This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
PATCH: [perl #133756] Failure to match properly
[perl5.git] / regcomp.c
CommitLineData
a0d0e21e
LW
1/* regcomp.c
2 */
3
4/*
4ac71550
TC
5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
6 *
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
a0d0e21e
LW
8 */
9
61296642
DM
10/* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
166f8a29 12 * a regular expression.
e4a054ea
DM
13 *
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
166f8a29
DM
18 */
19
a687059c
LW
20/* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
22 */
23
24/* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
27 */
28
e50aee73 29/* The names of the functions have been changed from regcomp and
3b753521 30 * regexec to pregcomp and pregexec in order to avoid conflicts
e50aee73
AD
31 * with the POSIX routines of the same names.
32*/
33
b9d5759e 34#ifdef PERL_EXT_RE_BUILD
54df2634 35#include "re_top.h"
b81d288d 36#endif
56953603 37
a687059c 38/*
e50aee73 39 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
40 *
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
43 *
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
47 *
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
50 * from defects in it.
51 *
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
54 *
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
57 *
58 *
59 **** Alterations to Henry's code are...
60 ****
4bb101f2 61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
1129b882
NC
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
a687059c 64 ****
9ef589d8
LW
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
67
a687059c
LW
68 *
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
72 */
73#include "EXTERN.h"
864dbfa3 74#define PERL_IN_REGCOMP_C
a687059c 75#include "perl.h"
d06ea78c 76
acfe0abc 77#ifndef PERL_IN_XSUB_RE
d06ea78c
GS
78# include "INTERN.h"
79#endif
c277df42
IZ
80
81#define REG_COMP_C
54df2634
NC
82#ifdef PERL_IN_XSUB_RE
83# include "re_comp.h"
afda64e8 84EXTERN_C const struct regexp_engine my_reg_engine;
54df2634
NC
85#else
86# include "regcomp.h"
87#endif
a687059c 88
f7e03a10 89#include "dquote_inline.h"
b992490d 90#include "invlist_inline.h"
1b0f46bf 91#include "unicode_constants.h"
04e98a4d 92
538e84ed
KW
93#define HAS_NONLATIN1_FOLD_CLOSURE(i) \
94 _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
f12c0118
KW
95#define HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(i) \
96 _HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
26faadbd 97#define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
2c61f163 98#define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
94dc5c2d 99
a687059c
LW
100#ifndef STATIC
101#define STATIC static
102#endif
103
3f910e62
YO
104/* this is a chain of data about sub patterns we are processing that
105 need to be handled separately/specially in study_chunk. Its so
106 we can simulate recursion without losing state. */
107struct scan_frame;
108typedef struct scan_frame {
109 regnode *last_regnode; /* last node to process in this frame */
110 regnode *next_regnode; /* next node to process when last is reached */
111 U32 prev_recursed_depth;
112 I32 stopparen; /* what stopparen do we use */
3f910e62
YO
113
114 struct scan_frame *this_prev_frame; /* this previous frame */
115 struct scan_frame *prev_frame; /* previous frame */
116 struct scan_frame *next_frame; /* next frame */
117} scan_frame;
118
f4ae5a27
KW
119/* Certain characters are output as a sequence with the first being a
120 * backslash. */
eb7486de 121#define isBACKSLASHED_PUNCT(c) strchr("-[]\\^", c)
f4ae5a27
KW
122
123
09b2b2e6 124struct RExC_state_t {
514a91f1
DM
125 U32 flags; /* RXf_* are we folding, multilining? */
126 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
830247a4 127 char *precomp; /* uncompiled string. */
711b303b 128 char *precomp_end; /* pointer to end of uncompiled string. */
288b8c02 129 REGEXP *rx_sv; /* The SV that is the regexp. */
f8fc2ecf 130 regexp *rx; /* perl core regexp structure */
538e84ed
KW
131 regexp_internal *rxi; /* internal data for regexp object
132 pprivate field */
fac92740 133 char *start; /* Start of input for compile */
830247a4
IZ
134 char *end; /* End of input for compile */
135 char *parse; /* Input-scan pointer. */
51684e3c
KW
136 char *copy_start; /* start of copy of input within
137 constructed parse string */
138 char *copy_start_in_input; /* Position in input string
139 corresponding to copy_start */
ea3daa5d 140 SSize_t whilem_seen; /* number of WHILEM in this expr */
fac92740 141 regnode *emit_start; /* Start of emitted-code area */
f55b7b26 142 regnode_offset emit; /* Code-emit pointer */
830247a4
IZ
143 I32 naughty; /* How bad is this pattern? */
144 I32 sawback; /* Did we see \1, ...? */
145 U32 seen;
88f063b4
KW
146 SSize_t size; /* Number of regnode equivalents in
147 pattern */
7c932d07
KW
148
149 /* position beyond 'precomp' of the warning message furthest away from
150 * 'precomp'. During the parse, no warnings are raised for any problems
151 * earlier in the parse than this position. This works if warnings are
152 * raised the first time a given spot is parsed, and if only one
153 * independent warning is raised for any given spot */
154 Size_t latest_warn_offset;
155
156 I32 npar; /* Capture buffer count so far in the
157 parse, (OPEN) plus one. ("par" 0 is
158 the whole pattern)*/
159 I32 total_par; /* During initial parse, is either 0,
160 or -1; the latter indicating a
161 reparse is needed. After that pass,
162 it is what 'npar' became after the
163 pass. Hence, it being > 0 indicates
164 we are in a reparse situation */
538e84ed
KW
165 I32 nestroot; /* root parens we are in - used by
166 accept */
830247a4 167 I32 seen_zerolen;
f55b7b26
KW
168 regnode_offset *open_parens; /* offsets to open parens */
169 regnode_offset *close_parens; /* offsets to close parens */
d5a00e4a 170 regnode *end_op; /* END node in program */
02daf0ab
YO
171 I32 utf8; /* whether the pattern is utf8 or not */
172 I32 orig_utf8; /* whether the pattern was originally in utf8 */
173 /* XXX use this for future optimisation of case
174 * where pattern must be upgraded to utf8. */
e40e74fe
KW
175 I32 uni_semantics; /* If a d charset modifier should use unicode
176 rules, even if the pattern is not in
177 utf8 */
81714fb9 178 HV *paren_names; /* Paren names */
538e84ed 179
40d049e4 180 regnode **recurse; /* Recurse regops */
88f063b4 181 I32 recurse_count; /* Number of recurse regops we have generated */
4286711a 182 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
538e84ed 183 through */
09a65838 184 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
b57e4118 185 I32 in_lookbehind;
4624b182 186 I32 contains_locale;
bb3f3ed2 187 I32 override_recoding;
b6d67071
KW
188#ifdef EBCDIC
189 I32 recode_x_to_native;
190#endif
9d53c457 191 I32 in_multi_char_class;
1acab4c5 192 struct reg_code_blocks *code_blocks;/* positions of literal (?{})
68e2671b 193 within pattern */
b1603ef8 194 int code_index; /* next code_blocks[] slot */
ee273784 195 SSize_t maxlen; /* mininum possible number of chars in string to match */
3f910e62
YO
196 scan_frame *frame_head;
197 scan_frame *frame_last;
198 U32 frame_count;
7eec73eb 199 AV *warn_text;
dc6d7f5c 200#ifdef ADD_TO_REGEXEC
830247a4
IZ
201 char *starttry; /* -Dr: where regtry was called. */
202#define RExC_starttry (pRExC_state->starttry)
203#endif
d24ca0c5 204 SV *runtime_code_qr; /* qr with the runtime code blocks */
3dab1dad 205#ifdef DEBUGGING
be8e71aa 206 const char *lastparse;
3dab1dad 207 I32 lastnum;
1f1031fe 208 AV *paren_name_list; /* idx -> name */
d9a72fcc 209 U32 study_chunk_recursed_count;
c9f0d54c
YO
210 SV *mysv1;
211 SV *mysv2;
88f063b4 212
3dab1dad
YO
213#define RExC_lastparse (pRExC_state->lastparse)
214#define RExC_lastnum (pRExC_state->lastnum)
1f1031fe 215#define RExC_paren_name_list (pRExC_state->paren_name_list)
d9a72fcc 216#define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
c9f0d54c
YO
217#define RExC_mysv (pRExC_state->mysv1)
218#define RExC_mysv1 (pRExC_state->mysv1)
219#define RExC_mysv2 (pRExC_state->mysv2)
220
3dab1dad 221#endif
bd0b62a6 222 bool seen_d_op;
911bd04e 223 bool strict;
da7cf1cc 224 bool study_started;
034602eb 225 bool in_script_run;
7c932d07 226 bool use_BRANCHJ;
09b2b2e6 227};
830247a4 228
e2509266 229#define RExC_flags (pRExC_state->flags)
514a91f1 230#define RExC_pm_flags (pRExC_state->pm_flags)
830247a4 231#define RExC_precomp (pRExC_state->precomp)
51684e3c
KW
232#define RExC_copy_start_in_input (pRExC_state->copy_start_in_input)
233#define RExC_copy_start_in_constructed (pRExC_state->copy_start)
711b303b 234#define RExC_precomp_end (pRExC_state->precomp_end)
288b8c02 235#define RExC_rx_sv (pRExC_state->rx_sv)
830247a4 236#define RExC_rx (pRExC_state->rx)
f8fc2ecf 237#define RExC_rxi (pRExC_state->rxi)
fac92740 238#define RExC_start (pRExC_state->start)
830247a4
IZ
239#define RExC_end (pRExC_state->end)
240#define RExC_parse (pRExC_state->parse)
7c932d07 241#define RExC_latest_warn_offset (pRExC_state->latest_warn_offset )
830247a4 242#define RExC_whilem_seen (pRExC_state->whilem_seen)
bd0b62a6
KW
243#define RExC_seen_d_op (pRExC_state->seen_d_op) /* Seen something that differs
244 under /d from /u ? */
512c0f5a 245
512c0f5a 246
7122b237 247#ifdef RE_TRACK_PATTERN_OFFSETS
44f36cd3 248# define RExC_offsets (RExC_rxi->u.offsets) /* I am not like the
538e84ed 249 others */
7122b237 250#endif
830247a4 251#define RExC_emit (pRExC_state->emit)
fac92740 252#define RExC_emit_start (pRExC_state->emit_start)
830247a4
IZ
253#define RExC_sawback (pRExC_state->sawback)
254#define RExC_seen (pRExC_state->seen)
255#define RExC_size (pRExC_state->size)
ee273784 256#define RExC_maxlen (pRExC_state->maxlen)
830247a4 257#define RExC_npar (pRExC_state->npar)
1533d1d2 258#define RExC_total_parens (pRExC_state->total_par)
e2e6a0f1 259#define RExC_nestroot (pRExC_state->nestroot)
830247a4 260#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
1aa99e6b 261#define RExC_utf8 (pRExC_state->utf8)
e40e74fe 262#define RExC_uni_semantics (pRExC_state->uni_semantics)
02daf0ab 263#define RExC_orig_utf8 (pRExC_state->orig_utf8)
40d049e4
YO
264#define RExC_open_parens (pRExC_state->open_parens)
265#define RExC_close_parens (pRExC_state->close_parens)
5bd2d46e 266#define RExC_end_op (pRExC_state->end_op)
81714fb9 267#define RExC_paren_names (pRExC_state->paren_names)
40d049e4
YO
268#define RExC_recurse (pRExC_state->recurse)
269#define RExC_recurse_count (pRExC_state->recurse_count)
09a65838 270#define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
538e84ed
KW
271#define RExC_study_chunk_recursed_bytes \
272 (pRExC_state->study_chunk_recursed_bytes)
b57e4118 273#define RExC_in_lookbehind (pRExC_state->in_lookbehind)
4624b182 274#define RExC_contains_locale (pRExC_state->contains_locale)
b6d67071
KW
275#ifdef EBCDIC
276# define RExC_recode_x_to_native (pRExC_state->recode_x_to_native)
277#endif
9d53c457 278#define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
3f910e62
YO
279#define RExC_frame_head (pRExC_state->frame_head)
280#define RExC_frame_last (pRExC_state->frame_last)
281#define RExC_frame_count (pRExC_state->frame_count)
67cdf558 282#define RExC_strict (pRExC_state->strict)
da7cf1cc 283#define RExC_study_started (pRExC_state->study_started)
7eec73eb 284#define RExC_warn_text (pRExC_state->warn_text)
034602eb 285#define RExC_in_script_run (pRExC_state->in_script_run)
7c932d07 286#define RExC_use_BRANCHJ (pRExC_state->use_BRANCHJ)
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; \
bd0b62a6
KW
358 if (RExC_seen_d_op && LIKELY(RExC_total_parens >= 0)) { \
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 BRANCH_MAX_OFFSET U16_MAX
370#define REQUIRE_BRANCHJ(flagp, restart_retval) \
371 STMT_START { \
372 RExC_use_BRANCHJ = 1; \
523a2924
KW
373 if (LIKELY(RExC_total_parens >= 0)) { \
374 /* No need to restart the parse immediately if we're \
375 * going to reparse anyway to count parens */ \
376 *flagp |= RESTART_PARSE; \
377 return restart_retval; \
378 } \
7c932d07
KW
379 } STMT_END
380
381#define REQUIRE_PARENS_PASS \
382 STMT_START { \
383 if (RExC_total_parens == 0) RExC_total_parens = -1; \
384 } STMT_END
385
f358228e
KW
386/* This is used to return failure (zero) early from the calling function if
387 * various flags in 'flags' are set. Two flags always cause a return:
388 * 'RESTART_PARSE' and 'NEED_UTF8'. 'extra' can be used to specify any
389 * additional flags that should cause a return; 0 if none. If the return will
390 * be done, '*flagp' is first set to be all of the flags that caused the
391 * return. */
392#define RETURN_FAIL_ON_RESTART_OR_FLAGS(flags,flagp,extra) \
fd7fc6ae 393 STMT_START { \
c36361b2
KW
394 if ((flags) & (RESTART_PARSE|NEED_UTF8|(extra))) { \
395 *(flagp) = (flags) & (RESTART_PARSE|NEED_UTF8|(extra)); \
f358228e 396 return 0; \
fd7fc6ae
YO
397 } \
398 } STMT_END
399
c36361b2 400#define MUST_RESTART(flags) ((flags) & (RESTART_PARSE))
fd7fc6ae 401
5d555bb8 402#define RETURN_FAIL_ON_RESTART(flags,flagp) \
f358228e 403 RETURN_FAIL_ON_RESTART_OR_FLAGS( flags, flagp, 0)
5d555bb8 404#define RETURN_FAIL_ON_RESTART_FLAGP(flagp) \
f358228e 405 if (MUST_RESTART(*(flagp))) return 0
fd7fc6ae 406
f19b1a63
KW
407/* This converts the named class defined in regcomp.h to its equivalent class
408 * number defined in handy.h. */
409#define namedclass_to_classnum(class) ((int) ((class) / 2))
410#define classnum_to_namedclass(classnum) ((classnum) * 2)
411
de92f5e6
KW
412#define _invlist_union_complement_2nd(a, b, output) \
413 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
414#define _invlist_intersection_complement_2nd(a, b, output) \
415 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
416
1de06328
YO
417/* About scan_data_t.
418
419 During optimisation we recurse through the regexp program performing
420 various inplace (keyhole style) optimisations. In addition study_chunk
421 and scan_commit populate this data structure with information about
538e84ed 422 what strings MUST appear in the pattern. We look for the longest
3b753521 423 string that must appear at a fixed location, and we look for the
1de06328
YO
424 longest string that may appear at a floating location. So for instance
425 in the pattern:
538e84ed 426
1de06328 427 /FOO[xX]A.*B[xX]BAR/
538e84ed 428
1de06328
YO
429 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
430 strings (because they follow a .* construct). study_chunk will identify
431 both FOO and BAR as being the longest fixed and floating strings respectively.
538e84ed 432
1de06328 433 The strings can be composites, for instance
538e84ed 434
1de06328 435 /(f)(o)(o)/
538e84ed 436
1de06328 437 will result in a composite fixed substring 'foo'.
538e84ed 438
1de06328 439 For each string some basic information is maintained:
538e84ed 440
6ea77169 441 - min_offset
1de06328
YO
442 This is the position the string must appear at, or not before.
443 It also implicitly (when combined with minlenp) tells us how many
3b753521
FN
444 characters must match before the string we are searching for.
445 Likewise when combined with minlenp and the length of the string it
538e84ed 446 tells us how many characters must appear after the string we have
1de06328 447 found.
538e84ed 448
1de06328
YO
449 - max_offset
450 Only used for floating strings. This is the rightmost point that
ea3daa5d 451 the string can appear at. If set to SSize_t_MAX it indicates that the
1de06328 452 string can occur infinitely far to the right.
6e12f832 453 For fixed strings, it is equal to min_offset.
538e84ed 454
1de06328 455 - minlenp
2d608413
KW
456 A pointer to the minimum number of characters of the pattern that the
457 string was found inside. This is important as in the case of positive
538e84ed 458 lookahead or positive lookbehind we can have multiple patterns
1de06328 459 involved. Consider
538e84ed 460
1de06328 461 /(?=FOO).*F/
538e84ed 462
1de06328
YO
463 The minimum length of the pattern overall is 3, the minimum length
464 of the lookahead part is 3, but the minimum length of the part that
538e84ed 465 will actually match is 1. So 'FOO's minimum length is 3, but the
1de06328 466 minimum length for the F is 1. This is important as the minimum length
538e84ed 467 is used to determine offsets in front of and behind the string being
1de06328 468 looked for. Since strings can be composites this is the length of the
486ec47a 469 pattern at the time it was committed with a scan_commit. Note that
1de06328 470 the length is calculated by study_chunk, so that the minimum lengths
538e84ed 471 are not known until the full pattern has been compiled, thus the
1de06328 472 pointer to the value.
538e84ed 473
1de06328 474 - lookbehind
538e84ed 475
1de06328 476 In the case of lookbehind the string being searched for can be
538e84ed 477 offset past the start point of the final matching string.
1de06328
YO
478 If this value was just blithely removed from the min_offset it would
479 invalidate some of the calculations for how many chars must match
480 before or after (as they are derived from min_offset and minlen and
538e84ed 481 the length of the string being searched for).
1de06328
YO
482 When the final pattern is compiled and the data is moved from the
483 scan_data_t structure into the regexp structure the information
538e84ed
KW
484 about lookbehind is factored in, with the information that would
485 have been lost precalculated in the end_shift field for the
1de06328
YO
486 associated string.
487
488 The fields pos_min and pos_delta are used to store the minimum offset
538e84ed 489 and the delta to the maximum offset at the current point in the pattern.
1de06328
YO
490
491*/
2c2d71f5 492
cb6aef8b
DM
493struct scan_data_substrs {
494 SV *str; /* longest substring found in pattern */
495 SSize_t min_offset; /* earliest point in string it can appear */
496 SSize_t max_offset; /* latest point in string it can appear */
497 SSize_t *minlenp; /* pointer to the minlen relevant to the string */
498 SSize_t lookbehind; /* is the pos of the string modified by LB */
11683ecb 499 I32 flags; /* per substring SF_* and SCF_* flags */
cb6aef8b
DM
500};
501
2c2d71f5 502typedef struct scan_data_t {
1de06328
YO
503 /*I32 len_min; unused */
504 /*I32 len_delta; unused */
49f55535 505 SSize_t pos_min;
ea3daa5d 506 SSize_t pos_delta;
2c2d71f5 507 SV *last_found;
ea3daa5d 508 SSize_t last_end; /* min value, <0 unless valid. */
49f55535 509 SSize_t last_start_min;
ea3daa5d 510 SSize_t last_start_max;
2099df82
DM
511 U8 cur_is_floating; /* whether the last_* values should be set as
512 * the next fixed (0) or floating (1)
513 * substring */
6ea77169 514
2099df82
DM
515 /* [0] is longest fixed substring so far, [1] is longest float so far */
516 struct scan_data_substrs substrs[2];
6ea77169 517
11683ecb 518 I32 flags; /* common SF_* and SCF_* flags */
2c2d71f5 519 I32 whilem_c;
49f55535 520 SSize_t *last_closep;
b8f7bb16 521 regnode_ssc *start_class;
2c2d71f5
JH
522} scan_data_t;
523
a687059c 524/*
e50aee73 525 * Forward declarations for pregcomp()'s friends.
a687059c 526 */
a0d0e21e 527
6ea77169 528static const scan_data_t zero_scan_data = {
a2ca2017 529 0, 0, NULL, 0, 0, 0, 0,
6ea77169 530 {
11683ecb
DM
531 { NULL, 0, 0, 0, 0, 0 },
532 { NULL, 0, 0, 0, 0, 0 },
6ea77169
DM
533 },
534 0, 0, NULL, NULL
535};
c277df42 536
11683ecb
DM
537/* study flags */
538
07be1b83
YO
539#define SF_BEFORE_SEOL 0x0001
540#define SF_BEFORE_MEOL 0x0002
11683ecb 541#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
c277df42 542
07be1b83
YO
543#define SF_IS_INF 0x0040
544#define SF_HAS_PAR 0x0080
545#define SF_IN_PAR 0x0100
546#define SF_HAS_EVAL 0x0200
bc604ad8
KW
547
548
549/* SCF_DO_SUBSTR is the flag that tells the regexp analyzer to track the
550 * longest substring in the pattern. When it is not set the optimiser keeps
551 * track of position, but does not keep track of the actual strings seen,
552 *
553 * So for instance /foo/ will be parsed with SCF_DO_SUBSTR being true, but
554 * /foo/i will not.
555 *
556 * Similarly, /foo.*(blah|erm|huh).*fnorble/ will have "foo" and "fnorble"
557 * parsed with SCF_DO_SUBSTR on, but while processing the (...) it will be
558 * turned off because of the alternation (BRANCH). */
07be1b83 559#define SCF_DO_SUBSTR 0x0400
bc604ad8 560
653099ff
GS
561#define SCF_DO_STCLASS_AND 0x0800
562#define SCF_DO_STCLASS_OR 0x1000
563#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
e1901655 564#define SCF_WHILEM_VISITED_POS 0x2000
c277df42 565
786e8c11 566#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
538e84ed 567#define SCF_SEEN_ACCEPT 0x8000
688e0391 568#define SCF_TRIE_DOING_RESTUDY 0x10000
4286711a
YO
569#define SCF_IN_DEFINE 0x20000
570
571
572
07be1b83 573
43fead97 574#define UTF cBOOL(RExC_utf8)
00b27cfc
KW
575
576/* The enums for all these are ordered so things work out correctly */
a62b1201 577#define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
538e84ed
KW
578#define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
579 == REGEX_DEPENDS_CHARSET)
00b27cfc 580#define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
538e84ed
KW
581#define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
582 >= REGEX_UNICODE_CHARSET)
583#define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
584 == REGEX_ASCII_RESTRICTED_CHARSET)
585#define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
586 >= REGEX_ASCII_RESTRICTED_CHARSET)
587#define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
588 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
a62b1201 589
43fead97 590#define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
a0ed51b3 591
f2c2a6ab
KW
592/* For programs that want to be strictly Unicode compatible by dying if any
593 * attempt is made to match a non-Unicode code point against a Unicode
594 * property. */
595#define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
596
93733859 597#define OOB_NAMEDCLASS -1
b8c5462f 598
8e661ac5
KW
599/* There is no code point that is out-of-bounds, so this is problematic. But
600 * its only current use is to initialize a variable that is always set before
601 * looked at. */
602#define OOB_UNICODE 0xDEADBEEF
603
a0ed51b3 604#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
a0ed51b3 605
8615cb43 606
b45f050a
JF
607/* length of regex to show in messages that don't mark a position within */
608#define RegexLengthToShowInErrorMessages 127
609
610/*
611 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
612 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
613 * op/pragma/warn/regcomp.
614 */
7253e4e3
RK
615#define MARKER1 "<-- HERE" /* marker as it appears in the description */
616#define MARKER2 " <-- HERE " /* marker as it appears within the regex */
b81d288d 617
538e84ed 618#define REPORT_LOCATION " in regex; marked by " MARKER1 \
147e3846 619 " in m/%" UTF8f MARKER2 "%" UTF8f "/"
b45f050a 620
285b5ca0
KW
621/* The code in this file in places uses one level of recursion with parsing
622 * rebased to an alternate string constructed by us in memory. This can take
623 * the form of something that is completely different from the input, or
624 * something that uses the input as part of the alternate. In the first case,
625 * there should be no possibility of an error, as we are in complete control of
51684e3c
KW
626 * the alternate string. But in the second case we don't completely control
627 * the input portion, so there may be errors in that. Here's an example:
285b5ca0
KW
628 * /[abc\x{DF}def]/ui
629 * is handled specially because \x{df} folds to a sequence of more than one
51684e3c 630 * character: 'ss'. What is done is to create and parse an alternate string,
285b5ca0
KW
631 * which looks like this:
632 * /(?:\x{DF}|[abc\x{DF}def])/ui
633 * where it uses the input unchanged in the middle of something it constructs,
634 * which is a branch for the DF outside the character class, and clustering
635 * parens around the whole thing. (It knows enough to skip the DF inside the
636 * class while in this substitute parse.) 'abc' and 'def' may have errors that
637 * need to be reported. The general situation looks like this:
638 *
51684e3c 639 * |<------- identical ------>|
285b5ca0 640 * sI tI xI eI
51684e3c 641 * Input: ---------------------------------------------------------------
285b5ca0
KW
642 * Constructed: ---------------------------------------------------
643 * sC tC xC eC EC
51684e3c 644 * |<------- identical ------>|
285b5ca0 645 *
51684e3c
KW
646 * sI..eI is the portion of the input pattern we are concerned with here.
647 * sC..EC is the constructed substitute parse string.
648 * sC..tC is constructed by us
649 * tC..eC is an exact duplicate of the portion of the input pattern tI..eI.
650 * In the diagram, these are vertically aligned.
651 * eC..EC is also constructed by us.
652 * xC is the position in the substitute parse string where we found a
653 * problem.
654 * xI is the position in the original pattern corresponding to xC.
285b5ca0 655 *
51684e3c
KW
656 * We want to display a message showing the real input string. Thus we need to
657 * translate from xC to xI. We know that xC >= tC, since the portion of the
658 * string sC..tC has been constructed by us, and so shouldn't have errors. We
659 * get:
660 * xI = tI + (xC - tC)
285b5ca0 661 *
51684e3c
KW
662 * When the substitute parse is constructed, the code needs to set:
663 * RExC_start (sC)
664 * RExC_end (eC)
665 * RExC_copy_start_in_input (tI)
666 * RExC_copy_start_in_constructed (tC)
667 * and restore them when done.
285b5ca0 668 *
51684e3c
KW
669 * During normal processing of the input pattern, both
670 * 'RExC_copy_start_in_input' and 'RExC_copy_start_in_constructed' are set to
671 * sI, so that xC equals xI.
285b5ca0
KW
672 */
673
51684e3c
KW
674#define sI RExC_precomp
675#define eI RExC_precomp_end
676#define sC RExC_start
677#define eC RExC_end
678#define tI RExC_copy_start_in_input
679#define tC RExC_copy_start_in_constructed
680#define xI(xC) (tI + (xC - tC))
681#define xI_offset(xC) (xI(xC) - sI)
285b5ca0
KW
682
683#define REPORT_LOCATION_ARGS(xC) \
684 UTF8fARG(UTF, \
51684e3c 685 (xI(xC) > eI) /* Don't run off end */ \
232b691f 686 ? eI - sI /* Length before the <--HERE */ \
bc929b61
KW
687 : ((xI_offset(xC) >= 0) \
688 ? xI_offset(xC) \
689 : (Perl_croak(aTHX_ "panic: %s: %d: negative offset: %" \
690 IVdf " trying to output message for " \
691 " pattern %.*s", \
51684e3c 692 __FILE__, __LINE__, (IV) xI_offset(xC), \
bc929b61 693 ((int) (eC - sC)), sC), 0)), \
51684e3c 694 sI), /* The input pattern printed up to the <--HERE */ \
285b5ca0 695 UTF8fARG(UTF, \
51684e3c
KW
696 (xI(xC) > eI) ? 0 : eI - xI(xC), /* Length after <--HERE */ \
697 (xI(xC) > eI) ? eI : xI(xC)) /* pattern after <--HERE */
c1d900c3 698
8a6d8ec6
HS
699/* Used to point after bad bytes for an error message, but avoid skipping
700 * past a nul byte. */
701#define SKIP_IF_CHAR(s) (!*(s) ? 0 : UTF ? UTF8SKIP(s) : 1)
702
ed68fe11
KW
703/* Set up to clean up after our imminent demise */
704#define PREPARE_TO_DIE \
705 STMT_START { \
706 if (RExC_rx_sv) \
707 SAVEFREESV(RExC_rx_sv); \
7c932d07
KW
708 if (RExC_open_parens) \
709 SAVEFREEPV(RExC_open_parens); \
710 if (RExC_close_parens) \
711 SAVEFREEPV(RExC_close_parens); \
ed68fe11
KW
712 } STMT_END
713
b45f050a
JF
714/*
715 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
716 * arg. Show regex, up to a maximum length. If it's too long, chop and add
717 * "...".
718 */
58e23c8d 719#define _FAIL(code) STMT_START { \
bfed75c6 720 const char *ellipses = ""; \
88f063b4 721 IV len = RExC_precomp_end - RExC_precomp; \
ccb2c380 722 \
ed68fe11 723 PREPARE_TO_DIE; \
ccb2c380
MP
724 if (len > RegexLengthToShowInErrorMessages) { \
725 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
726 len = RegexLengthToShowInErrorMessages - 10; \
727 ellipses = "..."; \
728 } \
58e23c8d 729 code; \
ccb2c380 730} STMT_END
8615cb43 731
58e23c8d 732#define FAIL(msg) _FAIL( \
147e3846 733 Perl_croak(aTHX_ "%s in regex m/%" UTF8f "%s/", \
c1d900c3 734 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d
YO
735
736#define FAIL2(msg,arg) _FAIL( \
147e3846 737 Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \
c1d900c3 738 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d 739
b45f050a 740/*
b45f050a
JF
741 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
742 */
ccb2c380 743#define Simple_vFAIL(m) STMT_START { \
ccb2c380 744 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
d528642a 745 m, REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 746} STMT_END
b45f050a
JF
747
748/*
749 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
750 */
ccb2c380 751#define vFAIL(m) STMT_START { \
ed68fe11 752 PREPARE_TO_DIE; \
ccb2c380
MP
753 Simple_vFAIL(m); \
754} STMT_END
b45f050a
JF
755
756/*
757 * Like Simple_vFAIL(), but accepts two arguments.
758 */
ccb2c380 759#define Simple_vFAIL2(m,a1) STMT_START { \
d528642a
KW
760 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
761 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 762} STMT_END
b45f050a
JF
763
764/*
765 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
766 */
ccb2c380 767#define vFAIL2(m,a1) STMT_START { \
ed68fe11 768 PREPARE_TO_DIE; \
ccb2c380
MP
769 Simple_vFAIL2(m, a1); \
770} STMT_END
b45f050a
JF
771
772
773/*
774 * Like Simple_vFAIL(), but accepts three arguments.
775 */
ccb2c380 776#define Simple_vFAIL3(m, a1, a2) STMT_START { \
c1d900c3 777 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 778 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 779} STMT_END
b45f050a
JF
780
781/*
782 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
783 */
ccb2c380 784#define vFAIL3(m,a1,a2) STMT_START { \
ed68fe11 785 PREPARE_TO_DIE; \
ccb2c380
MP
786 Simple_vFAIL3(m, a1, a2); \
787} STMT_END
b45f050a
JF
788
789/*
790 * Like Simple_vFAIL(), but accepts four arguments.
791 */
ccb2c380 792#define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
d528642a
KW
793 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
794 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 795} STMT_END
b45f050a 796
95db3ffa 797#define vFAIL4(m,a1,a2,a3) STMT_START { \
ed68fe11 798 PREPARE_TO_DIE; \
95db3ffa
KW
799 Simple_vFAIL4(m, a1, a2, a3); \
800} STMT_END
801
946095af 802/* A specialized version of vFAIL2 that works with UTF8f */
d528642a 803#define vFAIL2utf8f(m, a1) STMT_START { \
ed68fe11 804 PREPARE_TO_DIE; \
d528642a
KW
805 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
806 REPORT_LOCATION_ARGS(RExC_parse)); \
946095af
BF
807} STMT_END
808
3ba22297 809#define vFAIL3utf8f(m, a1, a2) STMT_START { \
ed68fe11 810 PREPARE_TO_DIE; \
3ba22297 811 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 812 REPORT_LOCATION_ARGS(RExC_parse)); \
3ba22297
KW
813} STMT_END
814
d62e749e
KW
815/* Setting this to NULL is a signal to not output warnings */
816#define TURN_OFF_WARNINGS_IN_SUBSTITUTE_PARSE RExC_copy_start_in_constructed = NULL
817#define RESTORE_WARNINGS RExC_copy_start_in_constructed = RExC_precomp
818
7c932d07
KW
819/* Since a warning can be generated multiple times as the input is reparsed, we
820 * output it the first time we come to that point in the parse, but suppress it
821 * otherwise. 'RExC_copy_start_in_constructed' being NULL is a flag to not
822 * generate any warnings */
55bcf857 823#define TO_OUTPUT_WARNINGS(loc) \
7c932d07
KW
824 ( RExC_copy_start_in_constructed \
825 && ((xI(loc)) - RExC_precomp) > (Ptrdiff_t) RExC_latest_warn_offset)
55bcf857 826
7c932d07
KW
827/* After we've emitted a warning, we save the position in the input so we don't
828 * output it again */
829#define UPDATE_WARNINGS_LOC(loc) \
830 STMT_START { \
831 if (TO_OUTPUT_WARNINGS(loc)) { \
832 RExC_latest_warn_offset = (xI(loc)) - RExC_precomp; \
833 } \
834 } STMT_END
77ccba3c 835
086f21bf 836/* 'warns' is the output of the packWARNx macro used in 'code' */
f8220fe2
KW
837#define _WARN_HELPER(loc, warns, code) \
838 STMT_START { \
d62e749e
KW
839 if (! RExC_copy_start_in_constructed) { \
840 Perl_croak( aTHX_ "panic! %s: %d: Tried to warn when none" \
841 " expected at '%s'", \
842 __FILE__, __LINE__, loc); \
843 } \
55bcf857 844 if (TO_OUTPUT_WARNINGS(loc)) { \
086f21bf
KW
845 if (ckDEAD(warns)) \
846 PREPARE_TO_DIE; \
338190a1 847 code; \
77ccba3c 848 UPDATE_WARNINGS_LOC(loc); \
338190a1 849 } \
f8220fe2 850 } STMT_END
946095af 851
5e0a247b 852/* m is not necessarily a "literal string", in this macro */
f8220fe2
KW
853#define reg_warn_non_literal_string(loc, m) \
854 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
855 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 856 "%s" REPORT_LOCATION, \
f8220fe2 857 m, REPORT_LOCATION_ARGS(loc)))
5e0a247b 858
f8220fe2
KW
859#define ckWARNreg(loc,m) \
860 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
861 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 862 m REPORT_LOCATION, \
f8220fe2 863 REPORT_LOCATION_ARGS(loc)))
ccb2c380 864
f8220fe2
KW
865#define vWARN(loc, m) \
866 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
867 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 868 m REPORT_LOCATION, \
f8220fe2 869 REPORT_LOCATION_ARGS(loc))) \
b927b7e9 870
f8220fe2
KW
871#define vWARN_dep(loc, m) \
872 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
873 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 874 m REPORT_LOCATION, \
f8220fe2 875 REPORT_LOCATION_ARGS(loc)))
0d6106aa 876
f8220fe2
KW
877#define ckWARNdep(loc,m) \
878 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
879 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 880 m REPORT_LOCATION, \
f8220fe2 881 REPORT_LOCATION_ARGS(loc)))
147508a2 882
f8220fe2
KW
883#define ckWARNregdep(loc,m) \
884 _WARN_HELPER(loc, packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
885 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, \
d528642a
KW
886 WARN_REGEXP), \
887 m REPORT_LOCATION, \
f8220fe2 888 REPORT_LOCATION_ARGS(loc)))
ccb2c380 889
f8220fe2
KW
890#define ckWARN2reg_d(loc,m, a1) \
891 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
892 Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
d528642a 893 m REPORT_LOCATION, \
f8220fe2 894 a1, REPORT_LOCATION_ARGS(loc)))
2335b3d3 895
f8220fe2
KW
896#define ckWARN2reg(loc, m, a1) \
897 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
898 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 899 m REPORT_LOCATION, \
f8220fe2 900 a1, REPORT_LOCATION_ARGS(loc)))
ccb2c380 901
f8220fe2
KW
902#define vWARN3(loc, m, a1, a2) \
903 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
904 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 905 m REPORT_LOCATION, \
f8220fe2 906 a1, a2, REPORT_LOCATION_ARGS(loc)))
ccb2c380 907
f8220fe2
KW
908#define ckWARN3reg(loc, m, a1, a2) \
909 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
910 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
911 m REPORT_LOCATION, \
912 a1, a2, \
f8220fe2 913 REPORT_LOCATION_ARGS(loc)))
668c081a 914
f8220fe2
KW
915#define vWARN4(loc, m, a1, a2, a3) \
916 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
917 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
918 m REPORT_LOCATION, \
919 a1, a2, a3, \
f8220fe2 920 REPORT_LOCATION_ARGS(loc)))
ccb2c380 921
f8220fe2
KW
922#define ckWARN4reg(loc, m, a1, a2, a3) \
923 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
924 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
925 m REPORT_LOCATION, \
926 a1, a2, a3, \
f8220fe2 927 REPORT_LOCATION_ARGS(loc)))
668c081a 928
f8220fe2
KW
929#define vWARN5(loc, m, a1, a2, a3, a4) \
930 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
931 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
932 m REPORT_LOCATION, \
933 a1, a2, a3, a4, \
f8220fe2 934 REPORT_LOCATION_ARGS(loc)))
9d1d55b5 935
aea5f181
KW
936#define ckWARNexperimental(loc, class, m) \
937 _WARN_HELPER(loc, packWARN(class), \
938 Perl_ck_warner_d(aTHX_ packWARN(class), \
939 m REPORT_LOCATION, \
940 REPORT_LOCATION_ARGS(loc)))
941
1a322bb4
KW
942/* Convert between a pointer to a node and its offset from the beginning of the
943 * program */
944#define REGNODE_p(offset) (RExC_emit_start + (offset))
945#define REGNODE_OFFSET(node) ((node) - RExC_emit_start)
946
538e84ed 947/* Macros for recording node offsets. 20001227 mjd@plover.com
fac92740
MJD
948 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
949 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
950 * Element 0 holds the number n.
07be1b83 951 * Position is 1 indexed.
fac92740 952 */
7122b237 953#ifndef RE_TRACK_PATTERN_OFFSETS
d7c442ce 954#define Set_Node_Offset_To_R(offset,byte)
7122b237
YO
955#define Set_Node_Offset(node,byte)
956#define Set_Cur_Node_Offset
957#define Set_Node_Length_To_R(node,len)
958#define Set_Node_Length(node,len)
6a86c6ad 959#define Set_Node_Cur_Length(node,start)
538e84ed
KW
960#define Node_Offset(n)
961#define Node_Length(n)
7122b237
YO
962#define Set_Node_Offset_Length(node,offset,len)
963#define ProgLen(ri) ri->u.proglen
964#define SetProgLen(ri,x) ri->u.proglen = x
4cbeff97 965#define Track_Code(code)
7122b237
YO
966#else
967#define ProgLen(ri) ri->u.offsets[0]
968#define SetProgLen(ri,x) ri->u.offsets[0] = x
d7c442ce 969#define Set_Node_Offset_To_R(offset,byte) STMT_START { \
ccb2c380 970 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
d7c442ce
KW
971 __LINE__, (int)(offset), (int)(byte))); \
972 if((offset) < 0) { \
538e84ed 973 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
d7c442ce 974 (int)(offset)); \
ccb2c380 975 } else { \
d7c442ce 976 RExC_offsets[2*(offset)-1] = (byte); \
ccb2c380 977 } \
ccb2c380
MP
978} STMT_END
979
88f063b4 980#define Set_Node_Offset(node,byte) \
1a322bb4 981 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (byte)-RExC_start)
ccb2c380
MP
982#define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
983
984#define Set_Node_Length_To_R(node,len) STMT_START { \
ccb2c380 985 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
551405c4 986 __LINE__, (int)(node), (int)(len))); \
ccb2c380 987 if((node) < 0) { \
538e84ed
KW
988 Perl_croak(aTHX_ "value of node is %d in Length macro", \
989 (int)(node)); \
ccb2c380
MP
990 } else { \
991 RExC_offsets[2*(node)] = (len); \
992 } \
ccb2c380
MP
993} STMT_END
994
995#define Set_Node_Length(node,len) \
1a322bb4 996 Set_Node_Length_To_R(REGNODE_OFFSET(node), len)
6a86c6ad
NC
997#define Set_Node_Cur_Length(node, start) \
998 Set_Node_Length(node, RExC_parse - start)
fac92740
MJD
999
1000/* Get offsets and lengths */
1a322bb4
KW
1001#define Node_Offset(n) (RExC_offsets[2*(REGNODE_OFFSET(n))-1])
1002#define Node_Length(n) (RExC_offsets[2*(REGNODE_OFFSET(n))])
fac92740 1003
07be1b83 1004#define Set_Node_Offset_Length(node,offset,len) STMT_START { \
1a322bb4
KW
1005 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (offset)); \
1006 Set_Node_Length_To_R(REGNODE_OFFSET(node), (len)); \
07be1b83 1007} STMT_END
4cbeff97
TC
1008
1009#define Track_Code(code) STMT_START { code } STMT_END
7122b237 1010#endif
07be1b83
YO
1011
1012#if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
1013#define EXPERIMENTAL_INPLACESCAN
f427392e 1014#endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
07be1b83 1015
cb41e5d6
YO
1016#ifdef DEBUGGING
1017int
6ad9a8ab 1018Perl_re_printf(pTHX_ const char *fmt, ...)
cb41e5d6
YO
1019{
1020 va_list ap;
1021 int result;
1022 PerlIO *f= Perl_debug_log;
1023 PERL_ARGS_ASSERT_RE_PRINTF;
1024 va_start(ap, fmt);
1025 result = PerlIO_vprintf(f, fmt, ap);
1026 va_end(ap);
1027 return result;
1028}
1029
1030int
7b031478 1031Perl_re_indentf(pTHX_ const char *fmt, U32 depth, ...)
cb41e5d6
YO
1032{
1033 va_list ap;
1034 int result;
1035 PerlIO *f= Perl_debug_log;
1036 PERL_ARGS_ASSERT_RE_INDENTF;
1037 va_start(ap, depth);
daeb874b 1038 PerlIO_printf(f, "%*s", ( (int)depth % 20 ) * 2, "");
cb41e5d6
YO
1039 result = PerlIO_vprintf(f, fmt, ap);
1040 va_end(ap);
1041 return result;
1042}
1043#endif /* DEBUGGING */
1044
7b031478 1045#define DEBUG_RExC_seen() \
538e84ed 1046 DEBUG_OPTIMISE_MORE_r({ \
88f063b4 1047 Perl_re_printf( aTHX_ "RExC_seen: "); \
538e84ed 1048 \
e384d5c1 1049 if (RExC_seen & REG_ZERO_LEN_SEEN) \
88f063b4 1050 Perl_re_printf( aTHX_ "REG_ZERO_LEN_SEEN "); \
538e84ed 1051 \
e384d5c1 1052 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
88f063b4 1053 Perl_re_printf( aTHX_ "REG_LOOKBEHIND_SEEN "); \
538e84ed 1054 \
e384d5c1 1055 if (RExC_seen & REG_GPOS_SEEN) \
88f063b4 1056 Perl_re_printf( aTHX_ "REG_GPOS_SEEN "); \
538e84ed 1057 \
e384d5c1 1058 if (RExC_seen & REG_RECURSE_SEEN) \
88f063b4 1059 Perl_re_printf( aTHX_ "REG_RECURSE_SEEN "); \
538e84ed 1060 \
7b031478 1061 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
88f063b4 1062 Perl_re_printf( aTHX_ "REG_TOP_LEVEL_BRANCHES_SEEN "); \
538e84ed 1063 \
e384d5c1 1064 if (RExC_seen & REG_VERBARG_SEEN) \
88f063b4 1065 Perl_re_printf( aTHX_ "REG_VERBARG_SEEN "); \
538e84ed 1066 \
e384d5c1 1067 if (RExC_seen & REG_CUTGROUP_SEEN) \
88f063b4 1068 Perl_re_printf( aTHX_ "REG_CUTGROUP_SEEN "); \
538e84ed 1069 \
e384d5c1 1070 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
88f063b4 1071 Perl_re_printf( aTHX_ "REG_RUN_ON_COMMENT_SEEN "); \
538e84ed 1072 \
e384d5c1 1073 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
88f063b4 1074 Perl_re_printf( aTHX_ "REG_UNFOLDED_MULTI_SEEN "); \
538e84ed 1075 \
7b031478 1076 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
88f063b4 1077 Perl_re_printf( aTHX_ "REG_UNBOUNDED_QUANTIFIER_SEEN "); \
ee273784 1078 \
88f063b4 1079 Perl_re_printf( aTHX_ "\n"); \
9e9ecfdf
YO
1080 });
1081
fdfb4f21 1082#define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
6ad9a8ab 1083 if ((flags) & flag) Perl_re_printf( aTHX_ "%s ", #flag)
fdfb4f21 1084
fdfb4f21 1085
f5a36d78
DM
1086#ifdef DEBUGGING
1087static void
1088S_debug_show_study_flags(pTHX_ U32 flags, const char *open_str,
1089 const char *close_str)
1090{
1091 if (!flags)
1092 return;
1093
1094 Perl_re_printf( aTHX_ "%s", open_str);
11683ecb
DM
1095 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_SEOL);
1096 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_MEOL);
f5a36d78
DM
1097 DEBUG_SHOW_STUDY_FLAG(flags, SF_IS_INF);
1098 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_PAR);
1099 DEBUG_SHOW_STUDY_FLAG(flags, SF_IN_PAR);
1100 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_EVAL);
1101 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_SUBSTR);
1102 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_AND);
1103 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_OR);
1104 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS);
1105 DEBUG_SHOW_STUDY_FLAG(flags, SCF_WHILEM_VISITED_POS);
1106 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_RESTUDY);
1107 DEBUG_SHOW_STUDY_FLAG(flags, SCF_SEEN_ACCEPT);
1108 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_DOING_RESTUDY);
1109 DEBUG_SHOW_STUDY_FLAG(flags, SCF_IN_DEFINE);
1110 Perl_re_printf( aTHX_ "%s", close_str);
1111}
1112
1113
1114static void
1115S_debug_studydata(pTHX_ const char *where, scan_data_t *data,
1116 U32 depth, int is_inf)
1117{
1118 GET_RE_DEBUG_FLAGS_DECL;
1119
1120 DEBUG_OPTIMISE_MORE_r({
1121 if (!data)
1122 return;
1123 Perl_re_indentf(aTHX_ "%s: Pos:%" IVdf "/%" IVdf " Flags: 0x%" UVXf,
1124 depth,
1125 where,
1126 (IV)data->pos_min,
1127 (IV)data->pos_delta,
1128 (UV)data->flags
1129 );
1130
1131 S_debug_show_study_flags(aTHX_ data->flags," [","]");
fdfb4f21 1132
f5a36d78
DM
1133 Perl_re_printf( aTHX_
1134 " Whilem_c: %" IVdf " Lcp: %" IVdf " %s",
1135 (IV)data->whilem_c,
1136 (IV)(data->last_closep ? *((data)->last_closep) : -1),
1137 is_inf ? "INF " : ""
1138 );
1139
11683ecb 1140 if (data->last_found) {
37b6262f 1141 int i;
11683ecb
DM
1142 Perl_re_printf(aTHX_
1143 "Last:'%s' %" IVdf ":%" IVdf "/%" IVdf,
1144 SvPVX_const(data->last_found),
1145 (IV)data->last_end,
1146 (IV)data->last_start_min,
1147 (IV)data->last_start_max
1148 );
1149
2099df82 1150 for (i = 0; i < 2; i++) {
37b6262f
DM
1151 Perl_re_printf(aTHX_
1152 " %s%s: '%s' @ %" IVdf "/%" IVdf,
2099df82
DM
1153 data->cur_is_floating == i ? "*" : "",
1154 i ? "Float" : "Fixed",
37b6262f
DM
1155 SvPVX_const(data->substrs[i].str),
1156 (IV)data->substrs[i].min_offset,
1157 (IV)data->substrs[i].max_offset
1158 );
1159 S_debug_show_study_flags(aTHX_ data->substrs[i].flags," [","]");
1160 }
11683ecb 1161 }
f5a36d78
DM
1162
1163 Perl_re_printf( aTHX_ "\n");
1164 });
1165}
1166
1167
1168static void
1169S_debug_peep(pTHX_ const char *str, const RExC_state_t *pRExC_state,
1170 regnode *scan, U32 depth, U32 flags)
1171{
1172 GET_RE_DEBUG_FLAGS_DECL;
1173
1174 DEBUG_OPTIMISE_r({
1175 regnode *Next;
1176
1177 if (!scan)
1178 return;
1179 Next = regnext(scan);
1180 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
1181 Perl_re_indentf( aTHX_ "%s>%3d: %s (%d)",
1182 depth,
1183 str,
1184 REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),
1185 Next ? (REG_NODE_NUM(Next)) : 0 );
1186 S_debug_show_study_flags(aTHX_ flags," [ ","]");
1187 Perl_re_printf( aTHX_ "\n");
1188 });
1189}
1190
1191
1192# define DEBUG_STUDYDATA(where, data, depth, is_inf) \
1193 S_debug_studydata(aTHX_ where, data, depth, is_inf)
1194
1195# define DEBUG_PEEP(str, scan, depth, flags) \
1196 S_debug_peep(aTHX_ str, pRExC_state, scan, depth, flags)
1197
1198#else
1199# define DEBUG_STUDYDATA(where, data, depth, is_inf) NOOP
1200# define DEBUG_PEEP(str, scan, depth, flags) NOOP
1201#endif
1de06328 1202
cb41e5d6 1203
c6871b76
KW
1204/* =========================================================
1205 * BEGIN edit_distance stuff.
1206 *
1207 * This calculates how many single character changes of any type are needed to
1208 * transform a string into another one. It is taken from version 3.1 of
1209 *
1210 * https://metacpan.org/pod/Text::Levenshtein::Damerau::XS
1211 */
1212
1213/* Our unsorted dictionary linked list. */
1214/* Note we use UVs, not chars. */
1215
1216struct dictionary{
1217 UV key;
1218 UV value;
1219 struct dictionary* next;
1220};
1221typedef struct dictionary item;
1222
1223
1224PERL_STATIC_INLINE item*
88f063b4 1225push(UV key, item* curr)
c6871b76
KW
1226{
1227 item* head;
d09f14bf 1228 Newx(head, 1, item);
c6871b76
KW
1229 head->key = key;
1230 head->value = 0;
1231 head->next = curr;
1232 return head;
1233}
1234
1235
1236PERL_STATIC_INLINE item*
1237find(item* head, UV key)
1238{
1239 item* iterator = head;
1240 while (iterator){
1241 if (iterator->key == key){
1242 return iterator;
1243 }
1244 iterator = iterator->next;
1245 }
1246
1247 return NULL;
1248}
1249
1250PERL_STATIC_INLINE item*
88f063b4 1251uniquePush(item* head, UV key)
c6871b76
KW
1252{
1253 item* iterator = head;
1254
1255 while (iterator){
1256 if (iterator->key == key) {
1257 return head;
1258 }
1259 iterator = iterator->next;
1260 }
1261
88f063b4 1262 return push(key, head);
c6871b76
KW
1263}
1264
1265PERL_STATIC_INLINE void
1266dict_free(item* head)
1267{
1268 item* iterator = head;
1269
1270 while (iterator) {
1271 item* temp = iterator;
1272 iterator = iterator->next;
1273 Safefree(temp);
1274 }
1275
1276 head = NULL;
1277}
1278
1279/* End of Dictionary Stuff */
1280
1281/* All calculations/work are done here */
1282STATIC int
1283S_edit_distance(const UV* src,
1284 const UV* tgt,
1285 const STRLEN x, /* length of src[] */
1286 const STRLEN y, /* length of tgt[] */
1287 const SSize_t maxDistance
1288)
1289{
1290 item *head = NULL;
88f063b4 1291 UV swapCount, swapScore, targetCharCount, i, j;
c6871b76
KW
1292 UV *scores;
1293 UV score_ceil = x + y;
1294
1295 PERL_ARGS_ASSERT_EDIT_DISTANCE;
1296
1297 /* intialize matrix start values */
d09f14bf 1298 Newx(scores, ( (x + 2) * (y + 2)), UV);
c6871b76
KW
1299 scores[0] = score_ceil;
1300 scores[1 * (y + 2) + 0] = score_ceil;
1301 scores[0 * (y + 2) + 1] = score_ceil;
1302 scores[1 * (y + 2) + 1] = 0;
88f063b4 1303 head = uniquePush(uniquePush(head, src[0]), tgt[0]);
c6871b76
KW
1304
1305 /* work loops */
1306 /* i = src index */
1307 /* j = tgt index */
1308 for (i=1;i<=x;i++) {
1309 if (i < x)
88f063b4 1310 head = uniquePush(head, src[i]);
c6871b76
KW
1311 scores[(i+1) * (y + 2) + 1] = i;
1312 scores[(i+1) * (y + 2) + 0] = score_ceil;
1313 swapCount = 0;
1314
1315 for (j=1;j<=y;j++) {
1316 if (i == 1) {
1317 if(j < y)
88f063b4 1318 head = uniquePush(head, tgt[j]);
c6871b76
KW
1319 scores[1 * (y + 2) + (j + 1)] = j;
1320 scores[0 * (y + 2) + (j + 1)] = score_ceil;
1321 }
1322
88f063b4 1323 targetCharCount = find(head, tgt[j-1])->value;
c6871b76
KW
1324 swapScore = scores[targetCharCount * (y + 2) + swapCount] + i - targetCharCount - 1 + j - swapCount;
1325
1326 if (src[i-1] != tgt[j-1]){
1327 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));
1328 }
1329 else {
1330 swapCount = j;
1331 scores[(i+1) * (y + 2) + (j + 1)] = MIN(scores[i * (y + 2) + j], swapScore);
1332 }
1333 }
1334
88f063b4 1335 find(head, src[i-1])->value = i;
c6871b76
KW
1336 }
1337
1338 {
1339 IV score = scores[(x+1) * (y + 2) + (y + 1)];
1340 dict_free(head);
1341 Safefree(scores);
1342 return (maxDistance != 0 && maxDistance < score)?(-1):score;
1343 }
1344}
1345
1346/* END of edit_distance() stuff
1347 * ========================================================= */
1348
8e35b056
KW
1349/* is c a control character for which we have a mnemonic? */
1350#define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
1351
549b4e78
KW
1352STATIC const char *
1353S_cntrl_to_mnemonic(const U8 c)
1354{
1355 /* Returns the mnemonic string that represents character 'c', if one
1356 * exists; NULL otherwise. The only ones that exist for the purposes of
1357 * this routine are a few control characters */
1358
1359 switch (c) {
1360 case '\a': return "\\a";
1361 case '\b': return "\\b";
1362 case ESC_NATIVE: return "\\e";
1363 case '\f': return "\\f";
1364 case '\n': return "\\n";
1365 case '\r': return "\\r";
1366 case '\t': return "\\t";
1367 }
1368
1369 return NULL;
1370}
1371
653099ff 1372/* Mark that we cannot extend a found fixed substring at this point.
37b6262f 1373 Update the longest found anchored substring or the longest found
653099ff
GS
1374 floating substrings if needed. */
1375
4327152a 1376STATIC void
ea3daa5d
FC
1377S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
1378 SSize_t *minlenp, int is_inf)
c277df42 1379{
e1ec3a88 1380 const STRLEN l = CHR_SVLEN(data->last_found);
2099df82 1381 SV * const longest_sv = data->substrs[data->cur_is_floating].str;
a2ca2017 1382 const STRLEN old_l = CHR_SVLEN(longest_sv);
1de06328 1383 GET_RE_DEBUG_FLAGS_DECL;
b81d288d 1384
7918f24d
NC
1385 PERL_ARGS_ASSERT_SCAN_COMMIT;
1386
c277df42 1387 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
2099df82 1388 const U8 i = data->cur_is_floating;
a2ca2017 1389 SvSetMagicSV(longest_sv, data->last_found);
37b6262f
DM
1390 data->substrs[i].min_offset = l ? data->last_start_min : data->pos_min;
1391
2099df82
DM
1392 if (!i) /* fixed */
1393 data->substrs[0].max_offset = data->substrs[0].min_offset;
1394 else { /* float */
1395 data->substrs[1].max_offset = (l
646e8787
DM
1396 ? data->last_start_max
1397 : (data->pos_delta > SSize_t_MAX - data->pos_min
ea3daa5d
FC
1398 ? SSize_t_MAX
1399 : data->pos_min + data->pos_delta));
1400 if (is_inf
2099df82
DM
1401 || (STRLEN)data->substrs[1].max_offset > (STRLEN)SSize_t_MAX)
1402 data->substrs[1].max_offset = SSize_t_MAX;
37b6262f 1403 }
11683ecb 1404
37b6262f
DM
1405 if (data->flags & SF_BEFORE_EOL)
1406 data->substrs[i].flags |= (data->flags & SF_BEFORE_EOL);
1407 else
1408 data->substrs[i].flags &= ~SF_BEFORE_EOL;
1409 data->substrs[i].minlenp = minlenp;
1410 data->substrs[i].lookbehind = 0;
c277df42 1411 }
37b6262f 1412
c277df42 1413 SvCUR_set(data->last_found, 0);
0eda9292 1414 {
a28509cc 1415 SV * const sv = data->last_found;
097eb12c
AL
1416 if (SvUTF8(sv) && SvMAGICAL(sv)) {
1417 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
1418 if (mg)
1419 mg->mg_len = 0;
1420 }
0eda9292 1421 }
c277df42
IZ
1422 data->last_end = -1;
1423 data->flags &= ~SF_BEFORE_EOL;
f5a36d78 1424 DEBUG_STUDYDATA("commit", data, 0, is_inf);
c277df42
IZ
1425}
1426
cdd87c1d
KW
1427/* An SSC is just a regnode_charclass_posix with an extra field: the inversion
1428 * list that describes which code points it matches */
1429
653099ff 1430STATIC void
3420edd7 1431S_ssc_anything(pTHX_ regnode_ssc *ssc)
653099ff 1432{
cdd87c1d
KW
1433 /* Set the SSC 'ssc' to match an empty string or any code point */
1434
557bd3fb 1435 PERL_ARGS_ASSERT_SSC_ANYTHING;
7918f24d 1436
71068078 1437 assert(is_ANYOF_SYNTHETIC(ssc));
3fffb88a 1438
0854ea0b
KW
1439 /* mortalize so won't leak */
1440 ssc->invlist = sv_2mortal(_add_range_to_invlist(NULL, 0, UV_MAX));
93e92956 1441 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
653099ff
GS
1442}
1443
653099ff 1444STATIC int
dc3bf405 1445S_ssc_is_anything(const regnode_ssc *ssc)
653099ff 1446{
c144baaa
KW
1447 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
1448 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
1449 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1450 * in any way, so there's no point in using it */
cdd87c1d
KW
1451
1452 UV start, end;
1453 bool ret;
653099ff 1454
557bd3fb 1455 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
7918f24d 1456
71068078 1457 assert(is_ANYOF_SYNTHETIC(ssc));
cdd87c1d 1458
93e92956 1459 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
cdd87c1d
KW
1460 return FALSE;
1461 }
1462
1463 /* See if the list consists solely of the range 0 - Infinity */
1464 invlist_iterinit(ssc->invlist);
1465 ret = invlist_iternext(ssc->invlist, &start, &end)
1466 && start == 0
1467 && end == UV_MAX;
1468
1469 invlist_iterfinish(ssc->invlist);
1470
1471 if (ret) {
1472 return TRUE;
1473 }
1474
1475 /* If e.g., both \w and \W are set, matches everything */
e0e1be5f 1476 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1477 int i;
1478 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1479 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1480 return TRUE;
1481 }
1482 }
1483 }
1484
1485 return FALSE;
653099ff
GS
1486}
1487
653099ff 1488STATIC void
cdd87c1d 1489S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
653099ff 1490{
cdd87c1d
KW
1491 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1492 * string, any code point, or any posix class under locale */
1493
557bd3fb 1494 PERL_ARGS_ASSERT_SSC_INIT;
7918f24d 1495
557bd3fb 1496 Zero(ssc, 1, regnode_ssc);
71068078 1497 set_ANYOF_SYNTHETIC(ssc);
93e92956 1498 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
3420edd7 1499 ssc_anything(ssc);
cdd87c1d 1500
2f306ab9
KW
1501 /* If any portion of the regex is to operate under locale rules that aren't
1502 * fully known at compile time, initialization includes it. The reason
1503 * this isn't done for all regexes is that the optimizer was written under
1504 * the assumption that locale was all-or-nothing. Given the complexity and
1505 * lack of documentation in the optimizer, and that there are inadequate
1506 * test cases for locale, many parts of it may not work properly, it is
1507 * safest to avoid locale unless necessary. */
cdd87c1d
KW
1508 if (RExC_contains_locale) {
1509 ANYOF_POSIXL_SETALL(ssc);
cdd87c1d
KW
1510 }
1511 else {
1512 ANYOF_POSIXL_ZERO(ssc);
1513 }
653099ff
GS
1514}
1515
b423522f 1516STATIC int
dc3bf405
BF
1517S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1518 const regnode_ssc *ssc)
b423522f
KW
1519{
1520 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1521 * to the list of code points matched, and locale posix classes; hence does
1522 * not check its flags) */
1523
1524 UV start, end;
1525 bool ret;
1526
1527 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1528
71068078 1529 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1530
1531 invlist_iterinit(ssc->invlist);
1532 ret = invlist_iternext(ssc->invlist, &start, &end)
1533 && start == 0
1534 && end == UV_MAX;
1535
1536 invlist_iterfinish(ssc->invlist);
1537
1538 if (! ret) {
1539 return FALSE;
1540 }
1541
e0e1be5f 1542 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
31f05a37 1543 return FALSE;
b423522f
KW
1544 }
1545
1546 return TRUE;
1547}
1548
1549STATIC SV*
1550S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
5c0f85ef 1551 const regnode_charclass* const node)
b423522f
KW
1552{
1553 /* Returns a mortal inversion list defining which code points are matched
1554 * by 'node', which is of type ANYOF. Handles complementing the result if
1555 * appropriate. If some code points aren't knowable at this time, the
31f05a37
KW
1556 * returned list must, and will, contain every code point that is a
1557 * possibility. */
b423522f 1558
e2506fa7 1559 SV* invlist = NULL;
1ee208c4 1560 SV* only_utf8_locale_invlist = NULL;
b423522f
KW
1561 unsigned int i;
1562 const U32 n = ARG(node);
31f05a37 1563 bool new_node_has_latin1 = FALSE;
b423522f
KW
1564
1565 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1566
1567 /* Look at the data structure created by S_set_ANYOF_arg() */
93e92956 1568 if (n != ANYOF_ONLY_HAS_BITMAP) {
b423522f
KW
1569 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1570 AV * const av = MUTABLE_AV(SvRV(rv));
1571 SV **const ary = AvARRAY(av);
1572 assert(RExC_rxi->data->what[n] == 's');
1573
1574 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
28118b9c 1575 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1]), NULL));
b423522f
KW
1576 }
1577 else if (ary[0] && ary[0] != &PL_sv_undef) {
1578
1579 /* Here, no compile-time swash, and there are things that won't be
1580 * known until runtime -- we have to assume it could be anything */
e2506fa7 1581 invlist = sv_2mortal(_new_invlist(1));
b423522f
KW
1582 return _add_range_to_invlist(invlist, 0, UV_MAX);
1583 }
1ee208c4 1584 else if (ary[3] && ary[3] != &PL_sv_undef) {
b423522f
KW
1585
1586 /* Here no compile-time swash, and no run-time only data. Use the
1587 * node's inversion list */
28118b9c 1588 invlist = sv_2mortal(invlist_clone(ary[3], NULL));
1ee208c4
KW
1589 }
1590
1591 /* Get the code points valid only under UTF-8 locales */
037715a6 1592 if ((ANYOF_FLAGS(node) & ANYOFL_FOLD)
1ee208c4
KW
1593 && ary[2] && ary[2] != &PL_sv_undef)
1594 {
1595 only_utf8_locale_invlist = ary[2];
b423522f
KW
1596 }
1597 }
1598
e2506fa7
KW
1599 if (! invlist) {
1600 invlist = sv_2mortal(_new_invlist(0));
1601 }
1602
dcb20b36
KW
1603 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1604 * code points, and an inversion list for the others, but if there are code
1605 * points that should match only conditionally on the target string being
1606 * UTF-8, those are placed in the inversion list, and not the bitmap.
1607 * Since there are circumstances under which they could match, they are
1608 * included in the SSC. But if the ANYOF node is to be inverted, we have
1609 * to exclude them here, so that when we invert below, the end result
1610 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1611 * have to do this here before we add the unconditionally matched code
1612 * points */
b423522f
KW
1613 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1614 _invlist_intersection_complement_2nd(invlist,
1615 PL_UpperLatin1,
1616 &invlist);
1617 }
1618
1619 /* Add in the points from the bit map */
5aa750d1 1620 if (OP(node) != ANYOFH) {
90973738
KW
1621 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
1622 if (ANYOF_BITMAP_TEST(node, i)) {
1623 unsigned int start = i++;
1624
1625 for (; i < NUM_ANYOF_CODE_POINTS
1626 && ANYOF_BITMAP_TEST(node, i); ++i)
1627 {
1628 /* empty */
1629 }
1630 invlist = _add_range_to_invlist(invlist, start, i-1);
1631 new_node_has_latin1 = TRUE;
6f8848d5 1632 }
b423522f
KW
1633 }
1634 }
1635
1636 /* If this can match all upper Latin1 code points, have to add them
ac33c516
KW
1637 * as well. But don't add them if inverting, as when that gets done below,
1638 * it would exclude all these characters, including the ones it shouldn't
1639 * that were added just above */
1640 if (! (ANYOF_FLAGS(node) & ANYOF_INVERT) && OP(node) == ANYOFD
f240c685
KW
1641 && (ANYOF_FLAGS(node) & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER))
1642 {
b423522f
KW
1643 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1644 }
1645
1646 /* Similarly for these */
93e92956 1647 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
e0a1ff7a 1648 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
b423522f
KW
1649 }
1650
1651 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1652 _invlist_invert(invlist);
1653 }
037715a6 1654 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOFL_FOLD) {
31f05a37
KW
1655
1656 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1657 * locale. We can skip this if there are no 0-255 at all. */
1658 _invlist_union(invlist, PL_Latin1, &invlist);
1659 }
1660
1ee208c4
KW
1661 /* Similarly add the UTF-8 locale possible matches. These have to be
1662 * deferred until after the non-UTF-8 locale ones are taken care of just
1663 * above, or it leads to wrong results under ANYOF_INVERT */
1664 if (only_utf8_locale_invlist) {
31f05a37 1665 _invlist_union_maybe_complement_2nd(invlist,
1ee208c4 1666 only_utf8_locale_invlist,
31f05a37
KW
1667 ANYOF_FLAGS(node) & ANYOF_INVERT,
1668 &invlist);
1669 }
b423522f
KW
1670
1671 return invlist;
1672}
1673
1051e1c4 1674/* These two functions currently do the exact same thing */
557bd3fb 1675#define ssc_init_zero ssc_init
653099ff 1676
cdd87c1d
KW
1677#define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1678#define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1679
557bd3fb 1680/* 'AND' a given class with another one. Can create false positives. 'ssc'
93e92956
KW
1681 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1682 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
cdd87c1d 1683
653099ff 1684STATIC void
b423522f 1685S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1686 const regnode_charclass *and_with)
653099ff 1687{
cdd87c1d
KW
1688 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1689 * another SSC or a regular ANYOF class. Can create false positives. */
40d049e4 1690
a0dd4231
KW
1691 SV* anded_cp_list;
1692 U8 anded_flags;
1e6ade67 1693
cdd87c1d 1694 PERL_ARGS_ASSERT_SSC_AND;
653099ff 1695
71068078 1696 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1697
1698 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1699 * the code point inversion list and just the relevant flags */
71068078 1700 if (is_ANYOF_SYNTHETIC(and_with)) {
7dcac5f6 1701 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
a0dd4231 1702 anded_flags = ANYOF_FLAGS(and_with);
e9b08962
KW
1703
1704 /* XXX This is a kludge around what appears to be deficiencies in the
1705 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1706 * there are paths through the optimizer where it doesn't get weeded
1707 * out when it should. And if we don't make some extra provision for
1708 * it like the code just below, it doesn't get added when it should.
1709 * This solution is to add it only when AND'ing, which is here, and
1710 * only when what is being AND'ed is the pristine, original node
1711 * matching anything. Thus it is like adding it to ssc_anything() but
1712 * only when the result is to be AND'ed. Probably the same solution
1713 * could be adopted for the same problem we have with /l matching,
1714 * which is solved differently in S_ssc_init(), and that would lead to
1715 * fewer false positives than that solution has. But if this solution
1716 * creates bugs, the consequences are only that a warning isn't raised
1717 * that should be; while the consequences for having /l bugs is
1718 * incorrect matches */
7dcac5f6 1719 if (ssc_is_anything((regnode_ssc *)and_with)) {
f240c685 1720 anded_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
e9b08962 1721 }
a0dd4231
KW
1722 }
1723 else {
5c0f85ef 1724 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
f240c685
KW
1725 if (OP(and_with) == ANYOFD) {
1726 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1727 }
1728 else {
1729 anded_flags = ANYOF_FLAGS(and_with)
1730 &( ANYOF_COMMON_FLAGS
108316fb
KW
1731 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1732 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1733 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(and_with))) {
1734 anded_flags &=
1735 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1736 }
f240c685 1737 }
a0dd4231
KW
1738 }
1739
1740 ANYOF_FLAGS(ssc) &= anded_flags;
cdd87c1d
KW
1741
1742 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1743 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1744 * 'and_with' may be inverted. When not inverted, we have the situation of
1745 * computing:
1746 * (C1 | P1) & (C2 | P2)
1747 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1748 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1749 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1750 * <= ((C1 & C2) | P1 | P2)
1751 * Alternatively, the last few steps could be:
1752 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1753 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1754 * <= (C1 | C2 | (P1 & P2))
1755 * We favor the second approach if either P1 or P2 is non-empty. This is
1756 * because these components are a barrier to doing optimizations, as what
1757 * they match cannot be known until the moment of matching as they are
1758 * dependent on the current locale, 'AND"ing them likely will reduce or
1759 * eliminate them.
1760 * But we can do better if we know that C1,P1 are in their initial state (a
1761 * frequent occurrence), each matching everything:
1762 * (<everything>) & (C2 | P2) = C2 | P2
1763 * Similarly, if C2,P2 are in their initial state (again a frequent
1764 * occurrence), the result is a no-op
1765 * (C1 | P1) & (<everything>) = C1 | P1
1766 *
1767 * Inverted, we have
1768 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1769 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1770 * <= (C1 & ~C2) | (P1 & ~P2)
1771 * */
1aa99e6b 1772
a0dd4231 1773 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
71068078 1774 && ! is_ANYOF_SYNTHETIC(and_with))
a0dd4231 1775 {
cdd87c1d 1776 unsigned int i;
8951c461 1777
cdd87c1d
KW
1778 ssc_intersection(ssc,
1779 anded_cp_list,
1780 FALSE /* Has already been inverted */
1781 );
c6b76537 1782
cdd87c1d
KW
1783 /* If either P1 or P2 is empty, the intersection will be also; can skip
1784 * the loop */
93e92956 1785 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
cdd87c1d
KW
1786 ANYOF_POSIXL_ZERO(ssc);
1787 }
e0e1be5f 1788 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1789
1790 /* Note that the Posix class component P from 'and_with' actually
1791 * looks like:
1792 * P = Pa | Pb | ... | Pn
1793 * where each component is one posix class, such as in [\w\s].
1794 * Thus
1795 * ~P = ~(Pa | Pb | ... | Pn)
1796 * = ~Pa & ~Pb & ... & ~Pn
1797 * <= ~Pa | ~Pb | ... | ~Pn
1798 * The last is something we can easily calculate, but unfortunately
1799 * is likely to have many false positives. We could do better
1800 * in some (but certainly not all) instances if two classes in
1801 * P have known relationships. For example
1802 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1803 * So
1804 * :lower: & :print: = :lower:
1805 * And similarly for classes that must be disjoint. For example,
1806 * since \s and \w can have no elements in common based on rules in
1807 * the POSIX standard,
1808 * \w & ^\S = nothing
1809 * Unfortunately, some vendor locales do not meet the Posix
1810 * standard, in particular almost everything by Microsoft.
1811 * The loop below just changes e.g., \w into \W and vice versa */
1812
1ee208c4 1813 regnode_charclass_posixl temp;
cdd87c1d
KW
1814 int add = 1; /* To calculate the index of the complement */
1815
b1234259 1816 Zero(&temp, 1, regnode_charclass_posixl);
cdd87c1d
KW
1817 ANYOF_POSIXL_ZERO(&temp);
1818 for (i = 0; i < ANYOF_MAX; i++) {
1819 assert(i % 2 != 0
7dcac5f6
KW
1820 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1821 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
cdd87c1d 1822
7dcac5f6 1823 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
cdd87c1d
KW
1824 ANYOF_POSIXL_SET(&temp, i + add);
1825 }
1826 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1827 }
1828 ANYOF_POSIXL_AND(&temp, ssc);
c6b76537 1829
cdd87c1d
KW
1830 } /* else ssc already has no posixes */
1831 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1832 in its initial state */
71068078 1833 else if (! is_ANYOF_SYNTHETIC(and_with)
7dcac5f6 1834 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
cdd87c1d
KW
1835 {
1836 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1837 * copy it over 'ssc' */
1838 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
71068078 1839 if (is_ANYOF_SYNTHETIC(and_with)) {
cdd87c1d
KW
1840 StructCopy(and_with, ssc, regnode_ssc);
1841 }
1842 else {
1843 ssc->invlist = anded_cp_list;
1844 ANYOF_POSIXL_ZERO(ssc);
93e92956 1845 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1846 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
cdd87c1d
KW
1847 }
1848 }
1849 }
e0e1be5f 1850 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
93e92956 1851 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
cdd87c1d
KW
1852 {
1853 /* One or the other of P1, P2 is non-empty. */
93e92956 1854 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1ea8b7fe
KW
1855 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1856 }
cdd87c1d
KW
1857 ssc_union(ssc, anded_cp_list, FALSE);
1858 }
1859 else { /* P1 = P2 = empty */
1860 ssc_intersection(ssc, anded_cp_list, FALSE);
1861 }
137165a6 1862 }
653099ff
GS
1863}
1864
653099ff 1865STATIC void
cdd87c1d 1866S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1867 const regnode_charclass *or_with)
653099ff 1868{
cdd87c1d
KW
1869 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1870 * another SSC or a regular ANYOF class. Can create false positives if
1871 * 'or_with' is to be inverted. */
7918f24d 1872
a0dd4231
KW
1873 SV* ored_cp_list;
1874 U8 ored_flags;
c6b76537 1875
cdd87c1d 1876 PERL_ARGS_ASSERT_SSC_OR;
c6b76537 1877
71068078 1878 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1879
1880 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1881 * the code point inversion list and just the relevant flags */
71068078 1882 if (is_ANYOF_SYNTHETIC(or_with)) {
7dcac5f6 1883 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
a0dd4231
KW
1884 ored_flags = ANYOF_FLAGS(or_with);
1885 }
1886 else {
5c0f85ef 1887 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
eff8b7dc 1888 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
f240c685
KW
1889 if (OP(or_with) != ANYOFD) {
1890 ored_flags
1891 |= ANYOF_FLAGS(or_with)
108316fb
KW
1892 & ( ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1893 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1894 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(or_with))) {
1895 ored_flags |=
1896 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1897 }
f240c685 1898 }
a0dd4231
KW
1899 }
1900
1901 ANYOF_FLAGS(ssc) |= ored_flags;
cdd87c1d
KW
1902
1903 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1904 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1905 * 'or_with' may be inverted. When not inverted, we have the simple
1906 * situation of computing:
1907 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1908 * If P1|P2 yields a situation with both a class and its complement are
1909 * set, like having both \w and \W, this matches all code points, and we
1910 * can delete these from the P component of the ssc going forward. XXX We
1911 * might be able to delete all the P components, but I (khw) am not certain
1912 * about this, and it is better to be safe.
1913 *
1914 * Inverted, we have
1915 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1916 * <= (C1 | P1) | ~C2
1917 * <= (C1 | ~C2) | P1
1918 * (which results in actually simpler code than the non-inverted case)
1919 * */
9826f543 1920
a0dd4231 1921 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
71068078 1922 && ! is_ANYOF_SYNTHETIC(or_with))
a0dd4231 1923 {
cdd87c1d 1924 /* We ignore P2, leaving P1 going forward */
1ea8b7fe 1925 } /* else Not inverted */
93e92956 1926 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1927 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
e0e1be5f 1928 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1929 unsigned int i;
1930 for (i = 0; i < ANYOF_MAX; i += 2) {
1931 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1932 {
1933 ssc_match_all_cp(ssc);
1934 ANYOF_POSIXL_CLEAR(ssc, i);
1935 ANYOF_POSIXL_CLEAR(ssc, i+1);
cdd87c1d
KW
1936 }
1937 }
1938 }
1aa99e6b 1939 }
cdd87c1d
KW
1940
1941 ssc_union(ssc,
1942 ored_cp_list,
1943 FALSE /* Already has been inverted */
1944 );
653099ff
GS
1945}
1946
b423522f
KW
1947PERL_STATIC_INLINE void
1948S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1949{
1950 PERL_ARGS_ASSERT_SSC_UNION;
1951
71068078 1952 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1953
1954 _invlist_union_maybe_complement_2nd(ssc->invlist,
1955 invlist,
1956 invert2nd,
1957 &ssc->invlist);
1958}
1959
1960PERL_STATIC_INLINE void
1961S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1962 SV* const invlist,
1963 const bool invert2nd)
1964{
1965 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1966
71068078 1967 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1968
1969 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1970 invlist,
1971 invert2nd,
1972 &ssc->invlist);
1973}
1974
1975PERL_STATIC_INLINE void
1976S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1977{
1978 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1979
71068078 1980 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1981
1982 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1983}
1984
1985PERL_STATIC_INLINE void
1986S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1987{
1988 /* AND just the single code point 'cp' into the SSC 'ssc' */
1989
1990 SV* cp_list = _new_invlist(2);
1991
1992 PERL_ARGS_ASSERT_SSC_CP_AND;
1993
71068078 1994 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1995
1996 cp_list = add_cp_to_invlist(cp_list, cp);
1997 ssc_intersection(ssc, cp_list,
1998 FALSE /* Not inverted */
1999 );
2000 SvREFCNT_dec_NN(cp_list);
2001}
2002
2003PERL_STATIC_INLINE void
dc3bf405 2004S_ssc_clear_locale(regnode_ssc *ssc)
b423522f
KW
2005{
2006 /* Set the SSC 'ssc' to not match any locale things */
b423522f
KW
2007 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
2008
71068078 2009 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
2010
2011 ANYOF_POSIXL_ZERO(ssc);
2012 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
2013}
2014
b35552de
KW
2015#define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
2016
2017STATIC bool
2018S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
2019{
2020 /* The synthetic start class is used to hopefully quickly winnow down
2021 * places where a pattern could start a match in the target string. If it
2022 * doesn't really narrow things down that much, there isn't much point to
2023 * having the overhead of using it. This function uses some very crude
2024 * heuristics to decide if to use the ssc or not.
2025 *
2026 * It returns TRUE if 'ssc' rules out more than half what it considers to
2027 * be the "likely" possible matches, but of course it doesn't know what the
2028 * actual things being matched are going to be; these are only guesses
2029 *
2030 * For /l matches, it assumes that the only likely matches are going to be
2031 * in the 0-255 range, uniformly distributed, so half of that is 127
2032 * For /a and /d matches, it assumes that the likely matches will be just
2033 * the ASCII range, so half of that is 63
2034 * For /u and there isn't anything matching above the Latin1 range, it
2035 * assumes that that is the only range likely to be matched, and uses
2036 * half that as the cut-off: 127. If anything matches above Latin1,
2037 * it assumes that all of Unicode could match (uniformly), except for
2038 * non-Unicode code points and things in the General Category "Other"
2039 * (unassigned, private use, surrogates, controls and formats). This
2040 * is a much large number. */
2041
b35552de
KW
2042 U32 count = 0; /* Running total of number of code points matched by
2043 'ssc' */
2044 UV start, end; /* Start and end points of current range in inversion
2045 list */
72400949
KW
2046 const U32 max_code_points = (LOC)
2047 ? 256
f6e8f31e 2048 : (( ! UNI_SEMANTICS
1a26cbcb 2049 || invlist_highest(ssc->invlist) < 256)
72400949
KW
2050 ? 128
2051 : NON_OTHER_COUNT);
2052 const U32 max_match = max_code_points / 2;
b35552de
KW
2053
2054 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
2055
2056 invlist_iterinit(ssc->invlist);
2057 while (invlist_iternext(ssc->invlist, &start, &end)) {
72400949
KW
2058 if (start >= max_code_points) {
2059 break;
b35552de 2060 }
72400949 2061 end = MIN(end, max_code_points - 1);
b35552de 2062 count += end - start + 1;
72400949 2063 if (count >= max_match) {
b35552de
KW
2064 invlist_iterfinish(ssc->invlist);
2065 return FALSE;
2066 }
2067 }
2068
2069 return TRUE;
2070}
2071
2072
b423522f
KW
2073STATIC void
2074S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
2075{
2076 /* The inversion list in the SSC is marked mortal; now we need a more
2077 * permanent copy, which is stored the same way that is done in a regular
dcb20b36
KW
2078 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
2079 * map */
b423522f 2080
28118b9c 2081 SV* invlist = invlist_clone(ssc->invlist, NULL);
b423522f
KW
2082
2083 PERL_ARGS_ASSERT_SSC_FINALIZE;
2084
71068078 2085 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f 2086
a0dd4231 2087 /* The code in this file assumes that all but these flags aren't relevant
93e92956
KW
2088 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
2089 * by the time we reach here */
f240c685
KW
2090 assert(! (ANYOF_FLAGS(ssc)
2091 & ~( ANYOF_COMMON_FLAGS
108316fb
KW
2092 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
2093 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)));
a0dd4231 2094
b423522f
KW
2095 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
2096
1ee208c4
KW
2097 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
2098 NULL, NULL, NULL, FALSE);
b423522f 2099
85c8e306
KW
2100 /* Make sure is clone-safe */
2101 ssc->invlist = NULL;
2102
e0e1be5f 2103 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
93e92956 2104 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
d156f5cb 2105 OP(ssc) = ANYOFPOSIXL;
e0e1be5f 2106 }
d156f5cb 2107 else if (RExC_contains_locale) {
b2e90ddf
KW
2108 OP(ssc) = ANYOFL;
2109 }
2110
1462525b 2111 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
b423522f
KW
2112}
2113
a3621e74
YO
2114#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
2115#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
2116#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
538e84ed
KW
2117#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
2118 ? (TRIE_LIST_CUR( idx ) - 1) \
2119 : 0 )
a3621e74 2120
3dab1dad
YO
2121
2122#ifdef DEBUGGING
07be1b83 2123/*
2b8b4781
NC
2124 dump_trie(trie,widecharmap,revcharmap)
2125 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
2126 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
2127
2128 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
2129 The _interim_ variants are used for debugging the interim
2130 tables that are used to generate the final compressed
2131 representation which is what dump_trie expects.
2132
486ec47a 2133 Part of the reason for their existence is to provide a form
3dab1dad 2134 of documentation as to how the different representations function.
07be1b83
YO
2135
2136*/
3dab1dad
YO
2137
2138/*
3dab1dad
YO
2139 Dumps the final compressed table form of the trie to Perl_debug_log.
2140 Used for debugging make_trie().
2141*/
b9a59e08 2142
3dab1dad 2143STATIC void
2b8b4781
NC
2144S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
2145 AV *revcharmap, U32 depth)
3dab1dad
YO
2146{
2147 U32 state;
ab3bbdeb 2148 SV *sv=sv_newmortal();
55eed653 2149 int colwidth= widecharmap ? 6 : 4;
2e64971a 2150 U16 word;
3dab1dad
YO
2151 GET_RE_DEBUG_FLAGS_DECL;
2152
7918f24d 2153 PERL_ARGS_ASSERT_DUMP_TRIE;
ab3bbdeb 2154
6ad9a8ab 2155 Perl_re_indentf( aTHX_ "Char : %-6s%-6s%-4s ",
1e37780e 2156 depth+1, "Match","Base","Ofs" );
3dab1dad
YO
2157
2158 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 2159 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 2160 if ( tmp ) {
6ad9a8ab 2161 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2162 colwidth,
538e84ed 2163 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2164 PL_colors[0], PL_colors[1],
2165 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2166 PERL_PV_ESCAPE_FIRSTCHAR
2167 )
ab3bbdeb 2168 );
3dab1dad
YO
2169 }
2170 }
1e37780e
YO
2171 Perl_re_printf( aTHX_ "\n");
2172 Perl_re_indentf( aTHX_ "State|-----------------------", depth+1);
3dab1dad
YO
2173
2174 for( state = 0 ; state < trie->uniquecharcount ; state++ )
6ad9a8ab
YO
2175 Perl_re_printf( aTHX_ "%.*s", colwidth, "--------");
2176 Perl_re_printf( aTHX_ "\n");
3dab1dad 2177
1e2e3d02 2178 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 2179 const U32 base = trie->states[ state ].trans.base;
3dab1dad 2180
147e3846 2181 Perl_re_indentf( aTHX_ "#%4" UVXf "|", depth+1, (UV)state);
3dab1dad
YO
2182
2183 if ( trie->states[ state ].wordnum ) {
1e37780e 2184 Perl_re_printf( aTHX_ " W%4X", trie->states[ state ].wordnum );
3dab1dad 2185 } else {
6ad9a8ab 2186 Perl_re_printf( aTHX_ "%6s", "" );
3dab1dad
YO
2187 }
2188
147e3846 2189 Perl_re_printf( aTHX_ " @%4" UVXf " ", (UV)base );
3dab1dad
YO
2190
2191 if ( base ) {
2192 U32 ofs = 0;
2193
2194 while( ( base + ofs < trie->uniquecharcount ) ||
2195 ( base + ofs - trie->uniquecharcount < trie->lasttrans
538e84ed
KW
2196 && trie->trans[ base + ofs - trie->uniquecharcount ].check
2197 != state))
3dab1dad
YO
2198 ofs++;
2199
147e3846 2200 Perl_re_printf( aTHX_ "+%2" UVXf "[ ", (UV)ofs);
3dab1dad
YO
2201
2202 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
538e84ed
KW
2203 if ( ( base + ofs >= trie->uniquecharcount )
2204 && ( base + ofs - trie->uniquecharcount
2205 < trie->lasttrans )
2206 && trie->trans[ base + ofs
2207 - trie->uniquecharcount ].check == state )
3dab1dad 2208 {
147e3846 2209 Perl_re_printf( aTHX_ "%*" UVXf, colwidth,
1e37780e
YO
2210 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next
2211 );
3dab1dad 2212 } else {
88f063b4 2213 Perl_re_printf( aTHX_ "%*s", colwidth," ." );
3dab1dad
YO
2214 }
2215 }
2216
6ad9a8ab 2217 Perl_re_printf( aTHX_ "]");
3dab1dad
YO
2218
2219 }
6ad9a8ab 2220 Perl_re_printf( aTHX_ "\n" );
3dab1dad 2221 }
6ad9a8ab 2222 Perl_re_indentf( aTHX_ "word_info N:(prev,len)=",
cb41e5d6 2223 depth);
2e64971a 2224 for (word=1; word <= trie->wordcount; word++) {
6ad9a8ab 2225 Perl_re_printf( aTHX_ " %d:(%d,%d)",
2e64971a
DM
2226 (int)word, (int)(trie->wordinfo[word].prev),
2227 (int)(trie->wordinfo[word].len));
2228 }
6ad9a8ab 2229 Perl_re_printf( aTHX_ "\n" );
538e84ed 2230}
3dab1dad 2231/*
3dab1dad 2232 Dumps a fully constructed but uncompressed trie in list form.
538e84ed 2233 List tries normally only are used for construction when the number of
3dab1dad
YO
2234 possible chars (trie->uniquecharcount) is very high.
2235 Used for debugging make_trie().
2236*/
2237STATIC void
55eed653 2238S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2239 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2240 U32 depth)
3dab1dad
YO
2241{
2242 U32 state;
ab3bbdeb 2243 SV *sv=sv_newmortal();
55eed653 2244 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2245 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2246
2247 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
2248
3dab1dad 2249 /* print out the table precompression. */
6ad9a8ab 2250 Perl_re_indentf( aTHX_ "State :Word | Transition Data\n",
cb41e5d6 2251 depth+1 );
6ad9a8ab 2252 Perl_re_indentf( aTHX_ "%s",
cb41e5d6 2253 depth+1, "------:-----+-----------------\n" );
538e84ed 2254
3dab1dad
YO
2255 for( state=1 ; state < next_alloc ; state ++ ) {
2256 U16 charid;
538e84ed 2257
147e3846 2258 Perl_re_indentf( aTHX_ " %4" UVXf " :",
cb41e5d6 2259 depth+1, (UV)state );
3dab1dad 2260 if ( ! trie->states[ state ].wordnum ) {
6ad9a8ab 2261 Perl_re_printf( aTHX_ "%5s| ","");
3dab1dad 2262 } else {
6ad9a8ab 2263 Perl_re_printf( aTHX_ "W%4x| ",
3dab1dad
YO
2264 trie->states[ state ].wordnum
2265 );
2266 }
2267 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
538e84ed 2268 SV ** const tmp = av_fetch( revcharmap,
88f063b4 2269 TRIE_LIST_ITEM(state, charid).forid, 0);
ab3bbdeb 2270 if ( tmp ) {
147e3846 2271 Perl_re_printf( aTHX_ "%*s:%3X=%4" UVXf " | ",
ab3bbdeb 2272 colwidth,
538e84ed
KW
2273 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
2274 colwidth,
2275 PL_colors[0], PL_colors[1],
2276 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2277 | PERL_PV_ESCAPE_FIRSTCHAR
ab3bbdeb 2278 ) ,
88f063b4
KW
2279 TRIE_LIST_ITEM(state, charid).forid,
2280 (UV)TRIE_LIST_ITEM(state, charid).newstate
1e2e3d02 2281 );
538e84ed 2282 if (!(charid % 10))
6ad9a8ab 2283 Perl_re_printf( aTHX_ "\n%*s| ",
664e119d 2284 (int)((depth * 2) + 14), "");
1e2e3d02 2285 }
ab3bbdeb 2286 }
6ad9a8ab 2287 Perl_re_printf( aTHX_ "\n");
3dab1dad 2288 }
538e84ed 2289}
3dab1dad
YO
2290
2291/*
3dab1dad 2292 Dumps a fully constructed but uncompressed trie in table form.
538e84ed
KW
2293 This is the normal DFA style state transition table, with a few
2294 twists to facilitate compression later.
3dab1dad
YO
2295 Used for debugging make_trie().
2296*/
2297STATIC void
55eed653 2298S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2299 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2300 U32 depth)
3dab1dad
YO
2301{
2302 U32 state;
2303 U16 charid;
ab3bbdeb 2304 SV *sv=sv_newmortal();
55eed653 2305 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2306 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2307
2308 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
538e84ed 2309
3dab1dad
YO
2310 /*
2311 print out the table precompression so that we can do a visual check
2312 that they are identical.
2313 */
538e84ed 2314
6ad9a8ab 2315 Perl_re_indentf( aTHX_ "Char : ", depth+1 );
3dab1dad
YO
2316
2317 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 2318 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 2319 if ( tmp ) {
6ad9a8ab 2320 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2321 colwidth,
538e84ed 2322 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2323 PL_colors[0], PL_colors[1],
2324 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2325 PERL_PV_ESCAPE_FIRSTCHAR
2326 )
ab3bbdeb 2327 );
3dab1dad
YO
2328 }
2329 }
2330
4aaafc03
YO
2331 Perl_re_printf( aTHX_ "\n");
2332 Perl_re_indentf( aTHX_ "State+-", depth+1 );
3dab1dad
YO
2333
2334 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
6ad9a8ab 2335 Perl_re_printf( aTHX_ "%.*s", colwidth,"--------");
3dab1dad
YO
2336 }
2337
6ad9a8ab 2338 Perl_re_printf( aTHX_ "\n" );
3dab1dad
YO
2339
2340 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
2341
147e3846 2342 Perl_re_indentf( aTHX_ "%4" UVXf " : ",
cb41e5d6 2343 depth+1,
3dab1dad
YO
2344 (UV)TRIE_NODENUM( state ) );
2345
2346 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
2347 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
2348 if (v)
147e3846 2349 Perl_re_printf( aTHX_ "%*" UVXf, colwidth, v );
ab3bbdeb 2350 else
6ad9a8ab 2351 Perl_re_printf( aTHX_ "%*s", colwidth, "." );
3dab1dad
YO
2352 }
2353 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
147e3846 2354 Perl_re_printf( aTHX_ " (%4" UVXf ")\n",
538e84ed 2355 (UV)trie->trans[ state ].check );
3dab1dad 2356 } else {
147e3846 2357 Perl_re_printf( aTHX_ " (%4" UVXf ") W%4X\n",
538e84ed 2358 (UV)trie->trans[ state ].check,
3dab1dad
YO
2359 trie->states[ TRIE_NODENUM( state ) ].wordnum );
2360 }
2361 }
07be1b83 2362}
3dab1dad
YO
2363
2364#endif
2365
2e64971a 2366
786e8c11
YO
2367/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
2368 startbranch: the first branch in the whole branch sequence
2369 first : start branch of sequence of branch-exact nodes.
2370 May be the same as startbranch
2371 last : Thing following the last branch.
2372 May be the same as tail.
2373 tail : item following the branch sequence
2374 count : words in the sequence
a4525e78 2375 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
786e8c11 2376 depth : indent depth
3dab1dad 2377
786e8c11 2378Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 2379
786e8c11
YO
2380A trie is an N'ary tree where the branches are determined by digital
2381decomposition of the key. IE, at the root node you look up the 1st character and
2382follow that branch repeat until you find the end of the branches. Nodes can be
2383marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 2384
786e8c11 2385 /he|she|his|hers/
72f13be8 2386
786e8c11
YO
2387would convert into the following structure. Numbers represent states, letters
2388following numbers represent valid transitions on the letter from that state, if
2389the number is in square brackets it represents an accepting state, otherwise it
2390will be in parenthesis.
07be1b83 2391
786e8c11
YO
2392 +-h->+-e->[3]-+-r->(8)-+-s->[9]
2393 | |
2394 | (2)
2395 | |
2396 (1) +-i->(6)-+-s->[7]
2397 |
2398 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 2399
786e8c11
YO
2400 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
2401
2402This shows that when matching against the string 'hers' we will begin at state 1
2403read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
2404then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
2405is also accepting. Thus we know that we can match both 'he' and 'hers' with a
2406single traverse. We store a mapping from accepting to state to which word was
2407matched, and then when we have multiple possibilities we try to complete the
b8fda935 2408rest of the regex in the order in which they occurred in the alternation.
786e8c11
YO
2409
2410The only prior NFA like behaviour that would be changed by the TRIE support is
2411the silent ignoring of duplicate alternations which are of the form:
2412
2413 / (DUPE|DUPE) X? (?{ ... }) Y /x
2414
4b714af6 2415Thus EVAL blocks following a trie may be called a different number of times with
786e8c11 2416and without the optimisation. With the optimisations dupes will be silently
486ec47a 2417ignored. This inconsistent behaviour of EVAL type nodes is well established as
786e8c11
YO
2418the following demonstrates:
2419
2420 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
2421
2422which prints out 'word' three times, but
2423
2424 'words'=~/(word|word|word)(?{ print $1 })S/
2425
2426which doesnt print it out at all. This is due to other optimisations kicking in.
2427
2428Example of what happens on a structural level:
2429
486ec47a 2430The regexp /(ac|ad|ab)+/ will produce the following debug output:
786e8c11
YO
2431
2432 1: CURLYM[1] {1,32767}(18)
2433 5: BRANCH(8)
2434 6: EXACT <ac>(16)
2435 8: BRANCH(11)
2436 9: EXACT <ad>(16)
2437 11: BRANCH(14)
2438 12: EXACT <ab>(16)
2439 16: SUCCEED(0)
2440 17: NOTHING(18)
2441 18: END(0)
2442
2443This would be optimizable with startbranch=5, first=5, last=16, tail=16
2444and should turn into:
2445
2446 1: CURLYM[1] {1,32767}(18)
2447 5: TRIE(16)
2448 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
2449 <ac>
2450 <ad>
2451 <ab>
2452 16: SUCCEED(0)
2453 17: NOTHING(18)
2454 18: END(0)
2455
2456Cases where tail != last would be like /(?foo|bar)baz/:
2457
2458 1: BRANCH(4)
2459 2: EXACT <foo>(8)
2460 4: BRANCH(7)
2461 5: EXACT <bar>(8)
2462 7: TAIL(8)
2463 8: EXACT <baz>(10)
2464 10: END(0)
2465
2466which would be optimizable with startbranch=1, first=1, last=7, tail=8
2467and would end up looking like:
2468
2469 1: TRIE(8)
2470 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
2471 <foo>
2472 <bar>
2473 7: TAIL(8)
2474 8: EXACT <baz>(10)
2475 10: END(0)
2476
c80e42f3 2477 d = uvchr_to_utf8_flags(d, uv, 0);
786e8c11
YO
2478
2479is the recommended Unicode-aware way of saying
2480
2481 *(d++) = uv;
2482*/
2483
fab2782b 2484#define TRIE_STORE_REVCHAR(val) \
786e8c11 2485 STMT_START { \
73031816 2486 if (UTF) { \
668fcfea 2487 SV *zlopp = newSV(UTF8_MAXBYTES); \
88c9ea1e 2488 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
c80e42f3 2489 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
73031816
NC
2490 SvCUR_set(zlopp, kapow - flrbbbbb); \
2491 SvPOK_on(zlopp); \
2492 SvUTF8_on(zlopp); \
2493 av_push(revcharmap, zlopp); \
2494 } else { \
fab2782b 2495 char ooooff = (char)val; \
73031816
NC
2496 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2497 } \
2498 } STMT_END
786e8c11 2499
914a25d5
KW
2500/* This gets the next character from the input, folding it if not already
2501 * folded. */
2502#define TRIE_READ_CHAR STMT_START { \
2503 wordlen++; \
2504 if ( UTF ) { \
2505 /* if it is UTF then it is either already folded, or does not need \
2506 * folding */ \
1c1d615a 2507 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
914a25d5
KW
2508 } \
2509 else if (folder == PL_fold_latin1) { \
7d006b13
KW
2510 /* This folder implies Unicode rules, which in the range expressible \
2511 * by not UTF is the lower case, with the two exceptions, one of \
2512 * which should have been taken care of before calling this */ \
2513 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2514 uvc = toLOWER_L1(*uc); \
2515 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2516 len = 1; \
914a25d5
KW
2517 } else { \
2518 /* raw data, will be folded later if needed */ \
2519 uvc = (U32)*uc; \
2520 len = 1; \
2521 } \
786e8c11
YO
2522} STMT_END
2523
2524
2525
2526#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2527 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
00195859 2528 U32 ging = TRIE_LIST_LEN( state ) * 2; \
f9003953 2529 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
00195859 2530 TRIE_LIST_LEN( state ) = ging; \
786e8c11
YO
2531 } \
2532 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2533 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2534 TRIE_LIST_CUR( state )++; \
2535} STMT_END
07be1b83 2536
786e8c11 2537#define TRIE_LIST_NEW(state) STMT_START { \
d09f14bf 2538 Newx( trie->states[ state ].trans.list, \
786e8c11
YO
2539 4, reg_trie_trans_le ); \
2540 TRIE_LIST_CUR( state ) = 1; \
2541 TRIE_LIST_LEN( state ) = 4; \
2542} STMT_END
07be1b83 2543
786e8c11
YO
2544#define TRIE_HANDLE_WORD(state) STMT_START { \
2545 U16 dupe= trie->states[ state ].wordnum; \
2546 regnode * const noper_next = regnext( noper ); \
2547 \
786e8c11
YO
2548 DEBUG_r({ \
2549 /* store the word for dumping */ \
2550 SV* tmp; \
2551 if (OP(noper) != NOTHING) \
740cce10 2552 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
786e8c11 2553 else \
740cce10 2554 tmp = newSVpvn_utf8( "", 0, UTF ); \
2b8b4781 2555 av_push( trie_words, tmp ); \
786e8c11
YO
2556 }); \
2557 \
2558 curword++; \
2e64971a
DM
2559 trie->wordinfo[curword].prev = 0; \
2560 trie->wordinfo[curword].len = wordlen; \
2561 trie->wordinfo[curword].accept = state; \
786e8c11
YO
2562 \
2563 if ( noper_next < tail ) { \
2564 if (!trie->jump) \
538e84ed
KW
2565 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2566 sizeof(U16) ); \
7f69552c 2567 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
2568 if (!jumper) \
2569 jumper = noper_next; \
2570 if (!nextbranch) \
2571 nextbranch= regnext(cur); \
2572 } \
2573 \
2574 if ( dupe ) { \
2e64971a
DM
2575 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2576 /* chain, so that when the bits of chain are later */\
2577 /* linked together, the dups appear in the chain */\
2578 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2579 trie->wordinfo[dupe].prev = curword; \
786e8c11
YO
2580 } else { \
2581 /* we haven't inserted this word yet. */ \
2582 trie->states[ state ].wordnum = curword; \
2583 } \
2584} STMT_END
07be1b83 2585
3dab1dad 2586
786e8c11
YO
2587#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2588 ( ( base + charid >= ucharcount \
2589 && base + charid < ubound \
2590 && state == trie->trans[ base - ucharcount + charid ].check \
2591 && trie->trans[ base - ucharcount + charid ].next ) \
2592 ? trie->trans[ base - ucharcount + charid ].next \
2593 : ( state==1 ? special : 0 ) \
2594 )
3dab1dad 2595
8bcafbf4
YO
2596#define TRIE_BITMAP_SET_FOLDED(trie, uvc, folder) \
2597STMT_START { \
2598 TRIE_BITMAP_SET(trie, uvc); \
2599 /* store the folded codepoint */ \
2600 if ( folder ) \
2601 TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); \
2602 \
2603 if ( !UTF ) { \
2604 /* store first byte of utf8 representation of */ \
2605 /* variant codepoints */ \
2606 if (! UVCHR_IS_INVARIANT(uvc)) { \
2607 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); \
2608 } \
2609 } \
2610} STMT_END
786e8c11
YO
2611#define MADE_TRIE 1
2612#define MADE_JUMP_TRIE 2
2613#define MADE_EXACT_TRIE 4
3dab1dad 2614
a3621e74 2615STATIC I32
538e84ed
KW
2616S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2617 regnode *first, regnode *last, regnode *tail,
2618 U32 word_count, U32 flags, U32 depth)
a3621e74
YO
2619{
2620 /* first pass, loop through and scan words */
2621 reg_trie_data *trie;
55eed653 2622 HV *widecharmap = NULL;
2b8b4781 2623 AV *revcharmap = newAV();
a3621e74 2624 regnode *cur;
a3621e74
YO
2625 STRLEN len = 0;
2626 UV uvc = 0;
2627 U16 curword = 0;
2628 U32 next_alloc = 0;
786e8c11
YO
2629 regnode *jumper = NULL;
2630 regnode *nextbranch = NULL;
7f69552c 2631 regnode *convert = NULL;
2e64971a 2632 U32 *prev_states; /* temp array mapping each state to previous one */
a3621e74 2633 /* we just use folder as a flag in utf8 */
1e696034 2634 const U8 * folder = NULL;
a3621e74 2635
3a611511
YO
2636 /* in the below add_data call we are storing either 'tu' or 'tuaa'
2637 * which stands for one trie structure, one hash, optionally followed
2638 * by two arrays */
2b8b4781 2639#ifdef DEBUGGING
3a611511 2640 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuaa"));
2b8b4781
NC
2641 AV *trie_words = NULL;
2642 /* along with revcharmap, this only used during construction but both are
2643 * useful during debugging so we store them in the struct when debugging.
8e11feef 2644 */
2b8b4781 2645#else
cf78de0b 2646 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
3dab1dad 2647 STRLEN trie_charcount=0;
3dab1dad 2648#endif
2b8b4781 2649 SV *re_trie_maxbuff;
a3621e74 2650 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2651
2652 PERL_ARGS_ASSERT_MAKE_TRIE;
72f13be8
YO
2653#ifndef DEBUGGING
2654 PERL_UNUSED_ARG(depth);
2655#endif
a3621e74 2656
1e696034 2657 switch (flags) {
f6b4b99d 2658 case EXACT: case EXACT_ONLY8: case EXACTL: break;
89829bb5 2659 case EXACTFAA:
0ea669f4 2660 case EXACTFUP:
a4525e78
KW
2661 case EXACTFU:
2662 case EXACTFLU8: folder = PL_fold_latin1; break;
1e696034 2663 case EXACTF: folder = PL_fold; break;
fab2782b 2664 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
1e696034
KW
2665 }
2666
c944940b 2667 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 2668 trie->refcount = 1;
3dab1dad 2669 trie->startstate = 1;
786e8c11 2670 trie->wordcount = word_count;
f8fc2ecf 2671 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 2672 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
f6b4b99d 2673 if (flags == EXACT || flags == EXACT_ONLY8 || flags == EXACTL)
c944940b 2674 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2e64971a
DM
2675 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2676 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2677
a3621e74 2678 DEBUG_r({
2b8b4781 2679 trie_words = newAV();
a3621e74 2680 });
a3621e74 2681
0111c4fd 2682 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
316ebaf2 2683 assert(re_trie_maxbuff);
a3621e74 2684 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 2685 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 2686 }
df826430 2687 DEBUG_TRIE_COMPILE_r({
6ad9a8ab 2688 Perl_re_indentf( aTHX_
cb41e5d6
YO
2689 "make_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2690 depth+1,
88f063b4 2691 REG_NODE_NUM(startbranch), REG_NODE_NUM(first),
538e84ed 2692 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
3dab1dad 2693 });
538e84ed 2694
7f69552c
YO
2695 /* Find the node we are going to overwrite */
2696 if ( first == startbranch && OP( last ) != BRANCH ) {
2697 /* whole branch chain */
2698 convert = first;
2699 } else {
2700 /* branch sub-chain */
2701 convert = NEXTOPER( first );
2702 }
538e84ed 2703
a3621e74
YO
2704 /* -- First loop and Setup --
2705
2706 We first traverse the branches and scan each word to determine if it
2707 contains widechars, and how many unique chars there are, this is
2708 important as we have to build a table with at least as many columns as we
2709 have unique chars.
2710
2711 We use an array of integers to represent the character codes 0..255
538e84ed
KW
2712 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2713 the native representation of the character value as the key and IV's for
2714 the coded index.
a3621e74
YO
2715
2716 *TODO* If we keep track of how many times each character is used we can
2717 remap the columns so that the table compression later on is more
3b753521 2718 efficient in terms of memory by ensuring the most common value is in the
a3621e74
YO
2719 middle and the least common are on the outside. IMO this would be better
2720 than a most to least common mapping as theres a decent chance the most
2721 common letter will share a node with the least common, meaning the node
486ec47a 2722 will not be compressible. With a middle is most common approach the worst
a3621e74
YO
2723 case is when we have the least common nodes twice.
2724
2725 */
2726
a3621e74 2727 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
df826430 2728 regnode *noper = NEXTOPER( cur );
944e05e3
YO
2729 const U8 *uc;
2730 const U8 *e;
bc031a7d 2731 int foldlen = 0;
07be1b83 2732 U32 wordlen = 0; /* required init */
bc031a7d
KW
2733 STRLEN minchars = 0;
2734 STRLEN maxchars = 0;
538e84ed
KW
2735 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2736 bitmap?*/
a3621e74 2737
3dab1dad 2738 if (OP(noper) == NOTHING) {
20ed8c88
YO
2739 /* skip past a NOTHING at the start of an alternation
2740 * eg, /(?:)a|(?:b)/ should be the same as /a|b/
2741 */
df826430 2742 regnode *noper_next= regnext(noper);
944e05e3
YO
2743 if (noper_next < tail)
2744 noper= noper_next;
2745 }
2746
f6b4b99d
KW
2747 if ( noper < tail
2748 && ( OP(noper) == flags
2749 || (flags == EXACT && OP(noper) == EXACT_ONLY8)
a9f8c7ac 2750 || (flags == EXACTFU && ( OP(noper) == EXACTFU_ONLY8
0ea669f4 2751 || OP(noper) == EXACTFUP))))
f6b4b99d 2752 {
944e05e3
YO
2753 uc= (U8*)STRING(noper);
2754 e= uc + STR_LEN(noper);
2755 } else {
2756 trie->minlen= 0;
2757 continue;
3dab1dad 2758 }
df826430 2759
944e05e3 2760
fab2782b 2761 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
02daf0ab
YO
2762 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2763 regardless of encoding */
0ea669f4 2764 if (OP( noper ) == EXACTFUP) {
fab2782b 2765 /* false positives are ok, so just set this */
0dc4a61d 2766 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
fab2782b
YO
2767 }
2768 }
dca5fc4c 2769
bc031a7d
KW
2770 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2771 branch */
3dab1dad 2772 TRIE_CHARCOUNT(trie)++;
a3621e74 2773 TRIE_READ_CHAR;
645de4ce 2774
bc031a7d
KW
2775 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2776 * is in effect. Under /i, this character can match itself, or
2777 * anything that folds to it. If not under /i, it can match just
2778 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2779 * all fold to k, and all are single characters. But some folds
2780 * expand to more than one character, so for example LATIN SMALL
2781 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2782 * the string beginning at 'uc' is 'ffi', it could be matched by
2783 * three characters, or just by the one ligature character. (It
2784 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2785 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2786 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2787 * match.) The trie needs to know the minimum and maximum number
2788 * of characters that could match so that it can use size alone to
2789 * quickly reject many match attempts. The max is simple: it is
2790 * the number of folded characters in this branch (since a fold is
2791 * never shorter than what folds to it. */
2792
2793 maxchars++;
2794
2795 /* And the min is equal to the max if not under /i (indicated by
2796 * 'folder' being NULL), or there are no multi-character folds. If
2797 * there is a multi-character fold, the min is incremented just
2798 * once, for the character that folds to the sequence. Each
2799 * character in the sequence needs to be added to the list below of
2800 * characters in the trie, but we count only the first towards the
2801 * min number of characters needed. This is done through the
2802 * variable 'foldlen', which is returned by the macros that look
2803 * for these sequences as the number of bytes the sequence
2804 * occupies. Each time through the loop, we decrement 'foldlen' by
2805 * how many bytes the current char occupies. Only when it reaches
2806 * 0 do we increment 'minchars' or look for another multi-character
2807 * sequence. */
2808 if (folder == NULL) {
2809 minchars++;
2810 }
2811 else if (foldlen > 0) {
2812 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
645de4ce
KW
2813 }
2814 else {
bc031a7d
KW
2815 minchars++;
2816
2817 /* See if *uc is the beginning of a multi-character fold. If
2818 * so, we decrement the length remaining to look at, to account
2819 * for the current character this iteration. (We can use 'uc'
2820 * instead of the fold returned by TRIE_READ_CHAR because for
2821 * non-UTF, the latin1_safe macro is smart enough to account
2822 * for all the unfolded characters, and because for UTF, the
2823 * string will already have been folded earlier in the
2824 * compilation process */
2825 if (UTF) {
2826 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2827 foldlen -= UTF8SKIP(uc);
645de4ce
KW
2828 }
2829 }
bc031a7d
KW
2830 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2831 foldlen--;
2832 }
645de4ce 2833 }
bc031a7d
KW
2834
2835 /* The current character (and any potential folds) should be added
2836 * to the possible matching characters for this position in this
2837 * branch */
a3621e74 2838 if ( uvc < 256 ) {
fab2782b
YO
2839 if ( folder ) {
2840 U8 folded= folder[ (U8) uvc ];
2841 if ( !trie->charmap[ folded ] ) {
2842 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2843 TRIE_STORE_REVCHAR( folded );
2844 }
2845 }
a3621e74
YO
2846 if ( !trie->charmap[ uvc ] ) {
2847 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
fab2782b 2848 TRIE_STORE_REVCHAR( uvc );
a3621e74 2849 }
02daf0ab 2850 if ( set_bit ) {
62012aee
KW
2851 /* store the codepoint in the bitmap, and its folded
2852 * equivalent. */
8bcafbf4 2853 TRIE_BITMAP_SET_FOLDED(trie, uvc, folder);
02daf0ab
YO
2854 set_bit = 0; /* We've done our bit :-) */
2855 }
a3621e74 2856 } else {
bc031a7d
KW
2857
2858 /* XXX We could come up with the list of code points that fold
2859 * to this using PL_utf8_foldclosures, except not for
2860 * multi-char folds, as there may be multiple combinations
2861 * there that could work, which needs to wait until runtime to
2862 * resolve (The comment about LIGATURE FFI above is such an
2863 * example */
2864
a3621e74 2865 SV** svpp;
55eed653
NC
2866 if ( !widecharmap )
2867 widecharmap = newHV();
a3621e74 2868
55eed653 2869 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
2870
2871 if ( !svpp )
147e3846 2872 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%" UVXf, uvc );
a3621e74
YO
2873
2874 if ( !SvTRUE( *svpp ) ) {
2875 sv_setiv( *svpp, ++trie->uniquecharcount );
fab2782b 2876 TRIE_STORE_REVCHAR(uvc);
a3621e74
YO
2877 }
2878 }
bc031a7d
KW
2879 } /* end loop through characters in this branch of the trie */
2880
2881 /* We take the min and max for this branch and combine to find the min
2882 * and max for all branches processed so far */
3dab1dad 2883 if( cur == first ) {
bc031a7d
KW
2884 trie->minlen = minchars;
2885 trie->maxlen = maxchars;
2886 } else if (minchars < trie->minlen) {
2887 trie->minlen = minchars;
2888 } else if (maxchars > trie->maxlen) {
2889 trie->maxlen = maxchars;
fab2782b 2890 }
a3621e74
YO
2891 } /* end first pass */
2892 DEBUG_TRIE_COMPILE_r(
6ad9a8ab 2893 Perl_re_indentf( aTHX_
cb41e5d6
YO
2894 "TRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2895 depth+1,
55eed653 2896 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
2897 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2898 (int)trie->minlen, (int)trie->maxlen )
a3621e74 2899 );
a3621e74
YO
2900
2901 /*
2902 We now know what we are dealing with in terms of unique chars and
2903 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
2904 representation using a flat table will take. If it's over a reasonable
2905 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
2906 conservative but potentially much slower representation using an array
2907 of lists.
2908
2909 At the end we convert both representations into the same compressed
2910 form that will be used in regexec.c for matching with. The latter
2911 is a form that cannot be used to construct with but has memory
2912 properties similar to the list form and access properties similar
2913 to the table form making it both suitable for fast searches and
2914 small enough that its feasable to store for the duration of a program.
2915
2916 See the comment in the code where the compressed table is produced
2917 inplace from the flat tabe representation for an explanation of how
2918 the compression works.
2919
2920 */
2921
2922
2e64971a
DM
2923 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2924 prev_states[1] = 0;
2925
538e84ed
KW
2926 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2927 > SvIV(re_trie_maxbuff) )
2928 {
a3621e74
YO
2929 /*
2930 Second Pass -- Array Of Lists Representation
2931
2932 Each state will be represented by a list of charid:state records
2933 (reg_trie_trans_le) the first such element holds the CUR and LEN
2934 points of the allocated array. (See defines above).
2935
2936 We build the initial structure using the lists, and then convert
2937 it into the compressed table form which allows faster lookups
2938 (but cant be modified once converted).
a3621e74
YO
2939 */
2940
a3621e74
YO
2941 STRLEN transcount = 1;
2942
6ad9a8ab 2943 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using list compiler\n",
cb41e5d6 2944 depth+1));
686b73d4 2945
c944940b
JH
2946 trie->states = (reg_trie_state *)
2947 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2948 sizeof(reg_trie_state) );
a3621e74
YO
2949 TRIE_LIST_NEW(1);
2950 next_alloc = 2;
2951
2952 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2953
df826430 2954 regnode *noper = NEXTOPER( cur );
c445ea15
AL
2955 U32 state = 1; /* required init */
2956 U16 charid = 0; /* sanity init */
07be1b83 2957 U32 wordlen = 0; /* required init */
c445ea15 2958
df826430
YO
2959 if (OP(noper) == NOTHING) {
2960 regnode *noper_next= regnext(noper);
944e05e3
YO
2961 if (noper_next < tail)
2962 noper= noper_next;
df826430
YO
2963 }
2964
f6b4b99d
KW
2965 if ( noper < tail
2966 && ( OP(noper) == flags
2967 || (flags == EXACT && OP(noper) == EXACT_ONLY8)
a9f8c7ac 2968 || (flags == EXACTFU && ( OP(noper) == EXACTFU_ONLY8
0ea669f4 2969 || OP(noper) == EXACTFUP))))
f6b4b99d 2970 {
944e05e3
YO
2971 const U8 *uc= (U8*)STRING(noper);
2972 const U8 *e= uc + STR_LEN(noper);
2973
786e8c11 2974 for ( ; uc < e ; uc += len ) {
c445ea15 2975
786e8c11 2976 TRIE_READ_CHAR;
c445ea15 2977
786e8c11
YO
2978 if ( uvc < 256 ) {
2979 charid = trie->charmap[ uvc ];
c445ea15 2980 } else {
538e84ed
KW
2981 SV** const svpp = hv_fetch( widecharmap,
2982 (char*)&uvc,
2983 sizeof( UV ),
2984 0);
786e8c11
YO
2985 if ( !svpp ) {
2986 charid = 0;
2987 } else {
2988 charid=(U16)SvIV( *svpp );
2989 }
c445ea15 2990 }
538e84ed
KW
2991 /* charid is now 0 if we dont know the char read, or
2992 * nonzero if we do */
786e8c11 2993 if ( charid ) {
a3621e74 2994
786e8c11
YO
2995 U16 check;
2996 U32 newstate = 0;
a3621e74 2997
786e8c11
YO
2998 charid--;
2999 if ( !trie->states[ state ].trans.list ) {
3000 TRIE_LIST_NEW( state );
c445ea15 3001 }
538e84ed
KW
3002 for ( check = 1;
3003 check <= TRIE_LIST_USED( state );
3004 check++ )
3005 {
3006 if ( TRIE_LIST_ITEM( state, check ).forid
3007 == charid )
3008 {
786e8c11
YO
3009 newstate = TRIE_LIST_ITEM( state, check ).newstate;
3010 break;
3011 }
3012 }
3013 if ( ! newstate ) {
3014 newstate = next_alloc++;
2e64971a 3015 prev_states[newstate] = state;
786e8c11
YO
3016 TRIE_LIST_PUSH( state, charid, newstate );
3017 transcount++;
3018 }
3019 state = newstate;
3020 } else {
147e3846 3021 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
c445ea15 3022 }
a28509cc 3023 }
c445ea15 3024 }
3dab1dad 3025 TRIE_HANDLE_WORD(state);
a3621e74
YO
3026
3027 } /* end second pass */
3028
1e2e3d02 3029 /* next alloc is the NEXT state to be allocated */
538e84ed 3030 trie->statecount = next_alloc;
c944940b
JH
3031 trie->states = (reg_trie_state *)
3032 PerlMemShared_realloc( trie->states,
3033 next_alloc
3034 * sizeof(reg_trie_state) );
a3621e74 3035
3dab1dad 3036 /* and now dump it out before we compress it */
2b8b4781
NC
3037 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
3038 revcharmap, next_alloc,
3039 depth+1)
1e2e3d02 3040 );
a3621e74 3041
c944940b
JH
3042 trie->trans = (reg_trie_trans *)
3043 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
3044 {
3045 U32 state;
a3621e74
YO
3046 U32 tp = 0;
3047 U32 zp = 0;
3048
3049
3050 for( state=1 ; state < next_alloc ; state ++ ) {
3051 U32 base=0;
3052
3053 /*
3054 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3055 Perl_re_printf( aTHX_ "tp: %d zp: %d ",tp,zp)
a3621e74
YO
3056 );
3057 */
3058
3059 if (trie->states[state].trans.list) {
3060 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
3061 U16 maxid=minid;
a28509cc 3062 U16 idx;
a3621e74
YO
3063
3064 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
3065 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
3066 if ( forid < minid ) {
3067 minid=forid;
3068 } else if ( forid > maxid ) {
3069 maxid=forid;
3070 }
a3621e74
YO
3071 }
3072 if ( transcount < tp + maxid - minid + 1) {
3073 transcount *= 2;
c944940b
JH
3074 trie->trans = (reg_trie_trans *)
3075 PerlMemShared_realloc( trie->trans,
446bd890
NC
3076 transcount
3077 * sizeof(reg_trie_trans) );
538e84ed
KW
3078 Zero( trie->trans + (transcount / 2),
3079 transcount / 2,
3080 reg_trie_trans );
a3621e74
YO
3081 }
3082 base = trie->uniquecharcount + tp - minid;
3083 if ( maxid == minid ) {
3084 U32 set = 0;
3085 for ( ; zp < tp ; zp++ ) {
3086 if ( ! trie->trans[ zp ].next ) {
3087 base = trie->uniquecharcount + zp - minid;
538e84ed
KW
3088 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
3089 1).newstate;
a3621e74
YO
3090 trie->trans[ zp ].check = state;
3091 set = 1;
3092 break;
3093 }
3094 }
3095 if ( !set ) {
538e84ed
KW
3096 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
3097 1).newstate;
a3621e74
YO
3098 trie->trans[ tp ].check = state;
3099 tp++;
3100 zp = tp;
3101 }
3102 } else {
3103 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
538e84ed
KW
3104 const U32 tid = base
3105 - trie->uniquecharcount
3106 + TRIE_LIST_ITEM( state, idx ).forid;
3107 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
3108 idx ).newstate;
a3621e74
YO
3109 trie->trans[ tid ].check = state;
3110 }
3111 tp += ( maxid - minid + 1 );
3112 }
3113 Safefree(trie->states[ state ].trans.list);
3114 }
3115 /*
3116 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3117 Perl_re_printf( aTHX_ " base: %d\n",base);
a3621e74
YO
3118 );
3119 */
3120 trie->states[ state ].trans.base=base;
3121 }
cc601c31 3122 trie->lasttrans = tp + 1;
a3621e74
YO
3123 }
3124 } else {
3125 /*
3126 Second Pass -- Flat Table Representation.
3127
b423522f
KW
3128 we dont use the 0 slot of either trans[] or states[] so we add 1 to
3129 each. We know that we will need Charcount+1 trans at most to store
3130 the data (one row per char at worst case) So we preallocate both
3131 structures assuming worst case.
a3621e74
YO
3132
3133 We then construct the trie using only the .next slots of the entry
3134 structs.
3135
b423522f
KW
3136 We use the .check field of the first entry of the node temporarily
3137 to make compression both faster and easier by keeping track of how
3138 many non zero fields are in the node.
a3621e74
YO
3139
3140 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
3141 transition.
3142
b423522f
KW
3143 There are two terms at use here: state as a TRIE_NODEIDX() which is
3144 a number representing the first entry of the node, and state as a
3145 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
3146 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
3147 if there are 2 entrys per node. eg:
a3621e74
YO
3148
3149 A B A B
3150 1. 2 4 1. 3 7
3151 2. 0 3 3. 0 5
3152 3. 0 0 5. 0 0
3153 4. 0 0 7. 0 0
3154
b423522f
KW
3155 The table is internally in the right hand, idx form. However as we
3156 also have to deal with the states array which is indexed by nodenum
3157 we have to use TRIE_NODENUM() to convert.
a3621e74
YO
3158
3159 */
6ad9a8ab 3160 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using table compiler\n",
cb41e5d6 3161 depth+1));
3dab1dad 3162
c944940b
JH
3163 trie->trans = (reg_trie_trans *)
3164 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
3165 * trie->uniquecharcount + 1,
3166 sizeof(reg_trie_trans) );
3167 trie->states = (reg_trie_state *)
3168 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
3169 sizeof(reg_trie_state) );
a3621e74
YO
3170 next_alloc = trie->uniquecharcount + 1;
3171
3dab1dad 3172
a3621e74
YO
3173 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
3174
df826430 3175 regnode *noper = NEXTOPER( cur );
a3621e74
YO
3176
3177 U32 state = 1; /* required init */
3178
3179 U16 charid = 0; /* sanity init */
3180 U32 accept_state = 0; /* sanity init */
a3621e74 3181
07be1b83 3182 U32 wordlen = 0; /* required init */
a3621e74 3183
df826430
YO
3184 if (OP(noper) == NOTHING) {
3185 regnode *noper_next= regnext(noper);
944e05e3
YO
3186 if (noper_next < tail)
3187 noper= noper_next;
df826430 3188 }
fab2782b 3189
f6b4b99d
KW
3190 if ( noper < tail
3191 && ( OP(noper) == flags
3192 || (flags == EXACT && OP(noper) == EXACT_ONLY8)
a9f8c7ac 3193 || (flags == EXACTFU && ( OP(noper) == EXACTFU_ONLY8
0ea669f4 3194 || OP(noper) == EXACTFUP))))
f6b4b99d 3195 {
944e05e3
YO
3196 const U8 *uc= (U8*)STRING(noper);
3197 const U8 *e= uc + STR_LEN(noper);
3198
786e8c11 3199 for ( ; uc < e ; uc += len ) {
a3621e74 3200
786e8c11 3201 TRIE_READ_CHAR;
a3621e74 3202
786e8c11
YO
3203 if ( uvc < 256 ) {
3204 charid = trie->charmap[ uvc ];
3205 } else {
538e84ed
KW
3206 SV* const * const svpp = hv_fetch( widecharmap,
3207 (char*)&uvc,
3208 sizeof( UV ),
3209 0);
786e8c11 3210 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 3211 }
786e8c11
YO
3212 if ( charid ) {
3213 charid--;
3214 if ( !trie->trans[ state + charid ].next ) {
3215 trie->trans[ state + charid ].next = next_alloc;
3216 trie->trans[ state ].check++;
2e64971a
DM
3217 prev_states[TRIE_NODENUM(next_alloc)]
3218 = TRIE_NODENUM(state);
786e8c11
YO
3219 next_alloc += trie->uniquecharcount;
3220 }
3221 state = trie->trans[ state + charid ].next;
3222 } else {
147e3846 3223 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
786e8c11 3224 }
538e84ed
KW
3225 /* charid is now 0 if we dont know the char read, or
3226 * nonzero if we do */
a3621e74 3227 }
a3621e74 3228 }
3dab1dad
YO
3229 accept_state = TRIE_NODENUM( state );
3230 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
3231
3232 } /* end second pass */
3233
3dab1dad 3234 /* and now dump it out before we compress it */
2b8b4781
NC
3235 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
3236 revcharmap,
3237 next_alloc, depth+1));
a3621e74 3238
a3621e74
YO
3239 {
3240 /*
3241 * Inplace compress the table.*
3242
3243 For sparse data sets the table constructed by the trie algorithm will
3244 be mostly 0/FAIL transitions or to put it another way mostly empty.
3245 (Note that leaf nodes will not contain any transitions.)
3246
3247 This algorithm compresses the tables by eliminating most such
3248 transitions, at the cost of a modest bit of extra work during lookup:
3249
3250 - Each states[] entry contains a .base field which indicates the
3251 index in the state[] array wheres its transition data is stored.
3252
3b753521 3253 - If .base is 0 there are no valid transitions from that node.
a3621e74
YO
3254
3255 - If .base is nonzero then charid is added to it to find an entry in
3256 the trans array.
3257
3258 -If trans[states[state].base+charid].check!=state then the
3259 transition is taken to be a 0/Fail transition. Thus if there are fail
3260 transitions at the front of the node then the .base offset will point
3261 somewhere inside the previous nodes data (or maybe even into a node
3262 even earlier), but the .check field determines if the transition is
3263 valid.
3264
786e8c11 3265 XXX - wrong maybe?
a3621e74 3266 The following process inplace converts the table to the compressed
3b753521 3267 table: We first do not compress the root node 1,and mark all its
a3621e74 3268 .check pointers as 1 and set its .base pointer as 1 as well. This
3b753521
FN
3269 allows us to do a DFA construction from the compressed table later,
3270 and ensures that any .base pointers we calculate later are greater
3271 than 0.
a3621e74
YO
3272
3273 - We set 'pos' to indicate the first entry of the second node.
3274
3275 - We then iterate over the columns of the node, finding the first and
3276 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
3277 and set the .check pointers accordingly, and advance pos
3278 appropriately and repreat for the next node. Note that when we copy
3279 the next pointers we have to convert them from the original
3280 NODEIDX form to NODENUM form as the former is not valid post
3281 compression.
3282
3283 - If a node has no transitions used we mark its base as 0 and do not
3284 advance the pos pointer.
3285
3286 - If a node only has one transition we use a second pointer into the
3287 structure to fill in allocated fail transitions from other states.
3288 This pointer is independent of the main pointer and scans forward
3289 looking for null transitions that are allocated to a state. When it
3290 finds one it writes the single transition into the "hole". If the
786e8c11 3291 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
3292
3293 - Once compressed we can Renew/realloc the structures to release the
3294 excess space.
3295
3296 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
3297 specifically Fig 3.47 and the associated pseudocode.
3298
3299 demq
3300 */
a3b680e6 3301 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 3302 U32 state, charid;
a3621e74 3303 U32 pos = 0, zp=0;
1e2e3d02 3304 trie->statecount = laststate;
a3621e74
YO
3305
3306 for ( state = 1 ; state < laststate ; state++ ) {
3307 U8 flag = 0;
a28509cc
AL
3308 const U32 stateidx = TRIE_NODEIDX( state );
3309 const U32 o_used = trie->trans[ stateidx ].check;
3310 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
3311 trie->trans[ stateidx ].check = 0;
3312
538e84ed
KW
3313 for ( charid = 0;
3314 used && charid < trie->uniquecharcount;
3315 charid++ )
3316 {
a3621e74
YO
3317 if ( flag || trie->trans[ stateidx + charid ].next ) {
3318 if ( trie->trans[ stateidx + charid ].next ) {
3319 if (o_used == 1) {
3320 for ( ; zp < pos ; zp++ ) {
3321 if ( ! trie->trans[ zp ].next ) {
3322 break;
3323 }
3324 }
538e84ed
KW
3325 trie->states[ state ].trans.base
3326 = zp
3327 + trie->uniquecharcount
3328 - charid ;
3329 trie->trans[ zp ].next
3330 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
3331 + charid ].next );
a3621e74
YO
3332 trie->trans[ zp ].check = state;
3333 if ( ++zp > pos ) pos = zp;
3334 break;
3335 }
3336 used--;
3337 }
3338 if ( !flag ) {
3339 flag = 1;
538e84ed
KW
3340 trie->states[ state ].trans.base
3341 = pos + trie->uniquecharcount - charid ;
a3621e74 3342 }
538e84ed
KW
3343 trie->trans[ pos ].next
3344 = SAFE_TRIE_NODENUM(
3345 trie->trans[ stateidx + charid ].next );
a3621e74
YO
3346 trie->trans[ pos ].check = state;
3347 pos++;
3348 }
3349 }
3350 }
cc601c31 3351 trie->lasttrans = pos + 1;
c944940b
JH
3352 trie->states = (reg_trie_state *)
3353 PerlMemShared_realloc( trie->states, laststate
3354 * sizeof(reg_trie_state) );
a3621e74 3355 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3356 Perl_re_indentf( aTHX_ "Alloc: %d Orig: %" IVdf " elements, Final:%" IVdf ". Savings of %%%5.2f\n",
cb41e5d6 3357 depth+1,
538e84ed
KW
3358 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
3359 + 1 ),
3360 (IV)next_alloc,
3361 (IV)pos,
3362 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
a3621e74
YO
3363 );
3364
3365 } /* end table compress */
3366 }
1e2e3d02 3367 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3368 Perl_re_indentf( aTHX_ "Statecount:%" UVxf " Lasttrans:%" UVxf "\n",
cb41e5d6 3369 depth+1,
1e2e3d02
YO
3370 (UV)trie->statecount,
3371 (UV)trie->lasttrans)
3372 );
cc601c31 3373 /* resize the trans array to remove unused space */
c944940b
JH
3374 trie->trans = (reg_trie_trans *)
3375 PerlMemShared_realloc( trie->trans, trie->lasttrans
3376 * sizeof(reg_trie_trans) );
a3621e74 3377
538e84ed 3378 { /* Modify the program and insert the new TRIE node */
3dab1dad
YO
3379 U8 nodetype =(U8)(flags & 0xFF);
3380 char *str=NULL;
538e84ed 3381
07be1b83 3382#ifdef DEBUGGING
e62cc96a 3383 regnode *optimize = NULL;
7122b237
YO
3384#ifdef RE_TRACK_PATTERN_OFFSETS
3385
b57a0404
JH
3386 U32 mjd_offset = 0;
3387 U32 mjd_nodelen = 0;
7122b237
YO
3388#endif /* RE_TRACK_PATTERN_OFFSETS */
3389#endif /* DEBUGGING */
a3621e74 3390 /*
3dab1dad
YO
3391 This means we convert either the first branch or the first Exact,
3392 depending on whether the thing following (in 'last') is a branch
3393 or not and whther first is the startbranch (ie is it a sub part of
3394 the alternation or is it the whole thing.)
3b753521 3395 Assuming its a sub part we convert the EXACT otherwise we convert
3dab1dad 3396 the whole branch sequence, including the first.
a3621e74 3397 */
3dab1dad 3398 /* Find the node we are going to overwrite */
7f69552c 3399 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 3400 /* branch sub-chain */
3dab1dad 3401 NEXT_OFF( first ) = (U16)(last - first);
7122b237 3402#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
3403 DEBUG_r({
3404 mjd_offset= Node_Offset((convert));
3405 mjd_nodelen= Node_Length((convert));
3406 });
7122b237 3407#endif
7f69552c 3408 /* whole branch chain */
7122b237
YO
3409 }
3410#ifdef RE_TRACK_PATTERN_OFFSETS
3411 else {
7f69552c
YO
3412 DEBUG_r({
3413 const regnode *nop = NEXTOPER( convert );
3414 mjd_offset= Node_Offset((nop));
3415 mjd_nodelen= Node_Length((nop));
3416 });
07be1b83
YO
3417 }
3418 DEBUG_OPTIMISE_r(
147e3846 3419 Perl_re_indentf( aTHX_ "MJD offset:%" UVuf " MJD length:%" UVuf "\n",
cb41e5d6 3420 depth+1,
786e8c11 3421 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 3422 );
7122b237 3423#endif
538e84ed 3424 /* But first we check to see if there is a common prefix we can
3dab1dad
YO
3425 split out as an EXACT and put in front of the TRIE node. */
3426 trie->startstate= 1;
55eed653 3427 if ( trie->bitmap && !widecharmap && !trie->jump ) {
5ee57374
YO
3428 /* we want to find the first state that has more than
3429 * one transition, if that state is not the first state
3430 * then we have a common prefix which we can remove.
3431