This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: Use safe UTF8SKIP
[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
c277df42 77#define REG_COMP_C
54df2634
NC
78#ifdef PERL_IN_XSUB_RE
79# include "re_comp.h"
afda64e8 80EXTERN_C const struct regexp_engine my_reg_engine;
54df2634
NC
81#else
82# include "regcomp.h"
83#endif
a687059c 84
f7e03a10 85#include "dquote_inline.h"
b992490d 86#include "invlist_inline.h"
1b0f46bf 87#include "unicode_constants.h"
04e98a4d 88
538e84ed
KW
89#define HAS_NONLATIN1_FOLD_CLOSURE(i) \
90 _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
f12c0118
KW
91#define HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(i) \
92 _HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
26faadbd 93#define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
2c61f163 94#define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
94dc5c2d 95
a687059c
LW
96#ifndef STATIC
97#define STATIC static
98#endif
99
3f910e62
YO
100/* this is a chain of data about sub patterns we are processing that
101 need to be handled separately/specially in study_chunk. Its so
102 we can simulate recursion without losing state. */
103struct scan_frame;
104typedef struct scan_frame {
105 regnode *last_regnode; /* last node to process in this frame */
106 regnode *next_regnode; /* next node to process when last is reached */
107 U32 prev_recursed_depth;
108 I32 stopparen; /* what stopparen do we use */
3f910e62
YO
109
110 struct scan_frame *this_prev_frame; /* this previous frame */
111 struct scan_frame *prev_frame; /* previous frame */
112 struct scan_frame *next_frame; /* next frame */
113} scan_frame;
114
f4ae5a27
KW
115/* Certain characters are output as a sequence with the first being a
116 * backslash. */
eb7486de 117#define isBACKSLASHED_PUNCT(c) strchr("-[]\\^", c)
f4ae5a27
KW
118
119
09b2b2e6 120struct RExC_state_t {
514a91f1
DM
121 U32 flags; /* RXf_* are we folding, multilining? */
122 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
830247a4 123 char *precomp; /* uncompiled string. */
711b303b 124 char *precomp_end; /* pointer to end of uncompiled string. */
288b8c02 125 REGEXP *rx_sv; /* The SV that is the regexp. */
f8fc2ecf 126 regexp *rx; /* perl core regexp structure */
538e84ed
KW
127 regexp_internal *rxi; /* internal data for regexp object
128 pprivate field */
fac92740 129 char *start; /* Start of input for compile */
830247a4
IZ
130 char *end; /* End of input for compile */
131 char *parse; /* Input-scan pointer. */
51684e3c
KW
132 char *copy_start; /* start of copy of input within
133 constructed parse string */
134 char *copy_start_in_input; /* Position in input string
135 corresponding to copy_start */
ea3daa5d 136 SSize_t whilem_seen; /* number of WHILEM in this expr */
fac92740 137 regnode *emit_start; /* Start of emitted-code area */
f55b7b26 138 regnode_offset emit; /* Code-emit pointer */
830247a4
IZ
139 I32 naughty; /* How bad is this pattern? */
140 I32 sawback; /* Did we see \1, ...? */
141 U32 seen;
88f063b4
KW
142 SSize_t size; /* Number of regnode equivalents in
143 pattern */
7c932d07
KW
144
145 /* position beyond 'precomp' of the warning message furthest away from
146 * 'precomp'. During the parse, no warnings are raised for any problems
147 * earlier in the parse than this position. This works if warnings are
148 * raised the first time a given spot is parsed, and if only one
149 * independent warning is raised for any given spot */
150 Size_t latest_warn_offset;
151
152 I32 npar; /* Capture buffer count so far in the
153 parse, (OPEN) plus one. ("par" 0 is
154 the whole pattern)*/
155 I32 total_par; /* During initial parse, is either 0,
156 or -1; the latter indicating a
157 reparse is needed. After that pass,
158 it is what 'npar' became after the
159 pass. Hence, it being > 0 indicates
160 we are in a reparse situation */
538e84ed
KW
161 I32 nestroot; /* root parens we are in - used by
162 accept */
830247a4 163 I32 seen_zerolen;
f55b7b26
KW
164 regnode_offset *open_parens; /* offsets to open parens */
165 regnode_offset *close_parens; /* offsets to close parens */
0275405c 166 I32 parens_buf_size; /* #slots malloced open/close_parens */
d5a00e4a 167 regnode *end_op; /* END node in program */
02daf0ab
YO
168 I32 utf8; /* whether the pattern is utf8 or not */
169 I32 orig_utf8; /* whether the pattern was originally in utf8 */
170 /* XXX use this for future optimisation of case
171 * where pattern must be upgraded to utf8. */
e40e74fe
KW
172 I32 uni_semantics; /* If a d charset modifier should use unicode
173 rules, even if the pattern is not in
174 utf8 */
81714fb9 175 HV *paren_names; /* Paren names */
538e84ed 176
40d049e4 177 regnode **recurse; /* Recurse regops */
88f063b4 178 I32 recurse_count; /* Number of recurse regops we have generated */
4286711a 179 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
538e84ed 180 through */
09a65838 181 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
b57e4118 182 I32 in_lookbehind;
4624b182 183 I32 contains_locale;
bb3f3ed2 184 I32 override_recoding;
b6d67071
KW
185#ifdef EBCDIC
186 I32 recode_x_to_native;
187#endif
9d53c457 188 I32 in_multi_char_class;
1acab4c5 189 struct reg_code_blocks *code_blocks;/* positions of literal (?{})
68e2671b 190 within pattern */
b1603ef8 191 int code_index; /* next code_blocks[] slot */
ee273784 192 SSize_t maxlen; /* mininum possible number of chars in string to match */
3f910e62
YO
193 scan_frame *frame_head;
194 scan_frame *frame_last;
195 U32 frame_count;
7eec73eb 196 AV *warn_text;
945961fd 197 HV *unlexed_names;
dc6d7f5c 198#ifdef ADD_TO_REGEXEC
830247a4
IZ
199 char *starttry; /* -Dr: where regtry was called. */
200#define RExC_starttry (pRExC_state->starttry)
201#endif
d24ca0c5 202 SV *runtime_code_qr; /* qr with the runtime code blocks */
3dab1dad 203#ifdef DEBUGGING
be8e71aa 204 const char *lastparse;
3dab1dad 205 I32 lastnum;
1f1031fe 206 AV *paren_name_list; /* idx -> name */
d9a72fcc 207 U32 study_chunk_recursed_count;
c9f0d54c
YO
208 SV *mysv1;
209 SV *mysv2;
88f063b4 210
3dab1dad
YO
211#define RExC_lastparse (pRExC_state->lastparse)
212#define RExC_lastnum (pRExC_state->lastnum)
1f1031fe 213#define RExC_paren_name_list (pRExC_state->paren_name_list)
d9a72fcc 214#define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
c9f0d54c
YO
215#define RExC_mysv (pRExC_state->mysv1)
216#define RExC_mysv1 (pRExC_state->mysv1)
217#define RExC_mysv2 (pRExC_state->mysv2)
218
3dab1dad 219#endif
bd0b62a6 220 bool seen_d_op;
911bd04e 221 bool strict;
da7cf1cc 222 bool study_started;
034602eb 223 bool in_script_run;
7c932d07 224 bool use_BRANCHJ;
09b2b2e6 225};
830247a4 226
e2509266 227#define RExC_flags (pRExC_state->flags)
514a91f1 228#define RExC_pm_flags (pRExC_state->pm_flags)
830247a4 229#define RExC_precomp (pRExC_state->precomp)
51684e3c
KW
230#define RExC_copy_start_in_input (pRExC_state->copy_start_in_input)
231#define RExC_copy_start_in_constructed (pRExC_state->copy_start)
711b303b 232#define RExC_precomp_end (pRExC_state->precomp_end)
288b8c02 233#define RExC_rx_sv (pRExC_state->rx_sv)
830247a4 234#define RExC_rx (pRExC_state->rx)
f8fc2ecf 235#define RExC_rxi (pRExC_state->rxi)
fac92740 236#define RExC_start (pRExC_state->start)
830247a4
IZ
237#define RExC_end (pRExC_state->end)
238#define RExC_parse (pRExC_state->parse)
7c932d07 239#define RExC_latest_warn_offset (pRExC_state->latest_warn_offset )
830247a4 240#define RExC_whilem_seen (pRExC_state->whilem_seen)
bd0b62a6
KW
241#define RExC_seen_d_op (pRExC_state->seen_d_op) /* Seen something that differs
242 under /d from /u ? */
512c0f5a 243
512c0f5a 244
7122b237 245#ifdef RE_TRACK_PATTERN_OFFSETS
44f36cd3 246# define RExC_offsets (RExC_rxi->u.offsets) /* I am not like the
538e84ed 247 others */
7122b237 248#endif
830247a4 249#define RExC_emit (pRExC_state->emit)
fac92740 250#define RExC_emit_start (pRExC_state->emit_start)
830247a4
IZ
251#define RExC_sawback (pRExC_state->sawback)
252#define RExC_seen (pRExC_state->seen)
253#define RExC_size (pRExC_state->size)
ee273784 254#define RExC_maxlen (pRExC_state->maxlen)
830247a4 255#define RExC_npar (pRExC_state->npar)
1533d1d2 256#define RExC_total_parens (pRExC_state->total_par)
0275405c 257#define RExC_parens_buf_size (pRExC_state->parens_buf_size)
e2e6a0f1 258#define RExC_nestroot (pRExC_state->nestroot)
830247a4 259#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
1aa99e6b 260#define RExC_utf8 (pRExC_state->utf8)
e40e74fe 261#define RExC_uni_semantics (pRExC_state->uni_semantics)
02daf0ab 262#define RExC_orig_utf8 (pRExC_state->orig_utf8)
40d049e4
YO
263#define RExC_open_parens (pRExC_state->open_parens)
264#define RExC_close_parens (pRExC_state->close_parens)
5bd2d46e 265#define RExC_end_op (pRExC_state->end_op)
81714fb9 266#define RExC_paren_names (pRExC_state->paren_names)
40d049e4
YO
267#define RExC_recurse (pRExC_state->recurse)
268#define RExC_recurse_count (pRExC_state->recurse_count)
09a65838 269#define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
538e84ed
KW
270#define RExC_study_chunk_recursed_bytes \
271 (pRExC_state->study_chunk_recursed_bytes)
b57e4118 272#define RExC_in_lookbehind (pRExC_state->in_lookbehind)
4624b182 273#define RExC_contains_locale (pRExC_state->contains_locale)
b6d67071
KW
274#ifdef EBCDIC
275# define RExC_recode_x_to_native (pRExC_state->recode_x_to_native)
276#endif
9d53c457 277#define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
3f910e62
YO
278#define RExC_frame_head (pRExC_state->frame_head)
279#define RExC_frame_last (pRExC_state->frame_last)
280#define RExC_frame_count (pRExC_state->frame_count)
67cdf558 281#define RExC_strict (pRExC_state->strict)
da7cf1cc 282#define RExC_study_started (pRExC_state->study_started)
7eec73eb 283#define RExC_warn_text (pRExC_state->warn_text)
034602eb 284#define RExC_in_script_run (pRExC_state->in_script_run)
7c932d07 285#define RExC_use_BRANCHJ (pRExC_state->use_BRANCHJ)
945961fd 286#define RExC_unlexed_names (pRExC_state->unlexed_names)
830247a4 287
99807a43
HS
288/* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
289 * a flag to disable back-off on the fixed/floating substrings - if it's
290 * a high complexity pattern we assume the benefit of avoiding a full match
291 * is worth the cost of checking for the substrings even if they rarely help.
292 */
293#define RExC_naughty (pRExC_state->naughty)
294#define TOO_NAUGHTY (10)
295#define MARK_NAUGHTY(add) \
296 if (RExC_naughty < TOO_NAUGHTY) \
297 RExC_naughty += (add)
298#define MARK_NAUGHTY_EXP(exp, add) \
299 if (RExC_naughty < TOO_NAUGHTY) \
300 RExC_naughty += RExC_naughty / (exp) + (add)
cde0cee5 301
a687059c
LW
302#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
303#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
412f55bb 304 ((*s) == '{' && regcurly(s)))
a687059c
LW
305
306/*
307 * Flags to be passed up and down.
308 */
a687059c 309#define WORST 0 /* Worst case. */
c4ead770
KW
310#define HASWIDTH 0x01 /* Known to not match null strings, could match
311 non-null ones. */
fda99bee 312
e64f369d 313/* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
2fd92675
KW
314 * character. (There needs to be a case: in the switch statement in regexec.c
315 * for any node marked SIMPLE.) Note that this is not the same thing as
316 * REGNODE_SIMPLE */
fda99bee 317#define SIMPLE 0x02
e64f369d 318#define SPSTART 0x04 /* Starts with * or + */
8d9c2815
NC
319#define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
320#define TRYAGAIN 0x10 /* Weeded out a declaration. */
c36361b2
KW
321#define RESTART_PARSE 0x20 /* Need to redo the parse */
322#define NEED_UTF8 0x40 /* In conjunction with RESTART_PARSE, need to
323 calcuate sizes as UTF-8 */
a687059c 324
3dab1dad
YO
325#define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
326
07be1b83
YO
327/* whether trie related optimizations are enabled */
328#if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
329#define TRIE_STUDY_OPT
786e8c11 330#define FULL_TRIE_STUDY
07be1b83
YO
331#define TRIE_STCLASS
332#endif
1de06328
YO
333
334
40d049e4
YO
335
336#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
337#define PBITVAL(paren) (1 << ((paren) & 7))
338#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
339#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
340#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
341
82a6ada4 342#define REQUIRE_UTF8(flagp) STMT_START { \
8d9c2815 343 if (!UTF) { \
c36361b2 344 *flagp = RESTART_PARSE|NEED_UTF8; \
f55b7b26 345 return 0; \
8d9c2815 346 } \
82a6ada4 347 } STMT_END
40d049e4 348
d63d75f5 349/* Change from /d into /u rules, and restart the parse. RExC_uni_semantics is
f6e8f31e
KW
350 * a flag that indicates we need to override /d with /u as a result of
351 * something in the pattern. It should only be used in regards to calling
352 * set_regex_charset() or get_regex_charse() */
512c0f5a
KW
353#define REQUIRE_UNI_RULES(flagp, restart_retval) \
354 STMT_START { \
355 if (DEPENDS_SEMANTICS) { \
512c0f5a
KW
356 set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); \
357 RExC_uni_semantics = 1; \
b76e1c04 358 if (RExC_seen_d_op && LIKELY(! IN_PARENS_PASS)) { \
bd0b62a6
KW
359 /* No need to restart the parse if we haven't seen \
360 * anything that differs between /u and /d, and no need \
361 * to restart immediately if we're going to reparse \
362 * anyway to count parens */ \
c36361b2 363 *flagp |= RESTART_PARSE; \
512c0f5a 364 return restart_retval; \
d1be211c 365 } \
512c0f5a
KW
366 } \
367 } STMT_END
368
7c932d07
KW
369#define REQUIRE_BRANCHJ(flagp, restart_retval) \
370 STMT_START { \
371 RExC_use_BRANCHJ = 1; \
b76e1c04 372 if (LIKELY(! IN_PARENS_PASS)) { \
523a2924
KW
373 /* No need to restart the parse immediately if we're \
374 * going to reparse anyway to count parens */ \
375 *flagp |= RESTART_PARSE; \
376 return restart_retval; \
377 } \
7c932d07
KW
378 } STMT_END
379
b76e1c04
KW
380/* Until we have completed the parse, we leave RExC_total_parens at 0 or
381 * less. After that, it must always be positive, because the whole re is
382 * considered to be surrounded by virtual parens. Setting it to negative
383 * indicates there is some construct that needs to know the actual number of
384 * parens to be properly handled. And that means an extra pass will be
385 * required after we've counted them all */
386#define ALL_PARENS_COUNTED (RExC_total_parens > 0)
7c932d07 387#define REQUIRE_PARENS_PASS \
b76e1c04
KW
388 STMT_START { /* No-op if have completed a pass */ \
389 if (! ALL_PARENS_COUNTED) RExC_total_parens = -1; \
7c932d07 390 } STMT_END
b76e1c04
KW
391#define IN_PARENS_PASS (RExC_total_parens < 0)
392
7c932d07 393
f358228e
KW
394/* This is used to return failure (zero) early from the calling function if
395 * various flags in 'flags' are set. Two flags always cause a return:
396 * 'RESTART_PARSE' and 'NEED_UTF8'. 'extra' can be used to specify any
397 * additional flags that should cause a return; 0 if none. If the return will
398 * be done, '*flagp' is first set to be all of the flags that caused the
399 * return. */
400#define RETURN_FAIL_ON_RESTART_OR_FLAGS(flags,flagp,extra) \
fd7fc6ae 401 STMT_START { \
c36361b2
KW
402 if ((flags) & (RESTART_PARSE|NEED_UTF8|(extra))) { \
403 *(flagp) = (flags) & (RESTART_PARSE|NEED_UTF8|(extra)); \
f358228e 404 return 0; \
fd7fc6ae
YO
405 } \
406 } STMT_END
407
c36361b2 408#define MUST_RESTART(flags) ((flags) & (RESTART_PARSE))
fd7fc6ae 409
5d555bb8 410#define RETURN_FAIL_ON_RESTART(flags,flagp) \
f358228e 411 RETURN_FAIL_ON_RESTART_OR_FLAGS( flags, flagp, 0)
5d555bb8 412#define RETURN_FAIL_ON_RESTART_FLAGP(flagp) \
f358228e 413 if (MUST_RESTART(*(flagp))) return 0
fd7fc6ae 414
f19b1a63
KW
415/* This converts the named class defined in regcomp.h to its equivalent class
416 * number defined in handy.h. */
417#define namedclass_to_classnum(class) ((int) ((class) / 2))
418#define classnum_to_namedclass(classnum) ((classnum) * 2)
419
de92f5e6
KW
420#define _invlist_union_complement_2nd(a, b, output) \
421 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
422#define _invlist_intersection_complement_2nd(a, b, output) \
423 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
424
1de06328
YO
425/* About scan_data_t.
426
427 During optimisation we recurse through the regexp program performing
428 various inplace (keyhole style) optimisations. In addition study_chunk
429 and scan_commit populate this data structure with information about
538e84ed 430 what strings MUST appear in the pattern. We look for the longest
3b753521 431 string that must appear at a fixed location, and we look for the
1de06328
YO
432 longest string that may appear at a floating location. So for instance
433 in the pattern:
538e84ed 434
1de06328 435 /FOO[xX]A.*B[xX]BAR/
538e84ed 436
1de06328
YO
437 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
438 strings (because they follow a .* construct). study_chunk will identify
439 both FOO and BAR as being the longest fixed and floating strings respectively.
538e84ed 440
1de06328 441 The strings can be composites, for instance
538e84ed 442
1de06328 443 /(f)(o)(o)/
538e84ed 444
1de06328 445 will result in a composite fixed substring 'foo'.
538e84ed 446
1de06328 447 For each string some basic information is maintained:
538e84ed 448
6ea77169 449 - min_offset
1de06328
YO
450 This is the position the string must appear at, or not before.
451 It also implicitly (when combined with minlenp) tells us how many
3b753521
FN
452 characters must match before the string we are searching for.
453 Likewise when combined with minlenp and the length of the string it
538e84ed 454 tells us how many characters must appear after the string we have
1de06328 455 found.
538e84ed 456
1de06328
YO
457 - max_offset
458 Only used for floating strings. This is the rightmost point that
ea3daa5d 459 the string can appear at. If set to SSize_t_MAX it indicates that the
1de06328 460 string can occur infinitely far to the right.
6e12f832 461 For fixed strings, it is equal to min_offset.
538e84ed 462
1de06328 463 - minlenp
2d608413
KW
464 A pointer to the minimum number of characters of the pattern that the
465 string was found inside. This is important as in the case of positive
538e84ed 466 lookahead or positive lookbehind we can have multiple patterns
1de06328 467 involved. Consider
538e84ed 468
1de06328 469 /(?=FOO).*F/
538e84ed 470
1de06328
YO
471 The minimum length of the pattern overall is 3, the minimum length
472 of the lookahead part is 3, but the minimum length of the part that
538e84ed 473 will actually match is 1. So 'FOO's minimum length is 3, but the
1de06328 474 minimum length for the F is 1. This is important as the minimum length
538e84ed 475 is used to determine offsets in front of and behind the string being
1de06328 476 looked for. Since strings can be composites this is the length of the
486ec47a 477 pattern at the time it was committed with a scan_commit. Note that
1de06328 478 the length is calculated by study_chunk, so that the minimum lengths
538e84ed 479 are not known until the full pattern has been compiled, thus the
1de06328 480 pointer to the value.
538e84ed 481
1de06328 482 - lookbehind
538e84ed 483
1de06328 484 In the case of lookbehind the string being searched for can be
538e84ed 485 offset past the start point of the final matching string.
1de06328
YO
486 If this value was just blithely removed from the min_offset it would
487 invalidate some of the calculations for how many chars must match
488 before or after (as they are derived from min_offset and minlen and
538e84ed 489 the length of the string being searched for).
1de06328
YO
490 When the final pattern is compiled and the data is moved from the
491 scan_data_t structure into the regexp structure the information
538e84ed
KW
492 about lookbehind is factored in, with the information that would
493 have been lost precalculated in the end_shift field for the
1de06328
YO
494 associated string.
495
496 The fields pos_min and pos_delta are used to store the minimum offset
538e84ed 497 and the delta to the maximum offset at the current point in the pattern.
1de06328
YO
498
499*/
2c2d71f5 500
cb6aef8b
DM
501struct scan_data_substrs {
502 SV *str; /* longest substring found in pattern */
503 SSize_t min_offset; /* earliest point in string it can appear */
504 SSize_t max_offset; /* latest point in string it can appear */
505 SSize_t *minlenp; /* pointer to the minlen relevant to the string */
506 SSize_t lookbehind; /* is the pos of the string modified by LB */
11683ecb 507 I32 flags; /* per substring SF_* and SCF_* flags */
cb6aef8b
DM
508};
509
2c2d71f5 510typedef struct scan_data_t {
1de06328
YO
511 /*I32 len_min; unused */
512 /*I32 len_delta; unused */
49f55535 513 SSize_t pos_min;
ea3daa5d 514 SSize_t pos_delta;
2c2d71f5 515 SV *last_found;
ea3daa5d 516 SSize_t last_end; /* min value, <0 unless valid. */
49f55535 517 SSize_t last_start_min;
ea3daa5d 518 SSize_t last_start_max;
2099df82
DM
519 U8 cur_is_floating; /* whether the last_* values should be set as
520 * the next fixed (0) or floating (1)
521 * substring */
6ea77169 522
2099df82
DM
523 /* [0] is longest fixed substring so far, [1] is longest float so far */
524 struct scan_data_substrs substrs[2];
6ea77169 525
11683ecb 526 I32 flags; /* common SF_* and SCF_* flags */
2c2d71f5 527 I32 whilem_c;
49f55535 528 SSize_t *last_closep;
b8f7bb16 529 regnode_ssc *start_class;
2c2d71f5
JH
530} scan_data_t;
531
a687059c 532/*
e50aee73 533 * Forward declarations for pregcomp()'s friends.
a687059c 534 */
a0d0e21e 535
6ea77169 536static const scan_data_t zero_scan_data = {
a2ca2017 537 0, 0, NULL, 0, 0, 0, 0,
6ea77169 538 {
11683ecb
DM
539 { NULL, 0, 0, 0, 0, 0 },
540 { NULL, 0, 0, 0, 0, 0 },
6ea77169
DM
541 },
542 0, 0, NULL, NULL
543};
c277df42 544
11683ecb
DM
545/* study flags */
546
07be1b83
YO
547#define SF_BEFORE_SEOL 0x0001
548#define SF_BEFORE_MEOL 0x0002
11683ecb 549#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
c277df42 550
07be1b83
YO
551#define SF_IS_INF 0x0040
552#define SF_HAS_PAR 0x0080
553#define SF_IN_PAR 0x0100
554#define SF_HAS_EVAL 0x0200
bc604ad8
KW
555
556
557/* SCF_DO_SUBSTR is the flag that tells the regexp analyzer to track the
558 * longest substring in the pattern. When it is not set the optimiser keeps
559 * track of position, but does not keep track of the actual strings seen,
560 *
561 * So for instance /foo/ will be parsed with SCF_DO_SUBSTR being true, but
562 * /foo/i will not.
563 *
564 * Similarly, /foo.*(blah|erm|huh).*fnorble/ will have "foo" and "fnorble"
565 * parsed with SCF_DO_SUBSTR on, but while processing the (...) it will be
566 * turned off because of the alternation (BRANCH). */
07be1b83 567#define SCF_DO_SUBSTR 0x0400
bc604ad8 568
653099ff
GS
569#define SCF_DO_STCLASS_AND 0x0800
570#define SCF_DO_STCLASS_OR 0x1000
571#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
e1901655 572#define SCF_WHILEM_VISITED_POS 0x2000
c277df42 573
786e8c11 574#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
538e84ed 575#define SCF_SEEN_ACCEPT 0x8000
688e0391 576#define SCF_TRIE_DOING_RESTUDY 0x10000
4286711a
YO
577#define SCF_IN_DEFINE 0x20000
578
579
580
07be1b83 581
43fead97 582#define UTF cBOOL(RExC_utf8)
00b27cfc
KW
583
584/* The enums for all these are ordered so things work out correctly */
a62b1201 585#define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
538e84ed
KW
586#define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
587 == REGEX_DEPENDS_CHARSET)
00b27cfc 588#define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
538e84ed
KW
589#define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
590 >= REGEX_UNICODE_CHARSET)
591#define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
592 == REGEX_ASCII_RESTRICTED_CHARSET)
593#define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
594 >= REGEX_ASCII_RESTRICTED_CHARSET)
595#define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
596 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
a62b1201 597
43fead97 598#define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
a0ed51b3 599
f2c2a6ab
KW
600/* For programs that want to be strictly Unicode compatible by dying if any
601 * attempt is made to match a non-Unicode code point against a Unicode
602 * property. */
603#define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
604
93733859 605#define OOB_NAMEDCLASS -1
b8c5462f 606
8e661ac5
KW
607/* There is no code point that is out-of-bounds, so this is problematic. But
608 * its only current use is to initialize a variable that is always set before
609 * looked at. */
610#define OOB_UNICODE 0xDEADBEEF
611
a0ed51b3 612#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
a0ed51b3 613
8615cb43 614
b45f050a
JF
615/* length of regex to show in messages that don't mark a position within */
616#define RegexLengthToShowInErrorMessages 127
617
618/*
619 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
620 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
621 * op/pragma/warn/regcomp.
622 */
7253e4e3
RK
623#define MARKER1 "<-- HERE" /* marker as it appears in the description */
624#define MARKER2 " <-- HERE " /* marker as it appears within the regex */
b81d288d 625
538e84ed 626#define REPORT_LOCATION " in regex; marked by " MARKER1 \
147e3846 627 " in m/%" UTF8f MARKER2 "%" UTF8f "/"
b45f050a 628
285b5ca0
KW
629/* The code in this file in places uses one level of recursion with parsing
630 * rebased to an alternate string constructed by us in memory. This can take
631 * the form of something that is completely different from the input, or
632 * something that uses the input as part of the alternate. In the first case,
633 * there should be no possibility of an error, as we are in complete control of
51684e3c
KW
634 * the alternate string. But in the second case we don't completely control
635 * the input portion, so there may be errors in that. Here's an example:
285b5ca0
KW
636 * /[abc\x{DF}def]/ui
637 * is handled specially because \x{df} folds to a sequence of more than one
51684e3c 638 * character: 'ss'. What is done is to create and parse an alternate string,
285b5ca0
KW
639 * which looks like this:
640 * /(?:\x{DF}|[abc\x{DF}def])/ui
641 * where it uses the input unchanged in the middle of something it constructs,
642 * which is a branch for the DF outside the character class, and clustering
643 * parens around the whole thing. (It knows enough to skip the DF inside the
644 * class while in this substitute parse.) 'abc' and 'def' may have errors that
645 * need to be reported. The general situation looks like this:
646 *
51684e3c 647 * |<------- identical ------>|
285b5ca0 648 * sI tI xI eI
51684e3c 649 * Input: ---------------------------------------------------------------
285b5ca0
KW
650 * Constructed: ---------------------------------------------------
651 * sC tC xC eC EC
51684e3c 652 * |<------- identical ------>|
285b5ca0 653 *
51684e3c
KW
654 * sI..eI is the portion of the input pattern we are concerned with here.
655 * sC..EC is the constructed substitute parse string.
656 * sC..tC is constructed by us
657 * tC..eC is an exact duplicate of the portion of the input pattern tI..eI.
658 * In the diagram, these are vertically aligned.
659 * eC..EC is also constructed by us.
660 * xC is the position in the substitute parse string where we found a
661 * problem.
662 * xI is the position in the original pattern corresponding to xC.
285b5ca0 663 *
51684e3c
KW
664 * We want to display a message showing the real input string. Thus we need to
665 * translate from xC to xI. We know that xC >= tC, since the portion of the
666 * string sC..tC has been constructed by us, and so shouldn't have errors. We
667 * get:
668 * xI = tI + (xC - tC)
285b5ca0 669 *
51684e3c
KW
670 * When the substitute parse is constructed, the code needs to set:
671 * RExC_start (sC)
672 * RExC_end (eC)
673 * RExC_copy_start_in_input (tI)
674 * RExC_copy_start_in_constructed (tC)
675 * and restore them when done.
285b5ca0 676 *
51684e3c
KW
677 * During normal processing of the input pattern, both
678 * 'RExC_copy_start_in_input' and 'RExC_copy_start_in_constructed' are set to
679 * sI, so that xC equals xI.
285b5ca0
KW
680 */
681
51684e3c
KW
682#define sI RExC_precomp
683#define eI RExC_precomp_end
684#define sC RExC_start
685#define eC RExC_end
686#define tI RExC_copy_start_in_input
687#define tC RExC_copy_start_in_constructed
688#define xI(xC) (tI + (xC - tC))
689#define xI_offset(xC) (xI(xC) - sI)
285b5ca0
KW
690
691#define REPORT_LOCATION_ARGS(xC) \
692 UTF8fARG(UTF, \
51684e3c 693 (xI(xC) > eI) /* Don't run off end */ \
232b691f 694 ? eI - sI /* Length before the <--HERE */ \
bc929b61
KW
695 : ((xI_offset(xC) >= 0) \
696 ? xI_offset(xC) \
697 : (Perl_croak(aTHX_ "panic: %s: %d: negative offset: %" \
698 IVdf " trying to output message for " \
699 " pattern %.*s", \
51684e3c 700 __FILE__, __LINE__, (IV) xI_offset(xC), \
bc929b61 701 ((int) (eC - sC)), sC), 0)), \
51684e3c 702 sI), /* The input pattern printed up to the <--HERE */ \
285b5ca0 703 UTF8fARG(UTF, \
51684e3c
KW
704 (xI(xC) > eI) ? 0 : eI - xI(xC), /* Length after <--HERE */ \
705 (xI(xC) > eI) ? eI : xI(xC)) /* pattern after <--HERE */
c1d900c3 706
8a6d8ec6
HS
707/* Used to point after bad bytes for an error message, but avoid skipping
708 * past a nul byte. */
e3304382 709#define SKIP_IF_CHAR(s, e) (!*(s) ? 0 : UTF ? UTF8_SAFE_SKIP(s, e) : 1)
8a6d8ec6 710
ed68fe11
KW
711/* Set up to clean up after our imminent demise */
712#define PREPARE_TO_DIE \
713 STMT_START { \
714 if (RExC_rx_sv) \
715 SAVEFREESV(RExC_rx_sv); \
7c932d07
KW
716 if (RExC_open_parens) \
717 SAVEFREEPV(RExC_open_parens); \
718 if (RExC_close_parens) \
719 SAVEFREEPV(RExC_close_parens); \
ed68fe11
KW
720 } STMT_END
721
b45f050a
JF
722/*
723 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
724 * arg. Show regex, up to a maximum length. If it's too long, chop and add
725 * "...".
726 */
58e23c8d 727#define _FAIL(code) STMT_START { \
bfed75c6 728 const char *ellipses = ""; \
88f063b4 729 IV len = RExC_precomp_end - RExC_precomp; \
ccb2c380 730 \
ed68fe11 731 PREPARE_TO_DIE; \
ccb2c380
MP
732 if (len > RegexLengthToShowInErrorMessages) { \
733 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
734 len = RegexLengthToShowInErrorMessages - 10; \
735 ellipses = "..."; \
736 } \
58e23c8d 737 code; \
ccb2c380 738} STMT_END
8615cb43 739
58e23c8d 740#define FAIL(msg) _FAIL( \
147e3846 741 Perl_croak(aTHX_ "%s in regex m/%" UTF8f "%s/", \
c1d900c3 742 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d
YO
743
744#define FAIL2(msg,arg) _FAIL( \
147e3846 745 Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \
c1d900c3 746 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d 747
b45f050a 748/*
b45f050a
JF
749 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
750 */
ccb2c380 751#define Simple_vFAIL(m) STMT_START { \
ccb2c380 752 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
d528642a 753 m, REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 754} STMT_END
b45f050a
JF
755
756/*
757 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
758 */
ccb2c380 759#define vFAIL(m) STMT_START { \
ed68fe11 760 PREPARE_TO_DIE; \
ccb2c380
MP
761 Simple_vFAIL(m); \
762} STMT_END
b45f050a
JF
763
764/*
765 * Like Simple_vFAIL(), but accepts two arguments.
766 */
ccb2c380 767#define Simple_vFAIL2(m,a1) STMT_START { \
d528642a
KW
768 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
769 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 770} STMT_END
b45f050a
JF
771
772/*
773 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
774 */
ccb2c380 775#define vFAIL2(m,a1) STMT_START { \
ed68fe11 776 PREPARE_TO_DIE; \
ccb2c380
MP
777 Simple_vFAIL2(m, a1); \
778} STMT_END
b45f050a
JF
779
780
781/*
782 * Like Simple_vFAIL(), but accepts three arguments.
783 */
ccb2c380 784#define Simple_vFAIL3(m, a1, a2) STMT_START { \
c1d900c3 785 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 786 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 787} STMT_END
b45f050a
JF
788
789/*
790 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
791 */
ccb2c380 792#define vFAIL3(m,a1,a2) STMT_START { \
ed68fe11 793 PREPARE_TO_DIE; \
ccb2c380
MP
794 Simple_vFAIL3(m, a1, a2); \
795} STMT_END
b45f050a
JF
796
797/*
798 * Like Simple_vFAIL(), but accepts four arguments.
799 */
ccb2c380 800#define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
d528642a
KW
801 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
802 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 803} STMT_END
b45f050a 804
95db3ffa 805#define vFAIL4(m,a1,a2,a3) STMT_START { \
ed68fe11 806 PREPARE_TO_DIE; \
95db3ffa
KW
807 Simple_vFAIL4(m, a1, a2, a3); \
808} STMT_END
809
946095af 810/* A specialized version of vFAIL2 that works with UTF8f */
d528642a 811#define vFAIL2utf8f(m, a1) STMT_START { \
ed68fe11 812 PREPARE_TO_DIE; \
d528642a
KW
813 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
814 REPORT_LOCATION_ARGS(RExC_parse)); \
946095af
BF
815} STMT_END
816
3ba22297 817#define vFAIL3utf8f(m, a1, a2) STMT_START { \
ed68fe11 818 PREPARE_TO_DIE; \
3ba22297 819 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 820 REPORT_LOCATION_ARGS(RExC_parse)); \
3ba22297
KW
821} STMT_END
822
d62e749e
KW
823/* Setting this to NULL is a signal to not output warnings */
824#define TURN_OFF_WARNINGS_IN_SUBSTITUTE_PARSE RExC_copy_start_in_constructed = NULL
825#define RESTORE_WARNINGS RExC_copy_start_in_constructed = RExC_precomp
826
7c932d07
KW
827/* Since a warning can be generated multiple times as the input is reparsed, we
828 * output it the first time we come to that point in the parse, but suppress it
829 * otherwise. 'RExC_copy_start_in_constructed' being NULL is a flag to not
830 * generate any warnings */
55bcf857 831#define TO_OUTPUT_WARNINGS(loc) \
7c932d07
KW
832 ( RExC_copy_start_in_constructed \
833 && ((xI(loc)) - RExC_precomp) > (Ptrdiff_t) RExC_latest_warn_offset)
55bcf857 834
7c932d07
KW
835/* After we've emitted a warning, we save the position in the input so we don't
836 * output it again */
837#define UPDATE_WARNINGS_LOC(loc) \
838 STMT_START { \
839 if (TO_OUTPUT_WARNINGS(loc)) { \
840 RExC_latest_warn_offset = (xI(loc)) - RExC_precomp; \
841 } \
842 } STMT_END
77ccba3c 843
086f21bf 844/* 'warns' is the output of the packWARNx macro used in 'code' */
f8220fe2
KW
845#define _WARN_HELPER(loc, warns, code) \
846 STMT_START { \
d62e749e
KW
847 if (! RExC_copy_start_in_constructed) { \
848 Perl_croak( aTHX_ "panic! %s: %d: Tried to warn when none" \
849 " expected at '%s'", \
850 __FILE__, __LINE__, loc); \
851 } \
55bcf857 852 if (TO_OUTPUT_WARNINGS(loc)) { \
086f21bf
KW
853 if (ckDEAD(warns)) \
854 PREPARE_TO_DIE; \
338190a1 855 code; \
77ccba3c 856 UPDATE_WARNINGS_LOC(loc); \
338190a1 857 } \
f8220fe2 858 } STMT_END
946095af 859
5e0a247b 860/* m is not necessarily a "literal string", in this macro */
f8220fe2
KW
861#define reg_warn_non_literal_string(loc, m) \
862 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
863 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 864 "%s" REPORT_LOCATION, \
f8220fe2 865 m, REPORT_LOCATION_ARGS(loc)))
5e0a247b 866
f8220fe2
KW
867#define ckWARNreg(loc,m) \
868 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
869 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 870 m REPORT_LOCATION, \
f8220fe2 871 REPORT_LOCATION_ARGS(loc)))
ccb2c380 872
f8220fe2
KW
873#define vWARN(loc, m) \
874 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
875 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 876 m REPORT_LOCATION, \
f8220fe2 877 REPORT_LOCATION_ARGS(loc))) \
b927b7e9 878
f8220fe2
KW
879#define vWARN_dep(loc, m) \
880 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
881 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 882 m REPORT_LOCATION, \
f8220fe2 883 REPORT_LOCATION_ARGS(loc)))
0d6106aa 884
f8220fe2
KW
885#define ckWARNdep(loc,m) \
886 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
887 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 888 m REPORT_LOCATION, \
f8220fe2 889 REPORT_LOCATION_ARGS(loc)))
147508a2 890
f8220fe2
KW
891#define ckWARNregdep(loc,m) \
892 _WARN_HELPER(loc, packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
893 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, \
d528642a
KW
894 WARN_REGEXP), \
895 m REPORT_LOCATION, \
f8220fe2 896 REPORT_LOCATION_ARGS(loc)))
ccb2c380 897
f8220fe2
KW
898#define ckWARN2reg_d(loc,m, a1) \
899 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
900 Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
d528642a 901 m REPORT_LOCATION, \
f8220fe2 902 a1, REPORT_LOCATION_ARGS(loc)))
2335b3d3 903
f8220fe2
KW
904#define ckWARN2reg(loc, m, a1) \
905 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
906 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 907 m REPORT_LOCATION, \
f8220fe2 908 a1, REPORT_LOCATION_ARGS(loc)))
ccb2c380 909
f8220fe2
KW
910#define vWARN3(loc, m, a1, a2) \
911 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
912 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 913 m REPORT_LOCATION, \
f8220fe2 914 a1, a2, REPORT_LOCATION_ARGS(loc)))
ccb2c380 915
f8220fe2
KW
916#define ckWARN3reg(loc, m, a1, a2) \
917 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
918 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
919 m REPORT_LOCATION, \
920 a1, a2, \
f8220fe2 921 REPORT_LOCATION_ARGS(loc)))
668c081a 922
f8220fe2
KW
923#define vWARN4(loc, m, a1, a2, a3) \
924 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
925 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
926 m REPORT_LOCATION, \
927 a1, a2, a3, \
f8220fe2 928 REPORT_LOCATION_ARGS(loc)))
ccb2c380 929
f8220fe2
KW
930#define ckWARN4reg(loc, m, a1, a2, a3) \
931 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
932 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
933 m REPORT_LOCATION, \
934 a1, a2, a3, \
f8220fe2 935 REPORT_LOCATION_ARGS(loc)))
668c081a 936
f8220fe2
KW
937#define vWARN5(loc, m, a1, a2, a3, a4) \
938 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
939 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
940 m REPORT_LOCATION, \
941 a1, a2, a3, a4, \
f8220fe2 942 REPORT_LOCATION_ARGS(loc)))
9d1d55b5 943
aea5f181
KW
944#define ckWARNexperimental(loc, class, m) \
945 _WARN_HELPER(loc, packWARN(class), \
946 Perl_ck_warner_d(aTHX_ packWARN(class), \
947 m REPORT_LOCATION, \
948 REPORT_LOCATION_ARGS(loc)))
949
1a322bb4
KW
950/* Convert between a pointer to a node and its offset from the beginning of the
951 * program */
952#define REGNODE_p(offset) (RExC_emit_start + (offset))
953#define REGNODE_OFFSET(node) ((node) - RExC_emit_start)
954
538e84ed 955/* Macros for recording node offsets. 20001227 mjd@plover.com
fac92740
MJD
956 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
957 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
958 * Element 0 holds the number n.
07be1b83 959 * Position is 1 indexed.
fac92740 960 */
7122b237 961#ifndef RE_TRACK_PATTERN_OFFSETS
d7c442ce 962#define Set_Node_Offset_To_R(offset,byte)
7122b237
YO
963#define Set_Node_Offset(node,byte)
964#define Set_Cur_Node_Offset
965#define Set_Node_Length_To_R(node,len)
966#define Set_Node_Length(node,len)
6a86c6ad 967#define Set_Node_Cur_Length(node,start)
538e84ed
KW
968#define Node_Offset(n)
969#define Node_Length(n)
7122b237
YO
970#define Set_Node_Offset_Length(node,offset,len)
971#define ProgLen(ri) ri->u.proglen
972#define SetProgLen(ri,x) ri->u.proglen = x
4cbeff97 973#define Track_Code(code)
7122b237
YO
974#else
975#define ProgLen(ri) ri->u.offsets[0]
976#define SetProgLen(ri,x) ri->u.offsets[0] = x
d7c442ce 977#define Set_Node_Offset_To_R(offset,byte) STMT_START { \
ccb2c380 978 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
d7c442ce
KW
979 __LINE__, (int)(offset), (int)(byte))); \
980 if((offset) < 0) { \
538e84ed 981 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
d7c442ce 982 (int)(offset)); \
ccb2c380 983 } else { \
d7c442ce 984 RExC_offsets[2*(offset)-1] = (byte); \
ccb2c380 985 } \
ccb2c380
MP
986} STMT_END
987
88f063b4 988#define Set_Node_Offset(node,byte) \
1a322bb4 989 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (byte)-RExC_start)
ccb2c380
MP
990#define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
991
992#define Set_Node_Length_To_R(node,len) STMT_START { \
ccb2c380 993 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
551405c4 994 __LINE__, (int)(node), (int)(len))); \
ccb2c380 995 if((node) < 0) { \
538e84ed
KW
996 Perl_croak(aTHX_ "value of node is %d in Length macro", \
997 (int)(node)); \
ccb2c380
MP
998 } else { \
999 RExC_offsets[2*(node)] = (len); \
1000 } \
ccb2c380
MP
1001} STMT_END
1002
1003#define Set_Node_Length(node,len) \
1a322bb4 1004 Set_Node_Length_To_R(REGNODE_OFFSET(node), len)
6a86c6ad
NC
1005#define Set_Node_Cur_Length(node, start) \
1006 Set_Node_Length(node, RExC_parse - start)
fac92740
MJD
1007
1008/* Get offsets and lengths */
1a322bb4
KW
1009#define Node_Offset(n) (RExC_offsets[2*(REGNODE_OFFSET(n))-1])
1010#define Node_Length(n) (RExC_offsets[2*(REGNODE_OFFSET(n))])
fac92740 1011
07be1b83 1012#define Set_Node_Offset_Length(node,offset,len) STMT_START { \
1a322bb4
KW
1013 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (offset)); \
1014 Set_Node_Length_To_R(REGNODE_OFFSET(node), (len)); \
07be1b83 1015} STMT_END
4cbeff97
TC
1016
1017#define Track_Code(code) STMT_START { code } STMT_END
7122b237 1018#endif
07be1b83
YO
1019
1020#if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
1021#define EXPERIMENTAL_INPLACESCAN
f427392e 1022#endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
07be1b83 1023
cb41e5d6
YO
1024#ifdef DEBUGGING
1025int
6ad9a8ab 1026Perl_re_printf(pTHX_ const char *fmt, ...)
cb41e5d6
YO
1027{
1028 va_list ap;
1029 int result;
1030 PerlIO *f= Perl_debug_log;
1031 PERL_ARGS_ASSERT_RE_PRINTF;
1032 va_start(ap, fmt);
1033 result = PerlIO_vprintf(f, fmt, ap);
1034 va_end(ap);
1035 return result;
1036}
1037
1038int
7b031478 1039Perl_re_indentf(pTHX_ const char *fmt, U32 depth, ...)
cb41e5d6
YO
1040{
1041 va_list ap;
1042 int result;
1043 PerlIO *f= Perl_debug_log;
1044 PERL_ARGS_ASSERT_RE_INDENTF;
1045 va_start(ap, depth);
daeb874b 1046 PerlIO_printf(f, "%*s", ( (int)depth % 20 ) * 2, "");
cb41e5d6
YO
1047 result = PerlIO_vprintf(f, fmt, ap);
1048 va_end(ap);
1049 return result;
1050}
1051#endif /* DEBUGGING */
1052
7b031478 1053#define DEBUG_RExC_seen() \
538e84ed 1054 DEBUG_OPTIMISE_MORE_r({ \
88f063b4 1055 Perl_re_printf( aTHX_ "RExC_seen: "); \
538e84ed 1056 \
e384d5c1 1057 if (RExC_seen & REG_ZERO_LEN_SEEN) \
88f063b4 1058 Perl_re_printf( aTHX_ "REG_ZERO_LEN_SEEN "); \
538e84ed 1059 \
e384d5c1 1060 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
88f063b4 1061 Perl_re_printf( aTHX_ "REG_LOOKBEHIND_SEEN "); \
538e84ed 1062 \
e384d5c1 1063 if (RExC_seen & REG_GPOS_SEEN) \
88f063b4 1064 Perl_re_printf( aTHX_ "REG_GPOS_SEEN "); \
538e84ed 1065 \
e384d5c1 1066 if (RExC_seen & REG_RECURSE_SEEN) \
88f063b4 1067 Perl_re_printf( aTHX_ "REG_RECURSE_SEEN "); \
538e84ed 1068 \
7b031478 1069 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
88f063b4 1070 Perl_re_printf( aTHX_ "REG_TOP_LEVEL_BRANCHES_SEEN "); \
538e84ed 1071 \
e384d5c1 1072 if (RExC_seen & REG_VERBARG_SEEN) \
88f063b4 1073 Perl_re_printf( aTHX_ "REG_VERBARG_SEEN "); \
538e84ed 1074 \
e384d5c1 1075 if (RExC_seen & REG_CUTGROUP_SEEN) \
88f063b4 1076 Perl_re_printf( aTHX_ "REG_CUTGROUP_SEEN "); \
538e84ed 1077 \
e384d5c1 1078 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
88f063b4 1079 Perl_re_printf( aTHX_ "REG_RUN_ON_COMMENT_SEEN "); \
538e84ed 1080 \
e384d5c1 1081 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
88f063b4 1082 Perl_re_printf( aTHX_ "REG_UNFOLDED_MULTI_SEEN "); \
538e84ed 1083 \
7b031478 1084 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
88f063b4 1085 Perl_re_printf( aTHX_ "REG_UNBOUNDED_QUANTIFIER_SEEN "); \
ee273784 1086 \
88f063b4 1087 Perl_re_printf( aTHX_ "\n"); \
9e9ecfdf
YO
1088 });
1089
fdfb4f21 1090#define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
6ad9a8ab 1091 if ((flags) & flag) Perl_re_printf( aTHX_ "%s ", #flag)
fdfb4f21 1092
fdfb4f21 1093
f5a36d78
DM
1094#ifdef DEBUGGING
1095static void
1096S_debug_show_study_flags(pTHX_ U32 flags, const char *open_str,
1097 const char *close_str)
1098{
1099 if (!flags)
1100 return;
1101
1102 Perl_re_printf( aTHX_ "%s", open_str);
11683ecb
DM
1103 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_SEOL);
1104 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_MEOL);
f5a36d78
DM
1105 DEBUG_SHOW_STUDY_FLAG(flags, SF_IS_INF);
1106 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_PAR);
1107 DEBUG_SHOW_STUDY_FLAG(flags, SF_IN_PAR);
1108 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_EVAL);
1109 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_SUBSTR);
1110 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_AND);
1111 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_OR);
1112 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS);
1113 DEBUG_SHOW_STUDY_FLAG(flags, SCF_WHILEM_VISITED_POS);
1114 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_RESTUDY);
1115 DEBUG_SHOW_STUDY_FLAG(flags, SCF_SEEN_ACCEPT);
1116 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_DOING_RESTUDY);
1117 DEBUG_SHOW_STUDY_FLAG(flags, SCF_IN_DEFINE);
1118 Perl_re_printf( aTHX_ "%s", close_str);
1119}
1120
1121
1122static void
1123S_debug_studydata(pTHX_ const char *where, scan_data_t *data,
1124 U32 depth, int is_inf)
1125{
1126 GET_RE_DEBUG_FLAGS_DECL;
1127
1128 DEBUG_OPTIMISE_MORE_r({
1129 if (!data)
1130 return;
1131 Perl_re_indentf(aTHX_ "%s: Pos:%" IVdf "/%" IVdf " Flags: 0x%" UVXf,
1132 depth,
1133 where,
1134 (IV)data->pos_min,
1135 (IV)data->pos_delta,
1136 (UV)data->flags
1137 );
1138
1139 S_debug_show_study_flags(aTHX_ data->flags," [","]");
fdfb4f21 1140
f5a36d78
DM
1141 Perl_re_printf( aTHX_
1142 " Whilem_c: %" IVdf " Lcp: %" IVdf " %s",
1143 (IV)data->whilem_c,
1144 (IV)(data->last_closep ? *((data)->last_closep) : -1),
1145 is_inf ? "INF " : ""
1146 );
1147
11683ecb 1148 if (data->last_found) {
37b6262f 1149 int i;
11683ecb
DM
1150 Perl_re_printf(aTHX_
1151 "Last:'%s' %" IVdf ":%" IVdf "/%" IVdf,
1152 SvPVX_const(data->last_found),
1153 (IV)data->last_end,
1154 (IV)data->last_start_min,
1155 (IV)data->last_start_max
1156 );
1157
2099df82 1158 for (i = 0; i < 2; i++) {
37b6262f
DM
1159 Perl_re_printf(aTHX_
1160 " %s%s: '%s' @ %" IVdf "/%" IVdf,
2099df82
DM
1161 data->cur_is_floating == i ? "*" : "",
1162 i ? "Float" : "Fixed",
37b6262f
DM
1163 SvPVX_const(data->substrs[i].str),
1164 (IV)data->substrs[i].min_offset,
1165 (IV)data->substrs[i].max_offset
1166 );
1167 S_debug_show_study_flags(aTHX_ data->substrs[i].flags," [","]");
1168 }
11683ecb 1169 }
f5a36d78
DM
1170
1171 Perl_re_printf( aTHX_ "\n");
1172 });
1173}
1174
1175
1176static void
1177S_debug_peep(pTHX_ const char *str, const RExC_state_t *pRExC_state,
1178 regnode *scan, U32 depth, U32 flags)
1179{
1180 GET_RE_DEBUG_FLAGS_DECL;
1181
1182 DEBUG_OPTIMISE_r({
1183 regnode *Next;
1184
1185 if (!scan)
1186 return;
1187 Next = regnext(scan);
1188 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
1189 Perl_re_indentf( aTHX_ "%s>%3d: %s (%d)",
1190 depth,
1191 str,
1192 REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),
1193 Next ? (REG_NODE_NUM(Next)) : 0 );
1194 S_debug_show_study_flags(aTHX_ flags," [ ","]");
1195 Perl_re_printf( aTHX_ "\n");
1196 });
1197}
1198
1199
1200# define DEBUG_STUDYDATA(where, data, depth, is_inf) \
1201 S_debug_studydata(aTHX_ where, data, depth, is_inf)
1202
1203# define DEBUG_PEEP(str, scan, depth, flags) \
1204 S_debug_peep(aTHX_ str, pRExC_state, scan, depth, flags)
1205
1206#else
1207# define DEBUG_STUDYDATA(where, data, depth, is_inf) NOOP
1208# define DEBUG_PEEP(str, scan, depth, flags) NOOP
1209#endif
1de06328 1210
cb41e5d6 1211
c6871b76
KW
1212/* =========================================================
1213 * BEGIN edit_distance stuff.
1214 *
1215 * This calculates how many single character changes of any type are needed to
1216 * transform a string into another one. It is taken from version 3.1 of
1217 *
1218 * https://metacpan.org/pod/Text::Levenshtein::Damerau::XS
1219 */
1220
1221/* Our unsorted dictionary linked list. */
1222/* Note we use UVs, not chars. */
1223
1224struct dictionary{
1225 UV key;
1226 UV value;
1227 struct dictionary* next;
1228};
1229typedef struct dictionary item;
1230
1231
1232PERL_STATIC_INLINE item*
88f063b4 1233push(UV key, item* curr)
c6871b76
KW
1234{
1235 item* head;
d09f14bf 1236 Newx(head, 1, item);
c6871b76
KW
1237 head->key = key;
1238 head->value = 0;
1239 head->next = curr;
1240 return head;
1241}
1242
1243
1244PERL_STATIC_INLINE item*
1245find(item* head, UV key)
1246{
1247 item* iterator = head;
1248 while (iterator){
1249 if (iterator->key == key){
1250 return iterator;
1251 }
1252 iterator = iterator->next;
1253 }
1254
1255 return NULL;
1256}
1257
1258PERL_STATIC_INLINE item*
88f063b4 1259uniquePush(item* head, UV key)
c6871b76
KW
1260{
1261 item* iterator = head;
1262
1263 while (iterator){
1264 if (iterator->key == key) {
1265 return head;
1266 }
1267 iterator = iterator->next;
1268 }
1269
88f063b4 1270 return push(key, head);
c6871b76
KW
1271}
1272
1273PERL_STATIC_INLINE void
1274dict_free(item* head)
1275{
1276 item* iterator = head;
1277
1278 while (iterator) {
1279 item* temp = iterator;
1280 iterator = iterator->next;
1281 Safefree(temp);
1282 }
1283
1284 head = NULL;
1285}
1286
1287/* End of Dictionary Stuff */
1288
1289/* All calculations/work are done here */
1290STATIC int
1291S_edit_distance(const UV* src,
1292 const UV* tgt,
1293 const STRLEN x, /* length of src[] */
1294 const STRLEN y, /* length of tgt[] */
1295 const SSize_t maxDistance
1296)
1297{
1298 item *head = NULL;
88f063b4 1299 UV swapCount, swapScore, targetCharCount, i, j;
c6871b76
KW
1300 UV *scores;
1301 UV score_ceil = x + y;
1302
1303 PERL_ARGS_ASSERT_EDIT_DISTANCE;
1304
1305 /* intialize matrix start values */
d09f14bf 1306 Newx(scores, ( (x + 2) * (y + 2)), UV);
c6871b76
KW
1307 scores[0] = score_ceil;
1308 scores[1 * (y + 2) + 0] = score_ceil;
1309 scores[0 * (y + 2) + 1] = score_ceil;
1310 scores[1 * (y + 2) + 1] = 0;
88f063b4 1311 head = uniquePush(uniquePush(head, src[0]), tgt[0]);
c6871b76
KW
1312
1313 /* work loops */
1314 /* i = src index */
1315 /* j = tgt index */
1316 for (i=1;i<=x;i++) {
1317 if (i < x)
88f063b4 1318 head = uniquePush(head, src[i]);
c6871b76
KW
1319 scores[(i+1) * (y + 2) + 1] = i;
1320 scores[(i+1) * (y + 2) + 0] = score_ceil;
1321 swapCount = 0;
1322
1323 for (j=1;j<=y;j++) {
1324 if (i == 1) {
1325 if(j < y)
88f063b4 1326 head = uniquePush(head, tgt[j]);
c6871b76
KW
1327 scores[1 * (y + 2) + (j + 1)] = j;
1328 scores[0 * (y + 2) + (j + 1)] = score_ceil;
1329 }
1330
88f063b4 1331 targetCharCount = find(head, tgt[j-1])->value;
c6871b76
KW
1332 swapScore = scores[targetCharCount * (y + 2) + swapCount] + i - targetCharCount - 1 + j - swapCount;
1333
1334 if (src[i-1] != tgt[j-1]){
1335 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));
1336 }
1337 else {
1338 swapCount = j;
1339 scores[(i+1) * (y + 2) + (j + 1)] = MIN(scores[i * (y + 2) + j], swapScore);
1340 }
1341 }
1342
88f063b4 1343 find(head, src[i-1])->value = i;
c6871b76
KW
1344 }
1345
1346 {
1347 IV score = scores[(x+1) * (y + 2) + (y + 1)];
1348 dict_free(head);
1349 Safefree(scores);
1350 return (maxDistance != 0 && maxDistance < score)?(-1):score;
1351 }
1352}
1353
1354/* END of edit_distance() stuff
1355 * ========================================================= */
1356
8e35b056
KW
1357/* is c a control character for which we have a mnemonic? */
1358#define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
1359
549b4e78
KW
1360STATIC const char *
1361S_cntrl_to_mnemonic(const U8 c)
1362{
1363 /* Returns the mnemonic string that represents character 'c', if one
1364 * exists; NULL otherwise. The only ones that exist for the purposes of
1365 * this routine are a few control characters */
1366
1367 switch (c) {
1368 case '\a': return "\\a";
1369 case '\b': return "\\b";
1370 case ESC_NATIVE: return "\\e";
1371 case '\f': return "\\f";
1372 case '\n': return "\\n";
1373 case '\r': return "\\r";
1374 case '\t': return "\\t";
1375 }
1376
1377 return NULL;
1378}
1379
653099ff 1380/* Mark that we cannot extend a found fixed substring at this point.
37b6262f 1381 Update the longest found anchored substring or the longest found
653099ff
GS
1382 floating substrings if needed. */
1383
4327152a 1384STATIC void
ea3daa5d
FC
1385S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
1386 SSize_t *minlenp, int is_inf)
c277df42 1387{
e1ec3a88 1388 const STRLEN l = CHR_SVLEN(data->last_found);
2099df82 1389 SV * const longest_sv = data->substrs[data->cur_is_floating].str;
a2ca2017 1390 const STRLEN old_l = CHR_SVLEN(longest_sv);
1de06328 1391 GET_RE_DEBUG_FLAGS_DECL;
b81d288d 1392
7918f24d
NC
1393 PERL_ARGS_ASSERT_SCAN_COMMIT;
1394
c277df42 1395 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
2099df82 1396 const U8 i = data->cur_is_floating;
a2ca2017 1397 SvSetMagicSV(longest_sv, data->last_found);
37b6262f
DM
1398 data->substrs[i].min_offset = l ? data->last_start_min : data->pos_min;
1399
2099df82
DM
1400 if (!i) /* fixed */
1401 data->substrs[0].max_offset = data->substrs[0].min_offset;
1402 else { /* float */
1403 data->substrs[1].max_offset = (l
646e8787
DM
1404 ? data->last_start_max
1405 : (data->pos_delta > SSize_t_MAX - data->pos_min
ea3daa5d
FC
1406 ? SSize_t_MAX
1407 : data->pos_min + data->pos_delta));
1408 if (is_inf
2099df82
DM
1409 || (STRLEN)data->substrs[1].max_offset > (STRLEN)SSize_t_MAX)
1410 data->substrs[1].max_offset = SSize_t_MAX;
37b6262f 1411 }
11683ecb 1412
37b6262f
DM
1413 if (data->flags & SF_BEFORE_EOL)
1414 data->substrs[i].flags |= (data->flags & SF_BEFORE_EOL);
1415 else
1416 data->substrs[i].flags &= ~SF_BEFORE_EOL;
1417 data->substrs[i].minlenp = minlenp;
1418 data->substrs[i].lookbehind = 0;
c277df42 1419 }
37b6262f 1420
c277df42 1421 SvCUR_set(data->last_found, 0);
0eda9292 1422 {
a28509cc 1423 SV * const sv = data->last_found;
097eb12c
AL
1424 if (SvUTF8(sv) && SvMAGICAL(sv)) {
1425 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
1426 if (mg)
1427 mg->mg_len = 0;
1428 }
0eda9292 1429 }
c277df42
IZ
1430 data->last_end = -1;
1431 data->flags &= ~SF_BEFORE_EOL;
f5a36d78 1432 DEBUG_STUDYDATA("commit", data, 0, is_inf);
c277df42
IZ
1433}
1434
cdd87c1d
KW
1435/* An SSC is just a regnode_charclass_posix with an extra field: the inversion
1436 * list that describes which code points it matches */
1437
653099ff 1438STATIC void
3420edd7 1439S_ssc_anything(pTHX_ regnode_ssc *ssc)
653099ff 1440{
cdd87c1d
KW
1441 /* Set the SSC 'ssc' to match an empty string or any code point */
1442
557bd3fb 1443 PERL_ARGS_ASSERT_SSC_ANYTHING;
7918f24d 1444
71068078 1445 assert(is_ANYOF_SYNTHETIC(ssc));
3fffb88a 1446
0854ea0b
KW
1447 /* mortalize so won't leak */
1448 ssc->invlist = sv_2mortal(_add_range_to_invlist(NULL, 0, UV_MAX));
93e92956 1449 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
653099ff
GS
1450}
1451
653099ff 1452STATIC int
dc3bf405 1453S_ssc_is_anything(const regnode_ssc *ssc)
653099ff 1454{
c144baaa
KW
1455 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
1456 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
1457 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1458 * in any way, so there's no point in using it */
cdd87c1d
KW
1459
1460 UV start, end;
1461 bool ret;
653099ff 1462
557bd3fb 1463 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
7918f24d 1464
71068078 1465 assert(is_ANYOF_SYNTHETIC(ssc));
cdd87c1d 1466
93e92956 1467 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
cdd87c1d
KW
1468 return FALSE;
1469 }
1470
1471 /* See if the list consists solely of the range 0 - Infinity */
1472 invlist_iterinit(ssc->invlist);
1473 ret = invlist_iternext(ssc->invlist, &start, &end)
1474 && start == 0
1475 && end == UV_MAX;
1476
1477 invlist_iterfinish(ssc->invlist);
1478
1479 if (ret) {
1480 return TRUE;
1481 }
1482
1483 /* If e.g., both \w and \W are set, matches everything */
e0e1be5f 1484 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1485 int i;
1486 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1487 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1488 return TRUE;
1489 }
1490 }
1491 }
1492
1493 return FALSE;
653099ff
GS
1494}
1495
653099ff 1496STATIC void
cdd87c1d 1497S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
653099ff 1498{
cdd87c1d
KW
1499 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1500 * string, any code point, or any posix class under locale */
1501
557bd3fb 1502 PERL_ARGS_ASSERT_SSC_INIT;
7918f24d 1503
557bd3fb 1504 Zero(ssc, 1, regnode_ssc);
71068078 1505 set_ANYOF_SYNTHETIC(ssc);
93e92956 1506 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
3420edd7 1507 ssc_anything(ssc);
cdd87c1d 1508
2f306ab9
KW
1509 /* If any portion of the regex is to operate under locale rules that aren't
1510 * fully known at compile time, initialization includes it. The reason
1511 * this isn't done for all regexes is that the optimizer was written under
1512 * the assumption that locale was all-or-nothing. Given the complexity and
1513 * lack of documentation in the optimizer, and that there are inadequate
1514 * test cases for locale, many parts of it may not work properly, it is
1515 * safest to avoid locale unless necessary. */
cdd87c1d
KW
1516 if (RExC_contains_locale) {
1517 ANYOF_POSIXL_SETALL(ssc);
cdd87c1d
KW
1518 }
1519 else {
1520 ANYOF_POSIXL_ZERO(ssc);
1521 }
653099ff
GS
1522}
1523
b423522f 1524STATIC int
dc3bf405
BF
1525S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1526 const regnode_ssc *ssc)
b423522f
KW
1527{
1528 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1529 * to the list of code points matched, and locale posix classes; hence does
1530 * not check its flags) */
1531
1532 UV start, end;
1533 bool ret;
1534
1535 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1536
71068078 1537 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1538
1539 invlist_iterinit(ssc->invlist);
1540 ret = invlist_iternext(ssc->invlist, &start, &end)
1541 && start == 0
1542 && end == UV_MAX;
1543
1544 invlist_iterfinish(ssc->invlist);
1545
1546 if (! ret) {
1547 return FALSE;
1548 }
1549
e0e1be5f 1550 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
31f05a37 1551 return FALSE;
b423522f
KW
1552 }
1553
1554 return TRUE;
1555}
1556
4ebed06a
KW
1557#define INVLIST_INDEX 0
1558#define ONLY_LOCALE_MATCHES_INDEX 1
1559#define DEFERRED_USER_DEFINED_INDEX 2
1560
b423522f
KW
1561STATIC SV*
1562S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
5c0f85ef 1563 const regnode_charclass* const node)
b423522f
KW
1564{
1565 /* Returns a mortal inversion list defining which code points are matched
1566 * by 'node', which is of type ANYOF. Handles complementing the result if
1567 * appropriate. If some code points aren't knowable at this time, the
31f05a37
KW
1568 * returned list must, and will, contain every code point that is a
1569 * possibility. */
b423522f 1570
1565c085 1571 dVAR;
e2506fa7 1572 SV* invlist = NULL;
1ee208c4 1573 SV* only_utf8_locale_invlist = NULL;
b423522f
KW
1574 unsigned int i;
1575 const U32 n = ARG(node);
31f05a37 1576 bool new_node_has_latin1 = FALSE;
b423522f
KW
1577
1578 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1579
1580 /* Look at the data structure created by S_set_ANYOF_arg() */
93e92956 1581 if (n != ANYOF_ONLY_HAS_BITMAP) {
b423522f
KW
1582 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1583 AV * const av = MUTABLE_AV(SvRV(rv));
1584 SV **const ary = AvARRAY(av);
1585 assert(RExC_rxi->data->what[n] == 's');
1586
4ebed06a 1587 if (av_tindex_skip_len_mg(av) >= DEFERRED_USER_DEFINED_INDEX) {
b423522f 1588
6ed02bb6
KW
1589 /* Here there are things that won't be known until runtime -- we
1590 * have to assume it could be anything */
e2506fa7 1591 invlist = sv_2mortal(_new_invlist(1));
b423522f
KW
1592 return _add_range_to_invlist(invlist, 0, UV_MAX);
1593 }
4ebed06a 1594 else if (ary[INVLIST_INDEX]) {
b423522f 1595
6ed02bb6 1596 /* Use the node's inversion list */
4ebed06a 1597 invlist = sv_2mortal(invlist_clone(ary[INVLIST_INDEX], NULL));
1ee208c4
KW
1598 }
1599
1600 /* Get the code points valid only under UTF-8 locales */
6ed02bb6 1601 if ( (ANYOF_FLAGS(node) & ANYOFL_FOLD)
4ebed06a 1602 && av_tindex_skip_len_mg(av) >= ONLY_LOCALE_MATCHES_INDEX)
1ee208c4 1603 {
4ebed06a 1604 only_utf8_locale_invlist = ary[ONLY_LOCALE_MATCHES_INDEX];
b423522f
KW
1605 }
1606 }
1607
e2506fa7
KW
1608 if (! invlist) {
1609 invlist = sv_2mortal(_new_invlist(0));
1610 }
1611
dcb20b36
KW
1612 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1613 * code points, and an inversion list for the others, but if there are code
1614 * points that should match only conditionally on the target string being
1615 * UTF-8, those are placed in the inversion list, and not the bitmap.
1616 * Since there are circumstances under which they could match, they are
1617 * included in the SSC. But if the ANYOF node is to be inverted, we have
1618 * to exclude them here, so that when we invert below, the end result
1619 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1620 * have to do this here before we add the unconditionally matched code
1621 * points */
b423522f
KW
1622 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1623 _invlist_intersection_complement_2nd(invlist,
1624 PL_UpperLatin1,
1625 &invlist);
1626 }
1627
1628 /* Add in the points from the bit map */
5aa750d1 1629 if (OP(node) != ANYOFH) {
90973738
KW
1630 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
1631 if (ANYOF_BITMAP_TEST(node, i)) {
1632 unsigned int start = i++;
1633
1634 for (; i < NUM_ANYOF_CODE_POINTS
1635 && ANYOF_BITMAP_TEST(node, i); ++i)
1636 {
1637 /* empty */
1638 }
1639 invlist = _add_range_to_invlist(invlist, start, i-1);
1640 new_node_has_latin1 = TRUE;
6f8848d5 1641 }
b423522f
KW
1642 }
1643 }
1644
1645 /* If this can match all upper Latin1 code points, have to add them
ac33c516
KW
1646 * as well. But don't add them if inverting, as when that gets done below,
1647 * it would exclude all these characters, including the ones it shouldn't
1648 * that were added just above */
1649 if (! (ANYOF_FLAGS(node) & ANYOF_INVERT) && OP(node) == ANYOFD
f240c685
KW
1650 && (ANYOF_FLAGS(node) & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER))
1651 {
b423522f
KW
1652 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1653 }
1654
1655 /* Similarly for these */
93e92956 1656 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
e0a1ff7a 1657 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
b423522f
KW
1658 }
1659
1660 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1661 _invlist_invert(invlist);
1662 }
35b8412f
KW
1663 else if (ANYOF_FLAGS(node) & ANYOFL_FOLD) {
1664 if (new_node_has_latin1) {
31f05a37 1665
26be5fe6
KW
1666 /* Under /li, any 0-255 could fold to any other 0-255, depending on
1667 * the locale. We can skip this if there are no 0-255 at all. */
1668 _invlist_union(invlist, PL_Latin1, &invlist);
35b8412f
KW
1669
1670 invlist = add_cp_to_invlist(invlist, LATIN_SMALL_LETTER_DOTLESS_I);
1671 invlist = add_cp_to_invlist(invlist, LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
1672 }
1673 else {
1674 if (_invlist_contains_cp(invlist, LATIN_SMALL_LETTER_DOTLESS_I)) {
1675 invlist = add_cp_to_invlist(invlist, 'I');
1676 }
1677 if (_invlist_contains_cp(invlist,
1678 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE))
1679 {
1680 invlist = add_cp_to_invlist(invlist, 'i');
1681 }
1682 }
31f05a37
KW
1683 }
1684
1ee208c4
KW
1685 /* Similarly add the UTF-8 locale possible matches. These have to be
1686 * deferred until after the non-UTF-8 locale ones are taken care of just
1687 * above, or it leads to wrong results under ANYOF_INVERT */
1688 if (only_utf8_locale_invlist) {
31f05a37 1689 _invlist_union_maybe_complement_2nd(invlist,
1ee208c4 1690 only_utf8_locale_invlist,
31f05a37
KW
1691 ANYOF_FLAGS(node) & ANYOF_INVERT,
1692 &invlist);
1693 }
b423522f
KW
1694
1695 return invlist;
1696}
1697
1051e1c4 1698/* These two functions currently do the exact same thing */
557bd3fb 1699#define ssc_init_zero ssc_init
653099ff 1700
cdd87c1d
KW
1701#define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1702#define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1703
557bd3fb 1704/* 'AND' a given class with another one. Can create false positives. 'ssc'
93e92956
KW
1705 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1706 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
cdd87c1d 1707
653099ff 1708STATIC void
b423522f 1709S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1710 const regnode_charclass *and_with)
653099ff 1711{
cdd87c1d
KW
1712 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1713 * another SSC or a regular ANYOF class. Can create false positives. */
40d049e4 1714
a0dd4231
KW
1715 SV* anded_cp_list;
1716 U8 anded_flags;
1e6ade67 1717
cdd87c1d 1718 PERL_ARGS_ASSERT_SSC_AND;
653099ff 1719
71068078 1720 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1721
1722 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1723 * the code point inversion list and just the relevant flags */
71068078 1724 if (is_ANYOF_SYNTHETIC(and_with)) {
7dcac5f6 1725 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
a0dd4231 1726 anded_flags = ANYOF_FLAGS(and_with);
e9b08962
KW
1727
1728 /* XXX This is a kludge around what appears to be deficiencies in the
1729 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1730 * there are paths through the optimizer where it doesn't get weeded
1731 * out when it should. And if we don't make some extra provision for
1732 * it like the code just below, it doesn't get added when it should.
1733 * This solution is to add it only when AND'ing, which is here, and
1734 * only when what is being AND'ed is the pristine, original node
1735 * matching anything. Thus it is like adding it to ssc_anything() but
1736 * only when the result is to be AND'ed. Probably the same solution
1737 * could be adopted for the same problem we have with /l matching,
1738 * which is solved differently in S_ssc_init(), and that would lead to
1739 * fewer false positives than that solution has. But if this solution
1740 * creates bugs, the consequences are only that a warning isn't raised
1741 * that should be; while the consequences for having /l bugs is
1742 * incorrect matches */
7dcac5f6 1743 if (ssc_is_anything((regnode_ssc *)and_with)) {
f240c685 1744 anded_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
e9b08962 1745 }
a0dd4231
KW
1746 }
1747 else {
5c0f85ef 1748 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
f240c685
KW
1749 if (OP(and_with) == ANYOFD) {
1750 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1751 }
1752 else {
1753 anded_flags = ANYOF_FLAGS(and_with)
1754 &( ANYOF_COMMON_FLAGS
108316fb
KW
1755 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1756 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1757 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(and_with))) {
1758 anded_flags &=
1759 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1760 }
f240c685 1761 }
a0dd4231
KW
1762 }
1763
1764 ANYOF_FLAGS(ssc) &= anded_flags;
cdd87c1d
KW
1765
1766 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1767 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1768 * 'and_with' may be inverted. When not inverted, we have the situation of
1769 * computing:
1770 * (C1 | P1) & (C2 | P2)
1771 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1772 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1773 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1774 * <= ((C1 & C2) | P1 | P2)
1775 * Alternatively, the last few steps could be:
1776 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1777 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1778 * <= (C1 | C2 | (P1 & P2))
1779 * We favor the second approach if either P1 or P2 is non-empty. This is
1780 * because these components are a barrier to doing optimizations, as what
1781 * they match cannot be known until the moment of matching as they are
1782 * dependent on the current locale, 'AND"ing them likely will reduce or
1783 * eliminate them.
1784 * But we can do better if we know that C1,P1 are in their initial state (a
1785 * frequent occurrence), each matching everything:
1786 * (<everything>) & (C2 | P2) = C2 | P2
1787 * Similarly, if C2,P2 are in their initial state (again a frequent
1788 * occurrence), the result is a no-op
1789 * (C1 | P1) & (<everything>) = C1 | P1
1790 *
1791 * Inverted, we have
1792 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1793 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1794 * <= (C1 & ~C2) | (P1 & ~P2)
1795 * */
1aa99e6b 1796
a0dd4231 1797 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
71068078 1798 && ! is_ANYOF_SYNTHETIC(and_with))
a0dd4231 1799 {
cdd87c1d 1800 unsigned int i;
8951c461 1801
cdd87c1d
KW
1802 ssc_intersection(ssc,
1803 anded_cp_list,
1804 FALSE /* Has already been inverted */
1805 );
c6b76537 1806
cdd87c1d
KW
1807 /* If either P1 or P2 is empty, the intersection will be also; can skip
1808 * the loop */
93e92956 1809 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
cdd87c1d
KW
1810 ANYOF_POSIXL_ZERO(ssc);
1811 }
e0e1be5f 1812 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1813
1814 /* Note that the Posix class component P from 'and_with' actually
1815 * looks like:
1816 * P = Pa | Pb | ... | Pn
1817 * where each component is one posix class, such as in [\w\s].
1818 * Thus
1819 * ~P = ~(Pa | Pb | ... | Pn)
1820 * = ~Pa & ~Pb & ... & ~Pn
1821 * <= ~Pa | ~Pb | ... | ~Pn
1822 * The last is something we can easily calculate, but unfortunately
1823 * is likely to have many false positives. We could do better
1824 * in some (but certainly not all) instances if two classes in
1825 * P have known relationships. For example
1826 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1827 * So
1828 * :lower: & :print: = :lower:
1829 * And similarly for classes that must be disjoint. For example,
1830 * since \s and \w can have no elements in common based on rules in
1831 * the POSIX standard,
1832 * \w & ^\S = nothing
1833 * Unfortunately, some vendor locales do not meet the Posix
1834 * standard, in particular almost everything by Microsoft.
1835 * The loop below just changes e.g., \w into \W and vice versa */
1836
1ee208c4 1837 regnode_charclass_posixl temp;
cdd87c1d
KW
1838 int add = 1; /* To calculate the index of the complement */
1839
b1234259 1840 Zero(&temp, 1, regnode_charclass_posixl);
cdd87c1d
KW
1841 ANYOF_POSIXL_ZERO(&temp);
1842 for (i = 0; i < ANYOF_MAX; i++) {
1843 assert(i % 2 != 0
7dcac5f6
KW
1844 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1845 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
cdd87c1d 1846
7dcac5f6 1847 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
cdd87c1d
KW
1848 ANYOF_POSIXL_SET(&temp, i + add);
1849 }
1850 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1851 }
1852 ANYOF_POSIXL_AND(&temp, ssc);
c6b76537 1853
cdd87c1d
KW
1854 } /* else ssc already has no posixes */
1855 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1856 in its initial state */
71068078 1857 else if (! is_ANYOF_SYNTHETIC(and_with)
7dcac5f6 1858 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
cdd87c1d
KW
1859 {
1860 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1861 * copy it over 'ssc' */
1862 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
71068078 1863 if (is_ANYOF_SYNTHETIC(and_with)) {
cdd87c1d
KW
1864 StructCopy(and_with, ssc, regnode_ssc);
1865 }
1866 else {
1867 ssc->invlist = anded_cp_list;
1868 ANYOF_POSIXL_ZERO(ssc);
93e92956 1869 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1870 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
cdd87c1d
KW
1871 }
1872 }
1873 }
e0e1be5f 1874 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
93e92956 1875 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
cdd87c1d
KW
1876 {
1877 /* One or the other of P1, P2 is non-empty. */
93e92956 1878 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1ea8b7fe
KW
1879 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1880 }
cdd87c1d
KW
1881 ssc_union(ssc, anded_cp_list, FALSE);
1882 }
1883 else { /* P1 = P2 = empty */
1884 ssc_intersection(ssc, anded_cp_list, FALSE);
1885 }
137165a6 1886 }
653099ff
GS
1887}
1888
653099ff 1889STATIC void
cdd87c1d 1890S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1891 const regnode_charclass *or_with)
653099ff 1892{
cdd87c1d
KW
1893 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1894 * another SSC or a regular ANYOF class. Can create false positives if
1895 * 'or_with' is to be inverted. */
7918f24d 1896
a0dd4231
KW
1897 SV* ored_cp_list;
1898 U8 ored_flags;
c6b76537 1899
cdd87c1d 1900 PERL_ARGS_ASSERT_SSC_OR;
c6b76537 1901
71068078 1902 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1903
1904 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1905 * the code point inversion list and just the relevant flags */
71068078 1906 if (is_ANYOF_SYNTHETIC(or_with)) {
7dcac5f6 1907 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
a0dd4231
KW
1908 ored_flags = ANYOF_FLAGS(or_with);
1909 }
1910 else {
5c0f85ef 1911 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
eff8b7dc 1912 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
f240c685
KW
1913 if (OP(or_with) != ANYOFD) {
1914 ored_flags
1915 |= ANYOF_FLAGS(or_with)
108316fb
KW
1916 & ( ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1917 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1918 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(or_with))) {
1919 ored_flags |=
1920 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1921 }
f240c685 1922 }
a0dd4231
KW
1923 }
1924
1925 ANYOF_FLAGS(ssc) |= ored_flags;
cdd87c1d
KW
1926
1927 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1928 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1929 * 'or_with' may be inverted. When not inverted, we have the simple
1930 * situation of computing:
1931 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1932 * If P1|P2 yields a situation with both a class and its complement are
1933 * set, like having both \w and \W, this matches all code points, and we
1934 * can delete these from the P component of the ssc going forward. XXX We
1935 * might be able to delete all the P components, but I (khw) am not certain
1936 * about this, and it is better to be safe.
1937 *
1938 * Inverted, we have
1939 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1940 * <= (C1 | P1) | ~C2
1941 * <= (C1 | ~C2) | P1
1942 * (which results in actually simpler code than the non-inverted case)
1943 * */
9826f543 1944
a0dd4231 1945 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
71068078 1946 && ! is_ANYOF_SYNTHETIC(or_with))
a0dd4231 1947 {
cdd87c1d 1948 /* We ignore P2, leaving P1 going forward */
1ea8b7fe 1949 } /* else Not inverted */
93e92956 1950 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1951 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
e0e1be5f 1952 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1953 unsigned int i;
1954 for (i = 0; i < ANYOF_MAX; i += 2) {
1955 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1956 {
1957 ssc_match_all_cp(ssc);
1958 ANYOF_POSIXL_CLEAR(ssc, i);
1959 ANYOF_POSIXL_CLEAR(ssc, i+1);
cdd87c1d
KW
1960 }
1961 }
1962 }
1aa99e6b 1963 }
cdd87c1d
KW
1964
1965 ssc_union(ssc,
1966 ored_cp_list,
1967 FALSE /* Already has been inverted */
1968 );
653099ff
GS
1969}
1970
b423522f
KW
1971PERL_STATIC_INLINE void
1972S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1973{
1974 PERL_ARGS_ASSERT_SSC_UNION;
1975
71068078 1976 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1977
1978 _invlist_union_maybe_complement_2nd(ssc->invlist,
1979 invlist,
1980 invert2nd,
1981 &ssc->invlist);
1982}
1983
1984PERL_STATIC_INLINE void
1985S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1986 SV* const invlist,
1987 const bool invert2nd)
1988{
1989 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1990
71068078 1991 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1992
1993 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1994 invlist,
1995 invert2nd,
1996 &ssc->invlist);
1997}
1998
1999PERL_STATIC_INLINE void
2000S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
2001{
2002 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
2003
71068078 2004 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
2005
2006 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
2007}
2008
2009PERL_STATIC_INLINE void
2010S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
2011{
2012 /* AND just the single code point 'cp' into the SSC 'ssc' */
2013
2014 SV* cp_list = _new_invlist(2);
2015
2016 PERL_ARGS_ASSERT_SSC_CP_AND;
2017
71068078 2018 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
2019
2020 cp_list = add_cp_to_invlist(cp_list, cp);
2021 ssc_intersection(ssc, cp_list,
2022 FALSE /* Not inverted */
2023 );
2024 SvREFCNT_dec_NN(cp_list);
2025}
2026
2027PERL_STATIC_INLINE void
dc3bf405 2028S_ssc_clear_locale(regnode_ssc *ssc)
b423522f
KW
2029{
2030 /* Set the SSC 'ssc' to not match any locale things */
b423522f
KW
2031 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
2032
71068078 2033 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
2034
2035 ANYOF_POSIXL_ZERO(ssc);
2036 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
2037}
2038
b35552de
KW
2039#define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
2040
2041STATIC bool
2042S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
2043{
2044 /* The synthetic start class is used to hopefully quickly winnow down
2045 * places where a pattern could start a match in the target string. If it
2046 * doesn't really narrow things down that much, there isn't much point to
2047 * having the overhead of using it. This function uses some very crude
2048 * heuristics to decide if to use the ssc or not.
2049 *
2050 * It returns TRUE if 'ssc' rules out more than half what it considers to
2051 * be the "likely" possible matches, but of course it doesn't know what the
2052 * actual things being matched are going to be; these are only guesses
2053 *
2054 * For /l matches, it assumes that the only likely matches are going to be
2055 * in the 0-255 range, uniformly distributed, so half of that is 127
2056 * For /a and /d matches, it assumes that the likely matches will be just
2057 * the ASCII range, so half of that is 63
2058 * For /u and there isn't anything matching above the Latin1 range, it
2059 * assumes that that is the only range likely to be matched, and uses
2060 * half that as the cut-off: 127. If anything matches above Latin1,
2061 * it assumes that all of Unicode could match (uniformly), except for
2062 * non-Unicode code points and things in the General Category "Other"
2063 * (unassigned, private use, surrogates, controls and formats). This
2064 * is a much large number. */
2065
b35552de
KW
2066 U32 count = 0; /* Running total of number of code points matched by
2067 'ssc' */
2068 UV start, end; /* Start and end points of current range in inversion
26be5fe6 2069 XXX outdated. UTF-8 locales are common, what about invert? list */
72400949
KW
2070 const U32 max_code_points = (LOC)
2071 ? 256
f6e8f31e 2072 : (( ! UNI_SEMANTICS
1a26cbcb 2073 || invlist_highest(ssc->invlist) < 256)
72400949
KW
2074 ? 128
2075 : NON_OTHER_COUNT);
2076 const U32 max_match = max_code_points / 2;
b35552de
KW
2077
2078 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
2079
2080 invlist_iterinit(ssc->invlist);
2081 while (invlist_iternext(ssc->invlist, &start, &end)) {
72400949
KW
2082 if (start >= max_code_points) {
2083 break;
b35552de 2084 }
72400949 2085 end = MIN(end, max_code_points - 1);
b35552de 2086 count += end - start + 1;
72400949 2087 if (count >= max_match) {
b35552de
KW
2088 invlist_iterfinish(ssc->invlist);
2089 return FALSE;
2090 }
2091 }
2092
2093 return TRUE;
2094}
2095
2096
b423522f
KW
2097STATIC void
2098S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
2099{
2100 /* The inversion list in the SSC is marked mortal; now we need a more
2101 * permanent copy, which is stored the same way that is done in a regular
dcb20b36
KW
2102 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
2103 * map */
b423522f 2104
28118b9c 2105 SV* invlist = invlist_clone(ssc->invlist, NULL);
b423522f
KW
2106
2107 PERL_ARGS_ASSERT_SSC_FINALIZE;
2108
71068078 2109 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f 2110
a0dd4231 2111 /* The code in this file assumes that all but these flags aren't relevant
93e92956
KW
2112 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
2113 * by the time we reach here */
f240c685
KW
2114 assert(! (ANYOF_FLAGS(ssc)
2115 & ~( ANYOF_COMMON_FLAGS
108316fb
KW
2116 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
2117 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)));
a0dd4231 2118
b423522f
KW
2119 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
2120
4c404f26 2121 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist, NULL, NULL);
b423522f 2122
85c8e306
KW
2123 /* Make sure is clone-safe */
2124 ssc->invlist = NULL;
2125
e0e1be5f 2126 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
93e92956 2127 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
d156f5cb 2128 OP(ssc) = ANYOFPOSIXL;
e0e1be5f 2129 }
d156f5cb 2130 else if (RExC_contains_locale) {
b2e90ddf
KW
2131 OP(ssc) = ANYOFL;
2132 }
2133
1462525b 2134 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
b423522f
KW
2135}
2136
a3621e74
YO
2137#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
2138#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
2139#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
538e84ed
KW
2140#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
2141 ? (TRIE_LIST_CUR( idx ) - 1) \
2142 : 0 )
a3621e74 2143
3dab1dad
YO
2144
2145#ifdef DEBUGGING
07be1b83 2146/*
2b8b4781
NC
2147 dump_trie(trie,widecharmap,revcharmap)
2148 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
2149 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
2150
2151 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
2152 The _interim_ variants are used for debugging the interim
2153 tables that are used to generate the final compressed
2154 representation which is what dump_trie expects.
2155
486ec47a 2156 Part of the reason for their existence is to provide a form
3dab1dad 2157 of documentation as to how the different representations function.
07be1b83
YO
2158
2159*/
3dab1dad
YO
2160
2161/*
3dab1dad
YO
2162 Dumps the final compressed table form of the trie to Perl_debug_log.
2163 Used for debugging make_trie().
2164*/
b9a59e08 2165
3dab1dad 2166STATIC void
2b8b4781
NC
2167S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
2168 AV *revcharmap, U32 depth)
3dab1dad
YO
2169{
2170 U32 state;
ab3bbdeb 2171 SV *sv=sv_newmortal();
55eed653 2172 int colwidth= widecharmap ? 6 : 4;
2e64971a 2173 U16 word;
3dab1dad
YO
2174 GET_RE_DEBUG_FLAGS_DECL;
2175
7918f24d 2176 PERL_ARGS_ASSERT_DUMP_TRIE;
ab3bbdeb 2177
6ad9a8ab 2178 Perl_re_indentf( aTHX_ "Char : %-6s%-6s%-4s ",
1e37780e 2179 depth+1, "Match","Base","Ofs" );
3dab1dad
YO
2180
2181 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 2182 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 2183 if ( tmp ) {
6ad9a8ab 2184 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2185 colwidth,
538e84ed 2186 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2187 PL_colors[0], PL_colors[1],
2188 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2189 PERL_PV_ESCAPE_FIRSTCHAR
2190 )
ab3bbdeb 2191 );
3dab1dad
YO
2192 }
2193 }
1e37780e
YO
2194 Perl_re_printf( aTHX_ "\n");
2195 Perl_re_indentf( aTHX_ "State|-----------------------", depth+1);
3dab1dad
YO
2196
2197 for( state = 0 ; state < trie->uniquecharcount ; state++ )
6ad9a8ab
YO
2198 Perl_re_printf( aTHX_ "%.*s", colwidth, "--------");
2199 Perl_re_printf( aTHX_ "\n");
3dab1dad 2200
1e2e3d02 2201 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 2202 const U32 base = trie->states[ state ].trans.base;
3dab1dad 2203
147e3846 2204 Perl_re_indentf( aTHX_ "#%4" UVXf "|", depth+1, (UV)state);
3dab1dad
YO
2205
2206 if ( trie->states[ state ].wordnum ) {
1e37780e 2207 Perl_re_printf( aTHX_ " W%4X", trie->states[ state ].wordnum );
3dab1dad 2208 } else {
6ad9a8ab 2209 Perl_re_printf( aTHX_ "%6s", "" );
3dab1dad
YO
2210 }
2211
147e3846 2212 Perl_re_printf( aTHX_ " @%4" UVXf " ", (UV)base );
3dab1dad
YO
2213
2214 if ( base ) {
2215 U32 ofs = 0;
2216
2217 while( ( base + ofs < trie->uniquecharcount ) ||
2218 ( base + ofs - trie->uniquecharcount < trie->lasttrans
538e84ed
KW
2219 && trie->trans[ base + ofs - trie->uniquecharcount ].check
2220 != state))
3dab1dad
YO
2221 ofs++;
2222
147e3846 2223 Perl_re_printf( aTHX_ "+%2" UVXf "[ ", (UV)ofs);
3dab1dad
YO
2224
2225 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
538e84ed
KW
2226 if ( ( base + ofs >= trie->uniquecharcount )
2227 && ( base + ofs - trie->uniquecharcount
2228 < trie->lasttrans )
2229 && trie->trans[ base + ofs
2230 - trie->uniquecharcount ].check == state )
3dab1dad 2231 {
147e3846 2232 Perl_re_printf( aTHX_ "%*" UVXf, colwidth,
1e37780e
YO
2233 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next
2234 );
3dab1dad 2235 } else {
88f063b4 2236 Perl_re_printf( aTHX_ "%*s", colwidth," ." );
3dab1dad
YO
2237 }
2238 }
2239
6ad9a8ab 2240 Perl_re_printf( aTHX_ "]");
3dab1dad
YO
2241
2242 }
6ad9a8ab 2243 Perl_re_printf( aTHX_ "\n" );
3dab1dad 2244 }
6ad9a8ab 2245 Perl_re_indentf( aTHX_ "word_info N:(prev,len)=",
cb41e5d6 2246 depth);
2e64971a 2247 for (word=1; word <= trie->wordcount; word++) {
6ad9a8ab 2248 Perl_re_printf( aTHX_ " %d:(%d,%d)",
2e64971a
DM
2249 (int)word, (int)(trie->wordinfo[word].prev),
2250 (int)(trie->wordinfo[word].len));
2251 }
6ad9a8ab 2252 Perl_re_printf( aTHX_ "\n" );
538e84ed 2253}
3dab1dad 2254/*
3dab1dad 2255 Dumps a fully constructed but uncompressed trie in list form.
538e84ed 2256 List tries normally only are used for construction when the number of
3dab1dad
YO
2257 possible chars (trie->uniquecharcount) is very high.
2258 Used for debugging make_trie().
2259*/
2260STATIC void
55eed653 2261S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2262 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2263 U32 depth)
3dab1dad
YO
2264{
2265 U32 state;
ab3bbdeb 2266 SV *sv=sv_newmortal();
55eed653 2267 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2268 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2269
2270 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
2271
3dab1dad 2272 /* print out the table precompression. */
6ad9a8ab 2273 Perl_re_indentf( aTHX_ "State :Word | Transition Data\n",
cb41e5d6 2274 depth+1 );
6ad9a8ab 2275 Perl_re_indentf( aTHX_ "%s",
cb41e5d6 2276 depth+1, "------:-----+-----------------\n" );
538e84ed 2277
3dab1dad
YO
2278 for( state=1 ; state < next_alloc ; state ++ ) {
2279 U16 charid;
538e84ed 2280
147e3846 2281 Perl_re_indentf( aTHX_ " %4" UVXf " :",
cb41e5d6 2282 depth+1, (UV)state );
3dab1dad 2283 if ( ! trie->states[ state ].wordnum ) {
6ad9a8ab 2284 Perl_re_printf( aTHX_ "%5s| ","");
3dab1dad 2285 } else {
6ad9a8ab 2286 Perl_re_printf( aTHX_ "W%4x| ",
3dab1dad
YO
2287 trie->states[ state ].wordnum
2288 );
2289 }
2290 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
538e84ed 2291 SV ** const tmp = av_fetch( revcharmap,
88f063b4 2292 TRIE_LIST_ITEM(state, charid).forid, 0);
ab3bbdeb 2293 if ( tmp ) {
147e3846 2294 Perl_re_printf( aTHX_ "%*s:%3X=%4" UVXf " | ",
ab3bbdeb 2295 colwidth,
538e84ed
KW
2296 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
2297 colwidth,
2298 PL_colors[0], PL_colors[1],
2299 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2300 | PERL_PV_ESCAPE_FIRSTCHAR
ab3bbdeb 2301 ) ,
88f063b4
KW
2302 TRIE_LIST_ITEM(state, charid).forid,
2303 (UV)TRIE_LIST_ITEM(state, charid).newstate
1e2e3d02 2304 );
538e84ed 2305 if (!(charid % 10))
6ad9a8ab 2306 Perl_re_printf( aTHX_ "\n%*s| ",
664e119d 2307 (int)((depth * 2) + 14), "");
1e2e3d02 2308 }
ab3bbdeb 2309 }
6ad9a8ab 2310 Perl_re_printf( aTHX_ "\n");
3dab1dad 2311 }
538e84ed 2312}
3dab1dad
YO
2313
2314/*
3dab1dad 2315 Dumps a fully constructed but uncompressed trie in table form.
538e84ed
KW
2316 This is the normal DFA style state transition table, with a few
2317 twists to facilitate compression later.
3dab1dad
YO
2318 Used for debugging make_trie().
2319*/
2320STATIC void
55eed653 2321S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2322 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2323 U32 depth)
3dab1dad
YO
2324{
2325 U32 state;
2326 U16 charid;
ab3bbdeb 2327 SV *sv=sv_newmortal();
55eed653 2328 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2329 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2330
2331 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
538e84ed 2332
3dab1dad
YO
2333 /*
2334 print out the table precompression so that we can do a visual check
2335 that they are identical.
2336 */
538e84ed 2337
6ad9a8ab 2338 Perl_re_indentf( aTHX_ "Char : ", depth+1 );
3dab1dad
YO
2339
2340 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 2341 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 2342 if ( tmp ) {
6ad9a8ab 2343 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2344 colwidth,
538e84ed 2345 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2346 PL_colors[0], PL_colors[1],
2347 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2348 PERL_PV_ESCAPE_FIRSTCHAR
2349 )
ab3bbdeb 2350 );
3dab1dad
YO
2351 }
2352 }
2353
4aaafc03
YO
2354 Perl_re_printf( aTHX_ "\n");
2355 Perl_re_indentf( aTHX_ "State+-", depth+1 );
3dab1dad
YO
2356
2357 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
6ad9a8ab 2358 Perl_re_printf( aTHX_ "%.*s", colwidth,"--------");
3dab1dad
YO
2359 }
2360
6ad9a8ab 2361 Perl_re_printf( aTHX_ "\n" );
3dab1dad
YO
2362
2363 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
2364
147e3846 2365 Perl_re_indentf( aTHX_ "%4" UVXf " : ",
cb41e5d6 2366 depth+1,
3dab1dad
YO
2367 (UV)TRIE_NODENUM( state ) );
2368
2369 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
2370 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
2371 if (v)
147e3846 2372 Perl_re_printf( aTHX_ "%*" UVXf, colwidth, v );
ab3bbdeb 2373 else
6ad9a8ab 2374 Perl_re_printf( aTHX_ "%*s", colwidth, "." );
3dab1dad
YO
2375 }
2376 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
147e3846 2377 Perl_re_printf( aTHX_ " (%4" UVXf ")\n",
538e84ed 2378 (UV)trie->trans[ state ].check );
3dab1dad 2379 } else {
147e3846 2380 Perl_re_printf( aTHX_ " (%4" UVXf ") W%4X\n",
538e84ed 2381 (UV)trie->trans[ state ].check,
3dab1dad
YO
2382 trie->states[ TRIE_NODENUM( state ) ].wordnum );
2383 }
2384 }
07be1b83 2385}
3dab1dad
YO
2386
2387#endif
2388
2e64971a 2389
786e8c11
YO
2390/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
2391 startbranch: the first branch in the whole branch sequence
2392 first : start branch of sequence of branch-exact nodes.
2393 May be the same as startbranch
2394 last : Thing following the last branch.
2395 May be the same as tail.
2396 tail : item following the branch sequence
2397 count : words in the sequence
a4525e78 2398 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
786e8c11 2399 depth : indent depth
3dab1dad 2400
786e8c11 2401Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 2402
786e8c11
YO
2403A trie is an N'ary tree where the branches are determined by digital
2404decomposition of the key. IE, at the root node you look up the 1st character and
2405follow that branch repeat until you find the end of the branches. Nodes can be
2406marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 2407
786e8c11 2408 /he|she|his|hers/
72f13be8 2409
786e8c11
YO
2410would convert into the following structure. Numbers represent states, letters
2411following numbers represent valid transitions on the letter from that state, if
2412the number is in square brackets it represents an accepting state, otherwise it
2413will be in parenthesis.
07be1b83 2414
786e8c11
YO
2415 +-h->+-e->[3]-+-r->(8)-+-s->[9]
2416 | |
2417 | (2)
2418 | |
2419 (1) +-i->(6)-+-s->[7]
2420 |
2421 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 2422
786e8c11
YO
2423 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
2424
2425This shows that when matching against the string 'hers' we will begin at state 1
2426read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
2427then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
2428is also accepting. Thus we know that we can match both 'he' and 'hers' with a
2429single traverse. We store a mapping from accepting to state to which word was
2430matched, and then when we have multiple possibilities we try to complete the
b8fda935 2431rest of the regex in the order in which they occurred in the alternation.
786e8c11
YO
2432
2433The only prior NFA like behaviour that would be changed by the TRIE support is
2434the silent ignoring of duplicate alternations which are of the form:
2435
2436 / (DUPE|DUPE) X? (?{ ... }) Y /x
2437
4b714af6 2438Thus EVAL blocks following a trie may be called a different number of times with
786e8c11 2439and without the optimisation. With the optimisations dupes will be silently
486ec47a 2440ignored. This inconsistent behaviour of EVAL type nodes is well established as
786e8c11
YO
2441the following demonstrates:
2442
2443 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
2444
2445which prints out 'word' three times, but
2446
2447 'words'=~/(word|word|word)(?{ print $1 })S/
2448
2449which doesnt print it out at all. This is due to other optimisations kicking in.
2450
2451Example of what happens on a structural level:
2452
486ec47a 2453The regexp /(ac|ad|ab)+/ will produce the following debug output:
786e8c11
YO
2454
2455 1: CURLYM[1] {1,32767}(18)
2456 5: BRANCH(8)
2457 6: EXACT <ac>(16)
2458 8: BRANCH(11)
2459 9: EXACT <ad>(16)
2460 11: BRANCH(14)
2461 12: EXACT <ab>(16)
2462 16: SUCCEED(0)
2463 17: NOTHING(18)
2464 18: END(0)
2465
2466This would be optimizable with startbranch=5, first=5, last=16, tail=16
2467and should turn into:
2468
2469 1: CURLYM[1] {1,32767}(18)
2470 5: TRIE(16)
2471 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
2472 <ac>
2473 <ad>
2474 <ab>
2475 16: SUCCEED(0)
2476 17: NOTHING(18)
2477 18: END(0)
2478
2479Cases where tail != last would be like /(?foo|bar)baz/:
2480
2481 1: BRANCH(4)
2482 2: EXACT <foo>(8)
2483 4: BRANCH(7)
2484 5: EXACT <bar>(8)
2485 7: TAIL(8)
2486 8: EXACT <baz>(10)
2487 10: END(0)
2488
2489which would be optimizable with startbranch=1, first=1, last=7, tail=8
2490and would end up looking like:
2491
2492 1: TRIE(8)
2493 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
2494 <foo>
2495 <bar>
2496 7: TAIL(8)
2497 8: EXACT <baz>(10)
2498 10: END(0)
2499
c80e42f3 2500 d = uvchr_to_utf8_flags(d, uv, 0);
786e8c11
YO
2501
2502is the recommended Unicode-aware way of saying
2503
2504 *(d++) = uv;
2505*/
2506
fab2782b 2507#define TRIE_STORE_REVCHAR(val) \
786e8c11 2508 STMT_START { \
73031816 2509 if (UTF) { \
668fcfea 2510 SV *zlopp = newSV(UTF8_MAXBYTES); \
88c9ea1e 2511 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
c80e42f3 2512 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
73031816
NC
2513 SvCUR_set(zlopp, kapow - flrbbbbb); \
2514 SvPOK_on(zlopp); \
2515 SvUTF8_on(zlopp); \
2516 av_push(revcharmap, zlopp); \
2517 } else { \
fab2782b 2518 char ooooff = (char)val; \
73031816
NC
2519 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2520 } \
2521 } STMT_END
786e8c11 2522
914a25d5
KW
2523/* This gets the next character from the input, folding it if not already
2524 * folded. */
2525#define TRIE_READ_CHAR STMT_START { \
2526 wordlen++; \
2527 if ( UTF ) { \
2528 /* if it is UTF then it is either already folded, or does not need \
2529 * folding */ \
1c1d615a 2530 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
914a25d5
KW
2531 } \
2532 else if (folder == PL_fold_latin1) { \
7d006b13
KW
2533 /* This folder implies Unicode rules, which in the range expressible \
2534 * by not UTF is the lower case, with the two exceptions, one of \
2535 * which should have been taken care of before calling this */ \
2536 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2537 uvc = toLOWER_L1(*uc); \
2538 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2539 len = 1; \
914a25d5
KW
2540 } else { \
2541 /* raw data, will be folded later if needed */ \
2542 uvc = (U32)*uc; \
2543 len = 1; \
2544 } \
786e8c11
YO
2545} STMT_END
2546
2547
2548
2549#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2550 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
00195859 2551 U32 ging = TRIE_LIST_LEN( state ) * 2; \
f9003953 2552 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
00195859 2553 TRIE_LIST_LEN( state ) = ging; \
786e8c11
YO
2554 } \
2555 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2556 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2557 TRIE_LIST_CUR( state )++; \
2558} STMT_END
07be1b83 2559
786e8c11 2560#define TRIE_LIST_NEW(state) STMT_START { \
d09f14bf 2561 Newx( trie->states[ state ].trans.list, \
786e8c11
YO
2562 4, reg_trie_trans_le ); \
2563 TRIE_LIST_CUR( state ) = 1; \
2564 TRIE_LIST_LEN( state ) = 4; \
2565} STMT_END
07be1b83 2566
786e8c11
YO
2567#define TRIE_HANDLE_WORD(state) STMT_START { \
2568 U16 dupe= trie->states[ state ].wordnum; \
2569 regnode * const noper_next = regnext( noper ); \
2570 \
786e8c11
YO
2571 DEBUG_r({ \
2572 /* store the word for dumping */ \
2573 SV* tmp; \
2574 if (OP(noper) != NOTHING) \
740cce10 2575 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
786e8c11 2576 else \
740cce10 2577 tmp = newSVpvn_utf8( "", 0, UTF ); \
2b8b4781 2578 av_push( trie_words, tmp ); \
786e8c11
YO
2579 }); \
2580 \
2581 curword++; \
2e64971a
DM
2582 trie->wordinfo[curword].prev = 0; \
2583 trie->wordinfo[curword].len = wordlen; \
2584 trie->wordinfo[curword].accept = state; \
786e8c11
YO
2585 \
2586 if ( noper_next < tail ) { \
2587 if (!trie->jump) \
538e84ed
KW
2588 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2589 sizeof(U16) ); \
7f69552c 2590 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
2591 if (!jumper) \
2592 jumper = noper_next; \
2593 if (!nextbranch) \
2594 nextbranch= regnext(cur); \
2595 } \
2596 \
2597 if ( dupe ) { \
2e64971a
DM
2598 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2599 /* chain, so that when the bits of chain are later */\
2600 /* linked together, the dups appear in the chain */\
2601 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2602 trie->wordinfo[dupe].prev = curword; \
786e8c11
YO
2603 } else { \
2604 /* we haven't inserted this word yet. */ \
2605 trie->states[ state ].wordnum = curword; \
2606 } \
2607} STMT_END
07be1b83 2608
3dab1dad 2609
786e8c11
YO
2610#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2611 ( ( base + charid >= ucharcount \
2612 && base + charid < ubound \
2613 && state == trie->trans[ base - ucharcount + charid ].check \
2614 && trie->trans[ base - ucharcount + charid ].next ) \
2615 ? trie->trans[ base - ucharcount + charid ].next \
2616 : ( state==1 ? special : 0 ) \
2617 )
3dab1dad 2618
8bcafbf4
YO
2619#define TRIE_BITMAP_SET_FOLDED(trie, uvc, folder) \
2620STMT_START { \
2621 TRIE_BITMAP_SET(trie, uvc); \
2622 /* store the folded codepoint */ \
2623 if ( folder ) \
2624 TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); \
2625 \
2626 if ( !UTF ) { \
2627 /* store first byte of utf8 representation of */ \
2628 /* variant codepoints */ \
2629 if (! UVCHR_IS_INVARIANT(uvc)) { \
2630 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); \
2631 } \
2632 } \
2633} STMT_END
786e8c11
YO
2634#define MADE_TRIE 1
2635#define MADE_JUMP_TRIE 2
2636#define MADE_EXACT_TRIE 4
3dab1dad 2637
a3621e74 2638STATIC I32
538e84ed
KW
2639S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2640 regnode *first, regnode *last, regnode *tail,
2641 U32 word_count, U32 flags, U32 depth)
a3621e74
YO
2642{
2643 /* first pass, loop through and scan words */
2644 reg_trie_data *trie;
55eed653 2645 HV *widecharmap = NULL;
2b8b4781 2646 AV *revcharmap = newAV();
a3621e74 2647 regnode *cur;
a3621e74
YO
2648 STRLEN len = 0;
2649 UV uvc = 0;
2650 U16 curword = 0;
2651 U32 next_alloc = 0;
786e8c11
YO
2652 regnode *jumper = NULL;
2653 regnode *nextbranch = NULL;
7f69552c 2654 regnode *convert = NULL;
2e64971a 2655 U32 *prev_states; /* temp array mapping each state to previous one */
a3621e74 2656 /* we just use folder as a flag in utf8 */
1e696034 2657 const U8 * folder = NULL;
a3621e74 2658
3a611511
YO
2659 /* in the below add_data call we are storing either 'tu' or 'tuaa'
2660 * which stands for one trie structure, one hash, optionally followed
2661 * by two arrays */
2b8b4781 2662#ifdef DEBUGGING
3a611511 2663 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuaa"));
2b8b4781
NC
2664 AV *trie_words = NULL;
2665 /* along with revcharmap, this only used during construction but both are
2666 * useful during debugging so we store them in the struct when debugging.
8e11feef 2667 */
2b8b4781 2668#else
cf78de0b 2669 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
3dab1dad 2670 STRLEN trie_charcount=0;
3dab1dad 2671#endif
2b8b4781 2672 SV *re_trie_maxbuff;
a3621e74 2673 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2674
2675 PERL_ARGS_ASSERT_MAKE_TRIE;
72f13be8
YO
2676#ifndef DEBUGGING
2677 PERL_UNUSED_ARG(depth);
2678#endif
a3621e74 2679
1e696034 2680 switch (flags) {
f6b4b99d 2681 case EXACT: case EXACT_ONLY8: case EXACTL: break;
89829bb5 2682 case EXACTFAA:
0ea669f4 2683 case EXACTFUP:
a4525e78
KW
2684 case EXACTFU:
2685 case EXACTFLU8: folder = PL_fold_latin1; break;
1e696034 2686 case EXACTF: folder = PL_fold; break;
fab2782b 2687 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
1e696034
KW
2688 }
2689
c944940b 2690 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 2691 trie->refcount = 1;
3dab1dad 2692 trie->startstate = 1;
786e8c11 2693 trie->wordcount = word_count;
f8fc2ecf 2694 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 2695 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
f6b4b99d 2696 if (flags == EXACT || flags == EXACT_ONLY8 || flags == EXACTL)
c944940b 2697 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2e64971a
DM
2698 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2699 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2700
a3621e74 2701 DEBUG_r({
2b8b4781 2702 trie_words = newAV();
a3621e74 2703 });
a3621e74 2704
3b89859a 2705 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, GV_ADD);
316ebaf2 2706 assert(re_trie_maxbuff);
a3621e74 2707 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 2708 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 2709 }
df826430 2710 DEBUG_TRIE_COMPILE_r({
6ad9a8ab 2711 Perl_re_indentf( aTHX_
cb41e5d6
YO
2712 "make_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2713 depth+1,
88f063b4 2714 REG_NODE_NUM(startbranch), REG_NODE_NUM(first),
538e84ed 2715 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
3dab1dad 2716 });
538e84ed 2717
7f69552c
YO
2718 /* Find the node we are going to overwrite */
2719 if ( first == startbranch && OP( last ) != BRANCH ) {
2720 /* whole branch chain */
2721 convert = first;
2722 } else {
2723 /* branch sub-chain */
2724 convert = NEXTOPER( first );
2725 }
538e84ed 2726
a3621e74
YO
2727 /* -- First loop and Setup --
2728
2729 We first traverse the branches and scan each word to determine if it
2730 contains widechars, and how many unique chars there are, this is
2731 important as we have to build a table with at least as many columns as we
2732 have unique chars.
2733
2734 We use an array of integers to represent the character codes 0..255
538e84ed
KW
2735 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2736 the native representation of the character value as the key and IV's for
2737 the coded index.
a3621e74
YO
2738
2739 *TODO* If we keep track of how many times each character is used we can
2740 remap the columns so that the table compression later on is more
3b753521 2741 efficient in terms of memory by ensuring the most common value is in the
a3621e74
YO
2742 middle and the least common are on the outside. IMO this would be better
2743 than a most to least common mapping as theres a decent chance the most
2744 common letter will share a node with the least common, meaning the node
486ec47a 2745 will not be compressible. With a middle is most common approach the worst
a3621e74
YO
2746 case is when we have the least common nodes twice.
2747
2748 */
2749
a3621e74 2750 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
df826430 2751 regnode *noper = NEXTOPER( cur );
944e05e3
YO
2752 const U8 *uc;
2753 const U8 *e;
bc031a7d 2754 int foldlen = 0;
07be1b83 2755 U32 wordlen = 0; /* required init */
bc031a7d
KW
2756 STRLEN minchars = 0;
2757 STRLEN maxchars = 0;
538e84ed
KW
2758 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2759 bitmap?*/
a3621e74 2760
3dab1dad 2761 if (OP(noper) == NOTHING) {
20ed8c88
YO
2762 /* skip past a NOTHING at the start of an alternation
2763 * eg, /(?:)a|(?:b)/ should be the same as /a|b/
2764 */
df826430 2765 regnode *noper_next= regnext(noper);
944e05e3
YO
2766 if (noper_next < tail)
2767 noper= noper_next;
2768 }
2769
f6b4b99d
KW
2770 if ( noper < tail
2771 && ( OP(noper) == flags
2772 || (flags == EXACT && OP(noper) == EXACT_ONLY8)
a9f8c7ac 2773 || (flags == EXACTFU && ( OP(noper) == EXACTFU_ONLY8
0ea669f4 2774 || OP(noper) == EXACTFUP))))
f6b4b99d 2775 {
944e05e3
YO
2776 uc= (U8*)STRING(noper);
2777 e= uc + STR_LEN(noper);
2778 } else {
2779 trie->minlen= 0;
2780 continue;
3dab1dad 2781 }
df826430 2782
944e05e3 2783
fab2782b 2784 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
02daf0ab
YO
2785 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2786 regardless of encoding */
0ea669f4 2787 if (OP( noper ) == EXACTFUP) {
fab2782b 2788 /* false positives are ok, so just set this */
0dc4a61d 2789 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
fab2782b
YO
2790 }
2791 }
dca5fc4c 2792
bc031a7d
KW
2793 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2794 branch */
3dab1dad 2795 TRIE_CHARCOUNT(trie)++;
a3621e74 2796 TRIE_READ_CHAR;
645de4ce 2797
bc031a7d
KW
2798 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2799 * is in effect. Under /i, this character can match itself, or
2800 * anything that folds to it. If not under /i, it can match just
2801 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2802 * all fold to k, and all are single characters. But some folds
2803 * expand to more than one character, so for example LATIN SMALL
2804 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2805 * the string beginning at 'uc' is 'ffi', it could be matched by
2806 * three characters, or just by the one ligature character. (It
2807 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2808 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2809 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2810 * match.) The trie needs to know the minimum and maximum number
2811 * of characters that could match so that it can use size alone to
2812 * quickly reject many match attempts. The max is simple: it is
2813 * the number of folded characters in this branch (since a fold is
2814 * never shorter than what folds to it. */
2815
2816 maxchars++;
2817
2818 /* And the min is equal to the max if not under /i (indicated by
2819 * 'folder' being NULL), or there are no multi-character folds. If
2820 * there is a multi-character fold, the min is incremented just
2821 * once, for the character that folds to the sequence. Each
2822 * character in the sequence needs to be added to the list below of
2823 * characters in the trie, but we count only the first towards the
2824 * min number of characters needed. This is done through the
2825 * variable 'foldlen', which is returned by the macros that look
2826 * for these sequences as the number of bytes the sequence
2827 * occupies. Each time through the loop, we decrement 'foldlen' by
2828 * how many bytes the current char occupies. Only when it reaches
2829 * 0 do we increment 'minchars' or look for another multi-character
2830 * sequence. */
2831 if (folder == NULL) {
2832 minchars++;
2833 }
2834 else if (foldlen > 0) {
2835 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
645de4ce
KW
2836 }
2837 else {
bc031a7d
KW
2838 minchars++;
2839
2840 /* See if *uc is the beginning of a multi-character fold. If
2841 * so, we decrement the length remaining to look at, to account
2842 * for the current character this iteration. (We can use 'uc'
2843 * instead of the fold returned by TRIE_READ_CHAR because for
2844 * non-UTF, the latin1_safe macro is smart enough to account
2845 * for all the unfolded characters, and because for UTF, the
2846 * string will already have been folded earlier in the
2847 * compilation process */
2848 if (UTF) {
2849 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2850 foldlen -= UTF8SKIP(uc);
645de4ce
KW
2851 }
2852 }
bc031a7d
KW
2853 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2854 foldlen--;
2855 }
645de4ce 2856 }
bc031a7d
KW
2857
2858 /* The current character (and any potential folds) should be added
2859 * to the possible matching characters for this position in this
2860 * branch */
a3621e74 2861 if ( uvc < 256 ) {
fab2782b
YO
2862 if ( folder ) {
2863 U8 folded= folder[ (U8) uvc ];
2864 if ( !trie->charmap[ folded ] ) {
2865 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2866 TRIE_STORE_REVCHAR( folded );
2867 }
2868 }
a3621e74
YO
2869 if ( !trie->charmap[ uvc ] ) {
2870 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
fab2782b 2871 TRIE_STORE_REVCHAR( uvc );
a3621e74 2872 }
02daf0ab 2873 if ( set_bit ) {
62012aee
KW
2874 /* store the codepoint in the bitmap, and its folded
2875 * equivalent. */
8bcafbf4 2876 TRIE_BITMAP_SET_FOLDED(trie, uvc, folder);
02daf0ab
YO
2877 set_bit = 0; /* We've done our bit :-) */
2878 }
a3621e74 2879 } else {
bc031a7d
KW
2880
2881 /* XXX We could come up with the list of code points that fold
2882 * to this using PL_utf8_foldclosures, except not for
2883 * multi-char folds, as there may be multiple combinations
2884 * there that could work, which needs to wait until runtime to
2885 * resolve (The comment about LIGATURE FFI above is such an
2886 * example */
2887
a3621e74 2888 SV** svpp;
55eed653
NC
2889 if ( !widecharmap )
2890 widecharmap = newHV();
a3621e74 2891
55eed653 2892 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
2893
2894 if ( !svpp )
147e3846 2895 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%" UVXf, uvc );
a3621e74
YO
2896
2897 if ( !SvTRUE( *svpp ) ) {
2898 sv_setiv( *svpp, ++trie->uniquecharcount );
fab2782b 2899 TRIE_STORE_REVCHAR(uvc);
a3621e74
YO
2900 }
2901 }
bc031a7d
KW
2902 } /* end loop through characters in this branch of the trie */
2903
2904 /* We take the min and max for this branch and combine to find the min
2905 * and max for all branches processed so far */
3dab1dad 2906 if( cur == first ) {
bc031a7d
KW
2907 trie->minlen = minchars;
2908 trie->maxlen = maxchars;
2909 } else if (minchars < trie->minlen) {
2910 trie->minlen = minchars;
2911 } else if (maxchars > trie->maxlen) {
2912 trie->maxlen = maxchars;
fab2782b 2913 }
a3621e74
YO
2914 } /* end first pass */
2915 DEBUG_TRIE_COMPILE_r(
6ad9a8ab 2916 Perl_re_indentf( aTHX_
cb41e5d6
YO
2917 "TRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2918 depth+1,
55eed653 2919 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
2920 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2921 (int)trie->minlen, (int)trie->maxlen )
a3621e74 2922 );
a3621e74
YO
2923
2924 /*
2925 We now know what we are dealing with in terms of unique chars and
2926 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
2927 representation using a flat table will take. If it's over a reasonable
2928 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
2929 conservative but potentially much slower representation using an array
2930 of lists.
2931
2932 At the end we convert both representations into the same compressed
2933 form that will be used in regexec.c for matching with. The latter
2934 is a form that cannot be used to construct with but has memory
2935 properties similar to the list form and access properties similar
2936 to the table form making it both suitable for fast searches and
2937 small enough that its feasable to store for the duration of a program.
2938
2939 See the comment in the code where the compressed table is produced
2940 inplace from the flat tabe representation for an explanation of how
2941 the compression works.
2942
2943 */
2944
2945
2e64971a
DM
2946 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2947 prev_states[1] = 0;
2948
538e84ed
KW
2949 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2950 > SvIV(re_trie_maxbuff) )
2951 {
a3621e74
YO
2952 /*
2953 Second Pass -- Array Of Lists Representation
2954
2955 Each state will be represented by a list of charid:state records
2956 (reg_trie_trans_le) the first such element holds the CUR and LEN
2957 points of the allocated array. (See defines above).
2958
2959 We build the initial structure using the lists, and then convert
2960 it into the compressed table form which allows faster lookups
2961 (but cant be modified once converted).
a3621e74
YO
2962 */
2963
a3621e74
YO
2964 STRLEN transcount = 1;
2965
6ad9a8ab 2966 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using list compiler\n",
cb41e5d6 2967 depth+1));
686b73d4 2968
c944940b
JH
2969 trie->states = (reg_trie_state *)
2970 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2971 sizeof(reg_trie_state) );
a3621e74
YO
2972 TRIE_LIST_NEW(1);
2973 next_alloc = 2;
2974
2975 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2976
df826430 2977 regnode *noper = NEXTOPER( cur );
c445ea15
AL
2978 U32 state = 1; /* required init */
2979 U16 charid = 0; /* sanity init */
07be1b83 2980 U32 wordlen = 0; /* required init */
c445ea15 2981
df826430
YO
2982 if (OP(noper) == NOTHING) {
2983 regnode *noper_next= regnext(noper);
944e05e3
YO
2984 if (noper_next < tail)
2985 noper= noper_next;
df826430
YO
2986 }
2987
f6b4b99d
KW
2988 if ( noper < tail
2989 && ( OP(noper) == flags
2990 || (flags == EXACT && OP(noper) == EXACT_ONLY8)
a9f8c7ac 2991 || (flags == EXACTFU && ( OP(noper) == EXACTFU_ONLY8
0ea669f4 2992 || OP(noper) == EXACTFUP))))
f6b4b99d 2993 {
944e05e3
YO
2994 const U8 *uc= (U8*)STRING(noper);
2995 const U8 *e= uc + STR_LEN(noper);
2996
786e8c11 2997 for ( ; uc < e ; uc += len ) {
c445ea15 2998
786e8c11 2999 TRIE_READ_CHAR;
c445ea15 3000
786e8c11
YO
3001 if ( uvc < 256 ) {
3002 charid = trie->charmap[ uvc ];
c445ea15 3003 } else {
538e84ed
KW
3004 SV** const svpp = hv_fetch( widecharmap,
3005 (char*)&uvc,
3006 sizeof( UV ),
3007 0);
786e8c11
YO
3008 if ( !svpp ) {
3009 charid = 0;
3010 } else {
3011 charid=(U16)SvIV( *svpp );
3012 }
c445ea15 3013 }
538e84ed
KW
3014 /* charid is now 0 if we dont know the char read, or
3015 * nonzero if we do */
786e8c11 3016 if ( charid ) {
a3621e74 3017
786e8c11
YO
3018 U16 check;
3019 U32 newstate = 0;
a3621e74 3020
786e8c11
YO
3021 charid--;
3022 if ( !trie->states[ state ].trans.list ) {
3023 TRIE_LIST_NEW( state );
c445ea15 3024 }
538e84ed
KW
3025 for ( check = 1;
3026 check <= TRIE_LIST_USED( state );
3027 check++ )
3028 {
3029 if ( TRIE_LIST_ITEM( state, check ).forid
3030 == charid )
3031 {
786e8c11
YO
3032 newstate = TRIE_LIST_ITEM( state, check ).newstate;
3033 break;
3034 }
3035 }
3036 if ( ! newstate ) {
3037 newstate = next_alloc++;
2e64971a 3038 prev_states[newstate] = state;
786e8c11
YO
3039 TRIE_LIST_PUSH( state, charid, newstate );
3040 transcount++;
3041 }
3042 state = newstate;
3043 } else {
147e3846 3044 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
c445ea15 3045 }
a28509cc 3046 }
c445ea15 3047 }
3dab1dad 3048 TRIE_HANDLE_WORD(state);
a3621e74
YO
3049
3050 } /* end second pass */
3051
1e2e3d02 3052 /* next alloc is the NEXT state to be allocated */
538e84ed 3053 trie->statecount = next_alloc;
c944940b
JH
3054 trie->states = (reg_trie_state *)
3055 PerlMemShared_realloc( trie->states,
3056 next_alloc
3057 * sizeof(reg_trie_state) );
a3621e74 3058
3dab1dad 3059 /* and now dump it out before we compress it */
2b8b4781
NC
3060 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
3061 revcharmap, next_alloc,
3062 depth+1)
1e2e3d02 3063 );
a3621e74 3064
c944940b
JH
3065 trie->trans = (reg_trie_trans *)
3066 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
3067 {
3068 U32 state;
a3621e74
YO
3069 U32 tp = 0;
3070 U32 zp = 0;
3071
3072
3073 for( state=1 ; state < next_alloc ; state ++ ) {
3074 U32 base=0;
3075
3076 /*
3077 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3078 Perl_re_printf( aTHX_ "tp: %d zp: %d ",tp,zp)
a3621e74
YO
3079 );
3080 */
3081
3082 if (trie->states[state].trans.list) {
3083 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
3084 U16 maxid=minid;
a28509cc 3085 U16 idx;
a3621e74
YO
3086
3087 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
3088 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
3089 if ( forid < minid ) {
3090 minid=forid;
3091 } else if ( forid > maxid ) {
3092 maxid=forid;
3093 }
a3621e74
YO
3094 }
3095 if ( transcount < tp + maxid - minid + 1) {
3096 transcount *= 2;
c944940b
JH
3097 trie->trans = (reg_trie_trans *)
3098 PerlMemShared_realloc( trie->trans,
446bd890
NC
3099 transcount
3100 * sizeof(reg_trie_trans) );
538e84ed
KW
3101 Zero( trie->trans + (transcount / 2),
3102 transcount / 2,
3103 reg_trie_trans );
a3621e74
YO
3104 }
3105 base = trie->uniquecharcount + tp - minid;
3106 if ( maxid == minid ) {
3107 U32 set = 0;
3108 for ( ; zp < tp ; zp++ ) {
3109 if ( ! trie->trans[ zp ].next ) {
3110 base = trie->uniquecharcount + zp - minid;
538e84ed
KW
3111 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
3112 1).newstate;
a3621e74
YO
3113 trie->trans[ zp ].check = state;
3114 set = 1;
3115 break;
3116 }
3117 }
3118 if ( !set ) {
538e84ed
KW
3119 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
3120 1).newstate;
a3621e74
YO
3121 trie->trans[ tp ].check = state;
3122 tp++;
3123 zp = tp;
3124 }
3125 } else {
3126 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
538e84ed
KW
3127 const U32 tid = base
3128 - trie->uniquecharcount
3129 + TRIE_LIST_ITEM( state, idx ).forid;
3130 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
3131 idx ).newstate;
a3621e74
YO
3132 trie->trans[ tid ].check = state;
3133 }
3134 tp += ( maxid - minid + 1 );
3135 }
3136 Safefree(trie->states[ state ].trans.list);
3137 }
3138 /*
3139 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3140 Perl_re_printf( aTHX_ " base: %d\n",base);
a3621e74
YO
3141 );
3142 */
3143 trie->states[ state ].trans.base=base;
3144 }
cc601c31 3145 trie->lasttrans = tp + 1;
a3621e74
YO
3146 }
3147 } else {
3148 /*
3149 Second Pass -- Flat Table Representation.
3150
b423522f
KW
3151 we dont use the 0 slot of either trans[] or states[] so we add 1 to
3152 each. We know that we will need Charcount+1 trans at most to store
3153 the data (one row per char at worst case) So we preallocate both
3154 structures assuming worst case.
a3621e74
YO
3155
3156 We then construct the trie using only the .next slots of the entry
3157 structs.
3158
b423522f
KW
3159 We use the .check field of the first entry of the node temporarily
3160 to make compression both faster and easier by keeping track of how
3161 many non zero fields are in the node.
a3621e74
YO
3162
3163 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
3164 transition.
3165
b423522f
KW
3166 There are two terms at use here: state as a TRIE_NODEIDX() which is
3167 a number representing the first entry of the node, and state as a
3168 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
3169 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
3170 if there are 2 entrys per node. eg:
a3621e74
YO
3171
3172 A B A B
3173 1. 2 4 1. 3 7
3174 2. 0 3 3. 0 5
3175 3. 0 0 5. 0 0
3176 4. 0 0 7. 0 0
3177
b423522f
KW
3178 The table is internally in the right hand, idx form. However as we
3179 also have to deal with the states array which is indexed by nodenum
3180 we have to use TRIE_NODENUM() to convert.
a3621e74
YO
3181
3182 */
6ad9a8ab 3183 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using table compiler\n",
cb41e5d6 3184 depth+1));
3dab1dad 3185
c944940b
JH
3186 trie->trans = (reg_trie_trans *)
3187 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
3188 * trie->uniquecharcount + 1,
3189 sizeof(reg_trie_trans) );
3190 trie->states = (reg_trie_state *)
3191 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
3192 sizeof(reg_trie_state) );
a3621e74
YO
3193 next_alloc = trie->uniquecharcount + 1;
3194
3dab1dad 3195
a3621e74
YO
3196 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
3197
df826430 3198 regnode *noper = NEXTOPER( cur );
a3621e74
YO
3199
3200 U32 state = 1; /* required init */
3201
3202 U16 charid = 0; /* sanity init */
3203 U32 accept_state = 0; /* sanity init */
a3621e74 3204
07be1b83 3205 U32 wordlen = 0; /* required init */
a3621e74 3206
df826430
YO
3207 if (OP(noper) == NOTHING) {
3208 regnode *noper_next= regnext(noper);
944e05e3
YO
3209 if (noper_next < tail)
3210 noper= noper_next;
df826430 3211 }
fab2782b 3212
f6b4b99d
KW
3213 if ( noper < tail
3214 && ( OP(noper) == flags
3215 || (flags == EXACT && OP(noper) == EXACT_ONLY8)
a9f8c7ac 3216 || (flags == EXACTFU && ( OP(noper) == EXACTFU_ONLY8
0ea669f4 3217 || OP(noper) == EXACTFUP))))
f6b4b99d 3218 {
944e05e3
YO
3219 const U8 *uc= (U8*)STRING(noper);
3220 const U8 *e= uc + STR_LEN(noper);
3221
786e8c11 3222 for ( ; uc < e ; uc += len ) {
a3621e74 3223
786e8c11 3224 TRIE_READ_CHAR;
a3621e74 3225
786e8c11
YO
3226 if ( uvc < 256 ) {
3227 charid = trie->charmap[ uvc ];
3228 } else {
538e84ed
KW
3229 SV* const * const svpp = hv_fetch( widecharmap,
3230 (char*)&uvc,
3231 sizeof( UV ),
3232 0);
786e8c11 3233 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 3234 }
786e8c11
YO
3235 if ( charid ) {
3236 charid--;
3237 if ( !trie->trans[ state + charid ].next ) {
3238 trie->trans[ state + charid ].next = next_alloc;
3239 trie->trans[ state ].check++;
2e64971a
DM
3240 prev_states[TRIE_NODENUM(next_alloc)]
3241 = TRIE_NODENUM(state);
786e8c11
YO
3242 next_alloc += trie->uniquecharcount;
3243 }
3244 state = trie->trans[ state + charid ].next;
3245 } else {
147e3846 3246 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
786e8c11 3247 }
538e84ed
KW
3248 /* charid is now 0 if we dont know the char read, or
3249 * nonzero if we do */
a3621e74 3250 }
a3621e74 3251 }
3dab1dad
YO
3252 accept_state = TRIE_NODENUM( state );
3253 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
3254
3255 } /* end second pass */
3256
3dab1dad 3257 /* and now dump it out before we compress it */
2b8b4781
NC
3258 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
3259 revcharmap,
3260 next_alloc, depth+1));
a3621e74 3261
a3621e74
YO
3262 {
3263 /*
3264 * Inplace compress the table.*
3265
3266 For sparse data sets the table constructed by the trie algorithm will
3267 be mostly 0/FAIL transitions or to put it another way mostly empty.
3268 (Note that leaf nodes will not contain any transitions.)
3269
3270 This algorithm compresses the tables by eliminating most such
3271 transitions, at the cost of a modest bit of extra work during lookup:
3272
3273 - Each states[] entry contains a .base field which indicates the
3274 index in the state[] array wheres its transition data is stored.
3275
3b753521 3276 - If .base is 0 there are no valid transitions from that node.
a3621e74
YO
3277
3278 - If .base is nonzero then charid is added to it to find an entry in
3279 the trans array.
3280
3281 -If trans[states[state].base+charid].check!=state then the
3282 transition is taken to be a 0/Fail transition. Thus if there are fail
3283 transitions at the front of the node then the .base offset will point
3284 somewhere inside the previous nodes data (or maybe even into a node
3285 even earlier), but the .check field determines if the transition is
3286 valid.
3287
786e8c11 3288 XXX - wrong maybe?
a3621e74 3289 The following process inplace converts the table to the compressed
3b753521 3290 table: We first do not compress the root node 1,and mark all its
a3621e74 3291 .check pointers as 1 and set its .base pointer as 1 as well. This
3b753521
FN
3292 allows us to do a DFA construction from the compressed table later,
3293 and ensures that any .base pointers we calculate later are greater
3294 than 0.
a3621e74
YO
3295
3296 - We set 'pos' to indicate the first entry of the second node.
3297
3298 - We then iterate over the columns of the node, finding the first and
3299 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
3300 and set the .check pointers accordingly, and advance pos
3301 appropriately and repreat for the next node. Note that when we copy
3302 the next pointers we have to convert them from the original
3303 NODEIDX form to NODENUM form as the former is not valid post
3304 compression.
3305
3306 - If a node has no transitions used we mark its base as 0 and do not
3307 advance the pos pointer.
3308
3309 - If a node only has one transition we use a second pointer into the
3310 structure to fill in allocated fail transitions from other states.
3311 This pointer is independent of the main pointer and scans forward
3312 looking for null transitions that are allocated to a state. When it
3313 finds one it writes the single transition into the "hole". If the
786e8c11 3314 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
3315
3316 - Once compressed we can Renew/realloc the structures to release the
3317 excess space.
3318
3319 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
3320 specifically Fig 3.47 and the associated pseudocode.
3321
3322 demq
3323 */
a3b680e6 3324 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 3325 U32 state, charid;
a3621e74 3326 U32 pos = 0, zp=0;
1e2e3d02 3327 trie->statecount = laststate;
a3621e74
YO
3328
3329 for ( state = 1 ; state < laststate ; state++ ) {
3330 U8 flag = 0;
a28509cc
AL
3331 const U32 stateidx = TRIE_NODEIDX( state );
3332 const U32 o_used = trie->trans[ stateidx ].check;
3333 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
3334 trie->trans[ stateidx ].check = 0;
3335
538e84ed
KW
3336 for ( charid = 0;
3337 used && charid < trie->uniquecharcount;
3338 charid++ )
3339 {
a3621e74
YO
3340 if ( flag || trie->trans[ stateidx + charid ].next ) {
3341 if ( trie->trans[ stateidx + charid ].next ) {
3342 if (o_used == 1) {
3343 for ( ; zp < pos ; zp++ ) {
3344 if ( ! trie->trans[ zp ].next ) {
3345 break;
3346 }
3347 }
538e84ed
KW
3348 trie->states[ state ].trans.base
3349 = zp
3350 + trie->uniquecharcount
3351 - charid ;
3352 trie->trans[ zp ].next
3353 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
3354 + charid ].next );
a3621e74
YO
3355 trie->trans[ zp ].check = state;
3356 if ( ++zp > pos ) pos = zp;
3357 break;
3358 }
3359 used--;
3360 }
3361 if ( !flag ) {
3362 flag = 1;
538e84ed
KW
3363 trie->states[ state ].trans.base
3364 = pos + trie->uniquecharcount - charid ;
a3621e74 3365 }
538e84ed
KW
3366 trie->trans[ pos ].next
3367 = SAFE_TRIE_NODENUM(
3368 trie->trans[ stateidx + charid ].next );
a3621e74
YO
3369 trie->trans[ pos ].check = state;
3370 pos++;
3371 }
3372 }
3373 }
cc601c31 3374 trie->lasttrans = pos + 1;
c944940b
JH
3375 trie->states = (reg_trie_state *)
3376 PerlMemShared_realloc( trie->states, laststate
3377 * sizeof(reg_trie_state) );
a3621e74 3378 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3379 Perl_re_indentf( aTHX_ "Alloc: %d Orig: %" IVdf " elements, Final:%" IVdf ". Savings of %%%5.2f\n",
cb41e5d6 3380 depth+1,
538e84ed
KW
3381 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
3382 + 1 ),
3383 (IV)next_alloc,
3384 (IV)pos,
3385 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
a3621e74
YO
3386 );
3387
3388 } /* end table compress */
3389 }
1e2e3d02 3390 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3391 Perl_re_indentf( aTHX_ "Statecount:%" UVxf " Lasttrans:%" UVxf "\n",
cb41e5d6 3392 depth+1,
1e2e3d02
YO
3393 (UV)trie->statecount,
3394 (UV)trie->lasttrans)
3395 );
cc601c31 3396 /* resize the trans array to remove unused space */
c944940b
JH
3397 trie->trans = (reg_trie_trans *)
3398 PerlMemShared_realloc( trie->trans, trie->lasttrans
3399 * sizeof(reg_trie_trans) );
a3621e74 3400
538e84ed 3401 { /* Modify the program and insert the new TRIE node */
3dab1dad
YO
3402 U8 nodetype =(U8)(flags & 0xFF);
3403 char *str=NULL;
538e84ed 3404
07be1b83 3405#ifdef DEBUGGING
e62cc96a 3406 regnode *optimize = NULL;
7122b237
YO
3407#ifdef RE_TRACK_PATTERN_OFFSETS
3408
b57a0404
JH
3409 U32 mjd_offset = 0;
3410 U32 mjd_nodelen = 0;
7122b237
YO
3411#endif /* RE_TRACK_PATTERN_OFFSETS */
3412#endif /* DEBUGGING */
a3621e74 3413 /*
3dab1dad
YO
3414 This means we convert either the first branch or the first Exact,
3415 depending on whether the thing following (in 'last') is a branch
3416 or not and whther first is the startbranch (ie is it a sub part of
3417 the alternation or is it the whole thing.)
3b753521 3418 Assuming its a sub part we convert the EXACT otherwise we convert
3dab1dad 3419 the whole branch sequence, including the first.
a3621e74 3420 */
3dab1dad 3421 /* Find the node we are going to overwrite */
7f69552c 3422 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 3423 /* branch sub-chain */
3dab1dad 3424 NEXT_OFF( first ) = (U16)(last - first);