This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: Consolidate 2nd pass for warnings
[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 */
538e84ed
KW
142 regnode *emit_bound; /* First regnode outside of the
143 allocated space */
f55b7b26 144 regnode_offset emit; /* Code-emit pointer */
830247a4
IZ
145 I32 naughty; /* How bad is this pattern? */
146 I32 sawback; /* Did we see \1, ...? */
147 U32 seen;
88f063b4
KW
148 SSize_t size; /* Number of regnode equivalents in
149 pattern */
e7f023ff 150 I32 npar; /* Capture buffer count, (OPEN) plus
538e84ed
KW
151 one. ("par" 0 is the whole
152 pattern)*/
153 I32 nestroot; /* root parens we are in - used by
154 accept */
830247a4
IZ
155 I32 extralen;
156 I32 seen_zerolen;
f55b7b26
KW
157 regnode_offset *open_parens; /* offsets to open parens */
158 regnode_offset *close_parens; /* offsets to close parens */
d5a00e4a 159 regnode *end_op; /* END node in program */
02daf0ab
YO
160 I32 utf8; /* whether the pattern is utf8 or not */
161 I32 orig_utf8; /* whether the pattern was originally in utf8 */
162 /* XXX use this for future optimisation of case
163 * where pattern must be upgraded to utf8. */
e40e74fe
KW
164 I32 uni_semantics; /* If a d charset modifier should use unicode
165 rules, even if the pattern is not in
166 utf8 */
81714fb9 167 HV *paren_names; /* Paren names */
538e84ed 168
40d049e4 169 regnode **recurse; /* Recurse regops */
88f063b4 170 I32 recurse_count; /* Number of recurse regops we have generated */
4286711a 171 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
538e84ed 172 through */
09a65838 173 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
b57e4118 174 I32 in_lookbehind;
4624b182 175 I32 contains_locale;
bb3f3ed2 176 I32 override_recoding;
b6d67071
KW
177#ifdef EBCDIC
178 I32 recode_x_to_native;
179#endif
9d53c457 180 I32 in_multi_char_class;
1acab4c5 181 struct reg_code_blocks *code_blocks;/* positions of literal (?{})
68e2671b 182 within pattern */
b1603ef8 183 int code_index; /* next code_blocks[] slot */
ee273784 184 SSize_t maxlen; /* mininum possible number of chars in string to match */
3f910e62
YO
185 scan_frame *frame_head;
186 scan_frame *frame_last;
187 U32 frame_count;
7eec73eb 188 AV *warn_text;
dc6d7f5c 189#ifdef ADD_TO_REGEXEC
830247a4
IZ
190 char *starttry; /* -Dr: where regtry was called. */
191#define RExC_starttry (pRExC_state->starttry)
192#endif
d24ca0c5 193 SV *runtime_code_qr; /* qr with the runtime code blocks */
3dab1dad 194#ifdef DEBUGGING
be8e71aa 195 const char *lastparse;
3dab1dad 196 I32 lastnum;
1f1031fe 197 AV *paren_name_list; /* idx -> name */
d9a72fcc 198 U32 study_chunk_recursed_count;
c9f0d54c
YO
199 SV *mysv1;
200 SV *mysv2;
88f063b4 201
3dab1dad
YO
202#define RExC_lastparse (pRExC_state->lastparse)
203#define RExC_lastnum (pRExC_state->lastnum)
1f1031fe 204#define RExC_paren_name_list (pRExC_state->paren_name_list)
d9a72fcc 205#define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
c9f0d54c
YO
206#define RExC_mysv (pRExC_state->mysv1)
207#define RExC_mysv1 (pRExC_state->mysv1)
208#define RExC_mysv2 (pRExC_state->mysv2)
209
3dab1dad 210#endif
512c0f5a 211 bool seen_unfolded_sharp_s;
911bd04e 212 bool strict;
da7cf1cc 213 bool study_started;
034602eb 214 bool in_script_run;
f55b7b26 215 bool pass1;
09b2b2e6 216};
830247a4 217
e2509266 218#define RExC_flags (pRExC_state->flags)
514a91f1 219#define RExC_pm_flags (pRExC_state->pm_flags)
830247a4 220#define RExC_precomp (pRExC_state->precomp)
51684e3c
KW
221#define RExC_copy_start_in_input (pRExC_state->copy_start_in_input)
222#define RExC_copy_start_in_constructed (pRExC_state->copy_start)
711b303b 223#define RExC_precomp_end (pRExC_state->precomp_end)
288b8c02 224#define RExC_rx_sv (pRExC_state->rx_sv)
830247a4 225#define RExC_rx (pRExC_state->rx)
f8fc2ecf 226#define RExC_rxi (pRExC_state->rxi)
fac92740 227#define RExC_start (pRExC_state->start)
830247a4
IZ
228#define RExC_end (pRExC_state->end)
229#define RExC_parse (pRExC_state->parse)
230#define RExC_whilem_seen (pRExC_state->whilem_seen)
512c0f5a
KW
231
232/* Set during the sizing pass when there is a LATIN SMALL LETTER SHARP S in any
233 * EXACTF node, hence was parsed under /di rules. If later in the parse,
234 * something forces the pattern into using /ui rules, the sharp s should be
235 * folded into the sequence 'ss', which takes up more space than previously
236 * calculated. This means that the sizing pass needs to be restarted. (The
237 * node also becomes an EXACTFU_SS.) For all other characters, an EXACTF node
238 * that gets converted to /ui (and EXACTFU) occupies the same amount of space,
239 * so there is no need to resize [perl #125990]. */
240#define RExC_seen_unfolded_sharp_s (pRExC_state->seen_unfolded_sharp_s)
241
7122b237 242#ifdef RE_TRACK_PATTERN_OFFSETS
44f36cd3 243# define RExC_offsets (RExC_rxi->u.offsets) /* I am not like the
538e84ed 244 others */
7122b237 245#endif
830247a4 246#define RExC_emit (pRExC_state->emit)
f55b7b26 247#define RExC_pass1 (pRExC_state->pass1)
fac92740 248#define RExC_emit_start (pRExC_state->emit_start)
3b57cd43 249#define RExC_emit_bound (pRExC_state->emit_bound)
830247a4
IZ
250#define RExC_sawback (pRExC_state->sawback)
251#define RExC_seen (pRExC_state->seen)
252#define RExC_size (pRExC_state->size)
ee273784 253#define RExC_maxlen (pRExC_state->maxlen)
830247a4 254#define RExC_npar (pRExC_state->npar)
e2e6a0f1 255#define RExC_nestroot (pRExC_state->nestroot)
830247a4
IZ
256#define RExC_extralen (pRExC_state->extralen)
257#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
1aa99e6b 258#define RExC_utf8 (pRExC_state->utf8)
e40e74fe 259#define RExC_uni_semantics (pRExC_state->uni_semantics)
02daf0ab 260#define RExC_orig_utf8 (pRExC_state->orig_utf8)
40d049e4
YO
261#define RExC_open_parens (pRExC_state->open_parens)
262#define RExC_close_parens (pRExC_state->close_parens)
5bd2d46e 263#define RExC_end_op (pRExC_state->end_op)
81714fb9 264#define RExC_paren_names (pRExC_state->paren_names)
40d049e4
YO
265#define RExC_recurse (pRExC_state->recurse)
266#define RExC_recurse_count (pRExC_state->recurse_count)
09a65838 267#define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
538e84ed
KW
268#define RExC_study_chunk_recursed_bytes \
269 (pRExC_state->study_chunk_recursed_bytes)
b57e4118 270#define RExC_in_lookbehind (pRExC_state->in_lookbehind)
4624b182 271#define RExC_contains_locale (pRExC_state->contains_locale)
b6d67071
KW
272#ifdef EBCDIC
273# define RExC_recode_x_to_native (pRExC_state->recode_x_to_native)
274#endif
9d53c457 275#define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
3f910e62
YO
276#define RExC_frame_head (pRExC_state->frame_head)
277#define RExC_frame_last (pRExC_state->frame_last)
278#define RExC_frame_count (pRExC_state->frame_count)
67cdf558 279#define RExC_strict (pRExC_state->strict)
da7cf1cc 280#define RExC_study_started (pRExC_state->study_started)
7eec73eb 281#define RExC_warn_text (pRExC_state->warn_text)
034602eb 282#define RExC_in_script_run (pRExC_state->in_script_run)
830247a4 283
99807a43
HS
284/* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
285 * a flag to disable back-off on the fixed/floating substrings - if it's
286 * a high complexity pattern we assume the benefit of avoiding a full match
287 * is worth the cost of checking for the substrings even if they rarely help.
288 */
289#define RExC_naughty (pRExC_state->naughty)
290#define TOO_NAUGHTY (10)
291#define MARK_NAUGHTY(add) \
292 if (RExC_naughty < TOO_NAUGHTY) \
293 RExC_naughty += (add)
294#define MARK_NAUGHTY_EXP(exp, add) \
295 if (RExC_naughty < TOO_NAUGHTY) \
296 RExC_naughty += RExC_naughty / (exp) + (add)
cde0cee5 297
a687059c
LW
298#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
299#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
412f55bb 300 ((*s) == '{' && regcurly(s)))
a687059c
LW
301
302/*
303 * Flags to be passed up and down.
304 */
a687059c 305#define WORST 0 /* Worst case. */
a3b492c3 306#define HASWIDTH 0x01 /* Known to match non-null strings. */
fda99bee 307
e64f369d 308/* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
2fd92675
KW
309 * character. (There needs to be a case: in the switch statement in regexec.c
310 * for any node marked SIMPLE.) Note that this is not the same thing as
311 * REGNODE_SIMPLE */
fda99bee 312#define SIMPLE 0x02
e64f369d 313#define SPSTART 0x04 /* Starts with * or + */
8d9c2815
NC
314#define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
315#define TRYAGAIN 0x10 /* Weeded out a declaration. */
c36361b2
KW
316#define RESTART_PARSE 0x20 /* Need to redo the parse */
317#define NEED_UTF8 0x40 /* In conjunction with RESTART_PARSE, need to
318 calcuate sizes as UTF-8 */
a687059c 319
3dab1dad
YO
320#define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
321
07be1b83
YO
322/* whether trie related optimizations are enabled */
323#if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
324#define TRIE_STUDY_OPT
786e8c11 325#define FULL_TRIE_STUDY
07be1b83
YO
326#define TRIE_STCLASS
327#endif
1de06328
YO
328
329
40d049e4
YO
330
331#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
332#define PBITVAL(paren) (1 << ((paren) & 7))
333#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
334#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
335#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
336
82a6ada4 337#define REQUIRE_UTF8(flagp) STMT_START { \
8d9c2815 338 if (!UTF) { \
82a6ada4 339 assert(PASS1); \
c36361b2 340 *flagp = RESTART_PARSE|NEED_UTF8; \
f55b7b26 341 return 0; \
8d9c2815 342 } \
82a6ada4 343 } STMT_END
40d049e4 344
512c0f5a
KW
345/* Change from /d into /u rules, and restart the parse if we've already seen
346 * something whose size would increase as a result, by setting *flagp and
347 * returning 'restart_retval'. RExC_uni_semantics is a flag that indicates
142ece99 348 * we've changed to /u during the parse. */
512c0f5a
KW
349#define REQUIRE_UNI_RULES(flagp, restart_retval) \
350 STMT_START { \
351 if (DEPENDS_SEMANTICS) { \
352 assert(PASS1); \
353 set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); \
354 RExC_uni_semantics = 1; \
355 if (RExC_seen_unfolded_sharp_s) { \
c36361b2 356 *flagp |= RESTART_PARSE; \
512c0f5a
KW
357 return restart_retval; \
358 } \
359 } \
360 } STMT_END
361
fd7fc6ae 362/* Executes a return statement with the value 'X', if 'flags' contains any of
c36361b2 363 * 'RESTART_PARSE', 'NEED_UTF8', or 'extra'. If so, *flagp is set to those
fd7fc6ae
YO
364 * flags */
365#define RETURN_X_ON_RESTART_OR_FLAGS(X, flags, flagp, extra) \
366 STMT_START { \
c36361b2
KW
367 if ((flags) & (RESTART_PARSE|NEED_UTF8|(extra))) { \
368 *(flagp) = (flags) & (RESTART_PARSE|NEED_UTF8|(extra)); \
fd7fc6ae
YO
369 return X; \
370 } \
371 } STMT_END
372
5d555bb8 373#define RETURN_FAIL_ON_RESTART_OR_FLAGS(flags,flagp,extra) \
f55b7b26 374 RETURN_X_ON_RESTART_OR_FLAGS(0,flags,flagp,extra)
fd7fc6ae
YO
375
376#define RETURN_X_ON_RESTART(X, flags,flagp) \
377 RETURN_X_ON_RESTART_OR_FLAGS( X, flags, flagp, 0)
378
379
5d555bb8 380#define RETURN_FAIL_ON_RESTART_FLAGP_OR_FLAGS(flagp,extra) \
c36361b2 381 if (*(flagp) & (RESTART_PARSE|(extra))) return 0
fd7fc6ae 382
c36361b2 383#define MUST_RESTART(flags) ((flags) & (RESTART_PARSE))
fd7fc6ae 384
5d555bb8 385#define RETURN_FAIL_ON_RESTART(flags,flagp) \
f55b7b26 386 RETURN_X_ON_RESTART(0, flags,flagp)
5d555bb8 387#define RETURN_FAIL_ON_RESTART_FLAGP(flagp) \
88f063b4 388 RETURN_FAIL_ON_RESTART_FLAGP_OR_FLAGS(flagp, 0)
fd7fc6ae 389
f19b1a63
KW
390/* This converts the named class defined in regcomp.h to its equivalent class
391 * number defined in handy.h. */
392#define namedclass_to_classnum(class) ((int) ((class) / 2))
393#define classnum_to_namedclass(classnum) ((classnum) * 2)
394
de92f5e6
KW
395#define _invlist_union_complement_2nd(a, b, output) \
396 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
397#define _invlist_intersection_complement_2nd(a, b, output) \
398 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
399
1de06328
YO
400/* About scan_data_t.
401
402 During optimisation we recurse through the regexp program performing
403 various inplace (keyhole style) optimisations. In addition study_chunk
404 and scan_commit populate this data structure with information about
538e84ed 405 what strings MUST appear in the pattern. We look for the longest
3b753521 406 string that must appear at a fixed location, and we look for the
1de06328
YO
407 longest string that may appear at a floating location. So for instance
408 in the pattern:
538e84ed 409
1de06328 410 /FOO[xX]A.*B[xX]BAR/
538e84ed 411
1de06328
YO
412 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
413 strings (because they follow a .* construct). study_chunk will identify
414 both FOO and BAR as being the longest fixed and floating strings respectively.
538e84ed 415
1de06328 416 The strings can be composites, for instance
538e84ed 417
1de06328 418 /(f)(o)(o)/
538e84ed 419
1de06328 420 will result in a composite fixed substring 'foo'.
538e84ed 421
1de06328 422 For each string some basic information is maintained:
538e84ed 423
6ea77169 424 - min_offset
1de06328
YO
425 This is the position the string must appear at, or not before.
426 It also implicitly (when combined with minlenp) tells us how many
3b753521
FN
427 characters must match before the string we are searching for.
428 Likewise when combined with minlenp and the length of the string it
538e84ed 429 tells us how many characters must appear after the string we have
1de06328 430 found.
538e84ed 431
1de06328
YO
432 - max_offset
433 Only used for floating strings. This is the rightmost point that
ea3daa5d 434 the string can appear at. If set to SSize_t_MAX it indicates that the
1de06328 435 string can occur infinitely far to the right.
6e12f832 436 For fixed strings, it is equal to min_offset.
538e84ed 437
1de06328 438 - minlenp
2d608413
KW
439 A pointer to the minimum number of characters of the pattern that the
440 string was found inside. This is important as in the case of positive
538e84ed 441 lookahead or positive lookbehind we can have multiple patterns
1de06328 442 involved. Consider
538e84ed 443
1de06328 444 /(?=FOO).*F/
538e84ed 445
1de06328
YO
446 The minimum length of the pattern overall is 3, the minimum length
447 of the lookahead part is 3, but the minimum length of the part that
538e84ed 448 will actually match is 1. So 'FOO's minimum length is 3, but the
1de06328 449 minimum length for the F is 1. This is important as the minimum length
538e84ed 450 is used to determine offsets in front of and behind the string being
1de06328 451 looked for. Since strings can be composites this is the length of the
486ec47a 452 pattern at the time it was committed with a scan_commit. Note that
1de06328 453 the length is calculated by study_chunk, so that the minimum lengths
538e84ed 454 are not known until the full pattern has been compiled, thus the
1de06328 455 pointer to the value.
538e84ed 456
1de06328 457 - lookbehind
538e84ed 458
1de06328 459 In the case of lookbehind the string being searched for can be
538e84ed 460 offset past the start point of the final matching string.
1de06328
YO
461 If this value was just blithely removed from the min_offset it would
462 invalidate some of the calculations for how many chars must match
463 before or after (as they are derived from min_offset and minlen and
538e84ed 464 the length of the string being searched for).
1de06328
YO
465 When the final pattern is compiled and the data is moved from the
466 scan_data_t structure into the regexp structure the information
538e84ed
KW
467 about lookbehind is factored in, with the information that would
468 have been lost precalculated in the end_shift field for the
1de06328
YO
469 associated string.
470
471 The fields pos_min and pos_delta are used to store the minimum offset
538e84ed 472 and the delta to the maximum offset at the current point in the pattern.
1de06328
YO
473
474*/
2c2d71f5 475
cb6aef8b
DM
476struct scan_data_substrs {
477 SV *str; /* longest substring found in pattern */
478 SSize_t min_offset; /* earliest point in string it can appear */
479 SSize_t max_offset; /* latest point in string it can appear */
480 SSize_t *minlenp; /* pointer to the minlen relevant to the string */
481 SSize_t lookbehind; /* is the pos of the string modified by LB */
11683ecb 482 I32 flags; /* per substring SF_* and SCF_* flags */
cb6aef8b
DM
483};
484
2c2d71f5 485typedef struct scan_data_t {
1de06328
YO
486 /*I32 len_min; unused */
487 /*I32 len_delta; unused */
49f55535 488 SSize_t pos_min;
ea3daa5d 489 SSize_t pos_delta;
2c2d71f5 490 SV *last_found;
ea3daa5d 491 SSize_t last_end; /* min value, <0 unless valid. */
49f55535 492 SSize_t last_start_min;
ea3daa5d 493 SSize_t last_start_max;
2099df82
DM
494 U8 cur_is_floating; /* whether the last_* values should be set as
495 * the next fixed (0) or floating (1)
496 * substring */
6ea77169 497
2099df82
DM
498 /* [0] is longest fixed substring so far, [1] is longest float so far */
499 struct scan_data_substrs substrs[2];
6ea77169 500
11683ecb 501 I32 flags; /* common SF_* and SCF_* flags */
2c2d71f5 502 I32 whilem_c;
49f55535 503 SSize_t *last_closep;
b8f7bb16 504 regnode_ssc *start_class;
2c2d71f5
JH
505} scan_data_t;
506
a687059c 507/*
e50aee73 508 * Forward declarations for pregcomp()'s friends.
a687059c 509 */
a0d0e21e 510
6ea77169 511static const scan_data_t zero_scan_data = {
a2ca2017 512 0, 0, NULL, 0, 0, 0, 0,
6ea77169 513 {
11683ecb
DM
514 { NULL, 0, 0, 0, 0, 0 },
515 { NULL, 0, 0, 0, 0, 0 },
6ea77169
DM
516 },
517 0, 0, NULL, NULL
518};
c277df42 519
11683ecb
DM
520/* study flags */
521
07be1b83
YO
522#define SF_BEFORE_SEOL 0x0001
523#define SF_BEFORE_MEOL 0x0002
11683ecb 524#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
c277df42 525
07be1b83
YO
526#define SF_IS_INF 0x0040
527#define SF_HAS_PAR 0x0080
528#define SF_IN_PAR 0x0100
529#define SF_HAS_EVAL 0x0200
bc604ad8
KW
530
531
532/* SCF_DO_SUBSTR is the flag that tells the regexp analyzer to track the
533 * longest substring in the pattern. When it is not set the optimiser keeps
534 * track of position, but does not keep track of the actual strings seen,
535 *
536 * So for instance /foo/ will be parsed with SCF_DO_SUBSTR being true, but
537 * /foo/i will not.
538 *
539 * Similarly, /foo.*(blah|erm|huh).*fnorble/ will have "foo" and "fnorble"
540 * parsed with SCF_DO_SUBSTR on, but while processing the (...) it will be
541 * turned off because of the alternation (BRANCH). */
07be1b83 542#define SCF_DO_SUBSTR 0x0400
bc604ad8 543
653099ff
GS
544#define SCF_DO_STCLASS_AND 0x0800
545#define SCF_DO_STCLASS_OR 0x1000
546#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
e1901655 547#define SCF_WHILEM_VISITED_POS 0x2000
c277df42 548
786e8c11 549#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
538e84ed 550#define SCF_SEEN_ACCEPT 0x8000
688e0391 551#define SCF_TRIE_DOING_RESTUDY 0x10000
4286711a
YO
552#define SCF_IN_DEFINE 0x20000
553
554
555
07be1b83 556
43fead97 557#define UTF cBOOL(RExC_utf8)
00b27cfc
KW
558
559/* The enums for all these are ordered so things work out correctly */
a62b1201 560#define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
538e84ed
KW
561#define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
562 == REGEX_DEPENDS_CHARSET)
00b27cfc 563#define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
538e84ed
KW
564#define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
565 >= REGEX_UNICODE_CHARSET)
566#define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
567 == REGEX_ASCII_RESTRICTED_CHARSET)
568#define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
569 >= REGEX_ASCII_RESTRICTED_CHARSET)
570#define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
571 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
a62b1201 572
43fead97 573#define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
a0ed51b3 574
f2c2a6ab
KW
575/* For programs that want to be strictly Unicode compatible by dying if any
576 * attempt is made to match a non-Unicode code point against a Unicode
577 * property. */
578#define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
579
93733859 580#define OOB_NAMEDCLASS -1
b8c5462f 581
8e661ac5
KW
582/* There is no code point that is out-of-bounds, so this is problematic. But
583 * its only current use is to initialize a variable that is always set before
584 * looked at. */
585#define OOB_UNICODE 0xDEADBEEF
586
a0ed51b3 587#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
a0ed51b3 588
8615cb43 589
b45f050a
JF
590/* length of regex to show in messages that don't mark a position within */
591#define RegexLengthToShowInErrorMessages 127
592
593/*
594 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
595 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
596 * op/pragma/warn/regcomp.
597 */
7253e4e3
RK
598#define MARKER1 "<-- HERE" /* marker as it appears in the description */
599#define MARKER2 " <-- HERE " /* marker as it appears within the regex */
b81d288d 600
538e84ed 601#define REPORT_LOCATION " in regex; marked by " MARKER1 \
147e3846 602 " in m/%" UTF8f MARKER2 "%" UTF8f "/"
b45f050a 603
285b5ca0
KW
604/* The code in this file in places uses one level of recursion with parsing
605 * rebased to an alternate string constructed by us in memory. This can take
606 * the form of something that is completely different from the input, or
607 * something that uses the input as part of the alternate. In the first case,
608 * there should be no possibility of an error, as we are in complete control of
51684e3c
KW
609 * the alternate string. But in the second case we don't completely control
610 * the input portion, so there may be errors in that. Here's an example:
285b5ca0
KW
611 * /[abc\x{DF}def]/ui
612 * is handled specially because \x{df} folds to a sequence of more than one
51684e3c 613 * character: 'ss'. What is done is to create and parse an alternate string,
285b5ca0
KW
614 * which looks like this:
615 * /(?:\x{DF}|[abc\x{DF}def])/ui
616 * where it uses the input unchanged in the middle of something it constructs,
617 * which is a branch for the DF outside the character class, and clustering
618 * parens around the whole thing. (It knows enough to skip the DF inside the
619 * class while in this substitute parse.) 'abc' and 'def' may have errors that
620 * need to be reported. The general situation looks like this:
621 *
51684e3c 622 * |<------- identical ------>|
285b5ca0 623 * sI tI xI eI
51684e3c 624 * Input: ---------------------------------------------------------------
285b5ca0
KW
625 * Constructed: ---------------------------------------------------
626 * sC tC xC eC EC
51684e3c 627 * |<------- identical ------>|
285b5ca0 628 *
51684e3c
KW
629 * sI..eI is the portion of the input pattern we are concerned with here.
630 * sC..EC is the constructed substitute parse string.
631 * sC..tC is constructed by us
632 * tC..eC is an exact duplicate of the portion of the input pattern tI..eI.
633 * In the diagram, these are vertically aligned.
634 * eC..EC is also constructed by us.
635 * xC is the position in the substitute parse string where we found a
636 * problem.
637 * xI is the position in the original pattern corresponding to xC.
285b5ca0 638 *
51684e3c
KW
639 * We want to display a message showing the real input string. Thus we need to
640 * translate from xC to xI. We know that xC >= tC, since the portion of the
641 * string sC..tC has been constructed by us, and so shouldn't have errors. We
642 * get:
643 * xI = tI + (xC - tC)
285b5ca0 644 *
51684e3c
KW
645 * When the substitute parse is constructed, the code needs to set:
646 * RExC_start (sC)
647 * RExC_end (eC)
648 * RExC_copy_start_in_input (tI)
649 * RExC_copy_start_in_constructed (tC)
650 * and restore them when done.
285b5ca0 651 *
51684e3c
KW
652 * During normal processing of the input pattern, both
653 * 'RExC_copy_start_in_input' and 'RExC_copy_start_in_constructed' are set to
654 * sI, so that xC equals xI.
285b5ca0
KW
655 */
656
51684e3c
KW
657#define sI RExC_precomp
658#define eI RExC_precomp_end
659#define sC RExC_start
660#define eC RExC_end
661#define tI RExC_copy_start_in_input
662#define tC RExC_copy_start_in_constructed
663#define xI(xC) (tI + (xC - tC))
664#define xI_offset(xC) (xI(xC) - sI)
285b5ca0
KW
665
666#define REPORT_LOCATION_ARGS(xC) \
667 UTF8fARG(UTF, \
51684e3c 668 (xI(xC) > eI) /* Don't run off end */ \
285b5ca0 669 ? eC - sC /* Length before the <--HERE */ \
bc929b61
KW
670 : ((xI_offset(xC) >= 0) \
671 ? xI_offset(xC) \
672 : (Perl_croak(aTHX_ "panic: %s: %d: negative offset: %" \
673 IVdf " trying to output message for " \
674 " pattern %.*s", \
51684e3c 675 __FILE__, __LINE__, (IV) xI_offset(xC), \
bc929b61 676 ((int) (eC - sC)), sC), 0)), \
51684e3c 677 sI), /* The input pattern printed up to the <--HERE */ \
285b5ca0 678 UTF8fARG(UTF, \
51684e3c
KW
679 (xI(xC) > eI) ? 0 : eI - xI(xC), /* Length after <--HERE */ \
680 (xI(xC) > eI) ? eI : xI(xC)) /* pattern after <--HERE */
c1d900c3 681
8a6d8ec6
HS
682/* Used to point after bad bytes for an error message, but avoid skipping
683 * past a nul byte. */
684#define SKIP_IF_CHAR(s) (!*(s) ? 0 : UTF ? UTF8SKIP(s) : 1)
685
b45f050a
JF
686/*
687 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
688 * arg. Show regex, up to a maximum length. If it's too long, chop and add
689 * "...".
690 */
58e23c8d 691#define _FAIL(code) STMT_START { \
bfed75c6 692 const char *ellipses = ""; \
88f063b4 693 IV len = RExC_precomp_end - RExC_precomp; \
ccb2c380
MP
694 \
695 if (!SIZE_ONLY) \
a5e7bc51 696 SAVEFREESV(RExC_rx_sv); \
ccb2c380
MP
697 if (len > RegexLengthToShowInErrorMessages) { \
698 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
699 len = RegexLengthToShowInErrorMessages - 10; \
700 ellipses = "..."; \
701 } \
58e23c8d 702 code; \
ccb2c380 703} STMT_END
8615cb43 704
58e23c8d 705#define FAIL(msg) _FAIL( \
147e3846 706 Perl_croak(aTHX_ "%s in regex m/%" UTF8f "%s/", \
c1d900c3 707 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d
YO
708
709#define FAIL2(msg,arg) _FAIL( \
147e3846 710 Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \
c1d900c3 711 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d 712
b45f050a 713/*
b45f050a
JF
714 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
715 */
ccb2c380 716#define Simple_vFAIL(m) STMT_START { \
ccb2c380 717 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
d528642a 718 m, REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 719} STMT_END
b45f050a
JF
720
721/*
722 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
723 */
ccb2c380
MP
724#define vFAIL(m) STMT_START { \
725 if (!SIZE_ONLY) \
a5e7bc51 726 SAVEFREESV(RExC_rx_sv); \
ccb2c380
MP
727 Simple_vFAIL(m); \
728} STMT_END
b45f050a
JF
729
730/*
731 * Like Simple_vFAIL(), but accepts two arguments.
732 */
ccb2c380 733#define Simple_vFAIL2(m,a1) STMT_START { \
d528642a
KW
734 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
735 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 736} STMT_END
b45f050a
JF
737
738/*
739 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
740 */
ccb2c380
MP
741#define vFAIL2(m,a1) STMT_START { \
742 if (!SIZE_ONLY) \
a5e7bc51 743 SAVEFREESV(RExC_rx_sv); \
ccb2c380
MP
744 Simple_vFAIL2(m, a1); \
745} STMT_END
b45f050a
JF
746
747
748/*
749 * Like Simple_vFAIL(), but accepts three arguments.
750 */
ccb2c380 751#define Simple_vFAIL3(m, a1, a2) STMT_START { \
c1d900c3 752 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 753 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 754} STMT_END
b45f050a
JF
755
756/*
757 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
758 */
ccb2c380
MP
759#define vFAIL3(m,a1,a2) STMT_START { \
760 if (!SIZE_ONLY) \
a5e7bc51 761 SAVEFREESV(RExC_rx_sv); \
ccb2c380
MP
762 Simple_vFAIL3(m, a1, a2); \
763} STMT_END
b45f050a
JF
764
765/*
766 * Like Simple_vFAIL(), but accepts four arguments.
767 */
ccb2c380 768#define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
d528642a
KW
769 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
770 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 771} STMT_END
b45f050a 772
95db3ffa
KW
773#define vFAIL4(m,a1,a2,a3) STMT_START { \
774 if (!SIZE_ONLY) \
775 SAVEFREESV(RExC_rx_sv); \
776 Simple_vFAIL4(m, a1, a2, a3); \
777} STMT_END
778
946095af 779/* A specialized version of vFAIL2 that works with UTF8f */
d528642a
KW
780#define vFAIL2utf8f(m, a1) STMT_START { \
781 if (!SIZE_ONLY) \
782 SAVEFREESV(RExC_rx_sv); \
783 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
784 REPORT_LOCATION_ARGS(RExC_parse)); \
946095af
BF
785} STMT_END
786
3ba22297 787#define vFAIL3utf8f(m, a1, a2) STMT_START { \
3ba22297
KW
788 if (!SIZE_ONLY) \
789 SAVEFREESV(RExC_rx_sv); \
790 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 791 REPORT_LOCATION_ARGS(RExC_parse)); \
3ba22297
KW
792} STMT_END
793
338190a1
KW
794/* Outputting warnings is generally deferred until the 2nd pass. This is
795 * because the first pass can be restarted, for example if the pattern has to
796 * be converted to UTF-8. If a warning had already been output earlier in the
797 * pass, it would be re-output after the restart. Pass 2 is never restarted,
798 * so the problem simply goes away if we defer the output to that pass. See
799 * [perl #122671]. */
f8220fe2
KW
800#define _WARN_HELPER(loc, warns, code) \
801 STMT_START { \
338190a1
KW
802 if (PASS2) { \
803 code; \
804 } \
f8220fe2 805 } STMT_END
946095af 806
5e0a247b 807/* m is not necessarily a "literal string", in this macro */
f8220fe2
KW
808#define reg_warn_non_literal_string(loc, m) \
809 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
810 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 811 "%s" REPORT_LOCATION, \
f8220fe2 812 m, REPORT_LOCATION_ARGS(loc)))
5e0a247b 813
f8220fe2
KW
814#define ckWARNreg(loc,m) \
815 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
816 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 817 m REPORT_LOCATION, \
f8220fe2 818 REPORT_LOCATION_ARGS(loc)))
ccb2c380 819
f8220fe2
KW
820#define vWARN(loc, m) \
821 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
822 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 823 m REPORT_LOCATION, \
f8220fe2 824 REPORT_LOCATION_ARGS(loc))) \
b927b7e9 825
f8220fe2
KW
826#define vWARN_dep(loc, m) \
827 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
828 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 829 m REPORT_LOCATION, \
f8220fe2 830 REPORT_LOCATION_ARGS(loc)))
0d6106aa 831
f8220fe2
KW
832#define ckWARNdep(loc,m) \
833 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
834 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 835 m REPORT_LOCATION, \
f8220fe2 836 REPORT_LOCATION_ARGS(loc)))
147508a2 837
f8220fe2
KW
838#define ckWARNregdep(loc,m) \
839 _WARN_HELPER(loc, packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
840 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, \
d528642a
KW
841 WARN_REGEXP), \
842 m REPORT_LOCATION, \
f8220fe2 843 REPORT_LOCATION_ARGS(loc)))
ccb2c380 844
f8220fe2
KW
845#define ckWARN2reg_d(loc,m, a1) \
846 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
847 Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
d528642a 848 m REPORT_LOCATION, \
f8220fe2 849 a1, REPORT_LOCATION_ARGS(loc)))
2335b3d3 850
f8220fe2
KW
851#define ckWARN2reg(loc, m, a1) \
852 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
853 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 854 m REPORT_LOCATION, \
f8220fe2 855 a1, REPORT_LOCATION_ARGS(loc)))
ccb2c380 856
f8220fe2
KW
857#define vWARN3(loc, m, a1, a2) \
858 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
859 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 860 m REPORT_LOCATION, \
f8220fe2 861 a1, a2, REPORT_LOCATION_ARGS(loc)))
ccb2c380 862
f8220fe2
KW
863#define ckWARN3reg(loc, m, a1, a2) \
864 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
865 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
866 m REPORT_LOCATION, \
867 a1, a2, \
f8220fe2 868 REPORT_LOCATION_ARGS(loc)))
668c081a 869
f8220fe2
KW
870#define vWARN4(loc, m, a1, a2, a3) \
871 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
872 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
873 m REPORT_LOCATION, \
874 a1, a2, a3, \
f8220fe2 875 REPORT_LOCATION_ARGS(loc)))
ccb2c380 876
f8220fe2
KW
877#define ckWARN4reg(loc, m, a1, a2, a3) \
878 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
879 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
880 m REPORT_LOCATION, \
881 a1, a2, a3, \
f8220fe2 882 REPORT_LOCATION_ARGS(loc)))
668c081a 883
f8220fe2
KW
884#define vWARN5(loc, m, a1, a2, a3, a4) \
885 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
886 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
887 m REPORT_LOCATION, \
888 a1, a2, a3, a4, \
f8220fe2 889 REPORT_LOCATION_ARGS(loc)))
9d1d55b5 890
aea5f181
KW
891#define ckWARNexperimental(loc, class, m) \
892 _WARN_HELPER(loc, packWARN(class), \
893 Perl_ck_warner_d(aTHX_ packWARN(class), \
894 m REPORT_LOCATION, \
895 REPORT_LOCATION_ARGS(loc)))
896
1a322bb4
KW
897/* Convert between a pointer to a node and its offset from the beginning of the
898 * program */
899#define REGNODE_p(offset) (RExC_emit_start + (offset))
900#define REGNODE_OFFSET(node) ((node) - RExC_emit_start)
901
538e84ed 902/* Macros for recording node offsets. 20001227 mjd@plover.com
fac92740
MJD
903 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
904 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
905 * Element 0 holds the number n.
07be1b83 906 * Position is 1 indexed.
fac92740 907 */
7122b237 908#ifndef RE_TRACK_PATTERN_OFFSETS
d7c442ce 909#define Set_Node_Offset_To_R(offset,byte)
7122b237
YO
910#define Set_Node_Offset(node,byte)
911#define Set_Cur_Node_Offset
912#define Set_Node_Length_To_R(node,len)
913#define Set_Node_Length(node,len)
6a86c6ad 914#define Set_Node_Cur_Length(node,start)
538e84ed
KW
915#define Node_Offset(n)
916#define Node_Length(n)
7122b237
YO
917#define Set_Node_Offset_Length(node,offset,len)
918#define ProgLen(ri) ri->u.proglen
919#define SetProgLen(ri,x) ri->u.proglen = x
920#else
921#define ProgLen(ri) ri->u.offsets[0]
922#define SetProgLen(ri,x) ri->u.offsets[0] = x
d7c442ce 923#define Set_Node_Offset_To_R(offset,byte) STMT_START { \
ccb2c380
MP
924 if (! SIZE_ONLY) { \
925 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
d7c442ce
KW
926 __LINE__, (int)(offset), (int)(byte))); \
927 if((offset) < 0) { \
538e84ed 928 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
d7c442ce 929 (int)(offset)); \
ccb2c380 930 } else { \
d7c442ce 931 RExC_offsets[2*(offset)-1] = (byte); \
ccb2c380
MP
932 } \
933 } \
934} STMT_END
935
88f063b4 936#define Set_Node_Offset(node,byte) \
1a322bb4 937 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (byte)-RExC_start)
ccb2c380
MP
938#define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
939
940#define Set_Node_Length_To_R(node,len) STMT_START { \
941 if (! SIZE_ONLY) { \
942 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
551405c4 943 __LINE__, (int)(node), (int)(len))); \
ccb2c380 944 if((node) < 0) { \
538e84ed
KW
945 Perl_croak(aTHX_ "value of node is %d in Length macro", \
946 (int)(node)); \
ccb2c380
MP
947 } else { \
948 RExC_offsets[2*(node)] = (len); \
949 } \
950 } \
951} STMT_END
952
953#define Set_Node_Length(node,len) \
1a322bb4 954 Set_Node_Length_To_R(REGNODE_OFFSET(node), len)
6a86c6ad
NC
955#define Set_Node_Cur_Length(node, start) \
956 Set_Node_Length(node, RExC_parse - start)
fac92740
MJD
957
958/* Get offsets and lengths */
1a322bb4
KW
959#define Node_Offset(n) (RExC_offsets[2*(REGNODE_OFFSET(n))-1])
960#define Node_Length(n) (RExC_offsets[2*(REGNODE_OFFSET(n))])
fac92740 961
07be1b83 962#define Set_Node_Offset_Length(node,offset,len) STMT_START { \
1a322bb4
KW
963 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (offset)); \
964 Set_Node_Length_To_R(REGNODE_OFFSET(node), (len)); \
07be1b83 965} STMT_END
7122b237 966#endif
07be1b83
YO
967
968#if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
969#define EXPERIMENTAL_INPLACESCAN
f427392e 970#endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
07be1b83 971
cb41e5d6
YO
972#ifdef DEBUGGING
973int
6ad9a8ab 974Perl_re_printf(pTHX_ const char *fmt, ...)
cb41e5d6
YO
975{
976 va_list ap;
977 int result;
978 PerlIO *f= Perl_debug_log;
979 PERL_ARGS_ASSERT_RE_PRINTF;
980 va_start(ap, fmt);
981 result = PerlIO_vprintf(f, fmt, ap);
982 va_end(ap);
983 return result;
984}
985
986int
7b031478 987Perl_re_indentf(pTHX_ const char *fmt, U32 depth, ...)
cb41e5d6
YO
988{
989 va_list ap;
990 int result;
991 PerlIO *f= Perl_debug_log;
992 PERL_ARGS_ASSERT_RE_INDENTF;
993 va_start(ap, depth);
daeb874b 994 PerlIO_printf(f, "%*s", ( (int)depth % 20 ) * 2, "");
cb41e5d6
YO
995 result = PerlIO_vprintf(f, fmt, ap);
996 va_end(ap);
997 return result;
998}
999#endif /* DEBUGGING */
1000
7b031478 1001#define DEBUG_RExC_seen() \
538e84ed 1002 DEBUG_OPTIMISE_MORE_r({ \
88f063b4 1003 Perl_re_printf( aTHX_ "RExC_seen: "); \
538e84ed 1004 \
e384d5c1 1005 if (RExC_seen & REG_ZERO_LEN_SEEN) \
88f063b4 1006 Perl_re_printf( aTHX_ "REG_ZERO_LEN_SEEN "); \
538e84ed 1007 \
e384d5c1 1008 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
88f063b4 1009 Perl_re_printf( aTHX_ "REG_LOOKBEHIND_SEEN "); \
538e84ed 1010 \
e384d5c1 1011 if (RExC_seen & REG_GPOS_SEEN) \
88f063b4 1012 Perl_re_printf( aTHX_ "REG_GPOS_SEEN "); \
538e84ed 1013 \
e384d5c1 1014 if (RExC_seen & REG_RECURSE_SEEN) \
88f063b4 1015 Perl_re_printf( aTHX_ "REG_RECURSE_SEEN "); \
538e84ed 1016 \
7b031478 1017 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
88f063b4 1018 Perl_re_printf( aTHX_ "REG_TOP_LEVEL_BRANCHES_SEEN "); \
538e84ed 1019 \
e384d5c1 1020 if (RExC_seen & REG_VERBARG_SEEN) \
88f063b4 1021 Perl_re_printf( aTHX_ "REG_VERBARG_SEEN "); \
538e84ed 1022 \
e384d5c1 1023 if (RExC_seen & REG_CUTGROUP_SEEN) \
88f063b4 1024 Perl_re_printf( aTHX_ "REG_CUTGROUP_SEEN "); \
538e84ed 1025 \
e384d5c1 1026 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
88f063b4 1027 Perl_re_printf( aTHX_ "REG_RUN_ON_COMMENT_SEEN "); \
538e84ed 1028 \
e384d5c1 1029 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
88f063b4 1030 Perl_re_printf( aTHX_ "REG_UNFOLDED_MULTI_SEEN "); \
538e84ed 1031 \
7b031478 1032 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
88f063b4 1033 Perl_re_printf( aTHX_ "REG_UNBOUNDED_QUANTIFIER_SEEN "); \
ee273784 1034 \
88f063b4 1035 Perl_re_printf( aTHX_ "\n"); \
9e9ecfdf
YO
1036 });
1037
fdfb4f21 1038#define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
6ad9a8ab 1039 if ((flags) & flag) Perl_re_printf( aTHX_ "%s ", #flag)
fdfb4f21 1040
fdfb4f21 1041
f5a36d78
DM
1042#ifdef DEBUGGING
1043static void
1044S_debug_show_study_flags(pTHX_ U32 flags, const char *open_str,
1045 const char *close_str)
1046{
1047 if (!flags)
1048 return;
1049
1050 Perl_re_printf( aTHX_ "%s", open_str);
11683ecb
DM
1051 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_SEOL);
1052 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_MEOL);
f5a36d78
DM
1053 DEBUG_SHOW_STUDY_FLAG(flags, SF_IS_INF);
1054 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_PAR);
1055 DEBUG_SHOW_STUDY_FLAG(flags, SF_IN_PAR);
1056 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_EVAL);
1057 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_SUBSTR);
1058 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_AND);
1059 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_OR);
1060 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS);
1061 DEBUG_SHOW_STUDY_FLAG(flags, SCF_WHILEM_VISITED_POS);
1062 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_RESTUDY);
1063 DEBUG_SHOW_STUDY_FLAG(flags, SCF_SEEN_ACCEPT);
1064 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_DOING_RESTUDY);
1065 DEBUG_SHOW_STUDY_FLAG(flags, SCF_IN_DEFINE);
1066 Perl_re_printf( aTHX_ "%s", close_str);
1067}
1068
1069
1070static void
1071S_debug_studydata(pTHX_ const char *where, scan_data_t *data,
1072 U32 depth, int is_inf)
1073{
1074 GET_RE_DEBUG_FLAGS_DECL;
1075
1076 DEBUG_OPTIMISE_MORE_r({
1077 if (!data)
1078 return;
1079 Perl_re_indentf(aTHX_ "%s: Pos:%" IVdf "/%" IVdf " Flags: 0x%" UVXf,
1080 depth,
1081 where,
1082 (IV)data->pos_min,
1083 (IV)data->pos_delta,
1084 (UV)data->flags
1085 );
1086
1087 S_debug_show_study_flags(aTHX_ data->flags," [","]");
fdfb4f21 1088
f5a36d78
DM
1089 Perl_re_printf( aTHX_
1090 " Whilem_c: %" IVdf " Lcp: %" IVdf " %s",
1091 (IV)data->whilem_c,
1092 (IV)(data->last_closep ? *((data)->last_closep) : -1),
1093 is_inf ? "INF " : ""
1094 );
1095
11683ecb 1096 if (data->last_found) {
37b6262f 1097 int i;
11683ecb
DM
1098 Perl_re_printf(aTHX_
1099 "Last:'%s' %" IVdf ":%" IVdf "/%" IVdf,
1100 SvPVX_const(data->last_found),
1101 (IV)data->last_end,
1102 (IV)data->last_start_min,
1103 (IV)data->last_start_max
1104 );
1105
2099df82 1106 for (i = 0; i < 2; i++) {
37b6262f
DM
1107 Perl_re_printf(aTHX_
1108 " %s%s: '%s' @ %" IVdf "/%" IVdf,
2099df82
DM
1109 data->cur_is_floating == i ? "*" : "",
1110 i ? "Float" : "Fixed",
37b6262f
DM
1111 SvPVX_const(data->substrs[i].str),
1112 (IV)data->substrs[i].min_offset,
1113 (IV)data->substrs[i].max_offset
1114 );
1115 S_debug_show_study_flags(aTHX_ data->substrs[i].flags," [","]");
1116 }
11683ecb 1117 }
f5a36d78
DM
1118
1119 Perl_re_printf( aTHX_ "\n");
1120 });
1121}
1122
1123
1124static void
1125S_debug_peep(pTHX_ const char *str, const RExC_state_t *pRExC_state,
1126 regnode *scan, U32 depth, U32 flags)
1127{
1128 GET_RE_DEBUG_FLAGS_DECL;
1129
1130 DEBUG_OPTIMISE_r({
1131 regnode *Next;
1132
1133 if (!scan)
1134 return;
1135 Next = regnext(scan);
1136 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
1137 Perl_re_indentf( aTHX_ "%s>%3d: %s (%d)",
1138 depth,
1139 str,
1140 REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),
1141 Next ? (REG_NODE_NUM(Next)) : 0 );
1142 S_debug_show_study_flags(aTHX_ flags," [ ","]");
1143 Perl_re_printf( aTHX_ "\n");
1144 });
1145}
1146
1147
1148# define DEBUG_STUDYDATA(where, data, depth, is_inf) \
1149 S_debug_studydata(aTHX_ where, data, depth, is_inf)
1150
1151# define DEBUG_PEEP(str, scan, depth, flags) \
1152 S_debug_peep(aTHX_ str, pRExC_state, scan, depth, flags)
1153
1154#else
1155# define DEBUG_STUDYDATA(where, data, depth, is_inf) NOOP
1156# define DEBUG_PEEP(str, scan, depth, flags) NOOP
1157#endif
1de06328 1158
cb41e5d6 1159
c6871b76
KW
1160/* =========================================================
1161 * BEGIN edit_distance stuff.
1162 *
1163 * This calculates how many single character changes of any type are needed to
1164 * transform a string into another one. It is taken from version 3.1 of
1165 *
1166 * https://metacpan.org/pod/Text::Levenshtein::Damerau::XS
1167 */
1168
1169/* Our unsorted dictionary linked list. */
1170/* Note we use UVs, not chars. */
1171
1172struct dictionary{
1173 UV key;
1174 UV value;
1175 struct dictionary* next;
1176};
1177typedef struct dictionary item;
1178
1179
1180PERL_STATIC_INLINE item*
88f063b4 1181push(UV key, item* curr)
c6871b76
KW
1182{
1183 item* head;
d09f14bf 1184 Newx(head, 1, item);
c6871b76
KW
1185 head->key = key;
1186 head->value = 0;
1187 head->next = curr;
1188 return head;
1189}
1190
1191
1192PERL_STATIC_INLINE item*
1193find(item* head, UV key)
1194{
1195 item* iterator = head;
1196 while (iterator){
1197 if (iterator->key == key){
1198 return iterator;
1199 }
1200 iterator = iterator->next;
1201 }
1202
1203 return NULL;
1204}
1205
1206PERL_STATIC_INLINE item*
88f063b4 1207uniquePush(item* head, UV key)
c6871b76
KW
1208{
1209 item* iterator = head;
1210
1211 while (iterator){
1212 if (iterator->key == key) {
1213 return head;
1214 }
1215 iterator = iterator->next;
1216 }
1217
88f063b4 1218 return push(key, head);
c6871b76
KW
1219}
1220
1221PERL_STATIC_INLINE void
1222dict_free(item* head)
1223{
1224 item* iterator = head;
1225
1226 while (iterator) {
1227 item* temp = iterator;
1228 iterator = iterator->next;
1229 Safefree(temp);
1230 }
1231
1232 head = NULL;
1233}
1234
1235/* End of Dictionary Stuff */
1236
1237/* All calculations/work are done here */
1238STATIC int
1239S_edit_distance(const UV* src,
1240 const UV* tgt,
1241 const STRLEN x, /* length of src[] */
1242 const STRLEN y, /* length of tgt[] */
1243 const SSize_t maxDistance
1244)
1245{
1246 item *head = NULL;
88f063b4 1247 UV swapCount, swapScore, targetCharCount, i, j;
c6871b76
KW
1248 UV *scores;
1249 UV score_ceil = x + y;
1250
1251 PERL_ARGS_ASSERT_EDIT_DISTANCE;
1252
1253 /* intialize matrix start values */
d09f14bf 1254 Newx(scores, ( (x + 2) * (y + 2)), UV);
c6871b76
KW
1255 scores[0] = score_ceil;
1256 scores[1 * (y + 2) + 0] = score_ceil;
1257 scores[0 * (y + 2) + 1] = score_ceil;
1258 scores[1 * (y + 2) + 1] = 0;
88f063b4 1259 head = uniquePush(uniquePush(head, src[0]), tgt[0]);
c6871b76
KW
1260
1261 /* work loops */
1262 /* i = src index */
1263 /* j = tgt index */
1264 for (i=1;i<=x;i++) {
1265 if (i < x)
88f063b4 1266 head = uniquePush(head, src[i]);
c6871b76
KW
1267 scores[(i+1) * (y + 2) + 1] = i;
1268 scores[(i+1) * (y + 2) + 0] = score_ceil;
1269 swapCount = 0;
1270
1271 for (j=1;j<=y;j++) {
1272 if (i == 1) {
1273 if(j < y)
88f063b4 1274 head = uniquePush(head, tgt[j]);
c6871b76
KW
1275 scores[1 * (y + 2) + (j + 1)] = j;
1276 scores[0 * (y + 2) + (j + 1)] = score_ceil;
1277 }
1278
88f063b4 1279 targetCharCount = find(head, tgt[j-1])->value;
c6871b76
KW
1280 swapScore = scores[targetCharCount * (y + 2) + swapCount] + i - targetCharCount - 1 + j - swapCount;
1281
1282 if (src[i-1] != tgt[j-1]){
1283 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));
1284 }
1285 else {
1286 swapCount = j;
1287 scores[(i+1) * (y + 2) + (j + 1)] = MIN(scores[i * (y + 2) + j], swapScore);
1288 }
1289 }
1290
88f063b4 1291 find(head, src[i-1])->value = i;
c6871b76
KW
1292 }
1293
1294 {
1295 IV score = scores[(x+1) * (y + 2) + (y + 1)];
1296 dict_free(head);
1297 Safefree(scores);
1298 return (maxDistance != 0 && maxDistance < score)?(-1):score;
1299 }
1300}
1301
1302/* END of edit_distance() stuff
1303 * ========================================================= */
1304
8e35b056
KW
1305/* is c a control character for which we have a mnemonic? */
1306#define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
1307
549b4e78
KW
1308STATIC const char *
1309S_cntrl_to_mnemonic(const U8 c)
1310{
1311 /* Returns the mnemonic string that represents character 'c', if one
1312 * exists; NULL otherwise. The only ones that exist for the purposes of
1313 * this routine are a few control characters */
1314
1315 switch (c) {
1316 case '\a': return "\\a";
1317 case '\b': return "\\b";
1318 case ESC_NATIVE: return "\\e";
1319 case '\f': return "\\f";
1320 case '\n': return "\\n";
1321 case '\r': return "\\r";
1322 case '\t': return "\\t";
1323 }
1324
1325 return NULL;
1326}
1327
653099ff 1328/* Mark that we cannot extend a found fixed substring at this point.
37b6262f 1329 Update the longest found anchored substring or the longest found
653099ff
GS
1330 floating substrings if needed. */
1331
4327152a 1332STATIC void
ea3daa5d
FC
1333S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
1334 SSize_t *minlenp, int is_inf)
c277df42 1335{
e1ec3a88 1336 const STRLEN l = CHR_SVLEN(data->last_found);
2099df82 1337 SV * const longest_sv = data->substrs[data->cur_is_floating].str;
a2ca2017 1338 const STRLEN old_l = CHR_SVLEN(longest_sv);
1de06328 1339 GET_RE_DEBUG_FLAGS_DECL;
b81d288d 1340
7918f24d
NC
1341 PERL_ARGS_ASSERT_SCAN_COMMIT;
1342
c277df42 1343 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
2099df82 1344 const U8 i = data->cur_is_floating;
a2ca2017 1345 SvSetMagicSV(longest_sv, data->last_found);
37b6262f
DM
1346 data->substrs[i].min_offset = l ? data->last_start_min : data->pos_min;
1347
2099df82
DM
1348 if (!i) /* fixed */
1349 data->substrs[0].max_offset = data->substrs[0].min_offset;
1350 else { /* float */
1351 data->substrs[1].max_offset = (l
646e8787
DM
1352 ? data->last_start_max
1353 : (data->pos_delta > SSize_t_MAX - data->pos_min
ea3daa5d
FC
1354 ? SSize_t_MAX
1355 : data->pos_min + data->pos_delta));
1356 if (is_inf
2099df82
DM
1357 || (STRLEN)data->substrs[1].max_offset > (STRLEN)SSize_t_MAX)
1358 data->substrs[1].max_offset = SSize_t_MAX;
37b6262f 1359 }
11683ecb 1360
37b6262f
DM
1361 if (data->flags & SF_BEFORE_EOL)
1362 data->substrs[i].flags |= (data->flags & SF_BEFORE_EOL);
1363 else
1364 data->substrs[i].flags &= ~SF_BEFORE_EOL;
1365 data->substrs[i].minlenp = minlenp;
1366 data->substrs[i].lookbehind = 0;
c277df42 1367 }
37b6262f 1368
c277df42 1369 SvCUR_set(data->last_found, 0);
0eda9292 1370 {
a28509cc 1371 SV * const sv = data->last_found;
097eb12c
AL
1372 if (SvUTF8(sv) && SvMAGICAL(sv)) {
1373 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
1374 if (mg)
1375 mg->mg_len = 0;
1376 }
0eda9292 1377 }
c277df42
IZ
1378 data->last_end = -1;
1379 data->flags &= ~SF_BEFORE_EOL;
f5a36d78 1380 DEBUG_STUDYDATA("commit", data, 0, is_inf);
c277df42
IZ
1381}
1382
cdd87c1d
KW
1383/* An SSC is just a regnode_charclass_posix with an extra field: the inversion
1384 * list that describes which code points it matches */
1385
653099ff 1386STATIC void
3420edd7 1387S_ssc_anything(pTHX_ regnode_ssc *ssc)
653099ff 1388{
cdd87c1d
KW
1389 /* Set the SSC 'ssc' to match an empty string or any code point */
1390
557bd3fb 1391 PERL_ARGS_ASSERT_SSC_ANYTHING;
7918f24d 1392
71068078 1393 assert(is_ANYOF_SYNTHETIC(ssc));
3fffb88a 1394
0854ea0b
KW
1395 /* mortalize so won't leak */
1396 ssc->invlist = sv_2mortal(_add_range_to_invlist(NULL, 0, UV_MAX));
93e92956 1397 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
653099ff
GS
1398}
1399
653099ff 1400STATIC int
dc3bf405 1401S_ssc_is_anything(const regnode_ssc *ssc)
653099ff 1402{
c144baaa
KW
1403 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
1404 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
1405 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1406 * in any way, so there's no point in using it */
cdd87c1d
KW
1407
1408 UV start, end;
1409 bool ret;
653099ff 1410
557bd3fb 1411 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
7918f24d 1412
71068078 1413 assert(is_ANYOF_SYNTHETIC(ssc));
cdd87c1d 1414
93e92956 1415 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
cdd87c1d
KW
1416 return FALSE;
1417 }
1418
1419 /* See if the list consists solely of the range 0 - Infinity */
1420 invlist_iterinit(ssc->invlist);
1421 ret = invlist_iternext(ssc->invlist, &start, &end)
1422 && start == 0
1423 && end == UV_MAX;
1424
1425 invlist_iterfinish(ssc->invlist);
1426
1427 if (ret) {
1428 return TRUE;
1429 }
1430
1431 /* If e.g., both \w and \W are set, matches everything */
e0e1be5f 1432 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1433 int i;
1434 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1435 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1436 return TRUE;
1437 }
1438 }
1439 }
1440
1441 return FALSE;
653099ff
GS
1442}
1443
653099ff 1444STATIC void
cdd87c1d 1445S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
653099ff 1446{
cdd87c1d
KW
1447 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1448 * string, any code point, or any posix class under locale */
1449
557bd3fb 1450 PERL_ARGS_ASSERT_SSC_INIT;
7918f24d 1451
557bd3fb 1452 Zero(ssc, 1, regnode_ssc);
71068078 1453 set_ANYOF_SYNTHETIC(ssc);
93e92956 1454 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
3420edd7 1455 ssc_anything(ssc);
cdd87c1d 1456
2f306ab9
KW
1457 /* If any portion of the regex is to operate under locale rules that aren't
1458 * fully known at compile time, initialization includes it. The reason
1459 * this isn't done for all regexes is that the optimizer was written under
1460 * the assumption that locale was all-or-nothing. Given the complexity and
1461 * lack of documentation in the optimizer, and that there are inadequate
1462 * test cases for locale, many parts of it may not work properly, it is
1463 * safest to avoid locale unless necessary. */
cdd87c1d
KW
1464 if (RExC_contains_locale) {
1465 ANYOF_POSIXL_SETALL(ssc);
cdd87c1d
KW
1466 }
1467 else {
1468 ANYOF_POSIXL_ZERO(ssc);
1469 }
653099ff
GS
1470}
1471
b423522f 1472STATIC int
dc3bf405
BF
1473S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1474 const regnode_ssc *ssc)
b423522f
KW
1475{
1476 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1477 * to the list of code points matched, and locale posix classes; hence does
1478 * not check its flags) */
1479
1480 UV start, end;
1481 bool ret;
1482
1483 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1484
71068078 1485 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1486
1487 invlist_iterinit(ssc->invlist);
1488 ret = invlist_iternext(ssc->invlist, &start, &end)
1489 && start == 0
1490 && end == UV_MAX;
1491
1492 invlist_iterfinish(ssc->invlist);
1493
1494 if (! ret) {
1495 return FALSE;
1496 }
1497
e0e1be5f 1498 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
31f05a37 1499 return FALSE;
b423522f
KW
1500 }
1501
1502 return TRUE;
1503}
1504
1505STATIC SV*
1506S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
5c0f85ef 1507 const regnode_charclass* const node)
b423522f
KW
1508{
1509 /* Returns a mortal inversion list defining which code points are matched
1510 * by 'node', which is of type ANYOF. Handles complementing the result if
1511 * appropriate. If some code points aren't knowable at this time, the
31f05a37
KW
1512 * returned list must, and will, contain every code point that is a
1513 * possibility. */
b423522f 1514
e2506fa7 1515 SV* invlist = NULL;
1ee208c4 1516 SV* only_utf8_locale_invlist = NULL;
b423522f
KW
1517 unsigned int i;
1518 const U32 n = ARG(node);
31f05a37 1519 bool new_node_has_latin1 = FALSE;
b423522f
KW
1520
1521 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1522
1523 /* Look at the data structure created by S_set_ANYOF_arg() */
93e92956 1524 if (n != ANYOF_ONLY_HAS_BITMAP) {
b423522f
KW
1525 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1526 AV * const av = MUTABLE_AV(SvRV(rv));
1527 SV **const ary = AvARRAY(av);
1528 assert(RExC_rxi->data->what[n] == 's');
1529
1530 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
28118b9c 1531 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1]), NULL));
b423522f
KW
1532 }
1533 else if (ary[0] && ary[0] != &PL_sv_undef) {
1534
1535 /* Here, no compile-time swash, and there are things that won't be
1536 * known until runtime -- we have to assume it could be anything */
e2506fa7 1537 invlist = sv_2mortal(_new_invlist(1));
b423522f
KW
1538 return _add_range_to_invlist(invlist, 0, UV_MAX);
1539 }
1ee208c4 1540 else if (ary[3] && ary[3] != &PL_sv_undef) {
b423522f
KW
1541
1542 /* Here no compile-time swash, and no run-time only data. Use the
1543 * node's inversion list */
28118b9c 1544 invlist = sv_2mortal(invlist_clone(ary[3], NULL));
1ee208c4
KW
1545 }
1546
1547 /* Get the code points valid only under UTF-8 locales */
037715a6 1548 if ((ANYOF_FLAGS(node) & ANYOFL_FOLD)
1ee208c4
KW
1549 && ary[2] && ary[2] != &PL_sv_undef)
1550 {
1551 only_utf8_locale_invlist = ary[2];
b423522f
KW
1552 }
1553 }
1554
e2506fa7
KW
1555 if (! invlist) {
1556 invlist = sv_2mortal(_new_invlist(0));
1557 }
1558
dcb20b36
KW
1559 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1560 * code points, and an inversion list for the others, but if there are code
1561 * points that should match only conditionally on the target string being
1562 * UTF-8, those are placed in the inversion list, and not the bitmap.
1563 * Since there are circumstances under which they could match, they are
1564 * included in the SSC. But if the ANYOF node is to be inverted, we have
1565 * to exclude them here, so that when we invert below, the end result
1566 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1567 * have to do this here before we add the unconditionally matched code
1568 * points */
b423522f
KW
1569 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1570 _invlist_intersection_complement_2nd(invlist,
1571 PL_UpperLatin1,
1572 &invlist);
1573 }
1574
1575 /* Add in the points from the bit map */
dcb20b36 1576 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
b423522f 1577 if (ANYOF_BITMAP_TEST(node, i)) {
6f8848d5
TC
1578 unsigned int start = i++;
1579
1580 for (; i < NUM_ANYOF_CODE_POINTS && ANYOF_BITMAP_TEST(node, i); ++i) {
1581 /* empty */
1582 }
1583 invlist = _add_range_to_invlist(invlist, start, i-1);
31f05a37 1584 new_node_has_latin1 = TRUE;
b423522f
KW
1585 }
1586 }
1587
1588 /* If this can match all upper Latin1 code points, have to add them
ac33c516
KW
1589 * as well. But don't add them if inverting, as when that gets done below,
1590 * it would exclude all these characters, including the ones it shouldn't
1591 * that were added just above */
1592 if (! (ANYOF_FLAGS(node) & ANYOF_INVERT) && OP(node) == ANYOFD
f240c685
KW
1593 && (ANYOF_FLAGS(node) & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER))
1594 {
b423522f
KW
1595 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1596 }
1597
1598 /* Similarly for these */
93e92956 1599 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
e0a1ff7a 1600 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
b423522f
KW
1601 }
1602
1603 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1604 _invlist_invert(invlist);
1605 }
037715a6 1606 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOFL_FOLD) {
31f05a37
KW
1607
1608 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1609 * locale. We can skip this if there are no 0-255 at all. */
1610 _invlist_union(invlist, PL_Latin1, &invlist);
1611 }
1612
1ee208c4
KW
1613 /* Similarly add the UTF-8 locale possible matches. These have to be
1614 * deferred until after the non-UTF-8 locale ones are taken care of just
1615 * above, or it leads to wrong results under ANYOF_INVERT */
1616 if (only_utf8_locale_invlist) {
31f05a37 1617 _invlist_union_maybe_complement_2nd(invlist,
1ee208c4 1618 only_utf8_locale_invlist,
31f05a37
KW
1619 ANYOF_FLAGS(node) & ANYOF_INVERT,
1620 &invlist);
1621 }
b423522f
KW
1622
1623 return invlist;
1624}
1625
1051e1c4 1626/* These two functions currently do the exact same thing */
557bd3fb 1627#define ssc_init_zero ssc_init
653099ff 1628
cdd87c1d
KW
1629#define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1630#define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1631
557bd3fb 1632/* 'AND' a given class with another one. Can create false positives. 'ssc'
93e92956
KW
1633 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1634 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
cdd87c1d 1635
653099ff 1636STATIC void
b423522f 1637S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1638 const regnode_charclass *and_with)
653099ff 1639{
cdd87c1d
KW
1640 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1641 * another SSC or a regular ANYOF class. Can create false positives. */
40d049e4 1642
a0dd4231
KW
1643 SV* anded_cp_list;
1644 U8 anded_flags;
1e6ade67 1645
cdd87c1d 1646 PERL_ARGS_ASSERT_SSC_AND;
653099ff 1647
71068078 1648 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1649
1650 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1651 * the code point inversion list and just the relevant flags */
71068078 1652 if (is_ANYOF_SYNTHETIC(and_with)) {
7dcac5f6 1653 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
a0dd4231 1654 anded_flags = ANYOF_FLAGS(and_with);
e9b08962
KW
1655
1656 /* XXX This is a kludge around what appears to be deficiencies in the
1657 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1658 * there are paths through the optimizer where it doesn't get weeded
1659 * out when it should. And if we don't make some extra provision for
1660 * it like the code just below, it doesn't get added when it should.
1661 * This solution is to add it only when AND'ing, which is here, and
1662 * only when what is being AND'ed is the pristine, original node
1663 * matching anything. Thus it is like adding it to ssc_anything() but
1664 * only when the result is to be AND'ed. Probably the same solution
1665 * could be adopted for the same problem we have with /l matching,
1666 * which is solved differently in S_ssc_init(), and that would lead to
1667 * fewer false positives than that solution has. But if this solution
1668 * creates bugs, the consequences are only that a warning isn't raised
1669 * that should be; while the consequences for having /l bugs is
1670 * incorrect matches */
7dcac5f6 1671 if (ssc_is_anything((regnode_ssc *)and_with)) {
f240c685 1672 anded_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
e9b08962 1673 }
a0dd4231
KW
1674 }
1675 else {
5c0f85ef 1676 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
f240c685
KW
1677 if (OP(and_with) == ANYOFD) {
1678 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1679 }
1680 else {
1681 anded_flags = ANYOF_FLAGS(and_with)
1682 &( ANYOF_COMMON_FLAGS
108316fb
KW
1683 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1684 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1685 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(and_with))) {
1686 anded_flags &=
1687 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1688 }
f240c685 1689 }
a0dd4231
KW
1690 }
1691
1692 ANYOF_FLAGS(ssc) &= anded_flags;
cdd87c1d
KW
1693
1694 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1695 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1696 * 'and_with' may be inverted. When not inverted, we have the situation of
1697 * computing:
1698 * (C1 | P1) & (C2 | P2)
1699 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1700 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1701 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1702 * <= ((C1 & C2) | P1 | P2)
1703 * Alternatively, the last few steps could be:
1704 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1705 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1706 * <= (C1 | C2 | (P1 & P2))
1707 * We favor the second approach if either P1 or P2 is non-empty. This is
1708 * because these components are a barrier to doing optimizations, as what
1709 * they match cannot be known until the moment of matching as they are
1710 * dependent on the current locale, 'AND"ing them likely will reduce or
1711 * eliminate them.
1712 * But we can do better if we know that C1,P1 are in their initial state (a
1713 * frequent occurrence), each matching everything:
1714 * (<everything>) & (C2 | P2) = C2 | P2
1715 * Similarly, if C2,P2 are in their initial state (again a frequent
1716 * occurrence), the result is a no-op
1717 * (C1 | P1) & (<everything>) = C1 | P1
1718 *
1719 * Inverted, we have
1720 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1721 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1722 * <= (C1 & ~C2) | (P1 & ~P2)
1723 * */
1aa99e6b 1724
a0dd4231 1725 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
71068078 1726 && ! is_ANYOF_SYNTHETIC(and_with))
a0dd4231 1727 {
cdd87c1d 1728 unsigned int i;
8951c461 1729
cdd87c1d
KW
1730 ssc_intersection(ssc,
1731 anded_cp_list,
1732 FALSE /* Has already been inverted */
1733 );
c6b76537 1734
cdd87c1d
KW
1735 /* If either P1 or P2 is empty, the intersection will be also; can skip
1736 * the loop */
93e92956 1737 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
cdd87c1d
KW
1738 ANYOF_POSIXL_ZERO(ssc);
1739 }
e0e1be5f 1740 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1741
1742 /* Note that the Posix class component P from 'and_with' actually
1743 * looks like:
1744 * P = Pa | Pb | ... | Pn
1745 * where each component is one posix class, such as in [\w\s].
1746 * Thus
1747 * ~P = ~(Pa | Pb | ... | Pn)
1748 * = ~Pa & ~Pb & ... & ~Pn
1749 * <= ~Pa | ~Pb | ... | ~Pn
1750 * The last is something we can easily calculate, but unfortunately
1751 * is likely to have many false positives. We could do better
1752 * in some (but certainly not all) instances if two classes in
1753 * P have known relationships. For example
1754 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1755 * So
1756 * :lower: & :print: = :lower:
1757 * And similarly for classes that must be disjoint. For example,
1758 * since \s and \w can have no elements in common based on rules in
1759 * the POSIX standard,
1760 * \w & ^\S = nothing
1761 * Unfortunately, some vendor locales do not meet the Posix
1762 * standard, in particular almost everything by Microsoft.
1763 * The loop below just changes e.g., \w into \W and vice versa */
1764
1ee208c4 1765 regnode_charclass_posixl temp;
cdd87c1d
KW
1766 int add = 1; /* To calculate the index of the complement */
1767
b1234259 1768 Zero(&temp, 1, regnode_charclass_posixl);
cdd87c1d
KW
1769 ANYOF_POSIXL_ZERO(&temp);
1770 for (i = 0; i < ANYOF_MAX; i++) {
1771 assert(i % 2 != 0
7dcac5f6
KW
1772 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1773 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
cdd87c1d 1774
7dcac5f6 1775 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
cdd87c1d
KW
1776 ANYOF_POSIXL_SET(&temp, i + add);
1777 }
1778 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1779 }
1780 ANYOF_POSIXL_AND(&temp, ssc);
c6b76537 1781
cdd87c1d
KW
1782 } /* else ssc already has no posixes */
1783 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1784 in its initial state */
71068078 1785 else if (! is_ANYOF_SYNTHETIC(and_with)
7dcac5f6 1786 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
cdd87c1d
KW
1787 {
1788 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1789 * copy it over 'ssc' */
1790 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
71068078 1791 if (is_ANYOF_SYNTHETIC(and_with)) {
cdd87c1d
KW
1792 StructCopy(and_with, ssc, regnode_ssc);
1793 }
1794 else {
1795 ssc->invlist = anded_cp_list;
1796 ANYOF_POSIXL_ZERO(ssc);
93e92956 1797 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1798 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
cdd87c1d
KW
1799 }
1800 }
1801 }
e0e1be5f 1802 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
93e92956 1803 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
cdd87c1d
KW
1804 {
1805 /* One or the other of P1, P2 is non-empty. */
93e92956 1806 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1ea8b7fe
KW
1807 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1808 }
cdd87c1d
KW
1809 ssc_union(ssc, anded_cp_list, FALSE);
1810 }
1811 else { /* P1 = P2 = empty */
1812 ssc_intersection(ssc, anded_cp_list, FALSE);
1813 }
137165a6 1814 }
653099ff
GS
1815}
1816
653099ff 1817STATIC void
cdd87c1d 1818S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1819 const regnode_charclass *or_with)
653099ff 1820{
cdd87c1d
KW
1821 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1822 * another SSC or a regular ANYOF class. Can create false positives if
1823 * 'or_with' is to be inverted. */
7918f24d 1824
a0dd4231
KW
1825 SV* ored_cp_list;
1826 U8 ored_flags;
c6b76537 1827
cdd87c1d 1828 PERL_ARGS_ASSERT_SSC_OR;
c6b76537 1829
71068078 1830 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1831
1832 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1833 * the code point inversion list and just the relevant flags */
71068078 1834 if (is_ANYOF_SYNTHETIC(or_with)) {
7dcac5f6 1835 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
a0dd4231
KW
1836 ored_flags = ANYOF_FLAGS(or_with);
1837 }
1838 else {
5c0f85ef 1839 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
eff8b7dc 1840 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
f240c685
KW
1841 if (OP(or_with) != ANYOFD) {
1842 ored_flags
1843 |= ANYOF_FLAGS(or_with)
108316fb
KW
1844 & ( ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1845 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1846 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(or_with))) {
1847 ored_flags |=
1848 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1849 }
f240c685 1850 }
a0dd4231
KW
1851 }
1852
1853 ANYOF_FLAGS(ssc) |= ored_flags;
cdd87c1d
KW
1854
1855 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1856 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1857 * 'or_with' may be inverted. When not inverted, we have the simple
1858 * situation of computing:
1859 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1860 * If P1|P2 yields a situation with both a class and its complement are
1861 * set, like having both \w and \W, this matches all code points, and we
1862 * can delete these from the P component of the ssc going forward. XXX We
1863 * might be able to delete all the P components, but I (khw) am not certain
1864 * about this, and it is better to be safe.
1865 *
1866 * Inverted, we have
1867 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1868 * <= (C1 | P1) | ~C2
1869 * <= (C1 | ~C2) | P1
1870 * (which results in actually simpler code than the non-inverted case)
1871 * */
9826f543 1872
a0dd4231 1873 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
71068078 1874 && ! is_ANYOF_SYNTHETIC(or_with))
a0dd4231 1875 {
cdd87c1d 1876 /* We ignore P2, leaving P1 going forward */
1ea8b7fe 1877 } /* else Not inverted */
93e92956 1878 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1879 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
e0e1be5f 1880 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1881 unsigned int i;
1882 for (i = 0; i < ANYOF_MAX; i += 2) {
1883 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1884 {
1885 ssc_match_all_cp(ssc);
1886 ANYOF_POSIXL_CLEAR(ssc, i);
1887 ANYOF_POSIXL_CLEAR(ssc, i+1);
cdd87c1d
KW
1888 }
1889 }
1890 }
1aa99e6b 1891 }
cdd87c1d
KW
1892
1893 ssc_union(ssc,
1894 ored_cp_list,
1895 FALSE /* Already has been inverted */
1896 );
653099ff
GS
1897}
1898
b423522f
KW
1899PERL_STATIC_INLINE void
1900S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1901{
1902 PERL_ARGS_ASSERT_SSC_UNION;
1903
71068078 1904 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1905
1906 _invlist_union_maybe_complement_2nd(ssc->invlist,
1907 invlist,
1908 invert2nd,
1909 &ssc->invlist);
1910}
1911
1912PERL_STATIC_INLINE void
1913S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1914 SV* const invlist,
1915 const bool invert2nd)
1916{
1917 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1918
71068078 1919 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1920
1921 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1922 invlist,
1923 invert2nd,
1924 &ssc->invlist);
1925}
1926
1927PERL_STATIC_INLINE void
1928S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1929{
1930 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1931
71068078 1932 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1933
1934 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1935}
1936
1937PERL_STATIC_INLINE void
1938S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1939{
1940 /* AND just the single code point 'cp' into the SSC 'ssc' */
1941
1942 SV* cp_list = _new_invlist(2);
1943
1944 PERL_ARGS_ASSERT_SSC_CP_AND;
1945
71068078 1946 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1947
1948 cp_list = add_cp_to_invlist(cp_list, cp);
1949 ssc_intersection(ssc, cp_list,
1950 FALSE /* Not inverted */
1951 );
1952 SvREFCNT_dec_NN(cp_list);
1953}
1954
1955PERL_STATIC_INLINE void
dc3bf405 1956S_ssc_clear_locale(regnode_ssc *ssc)
b423522f
KW
1957{
1958 /* Set the SSC 'ssc' to not match any locale things */
b423522f
KW
1959 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
1960
71068078 1961 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1962
1963 ANYOF_POSIXL_ZERO(ssc);
1964 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
1965}
1966
b35552de
KW
1967#define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
1968
1969STATIC bool
1970S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
1971{
1972 /* The synthetic start class is used to hopefully quickly winnow down
1973 * places where a pattern could start a match in the target string. If it
1974 * doesn't really narrow things down that much, there isn't much point to
1975 * having the overhead of using it. This function uses some very crude
1976 * heuristics to decide if to use the ssc or not.
1977 *
1978 * It returns TRUE if 'ssc' rules out more than half what it considers to
1979 * be the "likely" possible matches, but of course it doesn't know what the
1980 * actual things being matched are going to be; these are only guesses
1981 *
1982 * For /l matches, it assumes that the only likely matches are going to be
1983 * in the 0-255 range, uniformly distributed, so half of that is 127
1984 * For /a and /d matches, it assumes that the likely matches will be just
1985 * the ASCII range, so half of that is 63
1986 * For /u and there isn't anything matching above the Latin1 range, it
1987 * assumes that that is the only range likely to be matched, and uses
1988 * half that as the cut-off: 127. If anything matches above Latin1,
1989 * it assumes that all of Unicode could match (uniformly), except for
1990 * non-Unicode code points and things in the General Category "Other"
1991 * (unassigned, private use, surrogates, controls and formats). This
1992 * is a much large number. */
1993
b35552de
KW
1994 U32 count = 0; /* Running total of number of code points matched by
1995 'ssc' */
1996 UV start, end; /* Start and end points of current range in inversion
1997 list */
72400949
KW
1998 const U32 max_code_points = (LOC)
1999 ? 256
2000 : (( ! UNI_SEMANTICS
2001 || invlist_highest(ssc->invlist) < 256)
2002 ? 128
2003 : NON_OTHER_COUNT);
2004 const U32 max_match = max_code_points / 2;
b35552de
KW
2005
2006 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
2007
2008 invlist_iterinit(ssc->invlist);
2009 while (invlist_iternext(ssc->invlist, &start, &end)) {
72400949
KW
2010 if (start >= max_code_points) {
2011 break;
b35552de 2012 }
72400949 2013 end = MIN(end, max_code_points - 1);
b35552de 2014 count += end - start + 1;
72400949 2015 if (count >= max_match) {
b35552de
KW
2016 invlist_iterfinish(ssc->invlist);
2017 return FALSE;
2018 }
2019 }
2020
2021 return TRUE;
2022}
2023
2024
b423522f
KW
2025STATIC void
2026S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
2027{
2028 /* The inversion list in the SSC is marked mortal; now we need a more
2029 * permanent copy, which is stored the same way that is done in a regular
dcb20b36
KW
2030 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
2031 * map */
b423522f 2032
28118b9c 2033 SV* invlist = invlist_clone(ssc->invlist, NULL);
b423522f
KW
2034
2035 PERL_ARGS_ASSERT_SSC_FINALIZE;
2036
71068078 2037 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f 2038
a0dd4231 2039 /* The code in this file assumes that all but these flags aren't relevant
93e92956
KW
2040 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
2041 * by the time we reach here */
f240c685
KW
2042 assert(! (ANYOF_FLAGS(ssc)
2043 & ~( ANYOF_COMMON_FLAGS
108316fb
KW
2044 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
2045 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)));
a0dd4231 2046
b423522f
KW
2047 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
2048
1ee208c4
KW
2049 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
2050 NULL, NULL, NULL, FALSE);
b423522f 2051
85c8e306
KW
2052 /* Make sure is clone-safe */
2053 ssc->invlist = NULL;
2054
e0e1be5f 2055 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
93e92956 2056 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
d156f5cb 2057 OP(ssc) = ANYOFPOSIXL;
e0e1be5f 2058 }
d156f5cb 2059 else if (RExC_contains_locale) {
b2e90ddf
KW
2060 OP(ssc) = ANYOFL;
2061 }
2062
1462525b 2063 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
b423522f
KW
2064}
2065
a3621e74
YO
2066#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
2067#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
2068#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
538e84ed
KW
2069#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
2070 ? (TRIE_LIST_CUR( idx ) - 1) \
2071 : 0 )
a3621e74 2072
3dab1dad
YO
2073
2074#ifdef DEBUGGING
07be1b83 2075/*
2b8b4781
NC
2076 dump_trie(trie,widecharmap,revcharmap)
2077 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
2078 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
2079
2080 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
2081 The _interim_ variants are used for debugging the interim
2082 tables that are used to generate the final compressed
2083 representation which is what dump_trie expects.
2084
486ec47a 2085 Part of the reason for their existence is to provide a form
3dab1dad 2086 of documentation as to how the different representations function.
07be1b83
YO
2087
2088*/
3dab1dad
YO
2089
2090/*
3dab1dad
YO
2091 Dumps the final compressed table form of the trie to Perl_debug_log.
2092 Used for debugging make_trie().
2093*/
b9a59e08 2094
3dab1dad 2095STATIC void
2b8b4781
NC
2096S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
2097 AV *revcharmap, U32 depth)
3dab1dad
YO
2098{
2099 U32 state;
ab3bbdeb 2100 SV *sv=sv_newmortal();
55eed653 2101 int colwidth= widecharmap ? 6 : 4;
2e64971a 2102 U16 word;
3dab1dad
YO
2103 GET_RE_DEBUG_FLAGS_DECL;
2104
7918f24d 2105 PERL_ARGS_ASSERT_DUMP_TRIE;
ab3bbdeb 2106
6ad9a8ab 2107 Perl_re_indentf( aTHX_ "Char : %-6s%-6s%-4s ",
1e37780e 2108 depth+1, "Match","Base","Ofs" );
3dab1dad
YO
2109
2110 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 2111 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 2112 if ( tmp ) {
6ad9a8ab 2113 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2114 colwidth,
538e84ed 2115 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2116 PL_colors[0], PL_colors[1],
2117 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2118 PERL_PV_ESCAPE_FIRSTCHAR
2119 )
ab3bbdeb 2120 );
3dab1dad
YO
2121 }
2122 }
1e37780e
YO
2123 Perl_re_printf( aTHX_ "\n");
2124 Perl_re_indentf( aTHX_ "State|-----------------------", depth+1);
3dab1dad
YO
2125
2126 for( state = 0 ; state < trie->uniquecharcount ; state++ )
6ad9a8ab
YO
2127 Perl_re_printf( aTHX_ "%.*s", colwidth, "--------");
2128 Perl_re_printf( aTHX_ "\n");
3dab1dad 2129
1e2e3d02 2130 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 2131 const U32 base = trie->states[ state ].trans.base;
3dab1dad 2132
147e3846 2133 Perl_re_indentf( aTHX_ "#%4" UVXf "|", depth+1, (UV)state);
3dab1dad
YO
2134
2135 if ( trie->states[ state ].wordnum ) {
1e37780e 2136 Perl_re_printf( aTHX_ " W%4X", trie->states[ state ].wordnum );
3dab1dad 2137 } else {
6ad9a8ab 2138 Perl_re_printf( aTHX_ "%6s", "" );
3dab1dad
YO
2139 }
2140
147e3846 2141 Perl_re_printf( aTHX_ " @%4" UVXf " ", (UV)base );
3dab1dad
YO
2142
2143 if ( base ) {
2144 U32 ofs = 0;
2145
2146 while( ( base + ofs < trie->uniquecharcount ) ||
2147 ( base + ofs - trie->uniquecharcount < trie->lasttrans
538e84ed
KW
2148 && trie->trans[ base + ofs - trie->uniquecharcount ].check
2149 != state))
3dab1dad
YO
2150 ofs++;
2151
147e3846 2152 Perl_re_printf( aTHX_ "+%2" UVXf "[ ", (UV)ofs);
3dab1dad
YO
2153
2154 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
538e84ed
KW
2155 if ( ( base + ofs >= trie->uniquecharcount )
2156 && ( base + ofs - trie->uniquecharcount
2157 < trie->lasttrans )
2158 && trie->trans[ base + ofs
2159 - trie->uniquecharcount ].check == state )
3dab1dad 2160 {
147e3846 2161 Perl_re_printf( aTHX_ "%*" UVXf, colwidth,
1e37780e
YO
2162 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next
2163 );
3dab1dad 2164 } else {
88f063b4 2165 Perl_re_printf( aTHX_ "%*s", colwidth," ." );
3dab1dad
YO
2166 }
2167 }
2168
6ad9a8ab 2169 Perl_re_printf( aTHX_ "]");
3dab1dad
YO
2170
2171 }
6ad9a8ab 2172 Perl_re_printf( aTHX_ "\n" );
3dab1dad 2173 }
6ad9a8ab 2174 Perl_re_indentf( aTHX_ "word_info N:(prev,len)=",
cb41e5d6 2175 depth);
2e64971a 2176 for (word=1; word <= trie->wordcount; word++) {
6ad9a8ab 2177 Perl_re_printf( aTHX_ " %d:(%d,%d)",
2e64971a
DM
2178 (int)word, (int)(trie->wordinfo[word].prev),
2179 (int)(trie->wordinfo[word].len));
2180 }
6ad9a8ab 2181 Perl_re_printf( aTHX_ "\n" );
538e84ed 2182}
3dab1dad 2183/*
3dab1dad 2184 Dumps a fully constructed but uncompressed trie in list form.
538e84ed 2185 List tries normally only are used for construction when the number of
3dab1dad
YO
2186 possible chars (trie->uniquecharcount) is very high.
2187 Used for debugging make_trie().
2188*/
2189STATIC void
55eed653 2190S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2191 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2192 U32 depth)
3dab1dad
YO
2193{
2194 U32 state;
ab3bbdeb 2195 SV *sv=sv_newmortal();
55eed653 2196 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2197 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2198
2199 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
2200
3dab1dad 2201 /* print out the table precompression. */
6ad9a8ab 2202 Perl_re_indentf( aTHX_ "State :Word | Transition Data\n",
cb41e5d6 2203 depth+1 );
6ad9a8ab 2204 Perl_re_indentf( aTHX_ "%s",
cb41e5d6 2205 depth+1, "------:-----+-----------------\n" );
538e84ed 2206
3dab1dad
YO
2207 for( state=1 ; state < next_alloc ; state ++ ) {
2208 U16 charid;
538e84ed 2209
147e3846 2210 Perl_re_indentf( aTHX_ " %4" UVXf " :",
cb41e5d6 2211 depth+1, (UV)state );
3dab1dad 2212 if ( ! trie->states[ state ].wordnum ) {
6ad9a8ab 2213 Perl_re_printf( aTHX_ "%5s| ","");
3dab1dad 2214 } else {
6ad9a8ab 2215 Perl_re_printf( aTHX_ "W%4x| ",
3dab1dad
YO
2216 trie->states[ state ].wordnum
2217 );
2218 }
2219 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
538e84ed 2220 SV ** const tmp = av_fetch( revcharmap,
88f063b4 2221 TRIE_LIST_ITEM(state, charid).forid, 0);
ab3bbdeb 2222 if ( tmp ) {
147e3846 2223 Perl_re_printf( aTHX_ "%*s:%3X=%4" UVXf " | ",
ab3bbdeb 2224 colwidth,
538e84ed
KW
2225 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
2226 colwidth,
2227 PL_colors[0], PL_colors[1],
2228 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2229 | PERL_PV_ESCAPE_FIRSTCHAR
ab3bbdeb 2230 ) ,
88f063b4
KW
2231 TRIE_LIST_ITEM(state, charid).forid,
2232 (UV)TRIE_LIST_ITEM(state, charid).newstate
1e2e3d02 2233 );
538e84ed 2234 if (!(charid % 10))
6ad9a8ab 2235 Perl_re_printf( aTHX_ "\n%*s| ",
664e119d 2236 (int)((depth * 2) + 14), "");
1e2e3d02 2237 }
ab3bbdeb 2238 }
6ad9a8ab 2239 Perl_re_printf( aTHX_ "\n");
3dab1dad 2240 }
538e84ed 2241}
3dab1dad
YO
2242
2243/*
3dab1dad 2244 Dumps a fully constructed but uncompressed trie in table form.
538e84ed
KW
2245 This is the normal DFA style state transition table, with a few
2246 twists to facilitate compression later.
3dab1dad
YO
2247 Used for debugging make_trie().
2248*/
2249STATIC void
55eed653 2250S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2251 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2252 U32 depth)
3dab1dad
YO
2253{
2254 U32 state;
2255 U16 charid;
ab3bbdeb 2256 SV *sv=sv_newmortal();
55eed653 2257 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2258 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2259
2260 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
538e84ed 2261
3dab1dad
YO
2262 /*
2263 print out the table precompression so that we can do a visual check
2264 that they are identical.
2265 */
538e84ed 2266
6ad9a8ab 2267 Perl_re_indentf( aTHX_ "Char : ", depth+1 );
3dab1dad
YO
2268
2269 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 2270 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 2271 if ( tmp ) {
6ad9a8ab 2272 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2273 colwidth,
538e84ed 2274 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2275 PL_colors[0], PL_colors[1],
2276 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2277 PERL_PV_ESCAPE_FIRSTCHAR
2278 )
ab3bbdeb 2279 );
3dab1dad
YO
2280 }
2281 }
2282
4aaafc03
YO
2283 Perl_re_printf( aTHX_ "\n");
2284 Perl_re_indentf( aTHX_ "State+-", depth+1 );
3dab1dad
YO
2285
2286 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
6ad9a8ab 2287 Perl_re_printf( aTHX_ "%.*s", colwidth,"--------");
3dab1dad
YO
2288 }
2289
6ad9a8ab 2290 Perl_re_printf( aTHX_ "\n" );
3dab1dad
YO
2291
2292 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
2293
147e3846 2294 Perl_re_indentf( aTHX_ "%4" UVXf " : ",
cb41e5d6 2295 depth+1,
3dab1dad
YO
2296 (UV)TRIE_NODENUM( state ) );
2297
2298 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
2299 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
2300 if (v)
147e3846 2301 Perl_re_printf( aTHX_ "%*" UVXf, colwidth, v );
ab3bbdeb 2302 else
6ad9a8ab 2303 Perl_re_printf( aTHX_ "%*s", colwidth, "." );
3dab1dad
YO
2304 }
2305 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
147e3846 2306 Perl_re_printf( aTHX_ " (%4" UVXf ")\n",
538e84ed 2307 (UV)trie->trans[ state ].check );
3dab1dad 2308 } else {
147e3846 2309 Perl_re_printf( aTHX_ " (%4" UVXf ") W%4X\n",
538e84ed 2310 (UV)trie->trans[ state ].check,
3dab1dad
YO
2311 trie->states[ TRIE_NODENUM( state ) ].wordnum );
2312 }
2313 }
07be1b83 2314}
3dab1dad
YO
2315
2316#endif
2317
2e64971a 2318
786e8c11
YO
2319/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
2320 startbranch: the first branch in the whole branch sequence
2321 first : start branch of sequence of branch-exact nodes.
2322 May be the same as startbranch
2323 last : Thing following the last branch.
2324 May be the same as tail.
2325 tail : item following the branch sequence
2326 count : words in the sequence
a4525e78 2327 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
786e8c11 2328 depth : indent depth
3dab1dad 2329
786e8c11 2330Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 2331
786e8c11
YO
2332A trie is an N'ary tree where the branches are determined by digital
2333decomposition of the key. IE, at the root node you look up the 1st character and
2334follow that branch repeat until you find the end of the branches. Nodes can be
2335marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 2336
786e8c11 2337 /he|she|his|hers/
72f13be8 2338
786e8c11
YO
2339would convert into the following structure. Numbers represent states, letters
2340following numbers represent valid transitions on the letter from that state, if
2341the number is in square brackets it represents an accepting state, otherwise it
2342will be in parenthesis.
07be1b83 2343
786e8c11
YO
2344 +-h->+-e->[3]-+-r->(8)-+-s->[9]
2345 | |
2346 | (2)
2347 | |
2348 (1) +-i->(6)-+-s->[7]
2349 |
2350 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 2351
786e8c11
YO
2352 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
2353
2354This shows that when matching against the string 'hers' we will begin at state 1
2355read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
2356then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
2357is also accepting. Thus we know that we can match both 'he' and 'hers' with a
2358single traverse. We store a mapping from accepting to state to which word was
2359matched, and then when we have multiple possibilities we try to complete the
b8fda935 2360rest of the regex in the order in which they occurred in the alternation.
786e8c11
YO
2361
2362The only prior NFA like behaviour that would be changed by the TRIE support is
2363the silent ignoring of duplicate alternations which are of the form:
2364
2365 / (DUPE|DUPE) X? (?{ ... }) Y /x
2366
4b714af6 2367Thus EVAL blocks following a trie may be called a different number of times with
786e8c11 2368and without the optimisation. With the optimisations dupes will be silently
486ec47a 2369ignored. This inconsistent behaviour of EVAL type nodes is well established as
786e8c11
YO
2370the following demonstrates:
2371
2372 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
2373
2374which prints out 'word' three times, but
2375
2376 'words'=~/(word|word|word)(?{ print $1 })S/
2377
2378which doesnt print it out at all. This is due to other optimisations kicking in.
2379
2380Example of what happens on a structural level:
2381
486ec47a 2382The regexp /(ac|ad|ab)+/ will produce the following debug output:
786e8c11
YO
2383
2384 1: CURLYM[1] {1,32767}(18)
2385 5: BRANCH(8)
2386 6: EXACT <ac>(16)
2387 8: BRANCH(11)
2388 9: EXACT <ad>(16)
2389 11: BRANCH(14)
2390 12: EXACT <ab>(16)
2391 16: SUCCEED(0)
2392 17: NOTHING(18)
2393 18: END(0)
2394
2395This would be optimizable with startbranch=5, first=5, last=16, tail=16
2396and should turn into:
2397
2398 1: CURLYM[1] {1,32767}(18)
2399 5: TRIE(16)
2400 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
2401 <ac>
2402 <ad>
2403 <ab>
2404 16: SUCCEED(0)
2405 17: NOTHING(18)
2406 18: END(0)
2407
2408Cases where tail != last would be like /(?foo|bar)baz/:
2409
2410 1: BRANCH(4)
2411 2: EXACT <foo>(8)
2412 4: BRANCH(7)
2413 5: EXACT <bar>(8)
2414 7: TAIL(8)
2415 8: EXACT <baz>(10)
2416 10: END(0)
2417
2418which would be optimizable with startbranch=1, first=1, last=7, tail=8
2419and would end up looking like:
2420
2421 1: TRIE(8)
2422 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
2423 <foo>
2424 <bar>
2425 7: TAIL(8)
2426 8: EXACT <baz>(10)
2427 10: END(0)
2428
c80e42f3 2429 d = uvchr_to_utf8_flags(d, uv, 0);
786e8c11
YO
2430
2431is the recommended Unicode-aware way of saying
2432
2433 *(d++) = uv;
2434*/
2435
fab2782b 2436#define TRIE_STORE_REVCHAR(val) \
786e8c11 2437 STMT_START { \
73031816 2438 if (UTF) { \
668fcfea 2439 SV *zlopp = newSV(UTF8_MAXBYTES); \
88c9ea1e 2440 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
c80e42f3 2441 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
73031816
NC
2442 SvCUR_set(zlopp, kapow - flrbbbbb); \
2443 SvPOK_on(zlopp); \
2444 SvUTF8_on(zlopp); \
2445 av_push(revcharmap, zlopp); \
2446 } else { \
fab2782b 2447 char ooooff = (char)val; \
73031816
NC
2448 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2449 } \
2450 } STMT_END
786e8c11 2451
914a25d5
KW
2452/* This gets the next character from the input, folding it if not already
2453 * folded. */
2454#define TRIE_READ_CHAR STMT_START { \
2455 wordlen++; \
2456 if ( UTF ) { \
2457 /* if it is UTF then it is either already folded, or does not need \
2458 * folding */ \
1c1d615a 2459 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
914a25d5
KW
2460 } \
2461 else if (folder == PL_fold_latin1) { \
7d006b13
KW
2462 /* This folder implies Unicode rules, which in the range expressible \
2463 * by not UTF is the lower case, with the two exceptions, one of \
2464 * which should have been taken care of before calling this */ \
2465 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2466 uvc = toLOWER_L1(*uc); \
2467 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2468 len = 1; \
914a25d5
KW
2469 } else { \
2470 /* raw data, will be folded later if needed */ \
2471 uvc = (U32)*uc; \
2472 len = 1; \
2473 } \
786e8c11
YO
2474} STMT_END
2475
2476
2477
2478#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2479 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
00195859 2480 U32 ging = TRIE_LIST_LEN( state ) * 2; \
f9003953 2481 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
00195859 2482 TRIE_LIST_LEN( state ) = ging; \
786e8c11
YO
2483 } \
2484 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2485 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2486 TRIE_LIST_CUR( state )++; \
2487} STMT_END
07be1b83 2488
786e8c11 2489#define TRIE_LIST_NEW(state) STMT_START { \
d09f14bf 2490 Newx( trie->states[ state ].trans.list, \
786e8c11
YO
2491 4, reg_trie_trans_le ); \
2492 TRIE_LIST_CUR( state ) = 1; \
2493 TRIE_LIST_LEN( state ) = 4; \
2494} STMT_END
07be1b83 2495
786e8c11
YO
2496#define TRIE_HANDLE_WORD(state) STMT_START { \
2497 U16 dupe= trie->states[ state ].wordnum; \
2498 regnode * const noper_next = regnext( noper ); \
2499 \
786e8c11
YO
2500 DEBUG_r({ \
2501 /* store the word for dumping */ \
2502 SV* tmp; \
2503 if (OP(noper) != NOTHING) \
740cce10 2504 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
786e8c11 2505 else \
740cce10 2506 tmp = newSVpvn_utf8( "", 0, UTF ); \
2b8b4781 2507 av_push( trie_words, tmp ); \
786e8c11
YO
2508 }); \
2509 \
2510 curword++; \
2e64971a
DM
2511 trie->wordinfo[curword].prev = 0; \
2512 trie->wordinfo[curword].len = wordlen; \
2513 trie->wordinfo[curword].accept = state; \
786e8c11
YO
2514 \
2515 if ( noper_next < tail ) { \
2516 if (!trie->jump) \
538e84ed
KW
2517 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2518 sizeof(U16) ); \
7f69552c 2519 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
2520 if (!jumper) \
2521 jumper = noper_next; \
2522 if (!nextbranch) \
2523 nextbranch= regnext(cur); \
2524 } \
2525 \
2526 if ( dupe ) { \
2e64971a
DM
2527 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2528 /* chain, so that when the bits of chain are later */\
2529 /* linked together, the dups appear in the chain */\
2530 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2531 trie->wordinfo[dupe].prev = curword; \
786e8c11
YO
2532 } else { \
2533 /* we haven't inserted this word yet. */ \
2534 trie->states[ state ].wordnum = curword; \
2535 } \
2536} STMT_END
07be1b83 2537
3dab1dad 2538
786e8c11
YO
2539#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2540 ( ( base + charid >= ucharcount \
2541 && base + charid < ubound \
2542 && state == trie->trans[ base - ucharcount + charid ].check \
2543 && trie->trans[ base - ucharcount + charid ].next ) \
2544 ? trie->trans[ base - ucharcount + charid ].next \
2545 : ( state==1 ? special : 0 ) \
2546 )
3dab1dad 2547
8bcafbf4
YO
2548#define TRIE_BITMAP_SET_FOLDED(trie, uvc, folder) \
2549STMT_START { \
2550 TRIE_BITMAP_SET(trie, uvc); \
2551 /* store the folded codepoint */ \
2552 if ( folder ) \
2553 TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); \
2554 \
2555 if ( !UTF ) { \
2556 /* store first byte of utf8 representation of */ \
2557 /* variant codepoints */ \
2558 if (! UVCHR_IS_INVARIANT(uvc)) { \
2559 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); \
2560 } \
2561 } \
2562} STMT_END
786e8c11
YO
2563#define MADE_TRIE 1
2564#define MADE_JUMP_TRIE 2
2565#define MADE_EXACT_TRIE 4
3dab1dad 2566
a3621e74 2567STATIC I32
538e84ed
KW
2568S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2569 regnode *first, regnode *last, regnode *tail,
2570 U32 word_count, U32 flags, U32 depth)
a3621e74
YO
2571{
2572 /* first pass, loop through and scan words */
2573 reg_trie_data *trie;
55eed653 2574 HV *widecharmap = NULL;
2b8b4781 2575 AV *revcharmap = newAV();
a3621e74 2576 regnode *cur;
a3621e74
YO
2577 STRLEN len = 0;
2578 UV uvc = 0;
2579 U16 curword = 0;
2580 U32 next_alloc = 0;
786e8c11
YO
2581 regnode *jumper = NULL;
2582 regnode *nextbranch = NULL;
7f69552c 2583 regnode *convert = NULL;
2e64971a 2584 U32 *prev_states; /* temp array mapping each state to previous one */
a3621e74 2585 /* we just use folder as a flag in utf8 */
1e696034 2586 const U8 * folder = NULL;
a3621e74 2587
3a611511
YO
2588 /* in the below add_data call we are storing either 'tu' or 'tuaa'
2589 * which stands for one trie structure, one hash, optionally followed
2590 * by two arrays */
2b8b4781 2591#ifdef DEBUGGING
3a611511 2592 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuaa"));
2b8b4781
NC
2593 AV *trie_words = NULL;
2594 /* along with revcharmap, this only used during construction but both are
2595 * useful during debugging so we store them in the struct when debugging.
8e11feef 2596 */
2b8b4781 2597#else
cf78de0b 2598 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
3dab1dad 2599 STRLEN trie_charcount=0;
3dab1dad 2600#endif
2b8b4781 2601 SV *re_trie_maxbuff;
a3621e74 2602 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2603
2604 PERL_ARGS_ASSERT_MAKE_TRIE;
72f13be8
YO
2605#ifndef DEBUGGING
2606 PERL_UNUSED_ARG(depth);
2607#endif
a3621e74 2608
1e696034 2609 switch (flags) {
a4525e78 2610 case EXACT: case EXACTL: break;
89829bb5 2611 case EXACTFAA:
fab2782b 2612 case EXACTFU_SS:
a4525e78
KW
2613 case EXACTFU:
2614 case EXACTFLU8: folder = PL_fold_latin1; break;
1e696034 2615 case EXACTF: folder = PL_fold; break;
fab2782b 2616 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
1e696034
KW
2617 }
2618
c944940b 2619 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 2620 trie->refcount = 1;
3dab1dad 2621 trie->startstate = 1;
786e8c11 2622 trie->wordcount = word_count;
f8fc2ecf 2623 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 2624 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
a4525e78 2625 if (flags == EXACT || flags == EXACTL)
c944940b 2626 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2e64971a
DM
2627 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2628 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2629
a3621e74 2630 DEBUG_r({
2b8b4781 2631 trie_words = newAV();
a3621e74 2632 });
a3621e74 2633
0111c4fd 2634 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
316ebaf2 2635 assert(re_trie_maxbuff);
a3621e74 2636 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 2637 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 2638 }
df826430 2639 DEBUG_TRIE_COMPILE_r({
6ad9a8ab 2640 Perl_re_indentf( aTHX_
cb41e5d6
YO
2641 "make_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2642 depth+1,
88f063b4 2643 REG_NODE_NUM(startbranch), REG_NODE_NUM(first),
538e84ed 2644 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
3dab1dad 2645 });
538e84ed 2646
7f69552c
YO
2647 /* Find the node we are going to overwrite */
2648 if ( first == startbranch && OP( last ) != BRANCH ) {
2649 /* whole branch chain */
2650 convert = first;
2651 } else {
2652 /* branch sub-chain */
2653 convert = NEXTOPER( first );
2654 }
538e84ed 2655
a3621e74
YO
2656 /* -- First loop and Setup --
2657
2658 We first traverse the branches and scan each word to determine if it
2659 contains widechars, and how many unique chars there are, this is
2660 important as we have to build a table with at least as many columns as we
2661 have unique chars.
2662
2663 We use an array of integers to represent the character codes 0..255
538e84ed
KW
2664 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2665 the native representation of the character value as the key and IV's for
2666 the coded index.
a3621e74
YO
2667
2668 *TODO* If we keep track of how many times each character is used we can
2669 remap the columns so that the table compression later on is more
3b753521 2670 efficient in terms of memory by ensuring the most common value is in the
a3621e74
YO
2671 middle and the least common are on the outside. IMO this would be better
2672 than a most to least common mapping as theres a decent chance the most
2673 common letter will share a node with the least common, meaning the node
486ec47a 2674 will not be compressible. With a middle is most common approach the worst
a3621e74
YO
2675 case is when we have the least common nodes twice.
2676
2677 */
2678
a3621e74 2679 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
df826430 2680 regnode *noper = NEXTOPER( cur );
944e05e3
YO
2681 const U8 *uc;
2682 const U8 *e;
bc031a7d 2683 int foldlen = 0;
07be1b83 2684 U32 wordlen = 0; /* required init */
bc031a7d
KW
2685 STRLEN minchars = 0;
2686 STRLEN maxchars = 0;
538e84ed
KW
2687 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2688 bitmap?*/
a3621e74 2689
3dab1dad 2690 if (OP(noper) == NOTHING) {
20ed8c88
YO
2691 /* skip past a NOTHING at the start of an alternation
2692 * eg, /(?:)a|(?:b)/ should be the same as /a|b/
2693 */
df826430 2694 regnode *noper_next= regnext(noper);
944e05e3
YO
2695 if (noper_next < tail)
2696 noper= noper_next;
2697 }
2698
dca5fc4c
YO
2699 if ( noper < tail &&
2700 (
2701 OP(noper) == flags ||
2702 (
2703 flags == EXACTFU &&
2704 OP(noper) == EXACTFU_SS
2705 )
2706 )
2707 ) {
944e05e3
YO
2708 uc= (U8*)STRING(noper);
2709 e= uc + STR_LEN(noper);
2710 } else {
2711 trie->minlen= 0;
2712 continue;
3dab1dad 2713 }
df826430 2714
944e05e3 2715
fab2782b 2716 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
02daf0ab
YO
2717 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2718 regardless of encoding */
fab2782b
YO
2719 if (OP( noper ) == EXACTFU_SS) {
2720 /* false positives are ok, so just set this */
0dc4a61d 2721 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
fab2782b
YO
2722 }
2723 }
dca5fc4c 2724
bc031a7d
KW
2725 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2726 branch */
3dab1dad 2727 TRIE_CHARCOUNT(trie)++;
a3621e74 2728 TRIE_READ_CHAR;
645de4ce 2729
bc031a7d
KW
2730 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2731 * is in effect. Under /i, this character can match itself, or
2732 * anything that folds to it. If not under /i, it can match just
2733 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2734 * all fold to k, and all are single characters. But some folds
2735 * expand to more than one character, so for example LATIN SMALL
2736 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2737 * the string beginning at 'uc' is 'ffi', it could be matched by
2738 * three characters, or just by the one ligature character. (It
2739 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2740 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2741 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2742 * match.) The trie needs to know the minimum and maximum number
2743 * of characters that could match so that it can use size alone to
2744 * quickly reject many match attempts. The max is simple: it is
2745 * the number of folded characters in this branch (since a fold is
2746 * never shorter than what folds to it. */
2747
2748 maxchars++;
2749
2750 /* And the min is equal to the max if not under /i (indicated by
2751 * 'folder' being NULL), or there are no multi-character folds. If
2752 * there is a multi-character fold, the min is incremented just
2753 * once, for the character that folds to the sequence. Each
2754 * character in the sequence needs to be added to the list below of
2755 * characters in the trie, but we count only the first towards the
2756 * min number of characters needed. This is done through the
2757 * variable 'foldlen', which is returned by the macros that look
2758 * for these sequences as the number of bytes the sequence
2759 * occupies. Each time through the loop, we decrement 'foldlen' by
2760 * how many bytes the current char occupies. Only when it reaches
2761 * 0 do we increment 'minchars' or look for another multi-character
2762 * sequence. */
2763 if (folder == NULL) {
2764 minchars++;
2765 }
2766 else if (foldlen > 0) {
2767 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
645de4ce
KW
2768 }
2769 else {
bc031a7d
KW
2770 minchars++;
2771
2772 /* See if *uc is the beginning of a multi-character fold. If
2773 * so, we decrement the length remaining to look at, to account
2774 * for the current character this iteration. (We can use 'uc'
2775 * instead of the fold returned by TRIE_READ_CHAR because for
2776 * non-UTF, the latin1_safe macro is smart enough to account
2777 * for all the unfolded characters, and because for UTF, the
2778 * string will already have been folded earlier in the
2779 * compilation process */
2780 if (UTF) {
2781 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2782 foldlen -= UTF8SKIP(uc);
645de4ce
KW
2783 }
2784 }
bc031a7d
KW
2785 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2786 foldlen--;
2787 }
645de4ce 2788 }
bc031a7d
KW
2789
2790 /* The current character (and any potential folds) should be added
2791 * to the possible matching characters for this position in this
2792 * branch */
a3621e74 2793 if ( uvc < 256 ) {
fab2782b
YO
2794 if ( folder ) {
2795 U8 folded= folder[ (U8) uvc ];
2796 if ( !trie->charmap[ folded ] ) {
2797 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2798 TRIE_STORE_REVCHAR( folded );
2799 }
2800 }
a3621e74
YO
2801 if ( !trie->charmap[ uvc ] ) {
2802 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
fab2782b 2803 TRIE_STORE_REVCHAR( uvc );
a3621e74 2804 }
02daf0ab 2805 if ( set_bit ) {
62012aee
KW
2806 /* store the codepoint in the bitmap, and its folded
2807 * equivalent. */
8bcafbf4 2808 TRIE_BITMAP_SET_FOLDED(trie, uvc, folder);
02daf0ab
YO
2809 set_bit = 0; /* We've done our bit :-) */
2810 }
a3621e74 2811 } else {
bc031a7d
KW
2812
2813 /* XXX We could come up with the list of code points that fold
2814 * to this using PL_utf8_foldclosures, except not for
2815 * multi-char folds, as there may be multiple combinations
2816 * there that could work, which needs to wait until runtime to
2817 * resolve (The comment about LIGATURE FFI above is such an
2818 * example */
2819
a3621e74 2820 SV** svpp;
55eed653
NC
2821 if ( !widecharmap )
2822 widecharmap = newHV();
a3621e74 2823
55eed653 2824 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
2825
2826 if ( !svpp )
147e3846 2827 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%" UVXf, uvc );
a3621e74
YO
2828
2829 if ( !SvTRUE( *svpp ) ) {
2830 sv_setiv( *svpp, ++trie->uniquecharcount );
fab2782b 2831 TRIE_STORE_REVCHAR(uvc);
a3621e74
YO
2832 }
2833 }
bc031a7d
KW
2834 } /* end loop through characters in this branch of the trie */
2835
2836 /* We take the min and max for this branch and combine to find the min
2837 * and max for all branches processed so far */
3dab1dad 2838 if( cur == first ) {
bc031a7d
KW
2839 trie->minlen = minchars;
2840 trie->maxlen = maxchars;
2841 } else if (minchars < trie->minlen) {
2842 trie->minlen = minchars;
2843 } else if (maxchars > trie->maxlen) {
2844 trie->maxlen = maxchars;
fab2782b 2845 }
a3621e74
YO
2846 } /* end first pass */
2847 DEBUG_TRIE_COMPILE_r(
6ad9a8ab 2848 Perl_re_indentf( aTHX_
cb41e5d6
YO
2849 "TRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2850 depth+1,
55eed653 2851 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
2852 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2853 (int)trie->minlen, (int)trie->maxlen )
a3621e74 2854 );
a3621e74
YO
2855
2856 /*
2857 We now know what we are dealing with in terms of unique chars and
2858 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
2859 representation using a flat table will take. If it's over a reasonable
2860 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
2861 conservative but potentially much slower representation using an array
2862 of lists.
2863
2864 At the end we convert both representations into the same compressed
2865 form that will be used in regexec.c for matching with. The latter
2866 is a form that cannot be used to construct with but has memory
2867 properties similar to the list form and access properties similar
2868 to the table form making it both suitable for fast searches and
2869 small enough that its feasable to store for the duration of a program.
2870
2871 See the comment in the code where the compressed table is produced
2872 inplace from the flat tabe representation for an explanation of how
2873 the compression works.
2874
2875 */
2876
2877
2e64971a
DM
2878 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2879 prev_states[1] = 0;
2880
538e84ed
KW
2881 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2882 > SvIV(re_trie_maxbuff) )
2883 {
a3621e74
YO
2884 /*
2885 Second Pass -- Array Of Lists Representation
2886
2887 Each state will be represented by a list of charid:state records
2888 (reg_trie_trans_le) the first such element holds the CUR and LEN
2889 points of the allocated array. (See defines above).
2890
2891 We build the initial structure using the lists, and then convert
2892 it into the compressed table form which allows faster lookups
2893 (but cant be modified once converted).
a3621e74
YO
2894 */
2895
a3621e74
YO
2896 STRLEN transcount = 1;
2897
6ad9a8ab 2898 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using list compiler\n",
cb41e5d6 2899 depth+1));
686b73d4 2900
c944940b
JH
2901 trie->states = (reg_trie_state *)
2902 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2903 sizeof(reg_trie_state) );
a3621e74
YO
2904 TRIE_LIST_NEW(1);
2905 next_alloc = 2;
2906
2907 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2908
df826430 2909 regnode *noper = NEXTOPER( cur );
c445ea15
AL
2910 U32 state = 1; /* required init */
2911 U16 charid = 0; /* sanity init */
07be1b83 2912 U32 wordlen = 0; /* required init */
c445ea15 2913
df826430
YO
2914 if (OP(noper) == NOTHING) {
2915 regnode *noper_next= regnext(noper);
944e05e3
YO
2916 if (noper_next < tail)
2917 noper= noper_next;
df826430
YO
2918 }
2919
944e05e3
YO
2920 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
2921 const U8 *uc= (U8*)STRING(noper);
2922 const U8 *e= uc + STR_LEN(noper);
2923
786e8c11 2924 for ( ; uc < e ; uc += len ) {
c445ea15 2925
786e8c11 2926 TRIE_READ_CHAR;
c445ea15 2927
786e8c11
YO
2928 if ( uvc < 256 ) {
2929 charid = trie->charmap[ uvc ];
c445ea15 2930 } else {
538e84ed
KW
2931 SV** const svpp = hv_fetch( widecharmap,
2932 (char*)&uvc,
2933 sizeof( UV ),
2934 0);
786e8c11
YO
2935 if ( !svpp ) {
2936 charid = 0;
2937 } else {
2938 charid=(U16)SvIV( *svpp );
2939 }
c445ea15 2940 }
538e84ed
KW
2941 /* charid is now 0 if we dont know the char read, or
2942 * nonzero if we do */
786e8c11 2943 if ( charid ) {
a3621e74 2944
786e8c11
YO
2945 U16 check;
2946 U32 newstate = 0;
a3621e74 2947
786e8c11
YO
2948 charid--;
2949 if ( !trie->states[ state ].trans.list ) {
2950 TRIE_LIST_NEW( state );
c445ea15 2951 }
538e84ed
KW
2952 for ( check = 1;
2953 check <= TRIE_LIST_USED( state );
2954 check++ )
2955 {
2956 if ( TRIE_LIST_ITEM( state, check ).forid
2957 == charid )
2958 {
786e8c11
YO
2959 newstate = TRIE_LIST_ITEM( state, check ).newstate;
2960 break;
2961 }
2962 }
2963 if ( ! newstate ) {
2964 newstate = next_alloc++;
2e64971a 2965 prev_states[newstate] = state;
786e8c11
YO
2966 TRIE_LIST_PUSH( state, charid, newstate );
2967 transcount++;
2968 }
2969 state = newstate;
2970 } else {
147e3846 2971 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
c445ea15 2972 }
a28509cc 2973 }
c445ea15 2974 }
3dab1dad 2975 TRIE_HANDLE_WORD(state);
a3621e74
YO
2976
2977 } /* end second pass */
2978
1e2e3d02 2979 /* next alloc is the NEXT state to be allocated */
538e84ed 2980 trie->statecount = next_alloc;
c944940b
JH
2981 trie->states = (reg_trie_state *)
2982 PerlMemShared_realloc( trie->states,
2983 next_alloc
2984 * sizeof(reg_trie_state) );
a3621e74 2985
3dab1dad 2986 /* and now dump it out before we compress it */
2b8b4781
NC
2987 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
2988 revcharmap, next_alloc,
2989 depth+1)
1e2e3d02 2990 );
a3621e74 2991
c944940b
JH
2992 trie->trans = (reg_trie_trans *)
2993 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
2994 {
2995 U32 state;
a3621e74
YO
2996 U32 tp = 0;
2997 U32 zp = 0;
2998
2999
3000 for( state=1 ; state < next_alloc ; state ++ ) {
3001 U32 base=0;
3002
3003 /*
3004 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3005 Perl_re_printf( aTHX_ "tp: %d zp: %d ",tp,zp)
a3621e74
YO
3006 );
3007 */
3008
3009 if (trie->states[state].trans.list) {
3010 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
3011 U16 maxid=minid;
a28509cc 3012 U16 idx;
a3621e74
YO
3013
3014 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
3015 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
3016 if ( forid < minid ) {
3017 minid=forid;
3018 } else if ( forid > maxid ) {
3019 maxid=forid;
3020 }
a3621e74
YO
3021 }
3022 if ( transcount < tp + maxid - minid + 1) {
3023 transcount *= 2;
c944940b
JH
3024 trie->trans = (reg_trie_trans *)
3025 PerlMemShared_realloc( trie->trans,
446bd890
NC
3026 transcount
3027 * sizeof(reg_trie_trans) );
538e84ed
KW
3028 Zero( trie->trans + (transcount / 2),
3029 transcount / 2,
3030 reg_trie_trans );
a3621e74
YO
3031 }
3032 base = trie->uniquecharcount + tp - minid;
3033 if ( maxid == minid ) {
3034 U32 set = 0;
3035 for ( ; zp < tp ; zp++ ) {
3036 if ( ! trie->trans[ zp ].next ) {
3037 base = trie->uniquecharcount + zp - minid;
538e84ed
KW
3038 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
3039 1).newstate;
a3621e74
YO
3040 trie->trans[ zp ].check = state;
3041 set = 1;
3042 break;
3043 }
3044 }
3045 if ( !set ) {
538e84ed
KW
3046 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
3047 1).newstate;
a3621e74
YO
3048 trie->trans[ tp ].check = state;
3049 tp++;
3050 zp = tp;
3051 }
3052 } else {
3053 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
538e84ed
KW
3054 const U32 tid = base
3055 - trie->uniquecharcount
3056 + TRIE_LIST_ITEM( state, idx ).forid;
3057 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
3058 idx ).newstate;
a3621e74
YO
3059 trie->trans[ tid ].check = state;
3060 }
3061 tp += ( maxid - minid + 1 );
3062 }
3063 Safefree(trie->states[ state ].trans.list);
3064 }
3065 /*
3066 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3067 Perl_re_printf( aTHX_ " base: %d\n",base);
a3621e74
YO
3068 );
3069 */
3070 trie->states[ state ].trans.base=base;
3071 }
cc601c31 3072 trie->lasttrans = tp + 1;
a3621e74
YO
3073 }
3074 } else {
3075 /*
3076 Second Pass -- Flat Table Representation.
3077
b423522f
KW
3078 we dont use the 0 slot of either trans[] or states[] so we add 1 to
3079 each. We know that we will need Charcount+1 trans at most to store
3080 the data (one row per char at worst case) So we preallocate both
3081 structures assuming worst case.
a3621e74
YO
3082
3083 We then construct the trie using only the .next slots of the entry
3084 structs.
3085
b423522f
KW
3086 We use the .check field of the first entry of the node temporarily
3087 to make compression both faster and easier by keeping track of how
3088 many non zero fields are in the node.
a3621e74
YO
3089
3090 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
3091 transition.
3092
b423522f
KW
3093 There are two terms at use here: state as a TRIE_NODEIDX() which is
3094 a number representing the first entry of the node, and state as a
3095 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
3096 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
3097 if there are 2 entrys per node. eg:
a3621e74
YO
3098
3099 A B A B
3100 1. 2 4 1. 3 7
3101 2. 0 3 3. 0 5
3102 3. 0 0 5. 0 0
3103 4. 0 0 7. 0 0
3104
b423522f
KW
3105 The table is internally in the right hand, idx form. However as we
3106 also have to deal with the states array which is indexed by nodenum
3107 we have to use TRIE_NODENUM() to convert.
a3621e74
YO
3108
3109 */
6ad9a8ab 3110 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using table compiler\n",
cb41e5d6 3111 depth+1));
3dab1dad 3112
c944940b
JH
3113 trie->trans = (reg_trie_trans *)
3114 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
3115 * trie->uniquecharcount + 1,
3116 sizeof(reg_trie_trans) );
3117 trie->states = (reg_trie_state *)
3118 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
3119 sizeof(reg_trie_state) );
a3621e74
YO
3120 next_alloc = trie->uniquecharcount + 1;
3121
3dab1dad 3122
a3621e74
YO
3123 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
3124
df826430 3125 regnode *noper = NEXTOPER( cur );
a3621e74
YO
3126
3127 U32 state = 1; /* required init */
3128
3129 U16 charid = 0; /* sanity init */
3130 U32 accept_state = 0; /* sanity init */
a3621e74 3131
07be1b83 3132 U32 wordlen = 0; /* required init */
a3621e74 3133
df826430
YO
3134 if (OP(noper) == NOTHING) {
3135 regnode *noper_next= regnext(noper);
944e05e3
YO
3136 if (noper_next < tail)
3137 noper= noper_next;
df826430 3138 }
fab2782b 3139
944e05e3
YO
3140 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
3141 const U8 *uc= (U8*)STRING(noper);
3142 const U8 *e= uc + STR_LEN(noper);
3143
786e8c11 3144 for ( ; uc < e ; uc += len ) {
a3621e74 3145
786e8c11 3146 TRIE_READ_CHAR;
a3621e74 3147
786e8c11
YO
3148 if ( uvc < 256 ) {
3149 charid = trie->charmap[ uvc ];
3150 } else {
538e84ed
KW
3151 SV* const * const svpp = hv_fetch( widecharmap,
3152 (char*)&uvc,
3153 sizeof( UV ),
3154 0);
786e8c11 3155 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 3156 }
786e8c11
YO
3157 if ( charid ) {
3158 charid--;
3159 if ( !trie->trans[ state + charid ].next ) {
3160 trie->trans[ state + charid ].next = next_alloc;
3161 trie->trans[ state ].check++;
2e64971a
DM
3162 prev_states[TRIE_NODENUM(next_alloc)]
3163 = TRIE_NODENUM(state);
786e8c11
YO
3164 next_alloc += trie->uniquecharcount;
3165 }
3166 state = trie->trans[ state + charid ].next;
3167 } else {
147e3846 3168 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
786e8c11 3169 }
538e84ed
KW
3170 /* charid is now 0 if we dont know the char read, or
3171 * nonzero if we do */
a3621e74 3172 }
a3621e74 3173 }
3dab1dad
YO
3174 accept_state = TRIE_NODENUM( state );
3175 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
3176
3177 } /* end second pass */
3178
3dab1dad 3179 /* and now dump it out before we compress it */
2b8b4781
NC
3180 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
3181 revcharmap,
3182 next_alloc, depth+1));
a3621e74 3183
a3621e74
YO
3184 {
3185 /*
3186 * Inplace compress the table.*
3187
3188 For sparse data sets the table constructed by the trie algorithm will
3189 be mostly 0/FAIL transitions or to put it another way mostly empty.
3190 (Note that leaf nodes will not contain any transitions.)
3191
3192 This algorithm compresses the tables by eliminating most such
3193 transitions, at the cost of a modest bit of extra work during lookup:
3194
3195 - Each states[] entry contains a .base field which indicates the
3196 index in the state[] array wheres its transition data is stored.
3197
3b753521 3198 - If .base is 0 there are no valid transitions from that node.
a3621e74
YO
3199
3200 - If .base is nonzero then charid is added to it to find an entry in
3201 the trans array.
3202
3203 -If trans[states[state].base+charid].check!=state then the
3204 transition is taken to be a 0/Fail transition. Thus if there are fail
3205 transitions at the front of the node then the .base offset will point
3206 somewhere inside the previous nodes data (or maybe even into a node
3207 even earlier), but the .check field determines if the transition is
3208 valid.
3209
786e8c11 3210 XXX - wrong maybe?
a3621e74 3211 The following process inplace converts the table to the compressed
3b753521 3212 table: We first do not compress the root node 1,and mark all its
a3621e74 3213 .check pointers as 1 and set its .base pointer as 1 as well. This
3b753521
FN
3214 allows us to do a DFA construction from the compressed table later,
3215 and ensures that any .base pointers we calculate later are greater
3216 than 0.
a3621e74
YO
3217
3218 - We set 'pos' to indicate the first entry of the second node.
3219
3220 - We then iterate over the columns of the node, finding the first and
3221 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
3222 and set the .check pointers accordingly, and advance pos
3223 appropriately and repreat for the next node. Note that when we copy
3224 the next pointers we have to convert them from the original
3225 NODEIDX form to NODENUM form as the former is not valid post
3226 compression.
3227
3228 - If a node has no transitions used we mark its base as 0 and do not
3229 advance the pos pointer.
3230
3231 - If a node only has one transition we use a second pointer into the
3232 structure to fill in allocated fail transitions from other states.
3233 This pointer is independent of the main pointer and scans forward
3234 looking for null transitions that are allocated to a state. When it
3235 finds one it writes the single transition into the "hole". If the
786e8c11 3236 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
3237
3238 - Once compressed we can Renew/realloc the structures to release the
3239 excess space.
3240
3241 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
3242 specifically Fig 3.47 and the associated pseudocode.
3243
3244 demq
3245 */
a3b680e6 3246 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 3247 U32 state, charid;
a3621e74 3248 U32 pos = 0, zp=0;
1e2e3d02 3249 trie->statecount = laststate;
a3621e74
YO
3250
3251 for ( state = 1 ; state < laststate ; state++ ) {
3252 U8 flag = 0;
a28509cc
AL
3253 const U32 stateidx = TRIE_NODEIDX( state );
3254 const U32 o_used = trie->trans[ stateidx ].check;
3255 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
3256 trie->trans[ stateidx ].check = 0;
3257
538e84ed
KW
3258 for ( charid = 0;
3259 used && charid < trie->uniquecharcount;
3260 charid++ )
3261 {
a3621e74
YO
3262 if ( flag || trie->trans[ stateidx + charid ].next ) {
3263 if ( trie->trans[ stateidx + charid ].next ) {
3264 if (o_used == 1) {
3265 for ( ; zp < pos ; zp++ ) {
3266 if ( ! trie->trans[ zp ].next ) {
3267 break;
3268 }
3269 }
538e84ed
KW
3270 trie->states[ state ].trans.base
3271 = zp
3272 + trie->uniquecharcount
3273 - charid ;
3274 trie->trans[ zp ].next
3275 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
3276 + charid ].next );
a3621e74
YO
3277 trie->trans[ zp ].check = state;
3278 if ( ++zp > pos ) pos = zp;
3279 break;
3280 }
3281 used--;
3282 }
3283 if ( !flag ) {
3284 flag = 1;
538e84ed
KW
3285 trie->states[ state ].trans.base
3286 = pos + trie->uniquecharcount - charid ;
a3621e74 3287 }
538e84ed
KW
3288 trie->trans[ pos ].next
3289 = SAFE_TRIE_NODENUM(
3290 trie->trans[ stateidx + charid ].next );
a3621e74
YO
3291 trie->trans[ pos ].check = state;
3292 pos++;
3293 }
3294 }
3295 }
cc601c31 3296 trie->lasttrans = pos + 1;
c944940b
JH
3297 trie->states = (reg_trie_state *)
3298 PerlMemShared_realloc( trie->states, laststate
3299 * sizeof(reg_trie_state) );
a3621e74 3300 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3301 Perl_re_indentf( aTHX_ "Alloc: %d Orig: %" IVdf " elements, Final:%" IVdf ". Savings of %%%5.2f\n",
cb41e5d6 3302 depth+1,
538e84ed
KW
3303 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
3304 + 1 ),
3305 (IV)next_alloc,
3306 (IV)pos,
3307 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
a3621e74
YO
3308 );
3309
3310 } /* end table compress */
3311 }
1e2e3d02 3312 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3313 Perl_re_indentf( aTHX_ "Statecount:%" UVxf " Lasttrans:%" UVxf "\n",
cb41e5d6 3314 depth+1,
1e2e3d02
YO
3315 (UV)trie->statecount,
3316 (UV)trie->lasttrans)
3317 );
cc601c31 3318 /* resize the trans array to remove unused space */
c944940b
JH
3319 trie->trans = (reg_trie_trans *)
3320 PerlMemShared_realloc( trie->trans, trie->lasttrans
3321 * sizeof(reg_trie_trans) );
a3621e74 3322
538e84ed 3323 { /* Modify the program and insert the new TRIE node */
3dab1dad
YO
3324 U8 nodetype =(U8)(flags & 0xFF);
3325 char *str=NULL;
538e84ed 3326
07be1b83 3327#ifdef DEBUGGING
e62cc96a 3328 regnode *optimize = NULL;
7122b237
YO
3329#ifdef RE_TRACK_PATTERN_OFFSETS
3330
b57a0404
JH
3331 U32 mjd_offset = 0;
3332 U32 mjd_nodelen = 0;
7122b237
YO
3333#endif /* RE_TRACK_PATTERN_OFFSETS */
3334#endif /* DEBUGGING */
a3621e74 3335 /*
3dab1dad
YO
3336 This means we convert either the first branch or the first Exact,
3337 depending on whether the thing following (in 'last') is a branch
3338 or not and whther first is the startbranch (ie is it a sub part of
3339 the alternation or is it the whole thing.)
3b753521 3340 Assuming its a sub part we convert the EXACT otherwise we convert
3dab1dad 3341 the whole branch sequence, including the first.
a3621e74 3342 */
3dab1dad 3343 /* Find the node we are going to overwrite */
7f69552c 3344 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 3345 /* branch sub-chain */
3dab1dad 3346 NEXT_OFF( first ) = (U16)(last - first);
7122b237 3347#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
3348 DEBUG_r({
3349 mjd_offset= Node_Offset((convert));
3350 mjd_nodelen= Node_Length((convert));
3351 });
7122b237 3352#endif
7f69552c 3353 /* whole branch chain */
7122b237
YO
3354 }
3355#ifdef RE_TRACK_PATTERN_OFFSETS
3356 else {
7f69552c
YO
3357 DEBUG_r({
3358 const regnode *nop = NEXTOPER( convert );
3359 mjd_offset= Node_Offset((nop));
3360 mjd_nodelen= Node_Length((nop));
3361 });
07be1b83
YO
3362 }
3363 DEBUG_OPTIMISE_r(
147e3846 3364 Perl_re_indentf( aTHX_ "MJD offset:%" UVuf " MJD length:%" UVuf "\n",
cb41e5d6 3365 depth+1,
786e8c11 3366 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 3367 );
7122b237 3368#endif
538e84ed 3369 /* But first we check to see if there is a common prefix we can
3dab1dad
YO
3370 split out as an EXACT and put in front of the TRIE node. */
3371 trie->startstate= 1;
55eed653 3372 if ( trie->bitmap && !widecharmap && !trie->jump ) {
5ee57374
YO
3373 /* we want to find the first state that has more than
3374 * one transition, if that state is not the first state
3375 * then we have a common prefix which we can remove.
3376 */
3dab1dad 3377 U32 state;
1e2e3d02 3378 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
a3621e74 3379 U32 ofs = 0;
ee0dfd0b
YO
3380 I32 first_ofs = -1; /* keeps track of the ofs of the first
3381 transition, -1 means none */
8e11feef
RGS
3382 U32 count = 0;
3383 const U32 base = trie->states[ state ].trans.base;
a3621e74 3384
5ee57374 3385 /* does this state terminate an alternation? */
3dab1dad 3386 if ( trie->states[state].wordnum )
8e11feef 3387 count = 1;
a3621e74 3388
8e11feef 3389 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
cc601c31
YO
3390 if ( ( base + ofs >= trie->uniquecharcount ) &&
3391 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
a3621e74
YO
3392 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
3393 {
3dab1dad 3394 if ( ++count > 1 ) {
5ee57374 3395 /* we have more than one transition */
d3d91c7c
YO
3396 SV **tmp;
3397 U8 *ch;
5ee57374
YO
3398 /* if this is the first state there is no common prefix
3399 * to extract, so we can exit */
8e11feef 3400 if ( state == 1 ) break;
d3d91c7c
YO
3401 tmp = av_fetch( revcharmap, ofs, 0);
3402 ch = (U8*)SvPV_nolen_const( *tmp );
3403
5ee57374
YO
3404 /* if we are on count 2 then we need to initialize the
3405 * bitmap, and store the previous char if there was one
3406 * in it*/
3dab1dad 3407 if ( count == 2 ) {
5ee57374 3408 /* clear the bitmap */
3dab1dad
YO
3409 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
3410 DEBUG_OPTIMISE_r(
147e3846 3411 Perl_re_indentf( aTHX_ "New Start State=%" UVuf " Class: [",
cb41e5d6 3412 depth+1,
786e8c11 3413 (UV)state));
ee0dfd0b
YO
3414 if (first_ofs >= 0) {
3415 SV ** const tmp = av_fetch( revcharmap, first_ofs, 0);
be8e71aa 3416 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 3417
88f063b4 3418 TRIE_BITMAP_SET_FOLDED(trie,*ch, folder);
3dab1dad 3419 DEBUG_OPTIMISE_r(
6ad9a8ab 3420 Perl_re_printf( aTHX_ "%s", (char*)ch)
3dab1dad 3421 );
8e11feef
RGS
3422 }
3423 }
8bcafbf4 3424 /* store the current firstchar in the bitmap */
88f063b4 3425 TRIE_BITMAP_SET_FOLDED(trie,*ch, folder);
6ad9a8ab 3426 DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "%s", ch));
8e11feef 3427 }
ee0dfd0b 3428 first_ofs = ofs;
8e11feef 3429 }
3dab1dad
YO
3430 }
3431 if ( count == 1 ) {
5ee57374
YO
3432 /* This state has only one transition, its transition is part
3433 * of a common prefix - we need to concatenate the char it
3434 * represents to what we have so far. */