This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: Remove no longer useful code
[perl5.git] / regcomp.c
CommitLineData
a0d0e21e
LW
1/* regcomp.c
2 */
3
4/*
4ac71550
TC
5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
6 *
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
a0d0e21e
LW
8 */
9
61296642
DM
10/* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
166f8a29 12 * a regular expression.
e4a054ea
DM
13 *
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.
166f8a29
DM
18 */
19
a687059c
LW
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!
22 */
23
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.
27 */
28
e50aee73 29/* The names of the functions have been changed from regcomp and
3b753521 30 * regexec to pregcomp and pregexec in order to avoid conflicts
e50aee73
AD
31 * with the POSIX routines of the same names.
32*/
33
b9d5759e 34#ifdef PERL_EXT_RE_BUILD
54df2634 35#include "re_top.h"
b81d288d 36#endif
56953603 37
a687059c 38/*
e50aee73 39 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
40 *
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
43 *
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:
47 *
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
50 * from defects in it.
51 *
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
54 *
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
57 *
58 *
59 **** Alterations to Henry's code are...
60 ****
4bb101f2 61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
1129b882
NC
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
a687059c 64 ****
9ef589d8
LW
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.
67
a687059c
LW
68 *
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.
72 */
73#include "EXTERN.h"
864dbfa3 74#define PERL_IN_REGCOMP_C
a687059c 75#include "perl.h"
d06ea78c 76
acfe0abc 77#ifndef PERL_IN_XSUB_RE
d06ea78c
GS
78# include "INTERN.h"
79#endif
c277df42
IZ
80
81#define REG_COMP_C
54df2634
NC
82#ifdef PERL_IN_XSUB_RE
83# include "re_comp.h"
afda64e8 84EXTERN_C const struct regexp_engine my_reg_engine;
54df2634
NC
85#else
86# include "regcomp.h"
87#endif
a687059c 88
f7e03a10 89#include "dquote_inline.h"
b992490d 90#include "invlist_inline.h"
1b0f46bf 91#include "unicode_constants.h"
04e98a4d 92
538e84ed
KW
93#define HAS_NONLATIN1_FOLD_CLOSURE(i) \
94 _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
f12c0118
KW
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)
26faadbd 97#define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
2c61f163 98#define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
94dc5c2d 99
a687059c
LW
100#ifndef STATIC
101#define STATIC static
102#endif
103
3f910e62
YO
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. */
107struct scan_frame;
108typedef 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 */
3f910e62
YO
113
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 */
117} scan_frame;
118
f4ae5a27
KW
119/* Certain characters are output as a sequence with the first being a
120 * backslash. */
eb7486de 121#define isBACKSLASHED_PUNCT(c) strchr("-[]\\^", c)
f4ae5a27
KW
122
123
09b2b2e6 124struct RExC_state_t {
514a91f1
DM
125 U32 flags; /* RXf_* are we folding, multilining? */
126 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
830247a4 127 char *precomp; /* uncompiled string. */
711b303b 128 char *precomp_end; /* pointer to end of uncompiled string. */
288b8c02 129 REGEXP *rx_sv; /* The SV that is the regexp. */
f8fc2ecf 130 regexp *rx; /* perl core regexp structure */
538e84ed
KW
131 regexp_internal *rxi; /* internal data for regexp object
132 pprivate field */
fac92740 133 char *start; /* Start of input for compile */
830247a4
IZ
134 char *end; /* End of input for compile */
135 char *parse; /* Input-scan pointer. */
51684e3c
KW
136 char *copy_start; /* start of copy of input within
137 constructed parse string */
138 char *copy_start_in_input; /* Position in input string
139 corresponding to copy_start */
ea3daa5d 140 SSize_t whilem_seen; /* number of WHILEM in this expr */
fac92740 141 regnode *emit_start; /* Start of emitted-code area */
f55b7b26 142 regnode_offset emit; /* Code-emit pointer */
830247a4
IZ
143 I32 naughty; /* How bad is this pattern? */
144 I32 sawback; /* Did we see \1, ...? */
145 U32 seen;
88f063b4
KW
146 SSize_t size; /* Number of regnode equivalents in
147 pattern */
7c932d07
KW
148
149 /* position beyond 'precomp' of the warning message furthest away from
150 * 'precomp'. During the parse, no warnings are raised for any problems
151 * earlier in the parse than this position. This works if warnings are
152 * raised the first time a given spot is parsed, and if only one
153 * independent warning is raised for any given spot */
154 Size_t latest_warn_offset;
155
156 I32 npar; /* Capture buffer count so far in the
157 parse, (OPEN) plus one. ("par" 0 is
158 the whole pattern)*/
159 I32 total_par; /* During initial parse, is either 0,
160 or -1; the latter indicating a
161 reparse is needed. After that pass,
162 it is what 'npar' became after the
163 pass. Hence, it being > 0 indicates
164 we are in a reparse situation */
538e84ed
KW
165 I32 nestroot; /* root parens we are in - used by
166 accept */
830247a4 167 I32 seen_zerolen;
f55b7b26
KW
168 regnode_offset *open_parens; /* offsets to open parens */
169 regnode_offset *close_parens; /* offsets to close parens */
d5a00e4a 170 regnode *end_op; /* END node in program */
02daf0ab
YO
171 I32 utf8; /* whether the pattern is utf8 or not */
172 I32 orig_utf8; /* whether the pattern was originally in utf8 */
173 /* XXX use this for future optimisation of case
174 * where pattern must be upgraded to utf8. */
e40e74fe
KW
175 I32 uni_semantics; /* If a d charset modifier should use unicode
176 rules, even if the pattern is not in
177 utf8 */
81714fb9 178 HV *paren_names; /* Paren names */
538e84ed 179
40d049e4 180 regnode **recurse; /* Recurse regops */
88f063b4 181 I32 recurse_count; /* Number of recurse regops we have generated */
4286711a 182 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
538e84ed 183 through */
09a65838 184 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
b57e4118 185 I32 in_lookbehind;
4624b182 186 I32 contains_locale;
bb3f3ed2 187 I32 override_recoding;
b6d67071
KW
188#ifdef EBCDIC
189 I32 recode_x_to_native;
190#endif
9d53c457 191 I32 in_multi_char_class;
1acab4c5 192 struct reg_code_blocks *code_blocks;/* positions of literal (?{})
68e2671b 193 within pattern */
b1603ef8 194 int code_index; /* next code_blocks[] slot */
ee273784 195 SSize_t maxlen; /* mininum possible number of chars in string to match */
3f910e62
YO
196 scan_frame *frame_head;
197 scan_frame *frame_last;
198 U32 frame_count;
7eec73eb 199 AV *warn_text;
dc6d7f5c 200#ifdef ADD_TO_REGEXEC
830247a4
IZ
201 char *starttry; /* -Dr: where regtry was called. */
202#define RExC_starttry (pRExC_state->starttry)
203#endif
d24ca0c5 204 SV *runtime_code_qr; /* qr with the runtime code blocks */
3dab1dad 205#ifdef DEBUGGING
be8e71aa 206 const char *lastparse;
3dab1dad 207 I32 lastnum;
1f1031fe 208 AV *paren_name_list; /* idx -> name */
d9a72fcc 209 U32 study_chunk_recursed_count;
c9f0d54c
YO
210 SV *mysv1;
211 SV *mysv2;
88f063b4 212
3dab1dad
YO
213#define RExC_lastparse (pRExC_state->lastparse)
214#define RExC_lastnum (pRExC_state->lastnum)
1f1031fe 215#define RExC_paren_name_list (pRExC_state->paren_name_list)
d9a72fcc 216#define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
c9f0d54c
YO
217#define RExC_mysv (pRExC_state->mysv1)
218#define RExC_mysv1 (pRExC_state->mysv1)
219#define RExC_mysv2 (pRExC_state->mysv2)
220
3dab1dad 221#endif
512c0f5a 222 bool seen_unfolded_sharp_s;
911bd04e 223 bool strict;
da7cf1cc 224 bool study_started;
034602eb 225 bool in_script_run;
7c932d07 226 bool use_BRANCHJ;
09b2b2e6 227};
830247a4 228
e2509266 229#define RExC_flags (pRExC_state->flags)
514a91f1 230#define RExC_pm_flags (pRExC_state->pm_flags)
830247a4 231#define RExC_precomp (pRExC_state->precomp)
51684e3c
KW
232#define RExC_copy_start_in_input (pRExC_state->copy_start_in_input)
233#define RExC_copy_start_in_constructed (pRExC_state->copy_start)
711b303b 234#define RExC_precomp_end (pRExC_state->precomp_end)
288b8c02 235#define RExC_rx_sv (pRExC_state->rx_sv)
830247a4 236#define RExC_rx (pRExC_state->rx)
f8fc2ecf 237#define RExC_rxi (pRExC_state->rxi)
fac92740 238#define RExC_start (pRExC_state->start)
830247a4
IZ
239#define RExC_end (pRExC_state->end)
240#define RExC_parse (pRExC_state->parse)
7c932d07 241#define RExC_latest_warn_offset (pRExC_state->latest_warn_offset )
830247a4 242#define RExC_whilem_seen (pRExC_state->whilem_seen)
512c0f5a
KW
243
244/* Set during the sizing pass when there is a LATIN SMALL LETTER SHARP S in any
245 * EXACTF node, hence was parsed under /di rules. If later in the parse,
246 * something forces the pattern into using /ui rules, the sharp s should be
247 * folded into the sequence 'ss', which takes up more space than previously
248 * calculated. This means that the sizing pass needs to be restarted. (The
249 * node also becomes an EXACTFU_SS.) For all other characters, an EXACTF node
250 * that gets converted to /ui (and EXACTFU) occupies the same amount of space,
251 * so there is no need to resize [perl #125990]. */
252#define RExC_seen_unfolded_sharp_s (pRExC_state->seen_unfolded_sharp_s)
253
7122b237 254#ifdef RE_TRACK_PATTERN_OFFSETS
44f36cd3 255# define RExC_offsets (RExC_rxi->u.offsets) /* I am not like the
538e84ed 256 others */
7122b237 257#endif
830247a4 258#define RExC_emit (pRExC_state->emit)
fac92740 259#define RExC_emit_start (pRExC_state->emit_start)
830247a4
IZ
260#define RExC_sawback (pRExC_state->sawback)
261#define RExC_seen (pRExC_state->seen)
262#define RExC_size (pRExC_state->size)
ee273784 263#define RExC_maxlen (pRExC_state->maxlen)
830247a4 264#define RExC_npar (pRExC_state->npar)
1533d1d2 265#define RExC_total_parens (pRExC_state->total_par)
e2e6a0f1 266#define RExC_nestroot (pRExC_state->nestroot)
830247a4 267#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
1aa99e6b 268#define RExC_utf8 (pRExC_state->utf8)
e40e74fe 269#define RExC_uni_semantics (pRExC_state->uni_semantics)
02daf0ab 270#define RExC_orig_utf8 (pRExC_state->orig_utf8)
40d049e4
YO
271#define RExC_open_parens (pRExC_state->open_parens)
272#define RExC_close_parens (pRExC_state->close_parens)
5bd2d46e 273#define RExC_end_op (pRExC_state->end_op)
81714fb9 274#define RExC_paren_names (pRExC_state->paren_names)
40d049e4
YO
275#define RExC_recurse (pRExC_state->recurse)
276#define RExC_recurse_count (pRExC_state->recurse_count)
09a65838 277#define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
538e84ed
KW
278#define RExC_study_chunk_recursed_bytes \
279 (pRExC_state->study_chunk_recursed_bytes)
b57e4118 280#define RExC_in_lookbehind (pRExC_state->in_lookbehind)
4624b182 281#define RExC_contains_locale (pRExC_state->contains_locale)
b6d67071
KW
282#ifdef EBCDIC
283# define RExC_recode_x_to_native (pRExC_state->recode_x_to_native)
284#endif
9d53c457 285#define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
3f910e62
YO
286#define RExC_frame_head (pRExC_state->frame_head)
287#define RExC_frame_last (pRExC_state->frame_last)
288#define RExC_frame_count (pRExC_state->frame_count)
67cdf558 289#define RExC_strict (pRExC_state->strict)
da7cf1cc 290#define RExC_study_started (pRExC_state->study_started)
7eec73eb 291#define RExC_warn_text (pRExC_state->warn_text)
034602eb 292#define RExC_in_script_run (pRExC_state->in_script_run)
7c932d07 293#define RExC_use_BRANCHJ (pRExC_state->use_BRANCHJ)
830247a4 294
99807a43
HS
295/* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
296 * a flag to disable back-off on the fixed/floating substrings - if it's
297 * a high complexity pattern we assume the benefit of avoiding a full match
298 * is worth the cost of checking for the substrings even if they rarely help.
299 */
300#define RExC_naughty (pRExC_state->naughty)
301#define TOO_NAUGHTY (10)
302#define MARK_NAUGHTY(add) \
303 if (RExC_naughty < TOO_NAUGHTY) \
304 RExC_naughty += (add)
305#define MARK_NAUGHTY_EXP(exp, add) \
306 if (RExC_naughty < TOO_NAUGHTY) \
307 RExC_naughty += RExC_naughty / (exp) + (add)
cde0cee5 308
a687059c
LW
309#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
310#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
412f55bb 311 ((*s) == '{' && regcurly(s)))
a687059c
LW
312
313/*
314 * Flags to be passed up and down.
315 */
a687059c 316#define WORST 0 /* Worst case. */
a3b492c3 317#define HASWIDTH 0x01 /* Known to match non-null strings. */
fda99bee 318
e64f369d 319/* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
2fd92675
KW
320 * character. (There needs to be a case: in the switch statement in regexec.c
321 * for any node marked SIMPLE.) Note that this is not the same thing as
322 * REGNODE_SIMPLE */
fda99bee 323#define SIMPLE 0x02
e64f369d 324#define SPSTART 0x04 /* Starts with * or + */
8d9c2815
NC
325#define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
326#define TRYAGAIN 0x10 /* Weeded out a declaration. */
c36361b2
KW
327#define RESTART_PARSE 0x20 /* Need to redo the parse */
328#define NEED_UTF8 0x40 /* In conjunction with RESTART_PARSE, need to
329 calcuate sizes as UTF-8 */
a687059c 330
3dab1dad
YO
331#define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
332
07be1b83
YO
333/* whether trie related optimizations are enabled */
334#if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
335#define TRIE_STUDY_OPT
786e8c11 336#define FULL_TRIE_STUDY
07be1b83
YO
337#define TRIE_STCLASS
338#endif
1de06328
YO
339
340
40d049e4
YO
341
342#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
343#define PBITVAL(paren) (1 << ((paren) & 7))
344#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
345#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
346#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
347
82a6ada4 348#define REQUIRE_UTF8(flagp) STMT_START { \
8d9c2815 349 if (!UTF) { \
c36361b2 350 *flagp = RESTART_PARSE|NEED_UTF8; \
f55b7b26 351 return 0; \
8d9c2815 352 } \
82a6ada4 353 } STMT_END
40d049e4 354
512c0f5a
KW
355/* Change from /d into /u rules, and restart the parse if we've already seen
356 * something whose size would increase as a result, by setting *flagp and
357 * returning 'restart_retval'. RExC_uni_semantics is a flag that indicates
142ece99 358 * we've changed to /u during the parse. */
512c0f5a
KW
359#define REQUIRE_UNI_RULES(flagp, restart_retval) \
360 STMT_START { \
361 if (DEPENDS_SEMANTICS) { \
512c0f5a
KW
362 set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); \
363 RExC_uni_semantics = 1; \
c36361b2 364 *flagp |= RESTART_PARSE; \
512c0f5a 365 return restart_retval; \
512c0f5a
KW
366 } \
367 } STMT_END
368
7c932d07
KW
369#define BRANCH_MAX_OFFSET U16_MAX
370#define REQUIRE_BRANCHJ(flagp, restart_retval) \
371 STMT_START { \
372 RExC_use_BRANCHJ = 1; \
373 *flagp |= RESTART_PARSE; \
374 return restart_retval; \
375 } STMT_END
376
377#define REQUIRE_PARENS_PASS \
378 STMT_START { \
379 if (RExC_total_parens == 0) RExC_total_parens = -1; \
380 } STMT_END
381
fd7fc6ae 382/* Executes a return statement with the value 'X', if 'flags' contains any of
c36361b2 383 * 'RESTART_PARSE', 'NEED_UTF8', or 'extra'. If so, *flagp is set to those
fd7fc6ae
YO
384 * flags */
385#define RETURN_X_ON_RESTART_OR_FLAGS(X, flags, flagp, extra) \
386 STMT_START { \
c36361b2
KW
387 if ((flags) & (RESTART_PARSE|NEED_UTF8|(extra))) { \
388 *(flagp) = (flags) & (RESTART_PARSE|NEED_UTF8|(extra)); \
fd7fc6ae
YO
389 return X; \
390 } \
391 } STMT_END
392
5d555bb8 393#define RETURN_FAIL_ON_RESTART_OR_FLAGS(flags,flagp,extra) \
f55b7b26 394 RETURN_X_ON_RESTART_OR_FLAGS(0,flags,flagp,extra)
fd7fc6ae
YO
395
396#define RETURN_X_ON_RESTART(X, flags,flagp) \
397 RETURN_X_ON_RESTART_OR_FLAGS( X, flags, flagp, 0)
398
399
5d555bb8 400#define RETURN_FAIL_ON_RESTART_FLAGP_OR_FLAGS(flagp,extra) \
c36361b2 401 if (*(flagp) & (RESTART_PARSE|(extra))) return 0
fd7fc6ae 402
c36361b2 403#define MUST_RESTART(flags) ((flags) & (RESTART_PARSE))
fd7fc6ae 404
5d555bb8 405#define RETURN_FAIL_ON_RESTART(flags,flagp) \
f55b7b26 406 RETURN_X_ON_RESTART(0, flags,flagp)
5d555bb8 407#define RETURN_FAIL_ON_RESTART_FLAGP(flagp) \
88f063b4 408 RETURN_FAIL_ON_RESTART_FLAGP_OR_FLAGS(flagp, 0)
fd7fc6ae 409
f19b1a63
KW
410/* This converts the named class defined in regcomp.h to its equivalent class
411 * number defined in handy.h. */
412#define namedclass_to_classnum(class) ((int) ((class) / 2))
413#define classnum_to_namedclass(classnum) ((classnum) * 2)
414
de92f5e6
KW
415#define _invlist_union_complement_2nd(a, b, output) \
416 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
417#define _invlist_intersection_complement_2nd(a, b, output) \
418 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
419
1de06328
YO
420/* About scan_data_t.
421
422 During optimisation we recurse through the regexp program performing
423 various inplace (keyhole style) optimisations. In addition study_chunk
424 and scan_commit populate this data structure with information about
538e84ed 425 what strings MUST appear in the pattern. We look for the longest
3b753521 426 string that must appear at a fixed location, and we look for the
1de06328
YO
427 longest string that may appear at a floating location. So for instance
428 in the pattern:
538e84ed 429
1de06328 430 /FOO[xX]A.*B[xX]BAR/
538e84ed 431
1de06328
YO
432 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
433 strings (because they follow a .* construct). study_chunk will identify
434 both FOO and BAR as being the longest fixed and floating strings respectively.
538e84ed 435
1de06328 436 The strings can be composites, for instance
538e84ed 437
1de06328 438 /(f)(o)(o)/
538e84ed 439
1de06328 440 will result in a composite fixed substring 'foo'.
538e84ed 441
1de06328 442 For each string some basic information is maintained:
538e84ed 443
6ea77169 444 - min_offset
1de06328
YO
445 This is the position the string must appear at, or not before.
446 It also implicitly (when combined with minlenp) tells us how many
3b753521
FN
447 characters must match before the string we are searching for.
448 Likewise when combined with minlenp and the length of the string it
538e84ed 449 tells us how many characters must appear after the string we have
1de06328 450 found.
538e84ed 451
1de06328
YO
452 - max_offset
453 Only used for floating strings. This is the rightmost point that
ea3daa5d 454 the string can appear at. If set to SSize_t_MAX it indicates that the
1de06328 455 string can occur infinitely far to the right.
6e12f832 456 For fixed strings, it is equal to min_offset.
538e84ed 457
1de06328 458 - minlenp
2d608413
KW
459 A pointer to the minimum number of characters of the pattern that the
460 string was found inside. This is important as in the case of positive
538e84ed 461 lookahead or positive lookbehind we can have multiple patterns
1de06328 462 involved. Consider
538e84ed 463
1de06328 464 /(?=FOO).*F/
538e84ed 465
1de06328
YO
466 The minimum length of the pattern overall is 3, the minimum length
467 of the lookahead part is 3, but the minimum length of the part that
538e84ed 468 will actually match is 1. So 'FOO's minimum length is 3, but the
1de06328 469 minimum length for the F is 1. This is important as the minimum length
538e84ed 470 is used to determine offsets in front of and behind the string being
1de06328 471 looked for. Since strings can be composites this is the length of the
486ec47a 472 pattern at the time it was committed with a scan_commit. Note that
1de06328 473 the length is calculated by study_chunk, so that the minimum lengths
538e84ed 474 are not known until the full pattern has been compiled, thus the
1de06328 475 pointer to the value.
538e84ed 476
1de06328 477 - lookbehind
538e84ed 478
1de06328 479 In the case of lookbehind the string being searched for can be
538e84ed 480 offset past the start point of the final matching string.
1de06328
YO
481 If this value was just blithely removed from the min_offset it would
482 invalidate some of the calculations for how many chars must match
483 before or after (as they are derived from min_offset and minlen and
538e84ed 484 the length of the string being searched for).
1de06328
YO
485 When the final pattern is compiled and the data is moved from the
486 scan_data_t structure into the regexp structure the information
538e84ed
KW
487 about lookbehind is factored in, with the information that would
488 have been lost precalculated in the end_shift field for the
1de06328
YO
489 associated string.
490
491 The fields pos_min and pos_delta are used to store the minimum offset
538e84ed 492 and the delta to the maximum offset at the current point in the pattern.
1de06328
YO
493
494*/
2c2d71f5 495
cb6aef8b
DM
496struct scan_data_substrs {
497 SV *str; /* longest substring found in pattern */
498 SSize_t min_offset; /* earliest point in string it can appear */
499 SSize_t max_offset; /* latest point in string it can appear */
500 SSize_t *minlenp; /* pointer to the minlen relevant to the string */
501 SSize_t lookbehind; /* is the pos of the string modified by LB */
11683ecb 502 I32 flags; /* per substring SF_* and SCF_* flags */
cb6aef8b
DM
503};
504
2c2d71f5 505typedef struct scan_data_t {
1de06328
YO
506 /*I32 len_min; unused */
507 /*I32 len_delta; unused */
49f55535 508 SSize_t pos_min;
ea3daa5d 509 SSize_t pos_delta;
2c2d71f5 510 SV *last_found;
ea3daa5d 511 SSize_t last_end; /* min value, <0 unless valid. */
49f55535 512 SSize_t last_start_min;
ea3daa5d 513 SSize_t last_start_max;
2099df82
DM
514 U8 cur_is_floating; /* whether the last_* values should be set as
515 * the next fixed (0) or floating (1)
516 * substring */
6ea77169 517
2099df82
DM
518 /* [0] is longest fixed substring so far, [1] is longest float so far */
519 struct scan_data_substrs substrs[2];
6ea77169 520
11683ecb 521 I32 flags; /* common SF_* and SCF_* flags */
2c2d71f5 522 I32 whilem_c;
49f55535 523 SSize_t *last_closep;
b8f7bb16 524 regnode_ssc *start_class;
2c2d71f5
JH
525} scan_data_t;
526
a687059c 527/*
e50aee73 528 * Forward declarations for pregcomp()'s friends.
a687059c 529 */
a0d0e21e 530
6ea77169 531static const scan_data_t zero_scan_data = {
a2ca2017 532 0, 0, NULL, 0, 0, 0, 0,
6ea77169 533 {
11683ecb
DM
534 { NULL, 0, 0, 0, 0, 0 },
535 { NULL, 0, 0, 0, 0, 0 },
6ea77169
DM
536 },
537 0, 0, NULL, NULL
538};
c277df42 539
11683ecb
DM
540/* study flags */
541
07be1b83
YO
542#define SF_BEFORE_SEOL 0x0001
543#define SF_BEFORE_MEOL 0x0002
11683ecb 544#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
c277df42 545
07be1b83
YO
546#define SF_IS_INF 0x0040
547#define SF_HAS_PAR 0x0080
548#define SF_IN_PAR 0x0100
549#define SF_HAS_EVAL 0x0200
bc604ad8
KW
550
551
552/* SCF_DO_SUBSTR is the flag that tells the regexp analyzer to track the
553 * longest substring in the pattern. When it is not set the optimiser keeps
554 * track of position, but does not keep track of the actual strings seen,
555 *
556 * So for instance /foo/ will be parsed with SCF_DO_SUBSTR being true, but
557 * /foo/i will not.
558 *
559 * Similarly, /foo.*(blah|erm|huh).*fnorble/ will have "foo" and "fnorble"
560 * parsed with SCF_DO_SUBSTR on, but while processing the (...) it will be
561 * turned off because of the alternation (BRANCH). */
07be1b83 562#define SCF_DO_SUBSTR 0x0400
bc604ad8 563
653099ff
GS
564#define SCF_DO_STCLASS_AND 0x0800
565#define SCF_DO_STCLASS_OR 0x1000
566#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
e1901655 567#define SCF_WHILEM_VISITED_POS 0x2000
c277df42 568
786e8c11 569#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
538e84ed 570#define SCF_SEEN_ACCEPT 0x8000
688e0391 571#define SCF_TRIE_DOING_RESTUDY 0x10000
4286711a
YO
572#define SCF_IN_DEFINE 0x20000
573
574
575
07be1b83 576
43fead97 577#define UTF cBOOL(RExC_utf8)
00b27cfc
KW
578
579/* The enums for all these are ordered so things work out correctly */
a62b1201 580#define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
538e84ed
KW
581#define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
582 == REGEX_DEPENDS_CHARSET)
00b27cfc 583#define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
538e84ed
KW
584#define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
585 >= REGEX_UNICODE_CHARSET)
586#define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
587 == REGEX_ASCII_RESTRICTED_CHARSET)
588#define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
589 >= REGEX_ASCII_RESTRICTED_CHARSET)
590#define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
591 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
a62b1201 592
43fead97 593#define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
a0ed51b3 594
f2c2a6ab
KW
595/* For programs that want to be strictly Unicode compatible by dying if any
596 * attempt is made to match a non-Unicode code point against a Unicode
597 * property. */
598#define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
599
93733859 600#define OOB_NAMEDCLASS -1
b8c5462f 601
8e661ac5
KW
602/* There is no code point that is out-of-bounds, so this is problematic. But
603 * its only current use is to initialize a variable that is always set before
604 * looked at. */
605#define OOB_UNICODE 0xDEADBEEF
606
a0ed51b3 607#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
a0ed51b3 608
8615cb43 609
b45f050a
JF
610/* length of regex to show in messages that don't mark a position within */
611#define RegexLengthToShowInErrorMessages 127
612
613/*
614 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
615 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
616 * op/pragma/warn/regcomp.
617 */
7253e4e3
RK
618#define MARKER1 "<-- HERE" /* marker as it appears in the description */
619#define MARKER2 " <-- HERE " /* marker as it appears within the regex */
b81d288d 620
538e84ed 621#define REPORT_LOCATION " in regex; marked by " MARKER1 \
147e3846 622 " in m/%" UTF8f MARKER2 "%" UTF8f "/"
b45f050a 623
285b5ca0
KW
624/* The code in this file in places uses one level of recursion with parsing
625 * rebased to an alternate string constructed by us in memory. This can take
626 * the form of something that is completely different from the input, or
627 * something that uses the input as part of the alternate. In the first case,
628 * there should be no possibility of an error, as we are in complete control of
51684e3c
KW
629 * the alternate string. But in the second case we don't completely control
630 * the input portion, so there may be errors in that. Here's an example:
285b5ca0
KW
631 * /[abc\x{DF}def]/ui
632 * is handled specially because \x{df} folds to a sequence of more than one
51684e3c 633 * character: 'ss'. What is done is to create and parse an alternate string,
285b5ca0
KW
634 * which looks like this:
635 * /(?:\x{DF}|[abc\x{DF}def])/ui
636 * where it uses the input unchanged in the middle of something it constructs,
637 * which is a branch for the DF outside the character class, and clustering
638 * parens around the whole thing. (It knows enough to skip the DF inside the
639 * class while in this substitute parse.) 'abc' and 'def' may have errors that
640 * need to be reported. The general situation looks like this:
641 *
51684e3c 642 * |<------- identical ------>|
285b5ca0 643 * sI tI xI eI
51684e3c 644 * Input: ---------------------------------------------------------------
285b5ca0
KW
645 * Constructed: ---------------------------------------------------
646 * sC tC xC eC EC
51684e3c 647 * |<------- identical ------>|
285b5ca0 648 *
51684e3c
KW
649 * sI..eI is the portion of the input pattern we are concerned with here.
650 * sC..EC is the constructed substitute parse string.
651 * sC..tC is constructed by us
652 * tC..eC is an exact duplicate of the portion of the input pattern tI..eI.
653 * In the diagram, these are vertically aligned.
654 * eC..EC is also constructed by us.
655 * xC is the position in the substitute parse string where we found a
656 * problem.
657 * xI is the position in the original pattern corresponding to xC.
285b5ca0 658 *
51684e3c
KW
659 * We want to display a message showing the real input string. Thus we need to
660 * translate from xC to xI. We know that xC >= tC, since the portion of the
661 * string sC..tC has been constructed by us, and so shouldn't have errors. We
662 * get:
663 * xI = tI + (xC - tC)
285b5ca0 664 *
51684e3c
KW
665 * When the substitute parse is constructed, the code needs to set:
666 * RExC_start (sC)
667 * RExC_end (eC)
668 * RExC_copy_start_in_input (tI)
669 * RExC_copy_start_in_constructed (tC)
670 * and restore them when done.
285b5ca0 671 *
51684e3c
KW
672 * During normal processing of the input pattern, both
673 * 'RExC_copy_start_in_input' and 'RExC_copy_start_in_constructed' are set to
674 * sI, so that xC equals xI.
285b5ca0
KW
675 */
676
51684e3c
KW
677#define sI RExC_precomp
678#define eI RExC_precomp_end
679#define sC RExC_start
680#define eC RExC_end
681#define tI RExC_copy_start_in_input
682#define tC RExC_copy_start_in_constructed
683#define xI(xC) (tI + (xC - tC))
684#define xI_offset(xC) (xI(xC) - sI)
285b5ca0
KW
685
686#define REPORT_LOCATION_ARGS(xC) \
687 UTF8fARG(UTF, \
51684e3c 688 (xI(xC) > eI) /* Don't run off end */ \
232b691f 689 ? eI - sI /* Length before the <--HERE */ \
bc929b61
KW
690 : ((xI_offset(xC) >= 0) \
691 ? xI_offset(xC) \
692 : (Perl_croak(aTHX_ "panic: %s: %d: negative offset: %" \
693 IVdf " trying to output message for " \
694 " pattern %.*s", \
51684e3c 695 __FILE__, __LINE__, (IV) xI_offset(xC), \
bc929b61 696 ((int) (eC - sC)), sC), 0)), \
51684e3c 697 sI), /* The input pattern printed up to the <--HERE */ \
285b5ca0 698 UTF8fARG(UTF, \
51684e3c
KW
699 (xI(xC) > eI) ? 0 : eI - xI(xC), /* Length after <--HERE */ \
700 (xI(xC) > eI) ? eI : xI(xC)) /* pattern after <--HERE */
c1d900c3 701
8a6d8ec6
HS
702/* Used to point after bad bytes for an error message, but avoid skipping
703 * past a nul byte. */
704#define SKIP_IF_CHAR(s) (!*(s) ? 0 : UTF ? UTF8SKIP(s) : 1)
705
ed68fe11
KW
706/* Set up to clean up after our imminent demise */
707#define PREPARE_TO_DIE \
708 STMT_START { \
709 if (RExC_rx_sv) \
710 SAVEFREESV(RExC_rx_sv); \
7c932d07
KW
711 if (RExC_open_parens) \
712 SAVEFREEPV(RExC_open_parens); \
713 if (RExC_close_parens) \
714 SAVEFREEPV(RExC_close_parens); \
ed68fe11
KW
715 } STMT_END
716
b45f050a
JF
717/*
718 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
719 * arg. Show regex, up to a maximum length. If it's too long, chop and add
720 * "...".
721 */
58e23c8d 722#define _FAIL(code) STMT_START { \
bfed75c6 723 const char *ellipses = ""; \
88f063b4 724 IV len = RExC_precomp_end - RExC_precomp; \
ccb2c380 725 \
ed68fe11 726 PREPARE_TO_DIE; \
ccb2c380
MP
727 if (len > RegexLengthToShowInErrorMessages) { \
728 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
729 len = RegexLengthToShowInErrorMessages - 10; \
730 ellipses = "..."; \
731 } \
58e23c8d 732 code; \
ccb2c380 733} STMT_END
8615cb43 734
58e23c8d 735#define FAIL(msg) _FAIL( \
147e3846 736 Perl_croak(aTHX_ "%s in regex m/%" UTF8f "%s/", \
c1d900c3 737 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d
YO
738
739#define FAIL2(msg,arg) _FAIL( \
147e3846 740 Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \
c1d900c3 741 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d 742
b45f050a 743/*
b45f050a
JF
744 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
745 */
ccb2c380 746#define Simple_vFAIL(m) STMT_START { \
ccb2c380 747 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
d528642a 748 m, REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 749} STMT_END
b45f050a
JF
750
751/*
752 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
753 */
ccb2c380 754#define vFAIL(m) STMT_START { \
ed68fe11 755 PREPARE_TO_DIE; \
ccb2c380
MP
756 Simple_vFAIL(m); \
757} STMT_END
b45f050a
JF
758
759/*
760 * Like Simple_vFAIL(), but accepts two arguments.
761 */
ccb2c380 762#define Simple_vFAIL2(m,a1) STMT_START { \
d528642a
KW
763 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
764 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 765} STMT_END
b45f050a
JF
766
767/*
768 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
769 */
ccb2c380 770#define vFAIL2(m,a1) STMT_START { \
ed68fe11 771 PREPARE_TO_DIE; \
ccb2c380
MP
772 Simple_vFAIL2(m, a1); \
773} STMT_END
b45f050a
JF
774
775
776/*
777 * Like Simple_vFAIL(), but accepts three arguments.
778 */
ccb2c380 779#define Simple_vFAIL3(m, a1, a2) STMT_START { \
c1d900c3 780 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 781 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 782} STMT_END
b45f050a
JF
783
784/*
785 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
786 */
ccb2c380 787#define vFAIL3(m,a1,a2) STMT_START { \
ed68fe11 788 PREPARE_TO_DIE; \
ccb2c380
MP
789 Simple_vFAIL3(m, a1, a2); \
790} STMT_END
b45f050a
JF
791
792/*
793 * Like Simple_vFAIL(), but accepts four arguments.
794 */
ccb2c380 795#define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
d528642a
KW
796 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
797 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 798} STMT_END
b45f050a 799
95db3ffa 800#define vFAIL4(m,a1,a2,a3) STMT_START { \
ed68fe11 801 PREPARE_TO_DIE; \
95db3ffa
KW
802 Simple_vFAIL4(m, a1, a2, a3); \
803} STMT_END
804
946095af 805/* A specialized version of vFAIL2 that works with UTF8f */
d528642a 806#define vFAIL2utf8f(m, a1) STMT_START { \
ed68fe11 807 PREPARE_TO_DIE; \
d528642a
KW
808 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
809 REPORT_LOCATION_ARGS(RExC_parse)); \
946095af
BF
810} STMT_END
811
3ba22297 812#define vFAIL3utf8f(m, a1, a2) STMT_START { \
ed68fe11 813 PREPARE_TO_DIE; \
3ba22297 814 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 815 REPORT_LOCATION_ARGS(RExC_parse)); \
3ba22297
KW
816} STMT_END
817
d62e749e
KW
818/* Setting this to NULL is a signal to not output warnings */
819#define TURN_OFF_WARNINGS_IN_SUBSTITUTE_PARSE RExC_copy_start_in_constructed = NULL
820#define RESTORE_WARNINGS RExC_copy_start_in_constructed = RExC_precomp
821
7c932d07
KW
822/* Since a warning can be generated multiple times as the input is reparsed, we
823 * output it the first time we come to that point in the parse, but suppress it
824 * otherwise. 'RExC_copy_start_in_constructed' being NULL is a flag to not
825 * generate any warnings */
55bcf857 826#define TO_OUTPUT_WARNINGS(loc) \
7c932d07
KW
827 ( RExC_copy_start_in_constructed \
828 && ((xI(loc)) - RExC_precomp) > (Ptrdiff_t) RExC_latest_warn_offset)
55bcf857 829
7c932d07
KW
830/* After we've emitted a warning, we save the position in the input so we don't
831 * output it again */
832#define UPDATE_WARNINGS_LOC(loc) \
833 STMT_START { \
834 if (TO_OUTPUT_WARNINGS(loc)) { \
835 RExC_latest_warn_offset = (xI(loc)) - RExC_precomp; \
836 } \
837 } STMT_END
77ccba3c 838
086f21bf 839/* 'warns' is the output of the packWARNx macro used in 'code' */
f8220fe2
KW
840#define _WARN_HELPER(loc, warns, code) \
841 STMT_START { \
d62e749e
KW
842 if (! RExC_copy_start_in_constructed) { \
843 Perl_croak( aTHX_ "panic! %s: %d: Tried to warn when none" \
844 " expected at '%s'", \
845 __FILE__, __LINE__, loc); \
846 } \
55bcf857 847 if (TO_OUTPUT_WARNINGS(loc)) { \
086f21bf
KW
848 if (ckDEAD(warns)) \
849 PREPARE_TO_DIE; \
338190a1 850 code; \
77ccba3c 851 UPDATE_WARNINGS_LOC(loc); \
338190a1 852 } \
f8220fe2 853 } STMT_END
946095af 854
5e0a247b 855/* m is not necessarily a "literal string", in this macro */
f8220fe2
KW
856#define reg_warn_non_literal_string(loc, m) \
857 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
858 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 859 "%s" REPORT_LOCATION, \
f8220fe2 860 m, REPORT_LOCATION_ARGS(loc)))
5e0a247b 861
f8220fe2
KW
862#define ckWARNreg(loc,m) \
863 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
864 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 865 m REPORT_LOCATION, \
f8220fe2 866 REPORT_LOCATION_ARGS(loc)))
ccb2c380 867
f8220fe2
KW
868#define vWARN(loc, m) \
869 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
870 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 871 m REPORT_LOCATION, \
f8220fe2 872 REPORT_LOCATION_ARGS(loc))) \
b927b7e9 873
f8220fe2
KW
874#define vWARN_dep(loc, m) \
875 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
876 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 877 m REPORT_LOCATION, \
f8220fe2 878 REPORT_LOCATION_ARGS(loc)))
0d6106aa 879
f8220fe2
KW
880#define ckWARNdep(loc,m) \
881 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
882 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 883 m REPORT_LOCATION, \
f8220fe2 884 REPORT_LOCATION_ARGS(loc)))
147508a2 885
f8220fe2
KW
886#define ckWARNregdep(loc,m) \
887 _WARN_HELPER(loc, packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
888 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, \
d528642a
KW
889 WARN_REGEXP), \
890 m REPORT_LOCATION, \
f8220fe2 891 REPORT_LOCATION_ARGS(loc)))
ccb2c380 892
f8220fe2
KW
893#define ckWARN2reg_d(loc,m, a1) \
894 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
895 Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
d528642a 896 m REPORT_LOCATION, \
f8220fe2 897 a1, REPORT_LOCATION_ARGS(loc)))
2335b3d3 898
f8220fe2
KW
899#define ckWARN2reg(loc, m, a1) \
900 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
901 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 902 m REPORT_LOCATION, \
f8220fe2 903 a1, REPORT_LOCATION_ARGS(loc)))
ccb2c380 904
f8220fe2
KW
905#define vWARN3(loc, m, a1, a2) \
906 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
907 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 908 m REPORT_LOCATION, \
f8220fe2 909 a1, a2, REPORT_LOCATION_ARGS(loc)))
ccb2c380 910
f8220fe2
KW
911#define ckWARN3reg(loc, m, a1, a2) \
912 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
913 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
914 m REPORT_LOCATION, \
915 a1, a2, \
f8220fe2 916 REPORT_LOCATION_ARGS(loc)))
668c081a 917
f8220fe2
KW
918#define vWARN4(loc, m, a1, a2, a3) \
919 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
920 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
921 m REPORT_LOCATION, \
922 a1, a2, a3, \
f8220fe2 923 REPORT_LOCATION_ARGS(loc)))
ccb2c380 924
f8220fe2
KW
925#define ckWARN4reg(loc, m, a1, a2, a3) \
926 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
927 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
928 m REPORT_LOCATION, \
929 a1, a2, a3, \
f8220fe2 930 REPORT_LOCATION_ARGS(loc)))
668c081a 931
f8220fe2
KW
932#define vWARN5(loc, m, a1, a2, a3, a4) \
933 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
934 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
935 m REPORT_LOCATION, \
936 a1, a2, a3, a4, \
f8220fe2 937 REPORT_LOCATION_ARGS(loc)))
9d1d55b5 938
aea5f181
KW
939#define ckWARNexperimental(loc, class, m) \
940 _WARN_HELPER(loc, packWARN(class), \
941 Perl_ck_warner_d(aTHX_ packWARN(class), \
942 m REPORT_LOCATION, \
943 REPORT_LOCATION_ARGS(loc)))
944
1a322bb4
KW
945/* Convert between a pointer to a node and its offset from the beginning of the
946 * program */
947#define REGNODE_p(offset) (RExC_emit_start + (offset))
948#define REGNODE_OFFSET(node) ((node) - RExC_emit_start)
949
538e84ed 950/* Macros for recording node offsets. 20001227 mjd@plover.com
fac92740
MJD
951 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
952 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
953 * Element 0 holds the number n.
07be1b83 954 * Position is 1 indexed.
fac92740 955 */
7122b237 956#ifndef RE_TRACK_PATTERN_OFFSETS
d7c442ce 957#define Set_Node_Offset_To_R(offset,byte)
7122b237
YO
958#define Set_Node_Offset(node,byte)
959#define Set_Cur_Node_Offset
960#define Set_Node_Length_To_R(node,len)
961#define Set_Node_Length(node,len)
6a86c6ad 962#define Set_Node_Cur_Length(node,start)
538e84ed
KW
963#define Node_Offset(n)
964#define Node_Length(n)
7122b237
YO
965#define Set_Node_Offset_Length(node,offset,len)
966#define ProgLen(ri) ri->u.proglen
967#define SetProgLen(ri,x) ri->u.proglen = x
4cbeff97 968#define Track_Code(code)
7122b237
YO
969#else
970#define ProgLen(ri) ri->u.offsets[0]
971#define SetProgLen(ri,x) ri->u.offsets[0] = x
d7c442ce 972#define Set_Node_Offset_To_R(offset,byte) STMT_START { \
ccb2c380 973 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
d7c442ce
KW
974 __LINE__, (int)(offset), (int)(byte))); \
975 if((offset) < 0) { \
538e84ed 976 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
d7c442ce 977 (int)(offset)); \
ccb2c380 978 } else { \
d7c442ce 979 RExC_offsets[2*(offset)-1] = (byte); \
ccb2c380 980 } \
ccb2c380
MP
981} STMT_END
982
88f063b4 983#define Set_Node_Offset(node,byte) \
1a322bb4 984 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (byte)-RExC_start)
ccb2c380
MP
985#define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
986
987#define Set_Node_Length_To_R(node,len) STMT_START { \
ccb2c380 988 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
551405c4 989 __LINE__, (int)(node), (int)(len))); \
ccb2c380 990 if((node) < 0) { \
538e84ed
KW
991 Perl_croak(aTHX_ "value of node is %d in Length macro", \
992 (int)(node)); \
ccb2c380
MP
993 } else { \
994 RExC_offsets[2*(node)] = (len); \
995 } \
ccb2c380
MP
996} STMT_END
997
998#define Set_Node_Length(node,len) \
1a322bb4 999 Set_Node_Length_To_R(REGNODE_OFFSET(node), len)
6a86c6ad
NC
1000#define Set_Node_Cur_Length(node, start) \
1001 Set_Node_Length(node, RExC_parse - start)
fac92740
MJD
1002
1003/* Get offsets and lengths */
1a322bb4
KW
1004#define Node_Offset(n) (RExC_offsets[2*(REGNODE_OFFSET(n))-1])
1005#define Node_Length(n) (RExC_offsets[2*(REGNODE_OFFSET(n))])
fac92740 1006
07be1b83 1007#define Set_Node_Offset_Length(node,offset,len) STMT_START { \
1a322bb4
KW
1008 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (offset)); \
1009 Set_Node_Length_To_R(REGNODE_OFFSET(node), (len)); \
07be1b83 1010} STMT_END
4cbeff97
TC
1011
1012#define Track_Code(code) STMT_START { code } STMT_END
7122b237 1013#endif
07be1b83
YO
1014
1015#if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
1016#define EXPERIMENTAL_INPLACESCAN
f427392e 1017#endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
07be1b83 1018
cb41e5d6
YO
1019#ifdef DEBUGGING
1020int
6ad9a8ab 1021Perl_re_printf(pTHX_ const char *fmt, ...)
cb41e5d6
YO
1022{
1023 va_list ap;
1024 int result;
1025 PerlIO *f= Perl_debug_log;
1026 PERL_ARGS_ASSERT_RE_PRINTF;
1027 va_start(ap, fmt);
1028 result = PerlIO_vprintf(f, fmt, ap);
1029 va_end(ap);
1030 return result;
1031}
1032
1033int
7b031478 1034Perl_re_indentf(pTHX_ const char *fmt, U32 depth, ...)
cb41e5d6
YO
1035{
1036 va_list ap;
1037 int result;
1038 PerlIO *f= Perl_debug_log;
1039 PERL_ARGS_ASSERT_RE_INDENTF;
1040 va_start(ap, depth);
daeb874b 1041 PerlIO_printf(f, "%*s", ( (int)depth % 20 ) * 2, "");
cb41e5d6
YO
1042 result = PerlIO_vprintf(f, fmt, ap);
1043 va_end(ap);
1044 return result;
1045}
1046#endif /* DEBUGGING */
1047
7b031478 1048#define DEBUG_RExC_seen() \
538e84ed 1049 DEBUG_OPTIMISE_MORE_r({ \
88f063b4 1050 Perl_re_printf( aTHX_ "RExC_seen: "); \
538e84ed 1051 \
e384d5c1 1052 if (RExC_seen & REG_ZERO_LEN_SEEN) \
88f063b4 1053 Perl_re_printf( aTHX_ "REG_ZERO_LEN_SEEN "); \
538e84ed 1054 \
e384d5c1 1055 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
88f063b4 1056 Perl_re_printf( aTHX_ "REG_LOOKBEHIND_SEEN "); \
538e84ed 1057 \
e384d5c1 1058 if (RExC_seen & REG_GPOS_SEEN) \
88f063b4 1059 Perl_re_printf( aTHX_ "REG_GPOS_SEEN "); \
538e84ed 1060 \
e384d5c1 1061 if (RExC_seen & REG_RECURSE_SEEN) \
88f063b4 1062 Perl_re_printf( aTHX_ "REG_RECURSE_SEEN "); \
538e84ed 1063 \
7b031478 1064 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
88f063b4 1065 Perl_re_printf( aTHX_ "REG_TOP_LEVEL_BRANCHES_SEEN "); \
538e84ed 1066 \
e384d5c1 1067 if (RExC_seen & REG_VERBARG_SEEN) \
88f063b4 1068 Perl_re_printf( aTHX_ "REG_VERBARG_SEEN "); \
538e84ed 1069 \
e384d5c1 1070 if (RExC_seen & REG_CUTGROUP_SEEN) \
88f063b4 1071 Perl_re_printf( aTHX_ "REG_CUTGROUP_SEEN "); \
538e84ed 1072 \
e384d5c1 1073 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
88f063b4 1074 Perl_re_printf( aTHX_ "REG_RUN_ON_COMMENT_SEEN "); \
538e84ed 1075 \
e384d5c1 1076 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
88f063b4 1077 Perl_re_printf( aTHX_ "REG_UNFOLDED_MULTI_SEEN "); \
538e84ed 1078 \
7b031478 1079 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
88f063b4 1080 Perl_re_printf( aTHX_ "REG_UNBOUNDED_QUANTIFIER_SEEN "); \
ee273784 1081 \
88f063b4 1082 Perl_re_printf( aTHX_ "\n"); \
9e9ecfdf
YO
1083 });
1084
fdfb4f21 1085#define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
6ad9a8ab 1086 if ((flags) & flag) Perl_re_printf( aTHX_ "%s ", #flag)
fdfb4f21 1087
fdfb4f21 1088
f5a36d78
DM
1089#ifdef DEBUGGING
1090static void
1091S_debug_show_study_flags(pTHX_ U32 flags, const char *open_str,
1092 const char *close_str)
1093{
1094 if (!flags)
1095 return;
1096
1097 Perl_re_printf( aTHX_ "%s", open_str);
11683ecb
DM
1098 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_SEOL);
1099 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_MEOL);
f5a36d78
DM
1100 DEBUG_SHOW_STUDY_FLAG(flags, SF_IS_INF);
1101 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_PAR);
1102 DEBUG_SHOW_STUDY_FLAG(flags, SF_IN_PAR);
1103 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_EVAL);
1104 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_SUBSTR);
1105 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_AND);
1106 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_OR);
1107 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS);
1108 DEBUG_SHOW_STUDY_FLAG(flags, SCF_WHILEM_VISITED_POS);
1109 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_RESTUDY);
1110 DEBUG_SHOW_STUDY_FLAG(flags, SCF_SEEN_ACCEPT);
1111 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_DOING_RESTUDY);
1112 DEBUG_SHOW_STUDY_FLAG(flags, SCF_IN_DEFINE);
1113 Perl_re_printf( aTHX_ "%s", close_str);
1114}
1115
1116
1117static void
1118S_debug_studydata(pTHX_ const char *where, scan_data_t *data,
1119 U32 depth, int is_inf)
1120{
1121 GET_RE_DEBUG_FLAGS_DECL;
1122
1123 DEBUG_OPTIMISE_MORE_r({
1124 if (!data)
1125 return;
1126 Perl_re_indentf(aTHX_ "%s: Pos:%" IVdf "/%" IVdf " Flags: 0x%" UVXf,
1127 depth,
1128 where,
1129 (IV)data->pos_min,
1130 (IV)data->pos_delta,
1131 (UV)data->flags
1132 );
1133
1134 S_debug_show_study_flags(aTHX_ data->flags," [","]");
fdfb4f21 1135
f5a36d78
DM
1136 Perl_re_printf( aTHX_
1137 " Whilem_c: %" IVdf " Lcp: %" IVdf " %s",
1138 (IV)data->whilem_c,
1139 (IV)(data->last_closep ? *((data)->last_closep) : -1),
1140 is_inf ? "INF " : ""
1141 );
1142
11683ecb 1143 if (data->last_found) {
37b6262f 1144 int i;
11683ecb
DM
1145 Perl_re_printf(aTHX_
1146 "Last:'%s' %" IVdf ":%" IVdf "/%" IVdf,
1147 SvPVX_const(data->last_found),
1148 (IV)data->last_end,
1149 (IV)data->last_start_min,
1150 (IV)data->last_start_max
1151 );
1152
2099df82 1153 for (i = 0; i < 2; i++) {
37b6262f
DM
1154 Perl_re_printf(aTHX_
1155 " %s%s: '%s' @ %" IVdf "/%" IVdf,
2099df82
DM
1156 data->cur_is_floating == i ? "*" : "",
1157 i ? "Float" : "Fixed",
37b6262f
DM
1158 SvPVX_const(data->substrs[i].str),
1159 (IV)data->substrs[i].min_offset,
1160 (IV)data->substrs[i].max_offset
1161 );
1162 S_debug_show_study_flags(aTHX_ data->substrs[i].flags," [","]");
1163 }
11683ecb 1164 }
f5a36d78
DM
1165
1166 Perl_re_printf( aTHX_ "\n");
1167 });
1168}
1169
1170
1171static void
1172S_debug_peep(pTHX_ const char *str, const RExC_state_t *pRExC_state,
1173 regnode *scan, U32 depth, U32 flags)
1174{
1175 GET_RE_DEBUG_FLAGS_DECL;
1176
1177 DEBUG_OPTIMISE_r({
1178 regnode *Next;
1179
1180 if (!scan)
1181 return;
1182 Next = regnext(scan);
1183 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
1184 Perl_re_indentf( aTHX_ "%s>%3d: %s (%d)",
1185 depth,
1186 str,
1187 REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),
1188 Next ? (REG_NODE_NUM(Next)) : 0 );
1189 S_debug_show_study_flags(aTHX_ flags," [ ","]");
1190 Perl_re_printf( aTHX_ "\n");
1191 });
1192}
1193
1194
1195# define DEBUG_STUDYDATA(where, data, depth, is_inf) \
1196 S_debug_studydata(aTHX_ where, data, depth, is_inf)
1197
1198# define DEBUG_PEEP(str, scan, depth, flags) \
1199 S_debug_peep(aTHX_ str, pRExC_state, scan, depth, flags)
1200
1201#else
1202# define DEBUG_STUDYDATA(where, data, depth, is_inf) NOOP
1203# define DEBUG_PEEP(str, scan, depth, flags) NOOP
1204#endif
1de06328 1205
cb41e5d6 1206
c6871b76
KW
1207/* =========================================================
1208 * BEGIN edit_distance stuff.
1209 *
1210 * This calculates how many single character changes of any type are needed to
1211 * transform a string into another one. It is taken from version 3.1 of
1212 *
1213 * https://metacpan.org/pod/Text::Levenshtein::Damerau::XS
1214 */
1215
1216/* Our unsorted dictionary linked list. */
1217/* Note we use UVs, not chars. */
1218
1219struct dictionary{
1220 UV key;
1221 UV value;
1222 struct dictionary* next;
1223};
1224typedef struct dictionary item;
1225
1226
1227PERL_STATIC_INLINE item*
88f063b4 1228push(UV key, item* curr)
c6871b76
KW
1229{
1230 item* head;
d09f14bf 1231 Newx(head, 1, item);
c6871b76
KW
1232 head->key = key;
1233 head->value = 0;
1234 head->next = curr;
1235 return head;
1236}
1237
1238
1239PERL_STATIC_INLINE item*
1240find(item* head, UV key)
1241{
1242 item* iterator = head;
1243 while (iterator){
1244 if (iterator->key == key){
1245 return iterator;
1246 }
1247 iterator = iterator->next;
1248 }
1249
1250 return NULL;
1251}
1252
1253PERL_STATIC_INLINE item*
88f063b4 1254uniquePush(item* head, UV key)
c6871b76
KW
1255{
1256 item* iterator = head;
1257
1258 while (iterator){
1259 if (iterator->key == key) {
1260 return head;
1261 }
1262 iterator = iterator->next;
1263 }
1264
88f063b4 1265 return push(key, head);
c6871b76
KW
1266}
1267
1268PERL_STATIC_INLINE void
1269dict_free(item* head)
1270{
1271 item* iterator = head;
1272
1273 while (iterator) {
1274 item* temp = iterator;
1275 iterator = iterator->next;
1276 Safefree(temp);
1277 }
1278
1279 head = NULL;
1280}
1281
1282/* End of Dictionary Stuff */
1283
1284/* All calculations/work are done here */
1285STATIC int
1286S_edit_distance(const UV* src,
1287 const UV* tgt,
1288 const STRLEN x, /* length of src[] */
1289 const STRLEN y, /* length of tgt[] */
1290 const SSize_t maxDistance
1291)
1292{
1293 item *head = NULL;
88f063b4 1294 UV swapCount, swapScore, targetCharCount, i, j;
c6871b76
KW
1295 UV *scores;
1296 UV score_ceil = x + y;
1297
1298 PERL_ARGS_ASSERT_EDIT_DISTANCE;
1299
1300 /* intialize matrix start values */
d09f14bf 1301 Newx(scores, ( (x + 2) * (y + 2)), UV);
c6871b76
KW
1302 scores[0] = score_ceil;
1303 scores[1 * (y + 2) + 0] = score_ceil;
1304 scores[0 * (y + 2) + 1] = score_ceil;
1305 scores[1 * (y + 2) + 1] = 0;
88f063b4 1306 head = uniquePush(uniquePush(head, src[0]), tgt[0]);
c6871b76
KW
1307
1308 /* work loops */
1309 /* i = src index */
1310 /* j = tgt index */
1311 for (i=1;i<=x;i++) {
1312 if (i < x)
88f063b4 1313 head = uniquePush(head, src[i]);
c6871b76
KW
1314 scores[(i+1) * (y + 2) + 1] = i;
1315 scores[(i+1) * (y + 2) + 0] = score_ceil;
1316 swapCount = 0;
1317
1318 for (j=1;j<=y;j++) {
1319 if (i == 1) {
1320 if(j < y)
88f063b4 1321 head = uniquePush(head, tgt[j]);
c6871b76
KW
1322 scores[1 * (y + 2) + (j + 1)] = j;
1323 scores[0 * (y + 2) + (j + 1)] = score_ceil;
1324 }
1325
88f063b4 1326 targetCharCount = find(head, tgt[j-1])->value;
c6871b76
KW
1327 swapScore = scores[targetCharCount * (y + 2) + swapCount] + i - targetCharCount - 1 + j - swapCount;
1328
1329 if (src[i-1] != tgt[j-1]){
1330 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));
1331 }
1332 else {
1333 swapCount = j;
1334 scores[(i+1) * (y + 2) + (j + 1)] = MIN(scores[i * (y + 2) + j], swapScore);
1335 }
1336 }
1337
88f063b4 1338 find(head, src[i-1])->value = i;
c6871b76
KW
1339 }
1340
1341 {
1342 IV score = scores[(x+1) * (y + 2) + (y + 1)];
1343 dict_free(head);
1344 Safefree(scores);
1345 return (maxDistance != 0 && maxDistance < score)?(-1):score;
1346 }
1347}
1348
1349/* END of edit_distance() stuff
1350 * ========================================================= */
1351
8e35b056
KW
1352/* is c a control character for which we have a mnemonic? */
1353#define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
1354
549b4e78
KW
1355STATIC const char *
1356S_cntrl_to_mnemonic(const U8 c)
1357{
1358 /* Returns the mnemonic string that represents character 'c', if one
1359 * exists; NULL otherwise. The only ones that exist for the purposes of
1360 * this routine are a few control characters */
1361
1362 switch (c) {
1363 case '\a': return "\\a";
1364 case '\b': return "\\b";
1365 case ESC_NATIVE: return "\\e";
1366 case '\f': return "\\f";
1367 case '\n': return "\\n";
1368 case '\r': return "\\r";
1369 case '\t': return "\\t";
1370 }
1371
1372 return NULL;
1373}
1374
653099ff 1375/* Mark that we cannot extend a found fixed substring at this point.
37b6262f 1376 Update the longest found anchored substring or the longest found
653099ff
GS
1377 floating substrings if needed. */
1378
4327152a 1379STATIC void
ea3daa5d
FC
1380S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
1381 SSize_t *minlenp, int is_inf)
c277df42 1382{
e1ec3a88 1383 const STRLEN l = CHR_SVLEN(data->last_found);
2099df82 1384 SV * const longest_sv = data->substrs[data->cur_is_floating].str;
a2ca2017 1385 const STRLEN old_l = CHR_SVLEN(longest_sv);
1de06328 1386 GET_RE_DEBUG_FLAGS_DECL;
b81d288d 1387
7918f24d
NC
1388 PERL_ARGS_ASSERT_SCAN_COMMIT;
1389
c277df42 1390 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
2099df82 1391 const U8 i = data->cur_is_floating;
a2ca2017 1392 SvSetMagicSV(longest_sv, data->last_found);
37b6262f
DM
1393 data->substrs[i].min_offset = l ? data->last_start_min : data->pos_min;
1394
2099df82
DM
1395 if (!i) /* fixed */
1396 data->substrs[0].max_offset = data->substrs[0].min_offset;
1397 else { /* float */
1398 data->substrs[1].max_offset = (l
646e8787
DM
1399 ? data->last_start_max
1400 : (data->pos_delta > SSize_t_MAX - data->pos_min
ea3daa5d
FC
1401 ? SSize_t_MAX
1402 : data->pos_min + data->pos_delta));
1403 if (is_inf
2099df82
DM
1404 || (STRLEN)data->substrs[1].max_offset > (STRLEN)SSize_t_MAX)
1405 data->substrs[1].max_offset = SSize_t_MAX;
37b6262f 1406 }
11683ecb 1407
37b6262f
DM
1408 if (data->flags & SF_BEFORE_EOL)
1409 data->substrs[i].flags |= (data->flags & SF_BEFORE_EOL);
1410 else
1411 data->substrs[i].flags &= ~SF_BEFORE_EOL;
1412 data->substrs[i].minlenp = minlenp;
1413 data->substrs[i].lookbehind = 0;
c277df42 1414 }
37b6262f 1415
c277df42 1416 SvCUR_set(data->last_found, 0);
0eda9292 1417 {
a28509cc 1418 SV * const sv = data->last_found;
097eb12c
AL
1419 if (SvUTF8(sv) && SvMAGICAL(sv)) {
1420 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
1421 if (mg)
1422 mg->mg_len = 0;
1423 }
0eda9292 1424 }
c277df42
IZ
1425 data->last_end = -1;
1426 data->flags &= ~SF_BEFORE_EOL;
f5a36d78 1427 DEBUG_STUDYDATA("commit", data, 0, is_inf);
c277df42
IZ
1428}
1429
cdd87c1d
KW
1430/* An SSC is just a regnode_charclass_posix with an extra field: the inversion
1431 * list that describes which code points it matches */
1432
653099ff 1433STATIC void
3420edd7 1434S_ssc_anything(pTHX_ regnode_ssc *ssc)
653099ff 1435{
cdd87c1d
KW
1436 /* Set the SSC 'ssc' to match an empty string or any code point */
1437
557bd3fb 1438 PERL_ARGS_ASSERT_SSC_ANYTHING;
7918f24d 1439
71068078 1440 assert(is_ANYOF_SYNTHETIC(ssc));
3fffb88a 1441
0854ea0b
KW
1442 /* mortalize so won't leak */
1443 ssc->invlist = sv_2mortal(_add_range_to_invlist(NULL, 0, UV_MAX));
93e92956 1444 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
653099ff
GS
1445}
1446
653099ff 1447STATIC int
dc3bf405 1448S_ssc_is_anything(const regnode_ssc *ssc)
653099ff 1449{
c144baaa
KW
1450 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
1451 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
1452 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1453 * in any way, so there's no point in using it */
cdd87c1d
KW
1454
1455 UV start, end;
1456 bool ret;
653099ff 1457
557bd3fb 1458 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
7918f24d 1459
71068078 1460 assert(is_ANYOF_SYNTHETIC(ssc));
cdd87c1d 1461
93e92956 1462 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
cdd87c1d
KW
1463 return FALSE;
1464 }
1465
1466 /* See if the list consists solely of the range 0 - Infinity */
1467 invlist_iterinit(ssc->invlist);
1468 ret = invlist_iternext(ssc->invlist, &start, &end)
1469 && start == 0
1470 && end == UV_MAX;
1471
1472 invlist_iterfinish(ssc->invlist);
1473
1474 if (ret) {
1475 return TRUE;
1476 }
1477
1478 /* If e.g., both \w and \W are set, matches everything */
e0e1be5f 1479 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1480 int i;
1481 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1482 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1483 return TRUE;
1484 }
1485 }
1486 }
1487
1488 return FALSE;
653099ff
GS
1489}
1490
653099ff 1491STATIC void
cdd87c1d 1492S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
653099ff 1493{
cdd87c1d
KW
1494 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1495 * string, any code point, or any posix class under locale */
1496
557bd3fb 1497 PERL_ARGS_ASSERT_SSC_INIT;
7918f24d 1498
557bd3fb 1499 Zero(ssc, 1, regnode_ssc);
71068078 1500 set_ANYOF_SYNTHETIC(ssc);
93e92956 1501 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
3420edd7 1502 ssc_anything(ssc);
cdd87c1d 1503
2f306ab9
KW
1504 /* If any portion of the regex is to operate under locale rules that aren't
1505 * fully known at compile time, initialization includes it. The reason
1506 * this isn't done for all regexes is that the optimizer was written under
1507 * the assumption that locale was all-or-nothing. Given the complexity and
1508 * lack of documentation in the optimizer, and that there are inadequate
1509 * test cases for locale, many parts of it may not work properly, it is
1510 * safest to avoid locale unless necessary. */
cdd87c1d
KW
1511 if (RExC_contains_locale) {
1512 ANYOF_POSIXL_SETALL(ssc);
cdd87c1d
KW
1513 }
1514 else {
1515 ANYOF_POSIXL_ZERO(ssc);
1516 }
653099ff
GS
1517}
1518
b423522f 1519STATIC int
dc3bf405
BF
1520S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1521 const regnode_ssc *ssc)
b423522f
KW
1522{
1523 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1524 * to the list of code points matched, and locale posix classes; hence does
1525 * not check its flags) */
1526
1527 UV start, end;
1528 bool ret;
1529
1530 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1531
71068078 1532 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1533
1534 invlist_iterinit(ssc->invlist);
1535 ret = invlist_iternext(ssc->invlist, &start, &end)
1536 && start == 0
1537 && end == UV_MAX;
1538
1539 invlist_iterfinish(ssc->invlist);
1540
1541 if (! ret) {
1542 return FALSE;
1543 }
1544
e0e1be5f 1545 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
31f05a37 1546 return FALSE;
b423522f
KW
1547 }
1548
1549 return TRUE;
1550}
1551
1552STATIC SV*
1553S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
5c0f85ef 1554 const regnode_charclass* const node)
b423522f
KW
1555{
1556 /* Returns a mortal inversion list defining which code points are matched
1557 * by 'node', which is of type ANYOF. Handles complementing the result if
1558 * appropriate. If some code points aren't knowable at this time, the
31f05a37
KW
1559 * returned list must, and will, contain every code point that is a
1560 * possibility. */
b423522f 1561
e2506fa7 1562 SV* invlist = NULL;
1ee208c4 1563 SV* only_utf8_locale_invlist = NULL;
b423522f
KW
1564 unsigned int i;
1565 const U32 n = ARG(node);
31f05a37 1566 bool new_node_has_latin1 = FALSE;
b423522f
KW
1567
1568 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1569
1570 /* Look at the data structure created by S_set_ANYOF_arg() */
93e92956 1571 if (n != ANYOF_ONLY_HAS_BITMAP) {
b423522f
KW
1572 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1573 AV * const av = MUTABLE_AV(SvRV(rv));
1574 SV **const ary = AvARRAY(av);
1575 assert(RExC_rxi->data->what[n] == 's');
1576
1577 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
28118b9c 1578 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1]), NULL));
b423522f
KW
1579 }
1580 else if (ary[0] && ary[0] != &PL_sv_undef) {
1581
1582 /* Here, no compile-time swash, and there are things that won't be
1583 * known until runtime -- we have to assume it could be anything */
e2506fa7 1584 invlist = sv_2mortal(_new_invlist(1));
b423522f
KW
1585 return _add_range_to_invlist(invlist, 0, UV_MAX);
1586 }
1ee208c4 1587 else if (ary[3] && ary[3] != &PL_sv_undef) {
b423522f
KW
1588
1589 /* Here no compile-time swash, and no run-time only data. Use the
1590 * node's inversion list */
28118b9c 1591 invlist = sv_2mortal(invlist_clone(ary[3], NULL));
1ee208c4
KW
1592 }
1593
1594 /* Get the code points valid only under UTF-8 locales */
037715a6 1595 if ((ANYOF_FLAGS(node) & ANYOFL_FOLD)
1ee208c4
KW
1596 && ary[2] && ary[2] != &PL_sv_undef)
1597 {
1598 only_utf8_locale_invlist = ary[2];
b423522f
KW
1599 }
1600 }
1601
e2506fa7
KW
1602 if (! invlist) {
1603 invlist = sv_2mortal(_new_invlist(0));
1604 }
1605
dcb20b36
KW
1606 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1607 * code points, and an inversion list for the others, but if there are code
1608 * points that should match only conditionally on the target string being
1609 * UTF-8, those are placed in the inversion list, and not the bitmap.
1610 * Since there are circumstances under which they could match, they are
1611 * included in the SSC. But if the ANYOF node is to be inverted, we have
1612 * to exclude them here, so that when we invert below, the end result
1613 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1614 * have to do this here before we add the unconditionally matched code
1615 * points */
b423522f
KW
1616 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1617 _invlist_intersection_complement_2nd(invlist,
1618 PL_UpperLatin1,
1619 &invlist);
1620 }
1621
1622 /* Add in the points from the bit map */
dcb20b36 1623 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
b423522f 1624 if (ANYOF_BITMAP_TEST(node, i)) {
6f8848d5
TC
1625 unsigned int start = i++;
1626
1627 for (; i < NUM_ANYOF_CODE_POINTS && ANYOF_BITMAP_TEST(node, i); ++i) {
1628 /* empty */
1629 }
1630 invlist = _add_range_to_invlist(invlist, start, i-1);
31f05a37 1631 new_node_has_latin1 = TRUE;
b423522f
KW
1632 }
1633 }
1634
1635 /* If this can match all upper Latin1 code points, have to add them
ac33c516
KW
1636 * as well. But don't add them if inverting, as when that gets done below,
1637 * it would exclude all these characters, including the ones it shouldn't
1638 * that were added just above */
1639 if (! (ANYOF_FLAGS(node) & ANYOF_INVERT) && OP(node) == ANYOFD
f240c685
KW
1640 && (ANYOF_FLAGS(node) & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER))
1641 {
b423522f
KW
1642 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1643 }
1644
1645 /* Similarly for these */
93e92956 1646 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
e0a1ff7a 1647 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
b423522f
KW
1648 }
1649
1650 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1651 _invlist_invert(invlist);
1652 }
037715a6 1653 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOFL_FOLD) {
31f05a37
KW
1654
1655 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1656 * locale. We can skip this if there are no 0-255 at all. */
1657 _invlist_union(invlist, PL_Latin1, &invlist);
1658 }
1659
1ee208c4
KW
1660 /* Similarly add the UTF-8 locale possible matches. These have to be
1661 * deferred until after the non-UTF-8 locale ones are taken care of just
1662 * above, or it leads to wrong results under ANYOF_INVERT */
1663 if (only_utf8_locale_invlist) {
31f05a37 1664 _invlist_union_maybe_complement_2nd(invlist,
1ee208c4 1665 only_utf8_locale_invlist,
31f05a37
KW
1666 ANYOF_FLAGS(node) & ANYOF_INVERT,
1667 &invlist);
1668 }
b423522f
KW
1669
1670 return invlist;
1671}
1672
1051e1c4 1673/* These two functions currently do the exact same thing */
557bd3fb 1674#define ssc_init_zero ssc_init
653099ff 1675
cdd87c1d
KW
1676#define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1677#define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1678
557bd3fb 1679/* 'AND' a given class with another one. Can create false positives. 'ssc'
93e92956
KW
1680 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1681 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
cdd87c1d 1682
653099ff 1683STATIC void
b423522f 1684S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1685 const regnode_charclass *and_with)
653099ff 1686{
cdd87c1d
KW
1687 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1688 * another SSC or a regular ANYOF class. Can create false positives. */
40d049e4 1689
a0dd4231
KW
1690 SV* anded_cp_list;
1691 U8 anded_flags;
1e6ade67 1692
cdd87c1d 1693 PERL_ARGS_ASSERT_SSC_AND;
653099ff 1694
71068078 1695 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1696
1697 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1698 * the code point inversion list and just the relevant flags */
71068078 1699 if (is_ANYOF_SYNTHETIC(and_with)) {
7dcac5f6 1700 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
a0dd4231 1701 anded_flags = ANYOF_FLAGS(and_with);
e9b08962
KW
1702
1703 /* XXX This is a kludge around what appears to be deficiencies in the
1704 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1705 * there are paths through the optimizer where it doesn't get weeded
1706 * out when it should. And if we don't make some extra provision for
1707 * it like the code just below, it doesn't get added when it should.
1708 * This solution is to add it only when AND'ing, which is here, and
1709 * only when what is being AND'ed is the pristine, original node
1710 * matching anything. Thus it is like adding it to ssc_anything() but
1711 * only when the result is to be AND'ed. Probably the same solution
1712 * could be adopted for the same problem we have with /l matching,
1713 * which is solved differently in S_ssc_init(), and that would lead to
1714 * fewer false positives than that solution has. But if this solution
1715 * creates bugs, the consequences are only that a warning isn't raised
1716 * that should be; while the consequences for having /l bugs is
1717 * incorrect matches */
7dcac5f6 1718 if (ssc_is_anything((regnode_ssc *)and_with)) {
f240c685 1719 anded_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
e9b08962 1720 }
a0dd4231
KW
1721 }
1722 else {
5c0f85ef 1723 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
f240c685
KW
1724 if (OP(and_with) == ANYOFD) {
1725 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1726 }
1727 else {
1728 anded_flags = ANYOF_FLAGS(and_with)
1729 &( ANYOF_COMMON_FLAGS
108316fb
KW
1730 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1731 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1732 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(and_with))) {
1733 anded_flags &=
1734 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1735 }
f240c685 1736 }
a0dd4231
KW
1737 }
1738
1739 ANYOF_FLAGS(ssc) &= anded_flags;
cdd87c1d
KW
1740
1741 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1742 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1743 * 'and_with' may be inverted. When not inverted, we have the situation of
1744 * computing:
1745 * (C1 | P1) & (C2 | P2)
1746 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1747 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1748 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1749 * <= ((C1 & C2) | P1 | P2)
1750 * Alternatively, the last few steps could be:
1751 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1752 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1753 * <= (C1 | C2 | (P1 & P2))
1754 * We favor the second approach if either P1 or P2 is non-empty. This is
1755 * because these components are a barrier to doing optimizations, as what
1756 * they match cannot be known until the moment of matching as they are
1757 * dependent on the current locale, 'AND"ing them likely will reduce or
1758 * eliminate them.
1759 * But we can do better if we know that C1,P1 are in their initial state (a
1760 * frequent occurrence), each matching everything:
1761 * (<everything>) & (C2 | P2) = C2 | P2
1762 * Similarly, if C2,P2 are in their initial state (again a frequent
1763 * occurrence), the result is a no-op
1764 * (C1 | P1) & (<everything>) = C1 | P1
1765 *
1766 * Inverted, we have
1767 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1768 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1769 * <= (C1 & ~C2) | (P1 & ~P2)
1770 * */
1aa99e6b 1771
a0dd4231 1772 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
71068078 1773 && ! is_ANYOF_SYNTHETIC(and_with))
a0dd4231 1774 {
cdd87c1d 1775 unsigned int i;
8951c461 1776
cdd87c1d
KW
1777 ssc_intersection(ssc,
1778 anded_cp_list,
1779 FALSE /* Has already been inverted */
1780 );
c6b76537 1781
cdd87c1d
KW
1782 /* If either P1 or P2 is empty, the intersection will be also; can skip
1783 * the loop */
93e92956 1784 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
cdd87c1d
KW
1785 ANYOF_POSIXL_ZERO(ssc);
1786 }
e0e1be5f 1787 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1788
1789 /* Note that the Posix class component P from 'and_with' actually
1790 * looks like:
1791 * P = Pa | Pb | ... | Pn
1792 * where each component is one posix class, such as in [\w\s].
1793 * Thus
1794 * ~P = ~(Pa | Pb | ... | Pn)
1795 * = ~Pa & ~Pb & ... & ~Pn
1796 * <= ~Pa | ~Pb | ... | ~Pn
1797 * The last is something we can easily calculate, but unfortunately
1798 * is likely to have many false positives. We could do better
1799 * in some (but certainly not all) instances if two classes in
1800 * P have known relationships. For example
1801 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1802 * So
1803 * :lower: & :print: = :lower:
1804 * And similarly for classes that must be disjoint. For example,
1805 * since \s and \w can have no elements in common based on rules in
1806 * the POSIX standard,
1807 * \w & ^\S = nothing
1808 * Unfortunately, some vendor locales do not meet the Posix
1809 * standard, in particular almost everything by Microsoft.
1810 * The loop below just changes e.g., \w into \W and vice versa */
1811
1ee208c4 1812 regnode_charclass_posixl temp;
cdd87c1d
KW
1813 int add = 1; /* To calculate the index of the complement */
1814
b1234259 1815 Zero(&temp, 1, regnode_charclass_posixl);
cdd87c1d
KW
1816 ANYOF_POSIXL_ZERO(&temp);
1817 for (i = 0; i < ANYOF_MAX; i++) {
1818 assert(i % 2 != 0
7dcac5f6
KW
1819 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1820 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
cdd87c1d 1821
7dcac5f6 1822 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
cdd87c1d
KW
1823 ANYOF_POSIXL_SET(&temp, i + add);
1824 }
1825 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1826 }
1827 ANYOF_POSIXL_AND(&temp, ssc);
c6b76537 1828
cdd87c1d
KW
1829 } /* else ssc already has no posixes */
1830 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1831 in its initial state */
71068078 1832 else if (! is_ANYOF_SYNTHETIC(and_with)
7dcac5f6 1833 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
cdd87c1d
KW
1834 {
1835 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1836 * copy it over 'ssc' */
1837 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
71068078 1838 if (is_ANYOF_SYNTHETIC(and_with)) {
cdd87c1d
KW
1839 StructCopy(and_with, ssc, regnode_ssc);
1840 }
1841 else {
1842 ssc->invlist = anded_cp_list;
1843 ANYOF_POSIXL_ZERO(ssc);
93e92956 1844 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1845 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
cdd87c1d
KW
1846 }
1847 }
1848 }
e0e1be5f 1849 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
93e92956 1850 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
cdd87c1d
KW
1851 {
1852 /* One or the other of P1, P2 is non-empty. */
93e92956 1853 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1ea8b7fe
KW
1854 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1855 }
cdd87c1d
KW
1856 ssc_union(ssc, anded_cp_list, FALSE);
1857 }
1858 else { /* P1 = P2 = empty */
1859 ssc_intersection(ssc, anded_cp_list, FALSE);
1860 }
137165a6 1861 }
653099ff
GS
1862}
1863
653099ff 1864STATIC void
cdd87c1d 1865S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1866 const regnode_charclass *or_with)
653099ff 1867{
cdd87c1d
KW
1868 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1869 * another SSC or a regular ANYOF class. Can create false positives if
1870 * 'or_with' is to be inverted. */
7918f24d 1871
a0dd4231
KW
1872 SV* ored_cp_list;
1873 U8 ored_flags;
c6b76537 1874
cdd87c1d 1875 PERL_ARGS_ASSERT_SSC_OR;
c6b76537 1876
71068078 1877 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1878
1879 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1880 * the code point inversion list and just the relevant flags */
71068078 1881 if (is_ANYOF_SYNTHETIC(or_with)) {
7dcac5f6 1882 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
a0dd4231
KW
1883 ored_flags = ANYOF_FLAGS(or_with);
1884 }
1885 else {
5c0f85ef 1886 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
eff8b7dc 1887 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
f240c685
KW
1888 if (OP(or_with) != ANYOFD) {
1889 ored_flags
1890 |= ANYOF_FLAGS(or_with)
108316fb
KW
1891 & ( ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1892 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1893 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(or_with))) {
1894 ored_flags |=
1895 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1896 }
f240c685 1897 }
a0dd4231
KW
1898 }
1899
1900 ANYOF_FLAGS(ssc) |= ored_flags;
cdd87c1d
KW
1901
1902 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1903 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1904 * 'or_with' may be inverted. When not inverted, we have the simple
1905 * situation of computing:
1906 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1907 * If P1|P2 yields a situation with both a class and its complement are
1908 * set, like having both \w and \W, this matches all code points, and we
1909 * can delete these from the P component of the ssc going forward. XXX We
1910 * might be able to delete all the P components, but I (khw) am not certain
1911 * about this, and it is better to be safe.
1912 *
1913 * Inverted, we have
1914 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1915 * <= (C1 | P1) | ~C2
1916 * <= (C1 | ~C2) | P1
1917 * (which results in actually simpler code than the non-inverted case)
1918 * */
9826f543 1919
a0dd4231 1920 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
71068078 1921 && ! is_ANYOF_SYNTHETIC(or_with))
a0dd4231 1922 {
cdd87c1d 1923 /* We ignore P2, leaving P1 going forward */
1ea8b7fe 1924 } /* else Not inverted */
93e92956 1925 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1926 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
e0e1be5f 1927 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1928 unsigned int i;
1929 for (i = 0; i < ANYOF_MAX; i += 2) {
1930 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1931 {
1932 ssc_match_all_cp(ssc);
1933 ANYOF_POSIXL_CLEAR(ssc, i);
1934 ANYOF_POSIXL_CLEAR(ssc, i+1);
cdd87c1d
KW
1935 }
1936 }
1937 }
1aa99e6b 1938 }
cdd87c1d
KW
1939
1940 ssc_union(ssc,
1941 ored_cp_list,
1942 FALSE /* Already has been inverted */
1943 );
653099ff
GS
1944}
1945
b423522f
KW
1946PERL_STATIC_INLINE void
1947S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1948{
1949 PERL_ARGS_ASSERT_SSC_UNION;
1950
71068078 1951 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1952
1953 _invlist_union_maybe_complement_2nd(ssc->invlist,
1954 invlist,
1955 invert2nd,
1956 &ssc->invlist);
1957}
1958
1959PERL_STATIC_INLINE void
1960S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1961 SV* const invlist,
1962 const bool invert2nd)
1963{
1964 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1965
71068078 1966 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1967
1968 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1969 invlist,
1970 invert2nd,
1971 &ssc->invlist);
1972}
1973
1974PERL_STATIC_INLINE void
1975S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1976{
1977 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1978
71068078 1979 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1980
1981 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1982}
1983
1984PERL_STATIC_INLINE void
1985S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1986{
1987 /* AND just the single code point 'cp' into the SSC 'ssc' */
1988
1989 SV* cp_list = _new_invlist(2);
1990
1991 PERL_ARGS_ASSERT_SSC_CP_AND;
1992
71068078 1993 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1994
1995 cp_list = add_cp_to_invlist(cp_list, cp);
1996 ssc_intersection(ssc, cp_list,
1997 FALSE /* Not inverted */
1998 );
1999 SvREFCNT_dec_NN(cp_list);
2000}
2001
2002PERL_STATIC_INLINE void
dc3bf405 2003S_ssc_clear_locale(regnode_ssc *ssc)
b423522f
KW
2004{
2005 /* Set the SSC 'ssc' to not match any locale things */
b423522f
KW
2006 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
2007
71068078 2008 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
2009
2010 ANYOF_POSIXL_ZERO(ssc);
2011 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
2012}
2013
b35552de
KW
2014#define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
2015
2016STATIC bool
2017S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
2018{
2019 /* The synthetic start class is used to hopefully quickly winnow down
2020 * places where a pattern could start a match in the target string. If it
2021 * doesn't really narrow things down that much, there isn't much point to
2022 * having the overhead of using it. This function uses some very crude
2023 * heuristics to decide if to use the ssc or not.
2024 *
2025 * It returns TRUE if 'ssc' rules out more than half what it considers to
2026 * be the "likely" possible matches, but of course it doesn't know what the
2027 * actual things being matched are going to be; these are only guesses
2028 *
2029 * For /l matches, it assumes that the only likely matches are going to be
2030 * in the 0-255 range, uniformly distributed, so half of that is 127
2031 * For /a and /d matches, it assumes that the likely matches will be just
2032 * the ASCII range, so half of that is 63
2033 * For /u and there isn't anything matching above the Latin1 range, it
2034 * assumes that that is the only range likely to be matched, and uses
2035 * half that as the cut-off: 127. If anything matches above Latin1,
2036 * it assumes that all of Unicode could match (uniformly), except for
2037 * non-Unicode code points and things in the General Category "Other"
2038 * (unassigned, private use, surrogates, controls and formats). This
2039 * is a much large number. */
2040
b35552de
KW
2041 U32 count = 0; /* Running total of number of code points matched by
2042 'ssc' */
2043 UV start, end; /* Start and end points of current range in inversion
2044 list */
72400949
KW
2045 const U32 max_code_points = (LOC)
2046 ? 256
2047 : (( ! UNI_SEMANTICS
2048 || invlist_highest(ssc->invlist) < 256)
2049 ? 128
2050 : NON_OTHER_COUNT);
2051 const U32 max_match = max_code_points / 2;
b35552de
KW
2052
2053 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
2054
2055 invlist_iterinit(ssc->invlist);
2056 while (invlist_iternext(ssc->invlist, &start, &end)) {
72400949
KW
2057 if (start >= max_code_points) {
2058 break;
b35552de 2059 }
72400949 2060 end = MIN(end, max_code_points - 1);
b35552de 2061 count += end - start + 1;
72400949 2062 if (count >= max_match) {
b35552de
KW
2063 invlist_iterfinish(ssc->invlist);
2064 return FALSE;
2065 }
2066 }
2067
2068 return TRUE;
2069}
2070
2071
b423522f
KW
2072STATIC void
2073S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
2074{
2075 /* The inversion list in the SSC is marked mortal; now we need a more
2076 * permanent copy, which is stored the same way that is done in a regular
dcb20b36
KW
2077 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
2078 * map */
b423522f 2079
28118b9c 2080 SV* invlist = invlist_clone(ssc->invlist, NULL);
b423522f
KW
2081
2082 PERL_ARGS_ASSERT_SSC_FINALIZE;
2083
71068078 2084 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f 2085
a0dd4231 2086 /* The code in this file assumes that all but these flags aren't relevant
93e92956
KW
2087 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
2088 * by the time we reach here */
f240c685
KW
2089 assert(! (ANYOF_FLAGS(ssc)
2090 & ~( ANYOF_COMMON_FLAGS
108316fb
KW
2091 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
2092 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)));
a0dd4231 2093
b423522f
KW
2094 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
2095
1ee208c4
KW
2096 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
2097 NULL, NULL, NULL, FALSE);
b423522f 2098
85c8e306
KW
2099 /* Make sure is clone-safe */
2100 ssc->invlist = NULL;
2101
e0e1be5f 2102 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
93e92956 2103 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
d156f5cb 2104 OP(ssc) = ANYOFPOSIXL;
e0e1be5f 2105 }
d156f5cb 2106 else if (RExC_contains_locale) {
b2e90ddf
KW
2107 OP(ssc) = ANYOFL;
2108 }
2109
1462525b 2110 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
b423522f
KW
2111}
2112
a3621e74
YO
2113#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
2114#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
2115#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
538e84ed
KW
2116#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
2117 ? (TRIE_LIST_CUR( idx ) - 1) \
2118 : 0 )
a3621e74 2119
3dab1dad
YO
2120
2121#ifdef DEBUGGING
07be1b83 2122/*
2b8b4781
NC
2123 dump_trie(trie,widecharmap,revcharmap)
2124 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
2125 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
2126
2127 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
2128 The _interim_ variants are used for debugging the interim
2129 tables that are used to generate the final compressed
2130 representation which is what dump_trie expects.
2131
486ec47a 2132 Part of the reason for their existence is to provide a form
3dab1dad 2133 of documentation as to how the different representations function.
07be1b83
YO
2134
2135*/
3dab1dad
YO
2136
2137/*
3dab1dad
YO
2138 Dumps the final compressed table form of the trie to Perl_debug_log.
2139 Used for debugging make_trie().
2140*/
b9a59e08 2141
3dab1dad 2142STATIC void
2b8b4781
NC
2143S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
2144 AV *revcharmap, U32 depth)
3dab1dad
YO
2145{
2146 U32 state;
ab3bbdeb 2147 SV *sv=sv_newmortal();
55eed653 2148 int colwidth= widecharmap ? 6 : 4;
2e64971a 2149 U16 word;
3dab1dad
YO
2150 GET_RE_DEBUG_FLAGS_DECL;
2151
7918f24d 2152 PERL_ARGS_ASSERT_DUMP_TRIE;
ab3bbdeb 2153
6ad9a8ab 2154 Perl_re_indentf( aTHX_ "Char : %-6s%-6s%-4s ",
1e37780e 2155 depth+1, "Match","Base","Ofs" );
3dab1dad
YO
2156
2157 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 2158 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 2159 if ( tmp ) {
6ad9a8ab 2160 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2161 colwidth,
538e84ed 2162 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2163 PL_colors[0], PL_colors[1],
2164 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2165 PERL_PV_ESCAPE_FIRSTCHAR
2166 )
ab3bbdeb 2167 );
3dab1dad
YO
2168 }
2169 }
1e37780e
YO
2170 Perl_re_printf( aTHX_ "\n");
2171 Perl_re_indentf( aTHX_ "State|-----------------------", depth+1);
3dab1dad
YO
2172
2173 for( state = 0 ; state < trie->uniquecharcount ; state++ )
6ad9a8ab
YO
2174 Perl_re_printf( aTHX_ "%.*s", colwidth, "--------");
2175 Perl_re_printf( aTHX_ "\n");
3dab1dad 2176
1e2e3d02 2177 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 2178 const U32 base = trie->states[ state ].trans.base;
3dab1dad 2179
147e3846 2180 Perl_re_indentf( aTHX_ "#%4" UVXf "|", depth+1, (UV)state);
3dab1dad
YO
2181
2182 if ( trie->states[ state ].wordnum ) {
1e37780e 2183 Perl_re_printf( aTHX_ " W%4X", trie->states[ state ].wordnum );
3dab1dad 2184 } else {
6ad9a8ab 2185 Perl_re_printf( aTHX_ "%6s", "" );
3dab1dad
YO
2186 }
2187
147e3846 2188 Perl_re_printf( aTHX_ " @%4" UVXf " ", (UV)base );
3dab1dad
YO
2189
2190 if ( base ) {
2191 U32 ofs = 0;
2192
2193 while( ( base + ofs < trie->uniquecharcount ) ||
2194 ( base + ofs - trie->uniquecharcount < trie->lasttrans
538e84ed
KW
2195 && trie->trans[ base + ofs - trie->uniquecharcount ].check
2196 != state))
3dab1dad
YO
2197 ofs++;
2198
147e3846 2199 Perl_re_printf( aTHX_ "+%2" UVXf "[ ", (UV)ofs);
3dab1dad
YO
2200
2201 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
538e84ed
KW
2202 if ( ( base + ofs >= trie->uniquecharcount )
2203 && ( base + ofs - trie->uniquecharcount
2204 < trie->lasttrans )
2205 && trie->trans[ base + ofs
2206 - trie->uniquecharcount ].check == state )
3dab1dad 2207 {
147e3846 2208 Perl_re_printf( aTHX_ "%*" UVXf, colwidth,
1e37780e
YO
2209 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next
2210 );
3dab1dad 2211 } else {
88f063b4 2212 Perl_re_printf( aTHX_ "%*s", colwidth," ." );
3dab1dad
YO
2213 }
2214 }
2215
6ad9a8ab 2216 Perl_re_printf( aTHX_ "]");
3dab1dad
YO
2217
2218 }
6ad9a8ab 2219 Perl_re_printf( aTHX_ "\n" );
3dab1dad 2220 }
6ad9a8ab 2221 Perl_re_indentf( aTHX_ "word_info N:(prev,len)=",
cb41e5d6 2222 depth);
2e64971a 2223 for (word=1; word <= trie->wordcount; word++) {
6ad9a8ab 2224 Perl_re_printf( aTHX_ " %d:(%d,%d)",
2e64971a
DM
2225 (int)word, (int)(trie->wordinfo[word].prev),
2226 (int)(trie->wordinfo[word].len));
2227 }
6ad9a8ab 2228 Perl_re_printf( aTHX_ "\n" );
538e84ed 2229}
3dab1dad 2230/*
3dab1dad 2231 Dumps a fully constructed but uncompressed trie in list form.
538e84ed 2232 List tries normally only are used for construction when the number of
3dab1dad
YO
2233 possible chars (trie->uniquecharcount) is very high.
2234 Used for debugging make_trie().
2235*/
2236STATIC void
55eed653 2237S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2238 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2239 U32 depth)
3dab1dad
YO
2240{
2241 U32 state;
ab3bbdeb 2242 SV *sv=sv_newmortal();
55eed653 2243 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2244 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2245
2246 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
2247
3dab1dad 2248 /* print out the table precompression. */
6ad9a8ab 2249 Perl_re_indentf( aTHX_ "State :Word | Transition Data\n",
cb41e5d6 2250 depth+1 );
6ad9a8ab 2251 Perl_re_indentf( aTHX_ "%s",
cb41e5d6 2252 depth+1, "------:-----+-----------------\n" );
538e84ed 2253
3dab1dad
YO
2254 for( state=1 ; state < next_alloc ; state ++ ) {
2255 U16 charid;
538e84ed 2256
147e3846 2257 Perl_re_indentf( aTHX_ " %4" UVXf " :",
cb41e5d6 2258 depth+1, (UV)state );
3dab1dad 2259 if ( ! trie->states[ state ].wordnum ) {
6ad9a8ab 2260 Perl_re_printf( aTHX_ "%5s| ","");
3dab1dad 2261 } else {
6ad9a8ab 2262 Perl_re_printf( aTHX_ "W%4x| ",
3dab1dad
YO
2263 trie->states[ state ].wordnum
2264 );
2265 }
2266 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
538e84ed 2267 SV ** const tmp = av_fetch( revcharmap,
88f063b4 2268 TRIE_LIST_ITEM(state, charid).forid, 0);
ab3bbdeb 2269 if ( tmp ) {
147e3846 2270 Perl_re_printf( aTHX_ "%*s:%3X=%4" UVXf " | ",
ab3bbdeb 2271 colwidth,
538e84ed
KW
2272 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
2273 colwidth,
2274 PL_colors[0], PL_colors[1],
2275 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2276 | PERL_PV_ESCAPE_FIRSTCHAR
ab3bbdeb 2277 ) ,
88f063b4
KW
2278 TRIE_LIST_ITEM(state, charid).forid,
2279 (UV)TRIE_LIST_ITEM(state, charid).newstate
1e2e3d02 2280 );
538e84ed 2281 if (!(charid % 10))
6ad9a8ab 2282 Perl_re_printf( aTHX_ "\n%*s| ",
664e119d 2283 (int)((depth * 2) + 14), "");
1e2e3d02 2284 }
ab3bbdeb 2285 }
6ad9a8ab 2286 Perl_re_printf( aTHX_ "\n");
3dab1dad 2287 }
538e84ed 2288}
3dab1dad
YO
2289
2290/*
3dab1dad 2291 Dumps a fully constructed but uncompressed trie in table form.
538e84ed
KW
2292 This is the normal DFA style state transition table, with a few
2293 twists to facilitate compression later.
3dab1dad
YO
2294 Used for debugging make_trie().
2295*/
2296STATIC void
55eed653 2297S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2298 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2299 U32 depth)
3dab1dad
YO
2300{
2301 U32 state;
2302 U16 charid;
ab3bbdeb 2303 SV *sv=sv_newmortal();
55eed653 2304 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2305 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2306
2307 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
538e84ed 2308
3dab1dad
YO
2309 /*
2310 print out the table precompression so that we can do a visual check
2311 that they are identical.
2312 */
538e84ed 2313
6ad9a8ab 2314 Perl_re_indentf( aTHX_ "Char : ", depth+1 );
3dab1dad
YO
2315
2316 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 2317 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 2318 if ( tmp ) {
6ad9a8ab 2319 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2320 colwidth,
538e84ed 2321 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2322 PL_colors[0], PL_colors[1],
2323 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2324 PERL_PV_ESCAPE_FIRSTCHAR
2325 )
ab3bbdeb 2326 );
3dab1dad
YO
2327 }
2328 }
2329
4aaafc03
YO
2330 Perl_re_printf( aTHX_ "\n");
2331 Perl_re_indentf( aTHX_ "State+-", depth+1 );
3dab1dad
YO
2332
2333 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
6ad9a8ab 2334 Perl_re_printf( aTHX_ "%.*s", colwidth,"--------");
3dab1dad
YO
2335 }
2336
6ad9a8ab 2337 Perl_re_printf( aTHX_ "\n" );
3dab1dad
YO
2338
2339 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
2340
147e3846 2341 Perl_re_indentf( aTHX_ "%4" UVXf " : ",
cb41e5d6 2342 depth+1,
3dab1dad
YO
2343 (UV)TRIE_NODENUM( state ) );
2344
2345 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
2346 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
2347 if (v)
147e3846 2348 Perl_re_printf( aTHX_ "%*" UVXf, colwidth, v );
ab3bbdeb 2349 else
6ad9a8ab 2350 Perl_re_printf( aTHX_ "%*s", colwidth, "." );
3dab1dad
YO
2351 }
2352 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
147e3846 2353 Perl_re_printf( aTHX_ " (%4" UVXf ")\n",
538e84ed 2354 (UV)trie->trans[ state ].check );
3dab1dad 2355 } else {
147e3846 2356 Perl_re_printf( aTHX_ " (%4" UVXf ") W%4X\n",
538e84ed 2357 (UV)trie->trans[ state ].check,
3dab1dad
YO
2358 trie->states[ TRIE_NODENUM( state ) ].wordnum );
2359 }
2360 }
07be1b83 2361}
3dab1dad
YO
2362
2363#endif
2364
2e64971a 2365
786e8c11
YO
2366/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
2367 startbranch: the first branch in the whole branch sequence
2368 first : start branch of sequence of branch-exact nodes.
2369 May be the same as startbranch
2370 last : Thing following the last branch.
2371 May be the same as tail.
2372 tail : item following the branch sequence
2373 count : words in the sequence
a4525e78 2374 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
786e8c11 2375 depth : indent depth
3dab1dad 2376
786e8c11 2377Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 2378
786e8c11
YO
2379A trie is an N'ary tree where the branches are determined by digital
2380decomposition of the key. IE, at the root node you look up the 1st character and
2381follow that branch repeat until you find the end of the branches. Nodes can be
2382marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 2383
786e8c11 2384 /he|she|his|hers/
72f13be8 2385
786e8c11
YO
2386would convert into the following structure. Numbers represent states, letters
2387following numbers represent valid transitions on the letter from that state, if
2388the number is in square brackets it represents an accepting state, otherwise it
2389will be in parenthesis.
07be1b83 2390
786e8c11
YO
2391 +-h->+-e->[3]-+-r->(8)-+-s->[9]
2392 | |
2393 | (2)
2394 | |
2395 (1) +-i->(6)-+-s->[7]
2396 |
2397 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 2398
786e8c11
YO
2399 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
2400
2401This shows that when matching against the string 'hers' we will begin at state 1
2402read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
2403then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
2404is also accepting. Thus we know that we can match both 'he' and 'hers' with a
2405single traverse. We store a mapping from accepting to state to which word was
2406matched, and then when we have multiple possibilities we try to complete the
b8fda935 2407rest of the regex in the order in which they occurred in the alternation.
786e8c11
YO
2408
2409The only prior NFA like behaviour that would be changed by the TRIE support is
2410the silent ignoring of duplicate alternations which are of the form:
2411
2412 / (DUPE|DUPE) X? (?{ ... }) Y /x
2413
4b714af6 2414Thus EVAL blocks following a trie may be called a different number of times with
786e8c11 2415and without the optimisation. With the optimisations dupes will be silently
486ec47a 2416ignored. This inconsistent behaviour of EVAL type nodes is well established as
786e8c11
YO
2417the following demonstrates:
2418
2419 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
2420
2421which prints out 'word' three times, but
2422
2423 'words'=~/(word|word|word)(?{ print $1 })S/
2424
2425which doesnt print it out at all. This is due to other optimisations kicking in.
2426
2427Example of what happens on a structural level:
2428
486ec47a 2429The regexp /(ac|ad|ab)+/ will produce the following debug output:
786e8c11
YO
2430
2431 1: CURLYM[1] {1,32767}(18)
2432 5: BRANCH(8)
2433 6: EXACT <ac>(16)
2434 8: BRANCH(11)
2435 9: EXACT <ad>(16)
2436 11: BRANCH(14)
2437 12: EXACT <ab>(16)
2438 16: SUCCEED(0)
2439 17: NOTHING(18)
2440 18: END(0)
2441
2442This would be optimizable with startbranch=5, first=5, last=16, tail=16
2443and should turn into:
2444
2445 1: CURLYM[1] {1,32767}(18)
2446 5: TRIE(16)
2447 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
2448 <ac>
2449 <ad>
2450 <ab>
2451 16: SUCCEED(0)
2452 17: NOTHING(18)
2453 18: END(0)
2454
2455Cases where tail != last would be like /(?foo|bar)baz/:
2456
2457 1: BRANCH(4)
2458 2: EXACT <foo>(8)
2459 4: BRANCH(7)
2460 5: EXACT <bar>(8)
2461 7: TAIL(8)
2462 8: EXACT <baz>(10)
2463 10: END(0)
2464
2465which would be optimizable with startbranch=1, first=1, last=7, tail=8
2466and would end up looking like:
2467
2468 1: TRIE(8)
2469 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
2470 <foo>
2471 <bar>
2472 7: TAIL(8)
2473 8: EXACT <baz>(10)
2474 10: END(0)
2475
c80e42f3 2476 d = uvchr_to_utf8_flags(d, uv, 0);
786e8c11
YO
2477
2478is the recommended Unicode-aware way of saying
2479
2480 *(d++) = uv;
2481*/
2482
fab2782b 2483#define TRIE_STORE_REVCHAR(val) \
786e8c11 2484 STMT_START { \
73031816 2485 if (UTF) { \
668fcfea 2486 SV *zlopp = newSV(UTF8_MAXBYTES); \
88c9ea1e 2487 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
c80e42f3 2488 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
73031816
NC
2489 SvCUR_set(zlopp, kapow - flrbbbbb); \
2490 SvPOK_on(zlopp); \
2491 SvUTF8_on(zlopp); \
2492 av_push(revcharmap, zlopp); \
2493 } else { \
fab2782b 2494 char ooooff = (char)val; \
73031816
NC
2495 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2496 } \
2497 } STMT_END
786e8c11 2498
914a25d5
KW
2499/* This gets the next character from the input, folding it if not already
2500 * folded. */
2501#define TRIE_READ_CHAR STMT_START { \
2502 wordlen++; \
2503 if ( UTF ) { \
2504 /* if it is UTF then it is either already folded, or does not need \
2505 * folding */ \
1c1d615a 2506 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
914a25d5
KW
2507 } \
2508 else if (folder == PL_fold_latin1) { \
7d006b13
KW
2509 /* This folder implies Unicode rules, which in the range expressible \
2510 * by not UTF is the lower case, with the two exceptions, one of \
2511 * which should have been taken care of before calling this */ \
2512 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2513 uvc = toLOWER_L1(*uc); \
2514 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2515 len = 1; \
914a25d5
KW
2516 } else { \
2517 /* raw data, will be folded later if needed */ \
2518 uvc = (U32)*uc; \
2519 len = 1; \
2520 } \
786e8c11
YO
2521} STMT_END
2522
2523
2524
2525#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2526 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
00195859 2527 U32 ging = TRIE_LIST_LEN( state ) * 2; \
f9003953 2528 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
00195859 2529 TRIE_LIST_LEN( state ) = ging; \
786e8c11
YO
2530 } \
2531 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2532 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2533 TRIE_LIST_CUR( state )++; \
2534} STMT_END
07be1b83 2535
786e8c11 2536#define TRIE_LIST_NEW(state) STMT_START { \
d09f14bf 2537 Newx( trie->states[ state ].trans.list, \
786e8c11
YO
2538 4, reg_trie_trans_le ); \
2539 TRIE_LIST_CUR( state ) = 1; \
2540 TRIE_LIST_LEN( state ) = 4; \
2541} STMT_END
07be1b83 2542
786e8c11
YO
2543#define TRIE_HANDLE_WORD(state) STMT_START { \
2544 U16 dupe= trie->states[ state ].wordnum; \
2545 regnode * const noper_next = regnext( noper ); \
2546 \
786e8c11
YO
2547 DEBUG_r({ \
2548 /* store the word for dumping */ \
2549 SV* tmp; \
2550 if (OP(noper) != NOTHING) \
740cce10 2551 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
786e8c11 2552 else \
740cce10 2553 tmp = newSVpvn_utf8( "", 0, UTF ); \
2b8b4781 2554 av_push( trie_words, tmp ); \
786e8c11
YO
2555 }); \
2556 \
2557 curword++; \
2e64971a
DM
2558 trie->wordinfo[curword].prev = 0; \
2559 trie->wordinfo[curword].len = wordlen; \
2560 trie->wordinfo[curword].accept = state; \
786e8c11
YO
2561 \
2562 if ( noper_next < tail ) { \
2563 if (!trie->jump) \
538e84ed
KW
2564 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2565 sizeof(U16) ); \
7f69552c 2566 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
2567 if (!jumper) \
2568 jumper = noper_next; \
2569 if (!nextbranch) \
2570 nextbranch= regnext(cur); \
2571 } \
2572 \
2573 if ( dupe ) { \
2e64971a
DM
2574 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2575 /* chain, so that when the bits of chain are later */\
2576 /* linked together, the dups appear in the chain */\
2577 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2578 trie->wordinfo[dupe].prev = curword; \
786e8c11
YO
2579 } else { \
2580 /* we haven't inserted this word yet. */ \
2581 trie->states[ state ].wordnum = curword; \
2582 } \
2583} STMT_END
07be1b83 2584
3dab1dad 2585
786e8c11
YO
2586#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2587 ( ( base + charid >= ucharcount \
2588 && base + charid < ubound \
2589 && state == trie->trans[ base - ucharcount + charid ].check \
2590 && trie->trans[ base - ucharcount + charid ].next ) \
2591 ? trie->trans[ base - ucharcount + charid ].next \
2592 : ( state==1 ? special : 0 ) \
2593 )
3dab1dad 2594
8bcafbf4
YO
2595#define TRIE_BITMAP_SET_FOLDED(trie, uvc, folder) \
2596STMT_START { \
2597 TRIE_BITMAP_SET(trie, uvc); \
2598 /* store the folded codepoint */ \
2599 if ( folder ) \
2600 TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); \
2601 \
2602 if ( !UTF ) { \
2603 /* store first byte of utf8 representation of */ \
2604 /* variant codepoints */ \
2605 if (! UVCHR_IS_INVARIANT(uvc)) { \
2606 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); \
2607 } \
2608 } \
2609} STMT_END
786e8c11
YO
2610#define MADE_TRIE 1
2611#define MADE_JUMP_TRIE 2
2612#define MADE_EXACT_TRIE 4
3dab1dad 2613
a3621e74 2614STATIC I32
538e84ed
KW
2615S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2616 regnode *first, regnode *last, regnode *tail,
2617 U32 word_count, U32 flags, U32 depth)
a3621e74
YO
2618{
2619 /* first pass, loop through and scan words */
2620 reg_trie_data *trie;
55eed653 2621 HV *widecharmap = NULL;
2b8b4781 2622 AV *revcharmap = newAV();
a3621e74 2623 regnode *cur;
a3621e74
YO
2624 STRLEN len = 0;
2625 UV uvc = 0;
2626 U16 curword = 0;
2627 U32 next_alloc = 0;
786e8c11
YO
2628 regnode *jumper = NULL;
2629 regnode *nextbranch = NULL;
7f69552c 2630 regnode *convert = NULL;
2e64971a 2631 U32 *prev_states; /* temp array mapping each state to previous one */
a3621e74 2632 /* we just use folder as a flag in utf8 */
1e696034 2633 const U8 * folder = NULL;
a3621e74 2634
3a611511
YO
2635 /* in the below add_data call we are storing either 'tu' or 'tuaa'
2636 * which stands for one trie structure, one hash, optionally followed
2637 * by two arrays */
2b8b4781 2638#ifdef DEBUGGING
3a611511 2639 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuaa"));
2b8b4781
NC
2640 AV *trie_words = NULL;
2641 /* along with revcharmap, this only used during construction but both are
2642 * useful during debugging so we store them in the struct when debugging.
8e11feef 2643 */
2b8b4781 2644#else
cf78de0b 2645 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
3dab1dad 2646 STRLEN trie_charcount=0;
3dab1dad 2647#endif
2b8b4781 2648 SV *re_trie_maxbuff;
a3621e74 2649 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2650
2651 PERL_ARGS_ASSERT_MAKE_TRIE;
72f13be8
YO
2652#ifndef DEBUGGING
2653 PERL_UNUSED_ARG(depth);
2654#endif
a3621e74 2655
1e696034 2656 switch (flags) {
a4525e78 2657 case EXACT: case EXACTL: break;
89829bb5 2658 case EXACTFAA:
fab2782b 2659 case EXACTFU_SS:
a4525e78
KW
2660 case EXACTFU:
2661 case EXACTFLU8: folder = PL_fold_latin1; break;
1e696034 2662 case EXACTF: folder = PL_fold; break;
fab2782b 2663 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
1e696034
KW
2664 }
2665
c944940b 2666 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 2667 trie->refcount = 1;
3dab1dad 2668 trie->startstate = 1;
786e8c11 2669 trie->wordcount = word_count;
f8fc2ecf 2670 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 2671 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
a4525e78 2672 if (flags == EXACT || flags == EXACTL)
c944940b 2673 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2e64971a
DM
2674 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2675 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2676
a3621e74 2677 DEBUG_r({
2b8b4781 2678 trie_words = newAV();
a3621e74 2679 });
a3621e74 2680
0111c4fd 2681 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
316ebaf2 2682 assert(re_trie_maxbuff);
a3621e74 2683 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 2684 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 2685 }
df826430 2686 DEBUG_TRIE_COMPILE_r({
6ad9a8ab 2687 Perl_re_indentf( aTHX_
cb41e5d6
YO
2688 "make_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2689 depth+1,
88f063b4 2690 REG_NODE_NUM(startbranch), REG_NODE_NUM(first),
538e84ed 2691 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
3dab1dad 2692 });
538e84ed 2693
7f69552c
YO
2694 /* Find the node we are going to overwrite */
2695 if ( first == startbranch && OP( last ) != BRANCH ) {
2696 /* whole branch chain */
2697 convert = first;
2698 } else {
2699 /* branch sub-chain */
2700 convert = NEXTOPER( first );
2701 }
538e84ed 2702
a3621e74
YO
2703 /* -- First loop and Setup --
2704
2705 We first traverse the branches and scan each word to determine if it
2706 contains widechars, and how many unique chars there are, this is
2707 important as we have to build a table with at least as many columns as we
2708 have unique chars.
2709
2710 We use an array of integers to represent the character codes 0..255
538e84ed
KW
2711 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2712 the native representation of the character value as the key and IV's for
2713 the coded index.
a3621e74
YO
2714
2715 *TODO* If we keep track of how many times each character is used we can
2716 remap the columns so that the table compression later on is more
3b753521 2717 efficient in terms of memory by ensuring the most common value is in the
a3621e74
YO
2718 middle and the least common are on the outside. IMO this would be better
2719 than a most to least common mapping as theres a decent chance the most
2720 common letter will share a node with the least common, meaning the node
486ec47a 2721 will not be compressible. With a middle is most common approach the worst
a3621e74
YO
2722 case is when we have the least common nodes twice.
2723
2724 */
2725
a3621e74 2726 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
df826430 2727 regnode *noper = NEXTOPER( cur );
944e05e3
YO
2728 const U8 *uc;
2729 const U8 *e;
bc031a7d 2730 int foldlen = 0;
07be1b83 2731 U32 wordlen = 0; /* required init */
bc031a7d
KW
2732 STRLEN minchars = 0;
2733 STRLEN maxchars = 0;
538e84ed
KW
2734 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2735 bitmap?*/
a3621e74 2736
3dab1dad 2737 if (OP(noper) == NOTHING) {
20ed8c88
YO
2738 /* skip past a NOTHING at the start of an alternation
2739 * eg, /(?:)a|(?:b)/ should be the same as /a|b/
2740 */
df826430 2741 regnode *noper_next= regnext(noper);
944e05e3
YO
2742 if (noper_next < tail)
2743 noper= noper_next;
2744 }
2745
dca5fc4c
YO
2746 if ( noper < tail &&
2747 (
2748 OP(noper) == flags ||
2749 (
2750 flags == EXACTFU &&
2751 OP(noper) == EXACTFU_SS
2752 )
2753 )
2754 ) {
944e05e3
YO
2755 uc= (U8*)STRING(noper);
2756 e= uc + STR_LEN(noper);
2757 } else {
2758 trie->minlen= 0;
2759 continue;
3dab1dad 2760 }
df826430 2761
944e05e3 2762
fab2782b 2763 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
02daf0ab
YO
2764 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2765 regardless of encoding */
fab2782b
YO
2766 if (OP( noper ) == EXACTFU_SS) {
2767 /* false positives are ok, so just set this */
0dc4a61d 2768 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
fab2782b
YO
2769 }
2770 }
dca5fc4c 2771
bc031a7d
KW
2772 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2773 branch */
3dab1dad 2774 TRIE_CHARCOUNT(trie)++;
a3621e74 2775 TRIE_READ_CHAR;
645de4ce 2776
bc031a7d
KW
2777 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2778 * is in effect. Under /i, this character can match itself, or
2779 * anything that folds to it. If not under /i, it can match just
2780 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2781 * all fold to k, and all are single characters. But some folds
2782 * expand to more than one character, so for example LATIN SMALL
2783 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2784 * the string beginning at 'uc' is 'ffi', it could be matched by
2785 * three characters, or just by the one ligature character. (It
2786 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2787 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2788 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2789 * match.) The trie needs to know the minimum and maximum number
2790 * of characters that could match so that it can use size alone to
2791 * quickly reject many match attempts. The max is simple: it is
2792 * the number of folded characters in this branch (since a fold is
2793 * never shorter than what folds to it. */
2794
2795 maxchars++;
2796
2797 /* And the min is equal to the max if not under /i (indicated by
2798 * 'folder' being NULL), or there are no multi-character folds. If
2799 * there is a multi-character fold, the min is incremented just
2800 * once, for the character that folds to the sequence. Each
2801 * character in the sequence needs to be added to the list below of
2802 * characters in the trie, but we count only the first towards the
2803 * min number of characters needed. This is done through the
2804 * variable 'foldlen', which is returned by the macros that look
2805 * for these sequences as the number of bytes the sequence
2806 * occupies. Each time through the loop, we decrement 'foldlen' by
2807 * how many bytes the current char occupies. Only when it reaches
2808 * 0 do we increment 'minchars' or look for another multi-character
2809 * sequence. */
2810 if (folder == NULL) {
2811 minchars++;
2812 }
2813 else if (foldlen > 0) {
2814 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
645de4ce
KW
2815 }
2816 else {
bc031a7d
KW
2817 minchars++;
2818
2819 /* See if *uc is the beginning of a multi-character fold. If
2820 * so, we decrement the length remaining to look at, to account
2821 * for the current character this iteration. (We can use 'uc'
2822 * instead of the fold returned by TRIE_READ_CHAR because for
2823 * non-UTF, the latin1_safe macro is smart enough to account
2824 * for all the unfolded characters, and because for UTF, the
2825 * string will already have been folded earlier in the
2826 * compilation process */
2827 if (UTF) {
2828 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2829 foldlen -= UTF8SKIP(uc);
645de4ce
KW
2830 }
2831 }
bc031a7d
KW
2832 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2833 foldlen--;
2834 }
645de4ce 2835 }
bc031a7d
KW
2836
2837 /* The current character (and any potential folds) should be added
2838 * to the possible matching characters for this position in this
2839 * branch */
a3621e74 2840 if ( uvc < 256 ) {
fab2782b
YO
2841 if ( folder ) {
2842 U8 folded= folder[ (U8) uvc ];
2843 if ( !trie->charmap[ folded ] ) {
2844 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2845 TRIE_STORE_REVCHAR( folded );
2846 }
2847 }
a3621e74
YO
2848 if ( !trie->charmap[ uvc ] ) {
2849 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
fab2782b 2850 TRIE_STORE_REVCHAR( uvc );
a3621e74 2851 }
02daf0ab 2852 if ( set_bit ) {
62012aee
KW
2853 /* store the codepoint in the bitmap, and its folded
2854 * equivalent. */
8bcafbf4 2855 TRIE_BITMAP_SET_FOLDED(trie, uvc, folder);
02daf0ab
YO
2856 set_bit = 0; /* We've done our bit :-) */
2857 }
a3621e74 2858 } else {
bc031a7d
KW
2859
2860 /* XXX We could come up with the list of code points that fold
2861 * to this using PL_utf8_foldclosures, except not for
2862 * multi-char folds, as there may be multiple combinations
2863 * there that could work, which needs to wait until runtime to
2864 * resolve (The comment about LIGATURE FFI above is such an
2865 * example */
2866
a3621e74 2867 SV** svpp;
55eed653
NC
2868 if ( !widecharmap )
2869 widecharmap = newHV();
a3621e74 2870
55eed653 2871 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
2872
2873 if ( !svpp )
147e3846 2874 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%" UVXf, uvc );
a3621e74
YO
2875
2876 if ( !SvTRUE( *svpp ) ) {
2877 sv_setiv( *svpp, ++trie->uniquecharcount );
fab2782b 2878 TRIE_STORE_REVCHAR(uvc);
a3621e74
YO
2879 }
2880 }
bc031a7d
KW
2881 } /* end loop through characters in this branch of the trie */
2882
2883 /* We take the min and max for this branch and combine to find the min
2884 * and max for all branches processed so far */
3dab1dad 2885 if( cur == first ) {
bc031a7d
KW
2886 trie->minlen = minchars;
2887 trie->maxlen = maxchars;
2888 } else if (minchars < trie->minlen) {
2889 trie->minlen = minchars;
2890 } else if (maxchars > trie->maxlen) {
2891 trie->maxlen = maxchars;
fab2782b 2892 }
a3621e74
YO
2893 } /* end first pass */
2894 DEBUG_TRIE_COMPILE_r(
6ad9a8ab 2895 Perl_re_indentf( aTHX_
cb41e5d6
YO
2896 "TRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2897 depth+1,
55eed653 2898 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
2899 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2900 (int)trie->minlen, (int)trie->maxlen )
a3621e74 2901 );
a3621e74
YO
2902
2903 /*
2904 We now know what we are dealing with in terms of unique chars and
2905 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
2906 representation using a flat table will take. If it's over a reasonable
2907 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
2908 conservative but potentially much slower representation using an array
2909 of lists.
2910
2911 At the end we convert both representations into the same compressed
2912 form that will be used in regexec.c for matching with. The latter
2913 is a form that cannot be used to construct with but has memory
2914 properties similar to the list form and access properties similar
2915 to the table form making it both suitable for fast searches and
2916 small enough that its feasable to store for the duration of a program.
2917
2918 See the comment in the code where the compressed table is produced
2919 inplace from the flat tabe representation for an explanation of how
2920 the compression works.
2921
2922 */
2923
2924
2e64971a
DM
2925 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2926 prev_states[1] = 0;
2927
538e84ed
KW
2928 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2929 > SvIV(re_trie_maxbuff) )
2930 {
a3621e74
YO
2931 /*
2932 Second Pass -- Array Of Lists Representation
2933
2934 Each state will be represented by a list of charid:state records
2935 (reg_trie_trans_le) the first such element holds the CUR and LEN
2936 points of the allocated array. (See defines above).
2937
2938 We build the initial structure using the lists, and then convert
2939 it into the compressed table form which allows faster lookups
2940 (but cant be modified once converted).
a3621e74
YO
2941 */
2942
a3621e74
YO
2943 STRLEN transcount = 1;
2944
6ad9a8ab 2945 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using list compiler\n",
cb41e5d6 2946 depth+1));
686b73d4 2947
c944940b
JH
2948 trie->states = (reg_trie_state *)
2949 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2950 sizeof(reg_trie_state) );
a3621e74
YO
2951 TRIE_LIST_NEW(1);
2952 next_alloc = 2;
2953
2954 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2955
df826430 2956 regnode *noper = NEXTOPER( cur );
c445ea15
AL
2957 U32 state = 1; /* required init */
2958 U16 charid = 0; /* sanity init */
07be1b83 2959 U32 wordlen = 0; /* required init */
c445ea15 2960
df826430
YO
2961 if (OP(noper) == NOTHING) {
2962 regnode *noper_next= regnext(noper);
944e05e3
YO
2963 if (noper_next < tail)
2964 noper= noper_next;
df826430
YO
2965 }
2966
944e05e3
YO
2967 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
2968 const U8 *uc= (U8*)STRING(noper);
2969 const U8 *e= uc + STR_LEN(noper);
2970
786e8c11 2971 for ( ; uc < e ; uc += len ) {
c445ea15 2972
786e8c11 2973 TRIE_READ_CHAR;
c445ea15 2974
786e8c11
YO
2975 if ( uvc < 256 ) {
2976 charid = trie->charmap[ uvc ];
c445ea15 2977 } else {
538e84ed
KW
2978 SV** const svpp = hv_fetch( widecharmap,
2979 (char*)&uvc,
2980 sizeof( UV ),
2981 0);
786e8c11
YO
2982 if ( !svpp ) {
2983 charid = 0;
2984 } else {
2985 charid=(U16)SvIV( *svpp );
2986 }
c445ea15 2987 }
538e84ed
KW
2988 /* charid is now 0 if we dont know the char read, or
2989 * nonzero if we do */
786e8c11 2990 if ( charid ) {
a3621e74 2991
786e8c11
YO
2992 U16 check;
2993 U32 newstate = 0;
a3621e74 2994
786e8c11
YO
2995 charid--;
2996 if ( !trie->states[ state ].trans.list ) {
2997 TRIE_LIST_NEW( state );
c445ea15 2998 }
538e84ed
KW
2999 for ( check = 1;
3000 check <= TRIE_LIST_USED( state );
3001 check++ )
3002 {
3003 if ( TRIE_LIST_ITEM( state, check ).forid
3004 == charid )
3005 {
786e8c11
YO
3006 newstate = TRIE_LIST_ITEM( state, check ).newstate;
3007 break;
3008 }
3009 }
3010 if ( ! newstate ) {
3011 newstate = next_alloc++;
2e64971a 3012 prev_states[newstate] = state;
786e8c11
YO
3013 TRIE_LIST_PUSH( state, charid, newstate );
3014 transcount++;
3015 }
3016 state = newstate;
3017 } else {
147e3846 3018 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
c445ea15 3019 }
a28509cc 3020 }
c445ea15 3021 }
3dab1dad 3022 TRIE_HANDLE_WORD(state);
a3621e74
YO
3023
3024 } /* end second pass */
3025
1e2e3d02 3026 /* next alloc is the NEXT state to be allocated */
538e84ed 3027 trie->statecount = next_alloc;
c944940b
JH
3028 trie->states = (reg_trie_state *)
3029 PerlMemShared_realloc( trie->states,
3030 next_alloc
3031 * sizeof(reg_trie_state) );
a3621e74 3032
3dab1dad 3033 /* and now dump it out before we compress it */
2b8b4781
NC
3034 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
3035 revcharmap, next_alloc,
3036 depth+1)
1e2e3d02 3037 );
a3621e74 3038
c944940b
JH
3039 trie->trans = (reg_trie_trans *)
3040 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
3041 {
3042 U32 state;
a3621e74
YO
3043 U32 tp = 0;
3044 U32 zp = 0;
3045
3046
3047 for( state=1 ; state < next_alloc ; state ++ ) {
3048 U32 base=0;
3049
3050 /*
3051 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3052 Perl_re_printf( aTHX_ "tp: %d zp: %d ",tp,zp)
a3621e74
YO
3053 );
3054 */
3055
3056 if (trie->states[state].trans.list) {
3057 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
3058 U16 maxid=minid;
a28509cc 3059 U16 idx;
a3621e74
YO
3060
3061 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
3062 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
3063 if ( forid < minid ) {
3064 minid=forid;
3065 } else if ( forid > maxid ) {
3066 maxid=forid;
3067 }
a3621e74
YO
3068 }
3069 if ( transcount < tp + maxid - minid + 1) {
3070 transcount *= 2;
c944940b
JH
3071 trie->trans = (reg_trie_trans *)
3072 PerlMemShared_realloc( trie->trans,
446bd890
NC
3073 transcount
3074 * sizeof(reg_trie_trans) );
538e84ed
KW
3075 Zero( trie->trans + (transcount / 2),
3076 transcount / 2,
3077 reg_trie_trans );
a3621e74
YO
3078 }
3079 base = trie->uniquecharcount + tp - minid;
3080 if ( maxid == minid ) {
3081 U32 set = 0;
3082 for ( ; zp < tp ; zp++ ) {
3083 if ( ! trie->trans[ zp ].next ) {
3084 base = trie->uniquecharcount + zp - minid;
538e84ed
KW
3085 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
3086 1).newstate;
a3621e74
YO
3087 trie->trans[ zp ].check = state;
3088 set = 1;
3089 break;
3090 }
3091 }
3092 if ( !set ) {
538e84ed
KW
3093 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
3094 1).newstate;
a3621e74
YO
3095 trie->trans[ tp ].check = state;
3096 tp++;
3097 zp = tp;
3098 }
3099 } else {
3100 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
538e84ed
KW
3101 const U32 tid = base
3102 - trie->uniquecharcount
3103 + TRIE_LIST_ITEM( state, idx ).forid;
3104 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
3105 idx ).newstate;
a3621e74
YO
3106 trie->trans[ tid ].check = state;
3107 }
3108 tp += ( maxid - minid + 1 );
3109 }
3110 Safefree(trie->states[ state ].trans.list);
3111 }
3112 /*
3113 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3114 Perl_re_printf( aTHX_ " base: %d\n",base);
a3621e74
YO
3115 );
3116 */
3117 trie->states[ state ].trans.base=base;
3118 }
cc601c31 3119 trie->lasttrans = tp + 1;
a3621e74
YO
3120 }
3121 } else {
3122 /*
3123 Second Pass -- Flat Table Representation.
3124
b423522f
KW
3125 we dont use the 0 slot of either trans[] or states[] so we add 1 to
3126 each. We know that we will need Charcount+1 trans at most to store
3127 the data (one row per char at worst case) So we preallocate both
3128 structures assuming worst case.
a3621e74
YO
3129
3130 We then construct the trie using only the .next slots of the entry
3131 structs.
3132
b423522f
KW
3133 We use the .check field of the first entry of the node temporarily
3134 to make compression both faster and easier by keeping track of how
3135 many non zero fields are in the node.
a3621e74
YO
3136
3137 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
3138 transition.
3139
b423522f
KW
3140 There are two terms at use here: state as a TRIE_NODEIDX() which is
3141 a number representing the first entry of the node, and state as a
3142 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
3143 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
3144 if there are 2 entrys per node. eg:
a3621e74
YO
3145
3146 A B A B
3147 1. 2 4 1. 3 7
3148 2. 0 3 3. 0 5
3149 3. 0 0 5. 0 0
3150 4. 0 0 7. 0 0
3151
b423522f
KW
3152 The table is internally in the right hand, idx form. However as we
3153 also have to deal with the states array which is indexed by nodenum
3154 we have to use TRIE_NODENUM() to convert.
a3621e74
YO
3155
3156 */
6ad9a8ab 3157 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using table compiler\n",
cb41e5d6 3158 depth+1));
3dab1dad 3159
c944940b
JH
3160 trie->trans = (reg_trie_trans *)
3161 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
3162 * trie->uniquecharcount + 1,
3163 sizeof(reg_trie_trans) );
3164 trie->states = (reg_trie_state *)
3165 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
3166 sizeof(reg_trie_state) );
a3621e74
YO
3167 next_alloc = trie->uniquecharcount + 1;
3168
3dab1dad 3169
a3621e74
YO
3170 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
3171
df826430 3172 regnode *noper = NEXTOPER( cur );
a3621e74
YO
3173
3174 U32 state = 1; /* required init */
3175
3176 U16 charid = 0; /* sanity init */
3177 U32 accept_state = 0; /* sanity init */
a3621e74 3178
07be1b83 3179 U32 wordlen = 0; /* required init */
a3621e74 3180
df826430
YO
3181 if (OP(noper) == NOTHING) {
3182 regnode *noper_next= regnext(noper);
944e05e3
YO
3183 if (noper_next < tail)
3184 noper= noper_next;
df826430 3185 }
fab2782b 3186
944e05e3
YO
3187 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
3188 const U8 *uc= (U8*)STRING(noper);
3189 const U8 *e= uc + STR_LEN(noper);
3190
786e8c11 3191 for ( ; uc < e ; uc += len ) {
a3621e74 3192
786e8c11 3193 TRIE_READ_CHAR;
a3621e74 3194
786e8c11
YO
3195 if ( uvc < 256 ) {
3196 charid = trie->charmap[ uvc ];
3197 } else {
538e84ed
KW
3198 SV* const * const svpp = hv_fetch( widecharmap,
3199 (char*)&uvc,
3200 sizeof( UV ),
3201 0);
786e8c11 3202 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 3203 }
786e8c11
YO
3204 if ( charid ) {
3205 charid--;
3206 if ( !trie->trans[ state + charid ].next ) {
3207 trie->trans[ state + charid ].next = next_alloc;
3208 trie->trans[ state ].check++;
2e64971a
DM
3209 prev_states[TRIE_NODENUM(next_alloc)]
3210 = TRIE_NODENUM(state);
786e8c11
YO
3211 next_alloc += trie->uniquecharcount;
3212 }
3213 state = trie->trans[ state + charid ].next;
3214 } else {
147e3846 3215 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
786e8c11 3216 }
538e84ed
KW
3217 /* charid is now 0 if we dont know the char read, or
3218 * nonzero if we do */
a3621e74 3219 }
a3621e74 3220 }
3dab1dad
YO
3221 accept_state = TRIE_NODENUM( state );
3222 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
3223
3224 } /* end second pass */
3225
3dab1dad 3226 /* and now dump it out before we compress it */
2b8b4781
NC
3227 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
3228 revcharmap,
3229 next_alloc, depth+1));
a3621e74 3230
a3621e74
YO
3231 {
3232 /*
3233 * Inplace compress the table.*
3234
3235 For sparse data sets the table constructed by the trie algorithm will
3236 be mostly 0/FAIL transitions or to put it another way mostly empty.
3237 (Note that leaf nodes will not contain any transitions.)
3238
3239 This algorithm compresses the tables by eliminating most such
3240 transitions, at the cost of a modest bit of extra work during lookup:
3241
3242 - Each states[] entry contains a .base field which indicates the
3243 index in the state[] array wheres its transition data is stored.
3244
3b753521 3245 - If .base is 0 there are no valid transitions from that node.
a3621e74
YO
3246
3247 - If .base is nonzero then charid is added to it to find an entry in
3248 the trans array.
3249
3250 -If trans[states[state].base+charid].check!=state then the
3251 transition is taken to be a 0/Fail transition. Thus if there are fail
3252 transitions at the front of the node then the .base offset will point
3253 somewhere inside the previous nodes data (or maybe even into a node
3254 even earlier), but the .check field determines if the transition is
3255 valid.
3256
786e8c11 3257 XXX - wrong maybe?
a3621e74 3258 The following process inplace converts the table to the compressed
3b753521 3259 table: We first do not compress the root node 1,and mark all its
a3621e74 3260 .check pointers as 1 and set its .base pointer as 1 as well. This
3b753521
FN
3261 allows us to do a DFA construction from the compressed table later,
3262 and ensures that any .base pointers we calculate later are greater
3263 than 0.
a3621e74
YO
3264
3265 - We set 'pos' to indicate the first entry of the second node.
3266
3267 - We then iterate over the columns of the node, finding the first and
3268 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
3269 and set the .check pointers accordingly, and advance pos
3270 appropriately and repreat for the next node. Note that when we copy
3271 the next pointers we have to convert them from the original
3272 NODEIDX form to NODENUM form as the former is not valid post
3273 compression.
3274
3275 - If a node has no transitions used we mark its base as 0 and do not
3276 advance the pos pointer.
3277
3278 - If a node only has one transition we use a second pointer into the
3279 structure to fill in allocated fail transitions from other states.
3280 This pointer is independent of the main pointer and scans forward
3281 looking for null transitions that are allocated to a state. When it
3282 finds one it writes the single transition into the "hole". If the
786e8c11 3283 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
3284
3285 - Once compressed we can Renew/realloc the structures to release the
3286 excess space.
3287
3288 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
3289 specifically Fig 3.47 and the associated pseudocode.
3290
3291 demq
3292 */
a3b680e6 3293 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 3294 U32 state, charid;
a3621e74 3295 U32 pos = 0, zp=0;
1e2e3d02 3296 trie->statecount = laststate;
a3621e74
YO
3297
3298 for ( state = 1 ; state < laststate ; state++ ) {
3299 U8 flag = 0;
a28509cc
AL
3300 const U32 stateidx = TRIE_NODEIDX( state );
3301 const U32 o_used = trie->trans[ stateidx ].check;
3302 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
3303 trie->trans[ stateidx ].check = 0;
3304
538e84ed
KW
3305 for ( charid = 0;
3306 used && charid < trie->uniquecharcount;
3307 charid++ )
3308 {
a3621e74
YO
3309 if ( flag || trie->trans[ stateidx + charid ].next ) {
3310 if ( trie->trans[ stateidx + charid ].next ) {
3311 if (o_used == 1) {
3312 for ( ; zp < pos ; zp++ ) {
3313 if ( ! trie->trans[ zp ].next ) {
3314 break;
3315 }
3316 }
538e84ed
KW
3317 trie->states[ state ].trans.base
3318 = zp
3319 + trie->uniquecharcount
3320 - charid ;
3321 trie->trans[ zp ].next
3322 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
3323 + charid ].next );
a3621e74
YO
3324 trie->trans[ zp ].check = state;
3325 if ( ++zp > pos ) pos = zp;
3326 break;
3327 }
3328 used--;
3329 }
3330 if ( !flag ) {
3331 flag = 1;
538e84ed
KW
3332 trie->states[ state ].trans.base
3333 = pos + trie->uniquecharcount - charid ;
a3621e74 3334 }
538e84ed
KW
3335 trie->trans[ pos ].next
3336 = SAFE_TRIE_NODENUM(
3337 trie->trans[ stateidx + charid ].next );
a3621e74
YO
3338 trie->trans[ pos ].check = state;
3339 pos++;
3340 }
3341 }
3342 }
cc601c31 3343 trie->lasttrans = pos + 1;
c944940b
JH
3344 trie->states = (reg_trie_state *)
3345 PerlMemShared_realloc( trie->states, laststate
3346 * sizeof(reg_trie_state) );
a3621e74 3347 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3348 Perl_re_indentf( aTHX_ "Alloc: %d Orig: %" IVdf " elements, Final:%" IVdf ". Savings of %%%5.2f\n",
cb41e5d6 3349 depth+1,
538e84ed
KW
3350 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
3351 + 1 ),
3352 (IV)next_alloc,
3353 (IV)pos,
3354 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
a3621e74
YO
3355 );
3356
3357 } /* end table compress */
3358 }
1e2e3d02 3359 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3360 Perl_re_indentf( aTHX_ "Statecount:%" UVxf " Lasttrans:%" UVxf "\n",
cb41e5d6 3361 depth+1,
1e2e3d02
YO
3362 (UV)trie->statecount,
3363 (UV)trie->lasttrans)
3364 );
cc601c31 3365 /* resize the trans array to remove unused space */
c944940b
JH
3366 trie->trans = (reg_trie_trans *)
3367 PerlMemShared_realloc( trie->trans, trie->lasttrans
3368 * sizeof(reg_trie_trans) );
a3621e74 3369
538e84ed 3370 { /* Modify the program and insert the new TRIE node */
3dab1dad
YO
3371 U8 nodetype =(U8)(flags & 0xFF);
3372 char *str=NULL;
538e84ed 3373
07be1b83 3374#ifdef DEBUGGING
e62cc96a 3375 regnode *optimize = NULL;
7122b237
YO
3376#ifdef RE_TRACK_PATTERN_OFFSETS
3377
b57a0404
JH
3378 U32 mjd_offset = 0;
3379 U32 mjd_nodelen = 0;
7122b237
YO
3380#endif /* RE_TRACK_PATTERN_OFFSETS */
3381#endif /* DEBUGGING */
a3621e74 3382 /*
3dab1dad
YO
3383 This means we convert either the first branch or the first Exact,
3384 depending on whether the thing following (in 'last') is a branch
3385 or not and whther first is the startbranch (ie is it a sub part of
3386 the alternation or is it the whole thing.)
3b753521 3387 Assuming its a sub part we convert the EXACT otherwise we convert
3dab1dad 3388 the whole branch sequence, including the first.
a3621e74 3389 */
3dab1dad 3390 /* Find the node we are going to overwrite */
7f69552c 3391 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 3392 /* branch sub-chain */
3dab1dad 3393 NEXT_OFF( first ) = (U16)(last - first);
7122b237 3394#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
3395 DEBUG_r({
3396 mjd_offset= Node_Offset((convert));
3397 mjd_nodelen= Node_Length((convert));
3398 });
7122b237 3399#endif
7f69552c 3400 /* whole branch chain */
7122b237
YO
3401 }
3402#ifdef RE_TRACK_PATTERN_OFFSETS
3403 else {
7f69552c
YO
3404 DEBUG_r({
3405 const regnode *nop = NEXTOPER( convert );
3406 mjd_offset= Node_Offset((nop));
3407 mjd_nodelen= Node_Length((nop));
3408 });
07be1b83
YO
3409 }
3410 DEBUG_OPTIMISE_r(
147e3846 3411 Perl_re_indentf( aTHX_ "MJD offset:%" UVuf " MJD length:%" UVuf "\n",
cb41e5d6 3412 depth+1,
786e8c11 3413 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 3414 );
7122b237 3415#endif
538e84ed 3416 /* But first we check to see if there is a common prefix we can
3dab1dad
YO
3417 split out as an EXACT and put in front of the TRIE node. */
3418 trie->startstate= 1;
55eed653 3419 if ( trie->bitmap && !widecharmap && !trie->jump ) {
5ee57374
YO
3420 /* we want to find the first state that has more than
3421 * one transition, if that state is not the first state
3422 * then we have a common prefix which we can remove.
3423 */
3dab1dad 3424 U32 state;
1e2e3d02 3425 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
a3621e74 3426 U32 ofs = 0;
ee0dfd0b
YO
3427 I32 first_ofs = -1; /* keeps track of the ofs of the first
3428 transition, -1 means none */
8e11feef
RGS
3429 U32 count = 0;
3430 const U32 base = trie->states[ state ].trans.base;
a3621e74 3431
5ee57374 3432 /* does this state terminate an alternation? */
3dab1dad 3433 if ( trie->states[state].wordnum )
8e11feef 3434 count = 1;
a3621e74 3435
<