, colon_colon + 2)))
16798 { /* Has ::. We know this must be a user-defined
16801 final_n -= base_name - name;
16810 || base_name[0] != 'I'
16811 || (base_name[1] != 's' && base_name[1] != 'n')
16814 const char * const msg
16816 ? "Illegal user-defined property name"
16817 : "Can't find Unicode property definition";
16818 RExC_parse = e + 1;
16820 /* diag_listed_as: Can't find Unicode property definition "%s" */
16821 vFAIL3utf8f("%s \"%" UTF8f "\"",
16822 msg, UTF8fARG(UTF, n, name));
16825 /* If the property name doesn't already have a package
16826 * name, add the current one to it so that it can be
16827 * referred to outside it. [perl #121777] */
16828 if (! has_pkg && curpkg) {
16829 char* pkgname = HvNAME(curpkg);
16830 if (memNEs(pkgname, HvNAMELEN(curpkg), "main")) {
16831 char* full_name = Perl_form(aTHX_
16835 n = strlen(full_name);
16836 name = savepvn(full_name, n);
16840 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s%" UTF8f "%s\n",
16841 (value == 'p' ? '+' : '!'),
16842 (FOLD) ? "__" : "",
16843 UTF8fARG(UTF, n, name),
16844 (FOLD) ? "_i" : "");
16845 has_user_defined_property = TRUE;
16846 optimizable = FALSE; /* Will have to leave this an
16849 /* We don't know yet what this matches, so have to flag
16851 ANYOF_FLAGS(ret) |= ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP;
16855 /* Here, did get the swash and its inversion list. If
16856 * the swash is from a user-defined property, then this
16857 * whole character class should be regarded as such */
16858 if (swash_init_flags
16859 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)
16861 has_user_defined_property = TRUE;
16864 /* We warn on matching an above-Unicode code point
16865 * if the match would return true, except don't
16866 * warn for \p{All}, which has exactly one element
16868 (_invlist_contains_cp(invlist, 0x110000)
16869 && (! (_invlist_len(invlist) == 1
16870 && *invlist_array(invlist) == 0)))
16876 /* Invert if asking for the complement */
16877 if (value == 'P') {
16878 _invlist_union_complement_2nd(properties,
16882 /* The swash can't be used as-is, because we've
16883 * inverted things; delay removing it to here after
16884 * have copied its invlist above */
16885 SvREFCNT_dec_NN(swash);
16889 _invlist_union(properties, invlist, &properties);
16893 RExC_parse = e + 1;
16894 namedclass = ANYOF_UNIPROP; /* no official name, but it's
16897 /* \p means they want Unicode semantics */
16898 REQUIRE_UNI_RULES(flagp, NULL);
16901 case 'n': value = '\n'; break;
16902 case 'r': value = '\r'; break;
16903 case 't': value = '\t'; break;
16904 case 'f': value = '\f'; break;
16905 case 'b': value = '\b'; break;
16906 case 'e': value = ESC_NATIVE; break;
16907 case 'a': value = '\a'; break;
16909 RExC_parse--; /* function expects to be pointed at the 'o' */
16911 const char* error_msg;
16912 bool valid = grok_bslash_o(&RExC_parse,
16916 PASS2, /* warnings only in
16919 silence_non_portable,
16925 non_portable_endpoint++;
16928 RExC_parse--; /* function expects to be pointed at the 'x' */
16930 const char* error_msg;
16931 bool valid = grok_bslash_x(&RExC_parse,
16935 PASS2, /* Output warnings */
16937 silence_non_portable,
16943 non_portable_endpoint++;
16946 value = grok_bslash_c(*RExC_parse++, PASS2);
16947 non_portable_endpoint++;
16949 case '0': case '1': case '2': case '3': case '4':
16950 case '5': case '6': case '7':
16952 /* Take 1-3 octal digits */
16953 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
16954 numlen = (strict) ? 4 : 3;
16955 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
16956 RExC_parse += numlen;
16959 RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1;
16960 vFAIL("Need exactly 3 octal digits");
16962 else if (! SIZE_ONLY /* like \08, \178 */
16964 && RExC_parse < RExC_end
16965 && isDIGIT(*RExC_parse)
16966 && ckWARN(WARN_REGEXP))
16968 SAVEFREESV(RExC_rx_sv);
16969 reg_warn_non_literal_string(
16971 form_short_octal_warning(RExC_parse, numlen));
16972 (void)ReREFCNT_inc(RExC_rx_sv);
16975 non_portable_endpoint++;
16979 /* Allow \_ to not give an error */
16980 if (!SIZE_ONLY && isWORDCHAR(value) && value != '_') {
16982 vFAIL2("Unrecognized escape \\%c in character class",
16986 SAVEFREESV(RExC_rx_sv);
16987 ckWARN2reg(RExC_parse,
16988 "Unrecognized escape \\%c in character class passed through",
16990 (void)ReREFCNT_inc(RExC_rx_sv);
16994 } /* End of switch on char following backslash */
16995 } /* end of handling backslash escape sequences */
16997 /* Here, we have the current token in 'value' */
16999 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
17002 /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a
17003 * literal, as is the character that began the false range, i.e.
17004 * the 'a' in the examples */
17007 const int w = (RExC_parse >= rangebegin)
17008 ? RExC_parse - rangebegin
17012 "False [] range \"%" UTF8f "\"",
17013 UTF8fARG(UTF, w, rangebegin));
17016 SAVEFREESV(RExC_rx_sv); /* in case of fatal warnings */
17017 ckWARN2reg(RExC_parse,
17018 "False [] range \"%" UTF8f "\"",
17019 UTF8fARG(UTF, w, rangebegin));
17020 (void)ReREFCNT_inc(RExC_rx_sv);
17021 cp_list = add_cp_to_invlist(cp_list, '-');
17022 cp_foldable_list = add_cp_to_invlist(cp_foldable_list,
17027 range = 0; /* this was not a true range */
17028 element_count += 2; /* So counts for three values */
17031 classnum = namedclass_to_classnum(namedclass);
17033 if (LOC && namedclass < ANYOF_POSIXL_MAX
17034 #ifndef HAS_ISASCII
17035 && classnum != _CC_ASCII
17038 /* What the Posix classes (like \w, [:space:]) match in locale
17039 * isn't knowable under locale until actual match time. Room
17040 * must be reserved (one time per outer bracketed class) to
17041 * store such classes. The space will contain a bit for each
17042 * named class that is to be matched against. This isn't
17043 * needed for \p{} and pseudo-classes, as they are not affected
17044 * by locale, and hence are dealt with separately */
17045 if (! need_class) {
17048 RExC_size += ANYOF_POSIXL_SKIP - ANYOF_SKIP;
17051 RExC_emit += ANYOF_POSIXL_SKIP - ANYOF_SKIP;
17053 ANYOF_FLAGS(ret) |= ANYOF_MATCHES_POSIXL;
17054 ANYOF_POSIXL_ZERO(ret);
17056 /* We can't change this into some other type of node
17057 * (unless this is the only element, in which case there
17058 * are nodes that mean exactly this) as has runtime
17060 optimizable = FALSE;
17063 /* Coverity thinks it is possible for this to be negative; both
17064 * jhi and khw think it's not, but be safer */
17065 assert(! (ANYOF_FLAGS(ret) & ANYOF_MATCHES_POSIXL)
17066 || (namedclass + ((namedclass % 2) ? -1 : 1)) >= 0);
17068 /* See if it already matches the complement of this POSIX
17070 if ((ANYOF_FLAGS(ret) & ANYOF_MATCHES_POSIXL)
17071 && ANYOF_POSIXL_TEST(ret, namedclass + ((namedclass % 2)
17075 posixl_matches_all = TRUE;
17076 break; /* No need to continue. Since it matches both
17077 e.g., \w and \W, it matches everything, and the
17078 bracketed class can be optimized into qr/./s */
17081 /* Add this class to those that should be checked at runtime */
17082 ANYOF_POSIXL_SET(ret, namedclass);
17084 /* The above-Latin1 characters are not subject to locale rules.
17085 * Just add them, in the second pass, to the
17086 * unconditionally-matched list */
17088 SV* scratch_list = NULL;
17090 /* Get the list of the above-Latin1 code points this
17092 _invlist_intersection_maybe_complement_2nd(PL_AboveLatin1,
17093 PL_XPosix_ptrs[classnum],
17095 /* Odd numbers are complements, like
17096 * NDIGIT, NASCII, ... */
17097 namedclass % 2 != 0,
17099 /* Checking if 'cp_list' is NULL first saves an extra
17100 * clone. Its reference count will be decremented at the
17101 * next union, etc, or if this is the only instance, at the
17102 * end of the routine */
17104 cp_list = scratch_list;
17107 _invlist_union(cp_list, scratch_list, &cp_list);
17108 SvREFCNT_dec_NN(scratch_list);
17110 continue; /* Go get next character */
17113 else if (! SIZE_ONLY) {
17115 /* Here, not in pass1 (in that pass we skip calculating the
17116 * contents of this class), and is not /l, or is a POSIX class
17117 * for which /l doesn't matter (or is a Unicode property, which
17118 * is skipped here). */
17119 if (namedclass >= ANYOF_POSIXL_MAX) { /* If a special class */
17120 if (namedclass != ANYOF_UNIPROP) { /* UNIPROP = \p and \P */
17122 /* Here, should be \h, \H, \v, or \V. None of /d, /i
17123 * nor /l make a difference in what these match,
17124 * therefore we just add what they match to cp_list. */
17125 if (classnum != _CC_VERTSPACE) {
17126 assert( namedclass == ANYOF_HORIZWS
17127 || namedclass == ANYOF_NHORIZWS);
17129 /* It turns out that \h is just a synonym for
17131 classnum = _CC_BLANK;
17134 _invlist_union_maybe_complement_2nd(
17136 PL_XPosix_ptrs[classnum],
17137 namedclass % 2 != 0, /* Complement if odd
17138 (NHORIZWS, NVERTWS)
17143 else if ( UNI_SEMANTICS
17144 || classnum == _CC_ASCII
17145 || (DEPENDS_SEMANTICS && ( classnum == _CC_DIGIT
17146 || classnum == _CC_XDIGIT)))
17148 /* We usually have to worry about /d and /a affecting what
17149 * POSIX classes match, with special code needed for /d
17150 * because we won't know until runtime what all matches.
17151 * But there is no extra work needed under /u, and
17152 * [:ascii:] is unaffected by /a and /d; and :digit: and
17153 * :xdigit: don't have runtime differences under /d. So we
17154 * can special case these, and avoid some extra work below,
17155 * and at runtime. */
17156 _invlist_union_maybe_complement_2nd(
17158 PL_XPosix_ptrs[classnum],
17159 namedclass % 2 != 0,
17162 else { /* Garden variety class. If is NUPPER, NALPHA, ...
17163 complement and use nposixes */
17164 SV** posixes_ptr = namedclass % 2 == 0
17167 _invlist_union_maybe_complement_2nd(
17169 PL_XPosix_ptrs[classnum],
17170 namedclass % 2 != 0,
17174 } /* end of namedclass \blah */
17176 SKIP_BRACKETED_WHITE_SPACE(skip_white, RExC_parse);
17178 /* If 'range' is set, 'value' is the ending of a range--check its
17179 * validity. (If value isn't a single code point in the case of a
17180 * range, we should have figured that out above in the code that
17181 * catches false ranges). Later, we will handle each individual code
17182 * point in the range. If 'range' isn't set, this could be the
17183 * beginning of a range, so check for that by looking ahead to see if
17184 * the next real character to be processed is the range indicator--the
17189 /* For unicode ranges, we have to test that the Unicode as opposed
17190 * to the native values are not decreasing. (Above 255, there is
17191 * no difference between native and Unicode) */
17192 if (unicode_range && prevvalue < 255 && value < 255) {
17193 if (NATIVE_TO_LATIN1(prevvalue) > NATIVE_TO_LATIN1(value)) {
17194 goto backwards_range;
17199 if (prevvalue > value) /* b-a */ {
17204 w = RExC_parse - rangebegin;
17206 "Invalid [] range \"%" UTF8f "\"",
17207 UTF8fARG(UTF, w, rangebegin));
17208 NOT_REACHED; /* NOTREACHED */
17212 prevvalue = value; /* save the beginning of the potential range */
17213 if (! stop_at_1 /* Can't be a range if parsing just one thing */
17214 && *RExC_parse == '-')
17216 char* next_char_ptr = RExC_parse + 1;
17218 /* Get the next real char after the '-' */
17219 SKIP_BRACKETED_WHITE_SPACE(skip_white, next_char_ptr);
17221 /* If the '-' is at the end of the class (just before the ']',
17222 * it is a literal minus; otherwise it is a range */
17223 if (next_char_ptr < RExC_end && *next_char_ptr != ']') {
17224 RExC_parse = next_char_ptr;
17226 /* a bad range like \w-, [:word:]- ? */
17227 if (namedclass > OOB_NAMEDCLASS) {
17228 if (strict || (PASS2 && ckWARN(WARN_REGEXP))) {
17229 const int w = RExC_parse >= rangebegin
17230 ? RExC_parse - rangebegin
17233 vFAIL4("False [] range \"%*.*s\"",
17238 "False [] range \"%*.*s\"",
17243 cp_list = add_cp_to_invlist(cp_list, '-');
17247 range = 1; /* yeah, it's a range! */
17248 continue; /* but do it the next time */
17253 if (namedclass > OOB_NAMEDCLASS) {
17257 /* Here, we have a single value this time through the loop, and
17258 * <prevvalue> is the beginning of the range, if any; or <value> if
17261 /* non-Latin1 code point implies unicode semantics. Must be set in
17262 * pass1 so is there for the whole of pass 2 */
17264 REQUIRE_UNI_RULES(flagp, NULL);
17267 /* Ready to process either the single value, or the completed range.
17268 * For single-valued non-inverted ranges, we consider the possibility
17269 * of multi-char folds. (We made a conscious decision to not do this
17270 * for the other cases because it can often lead to non-intuitive
17271 * results. For example, you have the peculiar case that:
17272 * "s s" =~ /^[^\xDF]+$/i => Y
17273 * "ss" =~ /^[^\xDF]+$/i => N
17275 * See [perl #89750] */
17276 if (FOLD && allow_multi_folds && value == prevvalue) {
17277 if (value == LATIN_SMALL_LETTER_SHARP_S
17278 || (value > 255 && _invlist_contains_cp(PL_HasMultiCharFold,
17281 /* Here <value> is indeed a multi-char fold. Get what it is */
17283 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
17286 UV folded = _to_uni_fold_flags(
17290 FOLD_FLAGS_FULL | (ASCII_FOLD_RESTRICTED
17291 ? FOLD_FLAGS_NOMIX_ASCII
17295 /* Here, <folded> should be the first character of the
17296 * multi-char fold of <value>, with <foldbuf> containing the
17297 * whole thing. But, if this fold is not allowed (because of
17298 * the flags), <fold> will be the same as <value>, and should
17299 * be processed like any other character, so skip the special
17301 if (folded != value) {
17303 /* Skip if we are recursed, currently parsing the class
17304 * again. Otherwise add this character to the list of
17305 * multi-char folds. */
17306 if (! RExC_in_multi_char_class) {
17307 STRLEN cp_count = utf8_length(foldbuf,
17308 foldbuf + foldlen);
17309 SV* multi_fold = sv_2mortal(newSVpvs(""));
17311 Perl_sv_catpvf(aTHX_ multi_fold, "\\x{%" UVXf "}", value);
17314 = add_multi_match(multi_char_matches,
17320 /* This element should not be processed further in this
17323 value = save_value;
17324 prevvalue = save_prevvalue;
17330 if (strict && PASS2 && ckWARN(WARN_REGEXP)) {
17333 /* If the range starts above 255, everything is portable and
17334 * likely to be so for any forseeable character set, so don't
17336 if (unicode_range && non_portable_endpoint && prevvalue < 256) {
17337 vWARN(RExC_parse, "Both or neither range ends should be Unicode");
17339 else if (prevvalue != value) {
17341 /* Under strict, ranges that stop and/or end in an ASCII
17342 * printable should have each end point be a portable value
17343 * for it (preferably like 'A', but we don't warn if it is
17344 * a (portable) Unicode name or code point), and the range
17345 * must be be all digits or all letters of the same case.
17346 * Otherwise, the range is non-portable and unclear as to
17347 * what it contains */
17348 if ( (isPRINT_A(prevvalue) || isPRINT_A(value))
17349 && ( non_portable_endpoint
17350 || ! ( (isDIGIT_A(prevvalue) && isDIGIT_A(value))
17351 || (isLOWER_A(prevvalue) && isLOWER_A(value))
17352 || (isUPPER_A(prevvalue) && isUPPER_A(value))
17354 vWARN(RExC_parse, "Ranges of ASCII printables should"
17355 " be some subset of \"0-9\","
17356 " \"A-Z\", or \"a-z\"");
17358 else if (prevvalue >= FIRST_NON_ASCII_DECIMAL_DIGIT) {
17359 SSize_t index_start;
17360 SSize_t index_final;
17362 /* But the nature of Unicode and languages mean we
17363 * can't do the same checks for above-ASCII ranges,
17364 * except in the case of digit ones. These should
17365 * contain only digits from the same group of 10. The
17366 * ASCII case is handled just above. Hence here, the
17367 * range could be a range of digits. First some
17368 * unlikely special cases. Grandfather in that a range
17369 * ending in 19DA (NEW TAI LUE THAM DIGIT ONE) is bad
17370 * if its starting value is one of the 10 digits prior
17371 * to it. This is because it is an alternate way of
17372 * writing 19D1, and some people may expect it to be in
17373 * that group. But it is bad, because it won't give
17374 * the expected results. In Unicode 5.2 it was
17375 * considered to be in that group (of 11, hence), but
17376 * this was fixed in the next version */
17378 if (UNLIKELY(value == 0x19DA && prevvalue >= 0x19D0)) {
17379 goto warn_bad_digit_range;
17381 else if (UNLIKELY( prevvalue >= 0x1D7CE
17382 && value <= 0x1D7FF))
17384 /* This is the only other case currently in Unicode
17385 * where the algorithm below fails. The code
17386 * points just above are the end points of a single
17387 * range containing only decimal digits. It is 5
17388 * different series of 0-9. All other ranges of
17389 * digits currently in Unicode are just a single
17390 * series. (And mktables will notify us if a later
17391 * Unicode version breaks this.)
17393 * If the range being checked is at most 9 long,
17394 * and the digit values represented are in
17395 * numerical order, they are from the same series.
17397 if ( value - prevvalue > 9
17398 || ((( value - 0x1D7CE) % 10)
17399 <= (prevvalue - 0x1D7CE) % 10))
17401 goto warn_bad_digit_range;
17406 /* For all other ranges of digits in Unicode, the
17407 * algorithm is just to check if both end points
17408 * are in the same series, which is the same range.
17410 index_start = _invlist_search(
17411 PL_XPosix_ptrs[_CC_DIGIT],
17414 /* Warn if the range starts and ends with a digit,
17415 * and they are not in the same group of 10. */
17416 if ( index_start >= 0
17417 && ELEMENT_RANGE_MATCHES_INVLIST(index_start)
17419 _invlist_search(PL_XPosix_ptrs[_CC_DIGIT],
17420 value)) != index_start
17421 && index_final >= 0
17422 && ELEMENT_RANGE_MATCHES_INVLIST(index_final))
17424 warn_bad_digit_range:
17425 vWARN(RExC_parse, "Ranges of digits should be"
17426 " from the same group of"
17433 if ((! range || prevvalue == value) && non_portable_endpoint) {
17434 if (isPRINT_A(value)) {
17437 if (isBACKSLASHED_PUNCT(value)) {
17438 literal[d++] = '\\';
17440 literal[d++] = (char) value;
17441 literal[d++] = '\0';
17444 "\"%.*s\" is more clearly written simply as \"%s\"",
17445 (int) (RExC_parse - rangebegin),
17450 else if isMNEMONIC_CNTRL(value) {
17452 "\"%.*s\" is more clearly written simply as \"%s\"",
17453 (int) (RExC_parse - rangebegin),
17455 cntrl_to_mnemonic((U8) value)
17461 /* Deal with this element of the class */
17465 cp_foldable_list = _add_range_to_invlist(cp_foldable_list,
17468 /* On non-ASCII platforms, for ranges that span all of 0..255, and
17469 * ones that don't require special handling, we can just add the
17470 * range like we do for ASCII platforms */
17471 if ((UNLIKELY(prevvalue == 0) && value >= 255)
17472 || ! (prevvalue < 256
17474 || (! non_portable_endpoint
17475 && ((isLOWER_A(prevvalue) && isLOWER_A(value))
17476 || (isUPPER_A(prevvalue)
17477 && isUPPER_A(value)))))))
17479 cp_foldable_list = _add_range_to_invlist(cp_foldable_list,
17483 /* Here, requires special handling. This can be because it is
17484 * a range whose code points are considered to be Unicode, and
17485 * so must be individually translated into native, or because
17486 * its a subrange of 'A-Z' or 'a-z' which each aren't
17487 * contiguous in EBCDIC, but we have defined them to include
17488 * only the "expected" upper or lower case ASCII alphabetics.
17489 * Subranges above 255 are the same in native and Unicode, so
17490 * can be added as a range */
17491 U8 start = NATIVE_TO_LATIN1(prevvalue);
17493 U8 end = (value < 256) ? NATIVE_TO_LATIN1(value) : 255;
17494 for (j = start; j <= end; j++) {
17495 cp_foldable_list = add_cp_to_invlist(cp_foldable_list, LATIN1_TO_NATIVE(j));
17498 cp_foldable_list = _add_range_to_invlist(cp_foldable_list,
17505 range = 0; /* this range (if it was one) is done now */
17506 } /* End of loop through all the text within the brackets */
17509 if ( posix_warnings && av_tindex_skip_len_mg(posix_warnings) >= 0) {
17510 output_or_return_posix_warnings(pRExC_state, posix_warnings,
17511 return_posix_warnings);
17514 /* If anything in the class expands to more than one character, we have to
17515 * deal with them by building up a substitute parse string, and recursively
17516 * calling reg() on it, instead of proceeding */
17517 if (multi_char_matches) {
17518 SV * substitute_parse = newSVpvn_flags("?:", 2, SVs_TEMP);
17521 char *save_end = RExC_end;
17522 char *save_parse = RExC_parse;
17523 char *save_start = RExC_start;
17524 STRLEN prefix_end = 0; /* We copy the character class after a
17525 prefix supplied here. This is the size
17526 + 1 of that prefix */
17527 bool first_time = TRUE; /* First multi-char occurrence doesn't get
17532 assert(RExC_precomp_adj == 0); /* Only one level of recursion allowed */
17534 #if 0 /* Have decided not to deal with multi-char folds in inverted classes,
17535 because too confusing */
17537 sv_catpv(substitute_parse, "(?:");
17541 /* Look at the longest folds first */
17542 for (cp_count = av_tindex_skip_len_mg(multi_char_matches);
17547 if (av_exists(multi_char_matches, cp_count)) {
17548 AV** this_array_ptr;
17551 this_array_ptr = (AV**) av_fetch(multi_char_matches,
17553 while ((this_sequence = av_pop(*this_array_ptr)) !=
17556 if (! first_time) {
17557 sv_catpv(substitute_parse, "|");
17559 first_time = FALSE;
17561 sv_catpv(substitute_parse, SvPVX(this_sequence));
17566 /* If the character class contains anything else besides these
17567 * multi-character folds, have to include it in recursive parsing */
17568 if (element_count) {
17569 sv_catpv(substitute_parse, "|[");
17570 prefix_end = SvCUR(substitute_parse);
17571 sv_catpvn(substitute_parse, orig_parse, RExC_parse - orig_parse);
17573 /* Put in a closing ']' only if not going off the end, as otherwise
17574 * we are adding something that really isn't there */
17575 if (RExC_parse < RExC_end) {
17576 sv_catpv(substitute_parse, "]");
17580 sv_catpv(substitute_parse, ")");
17583 /* This is a way to get the parse to skip forward a whole named
17584 * sequence instead of matching the 2nd character when it fails the
17586 sv_catpv(substitute_parse, "(*THEN)(*SKIP)(*FAIL)|.)");
17590 /* Set up the data structure so that any errors will be properly
17591 * reported. See the comments at the definition of
17592 * REPORT_LOCATION_ARGS for details */
17593 RExC_precomp_adj = orig_parse - RExC_precomp;
17594 RExC_start = RExC_parse = SvPV(substitute_parse, len);
17595 RExC_adjusted_start = RExC_start + prefix_end;
17596 RExC_end = RExC_parse + len;
17597 RExC_in_multi_char_class = 1;
17598 RExC_emit = (regnode *)orig_emit;
17600 ret = reg(pRExC_state, 1, ®_flags, depth+1);
17602 *flagp |= reg_flags & (HASWIDTH|SIMPLE|SPSTART|POSTPONED|RESTART_PASS1|NEED_UTF8);
17604 /* And restore so can parse the rest of the pattern */
17605 RExC_parse = save_parse;
17606 RExC_start = RExC_adjusted_start = save_start;
17607 RExC_precomp_adj = 0;
17608 RExC_end = save_end;
17609 RExC_in_multi_char_class = 0;
17610 SvREFCNT_dec_NN(multi_char_matches);
17614 /* Here, we've gone through the entire class and dealt with multi-char
17615 * folds. We are now in a position that we can do some checks to see if we
17616 * can optimize this ANYOF node into a simpler one, even in Pass 1.
17617 * Currently we only do two checks:
17618 * 1) is in the unlikely event that the user has specified both, eg. \w and
17619 * \W under /l, then the class matches everything. (This optimization
17620 * is done only to make the optimizer code run later work.)
17621 * 2) if the character class contains only a single element (including a
17622 * single range), we see if there is an equivalent node for it.
17623 * Other checks are possible */
17625 && ! ret_invlist /* Can't optimize if returning the constructed
17627 && (UNLIKELY(posixl_matches_all) || element_count == 1))
17632 if (UNLIKELY(posixl_matches_all)) {
17635 else if (namedclass > OOB_NAMEDCLASS) { /* this is a single named
17636 class, like \w or [:digit:]
17639 /* All named classes are mapped into POSIXish nodes, with its FLAG
17640 * argument giving which class it is */
17641 switch ((I32)namedclass) {
17642 case ANYOF_UNIPROP:
17645 /* These don't depend on the charset modifiers. They always
17646 * match under /u rules */
17647 case ANYOF_NHORIZWS:
17648 case ANYOF_HORIZWS:
17649 namedclass = ANYOF_BLANK + namedclass - ANYOF_HORIZWS;
17652 case ANYOF_NVERTWS:
17657 /* The actual POSIXish node for all the rest depends on the
17658 * charset modifier. The ones in the first set depend only on
17659 * ASCII or, if available on this platform, also locale */
17670 /* (named_class - ANYOF_ASCII) is 0 or 1. xor'ing with
17671 * invert converts that to 1 or 0 */
17672 op = ASCII + ((namedclass - ANYOF_ASCII) ^ invert);
17675 /* The following don't have any matches in the upper Latin1
17676 * range, hence /d is equivalent to /u for them. Making it /u
17677 * saves some branches at runtime */
17681 case ANYOF_NXDIGIT:
17682 if (! DEPENDS_SEMANTICS) {
17683 goto treat_as_default;
17689 /* The following change to CASED under /i */
17695 namedclass = ANYOF_CASED + (namedclass % 2);
17699 /* The rest have more possibilities depending on the charset.
17700 * We take advantage of the enum ordering of the charset
17701 * modifiers to get the exact node type, */
17704 op = POSIXD + get_regex_charset(RExC_flags);
17705 if (op > POSIXA) { /* /aa is same as /a */
17710 /* The odd numbered ones are the complements of the
17711 * next-lower even number one */
17712 if (namedclass % 2 == 1) {
17716 arg = namedclass_to_classnum(namedclass);
17720 else if (value == prevvalue) {
17722 /* Here, the class consists of just a single code point */
17725 if (! LOC && value == '\n') {
17726 op = REG_ANY; /* Optimize [^\n] */
17727 *flagp |= HASWIDTH|SIMPLE;
17731 else if (value < 256 || UTF) {
17733 /* Optimize a single value into an EXACTish node, but not if it
17734 * would require converting the pattern to UTF-8. */
17735 op = compute_EXACTish(pRExC_state);
17737 } /* Otherwise is a range */
17738 else if (! LOC) { /* locale could vary these */
17739 if (prevvalue == '0') {
17740 if (value == '9') {
17745 else if (! FOLD || ASCII_FOLD_RESTRICTED) {
17746 /* We can optimize A-Z or a-z, but not if they could match
17747 * something like the KELVIN SIGN under /i. */
17748 if (prevvalue == 'A') {
17751 && ! non_portable_endpoint
17754 arg = (FOLD) ? _CC_ALPHA : _CC_UPPER;
17758 else if (prevvalue == 'a') {
17761 && ! non_portable_endpoint
17764 arg = (FOLD) ? _CC_ALPHA : _CC_LOWER;
17771 /* Here, we have changed <op> away from its initial value iff we found
17772 * an optimization */
17775 /* Throw away this ANYOF regnode, and emit the calculated one,
17776 * which should correspond to the beginning, not current, state of
17778 const char * cur_parse = RExC_parse;
17779 RExC_parse = (char *)orig_parse;
17783 /* To get locale nodes to not use the full ANYOF size would
17784 * require moving the code above that writes the portions
17785 * of it that aren't in other nodes to after this point.
17786 * e.g. ANYOF_POSIXL_SET */
17787 RExC_size = orig_size;
17791 RExC_emit = (regnode *)orig_emit;
17792 if (PL_regkind[op] == POSIXD) {
17793 if (op == POSIXL) {
17794 RExC_contains_locale = 1;
17797 op += NPOSIXD - POSIXD;
17802 ret = reg_node(pRExC_state, op);
17804 if (PL_regkind[op] == POSIXD || PL_regkind[op] == NPOSIXD) {
17808 *flagp |= HASWIDTH|SIMPLE;
17810 else if (PL_regkind[op] == EXACT) {
17811 alloc_maybe_populate_EXACT(pRExC_state, ret, flagp, 0, value,
17812 TRUE /* downgradable to EXACT */
17816 *flagp |= HASWIDTH|SIMPLE;
17819 RExC_parse = (char *) cur_parse;
17821 SvREFCNT_dec(posixes);
17822 SvREFCNT_dec(nposixes);
17823 SvREFCNT_dec(simple_posixes);
17824 SvREFCNT_dec(cp_list);
17825 SvREFCNT_dec(cp_foldable_list);
17832 /****** !SIZE_ONLY (Pass 2) AFTER HERE *********/
17834 /* If folding, we calculate all characters that could fold to or from the
17835 * ones already on the list */
17836 if (cp_foldable_list) {
17838 UV start, end; /* End points of code point ranges */
17840 SV* fold_intersection = NULL;
17843 /* Our calculated list will be for Unicode rules. For locale
17844 * matching, we have to keep a separate list that is consulted at
17845 * runtime only when the locale indicates Unicode rules. For
17846 * non-locale, we just use the general list */
17848 use_list = &only_utf8_locale_list;
17851 use_list = &cp_list;
17854 /* Only the characters in this class that participate in folds need
17855 * be checked. Get the intersection of this class and all the
17856 * possible characters that are foldable. This can quickly narrow
17857 * down a large class */
17858 _invlist_intersection(PL_utf8_foldable, cp_foldable_list,
17859 &fold_intersection);
17861 /* The folds for all the Latin1 characters are hard-coded into this
17862 * program, but we have to go out to disk to get the others. */
17863 if (invlist_highest(cp_foldable_list) >= 256) {
17865 /* This is a hash that for a particular fold gives all
17866 * characters that are involved in it */
17867 if (! PL_utf8_foldclosures) {
17868 _load_PL_utf8_foldclosures();
17872 /* Now look at the foldable characters in this class individually */
17873 invlist_iterinit(fold_intersection);
17874 while (invlist_iternext(fold_intersection, &start, &end)) {
17877 /* Look at every character in the range */
17878 for (j = start; j <= end; j++) {
17879 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
17885 if (IS_IN_SOME_FOLD_L1(j)) {
17887 /* ASCII is always matched; non-ASCII is matched
17888 * only under Unicode rules (which could happen
17889 * under /l if the locale is a UTF-8 one */
17890 if (isASCII(j) || ! DEPENDS_SEMANTICS) {
17891 *use_list = add_cp_to_invlist(*use_list,
17892 PL_fold_latin1[j]);
17895 has_upper_latin1_only_utf8_matches
17896 = add_cp_to_invlist(
17897 has_upper_latin1_only_utf8_matches,
17898 PL_fold_latin1[j]);
17902 if (HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(j)
17903 && (! isASCII(j) || ! ASCII_FOLD_RESTRICTED))
17905 add_above_Latin1_folds(pRExC_state,
17912 /* Here is an above Latin1 character. We don't have the
17913 * rules hard-coded for it. First, get its fold. This is
17914 * the simple fold, as the multi-character folds have been
17915 * handled earlier and separated out */
17916 _to_uni_fold_flags(j, foldbuf, &foldlen,
17917 (ASCII_FOLD_RESTRICTED)
17918 ? FOLD_FLAGS_NOMIX_ASCII
17921 /* Single character fold of above Latin1. Add everything in
17922 * its fold closure to the list that this node should match.
17923 * The fold closures data structure is a hash with the keys
17924 * being the UTF-8 of every character that is folded to, like
17925 * 'k', and the values each an array of all code points that
17926 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
17927 * Multi-character folds are not included */
17928 if ((listp = hv_fetch(PL_utf8_foldclosures,
17929 (char *) foldbuf, foldlen, FALSE)))
17931 AV* list = (AV*) *listp;
17933 for (k = 0; k <= av_tindex_skip_len_mg(list); k++) {
17934 SV** c_p = av_fetch(list, k, FALSE);
17940 /* /aa doesn't allow folds between ASCII and non- */
17941 if ((ASCII_FOLD_RESTRICTED
17942 && (isASCII(c) != isASCII(j))))
17947 /* Folds under /l which cross the 255/256 boundary
17948 * are added to a separate list. (These are valid
17949 * only when the locale is UTF-8.) */
17950 if (c < 256 && LOC) {
17951 *use_list = add_cp_to_invlist(*use_list, c);
17955 if (isASCII(c) || c > 255 || AT_LEAST_UNI_SEMANTICS)
17957 cp_list = add_cp_to_invlist(cp_list, c);
17960 /* Similarly folds involving non-ascii Latin1
17961 * characters under /d are added to their list */
17962 has_upper_latin1_only_utf8_matches
17963 = add_cp_to_invlist(
17964 has_upper_latin1_only_utf8_matches,
17971 SvREFCNT_dec_NN(fold_intersection);
17974 /* Now that we have finished adding all the folds, there is no reason
17975 * to keep the foldable list separate */
17976 _invlist_union(cp_list, cp_foldable_list, &cp_list);
17977 SvREFCNT_dec_NN(cp_foldable_list);
17980 /* And combine the result (if any) with any inversion lists from posix
17981 * classes. The lists are kept separate up to now because we don't want to
17982 * fold the classes (folding of those is automatically handled by the swash
17983 * fetching code) */
17984 if (simple_posixes) { /* These are the classes known to be unaffected by
17987 _invlist_union(cp_list, simple_posixes, &cp_list);
17988 SvREFCNT_dec_NN(simple_posixes);
17991 cp_list = simple_posixes;
17994 if (posixes || nposixes) {
17996 /* We have to adjust /a and /aa */
17997 if (AT_LEAST_ASCII_RESTRICTED) {
17999 /* Under /a and /aa, nothing above ASCII matches these */
18001 _invlist_intersection(posixes,
18002 PL_XPosix_ptrs[_CC_ASCII],
18006 /* Under /a and /aa, everything above ASCII matches these
18009 _invlist_union_complement_2nd(nposixes,
18010 PL_XPosix_ptrs[_CC_ASCII],
18015 if (! DEPENDS_SEMANTICS) {
18017 /* For everything but /d, we can just add the current 'posixes' and
18018 * 'nposixes' to the main list */
18021 _invlist_union(cp_list, posixes, &cp_list);
18022 SvREFCNT_dec_NN(posixes);
18030 _invlist_union(cp_list, nposixes, &cp_list);
18031 SvREFCNT_dec_NN(nposixes);
18034 cp_list = nposixes;
18039 /* Under /d, things like \w match upper Latin1 characters only if
18040 * the target string is in UTF-8. But things like \W match all the
18041 * upper Latin1 characters if the target string is not in UTF-8.
18043 * Handle the case where there something like \W separately */
18045 SV* only_non_utf8_list = invlist_clone(PL_UpperLatin1);
18047 /* A complemented posix class matches all upper Latin1
18048 * characters if not in UTF-8. And it matches just certain
18049 * ones when in UTF-8. That means those certain ones are
18050 * matched regardless, so can just be added to the
18051 * unconditional list */
18053 _invlist_union(cp_list, nposixes, &cp_list);
18054 SvREFCNT_dec_NN(nposixes);
18058 cp_list = nposixes;
18061 /* Likewise for 'posixes' */
18062 _invlist_union(posixes, cp_list, &cp_list);
18064 /* Likewise for anything else in the range that matched only
18066 if (has_upper_latin1_only_utf8_matches) {
18067 _invlist_union(cp_list,
18068 has_upper_latin1_only_utf8_matches,
18070 SvREFCNT_dec_NN(has_upper_latin1_only_utf8_matches);
18071 has_upper_latin1_only_utf8_matches = NULL;
18074 /* If we don't match all the upper Latin1 characters regardless
18075 * of UTF-8ness, we have to set a flag to match the rest when
18077 _invlist_subtract(only_non_utf8_list, cp_list,
18078 &only_non_utf8_list);
18079 if (_invlist_len(only_non_utf8_list) != 0) {
18080 ANYOF_FLAGS(ret) |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
18082 SvREFCNT_dec_NN(only_non_utf8_list);
18085 /* Here there were no complemented posix classes. That means
18086 * the upper Latin1 characters in 'posixes' match only when the
18087 * target string is in UTF-8. So we have to add them to the
18088 * list of those types of code points, while adding the
18089 * remainder to the unconditional list.
18091 * First calculate what they are */
18092 SV* nonascii_but_latin1_properties = NULL;
18093 _invlist_intersection(posixes, PL_UpperLatin1,
18094 &nonascii_but_latin1_properties);
18096 /* And add them to the final list of such characters. */
18097 _invlist_union(has_upper_latin1_only_utf8_matches,
18098 nonascii_but_latin1_properties,
18099 &has_upper_latin1_only_utf8_matches);
18101 /* Remove them from what now becomes the unconditional list */
18102 _invlist_subtract(posixes, nonascii_but_latin1_properties,
18105 /* And add those unconditional ones to the final list */
18107 _invlist_union(cp_list, posixes, &cp_list);
18108 SvREFCNT_dec_NN(posixes);
18115 SvREFCNT_dec(nonascii_but_latin1_properties);
18117 /* Get rid of any characters that we now know are matched
18118 * unconditionally from the conditional list, which may make
18119 * that list empty */
18120 _invlist_subtract(has_upper_latin1_only_utf8_matches,
18122 &has_upper_latin1_only_utf8_matches);
18123 if (_invlist_len(has_upper_latin1_only_utf8_matches) == 0) {
18124 SvREFCNT_dec_NN(has_upper_latin1_only_utf8_matches);
18125 has_upper_latin1_only_utf8_matches = NULL;
18131 /* And combine the result (if any) with any inversion list from properties.
18132 * The lists are kept separate up to now so that we can distinguish the two
18133 * in regards to matching above-Unicode. A run-time warning is generated
18134 * if a Unicode property is matched against a non-Unicode code point. But,
18135 * we allow user-defined properties to match anything, without any warning,
18136 * and we also suppress the warning if there is a portion of the character
18137 * class that isn't a Unicode property, and which matches above Unicode, \W
18138 * or [\x{110000}] for example.
18139 * (Note that in this case, unlike the Posix one above, there is no
18140 * <has_upper_latin1_only_utf8_matches>, because having a Unicode property
18141 * forces Unicode semantics */
18145 /* If it matters to the final outcome, see if a non-property
18146 * component of the class matches above Unicode. If so, the
18147 * warning gets suppressed. This is true even if just a single
18148 * such code point is specified, as, though not strictly correct if
18149 * another such code point is matched against, the fact that they
18150 * are using above-Unicode code points indicates they should know
18151 * the issues involved */
18153 warn_super = ! (invert
18154 ^ (invlist_highest(cp_list) > PERL_UNICODE_MAX));
18157 _invlist_union(properties, cp_list, &cp_list);
18158 SvREFCNT_dec_NN(properties);
18161 cp_list = properties;
18166 |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
18168 /* Because an ANYOF node is the only one that warns, this node
18169 * can't be optimized into something else */
18170 optimizable = FALSE;
18174 /* Here, we have calculated what code points should be in the character
18177 * Now we can see about various optimizations. Fold calculation (which we
18178 * did above) needs to take place before inversion. Otherwise /[^k]/i
18179 * would invert to include K, which under /i would match k, which it
18180 * shouldn't. Therefore we can't invert folded locale now, as it won't be
18181 * folded until runtime */
18183 /* If we didn't do folding, it's because some information isn't available
18184 * until runtime; set the run-time fold flag for these. (We don't have to
18185 * worry about properties folding, as that is taken care of by the swash
18186 * fetching). We know to set the flag if we have a non-NULL list for UTF-8
18187 * locales, or the class matches at least one 0-255 range code point */
18190 /* Some things on the list might be unconditionally included because of
18191 * other components. Remove them, and clean up the list if it goes to
18193 if (only_utf8_locale_list && cp_list) {
18194 _invlist_subtract(only_utf8_locale_list, cp_list,
18195 &only_utf8_locale_list);
18197 if (_invlist_len(only_utf8_locale_list) == 0) {
18198 SvREFCNT_dec_NN(only_utf8_locale_list);
18199 only_utf8_locale_list = NULL;
18202 if (only_utf8_locale_list) {
18205 |ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
18207 else if (cp_list) { /* Look to see if a 0-255 code point is in list */
18209 invlist_iterinit(cp_list);
18210 if (invlist_iternext(cp_list, &start, &end) && start < 256) {
18211 ANYOF_FLAGS(ret) |= ANYOFL_FOLD;
18213 invlist_iterfinish(cp_list);
18216 else if ( DEPENDS_SEMANTICS
18217 && ( has_upper_latin1_only_utf8_matches
18218 || (ANYOF_FLAGS(ret) & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)))
18221 optimizable = FALSE;
18225 /* Optimize inverted simple patterns (e.g. [^a-z]) when everything is known
18226 * at compile time. Besides not inverting folded locale now, we can't
18227 * invert if there are things such as \w, which aren't known until runtime
18231 && OP(ret) != ANYOFD
18232 && ! (ANYOF_FLAGS(ret) & (ANYOF_LOCALE_FLAGS))
18233 && ! HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION)
18235 _invlist_invert(cp_list);
18237 /* Any swash can't be used as-is, because we've inverted things */
18239 SvREFCNT_dec_NN(swash);
18243 /* Clear the invert flag since have just done it here */
18250 *ret_invlist = cp_list;
18251 SvREFCNT_dec(swash);
18253 /* Discard the generated node */
18255 RExC_size = orig_size;
18258 RExC_emit = orig_emit;
18263 /* Some character classes are equivalent to other nodes. Such nodes take
18264 * up less room and generally fewer operations to execute than ANYOF nodes.
18265 * Above, we checked for and optimized into some such equivalents for
18266 * certain common classes that are easy to test. Getting to this point in
18267 * the code means that the class didn't get optimized there. Since this
18268 * code is only executed in Pass 2, it is too late to save space--it has
18269 * been allocated in Pass 1, and currently isn't given back. XXX Why not?
18270 * But turning things into an EXACTish node can allow the optimizer to join
18271 * it to any adjacent such nodes. And if the class is equivalent to things
18272 * like /./, expensive run-time swashes can be avoided. Now that we have
18273 * more complete information, we can find things necessarily missed by the
18276 if (optimizable && cp_list && ! invert) {
18278 U8 op = END; /* The optimzation node-type */
18279 int posix_class = -1; /* Illegal value */
18280 const char * cur_parse= RExC_parse;
18281 U8 ANYOFM_mask = 0xFF;
18284 invlist_iterinit(cp_list);
18285 if (! invlist_iternext(cp_list, &start, &end)) {
18287 /* Here, the list is empty. This happens, for example, when a
18288 * Unicode property that doesn't match anything is the only element
18289 * in the character class (perluniprops.pod notes such properties).
18292 *flagp |= HASWIDTH|SIMPLE;
18294 else if (start == end) { /* The range is a single code point */
18295 if (! invlist_iternext(cp_list, &start, &end)
18297 /* Don't do this optimization if it would require changing
18298 * the pattern to UTF-8 */
18299 && (start < 256 || UTF))
18301 /* Here, the list contains a single code point. Can optimize
18302 * into an EXACTish node */
18313 /* A locale node under folding with one code point can be
18314 * an EXACTFL, as its fold won't be calculated until
18320 /* Here, we are generally folding, but there is only one
18321 * code point to match. If we have to, we use an EXACT
18322 * node, but it would be better for joining with adjacent
18323 * nodes in the optimization pass if we used the same
18324 * EXACTFish node that any such are likely to be. We can
18325 * do this iff the code point doesn't participate in any
18326 * folds. For example, an EXACTF of a colon is the same as
18327 * an EXACT one, since nothing folds to or from a colon. */
18329 if (IS_IN_SOME_FOLD_L1(value)) {
18334 if (_invlist_contains_cp(PL_utf8_foldable, value)) {
18339 /* If we haven't found the node type, above, it means we
18340 * can use the prevailing one */
18342 op = compute_EXACTish(pRExC_state);
18346 } /* End of first range contains just a single code point */
18347 else if (start == 0) {
18348 if (end == UV_MAX) {
18350 *flagp |= HASWIDTH|SIMPLE;
18353 else if (end == '\n' - 1
18354 && invlist_iternext(cp_list, &start, &end)
18355 && start == '\n' + 1 && end == UV_MAX)
18358 *flagp |= HASWIDTH|SIMPLE;
18362 invlist_iterfinish(cp_list);
18366 /* Here, didn't find an optimization. See if this matches any of
18367 * the POSIX classes. First try ASCII */
18369 if (_invlistEQ(cp_list, PL_XPosix_ptrs[_CC_ASCII], 0)) {
18371 *flagp |= HASWIDTH|SIMPLE;
18373 else if (_invlistEQ(cp_list, PL_XPosix_ptrs[_CC_ASCII], 1)) {
18375 *flagp |= HASWIDTH|SIMPLE;
18377 else if (invlist_highest(cp_list) >= 0x2029) {
18379 /* Then try the other POSIX classes. The POSIXA ones are about
18380 * the same speed as ANYOF ops, but the ones that have
18381 * above-Latin1 code point matches are somewhat faster than
18382 * ANYOF. So optimize those, but don't bother with the POSIXA
18383 * ones nor [:cntrl:] which has no above-Latin1 matches. If
18384 * this ANYOF node has a lower highest possible matching code
18385 * point than any of the XPosix ones, we know that it can't
18386 * possibly be the same as any of them, so we can avoid
18387 * executing this code. The 0x2029 above for the lowest max
18388 * was determined by manual inspection of the classes, and
18389 * comes from \v. Suppose Unicode in a later version adds a
18390 * higher code point to \v. All that means is that this code
18391 * can be executed unnecessarily. It will still give the
18392 * correct answer. */
18394 for (posix_class = 0;
18395 posix_class <= _HIGHEST_REGCOMP_DOT_H_SYNC;
18400 if (posix_class == _CC_CNTRL) {
18404 for (try_inverted = 0; try_inverted < 2; try_inverted++) {
18406 /* Check if matches normal or inverted */
18407 if (_invlistEQ(cp_list,
18408 PL_XPosix_ptrs[posix_class],
18411 op = (try_inverted)
18414 *flagp |= HASWIDTH|SIMPLE;
18422 /* If it didn't match a POSIX class, it might be able to be turned
18423 * into an ANYOFM node. Compare two different bytes, bit-by-bit.
18424 * In some positions, the bits in each will be 1; and in other
18425 * positions both will be 0; and in some positions the bit will be
18426 * 1 in one byte, and 0 in the other. Let 'n' be the number of
18427 * positions where the bits differ. We create a mask which has
18428 * exactly 'n' 0 bits, each in a position where the two bytes
18429 * differ. Now take the set of all bytes that when ANDed with the
18430 * mask yield the same result. That set has 2**n elements, and is
18431 * representable by just two 8 bit numbers: the result and the
18432 * mask. Importantly, matching the set can be vectorized by
18433 * creating a word full of the result bytes, and a word full of the
18434 * mask bytes, yielding a significant speed up. Here, see if this
18435 * node matches such a set. As a concrete example consider [01],
18436 * and the byte representing '0' which is 0x30 on ASCII machines.
18437 * It has the bits 0011 0000. Take the mask 1111 1110. If we AND
18438 * 0x31 and 0x30 with that mask we get 0x30. Any other bytes ANDed
18439 * yield something else. So [01], which is a common usage, is
18440 * optimizable into ANYOFM, and can benefit from the speed up. We
18441 * can only do this on UTF-8 invariant bytes, because the variance
18442 * would throw this off. */
18444 && invlist_highest(cp_list) <=
18451 Size_t cp_count = 0;
18452 bool first_time = TRUE;
18453 unsigned int lowest_cp = 0xFF;
18454 U8 bits_differing = 0;
18456 /* Only needed on EBCDIC, as there, variants and non- are mixed
18457 * together. Could #ifdef it out on ASCII, but probably the
18458 * compiler will optimize it out */
18459 bool has_variant = FALSE;
18461 /* Go through the bytes and find the bit positions that differ */
18462 invlist_iterinit(cp_list);
18463 while (invlist_iternext(cp_list, &start, &end)) {
18464 unsigned int i = start;
18466 cp_count += end - start + 1;
18469 if (! UVCHR_IS_INVARIANT(i)) {
18470 has_variant = TRUE;
18474 first_time = FALSE;
18480 /* Find the bit positions that differ from the lowest code
18481 * point in the node. Keep track of all such positions by
18483 for (; i <= end; i++) {
18484 if (! UVCHR_IS_INVARIANT(i)) {
18485 has_variant = TRUE;
18489 bits_differing |= i ^ lowest_cp;
18492 invlist_iterfinish(cp_list);
18494 /* At the end of the loop, we count how many bits differ from
18495 * the bits in lowest code point, call the count 'd'. If the
18496 * set we found contains 2**d elements, it is the closure of
18497 * all code points that differ only in those bit positions. To
18498 * convince yourself of that, first note that the number in the
18499 * closure must be a power of 2, which we test for. The only
18500 * way we could have that count and it be some differing set,
18501 * is if we got some code points that don't differ from the
18502 * lowest code point in any position, but do differ from each
18503 * other in some other position. That means one code point has
18504 * a 1 in that position, and another has a 0. But that would
18505 * mean that one of them differs from the lowest code point in
18506 * that position, which possibility we've already excluded. */
18508 && cp_count == 1U << PL_bitcount[bits_differing])
18510 assert(cp_count > 1);
18513 /* We need to make the bits that differ be 0's */
18514 ANYOFM_mask = ~ bits_differing; /* This goes into FLAGS */
18516 /* The argument is the lowest code point */
18517 anode_arg = lowest_cp;
18518 *flagp |= HASWIDTH|SIMPLE;
18524 RExC_parse = (char *)orig_parse;
18525 RExC_emit = (regnode *)orig_emit;
18527 if (regarglen[op]) {
18528 ret = reganode(pRExC_state, op, anode_arg);
18530 ret = reg_node(pRExC_state, op);
18533 RExC_parse = (char *)cur_parse;
18535 if (PL_regkind[op] == EXACT) {
18536 alloc_maybe_populate_EXACT(pRExC_state, ret, flagp, 0, value,
18537 TRUE /* downgradable to EXACT */
18540 else if (PL_regkind[op] == POSIXD || PL_regkind[op] == NPOSIXD) {
18541 FLAGS(ret) = posix_class;
18543 else if (PL_regkind[op] == ANYOFM) {
18544 FLAGS(ret) = ANYOFM_mask;
18547 SvREFCNT_dec_NN(cp_list);
18552 /* Here, <cp_list> contains all the code points we can determine at
18553 * compile time that match under all conditions. Go through it, and
18554 * for things that belong in the bitmap, put them there, and delete from
18555 * <cp_list>. While we are at it, see if everything above 255 is in the
18556 * list, and if so, set a flag to speed up execution */
18558 populate_ANYOF_from_invlist(ret, &cp_list);
18561 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
18564 /* Here, the bitmap has been populated with all the Latin1 code points that
18565 * always match. Can now add to the overall list those that match only
18566 * when the target string is UTF-8 (<has_upper_latin1_only_utf8_matches>).
18568 if (has_upper_latin1_only_utf8_matches) {
18570 _invlist_union(cp_list,
18571 has_upper_latin1_only_utf8_matches,
18573 SvREFCNT_dec_NN(has_upper_latin1_only_utf8_matches);
18576 cp_list = has_upper_latin1_only_utf8_matches;
18578 ANYOF_FLAGS(ret) |= ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP;
18581 /* If there is a swash and more than one element, we can't use the swash in
18582 * the optimization below. */
18583 if (swash && element_count > 1) {
18584 SvREFCNT_dec_NN(swash);
18588 /* Note that the optimization of using 'swash' if it is the only thing in
18589 * the class doesn't have us change swash at all, so it can include things
18590 * that are also in the bitmap; otherwise we have purposely deleted that
18591 * duplicate information */
18592 set_ANYOF_arg(pRExC_state, ret, cp_list,
18593 (HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION)
18595 only_utf8_locale_list,
18596 swash, has_user_defined_property);
18598 *flagp |= HASWIDTH|SIMPLE;
18600 if (ANYOF_FLAGS(ret) & ANYOF_LOCALE_FLAGS) {
18601 RExC_contains_locale = 1;
18607 #undef HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION
18610 S_set_ANYOF_arg(pTHX_ RExC_state_t* const pRExC_state,
18611 regnode* const node,
18613 SV* const runtime_defns,
18614 SV* const only_utf8_locale_list,
18616 const bool has_user_defined_property)
18618 /* Sets the arg field of an ANYOF-type node 'node', using information about
18619 * the node passed-in. If there is nothing outside the node's bitmap, the
18620 * arg is set to ANYOF_ONLY_HAS_BITMAP. Otherwise, it sets the argument to
18621 * the count returned by add_data(), having allocated and stored an array,
18622 * av, that that count references, as follows:
18623 * av[0] stores the character class description in its textual form.
18624 * This is used later (regexec.c:Perl_regclass_swash()) to
18625 * initialize the appropriate swash, and is also useful for dumping
18626 * the regnode. This is set to &PL_sv_undef if the textual
18627 * description is not needed at run-time (as happens if the other
18628 * elements completely define the class)
18629 * av[1] if &PL_sv_undef, is a placeholder to later contain the swash
18630 * computed from av[0]. But if no further computation need be done,
18631 * the swash is stored here now (and av[0] is &PL_sv_undef).
18632 * av[2] stores the inversion list of code points that match only if the
18633 * current locale is UTF-8
18634 * av[3] stores the cp_list inversion list for use in addition or instead
18635 * of av[0]; used only if cp_list exists and av[1] is &PL_sv_undef.
18636 * (Otherwise everything needed is already in av[0] and av[1])
18637 * av[4] is set if any component of the class is from a user-defined
18638 * property; used only if av[3] exists */
18642 PERL_ARGS_ASSERT_SET_ANYOF_ARG;
18644 if (! cp_list && ! runtime_defns && ! only_utf8_locale_list) {
18645 assert(! (ANYOF_FLAGS(node)
18646 & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP));
18647 ARG_SET(node, ANYOF_ONLY_HAS_BITMAP);
18650 AV * const av = newAV();
18653 av_store(av, 0, (runtime_defns)
18654 ? SvREFCNT_inc(runtime_defns) : &PL_sv_undef);
18657 av_store(av, 1, swash);
18658 SvREFCNT_dec_NN(cp_list);
18661 av_store(av, 1, &PL_sv_undef);
18663 av_store(av, 3, cp_list);
18664 av_store(av, 4, newSVuv(has_user_defined_property));
18668 if (only_utf8_locale_list) {
18669 av_store(av, 2, only_utf8_locale_list);
18672 av_store(av, 2, &PL_sv_undef);
18675 rv = newRV_noinc(MUTABLE_SV(av));
18676 n = add_data(pRExC_state, STR_WITH_LEN("s"));
18677 RExC_rxi->data->data[n] = (void*)rv;
18682 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
18684 Perl__get_regclass_nonbitmap_data(pTHX_ const regexp *prog,
18685 const regnode* node,
18688 SV** only_utf8_locale_ptr,
18689 SV** output_invlist)
18692 /* For internal core use only.
18693 * Returns the swash for the input 'node' in the regex 'prog'.
18694 * If <doinit> is 'true', will attempt to create the swash if not already
18696 * If <listsvp> is non-null, will return the printable contents of the
18697 * swash. This can be used to get debugging information even before the
18698 * swash exists, by calling this function with 'doinit' set to false, in
18699 * which case the components that will be used to eventually create the
18700 * swash are returned (in a printable form).
18701 * If <only_utf8_locale_ptr> is not NULL, it is where this routine is to
18702 * store an inversion list of code points that should match only if the
18703 * execution-time locale is a UTF-8 one.
18704 * If <output_invlist> is not NULL, it is where this routine is to store an
18705 * inversion list of the code points that would be instead returned in
18706 * <listsvp> if this were NULL. Thus, what gets output in <listsvp>
18707 * when this parameter is used, is just the non-code point data that
18708 * will go into creating the swash. This currently should be just
18709 * user-defined properties whose definitions were not known at compile
18710 * time. Using this parameter allows for easier manipulation of the
18711 * swash's data by the caller. It is illegal to call this function with
18712 * this parameter set, but not <listsvp>
18714 * Tied intimately to how S_set_ANYOF_arg sets up the data structure. Note
18715 * that, in spite of this function's name, the swash it returns may include
18716 * the bitmap data as well */
18719 SV *si = NULL; /* Input swash initialization string */
18720 SV* invlist = NULL;
18722 RXi_GET_DECL(prog,progi);
18723 const struct reg_data * const data = prog ? progi->data : NULL;
18725 PERL_ARGS_ASSERT__GET_REGCLASS_NONBITMAP_DATA;
18726 assert(! output_invlist || listsvp);
18728 if (data && data->count) {
18729 const U32 n = ARG(node);
18731 if (data->what[n] == 's') {
18732 SV * const rv = MUTABLE_SV(data->data[n]);
18733 AV * const av = MUTABLE_AV(SvRV(rv));
18734 SV **const ary = AvARRAY(av);
18735 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
18737 si = *ary; /* ary[0] = the string to initialize the swash with */
18739 if (av_tindex_skip_len_mg(av) >= 2) {
18740 if (only_utf8_locale_ptr
18742 && ary[2] != &PL_sv_undef)
18744 *only_utf8_locale_ptr = ary[2];
18747 assert(only_utf8_locale_ptr);
18748 *only_utf8_locale_ptr = NULL;
18751 /* Elements 3 and 4 are either both present or both absent. [3]
18752 * is any inversion list generated at compile time; [4]
18753 * indicates if that inversion list has any user-defined
18754 * properties in it. */
18755 if (av_tindex_skip_len_mg(av) >= 3) {
18757 if (SvUV(ary[4])) {
18758 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
18766 /* Element [1] is reserved for the set-up swash. If already there,
18767 * return it; if not, create it and store it there */
18768 if (ary[1] && SvROK(ary[1])) {
18771 else if (doinit && ((si && si != &PL_sv_undef)
18772 || (invlist && invlist != &PL_sv_undef))) {
18774 sw = _core_swash_init("utf8", /* the utf8 package */
18778 0, /* not from tr/// */
18780 &swash_init_flags);
18781 (void)av_store(av, 1, sw);
18786 /* If requested, return a printable version of what this swash matches */
18788 SV* matches_string = NULL;
18790 /* The swash should be used, if possible, to get the data, as it
18791 * contains the resolved data. But this function can be called at
18792 * compile-time, before everything gets resolved, in which case we
18793 * return the currently best available information, which is the string
18794 * that will eventually be used to do that resolving, 'si' */
18795 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
18796 && (si && si != &PL_sv_undef))
18798 /* Here, we only have 'si' (and possibly some passed-in data in
18799 * 'invlist', which is handled below) If the caller only wants
18800 * 'si', use that. */
18801 if (! output_invlist) {
18802 matches_string = newSVsv(si);
18805 /* But if the caller wants an inversion list of the node, we
18806 * need to parse 'si' and place as much as possible in the
18807 * desired output inversion list, making 'matches_string' only
18808 * contain the currently unresolvable things */
18809 const char *si_string = SvPVX(si);
18810 STRLEN remaining = SvCUR(si);
18814 /* Ignore everything before the first new-line */
18815 while (*si_string != '\n' && remaining > 0) {
18819 assert(remaining > 0);
18824 while (remaining > 0) {
18826 /* The data consists of just strings defining user-defined
18827 * property names, but in prior incarnations, and perhaps
18828 * somehow from pluggable regex engines, it could still
18829 * hold hex code point definitions. Each component of a
18830 * range would be separated by a tab, and each range by a
18831 * new-line. If these are found, instead add them to the
18832 * inversion list */
18833 I32 grok_flags = PERL_SCAN_SILENT_ILLDIGIT
18834 |PERL_SCAN_SILENT_NON_PORTABLE;
18835 STRLEN len = remaining;
18836 UV cp = grok_hex(si_string, &len, &grok_flags, NULL);
18838 /* If the hex decode routine found something, it should go
18839 * up to the next \n */
18840 if ( *(si_string + len) == '\n') {
18841 if (count) { /* 2nd code point on line */
18842 *output_invlist = _add_range_to_invlist(*output_invlist, prev_cp, cp);
18845 *output_invlist = add_cp_to_invlist(*output_invlist, cp);
18848 goto prepare_for_next_iteration;
18851 /* If the hex decode was instead for the lower range limit,
18852 * save it, and go parse the upper range limit */
18853 if (*(si_string + len) == '\t') {
18854 assert(count == 0);
18858 prepare_for_next_iteration:
18859 si_string += len + 1;
18860 remaining -= len + 1;
18864 /* Here, didn't find a legal hex number. Just add it from
18865 * here to the next \n */
18868 while (*(si_string + len) != '\n' && remaining > 0) {
18872 if (*(si_string + len) == '\n') {
18876 if (matches_string) {
18877 sv_catpvn(matches_string, si_string, len - 1);
18880 matches_string = newSVpvn(si_string, len - 1);
18883 sv_catpvs(matches_string, " ");
18884 } /* end of loop through the text */
18886 assert(matches_string);
18887 if (SvCUR(matches_string)) { /* Get rid of trailing blank */
18888 SvCUR_set(matches_string, SvCUR(matches_string) - 1);
18890 } /* end of has an 'si' but no swash */
18893 /* If we have a swash in place, its equivalent inversion list was above
18894 * placed into 'invlist'. If not, this variable may contain a stored
18895 * inversion list which is information beyond what is in 'si' */
18898 /* Again, if the caller doesn't want the output inversion list, put
18899 * everything in 'matches-string' */
18900 if (! output_invlist) {
18901 if ( ! matches_string) {
18902 matches_string = newSVpvs("\n");
18904 sv_catsv(matches_string, invlist_contents(invlist,
18905 TRUE /* traditional style */
18908 else if (! *output_invlist) {
18909 *output_invlist = invlist_clone(invlist);
18912 _invlist_union(*output_invlist, invlist, output_invlist);
18916 *listsvp = matches_string;
18921 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
18923 /* reg_skipcomment()
18925 Absorbs an /x style # comment from the input stream,
18926 returning a pointer to the first character beyond the comment, or if the
18927 comment terminates the pattern without anything following it, this returns
18928 one past the final character of the pattern (in other words, RExC_end) and
18929 sets the REG_RUN_ON_COMMENT_SEEN flag.
18931 Note it's the callers responsibility to ensure that we are
18932 actually in /x mode
18936 PERL_STATIC_INLINE char*
18937 S_reg_skipcomment(RExC_state_t *pRExC_state, char* p)
18939 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
18943 while (p < RExC_end) {
18944 if (*(++p) == '\n') {
18949 /* we ran off the end of the pattern without ending the comment, so we have
18950 * to add an \n when wrapping */
18951 RExC_seen |= REG_RUN_ON_COMMENT_SEEN;
18956 S_skip_to_be_ignored_text(pTHX_ RExC_state_t *pRExC_state,
18958 const bool force_to_xmod
18961 /* If the text at the current parse position '*p' is a '(?#...)' comment,
18962 * or if we are under /x or 'force_to_xmod' is TRUE, and the text at '*p'
18963 * is /x whitespace, advance '*p' so that on exit it points to the first
18964 * byte past all such white space and comments */
18966 const bool use_xmod = force_to_xmod || (RExC_flags & RXf_PMf_EXTENDED);
18968 PERL_ARGS_ASSERT_SKIP_TO_BE_IGNORED_TEXT;
18970 assert( ! UTF || UTF8_IS_INVARIANT(**p) || UTF8_IS_START(**p));
18973 if (RExC_end - (*p) >= 3
18975 && *(*p + 1) == '?'
18976 && *(*p + 2) == '#')
18978 while (*(*p) != ')') {
18979 if ((*p) == RExC_end)
18980 FAIL("Sequence (?#... not terminated");
18988 const char * save_p = *p;
18989 while ((*p) < RExC_end) {
18991 if ((len = is_PATWS_safe((*p), RExC_end, UTF))) {
18994 else if (*(*p) == '#') {
18995 (*p) = reg_skipcomment(pRExC_state, (*p));
19001 if (*p != save_p) {
19014 Advances the parse position by one byte, unless that byte is the beginning
19015 of a '(?#...)' style comment, or is /x whitespace and /x is in effect. In
19016 those two cases, the parse position is advanced beyond all such comments and
19019 This is the UTF, (?#...), and /x friendly way of saying RExC_parse++.
19023 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
19025 PERL_ARGS_ASSERT_NEXTCHAR;
19027 if (RExC_parse < RExC_end) {
19029 || UTF8_IS_INVARIANT(*RExC_parse)
19030 || UTF8_IS_START(*RExC_parse));
19032 RExC_parse += (UTF) ? UTF8SKIP(RExC_parse) : 1;
19034 skip_to_be_ignored_text(pRExC_state, &RExC_parse,
19035 FALSE /* Don't force /x */ );
19040 S_regnode_guts(pTHX_ RExC_state_t *pRExC_state, const U8 op, const STRLEN extra_size, const char* const name)
19042 /* Allocate a regnode for 'op' and returns it, with 'extra_size' extra
19043 * space. In pass1, it aligns and increments RExC_size; in pass2,
19046 regnode * const ret = RExC_emit;
19047 GET_RE_DEBUG_FLAGS_DECL;
19049 PERL_ARGS_ASSERT_REGNODE_GUTS;
19051 assert(extra_size >= regarglen[op]);
19054 SIZE_ALIGN(RExC_size);
19055 RExC_size += 1 + extra_size;
19058 if (RExC_emit >= RExC_emit_bound)
19059 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d, %p>=%p",
19060 op, (void*)RExC_emit, (void*)RExC_emit_bound);
19062 NODE_ALIGN_FILL(ret);
19063 #ifndef RE_TRACK_PATTERN_OFFSETS
19064 PERL_UNUSED_ARG(name);
19066 if (RExC_offsets) { /* MJD */
19068 ("%s:%d: (op %s) %s %" UVuf " (len %" UVuf ") (max %" UVuf ").\n",
19071 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
19072 ? "Overwriting end of array!\n" : "OK",
19073 (UV)(RExC_emit - RExC_emit_start),
19074 (UV)(RExC_parse - RExC_start),
19075 (UV)RExC_offsets[0]));
19076 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
19083 - reg_node - emit a node
19085 STATIC regnode * /* Location. */
19086 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
19088 regnode * const ret = regnode_guts(pRExC_state, op, regarglen[op], "reg_node");
19090 PERL_ARGS_ASSERT_REG_NODE;
19092 assert(regarglen[op] == 0);
19095 regnode *ptr = ret;
19096 FILL_ADVANCE_NODE(ptr, op);
19103 - reganode - emit a node with an argument
19105 STATIC regnode * /* Location. */
19106 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
19108 regnode * const ret = regnode_guts(pRExC_state, op, regarglen[op], "reganode");
19110 PERL_ARGS_ASSERT_REGANODE;
19112 assert(regarglen[op] == 1);
19115 regnode *ptr = ret;
19116 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
19123 S_reg2Lanode(pTHX_ RExC_state_t *pRExC_state, const U8 op, const U32 arg1, const I32 arg2)
19125 /* emit a node with U32 and I32 arguments */
19127 regnode * const ret = regnode_guts(pRExC_state, op, regarglen[op], "reg2Lanode");
19129 PERL_ARGS_ASSERT_REG2LANODE;
19131 assert(regarglen[op] == 2);
19134 regnode *ptr = ret;
19135 FILL_ADVANCE_NODE_2L_ARG(ptr, op, arg1, arg2);
19142 - reginsert - insert an operator in front of already-emitted operand
19144 * Means relocating the operand.
19146 * IMPORTANT NOTE - it is the *callers* responsibility to correctly
19147 * set up NEXT_OFF() of the inserted node if needed. Something like this:
19149 * reginsert(pRExC, OPFAIL, orig_emit, depth+1);
19151 * NEXT_OFF(orig_emit) = regarglen[OPFAIL] + NODE_STEP_REGNODE;
19153 * ALSO NOTE - operand->flags will be set to 0 as well.
19156 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *operand, U32 depth)
19161 const int offset = regarglen[(U8)op];
19162 const int size = NODE_STEP_REGNODE + offset;
19163 GET_RE_DEBUG_FLAGS_DECL;
19165 PERL_ARGS_ASSERT_REGINSERT;
19166 PERL_UNUSED_CONTEXT;
19167 PERL_UNUSED_ARG(depth);
19168 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
19169 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
19174 assert(!RExC_study_started); /* I believe we should never use reginsert once we have started
19175 studying. If this is wrong then we need to adjust RExC_recurse
19176 below like we do with RExC_open_parens/RExC_close_parens. */
19180 if (RExC_open_parens) {
19182 /*DEBUG_PARSE_FMT("inst"," - %" IVdf, (IV)RExC_npar);*/
19183 /* remember that RExC_npar is rex->nparens + 1,
19184 * iow it is 1 more than the number of parens seen in
19185 * the pattern so far. */
19186 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
19187 /* note, RExC_open_parens[0] is the start of the
19188 * regex, it can't move. RExC_close_parens[0] is the end
19189 * of the regex, it *can* move. */
19190 if ( paren && RExC_open_parens[paren] >= operand ) {
19191 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
19192 RExC_open_parens[paren] += size;
19194 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
19196 if ( RExC_close_parens[paren] >= operand ) {
19197 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
19198 RExC_close_parens[paren] += size;
19200 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
19205 RExC_end_op += size;
19207 while (src > operand) {
19208 StructCopy(--src, --dst, regnode);
19209 #ifdef RE_TRACK_PATTERN_OFFSETS
19210 if (RExC_offsets) { /* MJD 20010112 */
19212 ("%s(%d): (op %s) %s copy %" UVuf " -> %" UVuf " (max %" UVuf ").\n",
19216 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
19217 ? "Overwriting end of array!\n" : "OK",
19218 (UV)(src - RExC_emit_start),
19219 (UV)(dst - RExC_emit_start),
19220 (UV)RExC_offsets[0]));
19221 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
19222 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
19227 place = operand; /* Op node, where operand used to be. */
19228 #ifdef RE_TRACK_PATTERN_OFFSETS
19229 if (RExC_offsets) { /* MJD */
19231 ("%s(%d): (op %s) %s %" UVuf " <- %" UVuf " (max %" UVuf ").\n",
19235 (UV)(place - RExC_emit_start) > RExC_offsets[0]
19236 ? "Overwriting end of array!\n" : "OK",
19237 (UV)(place - RExC_emit_start),
19238 (UV)(RExC_parse - RExC_start),
19239 (UV)RExC_offsets[0]));
19240 Set_Node_Offset(place, RExC_parse);
19241 Set_Node_Length(place, 1);
19244 src = NEXTOPER(place);
19246 FILL_ADVANCE_NODE(place, op);
19247 Zero(src, offset, regnode);
19251 - regtail - set the next-pointer at the end of a node chain of p to val.
19252 - SEE ALSO: regtail_study
19255 S_regtail(pTHX_ RExC_state_t * pRExC_state,
19256 const regnode * const p,
19257 const regnode * const val,
19261 GET_RE_DEBUG_FLAGS_DECL;
19263 PERL_ARGS_ASSERT_REGTAIL;
19265 PERL_UNUSED_ARG(depth);
19271 /* Find last node. */
19272 scan = (regnode *) p;
19274 regnode * const temp = regnext(scan);
19276 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
19277 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
19278 Perl_re_printf( aTHX_ "~ %s (%d) %s %s\n",
19279 SvPV_nolen_const(RExC_mysv), REG_NODE_NUM(scan),
19280 (temp == NULL ? "->" : ""),
19281 (temp == NULL ? PL_reg_name[OP(val)] : "")
19289 if (reg_off_by_arg[OP(scan)]) {
19290 ARG_SET(scan, val - scan);
19293 NEXT_OFF(scan) = val - scan;
19299 - regtail_study - set the next-pointer at the end of a node chain of p to val.
19300 - Look for optimizable sequences at the same time.
19301 - currently only looks for EXACT chains.
19303 This is experimental code. The idea is to use this routine to perform
19304 in place optimizations on branches and groups as they are constructed,
19305 with the long term intention of removing optimization from study_chunk so
19306 that it is purely analytical.
19308 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
19309 to control which is which.
19312 /* TODO: All four parms should be const */
19315 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p,
19316 const regnode *val,U32 depth)
19320 #ifdef EXPERIMENTAL_INPLACESCAN
19323 GET_RE_DEBUG_FLAGS_DECL;
19325 PERL_ARGS_ASSERT_REGTAIL_STUDY;
19331 /* Find last node. */
19335 regnode * const temp = regnext(scan);
19336 #ifdef EXPERIMENTAL_INPLACESCAN
19337 if (PL_regkind[OP(scan)] == EXACT) {
19338 bool unfolded_multi_char; /* Unexamined in this routine */
19339 if (join_exact(pRExC_state, scan, &min,
19340 &unfolded_multi_char, 1, val, depth+1))
19345 switch (OP(scan)) {
19349 case EXACTFAA_NO_TRIE:
19355 if( exact == PSEUDO )
19357 else if ( exact != OP(scan) )
19366 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
19367 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
19368 Perl_re_printf( aTHX_ "~ %s (%d) -> %s\n",
19369 SvPV_nolen_const(RExC_mysv),
19370 REG_NODE_NUM(scan),
19371 PL_reg_name[exact]);
19378 DEBUG_PARSE_MSG("");
19379 regprop(RExC_rx, RExC_mysv, val, NULL, pRExC_state);
19380 Perl_re_printf( aTHX_
19381 "~ attach to %s (%" IVdf ") offset to %" IVdf "\n",
19382 SvPV_nolen_const(RExC_mysv),
19383 (IV)REG_NODE_NUM(val),
19387 if (reg_off_by_arg[OP(scan)]) {
19388 ARG_SET(scan, val - scan);
19391 NEXT_OFF(scan) = val - scan;
19399 S_get_ANYOFM_contents(pTHX_ const regnode * n) {
19401 /* Returns an inversion list of all the code points matched by the ANYOFM
19404 SV * cp_list = _new_invlist(-1);
19405 const U8 lowest = (U8) ARG(n);
19408 U8 needed = 1U << PL_bitcount[ (U8) ~ FLAGS(n)];
19410 PERL_ARGS_ASSERT_GET_ANYOFM_CONTENTS;
19412 /* Starting with the lowest code point, any code point that ANDed with the
19413 * mask yields the lowest code point is in the set */
19414 for (i = lowest; i <= 0xFF; i++) {
19415 if ((i & FLAGS(n)) == ARG(n)) {
19416 cp_list = add_cp_to_invlist(cp_list, i);
19419 /* We know how many code points (a power of two) that are in the
19420 * set. No use looking once we've got that number */
19421 if (count >= needed) break;
19429 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
19434 S_regdump_intflags(pTHX_ const char *lead, const U32 flags)
19439 ASSUME(REG_INTFLAGS_NAME_SIZE <= sizeof(flags)*8);
19441 for (bit=0; bit<REG_INTFLAGS_NAME_SIZE; bit++) {
19442 if (flags & (1<<bit)) {
19443 if (!set++ && lead)
19444 Perl_re_printf( aTHX_ "%s",lead);
19445 Perl_re_printf( aTHX_ "%s ",PL_reg_intflags_name[bit]);
19450 Perl_re_printf( aTHX_ "\n");
19452 Perl_re_printf( aTHX_ "%s[none-set]\n",lead);
19457 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
19463 ASSUME(REG_EXTFLAGS_NAME_SIZE <= sizeof(flags)*8);
19465 for (bit=0; bit<REG_EXTFLAGS_NAME_SIZE; bit++) {
19466 if (flags & (1<<bit)) {
19467 if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */
19470 if (!set++ && lead)
19471 Perl_re_printf( aTHX_ "%s",lead);
19472 Perl_re_printf( aTHX_ "%s ",PL_reg_extflags_name[bit]);
19475 if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) {
19476 if (!set++ && lead) {
19477 Perl_re_printf( aTHX_ "%s",lead);
19480 case REGEX_UNICODE_CHARSET:
19481 Perl_re_printf( aTHX_ "UNICODE");
19483 case REGEX_LOCALE_CHARSET:
19484 Perl_re_printf( aTHX_ "LOCALE");
19486 case REGEX_ASCII_RESTRICTED_CHARSET:
19487 Perl_re_printf( aTHX_ "ASCII-RESTRICTED");
19489 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
19490 Perl_re_printf( aTHX_ "ASCII-MORE_RESTRICTED");
19493 Perl_re_printf( aTHX_ "UNKNOWN CHARACTER SET");
19499 Perl_re_printf( aTHX_ "\n");
19501 Perl_re_printf( aTHX_ "%s[none-set]\n",lead);
19507 Perl_regdump(pTHX_ const regexp *r)
19511 SV * const sv = sv_newmortal();
19512 SV *dsv= sv_newmortal();
19513 RXi_GET_DECL(r,ri);
19514 GET_RE_DEBUG_FLAGS_DECL;
19516 PERL_ARGS_ASSERT_REGDUMP;
19518 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
19520 /* Header fields of interest. */
19521 for (i = 0; i < 2; i++) {
19522 if (r->substrs->data[i].substr) {
19523 RE_PV_QUOTED_DECL(s, 0, dsv,
19524 SvPVX_const(r->substrs->data[i].substr),
19525 RE_SV_DUMPLEN(r->substrs->data[i].substr),
19526 PL_dump_re_max_len);
19527 Perl_re_printf( aTHX_
19528 "%s %s%s at %" IVdf "..%" UVuf " ",
19529 i ? "floating" : "anchored",
19531 RE_SV_TAIL(r->substrs->data[i].substr),
19532 (IV)r->substrs->data[i].min_offset,
19533 (UV)r->substrs->data[i].max_offset);
19535 else if (r->substrs->data[i].utf8_substr) {
19536 RE_PV_QUOTED_DECL(s, 1, dsv,
19537 SvPVX_const(r->substrs->data[i].utf8_substr),
19538 RE_SV_DUMPLEN(r->substrs->data[i].utf8_substr),
19540 Perl_re_printf( aTHX_
19541 "%s utf8 %s%s at %" IVdf "..%" UVuf " ",
19542 i ? "floating" : "anchored",
19544 RE_SV_TAIL(r->substrs->data[i].utf8_substr),
19545 (IV)r->substrs->data[i].min_offset,
19546 (UV)r->substrs->data[i].max_offset);
19550 if (r->check_substr || r->check_utf8)
19551 Perl_re_printf( aTHX_
19553 ( r->check_substr == r->substrs->data[1].substr
19554 && r->check_utf8 == r->substrs->data[1].utf8_substr
19555 ? "(checking floating" : "(checking anchored"));
19556 if (r->intflags & PREGf_NOSCAN)
19557 Perl_re_printf( aTHX_ " noscan");
19558 if (r->extflags & RXf_CHECK_ALL)
19559 Perl_re_printf( aTHX_ " isall");
19560 if (r->check_substr || r->check_utf8)
19561 Perl_re_printf( aTHX_ ") ");
19563 if (ri->regstclass) {
19564 regprop(r, sv, ri->regstclass, NULL, NULL);
19565 Perl_re_printf( aTHX_ "stclass %s ", SvPVX_const(sv));
19567 if (r->intflags & PREGf_ANCH) {
19568 Perl_re_printf( aTHX_ "anchored");
19569 if (r->intflags & PREGf_ANCH_MBOL)
19570 Perl_re_printf( aTHX_ "(MBOL)");
19571 if (r->intflags & PREGf_ANCH_SBOL)
19572 Perl_re_printf( aTHX_ "(SBOL)");
19573 if (r->intflags & PREGf_ANCH_GPOS)
19574 Perl_re_printf( aTHX_ "(GPOS)");
19575 Perl_re_printf( aTHX_ " ");
19577 if (r->intflags & PREGf_GPOS_SEEN)
19578 Perl_re_printf( aTHX_ "GPOS:%" UVuf " ", (UV)r->gofs);
19579 if (r->intflags & PREGf_SKIP)
19580 Perl_re_printf( aTHX_ "plus ");
19581 if (r->intflags & PREGf_IMPLICIT)
19582 Perl_re_printf( aTHX_ "implicit ");
19583 Perl_re_printf( aTHX_ "minlen %" IVdf " ", (IV)r->minlen);
19584 if (r->extflags & RXf_EVAL_SEEN)
19585 Perl_re_printf( aTHX_ "with eval ");
19586 Perl_re_printf( aTHX_ "\n");
19588 regdump_extflags("r->extflags: ",r->extflags);
19589 regdump_intflags("r->intflags: ",r->intflags);
19592 PERL_ARGS_ASSERT_REGDUMP;
19593 PERL_UNUSED_CONTEXT;
19594 PERL_UNUSED_ARG(r);
19595 #endif /* DEBUGGING */
19598 /* Should be synchronized with ANYOF_ #defines in regcomp.h */
19601 # if _CC_WORDCHAR != 0 || _CC_DIGIT != 1 || _CC_ALPHA != 2 \
19602 || _CC_LOWER != 3 || _CC_UPPER != 4 || _CC_PUNCT != 5 \
19603 || _CC_PRINT != 6 || _CC_ALPHANUMERIC != 7 || _CC_GRAPH != 8 \
19604 || _CC_CASED != 9 || _CC_SPACE != 10 || _CC_BLANK != 11 \
19605 || _CC_XDIGIT != 12 || _CC_CNTRL != 13 || _CC_ASCII != 14 \
19606 || _CC_VERTSPACE != 15
19607 # error Need to adjust order of anyofs[]
19609 static const char * const anyofs[] = {
19646 - regprop - printable representation of opcode, with run time support
19650 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o, const regmatch_info *reginfo, const RExC_state_t *pRExC_state)
19654 RXi_GET_DECL(prog,progi);
19655 GET_RE_DEBUG_FLAGS_DECL;
19657 PERL_ARGS_ASSERT_REGPROP;
19661 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
19662 /* It would be nice to FAIL() here, but this may be called from
19663 regexec.c, and it would be hard to supply pRExC_state. */
19664 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d",
19665 (int)OP(o), (int)REGNODE_MAX);
19666 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
19668 k = PL_regkind[OP(o)];
19671 sv_catpvs(sv, " ");
19672 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
19673 * is a crude hack but it may be the best for now since
19674 * we have no flag "this EXACTish node was UTF-8"
19676 pv_pretty(sv, STRING(o), STR_LEN(o), PL_dump_re_max_len,
19677 PL_colors[0], PL_colors[1],
19678 PERL_PV_ESCAPE_UNI_DETECT |
19679 PERL_PV_ESCAPE_NONASCII |
19680 PERL_PV_PRETTY_ELLIPSES |
19681 PERL_PV_PRETTY_LTGT |
19682 PERL_PV_PRETTY_NOCLEAR
19684 } else if (k == TRIE) {
19685 /* print the details of the trie in dumpuntil instead, as
19686 * progi->data isn't available here */
19687 const char op = OP(o);
19688 const U32 n = ARG(o);
19689 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
19690 (reg_ac_data *)progi->data->data[n] :
19692 const reg_trie_data * const trie
19693 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
19695 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
19696 DEBUG_TRIE_COMPILE_r({
19698 sv_catpvs(sv, "(JUMP)");
19699 Perl_sv_catpvf(aTHX_ sv,
19700 "<S:%" UVuf "/%" IVdf " W:%" UVuf " L:%" UVuf "/%" UVuf " C:%" UVuf "/%" UVuf ">",
19701 (UV)trie->startstate,
19702 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
19703 (UV)trie->wordcount,
19706 (UV)TRIE_CHARCOUNT(trie),
19707 (UV)trie->uniquecharcount
19710 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
19711 sv_catpvs(sv, "[");
19712 (void) put_charclass_bitmap_innards(sv,
19713 ((IS_ANYOF_TRIE(op))
19715 : TRIE_BITMAP(trie)),
19721 sv_catpvs(sv, "]");
19723 } else if (k == CURLY) {
19724 U32 lo = ARG1(o), hi = ARG2(o);
19725 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
19726 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
19727 Perl_sv_catpvf(aTHX_ sv, "{%u,", (unsigned) lo);
19728 if (hi == REG_INFTY)
19729 sv_catpvs(sv, "INFTY");
19731 Perl_sv_catpvf(aTHX_ sv, "%u", (unsigned) hi);
19732 sv_catpvs(sv, "}");
19734 else if (k == WHILEM && o->flags) /* Ordinal/of */
19735 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
19736 else if (k == REF || k == OPEN || k == CLOSE
19737 || k == GROUPP || OP(o)==ACCEPT)
19739 AV *name_list= NULL;
19740 U32 parno= OP(o) == ACCEPT ? (U32)ARG2L(o) : ARG(o);
19741 Perl_sv_catpvf(aTHX_ sv, "%" UVuf, (UV)parno); /* Parenth number */
19742 if ( RXp_PAREN_NAMES(prog) ) {
19743 name_list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
19744 } else if ( pRExC_state ) {
19745 name_list= RExC_paren_name_list;
19748 if ( k != REF || (OP(o) < NREF)) {
19749 SV **name= av_fetch(name_list, parno, 0 );
19751 Perl_sv_catpvf(aTHX_ sv, " '%" SVf "'", SVfARG(*name));
19754 SV *sv_dat= MUTABLE_SV(progi->data->data[ parno ]);
19755 I32 *nums=(I32*)SvPVX(sv_dat);
19756 SV **name= av_fetch(name_list, nums[0], 0 );
19759 for ( n=0; n<SvIVX(sv_dat); n++ ) {
19760 Perl_sv_catpvf(aTHX_ sv, "%s%" IVdf,
19761 (n ? "," : ""), (IV)nums[n]);
19763 Perl_sv_catpvf(aTHX_ sv, " '%" SVf "'", SVfARG(*name));
19767 if ( k == REF && reginfo) {
19768 U32 n = ARG(o); /* which paren pair */
19769 I32 ln = prog->offs[n].start;
19770 if (prog->lastparen < n || ln == -1 || prog->offs[n].end == -1)
19771 Perl_sv_catpvf(aTHX_ sv, ": FAIL");
19772 else if (ln == prog->offs[n].end)
19773 Perl_sv_catpvf(aTHX_ sv, ": ACCEPT - EMPTY STRING");
19775 const char *s = reginfo->strbeg + ln;
19776 Perl_sv_catpvf(aTHX_ sv, ": ");
19777 Perl_pv_pretty( aTHX_ sv, s, prog->offs[n].end - prog->offs[n].start, 32, 0, 0,
19778 PERL_PV_ESCAPE_UNI_DETECT|PERL_PV_PRETTY_NOCLEAR|PERL_PV_PRETTY_ELLIPSES|PERL_PV_PRETTY_QUOTE );
19781 } else if (k == GOSUB) {
19782 AV *name_list= NULL;
19783 if ( RXp_PAREN_NAMES(prog) ) {
19784 name_list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
19785 } else if ( pRExC_state ) {
19786 name_list= RExC_paren_name_list;
19789 /* Paren and offset */
19790 Perl_sv_catpvf(aTHX_ sv, "%d[%+d:%d]", (int)ARG(o),(int)ARG2L(o),
19791 (int)((o + (int)ARG2L(o)) - progi->program) );
19793 SV **name= av_fetch(name_list, ARG(o), 0 );
19795 Perl_sv_catpvf(aTHX_ sv, " '%" SVf "'", SVfARG(*name));
19798 else if (k == LOGICAL)
19799 /* 2: embedded, otherwise 1 */
19800 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags);
19801 else if (k == ANYOF) {
19802 const U8 flags = ANYOF_FLAGS(o);
19803 bool do_sep = FALSE; /* Do we need to separate various components of
19805 /* Set if there is still an unresolved user-defined property */
19806 SV *unresolved = NULL;
19808 /* Things that are ignored except when the runtime locale is UTF-8 */
19809 SV *only_utf8_locale_invlist = NULL;
19811 /* Code points that don't fit in the bitmap */
19812 SV *nonbitmap_invlist = NULL;
19814 /* And things that aren't in the bitmap, but are small enough to be */
19815 SV* bitmap_range_not_in_bitmap = NULL;
19817 const bool inverted = flags & ANYOF_INVERT;
19819 if (OP(o) == ANYOFL) {
19820 if (ANYOFL_UTF8_LOCALE_REQD(flags)) {
19821 sv_catpvs(sv, "{utf8-locale-reqd}");
19823 if (flags & ANYOFL_FOLD) {
19824 sv_catpvs(sv, "{i}");
19828 /* If there is stuff outside the bitmap, get it */
19829 if (ARG(o) != ANYOF_ONLY_HAS_BITMAP) {
19830 (void) _get_regclass_nonbitmap_data(prog, o, FALSE,
19832 &only_utf8_locale_invlist,
19833 &nonbitmap_invlist);
19834 /* The non-bitmap data may contain stuff that could fit in the
19835 * bitmap. This could come from a user-defined property being
19836 * finally resolved when this call was done; or much more likely
19837 * because there are matches that require UTF-8 to be valid, and so
19838 * aren't in the bitmap. This is teased apart later */
19839 _invlist_intersection(nonbitmap_invlist,
19841 &bitmap_range_not_in_bitmap);
19842 /* Leave just the things that don't fit into the bitmap */
19843 _invlist_subtract(nonbitmap_invlist,
19845 &nonbitmap_invlist);
19848 /* Obey this flag to add all above-the-bitmap code points */
19849 if (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
19850 nonbitmap_invlist = _add_range_to_invlist(nonbitmap_invlist,
19851 NUM_ANYOF_CODE_POINTS,
19855 /* Ready to start outputting. First, the initial left bracket */
19856 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
19858 /* Then all the things that could fit in the bitmap */
19859 do_sep = put_charclass_bitmap_innards(sv,
19861 bitmap_range_not_in_bitmap,
19862 only_utf8_locale_invlist,
19865 /* Can't try inverting for a
19866 * better display if there are
19867 * things that haven't been
19869 unresolved != NULL);
19870 SvREFCNT_dec(bitmap_range_not_in_bitmap);
19872 /* If there are user-defined properties which haven't been defined yet,
19873 * output them. If the result is not to be inverted, it is clearest to
19874 * output them in a separate [] from the bitmap range stuff. If the
19875 * result is to be complemented, we have to show everything in one [],
19876 * as the inversion applies to the whole thing. Use {braces} to
19877 * separate them from anything in the bitmap and anything above the
19881 if (! do_sep) { /* If didn't output anything in the bitmap */
19882 sv_catpvs(sv, "^");
19884 sv_catpvs(sv, "{");
19887 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]);
19889 sv_catsv(sv, unresolved);
19891 sv_catpvs(sv, "}");
19893 do_sep = ! inverted;
19896 /* And, finally, add the above-the-bitmap stuff */
19897 if (nonbitmap_invlist && _invlist_len(nonbitmap_invlist)) {
19900 /* See if truncation size is overridden */
19901 const STRLEN dump_len = (PL_dump_re_max_len > 256)
19902 ? PL_dump_re_max_len
19905 /* This is output in a separate [] */
19907 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]);
19910 /* And, for easy of understanding, it is shown in the
19911 * uncomplemented form if possible. The one exception being if
19912 * there are unresolved items, where the inversion has to be
19913 * delayed until runtime */
19914 if (inverted && ! unresolved) {
19915 _invlist_invert(nonbitmap_invlist);
19916 _invlist_subtract(nonbitmap_invlist, PL_InBitmap, &nonbitmap_invlist);
19919 contents = invlist_contents(nonbitmap_invlist,
19920 FALSE /* output suitable for catsv */
19923 /* If the output is shorter than the permissible maximum, just do it. */
19924 if (SvCUR(contents) <= dump_len) {
19925 sv_catsv(sv, contents);
19928 const char * contents_string = SvPVX(contents);
19929 STRLEN i = dump_len;
19931 /* Otherwise, start at the permissible max and work back to the
19932 * first break possibility */
19933 while (i > 0 && contents_string[i] != ' ') {
19936 if (i == 0) { /* Fail-safe. Use the max if we couldn't
19937 find a legal break */
19941 sv_catpvn(sv, contents_string, i);
19942 sv_catpvs(sv, "...");
19945 SvREFCNT_dec_NN(contents);
19946 SvREFCNT_dec_NN(nonbitmap_invlist);
19949 /* And finally the matching, closing ']' */
19950 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
19952 SvREFCNT_dec(unresolved);
19954 else if (k == ANYOFM) {
19955 SV * cp_list = get_ANYOFM_contents(o);
19957 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
19958 put_charclass_bitmap_innards(sv, NULL, cp_list, NULL, NULL, TRUE);
19959 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
19961 SvREFCNT_dec(cp_list);
19963 else if (k == POSIXD || k == NPOSIXD) {
19964 U8 index = FLAGS(o) * 2;
19965 if (index < C_ARRAY_LENGTH(anyofs)) {
19966 if (*anyofs[index] != '[') {
19969 sv_catpv(sv, anyofs[index]);
19970 if (*anyofs[index] != '[') {
19975 Perl_sv_catpvf(aTHX_ sv, "[illegal type=%d])", index);
19978 else if (k == BOUND || k == NBOUND) {
19979 /* Must be synced with order of 'bound_type' in regcomp.h */
19980 const char * const bounds[] = {
19981 "", /* Traditional */
19987 assert(FLAGS(o) < C_ARRAY_LENGTH(bounds));
19988 sv_catpv(sv, bounds[FLAGS(o)]);
19990 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
19991 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
19992 else if (OP(o) == SBOL)
19993 Perl_sv_catpvf(aTHX_ sv, " /%s/", o->flags ? "\\A" : "^");
19995 /* add on the verb argument if there is one */
19996 if ( ( k == VERB || OP(o) == ACCEPT || OP(o) == OPFAIL ) && o->flags) {
19998 Perl_sv_catpvf(aTHX_ sv, ":%" SVf,
19999 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
20001 sv_catpvs(sv, ":NULL");
20004 PERL_UNUSED_CONTEXT;
20005 PERL_UNUSED_ARG(sv);
20006 PERL_UNUSED_ARG(o);
20007 PERL_UNUSED_ARG(prog);
20008 PERL_UNUSED_ARG(reginfo);
20009 PERL_UNUSED_ARG(pRExC_state);
20010 #endif /* DEBUGGING */
20016 Perl_re_intuit_string(pTHX_ REGEXP * const r)
20017 { /* Assume that RE_INTUIT is set */
20018 struct regexp *const prog = ReANY(r);
20019 GET_RE_DEBUG_FLAGS_DECL;
20021 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
20022 PERL_UNUSED_CONTEXT;
20026 const char * const s = SvPV_nolen_const(RX_UTF8(r)
20027 ? prog->check_utf8 : prog->check_substr);
20029 if (!PL_colorset) reginitcolors();
20030 Perl_re_printf( aTHX_
20031 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
20033 RX_UTF8(r) ? "utf8 " : "",
20034 PL_colors[5],PL_colors[0],
20037 (strlen(s) > PL_dump_re_max_len ? "..." : ""));
20040 /* use UTF8 check substring if regexp pattern itself is in UTF8 */
20041 return RX_UTF8(r) ? prog->check_utf8 : prog->check_substr;
20047 handles refcounting and freeing the perl core regexp structure. When
20048 it is necessary to actually free the structure the first thing it
20049 does is call the 'free' method of the regexp_engine associated to
20050 the regexp, allowing the handling of the void *pprivate; member
20051 first. (This routine is not overridable by extensions, which is why
20052 the extensions free is called first.)
20054 See regdupe and regdupe_internal if you change anything here.
20056 #ifndef PERL_IN_XSUB_RE
20058 Perl_pregfree(pTHX_ REGEXP *r)
20064 Perl_pregfree2(pTHX_ REGEXP *rx)
20066 struct regexp *const r = ReANY(rx);
20067 GET_RE_DEBUG_FLAGS_DECL;
20069 PERL_ARGS_ASSERT_PREGFREE2;
20071 if (r->mother_re) {
20072 ReREFCNT_dec(r->mother_re);
20074 CALLREGFREE_PVT(rx); /* free the private data */
20075 SvREFCNT_dec(RXp_PAREN_NAMES(r));
20079 for (i = 0; i < 2; i++) {
20080 SvREFCNT_dec(r->substrs->data[i].substr);
20081 SvREFCNT_dec(r->substrs->data[i].utf8_substr);
20083 Safefree(r->substrs);
20085 RX_MATCH_COPY_FREE(rx);
20086 #ifdef PERL_ANY_COW
20087 SvREFCNT_dec(r->saved_copy);
20090 SvREFCNT_dec(r->qr_anoncv);
20091 if (r->recurse_locinput)
20092 Safefree(r->recurse_locinput);
20098 Copy ssv to dsv, both of which should of type SVt_REGEXP or SVt_PVLV,
20099 except that dsv will be created if NULL.
20101 This function is used in two main ways. First to implement
20102 $r = qr/....; $s = $$r;
20104 Secondly, it is used as a hacky workaround to the structural issue of
20106 being stored in the regexp structure which is in turn stored in
20107 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
20108 could be PL_curpm in multiple contexts, and could require multiple
20109 result sets being associated with the pattern simultaneously, such
20110 as when doing a recursive match with (??{$qr})
20112 The solution is to make a lightweight copy of the regexp structure
20113 when a qr// is returned from the code executed by (??{$qr}) this
20114 lightweight copy doesn't actually own any of its data except for
20115 the starp/end and the actual regexp structure itself.
20121 Perl_reg_temp_copy(pTHX_ REGEXP *dsv, REGEXP *ssv)
20123 struct regexp *drx;
20124 struct regexp *const srx = ReANY(ssv);
20125 const bool islv = dsv && SvTYPE(dsv) == SVt_PVLV;
20127 PERL_ARGS_ASSERT_REG_TEMP_COPY;
20130 dsv = (REGEXP*) newSV_type(SVt_REGEXP);
20132 SvOK_off((SV *)dsv);
20134 /* For PVLVs, the head (sv_any) points to an XPVLV, while
20135 * the LV's xpvlenu_rx will point to a regexp body, which
20136 * we allocate here */
20137 REGEXP *temp = (REGEXP *)newSV_type(SVt_REGEXP);
20138 assert(!SvPVX(dsv));
20139 ((XPV*)SvANY(dsv))->xpv_len_u.xpvlenu_rx = temp->sv_any;
20140 temp->sv_any = NULL;
20141 SvFLAGS(temp) = (SvFLAGS(temp) & ~SVTYPEMASK) | SVt_NULL;
20142 SvREFCNT_dec_NN(temp);
20143 /* SvCUR still resides in the xpvlv struct, so the regexp copy-
20144 ing below will not set it. */
20145 SvCUR_set(dsv, SvCUR(ssv));
20148 /* This ensures that SvTHINKFIRST(sv) is true, and hence that
20149 sv_force_normal(sv) is called. */
20153 SvFLAGS(dsv) |= SvFLAGS(ssv) & (SVf_POK|SVp_POK|SVf_UTF8);
20154 SvPV_set(dsv, RX_WRAPPED(ssv));
20155 /* We share the same string buffer as the original regexp, on which we
20156 hold a reference count, incremented when mother_re is set below.
20157 The string pointer is copied here, being part of the regexp struct.
20159 memcpy(&(drx->xpv_cur), &(srx->xpv_cur),
20160 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
20164 const I32 npar = srx->nparens+1;
20165 Newx(drx->offs, npar, regexp_paren_pair);
20166 Copy(srx->offs, drx->offs, npar, regexp_paren_pair);
20168 if (srx->substrs) {
20170 Newx(drx->substrs, 1, struct reg_substr_data);
20171 StructCopy(srx->substrs, drx->substrs, struct reg_substr_data);
20173 for (i = 0; i < 2; i++) {
20174 SvREFCNT_inc_void(drx->substrs->data[i].substr);
20175 SvREFCNT_inc_void(drx->substrs->data[i].utf8_substr);
20178 /* check_substr and check_utf8, if non-NULL, point to either their
20179 anchored or float namesakes, and don't hold a second reference. */
20181 RX_MATCH_COPIED_off(dsv);
20182 #ifdef PERL_ANY_COW
20183 drx->saved_copy = NULL;
20185 drx->mother_re = ReREFCNT_inc(srx->mother_re ? srx->mother_re : ssv);
20186 SvREFCNT_inc_void(drx->qr_anoncv);
20187 if (srx->recurse_locinput)
20188 Newx(drx->recurse_locinput,srx->nparens + 1,char *);
20195 /* regfree_internal()
20197 Free the private data in a regexp. This is overloadable by
20198 extensions. Perl takes care of the regexp structure in pregfree(),
20199 this covers the *pprivate pointer which technically perl doesn't
20200 know about, however of course we have to handle the
20201 regexp_internal structure when no extension is in use.
20203 Note this is called before freeing anything in the regexp
20208 Perl_regfree_internal(pTHX_ REGEXP * const rx)
20210 struct regexp *const r = ReANY(rx);
20211 RXi_GET_DECL(r,ri);
20212 GET_RE_DEBUG_FLAGS_DECL;
20214 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
20220 SV *dsv= sv_newmortal();
20221 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
20222 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), PL_dump_re_max_len);
20223 Perl_re_printf( aTHX_ "%sFreeing REx:%s %s\n",
20224 PL_colors[4],PL_colors[5],s);
20227 #ifdef RE_TRACK_PATTERN_OFFSETS
20229 Safefree(ri->u.offsets); /* 20010421 MJD */
20231 if (ri->code_blocks)
20232 S_free_codeblocks(aTHX_ ri->code_blocks);
20235 int n = ri->data->count;
20238 /* If you add a ->what type here, update the comment in regcomp.h */
20239 switch (ri->data->what[n]) {
20245 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
20248 Safefree(ri->data->data[n]);
20254 { /* Aho Corasick add-on structure for a trie node.
20255 Used in stclass optimization only */
20257 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
20258 #ifdef USE_ITHREADS
20262 refcount = --aho->refcount;
20265 PerlMemShared_free(aho->states);
20266 PerlMemShared_free(aho->fail);
20267 /* do this last!!!! */
20268 PerlMemShared_free(ri->data->data[n]);
20269 /* we should only ever get called once, so
20270 * assert as much, and also guard the free
20271 * which /might/ happen twice. At the least
20272 * it will make code anlyzers happy and it
20273 * doesn't cost much. - Yves */
20274 assert(ri->regstclass);
20275 if (ri->regstclass) {
20276 PerlMemShared_free(ri->regstclass);
20277 ri->regstclass = 0;
20284 /* trie structure. */
20286 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
20287 #ifdef USE_ITHREADS
20291 refcount = --trie->refcount;
20294 PerlMemShared_free(trie->charmap);
20295 PerlMemShared_free(trie->states);
20296 PerlMemShared_free(trie->trans);
20298 PerlMemShared_free(trie->bitmap);
20300 PerlMemShared_free(trie->jump);
20301 PerlMemShared_free(trie->wordinfo);
20302 /* do this last!!!! */
20303 PerlMemShared_free(ri->data->data[n]);
20308 Perl_croak(aTHX_ "panic: regfree data code '%c'",
20309 ri->data->what[n]);
20312 Safefree(ri->data->what);
20313 Safefree(ri->data);
20319 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
20320 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
20321 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
20324 re_dup_guts - duplicate a regexp.
20326 This routine is expected to clone a given regexp structure. It is only
20327 compiled under USE_ITHREADS.
20329 After all of the core data stored in struct regexp is duplicated
20330 the regexp_engine.dupe method is used to copy any private data
20331 stored in the *pprivate pointer. This allows extensions to handle
20332 any duplication it needs to do.
20334 See pregfree() and regfree_internal() if you change anything here.
20336 #if defined(USE_ITHREADS)
20337 #ifndef PERL_IN_XSUB_RE
20339 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
20343 const struct regexp *r = ReANY(sstr);
20344 struct regexp *ret = ReANY(dstr);
20346 PERL_ARGS_ASSERT_RE_DUP_GUTS;
20348 npar = r->nparens+1;
20349 Newx(ret->offs, npar, regexp_paren_pair);
20350 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
20352 if (ret->substrs) {
20353 /* Do it this way to avoid reading from *r after the StructCopy().
20354 That way, if any of the sv_dup_inc()s dislodge *r from the L1
20355 cache, it doesn't matter. */
20357 const bool anchored = r->check_substr
20358 ? r->check_substr == r->substrs->data[0].substr
20359 : r->check_utf8 == r->substrs->data[0].utf8_substr;
20360 Newx(ret->substrs, 1, struct reg_substr_data);
20361 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
20363 for (i = 0; i < 2; i++) {
20364 ret->substrs->data[i].substr =
20365 sv_dup_inc(ret->substrs->data[i].substr, param);
20366 ret->substrs->data[i].utf8_substr =
20367 sv_dup_inc(ret->substrs->data[i].utf8_substr, param);
20370 /* check_substr and check_utf8, if non-NULL, point to either their
20371 anchored or float namesakes, and don't hold a second reference. */
20373 if (ret->check_substr) {
20375 assert(r->check_utf8 == r->substrs->data[0].utf8_substr);
20377 ret->check_substr = ret->substrs->data[0].substr;
20378 ret->check_utf8 = ret->substrs->data[0].utf8_substr;
20380 assert(r->check_substr == r->substrs->data[1].substr);
20381 assert(r->check_utf8 == r->substrs->data[1].utf8_substr);
20383 ret->check_substr = ret->substrs->data[1].substr;
20384 ret->check_utf8 = ret->substrs->data[1].utf8_substr;
20386 } else if (ret->check_utf8) {
20388 ret->check_utf8 = ret->substrs->data[0].utf8_substr;
20390 ret->check_utf8 = ret->substrs->data[1].utf8_substr;
20395 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
20396 ret->qr_anoncv = MUTABLE_CV(sv_dup_inc((const SV *)ret->qr_anoncv, param));
20397 if (r->recurse_locinput)
20398 Newx(ret->recurse_locinput,r->nparens + 1,char *);
20401 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
20403 if (RX_MATCH_COPIED(dstr))
20404 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
20406 ret->subbeg = NULL;
20407 #ifdef PERL_ANY_COW
20408 ret->saved_copy = NULL;
20411 /* Whether mother_re be set or no, we need to copy the string. We
20412 cannot refrain from copying it when the storage points directly to
20413 our mother regexp, because that's
20414 1: a buffer in a different thread
20415 2: something we no longer hold a reference on
20416 so we need to copy it locally. */
20417 RX_WRAPPED(dstr) = SAVEPVN(RX_WRAPPED_const(sstr), SvCUR(sstr)+1);
20418 ret->mother_re = NULL;
20420 #endif /* PERL_IN_XSUB_RE */
20425 This is the internal complement to regdupe() which is used to copy
20426 the structure pointed to by the *pprivate pointer in the regexp.
20427 This is the core version of the extension overridable cloning hook.
20428 The regexp structure being duplicated will be copied by perl prior
20429 to this and will be provided as the regexp *r argument, however
20430 with the /old/ structures pprivate pointer value. Thus this routine
20431 may override any copying normally done by perl.
20433 It returns a pointer to the new regexp_internal structure.
20437 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
20440 struct regexp *const r = ReANY(rx);
20441 regexp_internal *reti;
20443 RXi_GET_DECL(r,ri);
20445 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
20449 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode),
20450 char, regexp_internal);
20451 Copy(ri->program, reti->program, len+1, regnode);
20454 if (ri->code_blocks) {
20456 Newx(reti->code_blocks, 1, struct reg_code_blocks);
20457 Newx(reti->code_blocks->cb, ri->code_blocks->count,
20458 struct reg_code_block);
20459 Copy(ri->code_blocks->cb, reti->code_blocks->cb,
20460 ri->code_blocks->count, struct reg_code_block);
20461 for (n = 0; n < ri->code_blocks->count; n++)
20462 reti->code_blocks->cb[n].src_regex = (REGEXP*)
20463 sv_dup_inc((SV*)(ri->code_blocks->cb[n].src_regex), param);
20464 reti->code_blocks->count = ri->code_blocks->count;
20465 reti->code_blocks->refcnt = 1;
20468 reti->code_blocks = NULL;
20470 reti->regstclass = NULL;
20473 struct reg_data *d;
20474 const int count = ri->data->count;
20477 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
20478 char, struct reg_data);
20479 Newx(d->what, count, U8);
20482 for (i = 0; i < count; i++) {
20483 d->what[i] = ri->data->what[i];
20484 switch (d->what[i]) {
20485 /* see also regcomp.h and regfree_internal() */
20486 case 'a': /* actually an AV, but the dup function is identical.
20487 values seem to be "plain sv's" generally. */
20488 case 'r': /* a compiled regex (but still just another SV) */
20489 case 's': /* an RV (currently only used for an RV to an AV by the ANYOF code)
20490 this use case should go away, the code could have used
20491 'a' instead - see S_set_ANYOF_arg() for array contents. */
20492 case 'S': /* actually an SV, but the dup function is identical. */
20493 case 'u': /* actually an HV, but the dup function is identical.
20494 values are "plain sv's" */
20495 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
20498 /* Synthetic Start Class - "Fake" charclass we generate to optimize
20499 * patterns which could start with several different things. Pre-TRIE
20500 * this was more important than it is now, however this still helps
20501 * in some places, for instance /x?a+/ might produce a SSC equivalent
20502 * to [xa]. This is used by Perl_re_intuit_start() and S_find_byclass()
20505 /* This is cheating. */
20506 Newx(d->data[i], 1, regnode_ssc);
20507 StructCopy(ri->data->data[i], d->data[i], regnode_ssc);
20508 reti->regstclass = (regnode*)d->data[i];
20511 /* AHO-CORASICK fail table */
20512 /* Trie stclasses are readonly and can thus be shared
20513 * without duplication. We free the stclass in pregfree
20514 * when the corresponding reg_ac_data struct is freed.
20516 reti->regstclass= ri->regstclass;
20519 /* TRIE transition table */
20521 ((reg_trie_data*)ri->data->data[i])->refcount++;
20524 case 'l': /* (?{...}) or (??{ ... }) code (cb->block) */
20525 case 'L': /* same when RExC_pm_flags & PMf_HAS_CV and code
20526 is not from another regexp */
20527 d->data[i] = ri->data->data[i];
20530 Perl_croak(aTHX_ "panic: re_dup_guts unknown data code '%c'",
20531 ri->data->what[i]);
20540 reti->name_list_idx = ri->name_list_idx;
20542 #ifdef RE_TRACK_PATTERN_OFFSETS
20543 if (ri->u.offsets) {
20544 Newx(reti->u.offsets, 2*len+1, U32);
20545 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
20548 SetProgLen(reti,len);
20551 return (void*)reti;
20554 #endif /* USE_ITHREADS */
20556 #ifndef PERL_IN_XSUB_RE
20559 - regnext - dig the "next" pointer out of a node
20562 Perl_regnext(pTHX_ regnode *p)
20569 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
20570 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d",
20571 (int)OP(p), (int)REGNODE_MAX);
20574 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
20583 S_re_croak2(pTHX_ bool utf8, const char* pat1,const char* pat2,...)
20586 STRLEN l1 = strlen(pat1);
20587 STRLEN l2 = strlen(pat2);
20590 const char *message;
20592 PERL_ARGS_ASSERT_RE_CROAK2;
20598 Copy(pat1, buf, l1 , char);
20599 Copy(pat2, buf + l1, l2 , char);
20600 buf[l1 + l2] = '\n';
20601 buf[l1 + l2 + 1] = '\0';
20602 va_start(args, pat2);
20603 msv = vmess(buf, &args);
20605 message = SvPV_const(msv,l1);
20608 Copy(message, buf, l1 , char);
20609 /* l1-1 to avoid \n */
20610 Perl_croak(aTHX_ "%" UTF8f, UTF8fARG(utf8, l1-1, buf));
20613 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
20615 #ifndef PERL_IN_XSUB_RE
20617 Perl_save_re_context(pTHX)
20622 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
20625 const REGEXP * const rx = PM_GETRE(PL_curpm);
20627 nparens = RX_NPARENS(rx);
20630 /* RT #124109. This is a complete hack; in the SWASHNEW case we know
20631 * that PL_curpm will be null, but that utf8.pm and the modules it
20632 * loads will only use $1..$3.
20633 * The t/porting/re_context.t test file checks this assumption.
20638 for (i = 1; i <= nparens; i++) {
20639 char digits[TYPE_CHARS(long)];
20640 const STRLEN len = my_snprintf(digits, sizeof(digits),
20642 GV *const *const gvp
20643 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
20646 GV * const gv = *gvp;
20647 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
20657 S_put_code_point(pTHX_ SV *sv, UV c)
20659 PERL_ARGS_ASSERT_PUT_CODE_POINT;
20662 Perl_sv_catpvf(aTHX_ sv, "\\x{%04" UVXf "}", c);
20664 else if (isPRINT(c)) {
20665 const char string = (char) c;
20667 /* We use {phrase} as metanotation in the class, so also escape literal
20669 if (isBACKSLASHED_PUNCT(c) || c == '{' || c == '}')
20670 sv_catpvs(sv, "\\");
20671 sv_catpvn(sv, &string, 1);
20673 else if (isMNEMONIC_CNTRL(c)) {
20674 Perl_sv_catpvf(aTHX_ sv, "%s", cntrl_to_mnemonic((U8) c));
20677 Perl_sv_catpvf(aTHX_ sv, "\\x%02X", (U8) c);
20681 #define MAX_PRINT_A MAX_PRINT_A_FOR_USE_ONLY_BY_REGCOMP_DOT_C
20684 S_put_range(pTHX_ SV *sv, UV start, const UV end, const bool allow_literals)
20686 /* Appends to 'sv' a displayable version of the range of code points from
20687 * 'start' to 'end'. Mnemonics (like '\r') are used for the few controls
20688 * that have them, when they occur at the beginning or end of the range.
20689 * It uses hex to output the remaining code points, unless 'allow_literals'
20690 * is true, in which case the printable ASCII ones are output as-is (though
20691 * some of these will be escaped by put_code_point()).
20693 * NOTE: This is designed only for printing ranges of code points that fit
20694 * inside an ANYOF bitmap. Higher code points are simply suppressed
20697 const unsigned int min_range_count = 3;
20699 assert(start <= end);
20701 PERL_ARGS_ASSERT_PUT_RANGE;
20703 while (start <= end) {
20705 const char * format;
20707 if (end - start < min_range_count) {
20709 /* Output chars individually when they occur in short ranges */
20710 for (; start <= end; start++) {
20711 put_code_point(sv, start);
20716 /* If permitted by the input options, and there is a possibility that
20717 * this range contains a printable literal, look to see if there is
20719 if (allow_literals && start <= MAX_PRINT_A) {
20721 /* If the character at the beginning of the range isn't an ASCII
20722 * printable, effectively split the range into two parts:
20723 * 1) the portion before the first such printable,
20725 * and output them separately. */
20726 if (! isPRINT_A(start)) {
20727 UV temp_end = start + 1;
20729 /* There is no point looking beyond the final possible
20730 * printable, in MAX_PRINT_A */
20731 UV max = MIN(end, MAX_PRINT_A);
20733 while (temp_end <= max && ! isPRINT_A(temp_end)) {
20737 /* Here, temp_end points to one beyond the first printable if
20738 * found, or to one beyond 'max' if not. If none found, make
20739 * sure that we use the entire range */
20740 if (temp_end > MAX_PRINT_A) {
20741 temp_end = end + 1;
20744 /* Output the first part of the split range: the part that
20745 * doesn't have printables, with the parameter set to not look
20746 * for literals (otherwise we would infinitely recurse) */
20747 put_range(sv, start, temp_end - 1, FALSE);
20749 /* The 2nd part of the range (if any) starts here. */
20752 /* We do a continue, instead of dropping down, because even if
20753 * the 2nd part is non-empty, it could be so short that we want
20754 * to output it as individual characters, as tested for at the
20755 * top of this loop. */
20759 /* Here, 'start' is a printable ASCII. If it is an alphanumeric,
20760 * output a sub-range of just the digits or letters, then process
20761 * the remaining portion as usual. */
20762 if (isALPHANUMERIC_A(start)) {
20763 UV mask = (isDIGIT_A(start))
20768 UV temp_end = start + 1;
20770 /* Find the end of the sub-range that includes just the
20771 * characters in the same class as the first character in it */
20772 while (temp_end <= end && _generic_isCC_A(temp_end, mask)) {
20777 /* For short ranges, don't duplicate the code above to output
20778 * them; just call recursively */
20779 if (temp_end - start < min_range_count) {
20780 put_range(sv, start, temp_end, FALSE);
20782 else { /* Output as a range */
20783 put_code_point(sv, start);
20784 sv_catpvs(sv, "-");
20785 put_code_point(sv, temp_end);
20787 start = temp_end + 1;
20791 /* We output any other printables as individual characters */
20792 if (isPUNCT_A(start) || isSPACE_A(start)) {
20793 while (start <= end && (isPUNCT_A(start)
20794 || isSPACE_A(start)))
20796 put_code_point(sv, start);
20801 } /* End of looking for literals */
20803 /* Here is not to output as a literal. Some control characters have
20804 * mnemonic names. Split off any of those at the beginning and end of
20805 * the range to print mnemonically. It isn't possible for many of
20806 * these to be in a row, so this won't overwhelm with output */
20808 && (isMNEMONIC_CNTRL(start) || isMNEMONIC_CNTRL(end)))
20810 while (isMNEMONIC_CNTRL(start) && start <= end) {
20811 put_code_point(sv, start);
20815 /* If this didn't take care of the whole range ... */
20816 if (start <= end) {
20818 /* Look backwards from the end to find the final non-mnemonic
20821 while (isMNEMONIC_CNTRL(temp_end)) {
20825 /* And separately output the interior range that doesn't start
20826 * or end with mnemonics */
20827 put_range(sv, start, temp_end, FALSE);
20829 /* Then output the mnemonic trailing controls */
20830 start = temp_end + 1;
20831 while (start <= end) {
20832 put_code_point(sv, start);
20839 /* As a final resort, output the range or subrange as hex. */
20841 this_end = (end < NUM_ANYOF_CODE_POINTS)
20843 : NUM_ANYOF_CODE_POINTS - 1;
20844 #if NUM_ANYOF_CODE_POINTS > 256
20845 format = (this_end < 256)
20846 ? "\\x%02" UVXf "-\\x%02" UVXf
20847 : "\\x{%04" UVXf "}-\\x{%04" UVXf "}";
20849 format = "\\x%02" UVXf "-\\x%02" UVXf;
20851 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
20852 Perl_sv_catpvf(aTHX_ sv, format, start, this_end);
20853 GCC_DIAG_RESTORE_STMT;
20859 S_put_charclass_bitmap_innards_invlist(pTHX_ SV *sv, SV* invlist)
20861 /* Concatenate onto the PV in 'sv' a displayable form of the inversion list
20865 bool allow_literals = TRUE;
20867 PERL_ARGS_ASSERT_PUT_CHARCLASS_BITMAP_INNARDS_INVLIST;
20869 /* Generally, it is more readable if printable characters are output as
20870 * literals, but if a range (nearly) spans all of them, it's best to output
20871 * it as a single range. This code will use a single range if all but 2
20872 * ASCII printables are in it */
20873 invlist_iterinit(invlist);
20874 while (invlist_iternext(invlist, &start, &end)) {
20876 /* If the range starts beyond the final printable, it doesn't have any
20878 if (start > MAX_PRINT_A) {
20882 /* In both ASCII and EBCDIC, a SPACE is the lowest printable. To span
20883 * all but two, the range must start and end no later than 2 from
20885 if (start < ' ' + 2 && end > MAX_PRINT_A - 2) {
20886 if (end > MAX_PRINT_A) {
20892 if (end - start >= MAX_PRINT_A - ' ' - 2) {
20893 allow_literals = FALSE;
20898 invlist_iterfinish(invlist);
20900 /* Here we have figured things out. Output each range */
20901 invlist_iterinit(invlist);
20902 while (invlist_iternext(invlist, &start, &end)) {
20903 if (start >= NUM_ANYOF_CODE_POINTS) {
20906 put_range(sv, start, end, allow_literals);
20908 invlist_iterfinish(invlist);
20914 S_put_charclass_bitmap_innards_common(pTHX_
20915 SV* invlist, /* The bitmap */
20916 SV* posixes, /* Under /l, things like [:word:], \S */
20917 SV* only_utf8, /* Under /d, matches iff the target is UTF-8 */
20918 SV* not_utf8, /* /d, matches iff the target isn't UTF-8 */
20919 SV* only_utf8_locale, /* Under /l, matches if the locale is UTF-8 */
20920 const bool invert /* Is the result to be inverted? */
20923 /* Create and return an SV containing a displayable version of the bitmap
20924 * and associated information determined by the input parameters. If the
20925 * output would have been only the inversion indicator '^', NULL is instead
20930 PERL_ARGS_ASSERT_PUT_CHARCLASS_BITMAP_INNARDS_COMMON;
20933 output = newSVpvs("^");
20936 output = newSVpvs("");
20939 /* First, the code points in the bitmap that are unconditionally there */
20940 put_charclass_bitmap_innards_invlist(output, invlist);
20942 /* Traditionally, these have been placed after the main code points */
20944 sv_catsv(output, posixes);
20947 if (only_utf8 && _invlist_len(only_utf8)) {
20948 Perl_sv_catpvf(aTHX_ output, "%s{utf8}%s", PL_colors[1], PL_colors[0]);
20949 put_charclass_bitmap_innards_invlist(output, only_utf8);
20952 if (not_utf8 && _invlist_len(not_utf8)) {
20953 Perl_sv_catpvf(aTHX_ output, "%s{not utf8}%s", PL_colors[1], PL_colors[0]);
20954 put_charclass_bitmap_innards_invlist(output, not_utf8);
20957 if (only_utf8_locale && _invlist_len(only_utf8_locale)) {
20958 Perl_sv_catpvf(aTHX_ output, "%s{utf8 locale}%s", PL_colors[1], PL_colors[0]);
20959 put_charclass_bitmap_innards_invlist(output, only_utf8_locale);
20961 /* This is the only list in this routine that can legally contain code
20962 * points outside the bitmap range. The call just above to
20963 * 'put_charclass_bitmap_innards_invlist' will simply suppress them, so
20964 * output them here. There's about a half-dozen possible, and none in
20965 * contiguous ranges longer than 2 */
20966 if (invlist_highest(only_utf8_locale) >= NUM_ANYOF_CODE_POINTS) {
20968 SV* above_bitmap = NULL;
20970 _invlist_subtract(only_utf8_locale, PL_InBitmap, &above_bitmap);
20972 invlist_iterinit(above_bitmap);
20973 while (invlist_iternext(above_bitmap, &start, &end)) {
20976 for (i = start; i <= end; i++) {
20977 put_code_point(output, i);
20980 invlist_iterfinish(above_bitmap);
20981 SvREFCNT_dec_NN(above_bitmap);
20985 if (invert && SvCUR(output) == 1) {
20993 S_put_charclass_bitmap_innards(pTHX_ SV *sv,
20995 SV *nonbitmap_invlist,
20996 SV *only_utf8_locale_invlist,
20997 const regnode * const node,
20998 const bool force_as_is_display)
21000 /* Appends to 'sv' a displayable version of the innards of the bracketed
21001 * character class defined by the other arguments:
21002 * 'bitmap' points to the bitmap, or NULL if to ignore that.
21003 * 'nonbitmap_invlist' is an inversion list of the code points that are in
21004 * the bitmap range, but for some reason aren't in the bitmap; NULL if
21005 * none. The reasons for this could be that they require some
21006 * condition such as the target string being or not being in UTF-8
21007 * (under /d), or because they came from a user-defined property that
21008 * was not resolved at the time of the regex compilation (under /u)
21009 * 'only_utf8_locale_invlist' is an inversion list of the code points that
21010 * are valid only if the runtime locale is a UTF-8 one; NULL if none
21011 * 'node' is the regex pattern ANYOF node. It is needed only when the
21012 * above two parameters are not null, and is passed so that this
21013 * routine can tease apart the various reasons for them.
21014 * 'force_as_is_display' is TRUE if this routine should definitely NOT try
21015 * to invert things to see if that leads to a cleaner display. If
21016 * FALSE, this routine is free to use its judgment about doing this.
21018 * It returns TRUE if there was actually something output. (It may be that
21019 * the bitmap, etc is empty.)
21021 * When called for outputting the bitmap of a non-ANYOF node, just pass the
21022 * bitmap, with the succeeding parameters set to NULL, and the final one to
21026 /* In general, it tries to display the 'cleanest' representation of the
21027 * innards, choosing whether to display them inverted or not, regardless of
21028 * whether the class itself is to be inverted. However, there are some
21029 * cases where it can't try inverting, as what actually matches isn't known
21030 * until runtime, and hence the inversion isn't either. */
21031 bool inverting_allowed = ! force_as_is_display;
21034 STRLEN orig_sv_cur = SvCUR(sv);
21036 SV* invlist; /* Inversion list we accumulate of code points that
21037 are unconditionally matched */
21038 SV* only_utf8 = NULL; /* Under /d, list of matches iff the target is
21040 SV* not_utf8 = NULL; /* /d, list of matches iff the target isn't UTF-8
21042 SV* posixes = NULL; /* Under /l, string of things like [:word:], \D */
21043 SV* only_utf8_locale = NULL; /* Under /l, list of matches if the locale
21046 SV* as_is_display; /* The output string when we take the inputs
21048 SV* inverted_display; /* The output string when we invert the inputs */
21050 U8 flags = (node) ? ANYOF_FLAGS(node) : 0;
21052 bool invert = cBOOL(flags & ANYOF_INVERT); /* Is the input to be inverted
21054 /* We are biased in favor of displaying things without them being inverted,
21055 * as that is generally easier to understand */
21056 const int bias = 5;
21058 PERL_ARGS_ASSERT_PUT_CHARCLASS_BITMAP_INNARDS;
21060 /* Start off with whatever code points are passed in. (We clone, so we
21061 * don't change the caller's list) */
21062 if (nonbitmap_invlist) {
21063 assert(invlist_highest(nonbitmap_invlist) < NUM_ANYOF_CODE_POINTS);
21064 invlist = invlist_clone(nonbitmap_invlist);
21066 else { /* Worst case size is every other code point is matched */
21067 invlist = _new_invlist(NUM_ANYOF_CODE_POINTS / 2);
21071 if (OP(node) == ANYOFD) {
21073 /* This flag indicates that the code points below 0x100 in the
21074 * nonbitmap list are precisely the ones that match only when the
21075 * target is UTF-8 (they should all be non-ASCII). */
21076 if (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)
21078 _invlist_intersection(invlist, PL_UpperLatin1, &only_utf8);
21079 _invlist_subtract(invlist, only_utf8, &invlist);
21082 /* And this flag for matching all non-ASCII 0xFF and below */
21083 if (flags & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
21085 not_utf8 = invlist_clone(PL_UpperLatin1);
21088 else if (OP(node) == ANYOFL) {
21090 /* If either of these flags are set, what matches isn't
21091 * determinable except during execution, so don't know enough here
21093 if (flags & (ANYOFL_FOLD|ANYOF_MATCHES_POSIXL)) {
21094 inverting_allowed = FALSE;
21097 /* What the posix classes match also varies at runtime, so these
21098 * will be output symbolically. */
21099 if (ANYOF_POSIXL_TEST_ANY_SET(node)) {
21102 posixes = newSVpvs("");
21103 for (i = 0; i < ANYOF_POSIXL_MAX; i++) {
21104 if (ANYOF_POSIXL_TEST(node,i)) {
21105 sv_catpv(posixes, anyofs[i]);
21112 /* Accumulate the bit map into the unconditional match list */
21114 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
21115 if (BITMAP_TEST(bitmap, i)) {
21118 i < NUM_ANYOF_CODE_POINTS && BITMAP_TEST(bitmap, i);
21121 invlist = _add_range_to_invlist(invlist, start, i-1);
21126 /* Make sure that the conditional match lists don't have anything in them
21127 * that match unconditionally; otherwise the output is quite confusing.
21128 * This could happen if the code that populates these misses some
21131 _invlist_subtract(only_utf8, invlist, &only_utf8);
21134 _invlist_subtract(not_utf8, invlist, ¬_utf8);
21137 if (only_utf8_locale_invlist) {
21139 /* Since this list is passed in, we have to make a copy before
21141 only_utf8_locale = invlist_clone(only_utf8_locale_invlist);
21143 _invlist_subtract(only_utf8_locale, invlist, &only_utf8_locale);
21145 /* And, it can get really weird for us to try outputting an inverted
21146 * form of this list when it has things above the bitmap, so don't even
21148 if (invlist_highest(only_utf8_locale) >= NUM_ANYOF_CODE_POINTS) {
21149 inverting_allowed = FALSE;
21153 /* Calculate what the output would be if we take the input as-is */
21154 as_is_display = put_charclass_bitmap_innards_common(invlist,
21161 /* If have to take the output as-is, just do that */
21162 if (! inverting_allowed) {
21163 if (as_is_display) {
21164 sv_catsv(sv, as_is_display);
21165 SvREFCNT_dec_NN(as_is_display);
21168 else { /* But otherwise, create the output again on the inverted input, and
21169 use whichever version is shorter */
21171 int inverted_bias, as_is_bias;
21173 /* We will apply our bias to whichever of the the results doesn't have
21183 inverted_bias = bias;
21186 /* Now invert each of the lists that contribute to the output,
21187 * excluding from the result things outside the possible range */
21189 /* For the unconditional inversion list, we have to add in all the
21190 * conditional code points, so that when inverted, they will be gone
21192 _invlist_union(only_utf8, invlist, &invlist);
21193 _invlist_union(not_utf8, invlist, &invlist);
21194 _invlist_union(only_utf8_locale, invlist, &invlist);
21195 _invlist_invert(invlist);
21196 _invlist_intersection(invlist, PL_InBitmap, &invlist);
21199 _invlist_invert(only_utf8);
21200 _invlist_intersection(only_utf8, PL_UpperLatin1, &only_utf8);
21202 else if (not_utf8) {
21204 /* If a code point matches iff the target string is not in UTF-8,
21205 * then complementing the result has it not match iff not in UTF-8,
21206 * which is the same thing as matching iff it is UTF-8. */
21207 only_utf8 = not_utf8;
21211 if (only_utf8_locale) {
21212 _invlist_invert(only_utf8_locale);
21213 _invlist_intersection(only_utf8_locale,
21215 &only_utf8_locale);
21218 inverted_display = put_charclass_bitmap_innards_common(
21223 only_utf8_locale, invert);
21225 /* Use the shortest representation, taking into account our bias
21226 * against showing it inverted */
21227 if ( inverted_display
21228 && ( ! as_is_display
21229 || ( SvCUR(inverted_display) + inverted_bias
21230 < SvCUR(as_is_display) + as_is_bias)))
21232 sv_catsv(sv, inverted_display);
21234 else if (as_is_display) {
21235 sv_catsv(sv, as_is_display);
21238 SvREFCNT_dec(as_is_display);
21239 SvREFCNT_dec(inverted_display);
21242 SvREFCNT_dec_NN(invlist);
21243 SvREFCNT_dec(only_utf8);
21244 SvREFCNT_dec(not_utf8);
21245 SvREFCNT_dec(posixes);
21246 SvREFCNT_dec(only_utf8_locale);
21248 return SvCUR(sv) > orig_sv_cur;
21251 #define CLEAR_OPTSTART \
21252 if (optstart) STMT_START { \
21253 DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ \
21254 " (%" IVdf " nodes)\n", (IV)(node - optstart))); \
21258 #define DUMPUNTIL(b,e) \
21260 node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
21262 STATIC const regnode *
21263 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
21264 const regnode *last, const regnode *plast,
21265 SV* sv, I32 indent, U32 depth)
21267 U8 op = PSEUDO; /* Arbitrary non-END op. */
21268 const regnode *next;
21269 const regnode *optstart= NULL;
21271 RXi_GET_DECL(r,ri);
21272 GET_RE_DEBUG_FLAGS_DECL;
21274 PERL_ARGS_ASSERT_DUMPUNTIL;
21276 #ifdef DEBUG_DUMPUNTIL
21277 Perl_re_printf( aTHX_ "--- %d : %d - %d - %d\n",indent,node-start,
21278 last ? last-start : 0,plast ? plast-start : 0);
21281 if (plast && plast < last)
21284 while (PL_regkind[op] != END && (!last || node < last)) {
21286 /* While that wasn't END last time... */
21289 if (op == CLOSE || op == SRCLOSE || op == WHILEM)
21291 next = regnext((regnode *)node);
21294 if (OP(node) == OPTIMIZED) {
21295 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
21302 regprop(r, sv, node, NULL, NULL);
21303 Perl_re_printf( aTHX_ "%4" IVdf ":%*s%s", (IV)(node - start),
21304 (int)(2*indent + 1), "", SvPVX_const(sv));
21306 if (OP(node) != OPTIMIZED) {
21307 if (next == NULL) /* Next ptr. */
21308 Perl_re_printf( aTHX_ " (0)");
21309 else if (PL_regkind[(U8)op] == BRANCH
21310 && PL_regkind[OP(next)] != BRANCH )
21311 Perl_re_printf( aTHX_ " (FAIL)");
21313 Perl_re_printf( aTHX_ " (%" IVdf ")", (IV)(next - start));
21314 Perl_re_printf( aTHX_ "\n");
21318 if (PL_regkind[(U8)op] == BRANCHJ) {
21321 const regnode *nnode = (OP(next) == LONGJMP
21322 ? regnext((regnode *)next)
21324 if (last && nnode > last)
21326 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
21329 else if (PL_regkind[(U8)op] == BRANCH) {
21331 DUMPUNTIL(NEXTOPER(node), next);
21333 else if ( PL_regkind[(U8)op] == TRIE ) {
21334 const regnode *this_trie = node;
21335 const char op = OP(node);
21336 const U32 n = ARG(node);
21337 const reg_ac_data * const ac = op>=AHOCORASICK ?
21338 (reg_ac_data *)ri->data->data[n] :
21340 const reg_trie_data * const trie =
21341 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
21343 AV *const trie_words
21344 = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
21346 const regnode *nextbranch= NULL;
21349 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
21350 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
21352 Perl_re_indentf( aTHX_ "%s ",
21355 ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr),
21356 SvCUR(*elem_ptr), PL_dump_re_max_len,
21357 PL_colors[0], PL_colors[1],
21359 ? PERL_PV_ESCAPE_UNI
21361 | PERL_PV_PRETTY_ELLIPSES
21362 | PERL_PV_PRETTY_LTGT
21367 U16 dist= trie->jump[word_idx+1];
21368 Perl_re_printf( aTHX_ "(%" UVuf ")\n",
21369 (UV)((dist ? this_trie + dist : next) - start));
21372 nextbranch= this_trie + trie->jump[0];
21373 DUMPUNTIL(this_trie + dist, nextbranch);
21375 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
21376 nextbranch= regnext((regnode *)nextbranch);
21378 Perl_re_printf( aTHX_ "\n");
21381 if (last && next > last)
21386 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
21387 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
21388 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
21390 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
21392 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
21394 else if ( op == PLUS || op == STAR) {
21395 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
21397 else if (PL_regkind[(U8)op] == ANYOF) {
21398 /* arglen 1 + class block */
21399 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_MATCHES_POSIXL)
21400 ? ANYOF_POSIXL_SKIP
21402 node = NEXTOPER(node);
21404 else if (PL_regkind[(U8)op] == EXACT) {
21405 /* Literal string, where present. */
21406 node += NODE_SZ_STR(node) - 1;
21407 node = NEXTOPER(node);
21410 node = NEXTOPER(node);
21411 node += regarglen[(U8)op];
21413 if (op == CURLYX || op == OPEN || op == SROPEN)
21417 #ifdef DEBUG_DUMPUNTIL
21418 Perl_re_printf( aTHX_ "--- %d\n", (int)indent);
21423 #endif /* DEBUGGING */
21426 * ex: set ts=8 sts=4 sw=4 et: