This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: Use name consistently
[perl5.git] / regcomp.c
CommitLineData
a0d0e21e
LW
1/* regcomp.c
2 */
3
4/*
4ac71550
TC
5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
6 *
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
a0d0e21e
LW
8 */
9
61296642
DM
10/* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
166f8a29 12 * a regular expression.
e4a054ea
DM
13 *
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
166f8a29
DM
18 */
19
a687059c
LW
20/* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
22 */
23
24/* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
27 */
28
e50aee73 29/* The names of the functions have been changed from regcomp and
3b753521 30 * regexec to pregcomp and pregexec in order to avoid conflicts
e50aee73
AD
31 * with the POSIX routines of the same names.
32*/
33
b9d5759e 34#ifdef PERL_EXT_RE_BUILD
54df2634 35#include "re_top.h"
b81d288d 36#endif
56953603 37
a687059c 38/*
e50aee73 39 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
40 *
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
43 *
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
47 *
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
50 * from defects in it.
51 *
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
54 *
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
57 *
58 *
59 **** Alterations to Henry's code are...
60 ****
4bb101f2 61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
1129b882
NC
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
a687059c 64 ****
9ef589d8
LW
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
67
a687059c
LW
68 *
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
72 */
73#include "EXTERN.h"
864dbfa3 74#define PERL_IN_REGCOMP_C
a687059c 75#include "perl.h"
d06ea78c 76
acfe0abc 77#ifndef PERL_IN_XSUB_RE
d06ea78c
GS
78# include "INTERN.h"
79#endif
c277df42
IZ
80
81#define REG_COMP_C
54df2634
NC
82#ifdef PERL_IN_XSUB_RE
83# include "re_comp.h"
afda64e8 84EXTERN_C const struct regexp_engine my_reg_engine;
54df2634
NC
85#else
86# include "regcomp.h"
87#endif
a687059c 88
f7e03a10 89#include "dquote_inline.h"
b992490d 90#include "invlist_inline.h"
1b0f46bf 91#include "unicode_constants.h"
04e98a4d 92
538e84ed
KW
93#define HAS_NONLATIN1_FOLD_CLOSURE(i) \
94 _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
f12c0118
KW
95#define HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(i) \
96 _HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
26faadbd 97#define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
2c61f163 98#define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
94dc5c2d 99
a687059c
LW
100#ifndef STATIC
101#define STATIC static
102#endif
103
3f910e62
YO
104/* this is a chain of data about sub patterns we are processing that
105 need to be handled separately/specially in study_chunk. Its so
106 we can simulate recursion without losing state. */
107struct scan_frame;
108typedef struct scan_frame {
109 regnode *last_regnode; /* last node to process in this frame */
110 regnode *next_regnode; /* next node to process when last is reached */
111 U32 prev_recursed_depth;
112 I32 stopparen; /* what stopparen do we use */
3f910e62
YO
113
114 struct scan_frame *this_prev_frame; /* this previous frame */
115 struct scan_frame *prev_frame; /* previous frame */
116 struct scan_frame *next_frame; /* next frame */
117} scan_frame;
118
f4ae5a27
KW
119/* Certain characters are output as a sequence with the first being a
120 * backslash. */
eb7486de 121#define isBACKSLASHED_PUNCT(c) strchr("-[]\\^", c)
f4ae5a27
KW
122
123
09b2b2e6 124struct RExC_state_t {
514a91f1
DM
125 U32 flags; /* RXf_* are we folding, multilining? */
126 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
830247a4 127 char *precomp; /* uncompiled string. */
711b303b 128 char *precomp_end; /* pointer to end of uncompiled string. */
288b8c02 129 REGEXP *rx_sv; /* The SV that is the regexp. */
f8fc2ecf 130 regexp *rx; /* perl core regexp structure */
538e84ed
KW
131 regexp_internal *rxi; /* internal data for regexp object
132 pprivate field */
fac92740 133 char *start; /* Start of input for compile */
830247a4
IZ
134 char *end; /* End of input for compile */
135 char *parse; /* Input-scan pointer. */
51684e3c
KW
136 char *copy_start; /* start of copy of input within
137 constructed parse string */
138 char *copy_start_in_input; /* Position in input string
139 corresponding to copy_start */
ea3daa5d 140 SSize_t whilem_seen; /* number of WHILEM in this expr */
fac92740 141 regnode *emit_start; /* Start of emitted-code area */
538e84ed
KW
142 regnode *emit_bound; /* First regnode outside of the
143 allocated space */
f55b7b26 144 regnode_offset emit; /* Code-emit pointer */
830247a4
IZ
145 I32 naughty; /* How bad is this pattern? */
146 I32 sawback; /* Did we see \1, ...? */
147 U32 seen;
88f063b4
KW
148 SSize_t size; /* Number of regnode equivalents in
149 pattern */
e7f023ff 150 I32 npar; /* Capture buffer count, (OPEN) plus
538e84ed
KW
151 one. ("par" 0 is the whole
152 pattern)*/
153 I32 nestroot; /* root parens we are in - used by
154 accept */
830247a4
IZ
155 I32 extralen;
156 I32 seen_zerolen;
f55b7b26
KW
157 regnode_offset *open_parens; /* offsets to open parens */
158 regnode_offset *close_parens; /* offsets to close parens */
d5a00e4a 159 regnode *end_op; /* END node in program */
02daf0ab
YO
160 I32 utf8; /* whether the pattern is utf8 or not */
161 I32 orig_utf8; /* whether the pattern was originally in utf8 */
162 /* XXX use this for future optimisation of case
163 * where pattern must be upgraded to utf8. */
e40e74fe
KW
164 I32 uni_semantics; /* If a d charset modifier should use unicode
165 rules, even if the pattern is not in
166 utf8 */
81714fb9 167 HV *paren_names; /* Paren names */
538e84ed 168
40d049e4 169 regnode **recurse; /* Recurse regops */
88f063b4 170 I32 recurse_count; /* Number of recurse regops we have generated */
4286711a 171 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
538e84ed 172 through */
09a65838 173 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
b57e4118 174 I32 in_lookbehind;
4624b182 175 I32 contains_locale;
bb3f3ed2 176 I32 override_recoding;
b6d67071
KW
177#ifdef EBCDIC
178 I32 recode_x_to_native;
179#endif
9d53c457 180 I32 in_multi_char_class;
1acab4c5 181 struct reg_code_blocks *code_blocks;/* positions of literal (?{})
68e2671b 182 within pattern */
b1603ef8 183 int code_index; /* next code_blocks[] slot */
ee273784 184 SSize_t maxlen; /* mininum possible number of chars in string to match */
3f910e62
YO
185 scan_frame *frame_head;
186 scan_frame *frame_last;
187 U32 frame_count;
7eec73eb 188 AV *warn_text;
dc6d7f5c 189#ifdef ADD_TO_REGEXEC
830247a4
IZ
190 char *starttry; /* -Dr: where regtry was called. */
191#define RExC_starttry (pRExC_state->starttry)
192#endif
d24ca0c5 193 SV *runtime_code_qr; /* qr with the runtime code blocks */
3dab1dad 194#ifdef DEBUGGING
be8e71aa 195 const char *lastparse;
3dab1dad 196 I32 lastnum;
1f1031fe 197 AV *paren_name_list; /* idx -> name */
d9a72fcc 198 U32 study_chunk_recursed_count;
c9f0d54c
YO
199 SV *mysv1;
200 SV *mysv2;
88f063b4 201
3dab1dad
YO
202#define RExC_lastparse (pRExC_state->lastparse)
203#define RExC_lastnum (pRExC_state->lastnum)
1f1031fe 204#define RExC_paren_name_list (pRExC_state->paren_name_list)
d9a72fcc 205#define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
c9f0d54c
YO
206#define RExC_mysv (pRExC_state->mysv1)
207#define RExC_mysv1 (pRExC_state->mysv1)
208#define RExC_mysv2 (pRExC_state->mysv2)
209
3dab1dad 210#endif
512c0f5a 211 bool seen_unfolded_sharp_s;
911bd04e 212 bool strict;
da7cf1cc 213 bool study_started;
034602eb 214 bool in_script_run;
f55b7b26 215 bool pass1;
09b2b2e6 216};
830247a4 217
e2509266 218#define RExC_flags (pRExC_state->flags)
514a91f1 219#define RExC_pm_flags (pRExC_state->pm_flags)
830247a4 220#define RExC_precomp (pRExC_state->precomp)
51684e3c
KW
221#define RExC_copy_start_in_input (pRExC_state->copy_start_in_input)
222#define RExC_copy_start_in_constructed (pRExC_state->copy_start)
711b303b 223#define RExC_precomp_end (pRExC_state->precomp_end)
288b8c02 224#define RExC_rx_sv (pRExC_state->rx_sv)
830247a4 225#define RExC_rx (pRExC_state->rx)
f8fc2ecf 226#define RExC_rxi (pRExC_state->rxi)
fac92740 227#define RExC_start (pRExC_state->start)
830247a4
IZ
228#define RExC_end (pRExC_state->end)
229#define RExC_parse (pRExC_state->parse)
230#define RExC_whilem_seen (pRExC_state->whilem_seen)
512c0f5a
KW
231
232/* Set during the sizing pass when there is a LATIN SMALL LETTER SHARP S in any
233 * EXACTF node, hence was parsed under /di rules. If later in the parse,
234 * something forces the pattern into using /ui rules, the sharp s should be
235 * folded into the sequence 'ss', which takes up more space than previously
236 * calculated. This means that the sizing pass needs to be restarted. (The
237 * node also becomes an EXACTFU_SS.) For all other characters, an EXACTF node
238 * that gets converted to /ui (and EXACTFU) occupies the same amount of space,
239 * so there is no need to resize [perl #125990]. */
240#define RExC_seen_unfolded_sharp_s (pRExC_state->seen_unfolded_sharp_s)
241
7122b237 242#ifdef RE_TRACK_PATTERN_OFFSETS
44f36cd3 243# define RExC_offsets (RExC_rxi->u.offsets) /* I am not like the
538e84ed 244 others */
7122b237 245#endif
830247a4 246#define RExC_emit (pRExC_state->emit)
f55b7b26 247#define RExC_pass1 (pRExC_state->pass1)
fac92740 248#define RExC_emit_start (pRExC_state->emit_start)
3b57cd43 249#define RExC_emit_bound (pRExC_state->emit_bound)
830247a4
IZ
250#define RExC_sawback (pRExC_state->sawback)
251#define RExC_seen (pRExC_state->seen)
252#define RExC_size (pRExC_state->size)
ee273784 253#define RExC_maxlen (pRExC_state->maxlen)
830247a4 254#define RExC_npar (pRExC_state->npar)
e2e6a0f1 255#define RExC_nestroot (pRExC_state->nestroot)
830247a4
IZ
256#define RExC_extralen (pRExC_state->extralen)
257#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
1aa99e6b 258#define RExC_utf8 (pRExC_state->utf8)
e40e74fe 259#define RExC_uni_semantics (pRExC_state->uni_semantics)
02daf0ab 260#define RExC_orig_utf8 (pRExC_state->orig_utf8)
40d049e4
YO
261#define RExC_open_parens (pRExC_state->open_parens)
262#define RExC_close_parens (pRExC_state->close_parens)
5bd2d46e 263#define RExC_end_op (pRExC_state->end_op)
81714fb9 264#define RExC_paren_names (pRExC_state->paren_names)
40d049e4
YO
265#define RExC_recurse (pRExC_state->recurse)
266#define RExC_recurse_count (pRExC_state->recurse_count)
09a65838 267#define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
538e84ed
KW
268#define RExC_study_chunk_recursed_bytes \
269 (pRExC_state->study_chunk_recursed_bytes)
b57e4118 270#define RExC_in_lookbehind (pRExC_state->in_lookbehind)
4624b182 271#define RExC_contains_locale (pRExC_state->contains_locale)
b6d67071
KW
272#ifdef EBCDIC
273# define RExC_recode_x_to_native (pRExC_state->recode_x_to_native)
274#endif
9d53c457 275#define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
3f910e62
YO
276#define RExC_frame_head (pRExC_state->frame_head)
277#define RExC_frame_last (pRExC_state->frame_last)
278#define RExC_frame_count (pRExC_state->frame_count)
67cdf558 279#define RExC_strict (pRExC_state->strict)
da7cf1cc 280#define RExC_study_started (pRExC_state->study_started)
7eec73eb 281#define RExC_warn_text (pRExC_state->warn_text)
034602eb 282#define RExC_in_script_run (pRExC_state->in_script_run)
830247a4 283
99807a43
HS
284/* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
285 * a flag to disable back-off on the fixed/floating substrings - if it's
286 * a high complexity pattern we assume the benefit of avoiding a full match
287 * is worth the cost of checking for the substrings even if they rarely help.
288 */
289#define RExC_naughty (pRExC_state->naughty)
290#define TOO_NAUGHTY (10)
291#define MARK_NAUGHTY(add) \
292 if (RExC_naughty < TOO_NAUGHTY) \
293 RExC_naughty += (add)
294#define MARK_NAUGHTY_EXP(exp, add) \
295 if (RExC_naughty < TOO_NAUGHTY) \
296 RExC_naughty += RExC_naughty / (exp) + (add)
cde0cee5 297
a687059c
LW
298#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
299#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
412f55bb 300 ((*s) == '{' && regcurly(s)))
a687059c
LW
301
302/*
303 * Flags to be passed up and down.
304 */
a687059c 305#define WORST 0 /* Worst case. */
a3b492c3 306#define HASWIDTH 0x01 /* Known to match non-null strings. */
fda99bee 307
e64f369d 308/* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
2fd92675
KW
309 * character. (There needs to be a case: in the switch statement in regexec.c
310 * for any node marked SIMPLE.) Note that this is not the same thing as
311 * REGNODE_SIMPLE */
fda99bee 312#define SIMPLE 0x02
e64f369d 313#define SPSTART 0x04 /* Starts with * or + */
8d9c2815
NC
314#define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
315#define TRYAGAIN 0x10 /* Weeded out a declaration. */
b97943f7
KW
316#define RESTART_PASS1 0x20 /* Need to restart sizing pass */
317#define NEED_UTF8 0x40 /* In conjunction with RESTART_PASS1, need to
88f063b4 318 calculate sizes as UTF-8 */
a687059c 319
3dab1dad
YO
320#define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
321
07be1b83
YO
322/* whether trie related optimizations are enabled */
323#if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
324#define TRIE_STUDY_OPT
786e8c11 325#define FULL_TRIE_STUDY
07be1b83
YO
326#define TRIE_STCLASS
327#endif
1de06328
YO
328
329
40d049e4
YO
330
331#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
332#define PBITVAL(paren) (1 << ((paren) & 7))
333#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
334#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
335#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
336
82a6ada4 337#define REQUIRE_UTF8(flagp) STMT_START { \
8d9c2815 338 if (!UTF) { \
82a6ada4
KW
339 assert(PASS1); \
340 *flagp = RESTART_PASS1|NEED_UTF8; \
f55b7b26 341 return 0; \
8d9c2815 342 } \
82a6ada4 343 } STMT_END
40d049e4 344
512c0f5a
KW
345/* Change from /d into /u rules, and restart the parse if we've already seen
346 * something whose size would increase as a result, by setting *flagp and
347 * returning 'restart_retval'. RExC_uni_semantics is a flag that indicates
142ece99 348 * we've changed to /u during the parse. */
512c0f5a
KW
349#define REQUIRE_UNI_RULES(flagp, restart_retval) \
350 STMT_START { \
351 if (DEPENDS_SEMANTICS) { \
352 assert(PASS1); \
353 set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); \
354 RExC_uni_semantics = 1; \
355 if (RExC_seen_unfolded_sharp_s) { \
356 *flagp |= RESTART_PASS1; \
357 return restart_retval; \
358 } \
359 } \
360 } STMT_END
361
fd7fc6ae
YO
362/* Executes a return statement with the value 'X', if 'flags' contains any of
363 * 'RESTART_PASS1', 'NEED_UTF8', or 'extra'. If so, *flagp is set to those
364 * flags */
365#define RETURN_X_ON_RESTART_OR_FLAGS(X, flags, flagp, extra) \
366 STMT_START { \
367 if ((flags) & (RESTART_PASS1|NEED_UTF8|(extra))) { \
368 *(flagp) = (flags) & (RESTART_PASS1|NEED_UTF8|(extra)); \
369 return X; \
370 } \
371 } STMT_END
372
5d555bb8 373#define RETURN_FAIL_ON_RESTART_OR_FLAGS(flags,flagp,extra) \
f55b7b26 374 RETURN_X_ON_RESTART_OR_FLAGS(0,flags,flagp,extra)
fd7fc6ae
YO
375
376#define RETURN_X_ON_RESTART(X, flags,flagp) \
377 RETURN_X_ON_RESTART_OR_FLAGS( X, flags, flagp, 0)
378
379
5d555bb8 380#define RETURN_FAIL_ON_RESTART_FLAGP_OR_FLAGS(flagp,extra) \
f55b7b26 381 if (*(flagp) & (RESTART_PASS1|(extra))) return 0
fd7fc6ae
YO
382
383#define MUST_RESTART(flags) ((flags) & (RESTART_PASS1))
384
5d555bb8 385#define RETURN_FAIL_ON_RESTART(flags,flagp) \
f55b7b26 386 RETURN_X_ON_RESTART(0, flags,flagp)
5d555bb8 387#define RETURN_FAIL_ON_RESTART_FLAGP(flagp) \
88f063b4 388 RETURN_FAIL_ON_RESTART_FLAGP_OR_FLAGS(flagp, 0)
fd7fc6ae 389
f19b1a63
KW
390/* This converts the named class defined in regcomp.h to its equivalent class
391 * number defined in handy.h. */
392#define namedclass_to_classnum(class) ((int) ((class) / 2))
393#define classnum_to_namedclass(classnum) ((classnum) * 2)
394
de92f5e6
KW
395#define _invlist_union_complement_2nd(a, b, output) \
396 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
397#define _invlist_intersection_complement_2nd(a, b, output) \
398 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
399
1de06328
YO
400/* About scan_data_t.
401
402 During optimisation we recurse through the regexp program performing
403 various inplace (keyhole style) optimisations. In addition study_chunk
404 and scan_commit populate this data structure with information about
538e84ed 405 what strings MUST appear in the pattern. We look for the longest
3b753521 406 string that must appear at a fixed location, and we look for the
1de06328
YO
407 longest string that may appear at a floating location. So for instance
408 in the pattern:
538e84ed 409
1de06328 410 /FOO[xX]A.*B[xX]BAR/
538e84ed 411
1de06328
YO
412 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
413 strings (because they follow a .* construct). study_chunk will identify
414 both FOO and BAR as being the longest fixed and floating strings respectively.
538e84ed 415
1de06328 416 The strings can be composites, for instance
538e84ed 417
1de06328 418 /(f)(o)(o)/
538e84ed 419
1de06328 420 will result in a composite fixed substring 'foo'.
538e84ed 421
1de06328 422 For each string some basic information is maintained:
538e84ed 423
6ea77169 424 - min_offset
1de06328
YO
425 This is the position the string must appear at, or not before.
426 It also implicitly (when combined with minlenp) tells us how many
3b753521
FN
427 characters must match before the string we are searching for.
428 Likewise when combined with minlenp and the length of the string it
538e84ed 429 tells us how many characters must appear after the string we have
1de06328 430 found.
538e84ed 431
1de06328
YO
432 - max_offset
433 Only used for floating strings. This is the rightmost point that
ea3daa5d 434 the string can appear at. If set to SSize_t_MAX it indicates that the
1de06328 435 string can occur infinitely far to the right.
6e12f832 436 For fixed strings, it is equal to min_offset.
538e84ed 437
1de06328 438 - minlenp
2d608413
KW
439 A pointer to the minimum number of characters of the pattern that the
440 string was found inside. This is important as in the case of positive
538e84ed 441 lookahead or positive lookbehind we can have multiple patterns
1de06328 442 involved. Consider
538e84ed 443
1de06328 444 /(?=FOO).*F/
538e84ed 445
1de06328
YO
446 The minimum length of the pattern overall is 3, the minimum length
447 of the lookahead part is 3, but the minimum length of the part that
538e84ed 448 will actually match is 1. So 'FOO's minimum length is 3, but the
1de06328 449 minimum length for the F is 1. This is important as the minimum length
538e84ed 450 is used to determine offsets in front of and behind the string being
1de06328 451 looked for. Since strings can be composites this is the length of the
486ec47a 452 pattern at the time it was committed with a scan_commit. Note that
1de06328 453 the length is calculated by study_chunk, so that the minimum lengths
538e84ed 454 are not known until the full pattern has been compiled, thus the
1de06328 455 pointer to the value.
538e84ed 456
1de06328 457 - lookbehind
538e84ed 458
1de06328 459 In the case of lookbehind the string being searched for can be
538e84ed 460 offset past the start point of the final matching string.
1de06328
YO
461 If this value was just blithely removed from the min_offset it would
462 invalidate some of the calculations for how many chars must match
463 before or after (as they are derived from min_offset and minlen and
538e84ed 464 the length of the string being searched for).
1de06328
YO
465 When the final pattern is compiled and the data is moved from the
466 scan_data_t structure into the regexp structure the information
538e84ed
KW
467 about lookbehind is factored in, with the information that would
468 have been lost precalculated in the end_shift field for the
1de06328
YO
469 associated string.
470
471 The fields pos_min and pos_delta are used to store the minimum offset
538e84ed 472 and the delta to the maximum offset at the current point in the pattern.
1de06328
YO
473
474*/
2c2d71f5 475
cb6aef8b
DM
476struct scan_data_substrs {
477 SV *str; /* longest substring found in pattern */
478 SSize_t min_offset; /* earliest point in string it can appear */
479 SSize_t max_offset; /* latest point in string it can appear */
480 SSize_t *minlenp; /* pointer to the minlen relevant to the string */
481 SSize_t lookbehind; /* is the pos of the string modified by LB */
11683ecb 482 I32 flags; /* per substring SF_* and SCF_* flags */
cb6aef8b
DM
483};
484
2c2d71f5 485typedef struct scan_data_t {
1de06328
YO
486 /*I32 len_min; unused */
487 /*I32 len_delta; unused */
49f55535 488 SSize_t pos_min;
ea3daa5d 489 SSize_t pos_delta;
2c2d71f5 490 SV *last_found;
ea3daa5d 491 SSize_t last_end; /* min value, <0 unless valid. */
49f55535 492 SSize_t last_start_min;
ea3daa5d 493 SSize_t last_start_max;
2099df82
DM
494 U8 cur_is_floating; /* whether the last_* values should be set as
495 * the next fixed (0) or floating (1)
496 * substring */
6ea77169 497
2099df82
DM
498 /* [0] is longest fixed substring so far, [1] is longest float so far */
499 struct scan_data_substrs substrs[2];
6ea77169 500
11683ecb 501 I32 flags; /* common SF_* and SCF_* flags */
2c2d71f5 502 I32 whilem_c;
49f55535 503 SSize_t *last_closep;
b8f7bb16 504 regnode_ssc *start_class;
2c2d71f5
JH
505} scan_data_t;
506
a687059c 507/*
e50aee73 508 * Forward declarations for pregcomp()'s friends.
a687059c 509 */
a0d0e21e 510
6ea77169 511static const scan_data_t zero_scan_data = {
a2ca2017 512 0, 0, NULL, 0, 0, 0, 0,
6ea77169 513 {
11683ecb
DM
514 { NULL, 0, 0, 0, 0, 0 },
515 { NULL, 0, 0, 0, 0, 0 },
6ea77169
DM
516 },
517 0, 0, NULL, NULL
518};
c277df42 519
11683ecb
DM
520/* study flags */
521
07be1b83
YO
522#define SF_BEFORE_SEOL 0x0001
523#define SF_BEFORE_MEOL 0x0002
11683ecb 524#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
c277df42 525
07be1b83
YO
526#define SF_IS_INF 0x0040
527#define SF_HAS_PAR 0x0080
528#define SF_IN_PAR 0x0100
529#define SF_HAS_EVAL 0x0200
bc604ad8
KW
530
531
532/* SCF_DO_SUBSTR is the flag that tells the regexp analyzer to track the
533 * longest substring in the pattern. When it is not set the optimiser keeps
534 * track of position, but does not keep track of the actual strings seen,
535 *
536 * So for instance /foo/ will be parsed with SCF_DO_SUBSTR being true, but
537 * /foo/i will not.
538 *
539 * Similarly, /foo.*(blah|erm|huh).*fnorble/ will have "foo" and "fnorble"
540 * parsed with SCF_DO_SUBSTR on, but while processing the (...) it will be
541 * turned off because of the alternation (BRANCH). */
07be1b83 542#define SCF_DO_SUBSTR 0x0400
bc604ad8 543
653099ff
GS
544#define SCF_DO_STCLASS_AND 0x0800
545#define SCF_DO_STCLASS_OR 0x1000
546#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
e1901655 547#define SCF_WHILEM_VISITED_POS 0x2000
c277df42 548
786e8c11 549#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
538e84ed 550#define SCF_SEEN_ACCEPT 0x8000
688e0391 551#define SCF_TRIE_DOING_RESTUDY 0x10000
4286711a
YO
552#define SCF_IN_DEFINE 0x20000
553
554
555
07be1b83 556
43fead97 557#define UTF cBOOL(RExC_utf8)
00b27cfc
KW
558
559/* The enums for all these are ordered so things work out correctly */
a62b1201 560#define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
538e84ed
KW
561#define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
562 == REGEX_DEPENDS_CHARSET)
00b27cfc 563#define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
538e84ed
KW
564#define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
565 >= REGEX_UNICODE_CHARSET)
566#define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
567 == REGEX_ASCII_RESTRICTED_CHARSET)
568#define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
569 >= REGEX_ASCII_RESTRICTED_CHARSET)
570#define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
571 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
a62b1201 572
43fead97 573#define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
a0ed51b3 574
f2c2a6ab
KW
575/* For programs that want to be strictly Unicode compatible by dying if any
576 * attempt is made to match a non-Unicode code point against a Unicode
577 * property. */
578#define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
579
93733859 580#define OOB_NAMEDCLASS -1
b8c5462f 581
8e661ac5
KW
582/* There is no code point that is out-of-bounds, so this is problematic. But
583 * its only current use is to initialize a variable that is always set before
584 * looked at. */
585#define OOB_UNICODE 0xDEADBEEF
586
a0ed51b3 587#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
a0ed51b3 588
8615cb43 589
b45f050a
JF
590/* length of regex to show in messages that don't mark a position within */
591#define RegexLengthToShowInErrorMessages 127
592
593/*
594 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
595 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
596 * op/pragma/warn/regcomp.
597 */
7253e4e3
RK
598#define MARKER1 "<-- HERE" /* marker as it appears in the description */
599#define MARKER2 " <-- HERE " /* marker as it appears within the regex */
b81d288d 600
538e84ed 601#define REPORT_LOCATION " in regex; marked by " MARKER1 \
147e3846 602 " in m/%" UTF8f MARKER2 "%" UTF8f "/"
b45f050a 603
285b5ca0
KW
604/* The code in this file in places uses one level of recursion with parsing
605 * rebased to an alternate string constructed by us in memory. This can take
606 * the form of something that is completely different from the input, or
607 * something that uses the input as part of the alternate. In the first case,
608 * there should be no possibility of an error, as we are in complete control of
51684e3c
KW
609 * the alternate string. But in the second case we don't completely control
610 * the input portion, so there may be errors in that. Here's an example:
285b5ca0
KW
611 * /[abc\x{DF}def]/ui
612 * is handled specially because \x{df} folds to a sequence of more than one
51684e3c 613 * character: 'ss'. What is done is to create and parse an alternate string,
285b5ca0
KW
614 * which looks like this:
615 * /(?:\x{DF}|[abc\x{DF}def])/ui
616 * where it uses the input unchanged in the middle of something it constructs,
617 * which is a branch for the DF outside the character class, and clustering
618 * parens around the whole thing. (It knows enough to skip the DF inside the
619 * class while in this substitute parse.) 'abc' and 'def' may have errors that
620 * need to be reported. The general situation looks like this:
621 *
51684e3c 622 * |<------- identical ------>|
285b5ca0 623 * sI tI xI eI
51684e3c 624 * Input: ---------------------------------------------------------------
285b5ca0
KW
625 * Constructed: ---------------------------------------------------
626 * sC tC xC eC EC
51684e3c 627 * |<------- identical ------>|
285b5ca0 628 *
51684e3c
KW
629 * sI..eI is the portion of the input pattern we are concerned with here.
630 * sC..EC is the constructed substitute parse string.
631 * sC..tC is constructed by us
632 * tC..eC is an exact duplicate of the portion of the input pattern tI..eI.
633 * In the diagram, these are vertically aligned.
634 * eC..EC is also constructed by us.
635 * xC is the position in the substitute parse string where we found a
636 * problem.
637 * xI is the position in the original pattern corresponding to xC.
285b5ca0 638 *
51684e3c
KW
639 * We want to display a message showing the real input string. Thus we need to
640 * translate from xC to xI. We know that xC >= tC, since the portion of the
641 * string sC..tC has been constructed by us, and so shouldn't have errors. We
642 * get:
643 * xI = tI + (xC - tC)
285b5ca0 644 *
51684e3c
KW
645 * When the substitute parse is constructed, the code needs to set:
646 * RExC_start (sC)
647 * RExC_end (eC)
648 * RExC_copy_start_in_input (tI)
649 * RExC_copy_start_in_constructed (tC)
650 * and restore them when done.
285b5ca0 651 *
51684e3c
KW
652 * During normal processing of the input pattern, both
653 * 'RExC_copy_start_in_input' and 'RExC_copy_start_in_constructed' are set to
654 * sI, so that xC equals xI.
285b5ca0
KW
655 */
656
51684e3c
KW
657#define sI RExC_precomp
658#define eI RExC_precomp_end
659#define sC RExC_start
660#define eC RExC_end
661#define tI RExC_copy_start_in_input
662#define tC RExC_copy_start_in_constructed
663#define xI(xC) (tI + (xC - tC))
664#define xI_offset(xC) (xI(xC) - sI)
285b5ca0
KW
665
666#define REPORT_LOCATION_ARGS(xC) \
667 UTF8fARG(UTF, \
51684e3c 668 (xI(xC) > eI) /* Don't run off end */ \
285b5ca0 669 ? eC - sC /* Length before the <--HERE */ \
bc929b61
KW
670 : ((xI_offset(xC) >= 0) \
671 ? xI_offset(xC) \
672 : (Perl_croak(aTHX_ "panic: %s: %d: negative offset: %" \
673 IVdf " trying to output message for " \
674 " pattern %.*s", \
51684e3c 675 __FILE__, __LINE__, (IV) xI_offset(xC), \
bc929b61 676 ((int) (eC - sC)), sC), 0)), \
51684e3c 677 sI), /* The input pattern printed up to the <--HERE */ \
285b5ca0 678 UTF8fARG(UTF, \
51684e3c
KW
679 (xI(xC) > eI) ? 0 : eI - xI(xC), /* Length after <--HERE */ \
680 (xI(xC) > eI) ? eI : xI(xC)) /* pattern after <--HERE */
c1d900c3 681
8a6d8ec6
HS
682/* Used to point after bad bytes for an error message, but avoid skipping
683 * past a nul byte. */
684#define SKIP_IF_CHAR(s) (!*(s) ? 0 : UTF ? UTF8SKIP(s) : 1)
685
b45f050a
JF
686/*
687 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
688 * arg. Show regex, up to a maximum length. If it's too long, chop and add
689 * "...".
690 */
58e23c8d 691#define _FAIL(code) STMT_START { \
bfed75c6 692 const char *ellipses = ""; \
88f063b4 693 IV len = RExC_precomp_end - RExC_precomp; \
ccb2c380
MP
694 \
695 if (!SIZE_ONLY) \
a5e7bc51 696 SAVEFREESV(RExC_rx_sv); \
ccb2c380
MP
697 if (len > RegexLengthToShowInErrorMessages) { \
698 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
699 len = RegexLengthToShowInErrorMessages - 10; \
700 ellipses = "..."; \
701 } \
58e23c8d 702 code; \
ccb2c380 703} STMT_END
8615cb43 704
58e23c8d 705#define FAIL(msg) _FAIL( \
147e3846 706 Perl_croak(aTHX_ "%s in regex m/%" UTF8f "%s/", \
c1d900c3 707 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d
YO
708
709#define FAIL2(msg,arg) _FAIL( \
147e3846 710 Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \
c1d900c3 711 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d 712
b45f050a 713/*
b45f050a
JF
714 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
715 */
ccb2c380 716#define Simple_vFAIL(m) STMT_START { \
ccb2c380 717 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
d528642a 718 m, REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 719} STMT_END
b45f050a
JF
720
721/*
722 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
723 */
ccb2c380
MP
724#define vFAIL(m) STMT_START { \
725 if (!SIZE_ONLY) \
a5e7bc51 726 SAVEFREESV(RExC_rx_sv); \
ccb2c380
MP
727 Simple_vFAIL(m); \
728} STMT_END
b45f050a
JF
729
730/*
731 * Like Simple_vFAIL(), but accepts two arguments.
732 */
ccb2c380 733#define Simple_vFAIL2(m,a1) STMT_START { \
d528642a
KW
734 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
735 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 736} STMT_END
b45f050a
JF
737
738/*
739 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
740 */
ccb2c380
MP
741#define vFAIL2(m,a1) STMT_START { \
742 if (!SIZE_ONLY) \
a5e7bc51 743 SAVEFREESV(RExC_rx_sv); \
ccb2c380
MP
744 Simple_vFAIL2(m, a1); \
745} STMT_END
b45f050a
JF
746
747
748/*
749 * Like Simple_vFAIL(), but accepts three arguments.
750 */
ccb2c380 751#define Simple_vFAIL3(m, a1, a2) STMT_START { \
c1d900c3 752 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 753 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 754} STMT_END
b45f050a
JF
755
756/*
757 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
758 */
ccb2c380
MP
759#define vFAIL3(m,a1,a2) STMT_START { \
760 if (!SIZE_ONLY) \
a5e7bc51 761 SAVEFREESV(RExC_rx_sv); \
ccb2c380
MP
762 Simple_vFAIL3(m, a1, a2); \
763} STMT_END
b45f050a
JF
764
765/*
766 * Like Simple_vFAIL(), but accepts four arguments.
767 */
ccb2c380 768#define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
d528642a
KW
769 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
770 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 771} STMT_END
b45f050a 772
95db3ffa
KW
773#define vFAIL4(m,a1,a2,a3) STMT_START { \
774 if (!SIZE_ONLY) \
775 SAVEFREESV(RExC_rx_sv); \
776 Simple_vFAIL4(m, a1, a2, a3); \
777} STMT_END
778
946095af 779/* A specialized version of vFAIL2 that works with UTF8f */
d528642a
KW
780#define vFAIL2utf8f(m, a1) STMT_START { \
781 if (!SIZE_ONLY) \
782 SAVEFREESV(RExC_rx_sv); \
783 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
784 REPORT_LOCATION_ARGS(RExC_parse)); \
946095af
BF
785} STMT_END
786
3ba22297 787#define vFAIL3utf8f(m, a1, a2) STMT_START { \
3ba22297
KW
788 if (!SIZE_ONLY) \
789 SAVEFREESV(RExC_rx_sv); \
790 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 791 REPORT_LOCATION_ARGS(RExC_parse)); \
3ba22297
KW
792} STMT_END
793
499333dc
KW
794/* These have asserts in them because of [perl #122671] Many warnings in
795 * regcomp.c can occur twice. If they get output in pass1 and later in that
796 * pass, the pattern has to be converted to UTF-8 and the pass restarted, they
797 * would get output again. So they should be output in pass2, and these
798 * asserts make sure new warnings follow that paradigm. */
946095af 799
5e0a247b
KW
800/* m is not necessarily a "literal string", in this macro */
801#define reg_warn_non_literal_string(loc, m) STMT_START { \
d528642a
KW
802 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
803 "%s" REPORT_LOCATION, \
804 m, REPORT_LOCATION_ARGS(loc)); \
5e0a247b
KW
805} STMT_END
806
668c081a 807#define ckWARNreg(loc,m) STMT_START { \
d528642a
KW
808 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
809 m REPORT_LOCATION, \
810 REPORT_LOCATION_ARGS(loc)); \
ccb2c380
MP
811} STMT_END
812
b927b7e9 813#define vWARN(loc, m) STMT_START { \
d528642a
KW
814 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
815 m REPORT_LOCATION, \
816 REPORT_LOCATION_ARGS(loc)); \
b927b7e9
KW
817} STMT_END
818
0d6106aa 819#define vWARN_dep(loc, m) STMT_START { \
d528642a
KW
820 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), \
821 m REPORT_LOCATION, \
822 REPORT_LOCATION_ARGS(loc)); \
0d6106aa
KW
823} STMT_END
824
147508a2 825#define ckWARNdep(loc,m) STMT_START { \
d528642a
KW
826 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
827 m REPORT_LOCATION, \
828 REPORT_LOCATION_ARGS(loc)); \
147508a2
KW
829} STMT_END
830
d528642a
KW
831#define ckWARNregdep(loc,m) STMT_START { \
832 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, \
833 WARN_REGEXP), \
834 m REPORT_LOCATION, \
835 REPORT_LOCATION_ARGS(loc)); \
ccb2c380
MP
836} STMT_END
837
d528642a
KW
838#define ckWARN2reg_d(loc,m, a1) STMT_START { \
839 __ASSERT_(PASS2) Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
840 m REPORT_LOCATION, \
841 a1, REPORT_LOCATION_ARGS(loc)); \
2335b3d3
KW
842} STMT_END
843
d528642a
KW
844#define ckWARN2reg(loc, m, a1) STMT_START { \
845 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
846 m REPORT_LOCATION, \
847 a1, REPORT_LOCATION_ARGS(loc)); \
ccb2c380
MP
848} STMT_END
849
d528642a
KW
850#define vWARN3(loc, m, a1, a2) STMT_START { \
851 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
852 m REPORT_LOCATION, \
853 a1, a2, REPORT_LOCATION_ARGS(loc)); \
ccb2c380
MP
854} STMT_END
855
d528642a
KW
856#define ckWARN3reg(loc, m, a1, a2) STMT_START { \
857 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
858 m REPORT_LOCATION, \
859 a1, a2, \
860 REPORT_LOCATION_ARGS(loc)); \
668c081a
NC
861} STMT_END
862
ccb2c380 863#define vWARN4(loc, m, a1, a2, a3) STMT_START { \
d528642a
KW
864 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
865 m REPORT_LOCATION, \
866 a1, a2, a3, \
867 REPORT_LOCATION_ARGS(loc)); \
ccb2c380
MP
868} STMT_END
869
668c081a 870#define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
d528642a
KW
871 __ASSERT_(PASS2) Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
872 m REPORT_LOCATION, \
873 a1, a2, a3, \
874 REPORT_LOCATION_ARGS(loc)); \
668c081a
NC
875} STMT_END
876
ccb2c380 877#define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
d528642a
KW
878 __ASSERT_(PASS2) Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
879 m REPORT_LOCATION, \
880 a1, a2, a3, a4, \
881 REPORT_LOCATION_ARGS(loc)); \
ccb2c380 882} STMT_END
9d1d55b5 883
1a322bb4
KW
884/* Convert between a pointer to a node and its offset from the beginning of the
885 * program */
886#define REGNODE_p(offset) (RExC_emit_start + (offset))
887#define REGNODE_OFFSET(node) ((node) - RExC_emit_start)
888
538e84ed 889/* Macros for recording node offsets. 20001227 mjd@plover.com
fac92740
MJD
890 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
891 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
892 * Element 0 holds the number n.
07be1b83 893 * Position is 1 indexed.
fac92740 894 */
7122b237 895#ifndef RE_TRACK_PATTERN_OFFSETS
d7c442ce 896#define Set_Node_Offset_To_R(offset,byte)
7122b237
YO
897#define Set_Node_Offset(node,byte)
898#define Set_Cur_Node_Offset
899#define Set_Node_Length_To_R(node,len)
900#define Set_Node_Length(node,len)
6a86c6ad 901#define Set_Node_Cur_Length(node,start)
538e84ed
KW
902#define Node_Offset(n)
903#define Node_Length(n)
7122b237
YO
904#define Set_Node_Offset_Length(node,offset,len)
905#define ProgLen(ri) ri->u.proglen
906#define SetProgLen(ri,x) ri->u.proglen = x
907#else
908#define ProgLen(ri) ri->u.offsets[0]
909#define SetProgLen(ri,x) ri->u.offsets[0] = x
d7c442ce 910#define Set_Node_Offset_To_R(offset,byte) STMT_START { \
ccb2c380
MP
911 if (! SIZE_ONLY) { \
912 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
d7c442ce
KW
913 __LINE__, (int)(offset), (int)(byte))); \
914 if((offset) < 0) { \
538e84ed 915 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
d7c442ce 916 (int)(offset)); \
ccb2c380 917 } else { \
d7c442ce 918 RExC_offsets[2*(offset)-1] = (byte); \
ccb2c380
MP
919 } \
920 } \
921} STMT_END
922
88f063b4 923#define Set_Node_Offset(node,byte) \
1a322bb4 924 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (byte)-RExC_start)
ccb2c380
MP
925#define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
926
927#define Set_Node_Length_To_R(node,len) STMT_START { \
928 if (! SIZE_ONLY) { \
929 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
551405c4 930 __LINE__, (int)(node), (int)(len))); \
ccb2c380 931 if((node) < 0) { \
538e84ed
KW
932 Perl_croak(aTHX_ "value of node is %d in Length macro", \
933 (int)(node)); \
ccb2c380
MP
934 } else { \
935 RExC_offsets[2*(node)] = (len); \
936 } \
937 } \
938} STMT_END
939
940#define Set_Node_Length(node,len) \
1a322bb4 941 Set_Node_Length_To_R(REGNODE_OFFSET(node), len)
6a86c6ad
NC
942#define Set_Node_Cur_Length(node, start) \
943 Set_Node_Length(node, RExC_parse - start)
fac92740
MJD
944
945/* Get offsets and lengths */
1a322bb4
KW
946#define Node_Offset(n) (RExC_offsets[2*(REGNODE_OFFSET(n))-1])
947#define Node_Length(n) (RExC_offsets[2*(REGNODE_OFFSET(n))])
fac92740 948
07be1b83 949#define Set_Node_Offset_Length(node,offset,len) STMT_START { \
1a322bb4
KW
950 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (offset)); \
951 Set_Node_Length_To_R(REGNODE_OFFSET(node), (len)); \
07be1b83 952} STMT_END
7122b237 953#endif
07be1b83
YO
954
955#if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
956#define EXPERIMENTAL_INPLACESCAN
f427392e 957#endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
07be1b83 958
cb41e5d6
YO
959#ifdef DEBUGGING
960int
6ad9a8ab 961Perl_re_printf(pTHX_ const char *fmt, ...)
cb41e5d6
YO
962{
963 va_list ap;
964 int result;
965 PerlIO *f= Perl_debug_log;
966 PERL_ARGS_ASSERT_RE_PRINTF;
967 va_start(ap, fmt);
968 result = PerlIO_vprintf(f, fmt, ap);
969 va_end(ap);
970 return result;
971}
972
973int
7b031478 974Perl_re_indentf(pTHX_ const char *fmt, U32 depth, ...)
cb41e5d6
YO
975{
976 va_list ap;
977 int result;
978 PerlIO *f= Perl_debug_log;
979 PERL_ARGS_ASSERT_RE_INDENTF;
980 va_start(ap, depth);
daeb874b 981 PerlIO_printf(f, "%*s", ( (int)depth % 20 ) * 2, "");
cb41e5d6
YO
982 result = PerlIO_vprintf(f, fmt, ap);
983 va_end(ap);
984 return result;
985}
986#endif /* DEBUGGING */
987
7b031478 988#define DEBUG_RExC_seen() \
538e84ed 989 DEBUG_OPTIMISE_MORE_r({ \
88f063b4 990 Perl_re_printf( aTHX_ "RExC_seen: "); \
538e84ed 991 \
e384d5c1 992 if (RExC_seen & REG_ZERO_LEN_SEEN) \
88f063b4 993 Perl_re_printf( aTHX_ "REG_ZERO_LEN_SEEN "); \
538e84ed 994 \
e384d5c1 995 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
88f063b4 996 Perl_re_printf( aTHX_ "REG_LOOKBEHIND_SEEN "); \
538e84ed 997 \
e384d5c1 998 if (RExC_seen & REG_GPOS_SEEN) \
88f063b4 999 Perl_re_printf( aTHX_ "REG_GPOS_SEEN "); \
538e84ed 1000 \
e384d5c1 1001 if (RExC_seen & REG_RECURSE_SEEN) \
88f063b4 1002 Perl_re_printf( aTHX_ "REG_RECURSE_SEEN "); \
538e84ed 1003 \
7b031478 1004 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
88f063b4 1005 Perl_re_printf( aTHX_ "REG_TOP_LEVEL_BRANCHES_SEEN "); \
538e84ed 1006 \
e384d5c1 1007 if (RExC_seen & REG_VERBARG_SEEN) \
88f063b4 1008 Perl_re_printf( aTHX_ "REG_VERBARG_SEEN "); \
538e84ed 1009 \
e384d5c1 1010 if (RExC_seen & REG_CUTGROUP_SEEN) \
88f063b4 1011 Perl_re_printf( aTHX_ "REG_CUTGROUP_SEEN "); \
538e84ed 1012 \
e384d5c1 1013 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
88f063b4 1014 Perl_re_printf( aTHX_ "REG_RUN_ON_COMMENT_SEEN "); \
538e84ed 1015 \
e384d5c1 1016 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
88f063b4 1017 Perl_re_printf( aTHX_ "REG_UNFOLDED_MULTI_SEEN "); \
538e84ed 1018 \
7b031478 1019 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
88f063b4 1020 Perl_re_printf( aTHX_ "REG_UNBOUNDED_QUANTIFIER_SEEN "); \
ee273784 1021 \
88f063b4 1022 Perl_re_printf( aTHX_ "\n"); \
9e9ecfdf
YO
1023 });
1024
fdfb4f21 1025#define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
6ad9a8ab 1026 if ((flags) & flag) Perl_re_printf( aTHX_ "%s ", #flag)
fdfb4f21 1027
fdfb4f21 1028
f5a36d78
DM
1029#ifdef DEBUGGING
1030static void
1031S_debug_show_study_flags(pTHX_ U32 flags, const char *open_str,
1032 const char *close_str)
1033{
1034 if (!flags)
1035 return;
1036
1037 Perl_re_printf( aTHX_ "%s", open_str);
11683ecb
DM
1038 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_SEOL);
1039 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_MEOL);
f5a36d78
DM
1040 DEBUG_SHOW_STUDY_FLAG(flags, SF_IS_INF);
1041 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_PAR);
1042 DEBUG_SHOW_STUDY_FLAG(flags, SF_IN_PAR);
1043 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_EVAL);
1044 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_SUBSTR);
1045 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_AND);
1046 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_OR);
1047 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS);
1048 DEBUG_SHOW_STUDY_FLAG(flags, SCF_WHILEM_VISITED_POS);
1049 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_RESTUDY);
1050 DEBUG_SHOW_STUDY_FLAG(flags, SCF_SEEN_ACCEPT);
1051 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_DOING_RESTUDY);
1052 DEBUG_SHOW_STUDY_FLAG(flags, SCF_IN_DEFINE);
1053 Perl_re_printf( aTHX_ "%s", close_str);
1054}
1055
1056
1057static void
1058S_debug_studydata(pTHX_ const char *where, scan_data_t *data,
1059 U32 depth, int is_inf)
1060{
1061 GET_RE_DEBUG_FLAGS_DECL;
1062
1063 DEBUG_OPTIMISE_MORE_r({
1064 if (!data)
1065 return;
1066 Perl_re_indentf(aTHX_ "%s: Pos:%" IVdf "/%" IVdf " Flags: 0x%" UVXf,
1067 depth,
1068 where,
1069 (IV)data->pos_min,
1070 (IV)data->pos_delta,
1071 (UV)data->flags
1072 );
1073
1074 S_debug_show_study_flags(aTHX_ data->flags," [","]");
fdfb4f21 1075
f5a36d78
DM
1076 Perl_re_printf( aTHX_
1077 " Whilem_c: %" IVdf " Lcp: %" IVdf " %s",
1078 (IV)data->whilem_c,
1079 (IV)(data->last_closep ? *((data)->last_closep) : -1),
1080 is_inf ? "INF " : ""
1081 );
1082
11683ecb 1083 if (data->last_found) {
37b6262f 1084 int i;
11683ecb
DM
1085 Perl_re_printf(aTHX_
1086 "Last:'%s' %" IVdf ":%" IVdf "/%" IVdf,
1087 SvPVX_const(data->last_found),
1088 (IV)data->last_end,
1089 (IV)data->last_start_min,
1090 (IV)data->last_start_max
1091 );
1092
2099df82 1093 for (i = 0; i < 2; i++) {
37b6262f
DM
1094 Perl_re_printf(aTHX_
1095 " %s%s: '%s' @ %" IVdf "/%" IVdf,
2099df82
DM
1096 data->cur_is_floating == i ? "*" : "",
1097 i ? "Float" : "Fixed",
37b6262f
DM
1098 SvPVX_const(data->substrs[i].str),
1099 (IV)data->substrs[i].min_offset,
1100 (IV)data->substrs[i].max_offset
1101 );
1102 S_debug_show_study_flags(aTHX_ data->substrs[i].flags," [","]");
1103 }
11683ecb 1104 }
f5a36d78
DM
1105
1106 Perl_re_printf( aTHX_ "\n");
1107 });
1108}
1109
1110
1111static void
1112S_debug_peep(pTHX_ const char *str, const RExC_state_t *pRExC_state,
1113 regnode *scan, U32 depth, U32 flags)
1114{
1115 GET_RE_DEBUG_FLAGS_DECL;
1116
1117 DEBUG_OPTIMISE_r({
1118 regnode *Next;
1119
1120 if (!scan)
1121 return;
1122 Next = regnext(scan);
1123 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
1124 Perl_re_indentf( aTHX_ "%s>%3d: %s (%d)",
1125 depth,
1126 str,
1127 REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),
1128 Next ? (REG_NODE_NUM(Next)) : 0 );
1129 S_debug_show_study_flags(aTHX_ flags," [ ","]");
1130 Perl_re_printf( aTHX_ "\n");
1131 });
1132}
1133
1134
1135# define DEBUG_STUDYDATA(where, data, depth, is_inf) \
1136 S_debug_studydata(aTHX_ where, data, depth, is_inf)
1137
1138# define DEBUG_PEEP(str, scan, depth, flags) \
1139 S_debug_peep(aTHX_ str, pRExC_state, scan, depth, flags)
1140
1141#else
1142# define DEBUG_STUDYDATA(where, data, depth, is_inf) NOOP
1143# define DEBUG_PEEP(str, scan, depth, flags) NOOP
1144#endif
1de06328 1145
cb41e5d6 1146
c6871b76
KW
1147/* =========================================================
1148 * BEGIN edit_distance stuff.
1149 *
1150 * This calculates how many single character changes of any type are needed to
1151 * transform a string into another one. It is taken from version 3.1 of
1152 *
1153 * https://metacpan.org/pod/Text::Levenshtein::Damerau::XS
1154 */
1155
1156/* Our unsorted dictionary linked list. */
1157/* Note we use UVs, not chars. */
1158
1159struct dictionary{
1160 UV key;
1161 UV value;
1162 struct dictionary* next;
1163};
1164typedef struct dictionary item;
1165
1166
1167PERL_STATIC_INLINE item*
88f063b4 1168push(UV key, item* curr)
c6871b76
KW
1169{
1170 item* head;
d09f14bf 1171 Newx(head, 1, item);
c6871b76
KW
1172 head->key = key;
1173 head->value = 0;
1174 head->next = curr;
1175 return head;
1176}
1177
1178
1179PERL_STATIC_INLINE item*
1180find(item* head, UV key)
1181{
1182 item* iterator = head;
1183 while (iterator){
1184 if (iterator->key == key){
1185 return iterator;
1186 }
1187 iterator = iterator->next;
1188 }
1189
1190 return NULL;
1191}
1192
1193PERL_STATIC_INLINE item*
88f063b4 1194uniquePush(item* head, UV key)
c6871b76
KW
1195{
1196 item* iterator = head;
1197
1198 while (iterator){
1199 if (iterator->key == key) {
1200 return head;
1201 }
1202 iterator = iterator->next;
1203 }
1204
88f063b4 1205 return push(key, head);
c6871b76
KW
1206}
1207
1208PERL_STATIC_INLINE void
1209dict_free(item* head)
1210{
1211 item* iterator = head;
1212
1213 while (iterator) {
1214 item* temp = iterator;
1215 iterator = iterator->next;
1216 Safefree(temp);
1217 }
1218
1219 head = NULL;
1220}
1221
1222/* End of Dictionary Stuff */
1223
1224/* All calculations/work are done here */
1225STATIC int
1226S_edit_distance(const UV* src,
1227 const UV* tgt,
1228 const STRLEN x, /* length of src[] */
1229 const STRLEN y, /* length of tgt[] */
1230 const SSize_t maxDistance
1231)
1232{
1233 item *head = NULL;
88f063b4 1234 UV swapCount, swapScore, targetCharCount, i, j;
c6871b76
KW
1235 UV *scores;
1236 UV score_ceil = x + y;
1237
1238 PERL_ARGS_ASSERT_EDIT_DISTANCE;
1239
1240 /* intialize matrix start values */
d09f14bf 1241 Newx(scores, ( (x + 2) * (y + 2)), UV);
c6871b76
KW
1242 scores[0] = score_ceil;
1243 scores[1 * (y + 2) + 0] = score_ceil;
1244 scores[0 * (y + 2) + 1] = score_ceil;
1245 scores[1 * (y + 2) + 1] = 0;
88f063b4 1246 head = uniquePush(uniquePush(head, src[0]), tgt[0]);
c6871b76
KW
1247
1248 /* work loops */
1249 /* i = src index */
1250 /* j = tgt index */
1251 for (i=1;i<=x;i++) {
1252 if (i < x)
88f063b4 1253 head = uniquePush(head, src[i]);
c6871b76
KW
1254 scores[(i+1) * (y + 2) + 1] = i;
1255 scores[(i+1) * (y + 2) + 0] = score_ceil;
1256 swapCount = 0;
1257
1258 for (j=1;j<=y;j++) {
1259 if (i == 1) {
1260 if(j < y)
88f063b4 1261 head = uniquePush(head, tgt[j]);
c6871b76
KW
1262 scores[1 * (y + 2) + (j + 1)] = j;
1263 scores[0 * (y + 2) + (j + 1)] = score_ceil;
1264 }
1265
88f063b4 1266 targetCharCount = find(head, tgt[j-1])->value;
c6871b76
KW
1267 swapScore = scores[targetCharCount * (y + 2) + swapCount] + i - targetCharCount - 1 + j - swapCount;
1268
1269 if (src[i-1] != tgt[j-1]){
1270 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));
1271 }
1272 else {
1273 swapCount = j;
1274 scores[(i+1) * (y + 2) + (j + 1)] = MIN(scores[i * (y + 2) + j], swapScore);
1275 }
1276 }
1277
88f063b4 1278 find(head, src[i-1])->value = i;
c6871b76
KW
1279 }
1280
1281 {
1282 IV score = scores[(x+1) * (y + 2) + (y + 1)];
1283 dict_free(head);
1284 Safefree(scores);
1285 return (maxDistance != 0 && maxDistance < score)?(-1):score;
1286 }
1287}
1288
1289/* END of edit_distance() stuff
1290 * ========================================================= */
1291
8e35b056
KW
1292/* is c a control character for which we have a mnemonic? */
1293#define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
1294
549b4e78
KW
1295STATIC const char *
1296S_cntrl_to_mnemonic(const U8 c)
1297{
1298 /* Returns the mnemonic string that represents character 'c', if one
1299 * exists; NULL otherwise. The only ones that exist for the purposes of
1300 * this routine are a few control characters */
1301
1302 switch (c) {
1303 case '\a': return "\\a";
1304 case '\b': return "\\b";
1305 case ESC_NATIVE: return "\\e";
1306 case '\f': return "\\f";
1307 case '\n': return "\\n";
1308 case '\r': return "\\r";
1309 case '\t': return "\\t";
1310 }
1311
1312 return NULL;
1313}
1314
653099ff 1315/* Mark that we cannot extend a found fixed substring at this point.
37b6262f 1316 Update the longest found anchored substring or the longest found
653099ff
GS
1317 floating substrings if needed. */
1318
4327152a 1319STATIC void
ea3daa5d
FC
1320S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
1321 SSize_t *minlenp, int is_inf)
c277df42 1322{
e1ec3a88 1323 const STRLEN l = CHR_SVLEN(data->last_found);
2099df82 1324 SV * const longest_sv = data->substrs[data->cur_is_floating].str;
a2ca2017 1325 const STRLEN old_l = CHR_SVLEN(longest_sv);
1de06328 1326 GET_RE_DEBUG_FLAGS_DECL;
b81d288d 1327
7918f24d
NC
1328 PERL_ARGS_ASSERT_SCAN_COMMIT;
1329
c277df42 1330 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
2099df82 1331 const U8 i = data->cur_is_floating;
a2ca2017 1332 SvSetMagicSV(longest_sv, data->last_found);
37b6262f
DM
1333 data->substrs[i].min_offset = l ? data->last_start_min : data->pos_min;
1334
2099df82
DM
1335 if (!i) /* fixed */
1336 data->substrs[0].max_offset = data->substrs[0].min_offset;
1337 else { /* float */
1338 data->substrs[1].max_offset = (l
646e8787
DM
1339 ? data->last_start_max
1340 : (data->pos_delta > SSize_t_MAX - data->pos_min
ea3daa5d
FC
1341 ? SSize_t_MAX
1342 : data->pos_min + data->pos_delta));
1343 if (is_inf
2099df82
DM
1344 || (STRLEN)data->substrs[1].max_offset > (STRLEN)SSize_t_MAX)
1345 data->substrs[1].max_offset = SSize_t_MAX;
37b6262f 1346 }
11683ecb 1347
37b6262f
DM
1348 if (data->flags & SF_BEFORE_EOL)
1349 data->substrs[i].flags |= (data->flags & SF_BEFORE_EOL);
1350 else
1351 data->substrs[i].flags &= ~SF_BEFORE_EOL;
1352 data->substrs[i].minlenp = minlenp;
1353 data->substrs[i].lookbehind = 0;
c277df42 1354 }
37b6262f 1355
c277df42 1356 SvCUR_set(data->last_found, 0);
0eda9292 1357 {
a28509cc 1358 SV * const sv = data->last_found;
097eb12c
AL
1359 if (SvUTF8(sv) && SvMAGICAL(sv)) {
1360 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
1361 if (mg)
1362 mg->mg_len = 0;
1363 }
0eda9292 1364 }
c277df42
IZ
1365 data->last_end = -1;
1366 data->flags &= ~SF_BEFORE_EOL;
f5a36d78 1367 DEBUG_STUDYDATA("commit", data, 0, is_inf);
c277df42
IZ
1368}
1369
cdd87c1d
KW
1370/* An SSC is just a regnode_charclass_posix with an extra field: the inversion
1371 * list that describes which code points it matches */
1372
653099ff 1373STATIC void
3420edd7 1374S_ssc_anything(pTHX_ regnode_ssc *ssc)
653099ff 1375{
cdd87c1d
KW
1376 /* Set the SSC 'ssc' to match an empty string or any code point */
1377
557bd3fb 1378 PERL_ARGS_ASSERT_SSC_ANYTHING;
7918f24d 1379
71068078 1380 assert(is_ANYOF_SYNTHETIC(ssc));
3fffb88a 1381
0854ea0b
KW
1382 /* mortalize so won't leak */
1383 ssc->invlist = sv_2mortal(_add_range_to_invlist(NULL, 0, UV_MAX));
93e92956 1384 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
653099ff
GS
1385}
1386
653099ff 1387STATIC int
dc3bf405 1388S_ssc_is_anything(const regnode_ssc *ssc)
653099ff 1389{
c144baaa
KW
1390 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
1391 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
1392 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1393 * in any way, so there's no point in using it */
cdd87c1d
KW
1394
1395 UV start, end;
1396 bool ret;
653099ff 1397
557bd3fb 1398 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
7918f24d 1399
71068078 1400 assert(is_ANYOF_SYNTHETIC(ssc));
cdd87c1d 1401
93e92956 1402 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
cdd87c1d
KW
1403 return FALSE;
1404 }
1405
1406 /* See if the list consists solely of the range 0 - Infinity */
1407 invlist_iterinit(ssc->invlist);
1408 ret = invlist_iternext(ssc->invlist, &start, &end)
1409 && start == 0
1410 && end == UV_MAX;
1411
1412 invlist_iterfinish(ssc->invlist);
1413
1414 if (ret) {
1415 return TRUE;
1416 }
1417
1418 /* If e.g., both \w and \W are set, matches everything */
e0e1be5f 1419 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1420 int i;
1421 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1422 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1423 return TRUE;
1424 }
1425 }
1426 }
1427
1428 return FALSE;
653099ff
GS
1429}
1430
653099ff 1431STATIC void
cdd87c1d 1432S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
653099ff 1433{
cdd87c1d
KW
1434 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1435 * string, any code point, or any posix class under locale */
1436
557bd3fb 1437 PERL_ARGS_ASSERT_SSC_INIT;
7918f24d 1438
557bd3fb 1439 Zero(ssc, 1, regnode_ssc);
71068078 1440 set_ANYOF_SYNTHETIC(ssc);
93e92956 1441 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
3420edd7 1442 ssc_anything(ssc);
cdd87c1d 1443
2f306ab9
KW
1444 /* If any portion of the regex is to operate under locale rules that aren't
1445 * fully known at compile time, initialization includes it. The reason
1446 * this isn't done for all regexes is that the optimizer was written under
1447 * the assumption that locale was all-or-nothing. Given the complexity and
1448 * lack of documentation in the optimizer, and that there are inadequate
1449 * test cases for locale, many parts of it may not work properly, it is
1450 * safest to avoid locale unless necessary. */
cdd87c1d
KW
1451 if (RExC_contains_locale) {
1452 ANYOF_POSIXL_SETALL(ssc);
cdd87c1d
KW
1453 }
1454 else {
1455 ANYOF_POSIXL_ZERO(ssc);
1456 }
653099ff
GS
1457}
1458
b423522f 1459STATIC int
dc3bf405
BF
1460S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1461 const regnode_ssc *ssc)
b423522f
KW
1462{
1463 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1464 * to the list of code points matched, and locale posix classes; hence does
1465 * not check its flags) */
1466
1467 UV start, end;
1468 bool ret;
1469
1470 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1471
71068078 1472 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1473
1474 invlist_iterinit(ssc->invlist);
1475 ret = invlist_iternext(ssc->invlist, &start, &end)
1476 && start == 0
1477 && end == UV_MAX;
1478
1479 invlist_iterfinish(ssc->invlist);
1480
1481 if (! ret) {
1482 return FALSE;
1483 }
1484
e0e1be5f 1485 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
31f05a37 1486 return FALSE;
b423522f
KW
1487 }
1488
1489 return TRUE;
1490}
1491
1492STATIC SV*
1493S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
5c0f85ef 1494 const regnode_charclass* const node)
b423522f
KW
1495{
1496 /* Returns a mortal inversion list defining which code points are matched
1497 * by 'node', which is of type ANYOF. Handles complementing the result if
1498 * appropriate. If some code points aren't knowable at this time, the
31f05a37
KW
1499 * returned list must, and will, contain every code point that is a
1500 * possibility. */
b423522f 1501
e2506fa7 1502 SV* invlist = NULL;
1ee208c4 1503 SV* only_utf8_locale_invlist = NULL;
b423522f
KW
1504 unsigned int i;
1505 const U32 n = ARG(node);
31f05a37 1506 bool new_node_has_latin1 = FALSE;
b423522f
KW
1507
1508 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1509
1510 /* Look at the data structure created by S_set_ANYOF_arg() */
93e92956 1511 if (n != ANYOF_ONLY_HAS_BITMAP) {
b423522f
KW
1512 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1513 AV * const av = MUTABLE_AV(SvRV(rv));
1514 SV **const ary = AvARRAY(av);
1515 assert(RExC_rxi->data->what[n] == 's');
1516
1517 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
28118b9c 1518 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1]), NULL));
b423522f
KW
1519 }
1520 else if (ary[0] && ary[0] != &PL_sv_undef) {
1521
1522 /* Here, no compile-time swash, and there are things that won't be
1523 * known until runtime -- we have to assume it could be anything */
e2506fa7 1524 invlist = sv_2mortal(_new_invlist(1));
b423522f
KW
1525 return _add_range_to_invlist(invlist, 0, UV_MAX);
1526 }
1ee208c4 1527 else if (ary[3] && ary[3] != &PL_sv_undef) {
b423522f
KW
1528
1529 /* Here no compile-time swash, and no run-time only data. Use the
1530 * node's inversion list */
28118b9c 1531 invlist = sv_2mortal(invlist_clone(ary[3], NULL));
1ee208c4
KW
1532 }
1533
1534 /* Get the code points valid only under UTF-8 locales */
037715a6 1535 if ((ANYOF_FLAGS(node) & ANYOFL_FOLD)
1ee208c4
KW
1536 && ary[2] && ary[2] != &PL_sv_undef)
1537 {
1538 only_utf8_locale_invlist = ary[2];
b423522f
KW
1539 }
1540 }
1541
e2506fa7
KW
1542 if (! invlist) {
1543 invlist = sv_2mortal(_new_invlist(0));
1544 }
1545
dcb20b36
KW
1546 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1547 * code points, and an inversion list for the others, but if there are code
1548 * points that should match only conditionally on the target string being
1549 * UTF-8, those are placed in the inversion list, and not the bitmap.
1550 * Since there are circumstances under which they could match, they are
1551 * included in the SSC. But if the ANYOF node is to be inverted, we have
1552 * to exclude them here, so that when we invert below, the end result
1553 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1554 * have to do this here before we add the unconditionally matched code
1555 * points */
b423522f
KW
1556 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1557 _invlist_intersection_complement_2nd(invlist,
1558 PL_UpperLatin1,
1559 &invlist);
1560 }
1561
1562 /* Add in the points from the bit map */
dcb20b36 1563 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
b423522f 1564 if (ANYOF_BITMAP_TEST(node, i)) {
6f8848d5
TC
1565 unsigned int start = i++;
1566
1567 for (; i < NUM_ANYOF_CODE_POINTS && ANYOF_BITMAP_TEST(node, i); ++i) {
1568 /* empty */
1569 }
1570 invlist = _add_range_to_invlist(invlist, start, i-1);
31f05a37 1571 new_node_has_latin1 = TRUE;
b423522f
KW
1572 }
1573 }
1574
1575 /* If this can match all upper Latin1 code points, have to add them
ac33c516
KW
1576 * as well. But don't add them if inverting, as when that gets done below,
1577 * it would exclude all these characters, including the ones it shouldn't
1578 * that were added just above */
1579 if (! (ANYOF_FLAGS(node) & ANYOF_INVERT) && OP(node) == ANYOFD
f240c685
KW
1580 && (ANYOF_FLAGS(node) & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER))
1581 {
b423522f
KW
1582 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1583 }
1584
1585 /* Similarly for these */
93e92956 1586 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
e0a1ff7a 1587 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
b423522f
KW
1588 }
1589
1590 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1591 _invlist_invert(invlist);
1592 }
037715a6 1593 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOFL_FOLD) {
31f05a37
KW
1594
1595 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1596 * locale. We can skip this if there are no 0-255 at all. */
1597 _invlist_union(invlist, PL_Latin1, &invlist);
1598 }
1599
1ee208c4
KW
1600 /* Similarly add the UTF-8 locale possible matches. These have to be
1601 * deferred until after the non-UTF-8 locale ones are taken care of just
1602 * above, or it leads to wrong results under ANYOF_INVERT */
1603 if (only_utf8_locale_invlist) {
31f05a37 1604 _invlist_union_maybe_complement_2nd(invlist,
1ee208c4 1605 only_utf8_locale_invlist,
31f05a37
KW
1606 ANYOF_FLAGS(node) & ANYOF_INVERT,
1607 &invlist);
1608 }
b423522f
KW
1609
1610 return invlist;
1611}
1612
1051e1c4 1613/* These two functions currently do the exact same thing */
557bd3fb 1614#define ssc_init_zero ssc_init
653099ff 1615
cdd87c1d
KW
1616#define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1617#define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1618
557bd3fb 1619/* 'AND' a given class with another one. Can create false positives. 'ssc'
93e92956
KW
1620 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1621 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
cdd87c1d 1622
653099ff 1623STATIC void
b423522f 1624S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1625 const regnode_charclass *and_with)
653099ff 1626{
cdd87c1d
KW
1627 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1628 * another SSC or a regular ANYOF class. Can create false positives. */
40d049e4 1629
a0dd4231
KW
1630 SV* anded_cp_list;
1631 U8 anded_flags;
1e6ade67 1632
cdd87c1d 1633 PERL_ARGS_ASSERT_SSC_AND;
653099ff 1634
71068078 1635 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1636
1637 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1638 * the code point inversion list and just the relevant flags */
71068078 1639 if (is_ANYOF_SYNTHETIC(and_with)) {
7dcac5f6 1640 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
a0dd4231 1641 anded_flags = ANYOF_FLAGS(and_with);
e9b08962
KW
1642
1643 /* XXX This is a kludge around what appears to be deficiencies in the
1644 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1645 * there are paths through the optimizer where it doesn't get weeded
1646 * out when it should. And if we don't make some extra provision for
1647 * it like the code just below, it doesn't get added when it should.
1648 * This solution is to add it only when AND'ing, which is here, and
1649 * only when what is being AND'ed is the pristine, original node
1650 * matching anything. Thus it is like adding it to ssc_anything() but
1651 * only when the result is to be AND'ed. Probably the same solution
1652 * could be adopted for the same problem we have with /l matching,
1653 * which is solved differently in S_ssc_init(), and that would lead to
1654 * fewer false positives than that solution has. But if this solution
1655 * creates bugs, the consequences are only that a warning isn't raised
1656 * that should be; while the consequences for having /l bugs is
1657 * incorrect matches */
7dcac5f6 1658 if (ssc_is_anything((regnode_ssc *)and_with)) {
f240c685 1659 anded_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
e9b08962 1660 }
a0dd4231
KW
1661 }
1662 else {
5c0f85ef 1663 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
f240c685
KW
1664 if (OP(and_with) == ANYOFD) {
1665 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1666 }
1667 else {
1668 anded_flags = ANYOF_FLAGS(and_with)
1669 &( ANYOF_COMMON_FLAGS
108316fb
KW
1670 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1671 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1672 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(and_with))) {
1673 anded_flags &=
1674 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1675 }
f240c685 1676 }
a0dd4231
KW
1677 }
1678
1679 ANYOF_FLAGS(ssc) &= anded_flags;
cdd87c1d
KW
1680
1681 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1682 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1683 * 'and_with' may be inverted. When not inverted, we have the situation of
1684 * computing:
1685 * (C1 | P1) & (C2 | P2)
1686 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1687 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1688 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1689 * <= ((C1 & C2) | P1 | P2)
1690 * Alternatively, the last few steps could be:
1691 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1692 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1693 * <= (C1 | C2 | (P1 & P2))
1694 * We favor the second approach if either P1 or P2 is non-empty. This is
1695 * because these components are a barrier to doing optimizations, as what
1696 * they match cannot be known until the moment of matching as they are
1697 * dependent on the current locale, 'AND"ing them likely will reduce or
1698 * eliminate them.
1699 * But we can do better if we know that C1,P1 are in their initial state (a
1700 * frequent occurrence), each matching everything:
1701 * (<everything>) & (C2 | P2) = C2 | P2
1702 * Similarly, if C2,P2 are in their initial state (again a frequent
1703 * occurrence), the result is a no-op
1704 * (C1 | P1) & (<everything>) = C1 | P1
1705 *
1706 * Inverted, we have
1707 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1708 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1709 * <= (C1 & ~C2) | (P1 & ~P2)
1710 * */
1aa99e6b 1711
a0dd4231 1712 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
71068078 1713 && ! is_ANYOF_SYNTHETIC(and_with))
a0dd4231 1714 {
cdd87c1d 1715 unsigned int i;
8951c461 1716
cdd87c1d
KW
1717 ssc_intersection(ssc,
1718 anded_cp_list,
1719 FALSE /* Has already been inverted */
1720 );
c6b76537 1721
cdd87c1d
KW
1722 /* If either P1 or P2 is empty, the intersection will be also; can skip
1723 * the loop */
93e92956 1724 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
cdd87c1d
KW
1725 ANYOF_POSIXL_ZERO(ssc);
1726 }
e0e1be5f 1727 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1728
1729 /* Note that the Posix class component P from 'and_with' actually
1730 * looks like:
1731 * P = Pa | Pb | ... | Pn
1732 * where each component is one posix class, such as in [\w\s].
1733 * Thus
1734 * ~P = ~(Pa | Pb | ... | Pn)
1735 * = ~Pa & ~Pb & ... & ~Pn
1736 * <= ~Pa | ~Pb | ... | ~Pn
1737 * The last is something we can easily calculate, but unfortunately
1738 * is likely to have many false positives. We could do better
1739 * in some (but certainly not all) instances if two classes in
1740 * P have known relationships. For example
1741 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1742 * So
1743 * :lower: & :print: = :lower:
1744 * And similarly for classes that must be disjoint. For example,
1745 * since \s and \w can have no elements in common based on rules in
1746 * the POSIX standard,
1747 * \w & ^\S = nothing
1748 * Unfortunately, some vendor locales do not meet the Posix
1749 * standard, in particular almost everything by Microsoft.
1750 * The loop below just changes e.g., \w into \W and vice versa */
1751
1ee208c4 1752 regnode_charclass_posixl temp;
cdd87c1d
KW
1753 int add = 1; /* To calculate the index of the complement */
1754
b1234259 1755 Zero(&temp, 1, regnode_charclass_posixl);
cdd87c1d
KW
1756 ANYOF_POSIXL_ZERO(&temp);
1757 for (i = 0; i < ANYOF_MAX; i++) {
1758 assert(i % 2 != 0
7dcac5f6
KW
1759 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1760 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
cdd87c1d 1761
7dcac5f6 1762 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
cdd87c1d
KW
1763 ANYOF_POSIXL_SET(&temp, i + add);
1764 }
1765 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1766 }
1767 ANYOF_POSIXL_AND(&temp, ssc);
c6b76537 1768
cdd87c1d
KW
1769 } /* else ssc already has no posixes */
1770 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1771 in its initial state */
71068078 1772 else if (! is_ANYOF_SYNTHETIC(and_with)
7dcac5f6 1773 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
cdd87c1d
KW
1774 {
1775 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1776 * copy it over 'ssc' */
1777 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
71068078 1778 if (is_ANYOF_SYNTHETIC(and_with)) {
cdd87c1d
KW
1779 StructCopy(and_with, ssc, regnode_ssc);
1780 }
1781 else {
1782 ssc->invlist = anded_cp_list;
1783 ANYOF_POSIXL_ZERO(ssc);
93e92956 1784 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1785 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
cdd87c1d
KW
1786 }
1787 }
1788 }
e0e1be5f 1789 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
93e92956 1790 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
cdd87c1d
KW
1791 {
1792 /* One or the other of P1, P2 is non-empty. */
93e92956 1793 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1ea8b7fe
KW
1794 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1795 }
cdd87c1d
KW
1796 ssc_union(ssc, anded_cp_list, FALSE);
1797 }
1798 else { /* P1 = P2 = empty */
1799 ssc_intersection(ssc, anded_cp_list, FALSE);
1800 }
137165a6 1801 }
653099ff
GS
1802}
1803
653099ff 1804STATIC void
cdd87c1d 1805S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1806 const regnode_charclass *or_with)
653099ff 1807{
cdd87c1d
KW
1808 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1809 * another SSC or a regular ANYOF class. Can create false positives if
1810 * 'or_with' is to be inverted. */
7918f24d 1811
a0dd4231
KW
1812 SV* ored_cp_list;
1813 U8 ored_flags;
c6b76537 1814
cdd87c1d 1815 PERL_ARGS_ASSERT_SSC_OR;
c6b76537 1816
71068078 1817 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1818
1819 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1820 * the code point inversion list and just the relevant flags */
71068078 1821 if (is_ANYOF_SYNTHETIC(or_with)) {
7dcac5f6 1822 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
a0dd4231
KW
1823 ored_flags = ANYOF_FLAGS(or_with);
1824 }
1825 else {
5c0f85ef 1826 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
eff8b7dc 1827 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
f240c685
KW
1828 if (OP(or_with) != ANYOFD) {
1829 ored_flags
1830 |= ANYOF_FLAGS(or_with)
108316fb
KW
1831 & ( ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1832 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1833 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(or_with))) {
1834 ored_flags |=
1835 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1836 }
f240c685 1837 }
a0dd4231
KW
1838 }
1839
1840 ANYOF_FLAGS(ssc) |= ored_flags;
cdd87c1d
KW
1841
1842 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1843 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1844 * 'or_with' may be inverted. When not inverted, we have the simple
1845 * situation of computing:
1846 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1847 * If P1|P2 yields a situation with both a class and its complement are
1848 * set, like having both \w and \W, this matches all code points, and we
1849 * can delete these from the P component of the ssc going forward. XXX We
1850 * might be able to delete all the P components, but I (khw) am not certain
1851 * about this, and it is better to be safe.
1852 *
1853 * Inverted, we have
1854 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1855 * <= (C1 | P1) | ~C2
1856 * <= (C1 | ~C2) | P1
1857 * (which results in actually simpler code than the non-inverted case)
1858 * */
9826f543 1859
a0dd4231 1860 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
71068078 1861 && ! is_ANYOF_SYNTHETIC(or_with))
a0dd4231 1862 {
cdd87c1d 1863 /* We ignore P2, leaving P1 going forward */
1ea8b7fe 1864 } /* else Not inverted */
93e92956 1865 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1866 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
e0e1be5f 1867 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1868 unsigned int i;
1869 for (i = 0; i < ANYOF_MAX; i += 2) {
1870 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1871 {
1872 ssc_match_all_cp(ssc);
1873 ANYOF_POSIXL_CLEAR(ssc, i);
1874 ANYOF_POSIXL_CLEAR(ssc, i+1);
cdd87c1d
KW
1875 }
1876 }
1877 }
1aa99e6b 1878 }
cdd87c1d
KW
1879
1880 ssc_union(ssc,
1881 ored_cp_list,
1882 FALSE /* Already has been inverted */
1883 );
653099ff
GS
1884}
1885
b423522f
KW
1886PERL_STATIC_INLINE void
1887S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1888{
1889 PERL_ARGS_ASSERT_SSC_UNION;
1890
71068078 1891 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1892
1893 _invlist_union_maybe_complement_2nd(ssc->invlist,
1894 invlist,
1895 invert2nd,
1896 &ssc->invlist);
1897}
1898
1899PERL_STATIC_INLINE void
1900S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1901 SV* const invlist,
1902 const bool invert2nd)
1903{
1904 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1905
71068078 1906 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1907
1908 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1909 invlist,
1910 invert2nd,
1911 &ssc->invlist);
1912}
1913
1914PERL_STATIC_INLINE void
1915S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1916{
1917 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1918
71068078 1919 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1920
1921 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1922}
1923
1924PERL_STATIC_INLINE void
1925S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1926{
1927 /* AND just the single code point 'cp' into the SSC 'ssc' */
1928
1929 SV* cp_list = _new_invlist(2);
1930
1931 PERL_ARGS_ASSERT_SSC_CP_AND;
1932
71068078 1933 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1934
1935 cp_list = add_cp_to_invlist(cp_list, cp);
1936 ssc_intersection(ssc, cp_list,
1937 FALSE /* Not inverted */
1938 );
1939 SvREFCNT_dec_NN(cp_list);
1940}
1941
1942PERL_STATIC_INLINE void
dc3bf405 1943S_ssc_clear_locale(regnode_ssc *ssc)
b423522f
KW
1944{
1945 /* Set the SSC 'ssc' to not match any locale things */
b423522f
KW
1946 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
1947
71068078 1948 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1949
1950 ANYOF_POSIXL_ZERO(ssc);
1951 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
1952}
1953
b35552de
KW
1954#define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
1955
1956STATIC bool
1957S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
1958{
1959 /* The synthetic start class is used to hopefully quickly winnow down
1960 * places where a pattern could start a match in the target string. If it
1961 * doesn't really narrow things down that much, there isn't much point to
1962 * having the overhead of using it. This function uses some very crude
1963 * heuristics to decide if to use the ssc or not.
1964 *
1965 * It returns TRUE if 'ssc' rules out more than half what it considers to
1966 * be the "likely" possible matches, but of course it doesn't know what the
1967 * actual things being matched are going to be; these are only guesses
1968 *
1969 * For /l matches, it assumes that the only likely matches are going to be
1970 * in the 0-255 range, uniformly distributed, so half of that is 127
1971 * For /a and /d matches, it assumes that the likely matches will be just
1972 * the ASCII range, so half of that is 63
1973 * For /u and there isn't anything matching above the Latin1 range, it
1974 * assumes that that is the only range likely to be matched, and uses
1975 * half that as the cut-off: 127. If anything matches above Latin1,
1976 * it assumes that all of Unicode could match (uniformly), except for
1977 * non-Unicode code points and things in the General Category "Other"
1978 * (unassigned, private use, surrogates, controls and formats). This
1979 * is a much large number. */
1980
b35552de
KW
1981 U32 count = 0; /* Running total of number of code points matched by
1982 'ssc' */
1983 UV start, end; /* Start and end points of current range in inversion
1984 list */
72400949
KW
1985 const U32 max_code_points = (LOC)
1986 ? 256
1987 : (( ! UNI_SEMANTICS
1988 || invlist_highest(ssc->invlist) < 256)
1989 ? 128
1990 : NON_OTHER_COUNT);
1991 const U32 max_match = max_code_points / 2;
b35552de
KW
1992
1993 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
1994
1995 invlist_iterinit(ssc->invlist);
1996 while (invlist_iternext(ssc->invlist, &start, &end)) {
72400949
KW
1997 if (start >= max_code_points) {
1998 break;
b35552de 1999 }
72400949 2000 end = MIN(end, max_code_points - 1);
b35552de 2001 count += end - start + 1;
72400949 2002 if (count >= max_match) {
b35552de
KW
2003 invlist_iterfinish(ssc->invlist);
2004 return FALSE;
2005 }
2006 }
2007
2008 return TRUE;
2009}
2010
2011
b423522f
KW
2012STATIC void
2013S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
2014{
2015 /* The inversion list in the SSC is marked mortal; now we need a more
2016 * permanent copy, which is stored the same way that is done in a regular
dcb20b36
KW
2017 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
2018 * map */
b423522f 2019
28118b9c 2020 SV* invlist = invlist_clone(ssc->invlist, NULL);
b423522f
KW
2021
2022 PERL_ARGS_ASSERT_SSC_FINALIZE;
2023
71068078 2024 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f 2025
a0dd4231 2026 /* The code in this file assumes that all but these flags aren't relevant
93e92956
KW
2027 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
2028 * by the time we reach here */
f240c685
KW
2029 assert(! (ANYOF_FLAGS(ssc)
2030 & ~( ANYOF_COMMON_FLAGS
108316fb
KW
2031 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
2032 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)));
a0dd4231 2033
b423522f
KW
2034 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
2035
1ee208c4
KW
2036 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
2037 NULL, NULL, NULL, FALSE);
b423522f 2038
85c8e306
KW
2039 /* Make sure is clone-safe */
2040 ssc->invlist = NULL;
2041
e0e1be5f 2042 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
93e92956 2043 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
d156f5cb 2044 OP(ssc) = ANYOFPOSIXL;
e0e1be5f 2045 }
d156f5cb 2046 else if (RExC_contains_locale) {
b2e90ddf
KW
2047 OP(ssc) = ANYOFL;
2048 }
2049
1462525b 2050 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
b423522f
KW
2051}
2052
a3621e74
YO
2053#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
2054#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
2055#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
538e84ed
KW
2056#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
2057 ? (TRIE_LIST_CUR( idx ) - 1) \
2058 : 0 )
a3621e74 2059
3dab1dad
YO
2060
2061#ifdef DEBUGGING
07be1b83 2062/*
2b8b4781
NC
2063 dump_trie(trie,widecharmap,revcharmap)
2064 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
2065 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
2066
2067 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
2068 The _interim_ variants are used for debugging the interim
2069 tables that are used to generate the final compressed
2070 representation which is what dump_trie expects.
2071
486ec47a 2072 Part of the reason for their existence is to provide a form
3dab1dad 2073 of documentation as to how the different representations function.
07be1b83
YO
2074
2075*/
3dab1dad
YO
2076
2077/*
3dab1dad
YO
2078 Dumps the final compressed table form of the trie to Perl_debug_log.
2079 Used for debugging make_trie().
2080*/
b9a59e08 2081
3dab1dad 2082STATIC void
2b8b4781
NC
2083S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
2084 AV *revcharmap, U32 depth)
3dab1dad
YO
2085{
2086 U32 state;
ab3bbdeb 2087 SV *sv=sv_newmortal();
55eed653 2088 int colwidth= widecharmap ? 6 : 4;
2e64971a 2089 U16 word;
3dab1dad
YO
2090 GET_RE_DEBUG_FLAGS_DECL;
2091
7918f24d 2092 PERL_ARGS_ASSERT_DUMP_TRIE;
ab3bbdeb 2093
6ad9a8ab 2094 Perl_re_indentf( aTHX_ "Char : %-6s%-6s%-4s ",
1e37780e 2095 depth+1, "Match","Base","Ofs" );
3dab1dad
YO
2096
2097 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 2098 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 2099 if ( tmp ) {
6ad9a8ab 2100 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2101 colwidth,
538e84ed 2102 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2103 PL_colors[0], PL_colors[1],
2104 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2105 PERL_PV_ESCAPE_FIRSTCHAR
2106 )
ab3bbdeb 2107 );
3dab1dad
YO
2108 }
2109 }
1e37780e
YO
2110 Perl_re_printf( aTHX_ "\n");
2111 Perl_re_indentf( aTHX_ "State|-----------------------", depth+1);
3dab1dad
YO
2112
2113 for( state = 0 ; state < trie->uniquecharcount ; state++ )
6ad9a8ab
YO
2114 Perl_re_printf( aTHX_ "%.*s", colwidth, "--------");
2115 Perl_re_printf( aTHX_ "\n");
3dab1dad 2116
1e2e3d02 2117 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 2118 const U32 base = trie->states[ state ].trans.base;
3dab1dad 2119
147e3846 2120 Perl_re_indentf( aTHX_ "#%4" UVXf "|", depth+1, (UV)state);
3dab1dad
YO
2121
2122 if ( trie->states[ state ].wordnum ) {
1e37780e 2123 Perl_re_printf( aTHX_ " W%4X", trie->states[ state ].wordnum );
3dab1dad 2124 } else {
6ad9a8ab 2125 Perl_re_printf( aTHX_ "%6s", "" );
3dab1dad
YO
2126 }
2127
147e3846 2128 Perl_re_printf( aTHX_ " @%4" UVXf " ", (UV)base );
3dab1dad
YO
2129
2130 if ( base ) {
2131 U32 ofs = 0;
2132
2133 while( ( base + ofs < trie->uniquecharcount ) ||
2134 ( base + ofs - trie->uniquecharcount < trie->lasttrans
538e84ed
KW
2135 && trie->trans[ base + ofs - trie->uniquecharcount ].check
2136 != state))
3dab1dad
YO
2137 ofs++;
2138
147e3846 2139 Perl_re_printf( aTHX_ "+%2" UVXf "[ ", (UV)ofs);
3dab1dad
YO
2140
2141 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
538e84ed
KW
2142 if ( ( base + ofs >= trie->uniquecharcount )
2143 && ( base + ofs - trie->uniquecharcount
2144 < trie->lasttrans )
2145 && trie->trans[ base + ofs
2146 - trie->uniquecharcount ].check == state )
3dab1dad 2147 {
147e3846 2148 Perl_re_printf( aTHX_ "%*" UVXf, colwidth,
1e37780e
YO
2149 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next
2150 );
3dab1dad 2151 } else {
88f063b4 2152 Perl_re_printf( aTHX_ "%*s", colwidth," ." );
3dab1dad
YO
2153 }
2154 }
2155
6ad9a8ab 2156 Perl_re_printf( aTHX_ "]");
3dab1dad
YO
2157
2158 }
6ad9a8ab 2159 Perl_re_printf( aTHX_ "\n" );
3dab1dad 2160 }
6ad9a8ab 2161 Perl_re_indentf( aTHX_ "word_info N:(prev,len)=",
cb41e5d6 2162 depth);
2e64971a 2163 for (word=1; word <= trie->wordcount; word++) {
6ad9a8ab 2164 Perl_re_printf( aTHX_ " %d:(%d,%d)",
2e64971a
DM
2165 (int)word, (int)(trie->wordinfo[word].prev),
2166 (int)(trie->wordinfo[word].len));
2167 }
6ad9a8ab 2168 Perl_re_printf( aTHX_ "\n" );
538e84ed 2169}
3dab1dad 2170/*
3dab1dad 2171 Dumps a fully constructed but uncompressed trie in list form.
538e84ed 2172 List tries normally only are used for construction when the number of
3dab1dad
YO
2173 possible chars (trie->uniquecharcount) is very high.
2174 Used for debugging make_trie().
2175*/
2176STATIC void
55eed653 2177S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2178 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2179 U32 depth)
3dab1dad
YO
2180{
2181 U32 state;
ab3bbdeb 2182 SV *sv=sv_newmortal();
55eed653 2183 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2184 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2185
2186 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
2187
3dab1dad 2188 /* print out the table precompression. */
6ad9a8ab 2189 Perl_re_indentf( aTHX_ "State :Word | Transition Data\n",
cb41e5d6 2190 depth+1 );
6ad9a8ab 2191 Perl_re_indentf( aTHX_ "%s",
cb41e5d6 2192 depth+1, "------:-----+-----------------\n" );
538e84ed 2193
3dab1dad
YO
2194 for( state=1 ; state < next_alloc ; state ++ ) {
2195 U16 charid;
538e84ed 2196
147e3846 2197 Perl_re_indentf( aTHX_ " %4" UVXf " :",
cb41e5d6 2198 depth+1, (UV)state );
3dab1dad 2199 if ( ! trie->states[ state ].wordnum ) {
6ad9a8ab 2200 Perl_re_printf( aTHX_ "%5s| ","");
3dab1dad 2201 } else {
6ad9a8ab 2202 Perl_re_printf( aTHX_ "W%4x| ",
3dab1dad
YO
2203 trie->states[ state ].wordnum
2204 );
2205 }
2206 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
538e84ed 2207 SV ** const tmp = av_fetch( revcharmap,
88f063b4 2208 TRIE_LIST_ITEM(state, charid).forid, 0);
ab3bbdeb 2209 if ( tmp ) {
147e3846 2210 Perl_re_printf( aTHX_ "%*s:%3X=%4" UVXf " | ",
ab3bbdeb 2211 colwidth,
538e84ed
KW
2212 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
2213 colwidth,
2214 PL_colors[0], PL_colors[1],
2215 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2216 | PERL_PV_ESCAPE_FIRSTCHAR
ab3bbdeb 2217 ) ,
88f063b4
KW
2218 TRIE_LIST_ITEM(state, charid).forid,
2219 (UV)TRIE_LIST_ITEM(state, charid).newstate
1e2e3d02 2220 );
538e84ed 2221 if (!(charid % 10))
6ad9a8ab 2222 Perl_re_printf( aTHX_ "\n%*s| ",
664e119d 2223 (int)((depth * 2) + 14), "");
1e2e3d02 2224 }
ab3bbdeb 2225 }
6ad9a8ab 2226 Perl_re_printf( aTHX_ "\n");
3dab1dad 2227 }
538e84ed 2228}
3dab1dad
YO
2229
2230/*
3dab1dad 2231 Dumps a fully constructed but uncompressed trie in table form.
538e84ed
KW
2232 This is the normal DFA style state transition table, with a few
2233 twists to facilitate compression later.
3dab1dad
YO
2234 Used for debugging make_trie().
2235*/
2236STATIC void
55eed653 2237S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2238 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2239 U32 depth)
3dab1dad
YO
2240{
2241 U32 state;
2242 U16 charid;
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_TABLE;
538e84ed 2248
3dab1dad
YO
2249 /*
2250 print out the table precompression so that we can do a visual check
2251 that they are identical.
2252 */
538e84ed 2253
6ad9a8ab 2254 Perl_re_indentf( aTHX_ "Char : ", depth+1 );
3dab1dad
YO
2255
2256 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 2257 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 2258 if ( tmp ) {
6ad9a8ab 2259 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2260 colwidth,
538e84ed 2261 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2262 PL_colors[0], PL_colors[1],
2263 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2264 PERL_PV_ESCAPE_FIRSTCHAR
2265 )
ab3bbdeb 2266 );
3dab1dad
YO
2267 }
2268 }
2269
4aaafc03
YO
2270 Perl_re_printf( aTHX_ "\n");
2271 Perl_re_indentf( aTHX_ "State+-", depth+1 );
3dab1dad
YO
2272
2273 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
6ad9a8ab 2274 Perl_re_printf( aTHX_ "%.*s", colwidth,"--------");
3dab1dad
YO
2275 }
2276
6ad9a8ab 2277 Perl_re_printf( aTHX_ "\n" );
3dab1dad
YO
2278
2279 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
2280
147e3846 2281 Perl_re_indentf( aTHX_ "%4" UVXf " : ",
cb41e5d6 2282 depth+1,
3dab1dad
YO
2283 (UV)TRIE_NODENUM( state ) );
2284
2285 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
2286 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
2287 if (v)
147e3846 2288 Perl_re_printf( aTHX_ "%*" UVXf, colwidth, v );
ab3bbdeb 2289 else
6ad9a8ab 2290 Perl_re_printf( aTHX_ "%*s", colwidth, "." );
3dab1dad
YO
2291 }
2292 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
147e3846 2293 Perl_re_printf( aTHX_ " (%4" UVXf ")\n",
538e84ed 2294 (UV)trie->trans[ state ].check );
3dab1dad 2295 } else {
147e3846 2296 Perl_re_printf( aTHX_ " (%4" UVXf ") W%4X\n",
538e84ed 2297 (UV)trie->trans[ state ].check,
3dab1dad
YO
2298 trie->states[ TRIE_NODENUM( state ) ].wordnum );
2299 }
2300 }
07be1b83 2301}
3dab1dad
YO
2302
2303#endif
2304
2e64971a 2305
786e8c11
YO
2306/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
2307 startbranch: the first branch in the whole branch sequence
2308 first : start branch of sequence of branch-exact nodes.
2309 May be the same as startbranch
2310 last : Thing following the last branch.
2311 May be the same as tail.
2312 tail : item following the branch sequence
2313 count : words in the sequence
a4525e78 2314 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
786e8c11 2315 depth : indent depth
3dab1dad 2316
786e8c11 2317Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 2318
786e8c11
YO
2319A trie is an N'ary tree where the branches are determined by digital
2320decomposition of the key. IE, at the root node you look up the 1st character and
2321follow that branch repeat until you find the end of the branches. Nodes can be
2322marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 2323
786e8c11 2324 /he|she|his|hers/
72f13be8 2325
786e8c11
YO
2326would convert into the following structure. Numbers represent states, letters
2327following numbers represent valid transitions on the letter from that state, if
2328the number is in square brackets it represents an accepting state, otherwise it
2329will be in parenthesis.
07be1b83 2330
786e8c11
YO
2331 +-h->+-e->[3]-+-r->(8)-+-s->[9]
2332 | |
2333 | (2)
2334 | |
2335 (1) +-i->(6)-+-s->[7]
2336 |
2337 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 2338
786e8c11
YO
2339 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
2340
2341This shows that when matching against the string 'hers' we will begin at state 1
2342read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
2343then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
2344is also accepting. Thus we know that we can match both 'he' and 'hers' with a
2345single traverse. We store a mapping from accepting to state to which word was
2346matched, and then when we have multiple possibilities we try to complete the
b8fda935 2347rest of the regex in the order in which they occurred in the alternation.
786e8c11
YO
2348
2349The only prior NFA like behaviour that would be changed by the TRIE support is
2350the silent ignoring of duplicate alternations which are of the form:
2351
2352 / (DUPE|DUPE) X? (?{ ... }) Y /x
2353
4b714af6 2354Thus EVAL blocks following a trie may be called a different number of times with
786e8c11 2355and without the optimisation. With the optimisations dupes will be silently
486ec47a 2356ignored. This inconsistent behaviour of EVAL type nodes is well established as
786e8c11
YO
2357the following demonstrates:
2358
2359 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
2360
2361which prints out 'word' three times, but
2362
2363 'words'=~/(word|word|word)(?{ print $1 })S/
2364
2365which doesnt print it out at all. This is due to other optimisations kicking in.
2366
2367Example of what happens on a structural level:
2368
486ec47a 2369The regexp /(ac|ad|ab)+/ will produce the following debug output:
786e8c11
YO
2370
2371 1: CURLYM[1] {1,32767}(18)
2372 5: BRANCH(8)
2373 6: EXACT <ac>(16)
2374 8: BRANCH(11)
2375 9: EXACT <ad>(16)
2376 11: BRANCH(14)
2377 12: EXACT <ab>(16)
2378 16: SUCCEED(0)
2379 17: NOTHING(18)
2380 18: END(0)
2381
2382This would be optimizable with startbranch=5, first=5, last=16, tail=16
2383and should turn into:
2384
2385 1: CURLYM[1] {1,32767}(18)
2386 5: TRIE(16)
2387 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
2388 <ac>
2389 <ad>
2390 <ab>
2391 16: SUCCEED(0)
2392 17: NOTHING(18)
2393 18: END(0)
2394
2395Cases where tail != last would be like /(?foo|bar)baz/:
2396
2397 1: BRANCH(4)
2398 2: EXACT <foo>(8)
2399 4: BRANCH(7)
2400 5: EXACT <bar>(8)
2401 7: TAIL(8)
2402 8: EXACT <baz>(10)
2403 10: END(0)
2404
2405which would be optimizable with startbranch=1, first=1, last=7, tail=8
2406and would end up looking like:
2407
2408 1: TRIE(8)
2409 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
2410 <foo>
2411 <bar>
2412 7: TAIL(8)
2413 8: EXACT <baz>(10)
2414 10: END(0)
2415
c80e42f3 2416 d = uvchr_to_utf8_flags(d, uv, 0);
786e8c11
YO
2417
2418is the recommended Unicode-aware way of saying
2419
2420 *(d++) = uv;
2421*/
2422
fab2782b 2423#define TRIE_STORE_REVCHAR(val) \
786e8c11 2424 STMT_START { \
73031816 2425 if (UTF) { \
668fcfea 2426 SV *zlopp = newSV(UTF8_MAXBYTES); \
88c9ea1e 2427 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
c80e42f3 2428 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
73031816
NC
2429 SvCUR_set(zlopp, kapow - flrbbbbb); \
2430 SvPOK_on(zlopp); \
2431 SvUTF8_on(zlopp); \
2432 av_push(revcharmap, zlopp); \
2433 } else { \
fab2782b 2434 char ooooff = (char)val; \
73031816
NC
2435 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2436 } \
2437 } STMT_END
786e8c11 2438
914a25d5
KW
2439/* This gets the next character from the input, folding it if not already
2440 * folded. */
2441#define TRIE_READ_CHAR STMT_START { \
2442 wordlen++; \
2443 if ( UTF ) { \
2444 /* if it is UTF then it is either already folded, or does not need \
2445 * folding */ \
1c1d615a 2446 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
914a25d5
KW
2447 } \
2448 else if (folder == PL_fold_latin1) { \
7d006b13
KW
2449 /* This folder implies Unicode rules, which in the range expressible \
2450 * by not UTF is the lower case, with the two exceptions, one of \
2451 * which should have been taken care of before calling this */ \
2452 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2453 uvc = toLOWER_L1(*uc); \
2454 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2455 len = 1; \
914a25d5
KW
2456 } else { \
2457 /* raw data, will be folded later if needed */ \
2458 uvc = (U32)*uc; \
2459 len = 1; \
2460 } \
786e8c11
YO
2461} STMT_END
2462
2463
2464
2465#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2466 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
00195859 2467 U32 ging = TRIE_LIST_LEN( state ) * 2; \
f9003953 2468 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
00195859 2469 TRIE_LIST_LEN( state ) = ging; \
786e8c11
YO
2470 } \
2471 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2472 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2473 TRIE_LIST_CUR( state )++; \
2474} STMT_END
07be1b83 2475
786e8c11 2476#define TRIE_LIST_NEW(state) STMT_START { \
d09f14bf 2477 Newx( trie->states[ state ].trans.list, \
786e8c11
YO
2478 4, reg_trie_trans_le ); \
2479 TRIE_LIST_CUR( state ) = 1; \
2480 TRIE_LIST_LEN( state ) = 4; \
2481} STMT_END
07be1b83 2482
786e8c11
YO
2483#define TRIE_HANDLE_WORD(state) STMT_START { \
2484 U16 dupe= trie->states[ state ].wordnum; \
2485 regnode * const noper_next = regnext( noper ); \
2486 \
786e8c11
YO
2487 DEBUG_r({ \
2488 /* store the word for dumping */ \
2489 SV* tmp; \
2490 if (OP(noper) != NOTHING) \
740cce10 2491 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
786e8c11 2492 else \
740cce10 2493 tmp = newSVpvn_utf8( "", 0, UTF ); \
2b8b4781 2494 av_push( trie_words, tmp ); \
786e8c11
YO
2495 }); \
2496 \
2497 curword++; \
2e64971a
DM
2498 trie->wordinfo[curword].prev = 0; \
2499 trie->wordinfo[curword].len = wordlen; \
2500 trie->wordinfo[curword].accept = state; \
786e8c11
YO
2501 \
2502 if ( noper_next < tail ) { \
2503 if (!trie->jump) \
538e84ed
KW
2504 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2505 sizeof(U16) ); \
7f69552c 2506 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
2507 if (!jumper) \
2508 jumper = noper_next; \
2509 if (!nextbranch) \
2510 nextbranch= regnext(cur); \
2511 } \
2512 \
2513 if ( dupe ) { \
2e64971a
DM
2514 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2515 /* chain, so that when the bits of chain are later */\
2516 /* linked together, the dups appear in the chain */\
2517 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2518 trie->wordinfo[dupe].prev = curword; \
786e8c11
YO
2519 } else { \
2520 /* we haven't inserted this word yet. */ \
2521 trie->states[ state ].wordnum = curword; \
2522 } \
2523} STMT_END
07be1b83 2524
3dab1dad 2525
786e8c11
YO
2526#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2527 ( ( base + charid >= ucharcount \
2528 && base + charid < ubound \
2529 && state == trie->trans[ base - ucharcount + charid ].check \
2530 && trie->trans[ base - ucharcount + charid ].next ) \
2531 ? trie->trans[ base - ucharcount + charid ].next \
2532 : ( state==1 ? special : 0 ) \
2533 )
3dab1dad 2534
8bcafbf4
YO
2535#define TRIE_BITMAP_SET_FOLDED(trie, uvc, folder) \
2536STMT_START { \
2537 TRIE_BITMAP_SET(trie, uvc); \
2538 /* store the folded codepoint */ \
2539 if ( folder ) \
2540 TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); \
2541 \
2542 if ( !UTF ) { \
2543 /* store first byte of utf8 representation of */ \
2544 /* variant codepoints */ \
2545 if (! UVCHR_IS_INVARIANT(uvc)) { \
2546 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); \
2547 } \
2548 } \
2549} STMT_END
786e8c11
YO
2550#define MADE_TRIE 1
2551#define MADE_JUMP_TRIE 2
2552#define MADE_EXACT_TRIE 4
3dab1dad 2553
a3621e74 2554STATIC I32
538e84ed
KW
2555S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2556 regnode *first, regnode *last, regnode *tail,
2557 U32 word_count, U32 flags, U32 depth)
a3621e74
YO
2558{
2559 /* first pass, loop through and scan words */
2560 reg_trie_data *trie;
55eed653 2561 HV *widecharmap = NULL;
2b8b4781 2562 AV *revcharmap = newAV();
a3621e74 2563 regnode *cur;
a3621e74
YO
2564 STRLEN len = 0;
2565 UV uvc = 0;
2566 U16 curword = 0;
2567 U32 next_alloc = 0;
786e8c11
YO
2568 regnode *jumper = NULL;
2569 regnode *nextbranch = NULL;
7f69552c 2570 regnode *convert = NULL;
2e64971a 2571 U32 *prev_states; /* temp array mapping each state to previous one */
a3621e74 2572 /* we just use folder as a flag in utf8 */
1e696034 2573 const U8 * folder = NULL;
a3621e74 2574
3a611511
YO
2575 /* in the below add_data call we are storing either 'tu' or 'tuaa'
2576 * which stands for one trie structure, one hash, optionally followed
2577 * by two arrays */
2b8b4781 2578#ifdef DEBUGGING
3a611511 2579 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuaa"));
2b8b4781
NC
2580 AV *trie_words = NULL;
2581 /* along with revcharmap, this only used during construction but both are
2582 * useful during debugging so we store them in the struct when debugging.
8e11feef 2583 */
2b8b4781 2584#else
cf78de0b 2585 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
3dab1dad 2586 STRLEN trie_charcount=0;
3dab1dad 2587#endif
2b8b4781 2588 SV *re_trie_maxbuff;
a3621e74 2589 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2590
2591 PERL_ARGS_ASSERT_MAKE_TRIE;
72f13be8
YO
2592#ifndef DEBUGGING
2593 PERL_UNUSED_ARG(depth);
2594#endif
a3621e74 2595
1e696034 2596 switch (flags) {
a4525e78 2597 case EXACT: case EXACTL: break;
89829bb5 2598 case EXACTFAA:
fab2782b 2599 case EXACTFU_SS:
a4525e78
KW
2600 case EXACTFU:
2601 case EXACTFLU8: folder = PL_fold_latin1; break;
1e696034 2602 case EXACTF: folder = PL_fold; break;
fab2782b 2603 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
1e696034
KW
2604 }
2605
c944940b 2606 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 2607 trie->refcount = 1;
3dab1dad 2608 trie->startstate = 1;
786e8c11 2609 trie->wordcount = word_count;
f8fc2ecf 2610 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 2611 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
a4525e78 2612 if (flags == EXACT || flags == EXACTL)
c944940b 2613 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2e64971a
DM
2614 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2615 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2616
a3621e74 2617 DEBUG_r({
2b8b4781 2618 trie_words = newAV();
a3621e74 2619 });
a3621e74 2620
0111c4fd 2621 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
316ebaf2 2622 assert(re_trie_maxbuff);
a3621e74 2623 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 2624 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 2625 }
df826430 2626 DEBUG_TRIE_COMPILE_r({
6ad9a8ab 2627 Perl_re_indentf( aTHX_
cb41e5d6
YO
2628 "make_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2629 depth+1,
88f063b4 2630 REG_NODE_NUM(startbranch), REG_NODE_NUM(first),
538e84ed 2631 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
3dab1dad 2632 });
538e84ed 2633
7f69552c
YO
2634 /* Find the node we are going to overwrite */
2635 if ( first == startbranch && OP( last ) != BRANCH ) {
2636 /* whole branch chain */
2637 convert = first;
2638 } else {
2639 /* branch sub-chain */
2640 convert = NEXTOPER( first );
2641 }
538e84ed 2642
a3621e74
YO
2643 /* -- First loop and Setup --
2644
2645 We first traverse the branches and scan each word to determine if it
2646 contains widechars, and how many unique chars there are, this is
2647 important as we have to build a table with at least as many columns as we
2648 have unique chars.
2649
2650 We use an array of integers to represent the character codes 0..255
538e84ed
KW
2651 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2652 the native representation of the character value as the key and IV's for
2653 the coded index.
a3621e74
YO
2654
2655 *TODO* If we keep track of how many times each character is used we can
2656 remap the columns so that the table compression later on is more
3b753521 2657 efficient in terms of memory by ensuring the most common value is in the
a3621e74
YO
2658 middle and the least common are on the outside. IMO this would be better
2659 than a most to least common mapping as theres a decent chance the most
2660 common letter will share a node with the least common, meaning the node
486ec47a 2661 will not be compressible. With a middle is most common approach the worst
a3621e74
YO
2662 case is when we have the least common nodes twice.
2663
2664 */
2665
a3621e74 2666 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
df826430 2667 regnode *noper = NEXTOPER( cur );
944e05e3
YO
2668 const U8 *uc;
2669 const U8 *e;
bc031a7d 2670 int foldlen = 0;
07be1b83 2671 U32 wordlen = 0; /* required init */
bc031a7d
KW
2672 STRLEN minchars = 0;
2673 STRLEN maxchars = 0;
538e84ed
KW
2674 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2675 bitmap?*/
a3621e74 2676
3dab1dad 2677 if (OP(noper) == NOTHING) {
20ed8c88
YO
2678 /* skip past a NOTHING at the start of an alternation
2679 * eg, /(?:)a|(?:b)/ should be the same as /a|b/
2680 */
df826430 2681 regnode *noper_next= regnext(noper);
944e05e3
YO
2682 if (noper_next < tail)
2683 noper= noper_next;
2684 }
2685
dca5fc4c
YO
2686 if ( noper < tail &&
2687 (
2688 OP(noper) == flags ||
2689 (
2690 flags == EXACTFU &&
2691 OP(noper) == EXACTFU_SS
2692 )
2693 )
2694 ) {
944e05e3
YO
2695 uc= (U8*)STRING(noper);
2696 e= uc + STR_LEN(noper);
2697 } else {
2698 trie->minlen= 0;
2699 continue;
3dab1dad 2700 }
df826430 2701
944e05e3 2702
fab2782b 2703 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
02daf0ab
YO
2704 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2705 regardless of encoding */
fab2782b
YO
2706 if (OP( noper ) == EXACTFU_SS) {
2707 /* false positives are ok, so just set this */
0dc4a61d 2708 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
fab2782b
YO
2709 }
2710 }
dca5fc4c 2711
bc031a7d
KW
2712 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2713 branch */
3dab1dad 2714 TRIE_CHARCOUNT(trie)++;
a3621e74 2715 TRIE_READ_CHAR;
645de4ce 2716
bc031a7d
KW
2717 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2718 * is in effect. Under /i, this character can match itself, or
2719 * anything that folds to it. If not under /i, it can match just
2720 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2721 * all fold to k, and all are single characters. But some folds
2722 * expand to more than one character, so for example LATIN SMALL
2723 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2724 * the string beginning at 'uc' is 'ffi', it could be matched by
2725 * three characters, or just by the one ligature character. (It
2726 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2727 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2728 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2729 * match.) The trie needs to know the minimum and maximum number
2730 * of characters that could match so that it can use size alone to
2731 * quickly reject many match attempts. The max is simple: it is
2732 * the number of folded characters in this branch (since a fold is
2733 * never shorter than what folds to it. */
2734
2735 maxchars++;
2736
2737 /* And the min is equal to the max if not under /i (indicated by
2738 * 'folder' being NULL), or there are no multi-character folds. If
2739 * there is a multi-character fold, the min is incremented just
2740 * once, for the character that folds to the sequence. Each
2741 * character in the sequence needs to be added to the list below of
2742 * characters in the trie, but we count only the first towards the
2743 * min number of characters needed. This is done through the
2744 * variable 'foldlen', which is returned by the macros that look
2745 * for these sequences as the number of bytes the sequence
2746 * occupies. Each time through the loop, we decrement 'foldlen' by
2747 * how many bytes the current char occupies. Only when it reaches
2748 * 0 do we increment 'minchars' or look for another multi-character
2749 * sequence. */
2750 if (folder == NULL) {
2751 minchars++;
2752 }
2753 else if (foldlen > 0) {
2754 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
645de4ce
KW
2755 }
2756 else {
bc031a7d
KW
2757 minchars++;
2758
2759 /* See if *uc is the beginning of a multi-character fold. If
2760 * so, we decrement the length remaining to look at, to account
2761 * for the current character this iteration. (We can use 'uc'
2762 * instead of the fold returned by TRIE_READ_CHAR because for
2763 * non-UTF, the latin1_safe macro is smart enough to account
2764 * for all the unfolded characters, and because for UTF, the
2765 * string will already have been folded earlier in the
2766 * compilation process */
2767 if (UTF) {
2768 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2769 foldlen -= UTF8SKIP(uc);
645de4ce
KW
2770 }
2771 }
bc031a7d
KW
2772 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2773 foldlen--;
2774 }
645de4ce 2775 }
bc031a7d
KW
2776
2777 /* The current character (and any potential folds) should be added
2778 * to the possible matching characters for this position in this
2779 * branch */
a3621e74 2780 if ( uvc < 256 ) {
fab2782b
YO
2781 if ( folder ) {
2782 U8 folded= folder[ (U8) uvc ];
2783 if ( !trie->charmap[ folded ] ) {
2784 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2785 TRIE_STORE_REVCHAR( folded );
2786 }
2787 }
a3621e74
YO
2788 if ( !trie->charmap[ uvc ] ) {
2789 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
fab2782b 2790 TRIE_STORE_REVCHAR( uvc );
a3621e74 2791 }
02daf0ab 2792 if ( set_bit ) {
62012aee
KW
2793 /* store the codepoint in the bitmap, and its folded
2794 * equivalent. */
8bcafbf4 2795 TRIE_BITMAP_SET_FOLDED(trie, uvc, folder);
02daf0ab
YO
2796 set_bit = 0; /* We've done our bit :-) */
2797 }
a3621e74 2798 } else {
bc031a7d
KW
2799
2800 /* XXX We could come up with the list of code points that fold
2801 * to this using PL_utf8_foldclosures, except not for
2802 * multi-char folds, as there may be multiple combinations
2803 * there that could work, which needs to wait until runtime to
2804 * resolve (The comment about LIGATURE FFI above is such an
2805 * example */
2806
a3621e74 2807 SV** svpp;
55eed653
NC
2808 if ( !widecharmap )
2809 widecharmap = newHV();
a3621e74 2810
55eed653 2811 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
2812
2813 if ( !svpp )
147e3846 2814 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%" UVXf, uvc );
a3621e74
YO
2815
2816 if ( !SvTRUE( *svpp ) ) {
2817 sv_setiv( *svpp, ++trie->uniquecharcount );
fab2782b 2818 TRIE_STORE_REVCHAR(uvc);
a3621e74
YO
2819 }
2820 }
bc031a7d
KW
2821 } /* end loop through characters in this branch of the trie */
2822
2823 /* We take the min and max for this branch and combine to find the min
2824 * and max for all branches processed so far */
3dab1dad 2825 if( cur == first ) {
bc031a7d
KW
2826 trie->minlen = minchars;
2827 trie->maxlen = maxchars;
2828 } else if (minchars < trie->minlen) {
2829 trie->minlen = minchars;
2830 } else if (maxchars > trie->maxlen) {
2831 trie->maxlen = maxchars;
fab2782b 2832 }
a3621e74
YO
2833 } /* end first pass */
2834 DEBUG_TRIE_COMPILE_r(
6ad9a8ab 2835 Perl_re_indentf( aTHX_
cb41e5d6
YO
2836 "TRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2837 depth+1,
55eed653 2838 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
2839 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2840 (int)trie->minlen, (int)trie->maxlen )
a3621e74 2841 );
a3621e74
YO
2842
2843 /*
2844 We now know what we are dealing with in terms of unique chars and
2845 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
2846 representation using a flat table will take. If it's over a reasonable
2847 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
2848 conservative but potentially much slower representation using an array
2849 of lists.
2850
2851 At the end we convert both representations into the same compressed
2852 form that will be used in regexec.c for matching with. The latter
2853 is a form that cannot be used to construct with but has memory
2854 properties similar to the list form and access properties similar
2855 to the table form making it both suitable for fast searches and
2856 small enough that its feasable to store for the duration of a program.
2857
2858 See the comment in the code where the compressed table is produced
2859 inplace from the flat tabe representation for an explanation of how
2860 the compression works.
2861
2862 */
2863
2864
2e64971a
DM
2865 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2866 prev_states[1] = 0;
2867
538e84ed
KW
2868 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2869 > SvIV(re_trie_maxbuff) )
2870 {
a3621e74
YO
2871 /*
2872 Second Pass -- Array Of Lists Representation
2873
2874 Each state will be represented by a list of charid:state records
2875 (reg_trie_trans_le) the first such element holds the CUR and LEN
2876 points of the allocated array. (See defines above).
2877
2878 We build the initial structure using the lists, and then convert
2879 it into the compressed table form which allows faster lookups
2880 (but cant be modified once converted).
a3621e74
YO
2881 */
2882
a3621e74
YO
2883 STRLEN transcount = 1;
2884
6ad9a8ab 2885 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using list compiler\n",
cb41e5d6 2886 depth+1));
686b73d4 2887
c944940b
JH
2888 trie->states = (reg_trie_state *)
2889 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2890 sizeof(reg_trie_state) );
a3621e74
YO
2891 TRIE_LIST_NEW(1);
2892 next_alloc = 2;
2893
2894 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2895
df826430 2896 regnode *noper = NEXTOPER( cur );
c445ea15
AL
2897 U32 state = 1; /* required init */
2898 U16 charid = 0; /* sanity init */
07be1b83 2899 U32 wordlen = 0; /* required init */
c445ea15 2900
df826430
YO
2901 if (OP(noper) == NOTHING) {
2902 regnode *noper_next= regnext(noper);
944e05e3
YO
2903 if (noper_next < tail)
2904 noper= noper_next;
df826430
YO
2905 }
2906
944e05e3
YO
2907 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
2908 const U8 *uc= (U8*)STRING(noper);
2909 const U8 *e= uc + STR_LEN(noper);
2910
786e8c11 2911 for ( ; uc < e ; uc += len ) {
c445ea15 2912
786e8c11 2913 TRIE_READ_CHAR;
c445ea15 2914
786e8c11
YO
2915 if ( uvc < 256 ) {
2916 charid = trie->charmap[ uvc ];
c445ea15 2917 } else {
538e84ed
KW
2918 SV** const svpp = hv_fetch( widecharmap,
2919 (char*)&uvc,
2920 sizeof( UV ),
2921 0);
786e8c11
YO
2922 if ( !svpp ) {
2923 charid = 0;
2924 } else {
2925 charid=(U16)SvIV( *svpp );
2926 }
c445ea15 2927 }
538e84ed
KW
2928 /* charid is now 0 if we dont know the char read, or
2929 * nonzero if we do */
786e8c11 2930 if ( charid ) {
a3621e74 2931
786e8c11
YO
2932 U16 check;
2933 U32 newstate = 0;
a3621e74 2934
786e8c11
YO
2935 charid--;
2936 if ( !trie->states[ state ].trans.list ) {
2937 TRIE_LIST_NEW( state );
c445ea15 2938 }
538e84ed
KW
2939 for ( check = 1;
2940 check <= TRIE_LIST_USED( state );
2941 check++ )
2942 {
2943 if ( TRIE_LIST_ITEM( state, check ).forid
2944 == charid )
2945 {
786e8c11
YO
2946 newstate = TRIE_LIST_ITEM( state, check ).newstate;
2947 break;
2948 }
2949 }
2950 if ( ! newstate ) {
2951 newstate = next_alloc++;
2e64971a 2952 prev_states[newstate] = state;
786e8c11
YO
2953 TRIE_LIST_PUSH( state, charid, newstate );
2954 transcount++;
2955 }
2956 state = newstate;
2957 } else {
147e3846 2958 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
c445ea15 2959 }
a28509cc 2960 }
c445ea15 2961 }
3dab1dad 2962 TRIE_HANDLE_WORD(state);
a3621e74
YO
2963
2964 } /* end second pass */
2965
1e2e3d02 2966 /* next alloc is the NEXT state to be allocated */
538e84ed 2967 trie->statecount = next_alloc;
c944940b
JH
2968 trie->states = (reg_trie_state *)
2969 PerlMemShared_realloc( trie->states,
2970 next_alloc
2971 * sizeof(reg_trie_state) );
a3621e74 2972
3dab1dad 2973 /* and now dump it out before we compress it */
2b8b4781
NC
2974 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
2975 revcharmap, next_alloc,
2976 depth+1)
1e2e3d02 2977 );
a3621e74 2978
c944940b
JH
2979 trie->trans = (reg_trie_trans *)
2980 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
2981 {
2982 U32 state;
a3621e74
YO
2983 U32 tp = 0;
2984 U32 zp = 0;
2985
2986
2987 for( state=1 ; state < next_alloc ; state ++ ) {
2988 U32 base=0;
2989
2990 /*
2991 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 2992 Perl_re_printf( aTHX_ "tp: %d zp: %d ",tp,zp)
a3621e74
YO
2993 );
2994 */
2995
2996 if (trie->states[state].trans.list) {
2997 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
2998 U16 maxid=minid;
a28509cc 2999 U16 idx;
a3621e74
YO
3000
3001 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
3002 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
3003 if ( forid < minid ) {
3004 minid=forid;
3005 } else if ( forid > maxid ) {
3006 maxid=forid;
3007 }
a3621e74
YO
3008 }
3009 if ( transcount < tp + maxid - minid + 1) {
3010 transcount *= 2;
c944940b
JH
3011 trie->trans = (reg_trie_trans *)
3012 PerlMemShared_realloc( trie->trans,
446bd890
NC
3013 transcount
3014 * sizeof(reg_trie_trans) );
538e84ed
KW
3015 Zero( trie->trans + (transcount / 2),
3016 transcount / 2,
3017 reg_trie_trans );
a3621e74
YO
3018 }
3019 base = trie->uniquecharcount + tp - minid;
3020 if ( maxid == minid ) {
3021 U32 set = 0;
3022 for ( ; zp < tp ; zp++ ) {
3023 if ( ! trie->trans[ zp ].next ) {
3024 base = trie->uniquecharcount + zp - minid;
538e84ed
KW
3025 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
3026 1).newstate;
a3621e74
YO
3027 trie->trans[ zp ].check = state;
3028 set = 1;
3029 break;
3030 }
3031 }
3032 if ( !set ) {
538e84ed
KW
3033 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
3034 1).newstate;
a3621e74
YO
3035 trie->trans[ tp ].check = state;
3036 tp++;
3037 zp = tp;
3038 }
3039 } else {
3040 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
538e84ed
KW
3041 const U32 tid = base
3042 - trie->uniquecharcount
3043 + TRIE_LIST_ITEM( state, idx ).forid;
3044 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
3045 idx ).newstate;
a3621e74
YO
3046 trie->trans[ tid ].check = state;
3047 }
3048 tp += ( maxid - minid + 1 );
3049 }
3050 Safefree(trie->states[ state ].trans.list);
3051 }
3052 /*
3053 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3054 Perl_re_printf( aTHX_ " base: %d\n",base);
a3621e74
YO
3055 );
3056 */
3057 trie->states[ state ].trans.base=base;
3058 }
cc601c31 3059 trie->lasttrans = tp + 1;
a3621e74
YO
3060 }
3061 } else {
3062 /*
3063 Second Pass -- Flat Table Representation.
3064
b423522f
KW
3065 we dont use the 0 slot of either trans[] or states[] so we add 1 to
3066 each. We know that we will need Charcount+1 trans at most to store
3067 the data (one row per char at worst case) So we preallocate both
3068 structures assuming worst case.
a3621e74
YO
3069
3070 We then construct the trie using only the .next slots of the entry
3071 structs.
3072
b423522f
KW
3073 We use the .check field of the first entry of the node temporarily
3074 to make compression both faster and easier by keeping track of how
3075 many non zero fields are in the node.
a3621e74
YO
3076
3077 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
3078 transition.
3079
b423522f
KW
3080 There are two terms at use here: state as a TRIE_NODEIDX() which is
3081 a number representing the first entry of the node, and state as a
3082 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
3083 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
3084 if there are 2 entrys per node. eg:
a3621e74
YO
3085
3086 A B A B
3087 1. 2 4 1. 3 7
3088 2. 0 3 3. 0 5
3089 3. 0 0 5. 0 0
3090 4. 0 0 7. 0 0
3091
b423522f
KW
3092 The table is internally in the right hand, idx form. However as we
3093 also have to deal with the states array which is indexed by nodenum
3094 we have to use TRIE_NODENUM() to convert.
a3621e74
YO
3095
3096 */
6ad9a8ab 3097 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using table compiler\n",
cb41e5d6 3098 depth+1));
3dab1dad 3099
c944940b
JH
3100 trie->trans = (reg_trie_trans *)
3101 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
3102 * trie->uniquecharcount + 1,
3103 sizeof(reg_trie_trans) );
3104 trie->states = (reg_trie_state *)
3105 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
3106 sizeof(reg_trie_state) );
a3621e74
YO
3107 next_alloc = trie->uniquecharcount + 1;
3108
3dab1dad 3109
a3621e74
YO
3110 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
3111
df826430 3112 regnode *noper = NEXTOPER( cur );
a3621e74
YO
3113
3114 U32 state = 1; /* required init */
3115
3116 U16 charid = 0; /* sanity init */
3117 U32 accept_state = 0; /* sanity init */
a3621e74 3118
07be1b83 3119 U32 wordlen = 0; /* required init */
a3621e74 3120
df826430
YO
3121 if (OP(noper) == NOTHING) {
3122 regnode *noper_next= regnext(noper);
944e05e3
YO
3123 if (noper_next < tail)
3124 noper= noper_next;
df826430 3125 }
fab2782b 3126
944e05e3
YO
3127 if ( noper < tail && ( OP(noper) == flags || ( flags == EXACTFU && OP(noper) == EXACTFU_SS ) ) ) {
3128 const U8 *uc= (U8*)STRING(noper);
3129 const U8 *e= uc + STR_LEN(noper);
3130
786e8c11 3131 for ( ; uc < e ; uc += len ) {
a3621e74 3132
786e8c11 3133 TRIE_READ_CHAR;
a3621e74 3134
786e8c11
YO
3135 if ( uvc < 256 ) {
3136 charid = trie->charmap[ uvc ];
3137 } else {
538e84ed
KW
3138 SV* const * const svpp = hv_fetch( widecharmap,
3139 (char*)&uvc,
3140 sizeof( UV ),
3141 0);
786e8c11 3142 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 3143 }
786e8c11
YO
3144 if ( charid ) {
3145 charid--;
3146 if ( !trie->trans[ state + charid ].next ) {
3147 trie->trans[ state + charid ].next = next_alloc;
3148 trie->trans[ state ].check++;
2e64971a
DM
3149 prev_states[TRIE_NODENUM(next_alloc)]
3150 = TRIE_NODENUM(state);
786e8c11
YO
3151 next_alloc += trie->uniquecharcount;
3152 }
3153 state = trie->trans[ state + charid ].next;
3154 } else {
147e3846 3155 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
786e8c11 3156 }
538e84ed
KW
3157 /* charid is now 0 if we dont know the char read, or
3158 * nonzero if we do */
a3621e74 3159 }
a3621e74 3160 }
3dab1dad
YO
3161 accept_state = TRIE_NODENUM( state );
3162 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
3163
3164 } /* end second pass */
3165
3dab1dad 3166 /* and now dump it out before we compress it */
2b8b4781
NC
3167 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
3168 revcharmap,
3169 next_alloc, depth+1));
a3621e74 3170
a3621e74
YO
3171 {
3172 /*
3173 * Inplace compress the table.*
3174
3175 For sparse data sets the table constructed by the trie algorithm will
3176 be mostly 0/FAIL transitions or to put it another way mostly empty.
3177 (Note that leaf nodes will not contain any transitions.)
3178
3179 This algorithm compresses the tables by eliminating most such
3180 transitions, at the cost of a modest bit of extra work during lookup:
3181
3182 - Each states[] entry contains a .base field which indicates the
3183 index in the state[] array wheres its transition data is stored.
3184
3b753521 3185 - If .base is 0 there are no valid transitions from that node.
a3621e74
YO
3186
3187 - If .base is nonzero then charid is added to it to find an entry in
3188 the trans array.
3189
3190 -If trans[states[state].base+charid].check!=state then the
3191 transition is taken to be a 0/Fail transition. Thus if there are fail
3192 transitions at the front of the node then the .base offset will point
3193 somewhere inside the previous nodes data (or maybe even into a node
3194 even earlier), but the .check field determines if the transition is
3195 valid.
3196
786e8c11 3197 XXX - wrong maybe?
a3621e74 3198 The following process inplace converts the table to the compressed
3b753521 3199 table: We first do not compress the root node 1,and mark all its
a3621e74 3200 .check pointers as 1 and set its .base pointer as 1 as well. This
3b753521
FN
3201 allows us to do a DFA construction from the compressed table later,
3202 and ensures that any .base pointers we calculate later are greater
3203 than 0.
a3621e74
YO
3204
3205 - We set 'pos' to indicate the first entry of the second node.
3206
3207 - We then iterate over the columns of the node, finding the first and
3208 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
3209 and set the .check pointers accordingly, and advance pos
3210 appropriately and repreat for the next node. Note that when we copy
3211 the next pointers we have to convert them from the original
3212 NODEIDX form to NODENUM form as the former is not valid post
3213 compression.
3214
3215 - If a node has no transitions used we mark its base as 0 and do not
3216 advance the pos pointer.
3217
3218 - If a node only has one transition we use a second pointer into the
3219 structure to fill in allocated fail transitions from other states.
3220 This pointer is independent of the main pointer and scans forward
3221 looking for null transitions that are allocated to a state. When it
3222 finds one it writes the single transition into the "hole". If the
786e8c11 3223 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
3224
3225 - Once compressed we can Renew/realloc the structures to release the
3226 excess space.
3227
3228 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
3229 specifically Fig 3.47 and the associated pseudocode.
3230
3231 demq
3232 */
a3b680e6 3233 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 3234 U32 state, charid;
a3621e74 3235 U32 pos = 0, zp=0;
1e2e3d02 3236 trie->statecount = laststate;
a3621e74
YO
3237
3238 for ( state = 1 ; state < laststate ; state++ ) {
3239 U8 flag = 0;
a28509cc
AL
3240 const U32 stateidx = TRIE_NODEIDX( state );
3241 const U32 o_used = trie->trans[ stateidx ].check;
3242 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
3243 trie->trans[ stateidx ].check = 0;
3244
538e84ed
KW
3245 for ( charid = 0;
3246 used && charid < trie->uniquecharcount;
3247 charid++ )
3248 {
a3621e74
YO
3249 if ( flag || trie->trans[ stateidx + charid ].next ) {
3250 if ( trie->trans[ stateidx + charid ].next ) {
3251 if (o_used == 1) {
3252 for ( ; zp < pos ; zp++ ) {
3253 if ( ! trie->trans[ zp ].next ) {
3254 break;
3255 }
3256 }
538e84ed
KW
3257 trie->states[ state ].trans.base
3258 = zp
3259 + trie->uniquecharcount
3260 - charid ;
3261 trie->trans[ zp ].next
3262 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
3263 + charid ].next );
a3621e74
YO
3264 trie->trans[ zp ].check = state;
3265 if ( ++zp > pos ) pos = zp;
3266 break;
3267 }
3268 used--;
3269 }
3270 if ( !flag ) {
3271 flag = 1;
538e84ed
KW
3272 trie->states[ state ].trans.base
3273 = pos + trie->uniquecharcount - charid ;
a3621e74 3274 }
538e84ed
KW
3275 trie->trans[ pos ].next
3276 = SAFE_TRIE_NODENUM(
3277 trie->trans[ stateidx + charid ].next );
a3621e74
YO
3278 trie->trans[ pos ].check = state;
3279 pos++;
3280 }
3281 }
3282 }
cc601c31 3283 trie->lasttrans = pos + 1;
c944940b
JH
3284 trie->states = (reg_trie_state *)
3285 PerlMemShared_realloc( trie->states, laststate
3286 * sizeof(reg_trie_state) );
a3621e74 3287 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3288 Perl_re_indentf( aTHX_ "Alloc: %d Orig: %" IVdf " elements, Final:%" IVdf ". Savings of %%%5.2f\n",
cb41e5d6 3289 depth+1,
538e84ed
KW
3290 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
3291 + 1 ),
3292 (IV)next_alloc,
3293 (IV)pos,
3294 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
a3621e74
YO
3295 );
3296
3297 } /* end table compress */
3298 }
1e2e3d02 3299 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3300 Perl_re_indentf( aTHX_ "Statecount:%" UVxf " Lasttrans:%" UVxf "\n",
cb41e5d6 3301 depth+1,
1e2e3d02
YO
3302 (UV)trie->statecount,
3303 (UV)trie->lasttrans)
3304 );
cc601c31 3305 /* resize the trans array to remove unused space */
c944940b
JH
3306 trie->trans = (reg_trie_trans *)
3307 PerlMemShared_realloc( trie->trans, trie->lasttrans
3308 * sizeof(reg_trie_trans) );
a3621e74 3309
538e84ed 3310 { /* Modify the program and insert the new TRIE node */
3dab1dad
YO
3311 U8 nodetype =(U8)(flags & 0xFF);
3312 char *str=NULL;
538e84ed 3313
07be1b83 3314#ifdef DEBUGGING
e62cc96a 3315 regnode *optimize = NULL;
7122b237
YO
3316#ifdef RE_TRACK_PATTERN_OFFSETS
3317
b57a0404
JH
3318 U32 mjd_offset = 0;
3319 U32 mjd_nodelen = 0;
7122b237
YO
3320#endif /* RE_TRACK_PATTERN_OFFSETS */
3321#endif /* DEBUGGING */
a3621e74 3322 /*
3dab1dad
YO
3323 This means we convert either the first branch or the first Exact,
3324 depending on whether the thing following (in 'last') is a branch
3325 or not and whther first is the startbranch (ie is it a sub part of
3326 the alternation or is it the whole thing.)
3b753521 3327 Assuming its a sub part we convert the EXACT otherwise we convert
3dab1dad 3328 the whole branch sequence, including the first.
a3621e74 3329 */
3dab1dad 3330 /* Find the node we are going to overwrite */
7f69552c 3331 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 3332 /* branch sub-chain */
3dab1dad 3333 NEXT_OFF( first ) = (U16)(last - first);
7122b237 3334#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
3335 DEBUG_r({
3336 mjd_offset= Node_Offset((convert));
3337 mjd_nodelen= Node_Length((convert));
3338 });
7122b237 3339#endif
7f69552c 3340 /* whole branch chain */
7122b237
YO
3341 }
3342#ifdef RE_TRACK_PATTERN_OFFSETS
3343 else {
7f69552c
YO
3344 DEBUG_r({
3345 const regnode *nop = NEXTOPER( convert );
3346 mjd_offset= Node_Offset((nop));
3347 mjd_nodelen= Node_Length((nop));
3348 });
07be1b83
YO
3349 }
3350 DEBUG_OPTIMISE_r(
147e3846 3351 Perl_re_indentf( aTHX_ "MJD offset:%" UVuf " MJD length:%" UVuf "\n",
cb41e5d6 3352 depth+1,
786e8c11 3353 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 3354 );
7122b237 3355#endif
538e84ed 3356 /* But first we check to see if there is a common prefix we can
3dab1dad
YO
3357 split out as an EXACT and put in front of the TRIE node. */
3358 trie->startstate= 1;
55eed653 3359 if ( trie->bitmap && !widecharmap && !trie->jump ) {
5ee57374
YO
3360 /* we want to find the first state that has more than
3361 * one transition, if that state is not the first state
3362 * then we have a common prefix which we can remove.
3363 */
3dab1dad 3364 U32 state;
1e2e3d02 3365 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
a3621e74 3366 U32 ofs = 0;
ee0dfd0b
YO
3367 I32 first_ofs = -1; /* keeps track of the ofs of the first
3368 transition, -1 means none */
8e11feef
RGS
3369 U32 count = 0;
3370 const U32 base = trie->states[ state ].trans.base;
a3621e74 3371
5ee57374 3372 /* does this state terminate an alternation? */
3dab1dad 3373 if ( trie->states[state].wordnum )
8e11feef 3374 count = 1;
a3621e74 3375
8e11feef 3376 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
cc601c31
YO
3377 if ( ( base + ofs >= trie->uniquecharcount ) &&
3378 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
a3621e74
YO
3379 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
3380 {
3dab1dad 3381 if ( ++count > 1 ) {
5ee57374 3382 /* we have more than one transition */
d3d91c7c
YO
3383 SV **tmp;
3384 U8 *ch;
5ee57374
YO
3385 /* if this is the first state there is no common prefix
3386 * to extract, so we can exit */
8e11feef 3387 if ( state == 1 ) break;
d3d91c7c
YO
3388 tmp = av_fetch( revcharmap, ofs, 0);
3389 ch = (U8*)SvPV_nolen_const( *tmp );
3390
5ee57374
YO
3391 /* if we are on count 2 then we need to initialize the
3392 * bitmap, and store the previous char if there was one
3393 * in it*/
3dab1dad 3394 if ( count == 2 ) {
5ee57374 3395 /* clear the bitmap */
3dab1dad
YO
3396 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
3397 DEBUG_OPTIMISE_r(
147e3846 3398 Perl_re_indentf( aTHX_ "New Start State=%" UVuf " Class: [",
cb41e5d6 3399 depth+1,
786e8c11 3400 (UV)state));
ee0dfd0b
YO
3401 if (first_ofs >= 0) {
3402 SV ** const tmp = av_fetch( revcharmap, first_ofs, 0);
be8e71aa 3403 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 3404
88f063b4 3405 TRIE_BITMAP_SET_FOLDED(trie,*ch, folder);
3dab1dad 3406 DEBUG_OPTIMISE_r(
6ad9a8ab 3407 Perl_re_printf( aTHX_ "%s", (char*)ch)
3dab1dad 3408 );
8e11feef
RGS
3409 }
3410 }
8bcafbf4 3411 /* store the current firstchar in the bitmap */
88f063b4 3412 TRIE_BITMAP_SET_FOLDED(trie,*ch, folder);
6ad9a8ab 3413 DEBUG_OPTIMISE_r(Perl_re_printf( aTHX_ "%s", ch));
8e11feef 3414 }
ee0dfd0b 3415 first_ofs = ofs;
8e11feef 3416 }
3dab1dad
YO
3417 }
3418 if ( count == 1 ) {
5ee57374
YO
3419 /* This state has only one transition, its transition is part
3420 * of a common prefix - we need to concatenate the char it
3421 * represents to what we have so far. */
ee0dfd0b 3422 SV **tmp = av_fetch( revcharmap, first_ofs, 0);
c490c714
YO
3423 STRLEN len;
3424 char *ch = SvPV( *tmp, len );
de734bd5
A
3425 DEBUG_OPTIMISE_r({
3426 SV *sv=sv_newmortal();
147e3846 3427 Perl_re_indentf( aTHX_ "Prefix State: %" UVuf " Ofs:%" UVuf " Char='%s'\n",
cb41e5d6 3428 depth+1,
ee0dfd0b 3429 (UV)state, (UV)first_ofs,
538e84ed 3430 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
de734bd5
A
3431 PL_colors[0], PL_colors[1],
3432 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed 3433 PERL_PV_ESCAPE_FIRSTCHAR
de734bd5
A
3434 )
3435 );
3436 });
3dab1dad
YO
3437 if ( state==1 ) {
3438 OP( convert ) = nodetype;
3439 str=STRING(convert);
3440 STR_LEN(convert)=0;
3441 }
c490c714
YO
3442 STR_LEN(convert) += len;
3443 while (len--)
de734bd5 3444 *str++ = *ch++;
8e11feef 3445 } else {
538e84ed 3446#ifdef DEBUGGING
8e11feef 3447 if (state>1)
6ad9a8ab