Commit | Line | Data |
---|---|---|
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" | |
40b6423c | 84 | extern const struct regexp_engine my_reg_engine; |
54df2634 NC |
85 | #else |
86 | # include "regcomp.h" | |
87 | #endif | |
a687059c | 88 | |
04e98a4d | 89 | #include "dquote_static.c" |
26faadbd | 90 | #include "charclass_invlists.h" |
81e983c1 | 91 | #include "inline_invlist.c" |
1b0f46bf | 92 | #include "unicode_constants.h" |
04e98a4d | 93 | |
94dc5c2d | 94 | #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i) |
26faadbd | 95 | #define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c) |
94dc5c2d | 96 | |
d4cce5f1 | 97 | #ifdef op |
11343788 | 98 | #undef op |
d4cce5f1 | 99 | #endif /* op */ |
11343788 | 100 | |
fe14fcc3 | 101 | #ifdef MSDOS |
7e4e8c89 | 102 | # if defined(BUGGY_MSC6) |
fe14fcc3 | 103 | /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */ |
7e4e8c89 | 104 | # pragma optimize("a",off) |
fe14fcc3 | 105 | /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/ |
7e4e8c89 NC |
106 | # pragma optimize("w",on ) |
107 | # endif /* BUGGY_MSC6 */ | |
fe14fcc3 LW |
108 | #endif /* MSDOS */ |
109 | ||
a687059c LW |
110 | #ifndef STATIC |
111 | #define STATIC static | |
112 | #endif | |
113 | ||
b1603ef8 | 114 | |
830247a4 | 115 | typedef struct RExC_state_t { |
514a91f1 DM |
116 | U32 flags; /* RXf_* are we folding, multilining? */ |
117 | U32 pm_flags; /* PMf_* stuff from the calling PMOP */ | |
830247a4 | 118 | char *precomp; /* uncompiled string. */ |
288b8c02 | 119 | REGEXP *rx_sv; /* The SV that is the regexp. */ |
f8fc2ecf YO |
120 | regexp *rx; /* perl core regexp structure */ |
121 | regexp_internal *rxi; /* internal data for regexp object pprivate field */ | |
fac92740 | 122 | char *start; /* Start of input for compile */ |
830247a4 IZ |
123 | char *end; /* End of input for compile */ |
124 | char *parse; /* Input-scan pointer. */ | |
125 | I32 whilem_seen; /* number of WHILEM in this expr */ | |
fac92740 | 126 | regnode *emit_start; /* Start of emitted-code area */ |
3b57cd43 | 127 | regnode *emit_bound; /* First regnode outside of the allocated space */ |
ffc61ed2 | 128 | regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */ |
830247a4 IZ |
129 | I32 naughty; /* How bad is this pattern? */ |
130 | I32 sawback; /* Did we see \1, ...? */ | |
131 | U32 seen; | |
132 | I32 size; /* Code size. */ | |
c74340f9 YO |
133 | I32 npar; /* Capture buffer count, (OPEN). */ |
134 | I32 cpar; /* Capture buffer count, (CLOSE). */ | |
e2e6a0f1 | 135 | I32 nestroot; /* root parens we are in - used by accept */ |
830247a4 IZ |
136 | I32 extralen; |
137 | I32 seen_zerolen; | |
40d049e4 YO |
138 | regnode **open_parens; /* pointers to open parens */ |
139 | regnode **close_parens; /* pointers to close parens */ | |
140 | regnode *opend; /* END node in program */ | |
02daf0ab YO |
141 | I32 utf8; /* whether the pattern is utf8 or not */ |
142 | I32 orig_utf8; /* whether the pattern was originally in utf8 */ | |
143 | /* XXX use this for future optimisation of case | |
144 | * where pattern must be upgraded to utf8. */ | |
e40e74fe KW |
145 | I32 uni_semantics; /* If a d charset modifier should use unicode |
146 | rules, even if the pattern is not in | |
147 | utf8 */ | |
81714fb9 | 148 | HV *paren_names; /* Paren names */ |
1f1031fe | 149 | |
40d049e4 YO |
150 | regnode **recurse; /* Recurse regops */ |
151 | I32 recurse_count; /* Number of recurse regops */ | |
b57e4118 | 152 | I32 in_lookbehind; |
4624b182 | 153 | I32 contains_locale; |
bb3f3ed2 | 154 | I32 override_recoding; |
3d2bd50a | 155 | struct reg_code_block *code_blocks; /* positions of literal (?{}) |
68e2671b | 156 | within pattern */ |
b1603ef8 DM |
157 | int num_code_blocks; /* size of code_blocks[] */ |
158 | int code_index; /* next code_blocks[] slot */ | |
830247a4 IZ |
159 | #if ADD_TO_REGEXEC |
160 | char *starttry; /* -Dr: where regtry was called. */ | |
161 | #define RExC_starttry (pRExC_state->starttry) | |
162 | #endif | |
d24ca0c5 | 163 | SV *runtime_code_qr; /* qr with the runtime code blocks */ |
3dab1dad | 164 | #ifdef DEBUGGING |
be8e71aa | 165 | const char *lastparse; |
3dab1dad | 166 | I32 lastnum; |
1f1031fe | 167 | AV *paren_name_list; /* idx -> name */ |
3dab1dad YO |
168 | #define RExC_lastparse (pRExC_state->lastparse) |
169 | #define RExC_lastnum (pRExC_state->lastnum) | |
1f1031fe | 170 | #define RExC_paren_name_list (pRExC_state->paren_name_list) |
3dab1dad | 171 | #endif |
830247a4 IZ |
172 | } RExC_state_t; |
173 | ||
e2509266 | 174 | #define RExC_flags (pRExC_state->flags) |
514a91f1 | 175 | #define RExC_pm_flags (pRExC_state->pm_flags) |
830247a4 | 176 | #define RExC_precomp (pRExC_state->precomp) |
288b8c02 | 177 | #define RExC_rx_sv (pRExC_state->rx_sv) |
830247a4 | 178 | #define RExC_rx (pRExC_state->rx) |
f8fc2ecf | 179 | #define RExC_rxi (pRExC_state->rxi) |
fac92740 | 180 | #define RExC_start (pRExC_state->start) |
830247a4 IZ |
181 | #define RExC_end (pRExC_state->end) |
182 | #define RExC_parse (pRExC_state->parse) | |
183 | #define RExC_whilem_seen (pRExC_state->whilem_seen) | |
7122b237 YO |
184 | #ifdef RE_TRACK_PATTERN_OFFSETS |
185 | #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */ | |
186 | #endif | |
830247a4 | 187 | #define RExC_emit (pRExC_state->emit) |
fac92740 | 188 | #define RExC_emit_start (pRExC_state->emit_start) |
3b57cd43 | 189 | #define RExC_emit_bound (pRExC_state->emit_bound) |
830247a4 IZ |
190 | #define RExC_naughty (pRExC_state->naughty) |
191 | #define RExC_sawback (pRExC_state->sawback) | |
192 | #define RExC_seen (pRExC_state->seen) | |
193 | #define RExC_size (pRExC_state->size) | |
194 | #define RExC_npar (pRExC_state->npar) | |
e2e6a0f1 | 195 | #define RExC_nestroot (pRExC_state->nestroot) |
830247a4 IZ |
196 | #define RExC_extralen (pRExC_state->extralen) |
197 | #define RExC_seen_zerolen (pRExC_state->seen_zerolen) | |
1aa99e6b | 198 | #define RExC_utf8 (pRExC_state->utf8) |
e40e74fe | 199 | #define RExC_uni_semantics (pRExC_state->uni_semantics) |
02daf0ab | 200 | #define RExC_orig_utf8 (pRExC_state->orig_utf8) |
40d049e4 YO |
201 | #define RExC_open_parens (pRExC_state->open_parens) |
202 | #define RExC_close_parens (pRExC_state->close_parens) | |
203 | #define RExC_opend (pRExC_state->opend) | |
81714fb9 | 204 | #define RExC_paren_names (pRExC_state->paren_names) |
40d049e4 YO |
205 | #define RExC_recurse (pRExC_state->recurse) |
206 | #define RExC_recurse_count (pRExC_state->recurse_count) | |
b57e4118 | 207 | #define RExC_in_lookbehind (pRExC_state->in_lookbehind) |
4624b182 | 208 | #define RExC_contains_locale (pRExC_state->contains_locale) |
bb3f3ed2 | 209 | #define RExC_override_recoding (pRExC_state->override_recoding) |
830247a4 | 210 | |
cde0cee5 | 211 | |
a687059c LW |
212 | #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?') |
213 | #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \ | |
214 | ((*s) == '{' && regcurly(s))) | |
a687059c | 215 | |
35c8bce7 LW |
216 | #ifdef SPSTART |
217 | #undef SPSTART /* dratted cpp namespace... */ | |
218 | #endif | |
a687059c LW |
219 | /* |
220 | * Flags to be passed up and down. | |
221 | */ | |
a687059c | 222 | #define WORST 0 /* Worst case. */ |
a3b492c3 | 223 | #define HASWIDTH 0x01 /* Known to match non-null strings. */ |
fda99bee | 224 | |
18890cbd | 225 | /* Simple enough to be STAR/PLUS operand; in an EXACT node must be a single |
cf8c372d KW |
226 | * character, and if utf8, must be invariant. Note that this is not the same |
227 | * thing as REGNODE_SIMPLE */ | |
fda99bee | 228 | #define SIMPLE 0x02 |
a3b492c3 YO |
229 | #define SPSTART 0x04 /* Starts with * or +. */ |
230 | #define TRYAGAIN 0x08 /* Weeded out a declaration. */ | |
231 | #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */ | |
a687059c | 232 | |
3dab1dad YO |
233 | #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1) |
234 | ||
07be1b83 YO |
235 | /* whether trie related optimizations are enabled */ |
236 | #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION | |
237 | #define TRIE_STUDY_OPT | |
786e8c11 | 238 | #define FULL_TRIE_STUDY |
07be1b83 YO |
239 | #define TRIE_STCLASS |
240 | #endif | |
1de06328 YO |
241 | |
242 | ||
40d049e4 YO |
243 | |
244 | #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3] | |
245 | #define PBITVAL(paren) (1 << ((paren) & 7)) | |
246 | #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren)) | |
247 | #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren) | |
248 | #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren)) | |
249 | ||
bbd61b5f KW |
250 | /* If not already in utf8, do a longjmp back to the beginning */ |
251 | #define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */ | |
252 | #define REQUIRE_UTF8 STMT_START { \ | |
253 | if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \ | |
254 | } STMT_END | |
40d049e4 | 255 | |
1de06328 YO |
256 | /* About scan_data_t. |
257 | ||
258 | During optimisation we recurse through the regexp program performing | |
259 | various inplace (keyhole style) optimisations. In addition study_chunk | |
260 | and scan_commit populate this data structure with information about | |
261 | what strings MUST appear in the pattern. We look for the longest | |
3b753521 | 262 | string that must appear at a fixed location, and we look for the |
1de06328 YO |
263 | longest string that may appear at a floating location. So for instance |
264 | in the pattern: | |
265 | ||
266 | /FOO[xX]A.*B[xX]BAR/ | |
267 | ||
268 | Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating | |
269 | strings (because they follow a .* construct). study_chunk will identify | |
270 | both FOO and BAR as being the longest fixed and floating strings respectively. | |
271 | ||
272 | The strings can be composites, for instance | |
273 | ||
274 | /(f)(o)(o)/ | |
275 | ||
276 | will result in a composite fixed substring 'foo'. | |
277 | ||
278 | For each string some basic information is maintained: | |
279 | ||
280 | - offset or min_offset | |
281 | This is the position the string must appear at, or not before. | |
282 | It also implicitly (when combined with minlenp) tells us how many | |
3b753521 FN |
283 | characters must match before the string we are searching for. |
284 | Likewise when combined with minlenp and the length of the string it | |
1de06328 YO |
285 | tells us how many characters must appear after the string we have |
286 | found. | |
287 | ||
288 | - max_offset | |
289 | Only used for floating strings. This is the rightmost point that | |
3b753521 | 290 | the string can appear at. If set to I32 max it indicates that the |
1de06328 YO |
291 | string can occur infinitely far to the right. |
292 | ||
293 | - minlenp | |
294 | A pointer to the minimum length of the pattern that the string | |
295 | was found inside. This is important as in the case of positive | |
296 | lookahead or positive lookbehind we can have multiple patterns | |
297 | involved. Consider | |
298 | ||
299 | /(?=FOO).*F/ | |
300 | ||
301 | The minimum length of the pattern overall is 3, the minimum length | |
302 | of the lookahead part is 3, but the minimum length of the part that | |
303 | will actually match is 1. So 'FOO's minimum length is 3, but the | |
304 | minimum length for the F is 1. This is important as the minimum length | |
305 | is used to determine offsets in front of and behind the string being | |
306 | looked for. Since strings can be composites this is the length of the | |
486ec47a | 307 | pattern at the time it was committed with a scan_commit. Note that |
1de06328 YO |
308 | the length is calculated by study_chunk, so that the minimum lengths |
309 | are not known until the full pattern has been compiled, thus the | |
310 | pointer to the value. | |
311 | ||
312 | - lookbehind | |
313 | ||
314 | In the case of lookbehind the string being searched for can be | |
315 | offset past the start point of the final matching string. | |
316 | If this value was just blithely removed from the min_offset it would | |
317 | invalidate some of the calculations for how many chars must match | |
318 | before or after (as they are derived from min_offset and minlen and | |
319 | the length of the string being searched for). | |
320 | When the final pattern is compiled and the data is moved from the | |
321 | scan_data_t structure into the regexp structure the information | |
322 | about lookbehind is factored in, with the information that would | |
323 | have been lost precalculated in the end_shift field for the | |
324 | associated string. | |
325 | ||
326 | The fields pos_min and pos_delta are used to store the minimum offset | |
327 | and the delta to the maximum offset at the current point in the pattern. | |
328 | ||
329 | */ | |
2c2d71f5 JH |
330 | |
331 | typedef struct scan_data_t { | |
1de06328 YO |
332 | /*I32 len_min; unused */ |
333 | /*I32 len_delta; unused */ | |
2c2d71f5 JH |
334 | I32 pos_min; |
335 | I32 pos_delta; | |
336 | SV *last_found; | |
1de06328 | 337 | I32 last_end; /* min value, <0 unless valid. */ |
2c2d71f5 JH |
338 | I32 last_start_min; |
339 | I32 last_start_max; | |
1de06328 YO |
340 | SV **longest; /* Either &l_fixed, or &l_float. */ |
341 | SV *longest_fixed; /* longest fixed string found in pattern */ | |
342 | I32 offset_fixed; /* offset where it starts */ | |
486ec47a | 343 | I32 *minlen_fixed; /* pointer to the minlen relevant to the string */ |
1de06328 YO |
344 | I32 lookbehind_fixed; /* is the position of the string modfied by LB */ |
345 | SV *longest_float; /* longest floating string found in pattern */ | |
346 | I32 offset_float_min; /* earliest point in string it can appear */ | |
347 | I32 offset_float_max; /* latest point in string it can appear */ | |
486ec47a | 348 | I32 *minlen_float; /* pointer to the minlen relevant to the string */ |
1de06328 | 349 | I32 lookbehind_float; /* is the position of the string modified by LB */ |
2c2d71f5 JH |
350 | I32 flags; |
351 | I32 whilem_c; | |
cb434fcc | 352 | I32 *last_closep; |
653099ff | 353 | struct regnode_charclass_class *start_class; |
2c2d71f5 JH |
354 | } scan_data_t; |
355 | ||
a687059c | 356 | /* |
e50aee73 | 357 | * Forward declarations for pregcomp()'s friends. |
a687059c | 358 | */ |
a0d0e21e | 359 | |
27da23d5 | 360 | static const scan_data_t zero_scan_data = |
1de06328 | 361 | { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0}; |
c277df42 IZ |
362 | |
363 | #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL) | |
07be1b83 YO |
364 | #define SF_BEFORE_SEOL 0x0001 |
365 | #define SF_BEFORE_MEOL 0x0002 | |
c277df42 IZ |
366 | #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL) |
367 | #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL) | |
368 | ||
09b7f37c CB |
369 | #ifdef NO_UNARY_PLUS |
370 | # define SF_FIX_SHIFT_EOL (0+2) | |
371 | # define SF_FL_SHIFT_EOL (0+4) | |
372 | #else | |
373 | # define SF_FIX_SHIFT_EOL (+2) | |
374 | # define SF_FL_SHIFT_EOL (+4) | |
375 | #endif | |
c277df42 IZ |
376 | |
377 | #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL) | |
378 | #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL) | |
379 | ||
380 | #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL) | |
381 | #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */ | |
07be1b83 YO |
382 | #define SF_IS_INF 0x0040 |
383 | #define SF_HAS_PAR 0x0080 | |
384 | #define SF_IN_PAR 0x0100 | |
385 | #define SF_HAS_EVAL 0x0200 | |
386 | #define SCF_DO_SUBSTR 0x0400 | |
653099ff GS |
387 | #define SCF_DO_STCLASS_AND 0x0800 |
388 | #define SCF_DO_STCLASS_OR 0x1000 | |
389 | #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR) | |
e1901655 | 390 | #define SCF_WHILEM_VISITED_POS 0x2000 |
c277df42 | 391 | |
786e8c11 | 392 | #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */ |
e2e6a0f1 | 393 | #define SCF_SEEN_ACCEPT 0x8000 |
07be1b83 | 394 | |
43fead97 | 395 | #define UTF cBOOL(RExC_utf8) |
00b27cfc KW |
396 | |
397 | /* The enums for all these are ordered so things work out correctly */ | |
a62b1201 | 398 | #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET) |
cfaf538b | 399 | #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_DEPENDS_CHARSET) |
00b27cfc | 400 | #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET) |
cfaf538b KW |
401 | #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) >= REGEX_UNICODE_CHARSET) |
402 | #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_RESTRICTED_CHARSET) | |
2f7f8cb1 | 403 | #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) >= REGEX_ASCII_RESTRICTED_CHARSET) |
a725e29c | 404 | #define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_MORE_RESTRICTED_CHARSET) |
a62b1201 | 405 | |
43fead97 | 406 | #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD) |
a0ed51b3 | 407 | |
93733859 | 408 | #define OOB_NAMEDCLASS -1 |
b8c5462f | 409 | |
8e661ac5 KW |
410 | /* There is no code point that is out-of-bounds, so this is problematic. But |
411 | * its only current use is to initialize a variable that is always set before | |
412 | * looked at. */ | |
413 | #define OOB_UNICODE 0xDEADBEEF | |
414 | ||
a0ed51b3 LW |
415 | #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv)) |
416 | #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b) | |
417 | ||
8615cb43 | 418 | |
b45f050a JF |
419 | /* length of regex to show in messages that don't mark a position within */ |
420 | #define RegexLengthToShowInErrorMessages 127 | |
421 | ||
422 | /* | |
423 | * If MARKER[12] are adjusted, be sure to adjust the constants at the top | |
424 | * of t/op/regmesg.t, the tests in t/op/re_tests, and those in | |
425 | * op/pragma/warn/regcomp. | |
426 | */ | |
7253e4e3 RK |
427 | #define MARKER1 "<-- HERE" /* marker as it appears in the description */ |
428 | #define MARKER2 " <-- HERE " /* marker as it appears within the regex */ | |
b81d288d | 429 | |
7253e4e3 | 430 | #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/" |
b45f050a JF |
431 | |
432 | /* | |
433 | * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given | |
434 | * arg. Show regex, up to a maximum length. If it's too long, chop and add | |
435 | * "...". | |
436 | */ | |
58e23c8d | 437 | #define _FAIL(code) STMT_START { \ |
bfed75c6 | 438 | const char *ellipses = ""; \ |
ccb2c380 MP |
439 | IV len = RExC_end - RExC_precomp; \ |
440 | \ | |
441 | if (!SIZE_ONLY) \ | |
288b8c02 | 442 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
ccb2c380 MP |
443 | if (len > RegexLengthToShowInErrorMessages) { \ |
444 | /* chop 10 shorter than the max, to ensure meaning of "..." */ \ | |
445 | len = RegexLengthToShowInErrorMessages - 10; \ | |
446 | ellipses = "..."; \ | |
447 | } \ | |
58e23c8d | 448 | code; \ |
ccb2c380 | 449 | } STMT_END |
8615cb43 | 450 | |
58e23c8d YO |
451 | #define FAIL(msg) _FAIL( \ |
452 | Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \ | |
453 | msg, (int)len, RExC_precomp, ellipses)) | |
454 | ||
455 | #define FAIL2(msg,arg) _FAIL( \ | |
456 | Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \ | |
457 | arg, (int)len, RExC_precomp, ellipses)) | |
458 | ||
b45f050a | 459 | /* |
b45f050a JF |
460 | * Simple_vFAIL -- like FAIL, but marks the current location in the scan |
461 | */ | |
ccb2c380 | 462 | #define Simple_vFAIL(m) STMT_START { \ |
a28509cc | 463 | const IV offset = RExC_parse - RExC_precomp; \ |
ccb2c380 MP |
464 | Perl_croak(aTHX_ "%s" REPORT_LOCATION, \ |
465 | m, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
466 | } STMT_END | |
b45f050a JF |
467 | |
468 | /* | |
469 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL() | |
470 | */ | |
ccb2c380 MP |
471 | #define vFAIL(m) STMT_START { \ |
472 | if (!SIZE_ONLY) \ | |
288b8c02 | 473 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
ccb2c380 MP |
474 | Simple_vFAIL(m); \ |
475 | } STMT_END | |
b45f050a JF |
476 | |
477 | /* | |
478 | * Like Simple_vFAIL(), but accepts two arguments. | |
479 | */ | |
ccb2c380 | 480 | #define Simple_vFAIL2(m,a1) STMT_START { \ |
a28509cc | 481 | const IV offset = RExC_parse - RExC_precomp; \ |
ccb2c380 MP |
482 | S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \ |
483 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
484 | } STMT_END | |
b45f050a JF |
485 | |
486 | /* | |
487 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2(). | |
488 | */ | |
ccb2c380 MP |
489 | #define vFAIL2(m,a1) STMT_START { \ |
490 | if (!SIZE_ONLY) \ | |
288b8c02 | 491 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
ccb2c380 MP |
492 | Simple_vFAIL2(m, a1); \ |
493 | } STMT_END | |
b45f050a JF |
494 | |
495 | ||
496 | /* | |
497 | * Like Simple_vFAIL(), but accepts three arguments. | |
498 | */ | |
ccb2c380 | 499 | #define Simple_vFAIL3(m, a1, a2) STMT_START { \ |
a28509cc | 500 | const IV offset = RExC_parse - RExC_precomp; \ |
ccb2c380 MP |
501 | S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \ |
502 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
503 | } STMT_END | |
b45f050a JF |
504 | |
505 | /* | |
506 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3(). | |
507 | */ | |
ccb2c380 MP |
508 | #define vFAIL3(m,a1,a2) STMT_START { \ |
509 | if (!SIZE_ONLY) \ | |
288b8c02 | 510 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
ccb2c380 MP |
511 | Simple_vFAIL3(m, a1, a2); \ |
512 | } STMT_END | |
b45f050a JF |
513 | |
514 | /* | |
515 | * Like Simple_vFAIL(), but accepts four arguments. | |
516 | */ | |
ccb2c380 | 517 | #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \ |
a28509cc | 518 | const IV offset = RExC_parse - RExC_precomp; \ |
ccb2c380 MP |
519 | S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \ |
520 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
521 | } STMT_END | |
b45f050a | 522 | |
668c081a | 523 | #define ckWARNreg(loc,m) STMT_START { \ |
a28509cc | 524 | const IV offset = loc - RExC_precomp; \ |
f10f4c18 NC |
525 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
526 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
ccb2c380 MP |
527 | } STMT_END |
528 | ||
668c081a | 529 | #define ckWARNregdep(loc,m) STMT_START { \ |
a28509cc | 530 | const IV offset = loc - RExC_precomp; \ |
d1d15184 | 531 | Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \ |
f10f4c18 NC |
532 | m REPORT_LOCATION, \ |
533 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
ccb2c380 MP |
534 | } STMT_END |
535 | ||
2335b3d3 KW |
536 | #define ckWARN2regdep(loc,m, a1) STMT_START { \ |
537 | const IV offset = loc - RExC_precomp; \ | |
538 | Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \ | |
539 | m REPORT_LOCATION, \ | |
540 | a1, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
541 | } STMT_END | |
542 | ||
668c081a | 543 | #define ckWARN2reg(loc, m, a1) STMT_START { \ |
a28509cc | 544 | const IV offset = loc - RExC_precomp; \ |
668c081a | 545 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
ccb2c380 MP |
546 | a1, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
547 | } STMT_END | |
548 | ||
549 | #define vWARN3(loc, m, a1, a2) STMT_START { \ | |
a28509cc | 550 | const IV offset = loc - RExC_precomp; \ |
ccb2c380 MP |
551 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
552 | a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
553 | } STMT_END | |
554 | ||
668c081a NC |
555 | #define ckWARN3reg(loc, m, a1, a2) STMT_START { \ |
556 | const IV offset = loc - RExC_precomp; \ | |
557 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ | |
558 | a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
559 | } STMT_END | |
560 | ||
ccb2c380 | 561 | #define vWARN4(loc, m, a1, a2, a3) STMT_START { \ |
a28509cc | 562 | const IV offset = loc - RExC_precomp; \ |
ccb2c380 MP |
563 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
564 | a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
565 | } STMT_END | |
566 | ||
668c081a NC |
567 | #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \ |
568 | const IV offset = loc - RExC_precomp; \ | |
569 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ | |
570 | a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
571 | } STMT_END | |
572 | ||
ccb2c380 | 573 | #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \ |
a28509cc | 574 | const IV offset = loc - RExC_precomp; \ |
ccb2c380 MP |
575 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
576 | a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
577 | } STMT_END | |
9d1d55b5 | 578 | |
8615cb43 | 579 | |
cd439c50 | 580 | /* Allow for side effects in s */ |
ccb2c380 MP |
581 | #define REGC(c,s) STMT_START { \ |
582 | if (!SIZE_ONLY) *(s) = (c); else (void)(s); \ | |
583 | } STMT_END | |
cd439c50 | 584 | |
fac92740 MJD |
585 | /* Macros for recording node offsets. 20001227 mjd@plover.com |
586 | * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in | |
587 | * element 2*n-1 of the array. Element #2n holds the byte length node #n. | |
588 | * Element 0 holds the number n. | |
07be1b83 | 589 | * Position is 1 indexed. |
fac92740 | 590 | */ |
7122b237 YO |
591 | #ifndef RE_TRACK_PATTERN_OFFSETS |
592 | #define Set_Node_Offset_To_R(node,byte) | |
593 | #define Set_Node_Offset(node,byte) | |
594 | #define Set_Cur_Node_Offset | |
595 | #define Set_Node_Length_To_R(node,len) | |
596 | #define Set_Node_Length(node,len) | |
597 | #define Set_Node_Cur_Length(node) | |
598 | #define Node_Offset(n) | |
599 | #define Node_Length(n) | |
600 | #define Set_Node_Offset_Length(node,offset,len) | |
601 | #define ProgLen(ri) ri->u.proglen | |
602 | #define SetProgLen(ri,x) ri->u.proglen = x | |
603 | #else | |
604 | #define ProgLen(ri) ri->u.offsets[0] | |
605 | #define SetProgLen(ri,x) ri->u.offsets[0] = x | |
ccb2c380 MP |
606 | #define Set_Node_Offset_To_R(node,byte) STMT_START { \ |
607 | if (! SIZE_ONLY) { \ | |
608 | MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \ | |
2a49f0f5 | 609 | __LINE__, (int)(node), (int)(byte))); \ |
ccb2c380 | 610 | if((node) < 0) { \ |
551405c4 | 611 | Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \ |
ccb2c380 MP |
612 | } else { \ |
613 | RExC_offsets[2*(node)-1] = (byte); \ | |
614 | } \ | |
615 | } \ | |
616 | } STMT_END | |
617 | ||
618 | #define Set_Node_Offset(node,byte) \ | |
619 | Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start) | |
620 | #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse) | |
621 | ||
622 | #define Set_Node_Length_To_R(node,len) STMT_START { \ | |
623 | if (! SIZE_ONLY) { \ | |
624 | MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \ | |
551405c4 | 625 | __LINE__, (int)(node), (int)(len))); \ |
ccb2c380 | 626 | if((node) < 0) { \ |
551405c4 | 627 | Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \ |
ccb2c380 MP |
628 | } else { \ |
629 | RExC_offsets[2*(node)] = (len); \ | |
630 | } \ | |
631 | } \ | |
632 | } STMT_END | |
633 | ||
634 | #define Set_Node_Length(node,len) \ | |
635 | Set_Node_Length_To_R((node)-RExC_emit_start, len) | |
636 | #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len) | |
637 | #define Set_Node_Cur_Length(node) \ | |
638 | Set_Node_Length(node, RExC_parse - parse_start) | |
fac92740 MJD |
639 | |
640 | /* Get offsets and lengths */ | |
641 | #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1]) | |
642 | #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)]) | |
643 | ||
07be1b83 YO |
644 | #define Set_Node_Offset_Length(node,offset,len) STMT_START { \ |
645 | Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \ | |
646 | Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \ | |
647 | } STMT_END | |
7122b237 | 648 | #endif |
07be1b83 YO |
649 | |
650 | #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS | |
651 | #define EXPERIMENTAL_INPLACESCAN | |
f427392e | 652 | #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/ |
07be1b83 | 653 | |
304ee84b YO |
654 | #define DEBUG_STUDYDATA(str,data,depth) \ |
655 | DEBUG_OPTIMISE_MORE_r(if(data){ \ | |
1de06328 | 656 | PerlIO_printf(Perl_debug_log, \ |
304ee84b YO |
657 | "%*s" str "Pos:%"IVdf"/%"IVdf \ |
658 | " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \ | |
1de06328 YO |
659 | (int)(depth)*2, "", \ |
660 | (IV)((data)->pos_min), \ | |
661 | (IV)((data)->pos_delta), \ | |
304ee84b | 662 | (UV)((data)->flags), \ |
1de06328 | 663 | (IV)((data)->whilem_c), \ |
304ee84b YO |
664 | (IV)((data)->last_closep ? *((data)->last_closep) : -1), \ |
665 | is_inf ? "INF " : "" \ | |
1de06328 YO |
666 | ); \ |
667 | if ((data)->last_found) \ | |
668 | PerlIO_printf(Perl_debug_log, \ | |
669 | "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \ | |
670 | " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \ | |
671 | SvPVX_const((data)->last_found), \ | |
672 | (IV)((data)->last_end), \ | |
673 | (IV)((data)->last_start_min), \ | |
674 | (IV)((data)->last_start_max), \ | |
675 | ((data)->longest && \ | |
676 | (data)->longest==&((data)->longest_fixed)) ? "*" : "", \ | |
677 | SvPVX_const((data)->longest_fixed), \ | |
678 | (IV)((data)->offset_fixed), \ | |
679 | ((data)->longest && \ | |
680 | (data)->longest==&((data)->longest_float)) ? "*" : "", \ | |
681 | SvPVX_const((data)->longest_float), \ | |
682 | (IV)((data)->offset_float_min), \ | |
683 | (IV)((data)->offset_float_max) \ | |
684 | ); \ | |
685 | PerlIO_printf(Perl_debug_log,"\n"); \ | |
686 | }); | |
687 | ||
acfe0abc | 688 | static void clear_re(pTHX_ void *r); |
4327152a | 689 | |
653099ff | 690 | /* Mark that we cannot extend a found fixed substring at this point. |
786e8c11 | 691 | Update the longest found anchored substring and the longest found |
653099ff GS |
692 | floating substrings if needed. */ |
693 | ||
4327152a | 694 | STATIC void |
304ee84b | 695 | S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf) |
c277df42 | 696 | { |
e1ec3a88 AL |
697 | const STRLEN l = CHR_SVLEN(data->last_found); |
698 | const STRLEN old_l = CHR_SVLEN(*data->longest); | |
1de06328 | 699 | GET_RE_DEBUG_FLAGS_DECL; |
b81d288d | 700 | |
7918f24d NC |
701 | PERL_ARGS_ASSERT_SCAN_COMMIT; |
702 | ||
c277df42 | 703 | if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) { |
6b43b216 | 704 | SvSetMagicSV(*data->longest, data->last_found); |
c277df42 IZ |
705 | if (*data->longest == data->longest_fixed) { |
706 | data->offset_fixed = l ? data->last_start_min : data->pos_min; | |
707 | if (data->flags & SF_BEFORE_EOL) | |
b81d288d | 708 | data->flags |
c277df42 IZ |
709 | |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL); |
710 | else | |
711 | data->flags &= ~SF_FIX_BEFORE_EOL; | |
686b73d4 | 712 | data->minlen_fixed=minlenp; |
1de06328 | 713 | data->lookbehind_fixed=0; |
a0ed51b3 | 714 | } |
304ee84b | 715 | else { /* *data->longest == data->longest_float */ |
c277df42 | 716 | data->offset_float_min = l ? data->last_start_min : data->pos_min; |
b81d288d AB |
717 | data->offset_float_max = (l |
718 | ? data->last_start_max | |
c277df42 | 719 | : data->pos_min + data->pos_delta); |
304ee84b | 720 | if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX) |
9051bda5 | 721 | data->offset_float_max = I32_MAX; |
c277df42 | 722 | if (data->flags & SF_BEFORE_EOL) |
b81d288d | 723 | data->flags |
c277df42 IZ |
724 | |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL); |
725 | else | |
726 | data->flags &= ~SF_FL_BEFORE_EOL; | |
1de06328 YO |
727 | data->minlen_float=minlenp; |
728 | data->lookbehind_float=0; | |
c277df42 IZ |
729 | } |
730 | } | |
731 | SvCUR_set(data->last_found, 0); | |
0eda9292 | 732 | { |
a28509cc | 733 | SV * const sv = data->last_found; |
097eb12c AL |
734 | if (SvUTF8(sv) && SvMAGICAL(sv)) { |
735 | MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8); | |
736 | if (mg) | |
737 | mg->mg_len = 0; | |
738 | } | |
0eda9292 | 739 | } |
c277df42 IZ |
740 | data->last_end = -1; |
741 | data->flags &= ~SF_BEFORE_EOL; | |
bcdf7404 | 742 | DEBUG_STUDYDATA("commit: ",data,0); |
c277df42 IZ |
743 | } |
744 | ||
653099ff GS |
745 | /* Can match anything (initialization) */ |
746 | STATIC void | |
3fffb88a | 747 | S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl) |
653099ff | 748 | { |
7918f24d NC |
749 | PERL_ARGS_ASSERT_CL_ANYTHING; |
750 | ||
f8bef550 | 751 | ANYOF_BITMAP_SETALL(cl); |
dd58aee1 | 752 | cl->flags = ANYOF_CLASS|ANYOF_EOS|ANYOF_UNICODE_ALL |
3ad98780 | 753 | |ANYOF_LOC_NONBITMAP_FOLD|ANYOF_NON_UTF8_LATIN1_ALL; |
3fffb88a KW |
754 | |
755 | /* If any portion of the regex is to operate under locale rules, | |
756 | * initialization includes it. The reason this isn't done for all regexes | |
757 | * is that the optimizer was written under the assumption that locale was | |
758 | * all-or-nothing. Given the complexity and lack of documentation in the | |
759 | * optimizer, and that there are inadequate test cases for locale, so many | |
760 | * parts of it may not work properly, it is safest to avoid locale unless | |
761 | * necessary. */ | |
762 | if (RExC_contains_locale) { | |
9d7a1e63 | 763 | ANYOF_CLASS_SETALL(cl); /* /l uses class */ |
3fffb88a KW |
764 | cl->flags |= ANYOF_LOCALE; |
765 | } | |
9d7a1e63 KW |
766 | else { |
767 | ANYOF_CLASS_ZERO(cl); /* Only /l uses class now */ | |
768 | } | |
653099ff GS |
769 | } |
770 | ||
771 | /* Can match anything (initialization) */ | |
772 | STATIC int | |
5f66b61c | 773 | S_cl_is_anything(const struct regnode_charclass_class *cl) |
653099ff GS |
774 | { |
775 | int value; | |
776 | ||
7918f24d NC |
777 | PERL_ARGS_ASSERT_CL_IS_ANYTHING; |
778 | ||
aaa51d5e | 779 | for (value = 0; value <= ANYOF_MAX; value += 2) |
653099ff GS |
780 | if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1)) |
781 | return 1; | |
1aa99e6b IH |
782 | if (!(cl->flags & ANYOF_UNICODE_ALL)) |
783 | return 0; | |
10edeb5d | 784 | if (!ANYOF_BITMAP_TESTALLSET((const void*)cl)) |
f8bef550 | 785 | return 0; |
653099ff GS |
786 | return 1; |
787 | } | |
788 | ||
789 | /* Can match anything (initialization) */ | |
790 | STATIC void | |
e755fd73 | 791 | S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl) |
653099ff | 792 | { |
7918f24d NC |
793 | PERL_ARGS_ASSERT_CL_INIT; |
794 | ||
8ecf7187 | 795 | Zero(cl, 1, struct regnode_charclass_class); |
653099ff | 796 | cl->type = ANYOF; |
3fffb88a | 797 | cl_anything(pRExC_state, cl); |
1411dba4 | 798 | ARG_SET(cl, ANYOF_NONBITMAP_EMPTY); |
653099ff GS |
799 | } |
800 | ||
1051e1c4 KW |
801 | /* These two functions currently do the exact same thing */ |
802 | #define cl_init_zero S_cl_init | |
653099ff | 803 | |
dd58aee1 KW |
804 | /* 'AND' a given class with another one. Can create false positives. 'cl' |
805 | * should not be inverted. 'and_with->flags & ANYOF_CLASS' should be 0 if | |
806 | * 'and_with' is a regnode_charclass instead of a regnode_charclass_class. */ | |
653099ff | 807 | STATIC void |
5f66b61c | 808 | S_cl_and(struct regnode_charclass_class *cl, |
a28509cc | 809 | const struct regnode_charclass_class *and_with) |
653099ff | 810 | { |
7918f24d | 811 | PERL_ARGS_ASSERT_CL_AND; |
40d049e4 YO |
812 | |
813 | assert(and_with->type == ANYOF); | |
1e6ade67 | 814 | |
c6b76537 | 815 | /* I (khw) am not sure all these restrictions are necessary XXX */ |
1e6ade67 KW |
816 | if (!(ANYOF_CLASS_TEST_ANY_SET(and_with)) |
817 | && !(ANYOF_CLASS_TEST_ANY_SET(cl)) | |
653099ff | 818 | && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) |
39065660 KW |
819 | && !(and_with->flags & ANYOF_LOC_NONBITMAP_FOLD) |
820 | && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) { | |
653099ff GS |
821 | int i; |
822 | ||
823 | if (and_with->flags & ANYOF_INVERT) | |
824 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) | |
825 | cl->bitmap[i] &= ~and_with->bitmap[i]; | |
826 | else | |
827 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) | |
828 | cl->bitmap[i] &= and_with->bitmap[i]; | |
829 | } /* XXXX: logic is complicated otherwise, leave it along for a moment. */ | |
1aa99e6b | 830 | |
c6b76537 | 831 | if (and_with->flags & ANYOF_INVERT) { |
8951c461 | 832 | |
c6b76537 KW |
833 | /* Here, the and'ed node is inverted. Get the AND of the flags that |
834 | * aren't affected by the inversion. Those that are affected are | |
835 | * handled individually below */ | |
836 | U8 affected_flags = cl->flags & ~INVERSION_UNAFFECTED_FLAGS; | |
837 | cl->flags &= (and_with->flags & INVERSION_UNAFFECTED_FLAGS); | |
838 | cl->flags |= affected_flags; | |
839 | ||
840 | /* We currently don't know how to deal with things that aren't in the | |
841 | * bitmap, but we know that the intersection is no greater than what | |
842 | * is already in cl, so let there be false positives that get sorted | |
843 | * out after the synthetic start class succeeds, and the node is | |
844 | * matched for real. */ | |
845 | ||
846 | /* The inversion of these two flags indicate that the resulting | |
847 | * intersection doesn't have them */ | |
848 | if (and_with->flags & ANYOF_UNICODE_ALL) { | |
4713bfe1 KW |
849 | cl->flags &= ~ANYOF_UNICODE_ALL; |
850 | } | |
c6b76537 KW |
851 | if (and_with->flags & ANYOF_NON_UTF8_LATIN1_ALL) { |
852 | cl->flags &= ~ANYOF_NON_UTF8_LATIN1_ALL; | |
137165a6 | 853 | } |
1aa99e6b | 854 | } |
c6b76537 | 855 | else { /* and'd node is not inverted */ |
3ad98780 KW |
856 | U8 outside_bitmap_but_not_utf8; /* Temp variable */ |
857 | ||
137165a6 | 858 | if (! ANYOF_NONBITMAP(and_with)) { |
c6b76537 KW |
859 | |
860 | /* Here 'and_with' doesn't match anything outside the bitmap | |
861 | * (except possibly ANYOF_UNICODE_ALL), which means the | |
862 | * intersection can't either, except for ANYOF_UNICODE_ALL, in | |
863 | * which case we don't know what the intersection is, but it's no | |
864 | * greater than what cl already has, so can just leave it alone, | |
865 | * with possible false positives */ | |
866 | if (! (and_with->flags & ANYOF_UNICODE_ALL)) { | |
867 | ARG_SET(cl, ANYOF_NONBITMAP_EMPTY); | |
871d0d1a | 868 | cl->flags &= ~ANYOF_NONBITMAP_NON_UTF8; |
c6b76537 | 869 | } |
137165a6 | 870 | } |
c6b76537 KW |
871 | else if (! ANYOF_NONBITMAP(cl)) { |
872 | ||
873 | /* Here, 'and_with' does match something outside the bitmap, and cl | |
874 | * doesn't have a list of things to match outside the bitmap. If | |
875 | * cl can match all code points above 255, the intersection will | |
3ad98780 KW |
876 | * be those above-255 code points that 'and_with' matches. If cl |
877 | * can't match all Unicode code points, it means that it can't | |
878 | * match anything outside the bitmap (since the 'if' that got us | |
879 | * into this block tested for that), so we leave the bitmap empty. | |
880 | */ | |
c6b76537 KW |
881 | if (cl->flags & ANYOF_UNICODE_ALL) { |
882 | ARG_SET(cl, ARG(and_with)); | |
3ad98780 KW |
883 | |
884 | /* and_with's ARG may match things that don't require UTF8. | |
885 | * And now cl's will too, in spite of this being an 'and'. See | |
886 | * the comments below about the kludge */ | |
887 | cl->flags |= and_with->flags & ANYOF_NONBITMAP_NON_UTF8; | |
c6b76537 KW |
888 | } |
889 | } | |
890 | else { | |
891 | /* Here, both 'and_with' and cl match something outside the | |
892 | * bitmap. Currently we do not do the intersection, so just match | |
893 | * whatever cl had at the beginning. */ | |
894 | } | |
895 | ||
896 | ||
3ad98780 KW |
897 | /* Take the intersection of the two sets of flags. However, the |
898 | * ANYOF_NONBITMAP_NON_UTF8 flag is treated as an 'or'. This is a | |
899 | * kludge around the fact that this flag is not treated like the others | |
900 | * which are initialized in cl_anything(). The way the optimizer works | |
901 | * is that the synthetic start class (SSC) is initialized to match | |
902 | * anything, and then the first time a real node is encountered, its | |
903 | * values are AND'd with the SSC's with the result being the values of | |
904 | * the real node. However, there are paths through the optimizer where | |
905 | * the AND never gets called, so those initialized bits are set | |
906 | * inappropriately, which is not usually a big deal, as they just cause | |
907 | * false positives in the SSC, which will just mean a probably | |
908 | * imperceptible slow down in execution. However this bit has a | |
909 | * higher false positive consequence in that it can cause utf8.pm, | |
910 | * utf8_heavy.pl ... to be loaded when not necessary, which is a much | |
911 | * bigger slowdown and also causes significant extra memory to be used. | |
912 | * In order to prevent this, the code now takes a different tack. The | |
913 | * bit isn't set unless some part of the regular expression needs it, | |
914 | * but once set it won't get cleared. This means that these extra | |
915 | * modules won't get loaded unless there was some path through the | |
916 | * pattern that would have required them anyway, and so any false | |
917 | * positives that occur by not ANDing them out when they could be | |
918 | * aren't as severe as they would be if we treated this bit like all | |
919 | * the others */ | |
920 | outside_bitmap_but_not_utf8 = (cl->flags | and_with->flags) | |
921 | & ANYOF_NONBITMAP_NON_UTF8; | |
c6b76537 | 922 | cl->flags &= and_with->flags; |
3ad98780 | 923 | cl->flags |= outside_bitmap_but_not_utf8; |
137165a6 | 924 | } |
653099ff GS |
925 | } |
926 | ||
dd58aee1 KW |
927 | /* 'OR' a given class with another one. Can create false positives. 'cl' |
928 | * should not be inverted. 'or_with->flags & ANYOF_CLASS' should be 0 if | |
929 | * 'or_with' is a regnode_charclass instead of a regnode_charclass_class. */ | |
653099ff | 930 | STATIC void |
3fffb88a | 931 | S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with) |
653099ff | 932 | { |
7918f24d NC |
933 | PERL_ARGS_ASSERT_CL_OR; |
934 | ||
653099ff | 935 | if (or_with->flags & ANYOF_INVERT) { |
c6b76537 KW |
936 | |
937 | /* Here, the or'd node is to be inverted. This means we take the | |
938 | * complement of everything not in the bitmap, but currently we don't | |
939 | * know what that is, so give up and match anything */ | |
940 | if (ANYOF_NONBITMAP(or_with)) { | |
3fffb88a | 941 | cl_anything(pRExC_state, cl); |
c6b76537 | 942 | } |
653099ff GS |
943 | /* We do not use |
944 | * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2)) | |
945 | * <= (B1 | !B2) | (CL1 | !CL2) | |
946 | * which is wasteful if CL2 is small, but we ignore CL2: | |
947 | * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1 | |
948 | * XXXX Can we handle case-fold? Unclear: | |
949 | * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) = | |
950 | * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i')) | |
951 | */ | |
c6b76537 | 952 | else if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) |
39065660 KW |
953 | && !(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD) |
954 | && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD) ) { | |
653099ff GS |
955 | int i; |
956 | ||
957 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) | |
958 | cl->bitmap[i] |= ~or_with->bitmap[i]; | |
959 | } /* XXXX: logic is complicated otherwise */ | |
960 | else { | |
3fffb88a | 961 | cl_anything(pRExC_state, cl); |
653099ff | 962 | } |
c6b76537 KW |
963 | |
964 | /* And, we can just take the union of the flags that aren't affected | |
965 | * by the inversion */ | |
966 | cl->flags |= or_with->flags & INVERSION_UNAFFECTED_FLAGS; | |
967 | ||
968 | /* For the remaining flags: | |
969 | ANYOF_UNICODE_ALL and inverted means to not match anything above | |
970 | 255, which means that the union with cl should just be | |
971 | what cl has in it, so can ignore this flag | |
972 | ANYOF_NON_UTF8_LATIN1_ALL and inverted means if not utf8 and ord | |
973 | is 127-255 to match them, but then invert that, so the | |
974 | union with cl should just be what cl has in it, so can | |
975 | ignore this flag | |
976 | */ | |
977 | } else { /* 'or_with' is not inverted */ | |
653099ff GS |
978 | /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */ |
979 | if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) | |
39065660 KW |
980 | && (!(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD) |
981 | || (cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) ) { | |
653099ff GS |
982 | int i; |
983 | ||
984 | /* OR char bitmap and class bitmap separately */ | |
985 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) | |
986 | cl->bitmap[i] |= or_with->bitmap[i]; | |
1e6ade67 | 987 | if (ANYOF_CLASS_TEST_ANY_SET(or_with)) { |
653099ff GS |
988 | for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++) |
989 | cl->classflags[i] |= or_with->classflags[i]; | |
990 | cl->flags |= ANYOF_CLASS; | |
991 | } | |
992 | } | |
993 | else { /* XXXX: logic is complicated, leave it along for a moment. */ | |
3fffb88a | 994 | cl_anything(pRExC_state, cl); |
653099ff | 995 | } |
9826f543 | 996 | |
c6b76537 KW |
997 | if (ANYOF_NONBITMAP(or_with)) { |
998 | ||
999 | /* Use the added node's outside-the-bit-map match if there isn't a | |
1000 | * conflict. If there is a conflict (both nodes match something | |
1001 | * outside the bitmap, but what they match outside is not the same | |
1002 | * pointer, and hence not easily compared until XXX we extend | |
1003 | * inversion lists this far), give up and allow the start class to | |
d94b1d13 KW |
1004 | * match everything outside the bitmap. If that stuff is all above |
1005 | * 255, can just set UNICODE_ALL, otherwise caould be anything. */ | |
c6b76537 KW |
1006 | if (! ANYOF_NONBITMAP(cl)) { |
1007 | ARG_SET(cl, ARG(or_with)); | |
1008 | } | |
1009 | else if (ARG(cl) != ARG(or_with)) { | |
d94b1d13 KW |
1010 | |
1011 | if ((or_with->flags & ANYOF_NONBITMAP_NON_UTF8)) { | |
1012 | cl_anything(pRExC_state, cl); | |
1013 | } | |
1014 | else { | |
1015 | cl->flags |= ANYOF_UNICODE_ALL; | |
1016 | } | |
c6b76537 | 1017 | } |
4c34a693 | 1018 | } |
0b9668ee KW |
1019 | |
1020 | /* Take the union */ | |
1021 | cl->flags |= or_with->flags; | |
1aa99e6b | 1022 | } |
653099ff GS |
1023 | } |
1024 | ||
a3621e74 YO |
1025 | #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ] |
1026 | #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid ) | |
1027 | #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate ) | |
1028 | #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 ) | |
1029 | ||
3dab1dad YO |
1030 | |
1031 | #ifdef DEBUGGING | |
07be1b83 | 1032 | /* |
2b8b4781 NC |
1033 | dump_trie(trie,widecharmap,revcharmap) |
1034 | dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc) | |
1035 | dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc) | |
3dab1dad YO |
1036 | |
1037 | These routines dump out a trie in a somewhat readable format. | |
07be1b83 YO |
1038 | The _interim_ variants are used for debugging the interim |
1039 | tables that are used to generate the final compressed | |
1040 | representation which is what dump_trie expects. | |
1041 | ||
486ec47a | 1042 | Part of the reason for their existence is to provide a form |
3dab1dad | 1043 | of documentation as to how the different representations function. |
07be1b83 YO |
1044 | |
1045 | */ | |
3dab1dad YO |
1046 | |
1047 | /* | |
3dab1dad YO |
1048 | Dumps the final compressed table form of the trie to Perl_debug_log. |
1049 | Used for debugging make_trie(). | |
1050 | */ | |
b9a59e08 | 1051 | |
3dab1dad | 1052 | STATIC void |
2b8b4781 NC |
1053 | S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap, |
1054 | AV *revcharmap, U32 depth) | |
3dab1dad YO |
1055 | { |
1056 | U32 state; | |
ab3bbdeb | 1057 | SV *sv=sv_newmortal(); |
55eed653 | 1058 | int colwidth= widecharmap ? 6 : 4; |
2e64971a | 1059 | U16 word; |
3dab1dad YO |
1060 | GET_RE_DEBUG_FLAGS_DECL; |
1061 | ||
7918f24d | 1062 | PERL_ARGS_ASSERT_DUMP_TRIE; |
ab3bbdeb | 1063 | |
3dab1dad YO |
1064 | PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ", |
1065 | (int)depth * 2 + 2,"", | |
1066 | "Match","Base","Ofs" ); | |
1067 | ||
1068 | for( state = 0 ; state < trie->uniquecharcount ; state++ ) { | |
2b8b4781 | 1069 | SV ** const tmp = av_fetch( revcharmap, state, 0); |
3dab1dad | 1070 | if ( tmp ) { |
ab3bbdeb YO |
1071 | PerlIO_printf( Perl_debug_log, "%*s", |
1072 | colwidth, | |
ddc5bc0f | 1073 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
ab3bbdeb YO |
1074 | PL_colors[0], PL_colors[1], |
1075 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | | |
1076 | PERL_PV_ESCAPE_FIRSTCHAR | |
1077 | ) | |
1078 | ); | |
3dab1dad YO |
1079 | } |
1080 | } | |
1081 | PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------", | |
1082 | (int)depth * 2 + 2,""); | |
1083 | ||
1084 | for( state = 0 ; state < trie->uniquecharcount ; state++ ) | |
ab3bbdeb | 1085 | PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------"); |
3dab1dad YO |
1086 | PerlIO_printf( Perl_debug_log, "\n"); |
1087 | ||
1e2e3d02 | 1088 | for( state = 1 ; state < trie->statecount ; state++ ) { |
be8e71aa | 1089 | const U32 base = trie->states[ state ].trans.base; |
3dab1dad YO |
1090 | |
1091 | PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state); | |
1092 | ||
1093 | if ( trie->states[ state ].wordnum ) { | |
1094 | PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum ); | |
1095 | } else { | |
1096 | PerlIO_printf( Perl_debug_log, "%6s", "" ); | |
1097 | } | |
1098 | ||
1099 | PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base ); | |
1100 | ||
1101 | if ( base ) { | |
1102 | U32 ofs = 0; | |
1103 | ||
1104 | while( ( base + ofs < trie->uniquecharcount ) || | |
1105 | ( base + ofs - trie->uniquecharcount < trie->lasttrans | |
1106 | && trie->trans[ base + ofs - trie->uniquecharcount ].check != state)) | |
1107 | ofs++; | |
1108 | ||
1109 | PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs); | |
1110 | ||
1111 | for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) { | |
1112 | if ( ( base + ofs >= trie->uniquecharcount ) && | |
1113 | ( base + ofs - trie->uniquecharcount < trie->lasttrans ) && | |
1114 | trie->trans[ base + ofs - trie->uniquecharcount ].check == state ) | |
1115 | { | |
ab3bbdeb YO |
1116 | PerlIO_printf( Perl_debug_log, "%*"UVXf, |
1117 | colwidth, | |
3dab1dad YO |
1118 | (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next ); |
1119 | } else { | |
ab3bbdeb | 1120 | PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." ); |
3dab1dad YO |
1121 | } |
1122 | } | |
1123 | ||
1124 | PerlIO_printf( Perl_debug_log, "]"); | |
1125 | ||
1126 | } | |
1127 | PerlIO_printf( Perl_debug_log, "\n" ); | |
1128 | } | |
2e64971a DM |
1129 | PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, ""); |
1130 | for (word=1; word <= trie->wordcount; word++) { | |
1131 | PerlIO_printf(Perl_debug_log, " %d:(%d,%d)", | |
1132 | (int)word, (int)(trie->wordinfo[word].prev), | |
1133 | (int)(trie->wordinfo[word].len)); | |
1134 | } | |
1135 | PerlIO_printf(Perl_debug_log, "\n" ); | |
3dab1dad YO |
1136 | } |
1137 | /* | |
3dab1dad YO |
1138 | Dumps a fully constructed but uncompressed trie in list form. |
1139 | List tries normally only are used for construction when the number of | |
1140 | possible chars (trie->uniquecharcount) is very high. | |
1141 | Used for debugging make_trie(). | |
1142 | */ | |
1143 | STATIC void | |
55eed653 | 1144 | S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie, |
2b8b4781 NC |
1145 | HV *widecharmap, AV *revcharmap, U32 next_alloc, |
1146 | U32 depth) | |
3dab1dad YO |
1147 | { |
1148 | U32 state; | |
ab3bbdeb | 1149 | SV *sv=sv_newmortal(); |
55eed653 | 1150 | int colwidth= widecharmap ? 6 : 4; |
3dab1dad | 1151 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
1152 | |
1153 | PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST; | |
1154 | ||
3dab1dad | 1155 | /* print out the table precompression. */ |
ab3bbdeb YO |
1156 | PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s", |
1157 | (int)depth * 2 + 2,"", (int)depth * 2 + 2,"", | |
1158 | "------:-----+-----------------\n" ); | |
3dab1dad YO |
1159 | |
1160 | for( state=1 ; state < next_alloc ; state ++ ) { | |
1161 | U16 charid; | |
1162 | ||
ab3bbdeb | 1163 | PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :", |
3dab1dad YO |
1164 | (int)depth * 2 + 2,"", (UV)state ); |
1165 | if ( ! trie->states[ state ].wordnum ) { | |
1166 | PerlIO_printf( Perl_debug_log, "%5s| ",""); | |
1167 | } else { | |
1168 | PerlIO_printf( Perl_debug_log, "W%4x| ", | |
1169 | trie->states[ state ].wordnum | |
1170 | ); | |
1171 | } | |
1172 | for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) { | |
2b8b4781 | 1173 | SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0); |
ab3bbdeb YO |
1174 | if ( tmp ) { |
1175 | PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ", | |
1176 | colwidth, | |
ddc5bc0f | 1177 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
ab3bbdeb YO |
1178 | PL_colors[0], PL_colors[1], |
1179 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | | |
1180 | PERL_PV_ESCAPE_FIRSTCHAR | |
1181 | ) , | |
1e2e3d02 YO |
1182 | TRIE_LIST_ITEM(state,charid).forid, |
1183 | (UV)TRIE_LIST_ITEM(state,charid).newstate | |
1184 | ); | |
1185 | if (!(charid % 10)) | |
664e119d RGS |
1186 | PerlIO_printf(Perl_debug_log, "\n%*s| ", |
1187 | (int)((depth * 2) + 14), ""); | |
1e2e3d02 | 1188 | } |
ab3bbdeb YO |
1189 | } |
1190 | PerlIO_printf( Perl_debug_log, "\n"); | |
3dab1dad YO |
1191 | } |
1192 | } | |
1193 | ||
1194 | /* | |
3dab1dad YO |
1195 | Dumps a fully constructed but uncompressed trie in table form. |
1196 | This is the normal DFA style state transition table, with a few | |
1197 | twists to facilitate compression later. | |
1198 | Used for debugging make_trie(). | |
1199 | */ | |
1200 | STATIC void | |
55eed653 | 1201 | S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie, |
2b8b4781 NC |
1202 | HV *widecharmap, AV *revcharmap, U32 next_alloc, |
1203 | U32 depth) | |
3dab1dad YO |
1204 | { |
1205 | U32 state; | |
1206 | U16 charid; | |
ab3bbdeb | 1207 | SV *sv=sv_newmortal(); |
55eed653 | 1208 | int colwidth= widecharmap ? 6 : 4; |
3dab1dad | 1209 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
1210 | |
1211 | PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE; | |
3dab1dad YO |
1212 | |
1213 | /* | |
1214 | print out the table precompression so that we can do a visual check | |
1215 | that they are identical. | |
1216 | */ | |
1217 | ||
1218 | PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" ); | |
1219 | ||
1220 | for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { | |
2b8b4781 | 1221 | SV ** const tmp = av_fetch( revcharmap, charid, 0); |
3dab1dad | 1222 | if ( tmp ) { |
ab3bbdeb YO |
1223 | PerlIO_printf( Perl_debug_log, "%*s", |
1224 | colwidth, | |
ddc5bc0f | 1225 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
ab3bbdeb YO |
1226 | PL_colors[0], PL_colors[1], |
1227 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | | |
1228 | PERL_PV_ESCAPE_FIRSTCHAR | |
1229 | ) | |
1230 | ); | |
3dab1dad YO |
1231 | } |
1232 | } | |
1233 | ||
1234 | PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" ); | |
1235 | ||
1236 | for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) { | |
ab3bbdeb | 1237 | PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------"); |
3dab1dad YO |
1238 | } |
1239 | ||
1240 | PerlIO_printf( Perl_debug_log, "\n" ); | |
1241 | ||
1242 | for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) { | |
1243 | ||
1244 | PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ", | |
1245 | (int)depth * 2 + 2,"", | |
1246 | (UV)TRIE_NODENUM( state ) ); | |
1247 | ||
1248 | for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { | |
ab3bbdeb YO |
1249 | UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next ); |
1250 | if (v) | |
1251 | PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v ); | |
1252 | else | |
1253 | PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." ); | |
3dab1dad YO |
1254 | } |
1255 | if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) { | |
1256 | PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check ); | |
1257 | } else { | |
1258 | PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check, | |
1259 | trie->states[ TRIE_NODENUM( state ) ].wordnum ); | |
1260 | } | |
1261 | } | |
07be1b83 | 1262 | } |
3dab1dad YO |
1263 | |
1264 | #endif | |
1265 | ||
2e64971a | 1266 | |
786e8c11 YO |
1267 | /* make_trie(startbranch,first,last,tail,word_count,flags,depth) |
1268 | startbranch: the first branch in the whole branch sequence | |
1269 | first : start branch of sequence of branch-exact nodes. | |
1270 | May be the same as startbranch | |
1271 | last : Thing following the last branch. | |
1272 | May be the same as tail. | |
1273 | tail : item following the branch sequence | |
1274 | count : words in the sequence | |
1275 | flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/ | |
1276 | depth : indent depth | |
3dab1dad | 1277 | |
786e8c11 | 1278 | Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node. |
07be1b83 | 1279 | |
786e8c11 YO |
1280 | A trie is an N'ary tree where the branches are determined by digital |
1281 | decomposition of the key. IE, at the root node you look up the 1st character and | |
1282 | follow that branch repeat until you find the end of the branches. Nodes can be | |
1283 | marked as "accepting" meaning they represent a complete word. Eg: | |
07be1b83 | 1284 | |
786e8c11 | 1285 | /he|she|his|hers/ |
72f13be8 | 1286 | |
786e8c11 YO |
1287 | would convert into the following structure. Numbers represent states, letters |
1288 | following numbers represent valid transitions on the letter from that state, if | |
1289 | the number is in square brackets it represents an accepting state, otherwise it | |
1290 | will be in parenthesis. | |
07be1b83 | 1291 | |
786e8c11 YO |
1292 | +-h->+-e->[3]-+-r->(8)-+-s->[9] |
1293 | | | | |
1294 | | (2) | |
1295 | | | | |
1296 | (1) +-i->(6)-+-s->[7] | |
1297 | | | |
1298 | +-s->(3)-+-h->(4)-+-e->[5] | |
07be1b83 | 1299 | |
786e8c11 YO |
1300 | Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers) |
1301 | ||
1302 | This shows that when matching against the string 'hers' we will begin at state 1 | |
1303 | read 'h' and move to state 2, read 'e' and move to state 3 which is accepting, | |
1304 | then read 'r' and go to state 8 followed by 's' which takes us to state 9 which | |
1305 | is also accepting. Thus we know that we can match both 'he' and 'hers' with a | |
1306 | single traverse. We store a mapping from accepting to state to which word was | |
1307 | matched, and then when we have multiple possibilities we try to complete the | |
1308 | rest of the regex in the order in which they occured in the alternation. | |
1309 | ||
1310 | The only prior NFA like behaviour that would be changed by the TRIE support is | |
1311 | the silent ignoring of duplicate alternations which are of the form: | |
1312 | ||
1313 | / (DUPE|DUPE) X? (?{ ... }) Y /x | |
1314 | ||
4b714af6 | 1315 | Thus EVAL blocks following a trie may be called a different number of times with |
786e8c11 | 1316 | and without the optimisation. With the optimisations dupes will be silently |
486ec47a | 1317 | ignored. This inconsistent behaviour of EVAL type nodes is well established as |
786e8c11 YO |
1318 | the following demonstrates: |
1319 | ||
1320 | 'words'=~/(word|word|word)(?{ print $1 })[xyz]/ | |
1321 | ||
1322 | which prints out 'word' three times, but | |
1323 | ||
1324 | 'words'=~/(word|word|word)(?{ print $1 })S/ | |
1325 | ||
1326 | which doesnt print it out at all. This is due to other optimisations kicking in. | |
1327 | ||
1328 | Example of what happens on a structural level: | |
1329 | ||
486ec47a | 1330 | The regexp /(ac|ad|ab)+/ will produce the following debug output: |
786e8c11 YO |
1331 | |
1332 | 1: CURLYM[1] {1,32767}(18) | |
1333 | 5: BRANCH(8) | |
1334 | 6: EXACT <ac>(16) | |
1335 | 8: BRANCH(11) | |
1336 | 9: EXACT <ad>(16) | |
1337 | 11: BRANCH(14) | |
1338 | 12: EXACT <ab>(16) | |
1339 | 16: SUCCEED(0) | |
1340 | 17: NOTHING(18) | |
1341 | 18: END(0) | |
1342 | ||
1343 | This would be optimizable with startbranch=5, first=5, last=16, tail=16 | |
1344 | and should turn into: | |
1345 | ||
1346 | 1: CURLYM[1] {1,32767}(18) | |
1347 | 5: TRIE(16) | |
1348 | [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1] | |
1349 | <ac> | |
1350 | <ad> | |
1351 | <ab> | |
1352 | 16: SUCCEED(0) | |
1353 | 17: NOTHING(18) | |
1354 | 18: END(0) | |
1355 | ||
1356 | Cases where tail != last would be like /(?foo|bar)baz/: | |
1357 | ||
1358 | 1: BRANCH(4) | |
1359 | 2: EXACT <foo>(8) | |
1360 | 4: BRANCH(7) | |
1361 | 5: EXACT <bar>(8) | |
1362 | 7: TAIL(8) | |
1363 | 8: EXACT <baz>(10) | |
1364 | 10: END(0) | |
1365 | ||
1366 | which would be optimizable with startbranch=1, first=1, last=7, tail=8 | |
1367 | and would end up looking like: | |
1368 | ||
1369 | 1: TRIE(8) | |
1370 | [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1] | |
1371 | <foo> | |
1372 | <bar> | |
1373 | 7: TAIL(8) | |
1374 | 8: EXACT <baz>(10) | |
1375 | 10: END(0) | |
1376 | ||
1377 | d = uvuni_to_utf8_flags(d, uv, 0); | |
1378 | ||
1379 | is the recommended Unicode-aware way of saying | |
1380 | ||
1381 | *(d++) = uv; | |
1382 | */ | |
1383 | ||
fab2782b | 1384 | #define TRIE_STORE_REVCHAR(val) \ |
786e8c11 | 1385 | STMT_START { \ |
73031816 | 1386 | if (UTF) { \ |
fab2782b | 1387 | SV *zlopp = newSV(7); /* XXX: optimize me */ \ |
88c9ea1e | 1388 | unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \ |
fab2782b | 1389 | unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, val); \ |
73031816 NC |
1390 | SvCUR_set(zlopp, kapow - flrbbbbb); \ |
1391 | SvPOK_on(zlopp); \ | |
1392 | SvUTF8_on(zlopp); \ | |
1393 | av_push(revcharmap, zlopp); \ | |
1394 | } else { \ | |
fab2782b | 1395 | char ooooff = (char)val; \ |
73031816 NC |
1396 | av_push(revcharmap, newSVpvn(&ooooff, 1)); \ |
1397 | } \ | |
1398 | } STMT_END | |
786e8c11 | 1399 | |
fab2782b YO |
1400 | #define TRIE_READ_CHAR STMT_START { \ |
1401 | wordlen++; \ | |
1402 | if ( UTF ) { \ | |
1403 | /* if it is UTF then it is either already folded, or does not need folding */ \ | |
1404 | uvc = utf8n_to_uvuni( (const U8*) uc, UTF8_MAXLEN, &len, uniflags); \ | |
1405 | } \ | |
1406 | else if (folder == PL_fold_latin1) { \ | |
1407 | /* if we use this folder we have to obey unicode rules on latin-1 data */ \ | |
1408 | if ( foldlen > 0 ) { \ | |
1409 | uvc = utf8n_to_uvuni( (const U8*) scan, UTF8_MAXLEN, &len, uniflags ); \ | |
1410 | foldlen -= len; \ | |
1411 | scan += len; \ | |
1412 | len = 0; \ | |
1413 | } else { \ | |
1414 | len = 1; \ | |
1415 | uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, 1); \ | |
1416 | skiplen = UNISKIP(uvc); \ | |
1417 | foldlen -= skiplen; \ | |
1418 | scan = foldbuf + skiplen; \ | |
1419 | } \ | |
1420 | } else { \ | |
1421 | /* raw data, will be folded later if needed */ \ | |
1422 | uvc = (U32)*uc; \ | |
1423 | len = 1; \ | |
1424 | } \ | |
786e8c11 YO |
1425 | } STMT_END |
1426 | ||
1427 | ||
1428 | ||
1429 | #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \ | |
1430 | if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \ | |
f9003953 NC |
1431 | U32 ging = TRIE_LIST_LEN( state ) *= 2; \ |
1432 | Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \ | |
786e8c11 YO |
1433 | } \ |
1434 | TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \ | |
1435 | TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \ | |
1436 | TRIE_LIST_CUR( state )++; \ | |
1437 | } STMT_END | |
07be1b83 | 1438 | |
786e8c11 YO |
1439 | #define TRIE_LIST_NEW(state) STMT_START { \ |
1440 | Newxz( trie->states[ state ].trans.list, \ | |
1441 | 4, reg_trie_trans_le ); \ | |
1442 | TRIE_LIST_CUR( state ) = 1; \ | |
1443 | TRIE_LIST_LEN( state ) = 4; \ | |
1444 | } STMT_END | |
07be1b83 | 1445 | |
786e8c11 YO |
1446 | #define TRIE_HANDLE_WORD(state) STMT_START { \ |
1447 | U16 dupe= trie->states[ state ].wordnum; \ | |
1448 | regnode * const noper_next = regnext( noper ); \ | |
1449 | \ | |
786e8c11 YO |
1450 | DEBUG_r({ \ |
1451 | /* store the word for dumping */ \ | |
1452 | SV* tmp; \ | |
1453 | if (OP(noper) != NOTHING) \ | |
740cce10 | 1454 | tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \ |
786e8c11 | 1455 | else \ |
740cce10 | 1456 | tmp = newSVpvn_utf8( "", 0, UTF ); \ |
2b8b4781 | 1457 | av_push( trie_words, tmp ); \ |
786e8c11 YO |
1458 | }); \ |
1459 | \ | |
1460 | curword++; \ | |
2e64971a DM |
1461 | trie->wordinfo[curword].prev = 0; \ |
1462 | trie->wordinfo[curword].len = wordlen; \ | |
1463 | trie->wordinfo[curword].accept = state; \ | |
786e8c11 YO |
1464 | \ |
1465 | if ( noper_next < tail ) { \ | |
1466 | if (!trie->jump) \ | |
c944940b | 1467 | trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \ |
7f69552c | 1468 | trie->jump[curword] = (U16)(noper_next - convert); \ |
786e8c11 YO |
1469 | if (!jumper) \ |
1470 | jumper = noper_next; \ | |
1471 | if (!nextbranch) \ | |
1472 | nextbranch= regnext(cur); \ | |
1473 | } \ | |
1474 | \ | |
1475 | if ( dupe ) { \ | |
2e64971a DM |
1476 | /* It's a dupe. Pre-insert into the wordinfo[].prev */\ |
1477 | /* chain, so that when the bits of chain are later */\ | |
1478 | /* linked together, the dups appear in the chain */\ | |
1479 | trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \ | |
1480 | trie->wordinfo[dupe].prev = curword; \ | |
786e8c11 YO |
1481 | } else { \ |
1482 | /* we haven't inserted this word yet. */ \ | |
1483 | trie->states[ state ].wordnum = curword; \ | |
1484 | } \ | |
1485 | } STMT_END | |
07be1b83 | 1486 | |
3dab1dad | 1487 | |
786e8c11 YO |
1488 | #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \ |
1489 | ( ( base + charid >= ucharcount \ | |
1490 | && base + charid < ubound \ | |
1491 | && state == trie->trans[ base - ucharcount + charid ].check \ | |
1492 | && trie->trans[ base - ucharcount + charid ].next ) \ | |
1493 | ? trie->trans[ base - ucharcount + charid ].next \ | |
1494 | : ( state==1 ? special : 0 ) \ | |
1495 | ) | |
3dab1dad | 1496 | |
786e8c11 YO |
1497 | #define MADE_TRIE 1 |
1498 | #define MADE_JUMP_TRIE 2 | |
1499 | #define MADE_EXACT_TRIE 4 | |
3dab1dad | 1500 | |
a3621e74 | 1501 | STATIC I32 |
786e8c11 | 1502 | S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth) |
a3621e74 | 1503 | { |
27da23d5 | 1504 | dVAR; |
a3621e74 YO |
1505 | /* first pass, loop through and scan words */ |
1506 | reg_trie_data *trie; | |
55eed653 | 1507 | HV *widecharmap = NULL; |
2b8b4781 | 1508 | AV *revcharmap = newAV(); |
a3621e74 | 1509 | regnode *cur; |
9f7f3913 | 1510 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
a3621e74 YO |
1511 | STRLEN len = 0; |
1512 | UV uvc = 0; | |
1513 | U16 curword = 0; | |
1514 | U32 next_alloc = 0; | |
786e8c11 YO |
1515 | regnode *jumper = NULL; |
1516 | regnode *nextbranch = NULL; | |
7f69552c | 1517 | regnode *convert = NULL; |
2e64971a | 1518 | U32 *prev_states; /* temp array mapping each state to previous one */ |
a3621e74 | 1519 | /* we just use folder as a flag in utf8 */ |
1e696034 | 1520 | const U8 * folder = NULL; |
a3621e74 | 1521 | |
2b8b4781 NC |
1522 | #ifdef DEBUGGING |
1523 | const U32 data_slot = add_data( pRExC_state, 4, "tuuu" ); | |
1524 | AV *trie_words = NULL; | |
1525 | /* along with revcharmap, this only used during construction but both are | |
1526 | * useful during debugging so we store them in the struct when debugging. | |
8e11feef | 1527 | */ |
2b8b4781 NC |
1528 | #else |
1529 | const U32 data_slot = add_data( pRExC_state, 2, "tu" ); | |
3dab1dad | 1530 | STRLEN trie_charcount=0; |
3dab1dad | 1531 | #endif |
2b8b4781 | 1532 | SV *re_trie_maxbuff; |
a3621e74 | 1533 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
1534 | |
1535 | PERL_ARGS_ASSERT_MAKE_TRIE; | |
72f13be8 YO |
1536 | #ifndef DEBUGGING |
1537 | PERL_UNUSED_ARG(depth); | |
1538 | #endif | |
a3621e74 | 1539 | |
1e696034 | 1540 | switch (flags) { |
c46d03cf | 1541 | case EXACT: break; |
2f7f8cb1 | 1542 | case EXACTFA: |
fab2782b YO |
1543 | case EXACTFU_SS: |
1544 | case EXACTFU_TRICKYFOLD: | |
1e696034 KW |
1545 | case EXACTFU: folder = PL_fold_latin1; break; |
1546 | case EXACTF: folder = PL_fold; break; | |
1547 | case EXACTFL: folder = PL_fold_locale; break; | |
fab2782b | 1548 | default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] ); |
1e696034 KW |
1549 | } |
1550 | ||
c944940b | 1551 | trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) ); |
a3621e74 | 1552 | trie->refcount = 1; |
3dab1dad | 1553 | trie->startstate = 1; |
786e8c11 | 1554 | trie->wordcount = word_count; |
f8fc2ecf | 1555 | RExC_rxi->data->data[ data_slot ] = (void*)trie; |
c944940b | 1556 | trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) ); |
fab2782b | 1557 | if (flags == EXACT) |
c944940b | 1558 | trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 ); |
2e64971a DM |
1559 | trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc( |
1560 | trie->wordcount+1, sizeof(reg_trie_wordinfo)); | |
1561 | ||
a3621e74 | 1562 | DEBUG_r({ |
2b8b4781 | 1563 | trie_words = newAV(); |
a3621e74 | 1564 | }); |
a3621e74 | 1565 | |
0111c4fd | 1566 | re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1); |
a3621e74 | 1567 | if (!SvIOK(re_trie_maxbuff)) { |
0111c4fd | 1568 | sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); |
a3621e74 | 1569 | } |
df826430 | 1570 | DEBUG_TRIE_COMPILE_r({ |
3dab1dad | 1571 | PerlIO_printf( Perl_debug_log, |
786e8c11 | 1572 | "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n", |
3dab1dad YO |
1573 | (int)depth * 2 + 2, "", |
1574 | REG_NODE_NUM(startbranch),REG_NODE_NUM(first), | |
786e8c11 | 1575 | REG_NODE_NUM(last), REG_NODE_NUM(tail), |
85c3142d | 1576 | (int)depth); |
3dab1dad | 1577 | }); |
7f69552c YO |
1578 | |
1579 | /* Find the node we are going to overwrite */ | |
1580 | if ( first == startbranch && OP( last ) != BRANCH ) { | |
1581 | /* whole branch chain */ | |
1582 | convert = first; | |
1583 | } else { | |
1584 | /* branch sub-chain */ | |
1585 | convert = NEXTOPER( first ); | |
1586 | } | |
1587 | ||
a3621e74 YO |
1588 | /* -- First loop and Setup -- |
1589 | ||
1590 | We first traverse the branches and scan each word to determine if it | |
1591 | contains widechars, and how many unique chars there are, this is | |
1592 | important as we have to build a table with at least as many columns as we | |
1593 | have unique chars. | |
1594 | ||
1595 | We use an array of integers to represent the character codes 0..255 | |
38a44b82 | 1596 | (trie->charmap) and we use a an HV* to store Unicode characters. We use the |
a3621e74 YO |
1597 | native representation of the character value as the key and IV's for the |
1598 | coded index. | |
1599 | ||
1600 | *TODO* If we keep track of how many times each character is used we can | |
1601 | remap the columns so that the table compression later on is more | |
3b753521 | 1602 | efficient in terms of memory by ensuring the most common value is in the |
a3621e74 YO |
1603 | middle and the least common are on the outside. IMO this would be better |
1604 | than a most to least common mapping as theres a decent chance the most | |
1605 | common letter will share a node with the least common, meaning the node | |
486ec47a | 1606 | will not be compressible. With a middle is most common approach the worst |
a3621e74 YO |
1607 | case is when we have the least common nodes twice. |
1608 | ||
1609 | */ | |
1610 | ||
a3621e74 | 1611 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
df826430 | 1612 | regnode *noper = NEXTOPER( cur ); |
e1ec3a88 | 1613 | const U8 *uc = (U8*)STRING( noper ); |
df826430 | 1614 | const U8 *e = uc + STR_LEN( noper ); |
a3621e74 YO |
1615 | STRLEN foldlen = 0; |
1616 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; | |
fab2782b | 1617 | STRLEN skiplen = 0; |
2af232bd | 1618 | const U8 *scan = (U8*)NULL; |
07be1b83 | 1619 | U32 wordlen = 0; /* required init */ |
02daf0ab YO |
1620 | STRLEN chars = 0; |
1621 | bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/ | |
a3621e74 | 1622 | |
3dab1dad | 1623 | if (OP(noper) == NOTHING) { |
df826430 YO |
1624 | regnode *noper_next= regnext(noper); |
1625 | if (noper_next != tail && OP(noper_next) == flags) { | |
1626 | noper = noper_next; | |
1627 | uc= (U8*)STRING(noper); | |
1628 | e= uc + STR_LEN(noper); | |
1629 | trie->minlen= STR_LEN(noper); | |
1630 | } else { | |
1631 | trie->minlen= 0; | |
1632 | continue; | |
1633 | } | |
3dab1dad | 1634 | } |
df826430 | 1635 | |
fab2782b | 1636 | if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */ |
02daf0ab YO |
1637 | TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte |
1638 | regardless of encoding */ | |
fab2782b YO |
1639 | if (OP( noper ) == EXACTFU_SS) { |
1640 | /* false positives are ok, so just set this */ | |
1641 | TRIE_BITMAP_SET(trie,0xDF); | |
1642 | } | |
1643 | } | |
a3621e74 | 1644 | for ( ; uc < e ; uc += len ) { |
3dab1dad | 1645 | TRIE_CHARCOUNT(trie)++; |
a3621e74 | 1646 | TRIE_READ_CHAR; |
3dab1dad | 1647 | chars++; |
a3621e74 | 1648 | if ( uvc < 256 ) { |
fab2782b YO |
1649 | if ( folder ) { |
1650 | U8 folded= folder[ (U8) uvc ]; | |
1651 | if ( !trie->charmap[ folded ] ) { | |
1652 | trie->charmap[ folded ]=( ++trie->uniquecharcount ); | |
1653 | TRIE_STORE_REVCHAR( folded ); | |
1654 | } | |
1655 | } | |
a3621e74 YO |
1656 | if ( !trie->charmap[ uvc ] ) { |
1657 | trie->charmap[ uvc ]=( ++trie->uniquecharcount ); | |
fab2782b | 1658 | TRIE_STORE_REVCHAR( uvc ); |
a3621e74 | 1659 | } |
02daf0ab | 1660 | if ( set_bit ) { |
62012aee KW |
1661 | /* store the codepoint in the bitmap, and its folded |
1662 | * equivalent. */ | |
fab2782b | 1663 | TRIE_BITMAP_SET(trie, uvc); |
0921ee73 T |
1664 | |
1665 | /* store the folded codepoint */ | |
fab2782b | 1666 | if ( folder ) TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); |
0921ee73 T |
1667 | |
1668 | if ( !UTF ) { | |
1669 | /* store first byte of utf8 representation of | |
acdf4139 KW |
1670 | variant codepoints */ |
1671 | if (! UNI_IS_INVARIANT(uvc)) { | |
1672 | TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); | |
0921ee73 T |
1673 | } |
1674 | } | |
02daf0ab YO |
1675 | set_bit = 0; /* We've done our bit :-) */ |
1676 | } | |
a3621e74 YO |
1677 | } else { |
1678 | SV** svpp; | |
55eed653 NC |
1679 | if ( !widecharmap ) |
1680 | widecharmap = newHV(); | |
a3621e74 | 1681 | |
55eed653 | 1682 | svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 ); |
a3621e74 YO |
1683 | |
1684 | if ( !svpp ) | |
e4584336 | 1685 | Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc ); |
a3621e74 YO |
1686 | |
1687 | if ( !SvTRUE( *svpp ) ) { | |
1688 | sv_setiv( *svpp, ++trie->uniquecharcount ); | |
fab2782b | 1689 | TRIE_STORE_REVCHAR(uvc); |
a3621e74 YO |
1690 | } |
1691 | } | |
1692 | } | |
3dab1dad | 1693 | if( cur == first ) { |
fab2782b YO |
1694 | trie->minlen = chars; |
1695 | trie->maxlen = chars; | |
3dab1dad | 1696 | } else if (chars < trie->minlen) { |
fab2782b | 1697 | trie->minlen = chars; |
3dab1dad | 1698 | } else if (chars > trie->maxlen) { |
fab2782b YO |
1699 | trie->maxlen = chars; |
1700 | } | |
1701 | if (OP( noper ) == EXACTFU_SS) { | |
1702 | /* XXX: workaround - 'ss' could match "\x{DF}" so minlen could be 1 and not 2*/ | |
1703 | if (trie->minlen > 1) | |
1704 | trie->minlen= 1; | |
1705 | } | |
1706 | if (OP( noper ) == EXACTFU_TRICKYFOLD) { | |
1707 | /* XXX: workround - things like "\x{1FBE}\x{0308}\x{0301}" can match "\x{0390}" | |
1708 | * - We assume that any such sequence might match a 2 byte string */ | |
1709 | if (trie->minlen > 2 ) | |
1710 | trie->minlen= 2; | |
3dab1dad YO |
1711 | } |
1712 | ||
a3621e74 YO |
1713 | } /* end first pass */ |
1714 | DEBUG_TRIE_COMPILE_r( | |
3dab1dad YO |
1715 | PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n", |
1716 | (int)depth * 2 + 2,"", | |
55eed653 | 1717 | ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count, |
be8e71aa YO |
1718 | (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount, |
1719 | (int)trie->minlen, (int)trie->maxlen ) | |
a3621e74 | 1720 | ); |
a3621e74 YO |
1721 | |
1722 | /* | |
1723 | We now know what we are dealing with in terms of unique chars and | |
1724 | string sizes so we can calculate how much memory a naive | |
0111c4fd RGS |
1725 | representation using a flat table will take. If it's over a reasonable |
1726 | limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory | |
a3621e74 YO |
1727 | conservative but potentially much slower representation using an array |
1728 | of lists. | |
1729 | ||
1730 | At the end we convert both representations into the same compressed | |
1731 | form that will be used in regexec.c for matching with. The latter | |
1732 | is a form that cannot be used to construct with but has memory | |
1733 | properties similar to the list form and access properties similar | |
1734 | to the table form making it both suitable for fast searches and | |
1735 | small enough that its feasable to store for the duration of a program. | |
1736 | ||
1737 | See the comment in the code where the compressed table is produced | |
1738 | inplace from the flat tabe representation for an explanation of how | |
1739 | the compression works. | |
1740 | ||
1741 | */ | |
1742 | ||
1743 | ||
2e64971a DM |
1744 | Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32); |
1745 | prev_states[1] = 0; | |
1746 | ||
3dab1dad | 1747 | if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) { |
a3621e74 YO |
1748 | /* |
1749 | Second Pass -- Array Of Lists Representation | |
1750 | ||
1751 | Each state will be represented by a list of charid:state records | |
1752 | (reg_trie_trans_le) the first such element holds the CUR and LEN | |
1753 | points of the allocated array. (See defines above). | |
1754 | ||
1755 | We build the initial structure using the lists, and then convert | |
1756 | it into the compressed table form which allows faster lookups | |
1757 | (but cant be modified once converted). | |
a3621e74 YO |
1758 | */ |
1759 | ||
a3621e74 YO |
1760 | STRLEN transcount = 1; |
1761 | ||
1e2e3d02 YO |
1762 | DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log, |
1763 | "%*sCompiling trie using list compiler\n", | |
1764 | (int)depth * 2 + 2, "")); | |
686b73d4 | 1765 | |
c944940b JH |
1766 | trie->states = (reg_trie_state *) |
1767 | PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2, | |
1768 | sizeof(reg_trie_state) ); | |
a3621e74 YO |
1769 | TRIE_LIST_NEW(1); |
1770 | next_alloc = 2; | |
1771 | ||
1772 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { | |
1773 | ||
df826430 | 1774 | regnode *noper = NEXTOPER( cur ); |
c445ea15 | 1775 | U8 *uc = (U8*)STRING( noper ); |
df826430 | 1776 | const U8 *e = uc + STR_LEN( noper ); |
c445ea15 AL |
1777 | U32 state = 1; /* required init */ |
1778 | U16 charid = 0; /* sanity init */ | |
1779 | U8 *scan = (U8*)NULL; /* sanity init */ | |
1780 | STRLEN foldlen = 0; /* required init */ | |
07be1b83 | 1781 | U32 wordlen = 0; /* required init */ |
c445ea15 | 1782 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
fab2782b | 1783 | STRLEN skiplen = 0; |
c445ea15 | 1784 | |
df826430 YO |
1785 | if (OP(noper) == NOTHING) { |
1786 | regnode *noper_next= regnext(noper); | |
1787 | if (noper_next != tail && OP(noper_next) == flags) { | |
1788 | noper = noper_next; | |
1789 | uc= (U8*)STRING(noper); | |
1790 | e= uc + STR_LEN(noper); | |
1791 | } | |
1792 | } | |
1793 | ||
3dab1dad | 1794 | if (OP(noper) != NOTHING) { |
786e8c11 | 1795 | for ( ; uc < e ; uc += len ) { |
c445ea15 | 1796 | |
786e8c11 | 1797 | TRIE_READ_CHAR; |
c445ea15 | 1798 | |
786e8c11 YO |
1799 | if ( uvc < 256 ) { |
1800 | charid = trie->charmap[ uvc ]; | |
c445ea15 | 1801 | } else { |
55eed653 | 1802 | SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0); |
786e8c11 YO |
1803 | if ( !svpp ) { |
1804 | charid = 0; | |
1805 | } else { | |
1806 | charid=(U16)SvIV( *svpp ); | |
1807 | } | |
c445ea15 | 1808 | } |
786e8c11 YO |
1809 | /* charid is now 0 if we dont know the char read, or nonzero if we do */ |
1810 | if ( charid ) { | |
a3621e74 | 1811 | |
786e8c11 YO |
1812 | U16 check; |
1813 | U32 newstate = 0; | |
a3621e74 | 1814 | |
786e8c11 YO |
1815 | charid--; |
1816 | if ( !trie->states[ state ].trans.list ) { | |
1817 | TRIE_LIST_NEW( state ); | |
c445ea15 | 1818 | } |
786e8c11 YO |
1819 | for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) { |
1820 | if ( TRIE_LIST_ITEM( state, check ).forid == charid ) { | |
1821 | newstate = TRIE_LIST_ITEM( state, check ).newstate; | |
1822 | break; | |
1823 | } | |
1824 | } | |
1825 | if ( ! newstate ) { | |
1826 | newstate = next_alloc++; | |
2e64971a | 1827 | prev_states[newstate] = state; |
786e8c11 YO |
1828 | TRIE_LIST_PUSH( state, charid, newstate ); |
1829 | transcount++; | |
1830 | } | |
1831 | state = newstate; | |
1832 | } else { | |
1833 | Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc ); | |
c445ea15 | 1834 | } |
a28509cc | 1835 | } |
c445ea15 | 1836 | } |
3dab1dad | 1837 | TRIE_HANDLE_WORD(state); |
a3621e74 YO |
1838 | |
1839 | } /* end second pass */ | |
1840 | ||
1e2e3d02 YO |
1841 | /* next alloc is the NEXT state to be allocated */ |
1842 | trie->statecount = next_alloc; | |
c944940b JH |
1843 | trie->states = (reg_trie_state *) |
1844 | PerlMemShared_realloc( trie->states, | |
1845 | next_alloc | |
1846 | * sizeof(reg_trie_state) ); | |
a3621e74 | 1847 | |
3dab1dad | 1848 | /* and now dump it out before we compress it */ |
2b8b4781 NC |
1849 | DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap, |
1850 | revcharmap, next_alloc, | |
1851 | depth+1) | |
1e2e3d02 | 1852 | ); |
a3621e74 | 1853 | |
c944940b JH |
1854 | trie->trans = (reg_trie_trans *) |
1855 | PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) ); | |
a3621e74 YO |
1856 | { |
1857 | U32 state; | |
a3621e74 YO |
1858 | U32 tp = 0; |
1859 | U32 zp = 0; | |
1860 | ||
1861 | ||
1862 | for( state=1 ; state < next_alloc ; state ++ ) { | |
1863 | U32 base=0; | |
1864 | ||
1865 | /* | |
1866 | DEBUG_TRIE_COMPILE_MORE_r( | |
1867 | PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp) | |
1868 | ); | |
1869 | */ | |
1870 | ||
1871 | if (trie->states[state].trans.list) { | |
1872 | U16 minid=TRIE_LIST_ITEM( state, 1).forid; | |
1873 | U16 maxid=minid; | |
a28509cc | 1874 | U16 idx; |
a3621e74 YO |
1875 | |
1876 | for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) { | |
c445ea15 AL |
1877 | const U16 forid = TRIE_LIST_ITEM( state, idx).forid; |
1878 | if ( forid < minid ) { | |
1879 | minid=forid; | |
1880 | } else if ( forid > maxid ) { | |
1881 | maxid=forid; | |
1882 | } | |
a3621e74 YO |
1883 | } |
1884 | if ( transcount < tp + maxid - minid + 1) { | |
1885 | transcount *= 2; | |
c944940b JH |
1886 | trie->trans = (reg_trie_trans *) |
1887 | PerlMemShared_realloc( trie->trans, | |
446bd890 NC |
1888 | transcount |
1889 | * sizeof(reg_trie_trans) ); | |
a3621e74 YO |
1890 | Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans ); |
1891 | } | |
1892 | base = trie->uniquecharcount + tp - minid; | |
1893 | if ( maxid == minid ) { | |
1894 | U32 set = 0; | |
1895 | for ( ; zp < tp ; zp++ ) { | |
1896 | if ( ! trie->trans[ zp ].next ) { | |
1897 | base = trie->uniquecharcount + zp - minid; | |
1898 | trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate; | |
1899 | trie->trans[ zp ].check = state; | |
1900 | set = 1; | |
1901 | break; | |
1902 | } | |
1903 | } | |
1904 | if ( !set ) { | |
1905 | trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate; | |
1906 | trie->trans[ tp ].check = state; | |
1907 | tp++; | |
1908 | zp = tp; | |
1909 | } | |
1910 | } else { | |
1911 | for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) { | |
c445ea15 | 1912 | const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid; |
a3621e74 YO |
1913 | trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate; |
1914 | trie->trans[ tid ].check = state; | |
1915 | } | |
1916 | tp += ( maxid - minid + 1 ); | |
1917 | } | |
1918 | Safefree(trie->states[ state ].trans.list); | |
1919 | } | |
1920 | /* | |
1921 | DEBUG_TRIE_COMPILE_MORE_r( | |
1922 | PerlIO_printf( Perl_debug_log, " base: %d\n",base); | |
1923 | ); | |
1924 | */ | |
1925 | trie->states[ state ].trans.base=base; | |
1926 | } | |
cc601c31 | 1927 | trie->lasttrans = tp + 1; |
a3621e74 YO |
1928 | } |
1929 | } else { | |
1930 | /* | |
1931 | Second Pass -- Flat Table Representation. | |
1932 | ||
1933 | we dont use the 0 slot of either trans[] or states[] so we add 1 to each. | |
1934 | We know that we will need Charcount+1 trans at most to store the data | |
1935 | (one row per char at worst case) So we preallocate both structures | |
1936 | assuming worst case. | |
1937 | ||
1938 | We then construct the trie using only the .next slots of the entry | |
1939 | structs. | |
1940 | ||
3b753521 | 1941 | We use the .check field of the first entry of the node temporarily to |
a3621e74 YO |
1942 | make compression both faster and easier by keeping track of how many non |
1943 | zero fields are in the node. | |
1944 | ||
1945 | Since trans are numbered from 1 any 0 pointer in the table is a FAIL | |
1946 | transition. | |
1947 | ||
1948 | There are two terms at use here: state as a TRIE_NODEIDX() which is a | |
1949 | number representing the first entry of the node, and state as a | |
1950 | TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and | |
1951 | TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there | |
1952 | are 2 entrys per node. eg: | |
1953 | ||
1954 | A B A B | |
1955 | 1. 2 4 1. 3 7 | |
1956 | 2. 0 3 3. 0 5 | |
1957 | 3. 0 0 5. 0 0 | |
1958 | 4. 0 0 7. 0 0 | |
1959 | ||
1960 | The table is internally in the right hand, idx form. However as we also | |
1961 | have to deal with the states array which is indexed by nodenum we have to | |
1962 | use TRIE_NODENUM() to convert. | |
1963 | ||
1964 | */ | |
1e2e3d02 YO |
1965 | DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log, |
1966 | "%*sCompiling trie using table compiler\n", | |
1967 | (int)depth * 2 + 2, "")); | |
3dab1dad | 1968 | |
c944940b JH |
1969 | trie->trans = (reg_trie_trans *) |
1970 | PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 ) | |
1971 | * trie->uniquecharcount + 1, | |
1972 | sizeof(reg_trie_trans) ); | |
1973 | trie->states = (reg_trie_state *) | |
1974 | PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2, | |
1975 | sizeof(reg_trie_state) ); | |
a3621e74 YO |
1976 | next_alloc = trie->uniquecharcount + 1; |
1977 | ||
3dab1dad | 1978 | |
a3621e74 YO |
1979 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
1980 | ||
df826430 | 1981 | regnode *noper = NEXTOPER( cur ); |
a28509cc | 1982 | const U8 *uc = (U8*)STRING( noper ); |
df826430 | 1983 | const U8 *e = uc + STR_LEN( noper ); |
a3621e74 YO |
1984 | |
1985 | U32 state = 1; /* required init */ | |
1986 | ||
1987 | U16 charid = 0; /* sanity init */ | |
1988 | U32 accept_state = 0; /* sanity init */ | |
1989 | U8 *scan = (U8*)NULL; /* sanity init */ | |
1990 | ||
1991 | STRLEN foldlen = 0; /* required init */ | |
07be1b83 | 1992 | U32 wordlen = 0; /* required init */ |
fab2782b | 1993 | STRLEN skiplen = 0; |
a3621e74 YO |
1994 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
1995 | ||
df826430 YO |
1996 | if (OP(noper) == NOTHING) { |
1997 | regnode *noper_next= regnext(noper); | |
1998 | if (noper_next != tail && OP(noper_next) == flags) { | |
1999 | noper = noper_next; | |
2000 | uc= (U8*)STRING(noper); | |
2001 | e= uc + STR_LEN(noper); | |
2002 | } | |
2003 | } | |
fab2782b | 2004 | |
3dab1dad | 2005 | if ( OP(noper) != NOTHING ) { |
786e8c11 | 2006 | for ( ; uc < e ; uc += len ) { |
a3621e74 | 2007 | |
786e8c11 | 2008 | TRIE_READ_CHAR; |
a3621e74 | 2009 | |
786e8c11 YO |
2010 | if ( uvc < 256 ) { |
2011 | charid = trie->charmap[ uvc ]; | |
2012 | } else { | |
55eed653 | 2013 | SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0); |
786e8c11 | 2014 | charid = svpp ? (U16)SvIV(*svpp) : 0; |
a3621e74 | 2015 | } |
786e8c11 YO |
2016 | if ( charid ) { |
2017 | charid--; | |
2018 | if ( !trie->trans[ state + charid ].next ) { | |
2019 | trie->trans[ state + charid ].next = next_alloc; | |
2020 | trie->trans[ state ].check++; | |
2e64971a DM |
2021 | prev_states[TRIE_NODENUM(next_alloc)] |
2022 | = TRIE_NODENUM(state); | |
786e8c11 YO |
2023 | next_alloc += trie->uniquecharcount; |
2024 | } | |
2025 | state = trie->trans[ state + charid ].next; | |
2026 | } else { | |
2027 | Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc ); | |
2028 | } | |
2029 | /* charid is now 0 if we dont know the char read, or nonzero if we do */ | |
a3621e74 | 2030 | } |
a3621e74 | 2031 | } |
3dab1dad YO |
2032 | accept_state = TRIE_NODENUM( state ); |
2033 | TRIE_HANDLE_WORD(accept_state); | |
a3621e74 YO |
2034 | |
2035 | } /* end second pass */ | |
2036 | ||
3dab1dad | 2037 | /* and now dump it out before we compress it */ |
2b8b4781 NC |
2038 | DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap, |
2039 | revcharmap, | |
2040 | next_alloc, depth+1)); | |
a3621e74 | 2041 | |
a3621e74 YO |
2042 | { |
2043 | /* | |
2044 | * Inplace compress the table.* | |
2045 | ||
2046 | For sparse data sets the table constructed by the trie algorithm will | |
2047 | be mostly 0/FAIL transitions or to put it another way mostly empty. | |
2048 | (Note that leaf nodes will not contain any transitions.) | |
2049 | ||
2050 | This algorithm compresses the tables by eliminating most such | |
2051 | transitions, at the cost of a modest bit of extra work during lookup: | |
2052 | ||
2053 | - Each states[] entry contains a .base field which indicates the | |
2054 | index in the state[] array wheres its transition data is stored. | |
2055 | ||
3b753521 | 2056 | - If .base is 0 there are no valid transitions from that node. |
a3621e74 YO |
2057 | |
2058 | - If .base is nonzero then charid is added to it to find an entry in | |
2059 | the trans array. | |
2060 | ||
2061 | -If trans[states[state].base+charid].check!=state then the | |
2062 | transition is taken to be a 0/Fail transition. Thus if there are fail | |
2063 | transitions at the front of the node then the .base offset will point | |
2064 | somewhere inside the previous nodes data (or maybe even into a node | |
2065 | even earlier), but the .check field determines if the transition is | |
2066 | valid. | |
2067 | ||
786e8c11 | 2068 | XXX - wrong maybe? |
a3621e74 | 2069 | The following process inplace converts the table to the compressed |
3b753521 | 2070 | table: We first do not compress the root node 1,and mark all its |
a3621e74 | 2071 | .check pointers as 1 and set its .base pointer as 1 as well. This |
3b753521 FN |
2072 | allows us to do a DFA construction from the compressed table later, |
2073 | and ensures that any .base pointers we calculate later are greater | |
2074 | than 0. | |
a3621e74 YO |
2075 | |
2076 | - We set 'pos' to indicate the first entry of the second node. | |
2077 | ||
2078 | - We then iterate over the columns of the node, finding the first and | |
2079 | last used entry at l and m. We then copy l..m into pos..(pos+m-l), | |
2080 | and set the .check pointers accordingly, and advance pos | |
2081 | appropriately and repreat for the next node. Note that when we copy | |
2082 | the next pointers we have to convert them from the original | |
2083 | NODEIDX form to NODENUM form as the former is not valid post | |
2084 | compression. | |
2085 | ||
2086 | - If a node has no transitions used we mark its base as 0 and do not | |
2087 | advance the pos pointer. | |
2088 | ||
2089 | - If a node only has one transition we use a second pointer into the | |
2090 | structure to fill in allocated fail transitions from other states. | |
2091 | This pointer is independent of the main pointer and scans forward | |
2092 | looking for null transitions that are allocated to a state. When it | |
2093 | finds one it writes the single transition into the "hole". If the | |
786e8c11 | 2094 | pointer doesnt find one the single transition is appended as normal. |
a3621e74 YO |
2095 | |
2096 | - Once compressed we can Renew/realloc the structures to release the | |
2097 | excess space. | |
2098 | ||
2099 | See "Table-Compression Methods" in sec 3.9 of the Red Dragon, | |
2100 | specifically Fig 3.47 and the associated pseudocode. | |
2101 | ||
2102 | demq | |
2103 | */ | |
a3b680e6 | 2104 | const U32 laststate = TRIE_NODENUM( next_alloc ); |
a28509cc | 2105 | U32 state, charid; |
a3621e74 | 2106 | U32 pos = 0, zp=0; |
1e2e3d02 | 2107 | trie->statecount = laststate; |
a3621e74 YO |
2108 | |
2109 | for ( state = 1 ; state < laststate ; state++ ) { | |
2110 | U8 flag = 0; | |
a28509cc AL |
2111 | const U32 stateidx = TRIE_NODEIDX( state ); |
2112 | const U32 o_used = trie->trans[ stateidx ].check; | |
2113 | U32 used = trie->trans[ stateidx ].check; | |
a3621e74 YO |
2114 | trie->trans[ stateidx ].check = 0; |
2115 | ||
2116 | for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) { | |
2117 | if ( flag || trie->trans[ stateidx + charid ].next ) { | |
2118 | if ( trie->trans[ stateidx + charid ].next ) { | |
2119 | if (o_used == 1) { | |
2120 | for ( ; zp < pos ; zp++ ) { | |
2121 | if ( ! trie->trans[ zp ].next ) { | |
2122 | break; | |
2123 | } | |
2124 | } | |
2125 | trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ; | |
2126 | trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next ); | |
2127 | trie->trans[ zp ].check = state; | |
2128 | if ( ++zp > pos ) pos = zp; | |
2129 | break; | |
2130 | } | |
2131 | used--; | |
2132 | } | |
2133 | if ( !flag ) { | |
2134 | flag = 1; | |
2135 | trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ; | |
2136 | } | |
2137 | trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next ); | |
2138 | trie->trans[ pos ].check = state; | |
2139 | pos++; | |
2140 | } | |
2141 | } | |
2142 | } | |
cc601c31 | 2143 | trie->lasttrans = pos + 1; |
c944940b JH |
2144 | trie->states = (reg_trie_state *) |
2145 | PerlMemShared_realloc( trie->states, laststate | |
2146 | * sizeof(reg_trie_state) ); | |
a3621e74 | 2147 | DEBUG_TRIE_COMPILE_MORE_r( |
e4584336 | 2148 | PerlIO_printf( Perl_debug_log, |
3dab1dad YO |
2149 | "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n", |
2150 | (int)depth * 2 + 2,"", | |
2151 | (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ), | |
5d7488b2 AL |
2152 | (IV)next_alloc, |
2153 | (IV)pos, | |
a3621e74 YO |
2154 | ( ( next_alloc - pos ) * 100 ) / (double)next_alloc ); |
2155 | ); | |
2156 | ||
2157 | } /* end table compress */ | |
2158 | } | |
1e2e3d02 YO |
2159 | DEBUG_TRIE_COMPILE_MORE_r( |
2160 | PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n", | |
2161 | (int)depth * 2 + 2, "", | |
2162 | (UV)trie->statecount, | |
2163 | (UV)trie->lasttrans) | |
2164 | ); | |
cc601c31 | 2165 | /* resize the trans array to remove unused space */ |
c944940b JH |
2166 | trie->trans = (reg_trie_trans *) |
2167 | PerlMemShared_realloc( trie->trans, trie->lasttrans | |
2168 | * sizeof(reg_trie_trans) ); | |
a3621e74 | 2169 | |
3b753521 | 2170 | { /* Modify the program and insert the new TRIE node */ |
3dab1dad YO |
2171 | U8 nodetype =(U8)(flags & 0xFF); |
2172 | char *str=NULL; | |
786e8c11 | 2173 | |
07be1b83 | 2174 | #ifdef DEBUGGING |
e62cc96a | 2175 | regnode *optimize = NULL; |
7122b237 YO |
2176 | #ifdef RE_TRACK_PATTERN_OFFSETS |
2177 | ||
b57a0404 JH |
2178 | U32 mjd_offset = 0; |
2179 | U32 mjd_nodelen = 0; | |
7122b237 YO |
2180 | #endif /* RE_TRACK_PATTERN_OFFSETS */ |
2181 | #endif /* DEBUGGING */ | |
a3621e74 | 2182 | /* |
3dab1dad YO |
2183 | This means we convert either the first branch or the first Exact, |
2184 | depending on whether the thing following (in 'last') is a branch | |
2185 | or not and whther first is the startbranch (ie is it a sub part of | |
2186 | the alternation or is it the whole thing.) | |
3b753521 | 2187 | Assuming its a sub part we convert the EXACT otherwise we convert |
3dab1dad | 2188 | the whole branch sequence, including the first. |
a3621e74 | 2189 | */ |
3dab1dad | 2190 | /* Find the node we are going to overwrite */ |
7f69552c | 2191 | if ( first != startbranch || OP( last ) == BRANCH ) { |
07be1b83 | 2192 | /* branch sub-chain */ |
3dab1dad | 2193 | NEXT_OFF( first ) = (U16)(last - first); |
7122b237 | 2194 | #ifdef RE_TRACK_PATTERN_OFFSETS |
07be1b83 YO |
2195 | DEBUG_r({ |
2196 | mjd_offset= Node_Offset((convert)); | |
2197 | mjd_nodelen= Node_Length((convert)); | |
2198 | }); | |
7122b237 | 2199 | #endif |
7f69552c | 2200 | /* whole branch chain */ |
7122b237 YO |
2201 | } |
2202 | #ifdef RE_TRACK_PATTERN_OFFSETS | |
2203 | else { | |
7f69552c YO |
2204 | DEBUG_r({ |
2205 | const regnode *nop = NEXTOPER( convert ); | |
2206 | mjd_offset= Node_Offset((nop)); | |
2207 | mjd_nodelen= Node_Length((nop)); | |
2208 | }); | |
07be1b83 YO |
2209 | } |
2210 | DEBUG_OPTIMISE_r( | |
2211 | PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n", | |
2212 | (int)depth * 2 + 2, "", | |
786e8c11 | 2213 | (UV)mjd_offset, (UV)mjd_nodelen) |
07be1b83 | 2214 | ); |
7122b237 | 2215 | #endif |
3dab1dad YO |
2216 | /* But first we check to see if there is a common prefix we can |
2217 | split out as an EXACT and put in front of the TRIE node. */ | |
2218 | trie->startstate= 1; | |
55eed653 | 2219 | if ( trie->bitmap && !widecharmap && !trie->jump ) { |
3dab1dad | 2220 | U32 state; |
1e2e3d02 | 2221 | for ( state = 1 ; state < trie->statecount-1 ; state++ ) { |
a3621e74 | 2222 | U32 ofs = 0; |
8e11feef RGS |
2223 | I32 idx = -1; |
2224 | U32 count = 0; | |
2225 | const U32 base = trie->states[ state ].trans.base; | |
a3621e74 | 2226 | |
3dab1dad | 2227 | if ( trie->states[state].wordnum ) |
8e11feef | 2228 | count = 1; |
a3621e74 | 2229 | |
8e11feef | 2230 | for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) { |
cc601c31 YO |
2231 | if ( ( base + ofs >= trie->uniquecharcount ) && |
2232 | ( base + ofs - trie->uniquecharcount < trie->lasttrans ) && | |
a3621e74 YO |
2233 | trie->trans[ base + ofs - trie->uniquecharcount ].check == state ) |
2234 | { | |
3dab1dad | 2235 | if ( ++count > 1 ) { |
2b8b4781 | 2236 | SV **tmp = av_fetch( revcharmap, ofs, 0); |
07be1b83 | 2237 | const U8 *ch = (U8*)SvPV_nolen_const( *tmp ); |
8e11feef | 2238 | if ( state == 1 ) break; |
3dab1dad YO |
2239 | if ( count == 2 ) { |
2240 | Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char); | |
2241 | DEBUG_OPTIMISE_r( | |
8e11feef RGS |
2242 | PerlIO_printf(Perl_debug_log, |
2243 | "%*sNew Start State=%"UVuf" Class: [", | |
2244 | (int)depth * 2 + 2, "", | |
786e8c11 | 2245 | (UV)state)); |
be8e71aa | 2246 | if (idx >= 0) { |
2b8b4781 | 2247 | SV ** const tmp = av_fetch( revcharmap, idx, 0); |
be8e71aa | 2248 | const U8 * const ch = (U8*)SvPV_nolen_const( *tmp ); |
8e11feef | 2249 | |
3dab1dad | 2250 | TRIE_BITMAP_SET(trie,*ch); |
8e11feef RGS |
2251 | if ( folder ) |
2252 | TRIE_BITMAP_SET(trie, folder[ *ch ]); | |
3dab1dad | 2253 | DEBUG_OPTIMISE_r( |
f1f66076 | 2254 | PerlIO_printf(Perl_debug_log, "%s", (char*)ch) |
3dab1dad | 2255 | ); |
8e11feef RGS |
2256 | } |
2257 | } | |
2258 | TRIE_BITMAP_SET(trie,*ch); | |
2259 | if ( folder ) | |
2260 | TRIE_BITMAP_SET(trie,folder[ *ch ]); | |
2261 | DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch)); | |
2262 | } | |
2263 | idx = ofs; | |
2264 | } | |
3dab1dad YO |
2265 | } |
2266 | if ( count == 1 ) { | |
2b8b4781 | 2267 | SV **tmp = av_fetch( revcharmap, idx, 0); |
c490c714 YO |
2268 | STRLEN len; |
2269 | char *ch = SvPV( *tmp, len ); | |
de734bd5 A |
2270 | DEBUG_OPTIMISE_r({ |
2271 | SV *sv=sv_newmortal(); | |
8e11feef RGS |
2272 | PerlIO_printf( Perl_debug_log, |
2273 | "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n", | |
2274 | (int)depth * 2 + 2, "", | |
de734bd5 A |
2275 | (UV)state, (UV)idx, |
2276 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6, | |
2277 | PL_colors[0], PL_colors[1], | |
2278 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | | |
2279 | PERL_PV_ESCAPE_FIRSTCHAR | |
2280 | ) | |
2281 | ); | |
2282 | }); | |
3dab1dad YO |
2283 | if ( state==1 ) { |
2284 | OP( convert ) = nodetype; | |
2285 | str=STRING(convert); | |
2286 | STR_LEN(convert)=0; | |
2287 | } | |
c490c714 YO |
2288 | STR_LEN(convert) += len; |
2289 | while (len--) | |
de734bd5 | 2290 | *str++ = *ch++; |
8e11feef | 2291 | } else { |
f9049ba1 | 2292 | #ifdef DEBUGGING |
8e11feef RGS |
2293 | if (state>1) |
2294 | DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n")); | |
f9049ba1 | 2295 | #endif |
8e11feef RGS |
2296 | break; |
2297 | } | |
2298 | } | |
2e64971a | 2299 | trie->prefixlen = (state-1); |
3dab1dad | 2300 | if (str) { |
8e11feef | 2301 | regnode *n = convert+NODE_SZ_STR(convert); |
07be1b83 | 2302 | NEXT_OFF(convert) = NODE_SZ_STR(convert); |
8e11feef | 2303 | trie->startstate = state; |
07be1b83 YO |
2304 | trie->minlen -= (state - 1); |
2305 | trie->maxlen -= (state - 1); | |
33809eae JH |
2306 | #ifdef DEBUGGING |
2307 | /* At least the UNICOS C compiler choked on this | |
2308 | * being argument to DEBUG_r(), so let's just have | |
2309 | * it right here. */ | |
2310 | if ( | |
2311 | #ifdef PERL_EXT_RE_BUILD | |
2312 | 1 | |
2313 | #else | |
2314 | DEBUG_r_TEST | |
2315 | #endif | |
2316 | ) { | |
2317 | regnode *fix = convert; | |
2318 | U32 word = trie->wordcount; | |
2319 | mjd_nodelen++; | |
2320 | Set_Node_Offset_Length(convert, mjd_offset, state - 1); | |
2321 | while( ++fix < n ) { | |
2322 | Set_Node_Offset_Length(fix, 0, 0); | |
2323 | } | |
2324 | while (word--) { | |
2325 | SV ** const tmp = av_fetch( trie_words, word, 0 ); | |
2326 | if (tmp) { | |
2327 | if ( STR_LEN(convert) <= SvCUR(*tmp) ) | |
2328 | sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert)); | |
2329 | else | |
2330 | sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp)); | |
2331 | } | |
2332 | } | |
2333 | } | |
2334 | #endif | |
8e11feef RGS |
2335 | if (trie->maxlen) { |
2336 | convert = n; | |
2337 | } else { | |
3dab1dad | 2338 | NEXT_OFF(convert) = (U16)(tail - convert); |
a5ca303d | 2339 | DEBUG_r(optimize= n); |
3dab1dad YO |
2340 | } |
2341 | } | |
2342 | } | |
a5ca303d YO |
2343 | if (!jumper) |
2344 | jumper = last; | |
3dab1dad | 2345 | if ( trie->maxlen ) { |
8e11feef RGS |
2346 | NEXT_OFF( convert ) = (U16)(tail - convert); |
2347 | ARG_SET( convert, data_slot ); | |
786e8c11 YO |
2348 | /* Store the offset to the first unabsorbed branch in |
2349 | jump[0], which is otherwise unused by the jump logic. | |
2350 | We use this when dumping a trie and during optimisation. */ | |
2351 | if (trie->jump) | |
7f69552c | 2352 | trie->jump[0] = (U16)(nextbranch - convert); |
a5ca303d | 2353 | |
6c48061a YO |
2354 | /* If the start state is not accepting (meaning there is no empty string/NOTHING) |
2355 | * and there is a bitmap | |
2356 | * and the first "jump target" node we found leaves enough room | |
2357 | * then convert the TRIE node into a TRIEC node, with the bitmap | |
2358 | * embedded inline in the opcode - this is hypothetically faster. | |
2359 | */ | |
2360 | if ( !trie->states[trie->startstate].wordnum | |
2361 | && trie->bitmap | |
2362 | && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) ) | |
786e8c11 YO |
2363 | { |
2364 | OP( convert ) = TRIEC; | |
2365 | Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char); | |
446bd890 | 2366 | PerlMemShared_free(trie->bitmap); |
786e8c11 YO |
2367 | trie->bitmap= NULL; |
2368 | } else | |
2369 | OP( convert ) = TRIE; | |
a3621e74 | 2370 | |
3dab1dad YO |
2371 | /* store the type in the flags */ |
2372 | convert->flags = nodetype; | |
a5ca303d YO |
2373 | DEBUG_r({ |
2374 | optimize = convert | |
2375 | + NODE_STEP_REGNODE | |
2376 | + regarglen[ OP( convert ) ]; | |
2377 | }); | |
2378 | /* XXX We really should free up the resource in trie now, | |
2379 | as we won't use them - (which resources?) dmq */ | |
3dab1dad | 2380 | } |
a3621e74 | 2381 | /* needed for dumping*/ |
e62cc96a | 2382 | DEBUG_r(if (optimize) { |
07be1b83 | 2383 | regnode *opt = convert; |
bcdf7404 | 2384 | |
e62cc96a | 2385 | while ( ++opt < optimize) { |
07be1b83 YO |
2386 | Set_Node_Offset_Length(opt,0,0); |
2387 | } | |
786e8c11 YO |
2388 | /* |
2389 | Try to clean up some of the debris left after the | |
2390 | optimisation. | |
a3621e74 | 2391 | */ |
786e8c11 | 2392 | while( optimize < jumper ) { |
07be1b83 | 2393 | mjd_nodelen += Node_Length((optimize)); |
a3621e74 | 2394 | OP( optimize ) = OPTIMIZED; |
07be1b83 | 2395 | Set_Node_Offset_Length(optimize,0,0); |
a3621e74 YO |
2396 | optimize++; |
2397 | } | |
07be1b83 | 2398 | Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen); |
a3621e74 YO |
2399 | }); |
2400 | } /* end node insert */ | |
2e64971a DM |
2401 | |
2402 | /* Finish populating the prev field of the wordinfo array. Walk back | |
2403 | * from each accept state until we find another accept state, and if | |
2404 | * so, point the first word's .prev field at the second word. If the | |
2405 | * second already has a .prev field set, stop now. This will be the | |
2406 | * case either if we've already processed that word's accept state, | |
3b753521 FN |
2407 | * or that state had multiple words, and the overspill words were |
2408 | * already linked up earlier. | |
2e64971a DM |
2409 | */ |
2410 | { | |
2411 | U16 word; | |
2412 | U32 state; | |
2413 | U16 prev; | |
2414 | ||
2415 | for (word=1; word <= trie->wordcount; word++) { | |
2416 | prev = 0; | |
2417 | if (trie->wordinfo[word].prev) | |
2418 | continue; | |
2419 | state = trie->wordinfo[word].accept; | |
2420 | while (state) { | |
2421 | state = prev_states[state]; | |
2422 | if (!state) | |
2423 | break; | |
2424 | prev = trie->states[state].wordnum; | |
2425 | if (prev) | |
2426 | break; | |
2427 | } | |
2428 | trie->wordinfo[word].prev = prev; | |
2429 | } | |
2430 | Safefree(prev_states); | |
2431 | } | |
2432 | ||
2433 | ||
2434 | /* and now dump out the compressed format */ | |
2435 | DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1)); | |
2436 | ||
55eed653 | 2437 | RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap; |
2b8b4781 NC |
2438 | #ifdef DEBUGGING |
2439 | RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words; | |
2440 | RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap; | |
2441 | #else | |
2442 | SvREFCNT_dec(revcharmap); | |
07be1b83 | 2443 | #endif |
786e8c11 YO |
2444 | return trie->jump |
2445 | ? MADE_JUMP_TRIE | |
2446 | : trie->startstate>1 | |
2447 | ? MADE_EXACT_TRIE | |
2448 | : MADE_TRIE; | |
2449 | } | |
2450 | ||
2451 | STATIC void | |
2452 | S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth) | |
2453 | { | |
3b753521 | 2454 | /* The Trie is constructed and compressed now so we can build a fail array if it's needed |
786e8c11 YO |
2455 | |
2456 | This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the | |
2457 | "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88 | |
2458 | ISBN 0-201-10088-6 | |
2459 | ||
2460 | We find the fail state for each state in the trie, this state is the longest proper | |
3b753521 FN |
2461 | suffix of the current state's 'word' that is also a proper prefix of another word in our |
2462 | trie. State 1 represents the word '' and is thus the default fail state. This allows | |
786e8c11 YO |
2463 | the DFA not to have to restart after its tried and failed a word at a given point, it |
2464 | simply continues as though it had been matching the other word in the first place. | |
2465 | Consider | |
2466 | 'abcdgu'=~/abcdefg|cdgu/ | |
2467 | When we get to 'd' we are still matching the first word, we would encounter 'g' which would | |
3b753521 FN |
2468 | fail, which would bring us to the state representing 'd' in the second word where we would |
2469 | try 'g' and succeed, proceeding to match 'cdgu'. | |
786e8c11 YO |
2470 | */ |
2471 | /* add a fail transition */ | |
3251b653 NC |
2472 | const U32 trie_offset = ARG(source); |
2473 | reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset]; | |
786e8c11 YO |
2474 | U32 *q; |
2475 | const U32 ucharcount = trie->uniquecharcount; | |
1e2e3d02 | 2476 | const U32 numstates = trie->statecount; |
786e8c11 YO |
2477 | const U32 ubound = trie->lasttrans + ucharcount; |
2478 | U32 q_read = 0; | |
2479 | U32 q_write = 0; | |
2480 | U32 charid; | |
2481 | U32 base = trie->states[ 1 ].trans.base; | |
2482 | U32 *fail; | |
2483 | reg_ac_data *aho; | |
2484 | const U32 data_slot = add_data( pRExC_state, 1, "T" ); | |
2485 | GET_RE_DEBUG_FLAGS_DECL; | |
7918f24d NC |
2486 | |
2487 | PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE; | |
786e8c11 YO |
2488 | #ifndef DEBUGGING |
2489 | PERL_UNUSED_ARG(depth); | |
2490 | #endif | |
2491 | ||
2492 | ||
2493 | ARG_SET( stclass, data_slot ); | |
c944940b | 2494 | aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) ); |
f8fc2ecf | 2495 | RExC_rxi->data->data[ data_slot ] = (void*)aho; |
3251b653 | 2496 | aho->trie=trie_offset; |
446bd890 NC |
2497 | aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) ); |
2498 | Copy( trie->states, aho->states, numstates, reg_trie_state ); | |
786e8c11 | 2499 | Newxz( q, numstates, U32); |
c944940b | 2500 | aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) ); |
786e8c11 YO |
2501 | aho->refcount = 1; |
2502 | fail = aho->fail; | |
2503 | /* initialize fail[0..1] to be 1 so that we always have | |
2504 | a valid final fail state */ | |
2505 | fail[ 0 ] = fail[ 1 ] = 1; | |
2506 | ||
2507 | for ( charid = 0; charid < ucharcount ; charid++ ) { | |
2508 | const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 ); | |
2509 | if ( newstate ) { | |
2510 | q[ q_write ] = newstate; | |
2511 | /* set to point at the root */ | |
2512 | fail[ q[ q_write++ ] ]=1; | |
2513 | } | |
2514 | } | |
2515 | while ( q_read < q_write) { | |
2516 | const U32 cur = q[ q_read++ % numstates ]; | |
2517 | base = trie->states[ cur ].trans.base; | |
2518 | ||
2519 | for ( charid = 0 ; charid < ucharcount ; charid++ ) { | |
2520 | const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 ); | |
2521 | if (ch_state) { | |
2522 | U32 fail_state = cur; | |
2523 | U32 fail_base; | |
2524 | do { | |
2525 | fail_state = fail[ fail_state ]; | |
2526 | fail_base = aho->states[ fail_state ].trans.base; | |
2527 | } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) ); | |
2528 | ||
2529 | fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ); | |
2530 | fail[ ch_state ] = fail_state; | |
2531 | if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum ) | |
2532 | { | |
2533 | aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum; | |
2534 | } | |
2535 | q[ q_write++ % numstates] = ch_state; | |
2536 | } | |
2537 | } | |
2538 | } | |
2539 | /* restore fail[0..1] to 0 so that we "fall out" of the AC loop | |
2540 | when we fail in state 1, this allows us to use the | |
2541 | charclass scan to find a valid start char. This is based on the principle | |
2542 | that theres a good chance the string being searched contains lots of stuff | |
2543 | that cant be a start char. | |
2544 | */ | |
2545 | fail[ 0 ] = fail[ 1 ] = 0; | |
2546 | DEBUG_TRIE_COMPILE_r({ | |
6d99fb9b JH |
2547 | PerlIO_printf(Perl_debug_log, |
2548 | "%*sStclass Failtable (%"UVuf" states): 0", | |
2549 | (int)(depth * 2), "", (UV)numstates | |
1e2e3d02 | 2550 | ); |
786e8c11 YO |
2551 | for( q_read=1; q_read<numstates; q_read++ ) { |
2552 | PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]); | |
2553 | } | |
2554 | PerlIO_printf(Perl_debug_log, "\n"); | |
2555 | }); | |
2556 | Safefree(q); | |
2557 | /*RExC_seen |= REG_SEEN_TRIEDFA;*/ | |
a3621e74 YO |
2558 | } |
2559 | ||
786e8c11 | 2560 | |
a3621e74 | 2561 | /* |
5d1c421c JH |
2562 | * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2. |
2563 | * These need to be revisited when a newer toolchain becomes available. | |
2564 | */ | |
2565 | #if defined(__sparc64__) && defined(__GNUC__) | |
2566 | # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96) | |
2567 | # undef SPARC64_GCC_WORKAROUND | |
2568 | # define SPARC64_GCC_WORKAROUND 1 | |
2569 | # endif | |
2570 | #endif | |
2571 | ||
07be1b83 | 2572 | #define DEBUG_PEEP(str,scan,depth) \ |
b515a41d | 2573 | DEBUG_OPTIMISE_r({if (scan){ \ |
07be1b83 YO |
2574 | SV * const mysv=sv_newmortal(); \ |
2575 | regnode *Next = regnext(scan); \ | |
2576 | regprop(RExC_rx, mysv, scan); \ | |
7f69552c | 2577 | PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \ |
07be1b83 YO |
2578 | (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\ |
2579 | Next ? (REG_NODE_NUM(Next)) : 0 ); \ | |
b515a41d | 2580 | }}); |
07be1b83 | 2581 | |
1de06328 | 2582 | |
bb914485 KW |
2583 | /* The below joins as many adjacent EXACTish nodes as possible into a single |
2584 | * one, and looks for problematic sequences of characters whose folds vs. | |
2585 | * non-folds have sufficiently different lengths, that the optimizer would be | |
2586 | * fooled into rejecting legitimate matches of them, and the trie construction | |
87b8b349 | 2587 | * code needs to handle specially. The joining is only done if: |
bb914485 KW |
2588 | * 1) there is room in the current conglomerated node to entirely contain the |
2589 | * next one. | |
2590 | * 2) they are the exact same node type | |
2591 | * | |
87b8b349 | 2592 | * The adjacent nodes actually may be separated by NOTHING-kind nodes, and |
bb914485 KW |
2593 | * these get optimized out |
2594 | * | |
9d071ca8 KW |
2595 | * If there are problematic code sequences, *min_subtract is set to the delta |
2596 | * that the minimum size of the node can be less than its actual size. And, | |
2597 | * the node type of the result is changed to reflect that it contains these | |
bb914485 KW |
2598 | * sequences. |
2599 | * | |
a0c4c608 KW |
2600 | * And *has_exactf_sharp_s is set to indicate whether or not the node is EXACTF |
2601 | * and contains LATIN SMALL LETTER SHARP S | |
f758bddf | 2602 | * |
bb914485 KW |
2603 | * This is as good a place as any to discuss the design of handling these |
2604 | * problematic sequences. It's been wrong in Perl for a very long time. There | |
87b8b349 KW |
2605 | * are three code points currently in Unicode whose folded lengths differ so |
2606 | * much from the un-folded lengths that it causes problems for the optimizer | |
2607 | * and trie construction. Why only these are problematic, and not others where | |
2608 | * lengths also differ is something I (khw) do not understand. New versions of | |
2609 | * Unicode might add more such code points. Hopefully the logic in | |
2610 | * fold_grind.t that figures out what to test (in part by verifying that each | |
2611 | * size-combination gets tested) will catch any that do come along, so they can | |
2612 | * be added to the special handling below. The chances of new ones are | |
2613 | * actually rather small, as most, if not all, of the world's scripts that have | |
2614 | * casefolding have already been encoded by Unicode. Also, a number of | |
2615 | * Unicode's decisions were made to allow compatibility with pre-existing | |
2616 | * standards, and almost all of those have already been dealt with. These | |
2617 | * would otherwise be the most likely candidates for generating further tricky | |
2618 | * sequences. In other words, Unicode by itself is unlikely to add new ones | |
2619 | * unless it is for compatibility with pre-existing standards, and there aren't | |
2620 | * many of those left. | |
bb914485 KW |
2621 | * |
2622 | * The previous designs for dealing with these involved assigning a special | |
2623 | * node for them. This approach doesn't work, as evidenced by this example: | |
a0c4c608 | 2624 | * "\xDFs" =~ /s\xDF/ui # Used to fail before these patches |
87b8b349 | 2625 | * Both these fold to "sss", but if the pattern is parsed to create a node |
bb914485 KW |
2626 | * that would match just the \xDF, it won't be able to handle the case where a |
2627 | * successful match would have to cross the node's boundary. The new approach | |
2628 | * that hopefully generally solves the problem generates an EXACTFU_SS node | |
2629 | * that is "sss". | |
2630 | * | |
2631 | * There are a number of components to the approach (a lot of work for just | |
2632 | * three code points!): | |
2633 | * 1) This routine examines each EXACTFish node that could contain the | |
9d071ca8 KW |
2634 | * problematic sequences. It returns in *min_subtract how much to |
2635 | * subtract from the the actual length of the string to get a real minimum | |
2636 | * for one that could match it. This number is usually 0 except for the | |
2637 | * problematic sequences. This delta is used by the caller to adjust the | |
2638 | * min length of the match, and the delta between min and max, so that the | |
2639 | * optimizer doesn't reject these possibilities based on size constraints. | |
87b8b349 KW |
2640 | * 2) These sequences require special handling by the trie code, so this code |
2641 | * changes the joined node type to special ops: EXACTFU_TRICKYFOLD and | |
2642 | * EXACTFU_SS. | |
bb914485 KW |
2643 | * 3) This is sufficient for the two Greek sequences (described below), but |
2644 | * the one involving the Sharp s (\xDF) needs more. The node type | |
2645 | * EXACTFU_SS is used for an EXACTFU node that contains at least one "ss" | |
2646 | * sequence in it. For non-UTF-8 patterns and strings, this is the only | |
2647 | * case where there is a possible fold length change. That means that a | |
2648 | * regular EXACTFU node without UTF-8 involvement doesn't have to concern | |
2649 | * itself with length changes, and so can be processed faster. regexec.c | |
2650 | * takes advantage of this. Generally, an EXACTFish node that is in UTF-8 | |
2651 | * is pre-folded by regcomp.c. This saves effort in regex matching. | |
87b8b349 KW |
2652 | * However, the pre-folding isn't done for non-UTF8 patterns because the |
2653 | * fold of the MICRO SIGN requires UTF-8. Also what EXACTF and EXACTFL | |
2654 | * nodes fold to isn't known until runtime. The fold possibilities for | |
2655 | * the non-UTF8 patterns are quite simple, except for the sharp s. All | |
2656 | * the ones that don't involve a UTF-8 target string are members of a | |
2657 | * fold-pair, and arrays are set up for all of them so that the other | |
2658 | * member of the pair can be found quickly. Code elsewhere in this file | |
2659 | * makes sure that in EXACTFU nodes, the sharp s gets folded to 'ss', even | |
2660 | * if the pattern isn't UTF-8. This avoids the issues described in the | |
2661 | * next item. | |
bb914485 KW |
2662 | * 4) A problem remains for the sharp s in EXACTF nodes. Whether it matches |
2663 | * 'ss' or not is not knowable at compile time. It will match iff the | |
2664 | * target string is in UTF-8, unlike the EXACTFU nodes, where it always | |
2665 | * matches; and the EXACTFL and EXACTFA nodes where it never does. Thus | |
6012a526 | 2666 | * it can't be folded to "ss" at compile time, unlike EXACTFU does (as |
bb914485 KW |
2667 | * described in item 3). An assumption that the optimizer part of |
2668 | * regexec.c (probably unwittingly) makes is that a character in the | |
2669 | * pattern corresponds to at most a single character in the target string. | |
2670 | * (And I do mean character, and not byte here, unlike other parts of the | |
2671 | * documentation that have never been updated to account for multibyte | |
2672 | * Unicode.) This assumption is wrong only in this case, as all other | |
2673 | * cases are either 1-1 folds when no UTF-8 is involved; or is true by | |
2674 | * virtue of having this file pre-fold UTF-8 patterns. I'm | |
2675 | * reluctant to try to change this assumption, so instead the code punts. | |
9d071ca8 KW |
2676 | * This routine examines EXACTF nodes for the sharp s, and returns a |
2677 | * boolean indicating whether or not the node is an EXACTF node that | |
2678 | * contains a sharp s. When it is true, the caller sets a flag that later | |
2679 | * causes the optimizer in this file to not set values for the floating | |
2680 | * and fixed string lengths, and thus avoids the optimizer code in | |
2681 | * regexec.c that makes the invalid assumption. Thus, there is no | |
2682 | * optimization based on string lengths for EXACTF nodes that contain the | |
2683 | * sharp s. This only happens for /id rules (which means the pattern | |
2684 | * isn't in UTF-8). | |
bb914485 | 2685 | */ |
1de06328 | 2686 | |
9d071ca8 | 2687 | #define JOIN_EXACT(scan,min_subtract,has_exactf_sharp_s, flags) \ |
07be1b83 | 2688 | if (PL_regkind[OP(scan)] == EXACT) \ |
9d071ca8 | 2689 | join_exact(pRExC_state,(scan),(min_subtract),has_exactf_sharp_s, (flags),NULL,depth+1) |
07be1b83 | 2690 | |
be8e71aa | 2691 | STATIC U32 |
9d071ca8 | 2692 | S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, UV *min_subtract, bool *has_exactf_sharp_s, U32 flags,regnode *val, U32 depth) { |
07be1b83 YO |
2693 | /* Merge several consecutive EXACTish nodes into one. */ |
2694 | regnode *n = regnext(scan); | |
2695 | U32 stringok = 1; | |
2696 | regnode *next = scan + NODE_SZ_STR(scan); | |
2697 | U32 merged = 0; | |
2698 | U32 stopnow = 0; | |
2699 | #ifdef DEBUGGING | |
2700 | regnode *stop = scan; | |
72f13be8 | 2701 | GET_RE_DEBUG_FLAGS_DECL; |
f9049ba1 | 2702 | #else |
d47053eb RGS |
2703 | PERL_UNUSED_ARG(depth); |
2704 | #endif | |
7918f24d NC |
2705 | |
2706 | PERL_ARGS_ASSERT_JOIN_EXACT; | |
d47053eb | 2707 | #ifndef EXPERIMENTAL_INPLACESCAN |
f9049ba1 SP |
2708 | PERL_UNUSED_ARG(flags); |
2709 | PERL_UNUSED_ARG(val); | |
07be1b83 | 2710 | #endif |
07be1b83 | 2711 | DEBUG_PEEP("join",scan,depth); |
bb914485 | 2712 | |
3f410cf6 KW |
2713 | /* Look through the subsequent nodes in the chain. Skip NOTHING, merge |
2714 | * EXACT ones that are mergeable to the current one. */ | |
2715 | while (n | |
2716 | && (PL_regkind[OP(n)] == NOTHING | |
2717 | || (stringok && OP(n) == OP(scan))) | |
07be1b83 | 2718 | && NEXT_OFF(n) |
3f410cf6 KW |
2719 | && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) |
2720 | { | |
07be1b83 YO |
2721 | |
2722 | if (OP(n) == TAIL || n > next) | |
2723 | stringok = 0; | |
2724 | if (PL_regkind[OP(n)] == NOTHING) { | |
07be1b83 YO |
2725 | DEBUG_PEEP("skip:",n,depth); |
2726 | NEXT_OFF(scan) += NEXT_OFF(n); | |
2727 | next = n + NODE_STEP_REGNODE; | |
2728 | #ifdef DEBUGGING | |
2729 | if (stringok) | |
2730 | stop = n; | |
2731 | #endif | |
2732 | n = regnext(n); | |
2733 | } | |
2734 | else if (stringok) { | |
786e8c11 | 2735 | const unsigned int oldl = STR_LEN(scan); |
07be1b83 | 2736 | regnode * const nnext = regnext(n); |
b2230d39 | 2737 | |
87b8b349 KW |
2738 | /* XXX I (khw) kind of doubt that this works on platforms where |
2739 | * U8_MAX is above 255 because of lots of other assumptions */ | |
b2230d39 KW |
2740 | if (oldl + STR_LEN(n) > U8_MAX) |
2741 | break; | |
07be1b83 YO |
2742 | |
2743 | DEBUG_PEEP("merg",n,depth); | |
07be1b83 | 2744 | merged++; |
b2230d39 | 2745 | |
07be1b83 YO |
2746 | NEXT_OFF(scan) += NEXT_OFF(n); |
2747 | STR_LEN(scan) += STR_LEN(n); | |
2748 | next = n + NODE_SZ_STR(n); | |
2749 | /* Now we can overwrite *n : */ | |
2750 | Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char); | |
2751 | #ifdef DEBUGGING | |
2752 | stop = next - 1; | |
2753 | #endif | |
2754 | n = nnext; | |
2755 | if (stopnow) break; | |
2756 | } | |
2757 | ||
d47053eb RGS |
2758 | #ifdef EXPERIMENTAL_INPLACESCAN |
2759 | if (flags && !NEXT_OFF(n)) { | |
2760 | DEBUG_PEEP("atch", val, depth); | |
2761 | if (reg_off_by_arg[OP(n)]) { | |
2762 | ARG_SET(n, val - n); | |
2763 | } | |
2764 | else { | |
2765 | NEXT_OFF(n) = val - n; | |
2766 | } | |
2767 | stopnow = 1; | |
2768 | } | |
07be1b83 YO |
2769 | #endif |
2770 | } | |
2c2b7f86 | 2771 | |
9d071ca8 | 2772 | *min_subtract = 0; |
f758bddf | 2773 | *has_exactf_sharp_s = FALSE; |
f646642f | 2774 | |
3f410cf6 KW |
2775 | /* Here, all the adjacent mergeable EXACTish nodes have been merged. We |
2776 | * can now analyze for sequences of problematic code points. (Prior to | |
2777 | * this final joining, sequences could have been split over boundaries, and | |
a0c4c608 KW |
2778 | * hence missed). The sequences only happen in folding, hence for any |
2779 | * non-EXACT EXACTish node */ | |
86d6fcad | 2780 | if (OP(scan) != EXACT) { |
f758bddf KW |
2781 | U8 *s; |
2782 | U8 * s0 = (U8*) STRING(scan); | |
2783 | U8 * const s_end = s0 + STR_LEN(scan); | |
2784 | ||
2785 | /* The below is perhaps overboard, but this allows us to save a test | |
2786 | * each time through the loop at the expense of a mask. This is | |
2787 | * because on both EBCDIC and ASCII machines, 'S' and 's' differ by a | |
2788 | * single bit. On ASCII they are 32 apart; on EBCDIC, they are 64. | |
2789 | * This uses an exclusive 'or' to find that bit and then inverts it to | |
2790 | * form a mask, with just a single 0, in the bit position where 'S' and | |
2791 | * 's' differ. */ | |
dbeb8947 | 2792 | const U8 S_or_s_mask = (U8) ~ ('S' ^ 's'); |
f758bddf KW |
2793 | const U8 s_masked = 's' & S_or_s_mask; |
2794 | ||
2795 | /* One pass is made over the node's string looking for all the | |
2796 | * possibilities. to avoid some tests in the loop, there are two main | |
2797 | * cases, for UTF-8 patterns (which can't have EXACTF nodes) and | |
2798 | * non-UTF-8 */ | |
2799 | if (UTF) { | |
86d6fcad | 2800 | |
f758bddf KW |
2801 | /* There are two problematic Greek code points in Unicode |
2802 | * casefolding | |
86d6fcad KW |
2803 | * |
2804 | * U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS | |
2805 | * U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS | |
2806 | * | |
2807 | * which casefold to | |
2808 | * | |
2809 | * Unicode UTF-8 | |
2810 | * | |
2811 | * U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81 | |
2812 | * U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81 | |
2813 | * | |
2814 | * This means that in case-insensitive matching (or "loose | |
2815 | * matching", as Unicode calls it), an EXACTF of length six (the | |
2816 | * UTF-8 encoded byte length of the above casefolded versions) can | |
2817 | * match a target string of length two (the byte length of UTF-8 | |
2818 | * encoded U+0390 or U+03B0). This would rather mess up the | |
2819 | * minimum length computation. (there are other code points that | |
2820 | * also fold to these two sequences, but the delta is smaller) | |
2821 | * | |
f758bddf KW |
2822 | * If these sequences are found, the minimum length is decreased by |
2823 | * four (six minus two). | |
86d6fcad | 2824 | * |
f758bddf KW |
2825 | * Similarly, 'ss' may match the single char and byte LATIN SMALL |
2826 | * LETTER SHARP S. We decrease the min length by 1 for each | |
2827 | * occurrence of 'ss' found */ | |
3f410cf6 | 2828 | |
61dad979 KW |
2829 | #define U390_FIRST_BYTE GREEK_SMALL_LETTER_IOTA_UTF8_FIRST_BYTE |
2830 | #define U3B0_FIRST_BYTE GREEK_SMALL_LETTER_UPSILON_UTF8_FIRST_BYTE | |
2831 | const U8 U390_tail[] = GREEK_SMALL_LETTER_IOTA_UTF8_TAIL | |
2832 | COMBINING_DIAERESIS_UTF8 | |
2833 | COMBINING_ACUTE_ACCENT_UTF8; | |
2834 | const U8 U3B0_tail[] = GREEK_SMALL_LETTER_UPSILON_UTF8_TAIL | |
2835 | COMBINING_DIAERESIS_UTF8 | |
2836 | COMBINING_ACUTE_ACCENT_UTF8; | |
2837 | const U8 len = sizeof(U390_tail); /* (-1 for NUL; +1 for 1st byte; | |
f758bddf KW |
2838 | yields a net of 0 */ |
2839 | /* Examine the string for one of the problematic sequences */ | |
2840 | for (s = s0; | |
2841 | s < s_end - 1; /* Can stop 1 before the end, as minimum length | |
2842 | * sequence we are looking for is 2 */ | |
2843 | s += UTF8SKIP(s)) | |
86d6fcad | 2844 | { |
bb914485 | 2845 | |
f758bddf KW |
2846 | /* Look for the first byte in each problematic sequence */ |
2847 | switch (*s) { | |
2848 | /* We don't have to worry about other things that fold to | |
2849 | * 's' (such as the long s, U+017F), as all above-latin1 | |
2850 | * code points have been pre-folded */ | |
2851 | case 's': | |
2852 | case 'S': | |
2853 | ||
a0c4c608 KW |
2854 | /* Current character is an 's' or 'S'. If next one is |
2855 | * as well, we have the dreaded sequence */ | |
f758bddf KW |
2856 | if (((*(s+1) & S_or_s_mask) == s_masked) |
2857 | /* These two node types don't have special handling | |
2858 | * for 'ss' */ | |
2859 | && OP(scan) != EXACTFL && OP(scan) != EXACTFA) | |
2860 | { | |
9d071ca8 | 2861 | *min_subtract += 1; |
f758bddf KW |
2862 | OP(scan) = EXACTFU_SS; |
2863 | s++; /* No need to look at this character again */ | |
2864 | } | |
2865 | break; | |
2866 | ||
61dad979 | 2867 | case U390_FIRST_BYTE: |
f758bddf KW |
2868 | if (s_end - s >= len |
2869 | ||
2870 | /* The 1's are because are skipping comparing the | |
2871 | * first byte */ | |
2872 | && memEQ(s + 1, U390_tail, len - 1)) | |
2873 | { | |
2874 | goto greek_sequence; | |
2875 | } | |
2876 | break; | |
2877 | ||
61dad979 | 2878 | case U3B0_FIRST_BYTE: |
f758bddf KW |
2879 | if (! (s_end - s >= len |
2880 | && memEQ(s + 1, U3B0_tail, len - 1))) | |
2881 | { | |
2882 | break; | |
2883 | } | |
2884 | greek_sequence: | |
9d071ca8 | 2885 | *min_subtract += 4; |
f758bddf | 2886 | |
87b8b349 | 2887 | /* This requires special handling by trie's, so change |
f758bddf KW |
2888 | * the node type to indicate this. If EXACTFA and |
2889 | * EXACTFL were ever to be handled by trie's, this | |
2890 | * would have to be changed. If this node has already | |
2891 | * been changed to EXACTFU_SS in this loop, leave it as | |
2892 | * is. (I (khw) think it doesn't matter in regexec.c | |
2893 | * for UTF patterns, but no need to change it */ | |
2894 | if (OP(scan) == EXACTFU) { | |
fab2782b | 2895 | OP(scan) = EXACTFU_TRICKYFOLD; |
f758bddf KW |
2896 | } |
2897 | s += 6; /* We already know what this sequence is. Skip | |
2898 | the rest of it */ | |
2899 | break; | |
bb914485 KW |
2900 | } |
2901 | } | |
2902 | } | |
f758bddf | 2903 | else if (OP(scan) != EXACTFL && OP(scan) != EXACTFA) { |
bb914485 | 2904 | |
f758bddf KW |
2905 | /* Here, the pattern is not UTF-8. We need to look only for the |
2906 | * 'ss' sequence, and in the EXACTF case, the sharp s, which can be | |
2907 | * in the final position. Otherwise we can stop looking 1 byte | |
2908 | * earlier because have to find both the first and second 's' */ | |
2909 | const U8* upper = (OP(scan) == EXACTF) ? s_end : s_end -1; | |
2910 | ||
2911 | for (s = s0; s < upper; s++) { | |
2912 | switch (*s) { | |
2913 | case 'S': | |
2914 | case 's': | |
2915 | if (s_end - s > 1 | |
2916 | && ((*(s+1) & S_or_s_mask) == s_masked)) | |
2917 | { | |
9d071ca8 | 2918 | *min_subtract += 1; |
f758bddf KW |
2919 | |
2920 | /* EXACTF nodes need to know that the minimum | |
2921 | * length changed so that a sharp s in the string | |
2922 | * can match this ss in the pattern, but they | |
87b8b349 KW |
2923 | * remain EXACTF nodes, as they won't match this |
2924 | * unless the target string is is UTF-8, which we | |
2925 | * don't know until runtime */ | |
f758bddf KW |
2926 | if (OP(scan) != EXACTF) { |
2927 | OP(scan) = EXACTFU_SS; | |
2928 | } | |
2929 | s++; | |
2930 | } | |
2931 | break; | |
2932 | case LATIN_SMALL_LETTER_SHARP_S: | |
2933 | if (OP(scan) == EXACTF) { | |
2934 | *has_exactf_sharp_s = TRUE; | |
2935 | } | |
2936 | break; | |
86d6fcad KW |
2937 | } |
2938 | } | |
2939 | } | |
07be1b83 | 2940 | } |
3f410cf6 | 2941 | |
07be1b83 | 2942 | #ifdef DEBUGGING |
bb789b09 DM |
2943 | /* Allow dumping but overwriting the collection of skipped |
2944 | * ops and/or strings with fake optimized ops */ | |
07be1b83 YO |
2945 | n = scan + NODE_SZ_STR(scan); |
2946 | while (n <= stop) { | |
bb789b09 DM |
2947 | OP(n) = OPTIMIZED; |
2948 | FLAGS(n) = 0; | |
2949 | NEXT_OFF(n) = 0; | |
07be1b83 YO |
2950 | n++; |
2951 | } | |
2952 | #endif | |
2953 | DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)}); | |
2954 | return stopnow; | |
2955 | } | |
2956 | ||
486ec47a | 2957 | /* REx optimizer. Converts nodes into quicker variants "in place". |
653099ff GS |
2958 | Finds fixed substrings. */ |
2959 | ||
a0288114 | 2960 | /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set |
c277df42 IZ |
2961 | to the position after last scanned or to NULL. */ |
2962 | ||
40d049e4 YO |
2963 | #define INIT_AND_WITHP \ |
2964 | assert(!and_withp); \ | |
2965 | Newx(and_withp,1,struct regnode_charclass_class); \ | |
2966 | SAVEFREEPV(and_withp) | |
07be1b83 | 2967 | |
b515a41d | 2968 | /* this is a chain of data about sub patterns we are processing that |
486ec47a | 2969 | need to be handled separately/specially in study_chunk. Its so |
b515a41d YO |
2970 | we can simulate recursion without losing state. */ |
2971 | struct scan_frame; | |
2972 | typedef struct scan_frame { | |
2973 | regnode *last; /* last node to process in this frame */ | |
2974 | regnode *next; /* next node to process when last is reached */ | |
2975 | struct scan_frame *prev; /*previous frame*/ | |
2976 | I32 stop; /* what stopparen do we use */ | |
2977 | } scan_frame; | |
2978 | ||
304ee84b YO |
2979 | |
2980 | #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf) | |
2981 | ||
e1d1eefb YO |
2982 | #define CASE_SYNST_FNC(nAmE) \ |
2983 | case nAmE: \ | |
2984 | if (flags & SCF_DO_STCLASS_AND) { \ | |
2985 | for (value = 0; value < 256; value++) \ | |
2986 | if (!is_ ## nAmE ## _cp(value)) \ | |
2987 | ANYOF_BITMAP_CLEAR(data->start_class, value); \ | |
2988 | } \ | |
2989 | else { \ | |
2990 | for (value = 0; value < 256; value++) \ | |
2991 | if (is_ ## nAmE ## _cp(value)) \ | |
2992 | ANYOF_BITMAP_SET(data->start_class, value); \ | |
2993 | } \ | |
2994 | break; \ | |
2995 | case N ## nAmE: \ | |
2996 | if (flags & SCF_DO_STCLASS_AND) { \ | |
2997 | for (value = 0; value < 256; value++) \ | |
2998 | if (is_ ## nAmE ## _cp(value)) \ | |
2999 | ANYOF_BITMAP_CLEAR(data->start_class, value); \ | |
3000 | } \ | |
3001 | else { \ | |
3002 | for (value = 0; value < 256; value++) \ | |
3003 | if (!is_ ## nAmE ## _cp(value)) \ | |
3004 | ANYOF_BITMAP_SET(data->start_class, value); \ | |
3005 | } \ | |
3006 | break | |
3007 | ||
3008 | ||
3009 | ||
76e3520e | 3010 | STATIC I32 |
40d049e4 | 3011 | S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, |
1de06328 | 3012 | I32 *minlenp, I32 *deltap, |
40d049e4 YO |
3013 | regnode *last, |
3014 | scan_data_t *data, | |
3015 | I32 stopparen, | |
3016 | U8* recursed, | |
3017 | struct regnode_charclass_class *and_withp, | |
3018 | U32 flags, U32 depth) | |
c277df42 IZ |
3019 | /* scanp: Start here (read-write). */ |
3020 | /* deltap: Write maxlen-minlen here. */ | |
3021 | /* last: Stop before this one. */ | |
40d049e4 YO |
3022 | /* data: string data about the pattern */ |
3023 | /* stopparen: treat close N as END */ | |
3024 | /* recursed: which subroutines have we recursed into */ | |
3025 | /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */ | |
c277df42 | 3026 | { |
97aff369 | 3027 | dVAR; |
c277df42 IZ |
3028 | I32 min = 0, pars = 0, code; |
3029 | regnode *scan = *scanp, *next; | |
3030 | I32 delta = 0; | |
3031 | int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF); | |
aca2d497 | 3032 | int is_inf_internal = 0; /* The studied chunk is infinite */ |
c277df42 IZ |
3033 | I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0; |
3034 | scan_data_t data_fake; | |
a3621e74 | 3035 | SV *re_trie_maxbuff = NULL; |
786e8c11 | 3036 | regnode *first_non_open = scan; |
e2e6a0f1 | 3037 | I32 stopmin = I32_MAX; |
8aa23a47 | 3038 | scan_frame *frame = NULL; |
a3621e74 | 3039 | GET_RE_DEBUG_FLAGS_DECL; |
8aa23a47 | 3040 | |
7918f24d NC |
3041 | PERL_ARGS_ASSERT_STUDY_CHUNK; |
3042 | ||
13a24bad | 3043 | #ifdef DEBUGGING |
40d049e4 | 3044 | StructCopy(&zero_scan_data, &data_fake, scan_data_t); |
13a24bad | 3045 | #endif |
40d049e4 | 3046 | |
786e8c11 | 3047 | if ( depth == 0 ) { |
40d049e4 | 3048 | while (first_non_open && OP(first_non_open) == OPEN) |
786e8c11 YO |
3049 | first_non_open=regnext(first_non_open); |
3050 | } | |
3051 | ||
b81d288d | 3052 | |
8aa23a47 YO |
3053 | fake_study_recurse: |
3054 | while ( scan && OP(scan) != END && scan < last ){ | |
9d071ca8 KW |
3055 | UV min_subtract = 0; /* How much to subtract from the minimum node |
3056 | length to get a real minimum (because the | |
3057 | folded version may be shorter) */ | |
f758bddf | 3058 | bool has_exactf_sharp_s = FALSE; |
8aa23a47 | 3059 | /* Peephole optimizer: */ |
304ee84b | 3060 | DEBUG_STUDYDATA("Peep:", data,depth); |
8aa23a47 | 3061 | DEBUG_PEEP("Peep",scan,depth); |
a0c4c608 KW |
3062 | |
3063 | /* Its not clear to khw or hv why this is done here, and not in the | |
3064 | * clauses that deal with EXACT nodes. khw's guess is that it's | |
3065 | * because of a previous design */ | |
9d071ca8 | 3066 | JOIN_EXACT(scan,&min_subtract, &has_exactf_sharp_s, 0); |
8aa23a47 YO |
3067 | |
3068 | /* Follow the next-chain of the current node and optimize | |
3069 | away all the NOTHINGs from it. */ | |
3070 | if (OP(scan) != CURLYX) { | |
3071 | const int max = (reg_off_by_arg[OP(scan)] | |
3072 | ? I32_MAX | |
3073 | /* I32 may be smaller than U16 on CRAYs! */ | |
3074 | : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX)); | |
3075 | int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan)); | |
3076 | int noff; | |
3077 | regnode *n = scan; | |
686b73d4 | 3078 | |
8aa23a47 YO |
3079 | /* Skip NOTHING and LONGJMP. */ |
3080 | while ((n = regnext(n)) | |
3081 | && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n))) | |
3082 | || ((OP(n) == LONGJMP) && (noff = ARG(n)))) | |
3083 | && off + noff < max) | |
3084 | off += noff; | |
3085 | if (reg_off_by_arg[OP(scan)]) | |
3086 | ARG(scan) = off; | |
3087 | else | |
3088 | NEXT_OFF(scan) = off; | |
3089 | } | |
a3621e74 | 3090 | |
c277df42 | 3091 | |
8aa23a47 YO |
3092 | |
3093 | /* The principal pseudo-switch. Cannot be a switch, since we | |
3094 | look into several different things. */ | |
3095 | if (OP(scan) == BRANCH || OP(scan) == BRANCHJ | |
3096 | || OP(scan) == IFTHEN) { | |
3097 | next = regnext(scan); | |
3098 | code = OP(scan); | |
3099 | /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */ | |
686b73d4 | 3100 | |
8aa23a47 YO |
3101 | if (OP(next) == code || code == IFTHEN) { |
3102 | /* NOTE - There is similar code to this block below for handling | |
3103 | TRIE nodes on a re-study. If you change stuff here check there | |
3104 | too. */ | |
3105 | I32 max1 = 0, min1 = I32_MAX, num = 0; | |
3106 | struct regnode_charclass_class accum; | |
3107 | regnode * const startbranch=scan; | |
686b73d4 | 3108 | |
8aa23a47 | 3109 | if (flags & SCF_DO_SUBSTR) |
304ee84b | 3110 | SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */ |
8aa23a47 | 3111 | if (flags & SCF_DO_STCLASS) |
e755fd73 | 3112 | cl_init_zero(pRExC_state, &accum); |
8aa23a47 YO |
3113 | |
3114 | while (OP(scan) == code) { | |
3115 | I32 deltanext, minnext, f = 0, fake; | |
3116 | struct regnode_charclass_class this_class; | |
3117 | ||
3118 | num++; | |
3119 | data_fake.flags = 0; | |
3120 | if (data) { | |
3121 | data_fake.whilem_c = data->whilem_c; | |
3122 | data_fake.last_closep = data->last_closep; | |
3123 | } | |
3124 | else | |
3125 | data_fake.last_closep = &fake; | |
58e23c8d YO |
3126 | |
3127 | data_fake.pos_delta = delta; | |
8aa23a47 YO |
3128 | next = regnext(scan); |
3129 | scan = NEXTOPER(scan); | |
3130 | if (code != BRANCH) | |
c277df42 | 3131 | scan = NEXTOPER(scan); |
8aa23a47 | 3132 | if (flags & SCF_DO_STCLASS) { |
e755fd73 | 3133 | cl_init(pRExC_state, &this_class); |
8aa23a47 YO |
3134 | data_fake.start_class = &this_class; |
3135 | f = SCF_DO_STCLASS_AND; | |
58e23c8d | 3136 | } |
8aa23a47 YO |
3137 | if (flags & SCF_WHILEM_VISITED_POS) |
3138 | f |= SCF_WHILEM_VISITED_POS; | |
3139 | ||
3140 | /* we suppose the run is continuous, last=next...*/ | |
3141 | minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, | |
3142 | next, &data_fake, | |
3143 | stopparen, recursed, NULL, f,depth+1); | |
3144 | if (min1 > minnext) | |
3145 | min1 = minnext; | |
3146 | if (max1 < minnext + deltanext) | |
3147 | max1 = minnext + deltanext; | |
3148 | if (deltanext == I32_MAX) | |
3149 | is_inf = is_inf_internal = 1; | |
3150 | scan = next; | |
3151 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) | |
3152 | pars++; | |
3153 | if (data_fake.flags & SCF_SEEN_ACCEPT) { | |
3154 | if ( stopmin > minnext) | |
3155 | stopmin = min + min1; | |
3156 | flags &= ~SCF_DO_SUBSTR; | |
3157 | if (data) | |
3158 | data->flags |= SCF_SEEN_ACCEPT; | |
3159 | } | |
3160 | if (data) { | |
3161 | if (data_fake.flags & SF_HAS_EVAL) | |
3162 | data->flags |= SF_HAS_EVAL; | |
3163 | data->whilem_c = data_fake.whilem_c; | |
3dab1dad | 3164 | } |
8aa23a47 | 3165 | if (flags & SCF_DO_STCLASS) |
3fffb88a | 3166 | cl_or(pRExC_state, &accum, &this_class); |
8aa23a47 YO |
3167 | } |
3168 | if (code == IFTHEN && num < 2) /* Empty ELSE branch */ | |
3169 | min1 = 0; | |
3170 | if (flags & SCF_DO_SUBSTR) { | |
3171 | data->pos_min += min1; | |
3172 | data->pos_delta += max1 - min1; | |
3173 | if (max1 != min1 || is_inf) | |
3174 | data->longest = &(data->longest_float); | |
3175 | } | |
3176 | min += min1; | |
3177 | delta += max1 - min1; | |
3178 | if (flags & SCF_DO_STCLASS_OR) { | |
3fffb88a | 3179 | cl_or(pRExC_state, data->start_class, &accum); |
8aa23a47 YO |
3180 | if (min1) { |
3181 | cl_and(data->start_class, and_withp); | |
3182 | flags &= ~SCF_DO_STCLASS; | |
653099ff | 3183 | } |
8aa23a47 YO |
3184 | } |
3185 | else if (flags & SCF_DO_STCLASS_AND) { | |
3186 | if (min1) { | |
3187 | cl_and(data->start_class, &accum); | |
3188 | flags &= ~SCF_DO_STCLASS; | |
de0c8cb8 | 3189 | } |
8aa23a47 YO |
3190 | else { |
3191 | /* Switch to OR mode: cache the old value of | |
3192 | * data->start_class */ | |
3193 | INIT_AND_WITHP; | |
3194 | StructCopy(data->start_class, and_withp, | |
3195 | struct regnode_charclass_class); | |
3196 | flags &= ~SCF_DO_STCLASS_AND; | |
3197 | StructCopy(&accum, data->start_class, | |
3198 | struct regnode_charclass_class); | |
3199 | flags |= SCF_DO_STCLASS_OR; | |
3200 | data->start_class->flags |= ANYOF_EOS; | |
de0c8cb8 | 3201 | } |
8aa23a47 | 3202 | } |
a3621e74 | 3203 | |
8aa23a47 YO |
3204 | if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) { |
3205 | /* demq. | |
a3621e74 | 3206 | |
8aa23a47 YO |
3207 | Assuming this was/is a branch we are dealing with: 'scan' now |
3208 | points at the item that follows the branch sequence, whatever | |
3209 | it is. We now start at the beginning of the sequence and look | |
3210 | for subsequences of | |
a3621e74 | 3211 | |
8aa23a47 YO |
3212 | BRANCH->EXACT=>x1 |
3213 | BRANCH->EXACT=>x2 | |
3214 | tail | |
a3621e74 | 3215 | |
8aa23a47 | 3216 | which would be constructed from a pattern like /A|LIST|OF|WORDS/ |
a3621e74 | 3217 | |
486ec47a | 3218 | If we can find such a subsequence we need to turn the first |
8aa23a47 YO |
3219 | element into a trie and then add the subsequent branch exact |
3220 | strings to the trie. | |
a3621e74 | 3221 | |
8aa23a47 | 3222 | We have two cases |
a3621e74 | 3223 | |
3b753521 | 3224 | 1. patterns where the whole set of branches can be converted. |
a3621e74 | 3225 | |
8aa23a47 | 3226 | 2. patterns where only a subset can be converted. |
a3621e74 | 3227 | |
8aa23a47 YO |
3228 | In case 1 we can replace the whole set with a single regop |
3229 | for the trie. In case 2 we need to keep the start and end | |
3b753521 | 3230 | branches so |
a3621e74 | 3231 | |
8aa23a47 YO |
3232 | 'BRANCH EXACT; BRANCH EXACT; BRANCH X' |
3233 | becomes BRANCH TRIE; BRANCH X; | |
786e8c11 | 3234 | |
8aa23a47 YO |
3235 | There is an additional case, that being where there is a |
3236 | common prefix, which gets split out into an EXACT like node | |
3237 | preceding the TRIE node. | |
a3621e74 | 3238 | |
8aa23a47 YO |
3239 | If x(1..n)==tail then we can do a simple trie, if not we make |
3240 | a "jump" trie, such that when we match the appropriate word | |
486ec47a | 3241 | we "jump" to the appropriate tail node. Essentially we turn |
8aa23a47 | 3242 | a nested if into a case structure of sorts. |
b515a41d | 3243 | |
8aa23a47 | 3244 | */ |
686b73d4 | 3245 | |
8aa23a47 YO |
3246 | int made=0; |
3247 | if (!re_trie_maxbuff) { | |
3248 | re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1); | |
3249 | if (!SvIOK(re_trie_maxbuff)) | |
3250 | sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); | |
3251 | } | |
3252 | if ( SvIV(re_trie_maxbuff)>=0 ) { | |
3253 | regnode *cur; | |
3254 | regnode *first = (regnode *)NULL; | |
3255 | regnode *last = (regnode *)NULL; | |
3256 | regnode *tail = scan; | |
fab2782b | 3257 | U8 trietype = 0; |
8aa23a47 | 3258 | U32 count=0; |
a3621e74 YO |
3259 | |
3260 | #ifdef DEBUGGING | |
8aa23a47 | 3261 | SV * const mysv = sv_newmortal(); /* for dumping */ |
a3621e74 | 3262 | #endif |
8aa23a47 YO |
3263 | /* var tail is used because there may be a TAIL |
3264 | regop in the way. Ie, the exacts will point to the | |
3265 | thing following the TAIL, but the last branch will | |
3266 | point at the TAIL. So we advance tail. If we | |
3267 | have nested (?:) we may have to move through several | |
3268 | tails. | |
3269 | */ | |
3270 | ||
3271 | while ( OP( tail ) == TAIL ) { | |
3272 | /* this is the TAIL generated by (?:) */ | |
3273 | tail = regnext( tail ); | |
3274 | } | |
a3621e74 | 3275 | |
8aa23a47 | 3276 | |
df826430 | 3277 | DEBUG_TRIE_COMPILE_r({ |
8aa23a47 YO |
3278 | regprop(RExC_rx, mysv, tail ); |
3279 | PerlIO_printf( Perl_debug_log, "%*s%s%s\n", | |
3280 | (int)depth * 2 + 2, "", | |
3281 | "Looking for TRIE'able sequences. Tail node is: ", | |
3282 | SvPV_nolen_const( mysv ) | |
3283 | ); | |
3284 | }); | |
3285 | ||
3286 | /* | |
3287 | ||
fab2782b YO |
3288 | Step through the branches |
3289 | cur represents each branch, | |
3290 | noper is the first thing to be matched as part of that branch | |
3291 | noper_next is the regnext() of that node. | |
3292 | ||
3293 | We normally handle a case like this /FOO[xyz]|BAR[pqr]/ | |
3294 | via a "jump trie" but we also support building with NOJUMPTRIE, | |
3295 | which restricts the trie logic to structures like /FOO|BAR/. | |
3296 | ||
3297 | If noper is a trieable nodetype then the branch is a possible optimization | |
3298 | target. If we are building under NOJUMPTRIE then we require that noper_next | |
3299 | is the same as scan (our current position in the regex program). | |
3300 | ||
3301 | Once we have two or more consecutive such branches we can create a | |
3302 | trie of the EXACT's contents and stitch it in place into the program. | |
3303 | ||
3304 | If the sequence represents all of the branches in the alternation we | |
3305 | replace the entire thing with a single TRIE node. | |
3306 | ||
3307 | Otherwise when it is a subsequence we need to stitch it in place and | |
3308 | replace only the relevant branches. This means the first branch has | |
3309 | to remain as it is used by the alternation logic, and its next pointer, | |
3310 | and needs to be repointed at the item on the branch chain following | |
3311 | the last branch we have optimized away. | |
3312 | ||
3313 | This could be either a BRANCH, in which case the subsequence is internal, | |
3314 | or it could be the item following the branch sequence in which case the | |
3315 | subsequence is at the end (which does not necessarily mean the first node | |
3316 | is the start of the alternation). | |
3317 | ||
3318 | TRIE_TYPE(X) is a define which maps the optype to a trietype. | |
3319 | ||
3320 | optype | trietype | |
3321 | ----------------+----------- | |
3322 | NOTHING | NOTHING | |
3323 | EXACT | EXACT | |
3324 | EXACTFU | EXACTFU | |
3325 | EXACTFU_SS | EXACTFU | |
3326 | EXACTFU_TRICKYFOLD | EXACTFU | |
3327 | EXACTFA | 0 | |
3328 | ||
8aa23a47 YO |
3329 | |
3330 | */ | |
fab2782b YO |
3331 | #define TRIE_TYPE(X) ( ( NOTHING == (X) ) ? NOTHING : \ |
3332 | ( EXACT == (X) ) ? EXACT : \ | |
3333 | ( EXACTFU == (X) || EXACTFU_SS == (X) || EXACTFU_TRICKYFOLD == (X) ) ? EXACTFU : \ | |
3334 | 0 ) | |
8aa23a47 YO |
3335 | |
3336 | /* dont use tail as the end marker for this traverse */ | |
3337 | for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) { | |
3338 | regnode * const noper = NEXTOPER( cur ); | |
fab2782b YO |
3339 | U8 noper_type = OP( noper ); |
3340 | U8 noper_trietype = TRIE_TYPE( noper_type ); | |
b515a41d | 3341 | #if defined(DEBUGGING) || defined(NOJUMPTRIE) |
8aa23a47 | 3342 | regnode * const noper_next = regnext( noper ); |
df826430 YO |
3343 | U8 noper_next_type = (noper_next && noper_next != tail) ? OP(noper_next) : 0; |
3344 | U8 noper_next_trietype = (noper_next && noper_next != tail) ? TRIE_TYPE( noper_next_type ) :0; | |
b515a41d YO |
3345 | #endif |
3346 | ||
df826430 | 3347 | DEBUG_TRIE_COMPILE_r({ |
8aa23a47 YO |
3348 | regprop(RExC_rx, mysv, cur); |
3349 | PerlIO_printf( Perl_debug_log, "%*s- %s (%d)", | |
3350 | (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) ); | |
3351 | ||
3352 | regprop(RExC_rx, mysv, noper); | |
3353 | PerlIO_printf( Perl_debug_log, " -> %s", | |
3354 | SvPV_nolen_const(mysv)); | |
3355 | ||
3356 | if ( noper_next ) { | |
3357 | regprop(RExC_rx, mysv, noper_next ); | |
3358 | PerlIO_printf( Perl_debug_log,"\t=> %s\t", | |
3359 | SvPV_nolen_const(mysv)); | |
3360 | } | |
df826430 YO |
3361 | PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d,tt==%s,nt==%s,nnt==%s)\n", |
3362 | REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur), | |
3363 | PL_reg_name[trietype], PL_reg_name[noper_trietype], PL_reg_name[noper_next_trietype] | |
3364 | ); | |
8aa23a47 | 3365 | }); |
fab2782b YO |
3366 | |
3367 | /* Is noper a trieable nodetype that can be merged with the | |
3368 | * current trie (if there is one)? */ | |
3369 | if ( noper_trietype | |
3370 | && | |
3371 | ( | |
df826430 YO |
3372 | ( noper_trietype == NOTHING) |
3373 | || ( trietype == NOTHING ) | |
a40630bf | 3374 | || ( trietype == noper_trietype ) |
fab2782b | 3375 | ) |
786e8c11 | 3376 | #ifdef NOJUMPTRIE |
8aa23a47 | 3377 | && noper_next == tail |
786e8c11 | 3378 | #endif |
8aa23a47 YO |
3379 | && count < U16_MAX) |
3380 | { | |
fab2782b YO |
3381 | /* Handle mergable triable node |
3382 | * Either we are the first node in a new trieable sequence, | |
3383 | * in which case we do some bookkeeping, otherwise we update | |
3384 | * the end pointer. */ | |
fab2782b | 3385 | if ( !first ) { |
3b6759a6 | 3386 | first = cur; |
df826430 YO |
3387 | if ( noper_trietype == NOTHING ) { |
3388 | #if !defined(DEBUGGING) && !defined(NOJUMPTRIE) | |
3389 | regnode * const noper_next = regnext( noper ); | |
3b6759a6 | 3390 | U8 noper_next_type = (noper_next && noper_next!=tail) ? OP(noper_next) : 0; |
df826430 YO |
3391 | U8 noper_next_trietype = noper_next_type ? TRIE_TYPE( noper_next_type ) :0; |
3392 | #endif | |
3393 | ||
190c1910 | 3394 | if ( noper_next_trietype ) { |
df826430 | 3395 | trietype = noper_next_trietype; |
190c1910 YO |
3396 | } else if (noper_next_type) { |
3397 | /* a NOTHING regop is 1 regop wide. We need at least two | |
3398 | * for a trie so we can't merge this in */ | |
3399 | first = NULL; | |
3400 | } | |
3401 | } else { | |
3402 | trietype = noper_trietype; | |
3b6759a6 | 3403 | } |
8aa23a47 | 3404 | } else { |
fab2782b YO |
3405 | if ( trietype == NOTHING ) |
3406 | trietype = noper_trietype; | |
8aa23a47 YO |
3407 | last = cur; |
3408 | } | |
df826430 YO |
3409 | if (first) |
3410 | count++; | |
fab2782b YO |
3411 | } /* end handle mergable triable node */ |
3412 | else { | |
3413 | /* handle unmergable node - | |
3414 | * noper may either be a triable node which can not be tried | |
3415 | * together with the current trie, or a non triable node */ | |
729aaeb5 YO |
3416 | if ( last ) { |
3417 | /* If last is set and trietype is not NOTHING then we have found | |
3418 | * at least two triable branch sequences in a row of a similar | |
3419 | * trietype so we can turn them into a trie. If/when we | |
3420 | * allow NOTHING to start a trie sequence this condition will be | |
3421 | * required, and it isn't expensive so we leave it in for now. */ | |
3422 | if ( trietype != NOTHING ) | |
3423 | make_trie( pRExC_state, | |
3424 | startbranch, first, cur, tail, count, | |
3425 | trietype, depth+1 ); | |
fab2782b | 3426 | last = NULL; /* note: we clear/update first, trietype etc below, so we dont do it here */ |
8aa23a47 | 3427 | } |
fab2782b | 3428 | if ( noper_trietype |
786e8c11 | 3429 | #ifdef NOJUMPTRIE |
8aa23a47 | 3430 | && noper_next == tail |
786e8c11 | 3431 | #endif |
8aa23a47 | 3432 | ){ |
fab2782b | 3433 | /* noper is triable, so we can start a new trie sequence */ |
8aa23a47 YO |
3434 | count = 1; |
3435 | first = cur; | |
fab2782b YO |
3436 | trietype = noper_trietype; |
3437 | } else if (first) { | |
3438 | /* if we already saw a first but the current node is not triable then we have | |
3439 | * to reset the first information. */ | |
8aa23a47 YO |
3440 | count = 0; |
3441 | first = NULL; | |
fab2782b | 3442 | trietype = 0; |
8aa23a47 | 3443 | } |
fab2782b YO |
3444 | } /* end handle unmergable node */ |
3445 | } /* loop over branches */ | |
df826430 | 3446 | DEBUG_TRIE_COMPILE_r({ |
8aa23a47 YO |
3447 | regprop(RExC_rx, mysv, cur); |
3448 | PerlIO_printf( Perl_debug_log, | |
3449 | "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2, | |
3450 | "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur)); | |
3451 | ||
3452 | }); | |
3b6759a6 YO |
3453 | if ( last ) { |
3454 | if ( trietype != NOTHING ) { | |
3455 | /* the last branch of the sequence was part of a trie, | |
3456 | * so we have to construct it here outside of the loop | |
3457 | */ | |
3458 | made= make_trie( pRExC_state, startbranch, first, scan, tail, count, trietype, depth+1 ); | |
686b73d4 | 3459 | #ifdef TRIE_STUDY_OPT |
3b6759a6 YO |
3460 | if ( ((made == MADE_EXACT_TRIE && |
3461 | startbranch == first) | |
3462 | || ( first_non_open == first )) && | |
3463 | depth==0 ) { | |
3464 | flags |= SCF_TRIE_RESTUDY; | |
3465 | if ( startbranch == first | |
3466 | && scan == tail ) | |
3467 | { | |
3468 | RExC_seen &=~REG_TOP_LEVEL_BRANCHES; | |
3469 | } | |
8aa23a47 | 3470 | } |
3dab1dad | 3471 | #endif |
3b6759a6 YO |
3472 | } else { |
3473 | /* at this point we know whatever we have is a NOTHING sequence/branch | |
3474 | * AND if 'startbranch' is 'first' then we can turn the whole thing into a NOTHING | |
3475 | */ | |
3476 | if ( startbranch == first ) { | |
3477 | regnode *opt; | |
3478 | /* the entire thing is a NOTHING sequence, something like this: | |
3479 | * (?:|) So we can turn it into a plain NOTHING op. */ | |
3480 | DEBUG_TRIE_COMPILE_r({ | |
3481 | regprop(RExC_rx, mysv, cur); | |
3482 | PerlIO_printf( Perl_debug_log, | |
3483 | "%*s- %s (%d) <NOTHING BRANCH SEQUENCE>\n", (int)depth * 2 + 2, | |
3484 | "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur)); | |
3485 | ||
3486 | }); | |
3487 | OP(startbranch)= NOTHING; | |
3488 | NEXT_OFF(startbranch)= tail - startbranch; | |
3489 | for ( opt= startbranch + 1; opt < tail ; opt++ ) | |
3490 | OP(opt)= OPTIMIZED; | |
3491 | } | |
3492 | } | |
fab2782b YO |
3493 | } /* end if ( last) */ |
3494 | } /* TRIE_MAXBUF is non zero */ | |
8aa23a47 YO |
3495 | |
3496 | } /* do trie */ | |
3497 | ||
653099ff | 3498 | } |
8aa23a47 YO |
3499 | else if ( code == BRANCHJ ) { /* single branch is optimized. */ |
3500 | scan = NEXTOPER(NEXTOPER(scan)); | |
3501 | } else /* single branch is optimized. */ | |
3502 | scan = NEXTOPER(scan); | |
3503 | continue; | |
3504 | } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) { | |
3505 | scan_frame *newframe = NULL; | |
3506 | I32 paren; | |
3507 | regnode *start; | |
3508 | regnode *end; | |
3509 | ||
3510 | if (OP(scan) != SUSPEND) { | |
3511 | /* set the pointer */ | |
3512 | if (OP(scan) == GOSUB) { | |
3513 | paren = ARG(scan); | |
3514 | RExC_recurse[ARG2L(scan)] = scan; | |
3515 | start = RExC_open_parens[paren-1]; | |
3516 | end = RExC_close_parens[paren-1]; | |
3517 | } else { | |
3518 | paren = 0; | |
f8fc2ecf | 3519 | start = RExC_rxi->program + 1; |
8aa23a47 YO |
3520 | end = RExC_opend; |
3521 | } | |
3522 | if (!recursed) { | |
3523 | Newxz(recursed, (((RExC_npar)>>3) +1), U8); | |
3524 | SAVEFREEPV(recursed); | |
3525 | } | |
3526 | if (!PAREN_TEST(recursed,paren+1)) { | |
3527 | PAREN_SET(recursed,paren+1); | |
3528 | Newx(newframe,1,scan_frame); | |
3529 | } else { | |
3530 | if (flags & SCF_DO_SUBSTR) { | |
304ee84b | 3531 | SCAN_COMMIT(pRExC_state,data,minlenp); |
8aa23a47 YO |
3532 | data->longest = &(data->longest_float); |
3533 | } | |
3534 | is_inf = is_inf_internal = 1; | |
3535 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ | |
3fffb88a | 3536 | cl_anything(pRExC_state, data->start_class); |
8aa23a47 YO |
3537 | flags &= ~SCF_DO_STCLASS; |
3538 | } | |
3539 | } else { | |
3540 | Newx(newframe,1,scan_frame); | |
3541 | paren = stopparen; | |
3542 | start = scan+2; | |
3543 | end = regnext(scan); | |
3544 | } | |
3545 | if (newframe) { | |
3546 | assert(start); | |
3547 | assert(end); | |
3548 | SAVEFREEPV(newframe); | |
3549 | newframe->next = regnext(scan); | |
3550 | newframe->last = last; | |
3551 | newframe->stop = stopparen; | |
3552 | newframe->prev = frame; | |
3553 | ||
3554 | frame = newframe; | |
3555 | scan = start; | |
3556 | stopparen = paren; | |
3557 | last = end; | |
3558 | ||
3559 | continue; | |
3560 | } | |
3561 | } | |
3562 | else if (OP(scan) == EXACT) { | |
3563 | I32 l = STR_LEN(scan); | |
3564 | UV uc; | |
3565 | if (UTF) { | |
3566 | const U8 * const s = (U8*)STRING(scan); | |
4b88fb76 | 3567 | uc = utf8_to_uvchr_buf(s, s + l, NULL); |
8aa23a47 | 3568 | l = utf8_length(s, s + l); |
8aa23a47 YO |
3569 | } else { |
3570 | uc = *((U8*)STRING(scan)); | |
3571 | } | |
3572 | min += l; | |
3573 | if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */ | |
3574 | /* The code below prefers earlier match for fixed | |
3575 | offset, later match for variable offset. */ | |
3576 | if (data->last_end == -1) { /* Update the start info. */ | |
3577 | data->last_start_min = data->pos_min; | |
3578 | data->last_start_max = is_inf | |
3579 | ? I32_MAX : data->pos_min + data->pos_delta; | |
b515a41d | 3580 | } |
8aa23a47 YO |
3581 | sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan)); |
3582 | if (UTF) | |
3583 | SvUTF8_on(data->last_found); | |
3584 | { | |
3585 | SV * const sv = data->last_found; | |
3586 | MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ? | |
3587 | mg_find(sv, PERL_MAGIC_utf8) : NULL; | |
3588 | if (mg && mg->mg_len >= 0) | |
3589 | mg->mg_len += utf8_length((U8*)STRING(scan), | |
3590 | (U8*)STRING(scan)+STR_LEN(scan)); | |
b515a41d | 3591 | } |
8aa23a47 YO |
3592 | data->last_end = data->pos_min + l; |
3593 | data->pos_min += l; /* As in the first entry. */ | |
3594 | data->flags &= ~SF_BEFORE_EOL; | |
3595 | } | |
3596 | if (flags & SCF_DO_STCLASS_AND) { | |
3597 | /* Check whether it is compatible with what we know already! */ | |
3598 | int compat = 1; | |
3599 | ||
54251c2e | 3600 | |
486ec47a | 3601 | /* If compatible, we or it in below. It is compatible if is |
54251c2e KW |
3602 | * in the bitmp and either 1) its bit or its fold is set, or 2) |
3603 | * it's for a locale. Even if there isn't unicode semantics | |
3604 | * here, at runtime there may be because of matching against a | |
3605 | * utf8 string, so accept a possible false positive for | |
3606 | * latin1-range folds */ | |
8aa23a47 YO |
3607 | if (uc >= 0x100 || |
3608 | (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE)) | |
3609 | && !ANYOF_BITMAP_TEST(data->start_class, uc) | |
39065660 | 3610 | && (!(data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) |
54251c2e | 3611 | || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc]))) |
8aa23a47 | 3612 | ) |
d18bf9dc | 3613 | { |
8aa23a47 | 3614 | compat = 0; |
d18bf9dc | 3615 | } |
8aa23a47 YO |
3616 | ANYOF_CLASS_ZERO(data->start_class); |
3617 | ANYOF_BITMAP_ZERO(data->start_class); | |
3618 | if (compat) | |
3619 | ANYOF_BITMAP_SET(data->start_class, uc); | |
d18bf9dc KW |
3620 | else if (uc >= 0x100) { |
3621 | int i; | |
3622 | ||
3623 | /* Some Unicode code points fold to the Latin1 range; as | |
3624 | * XXX temporary code, instead of figuring out if this is | |
3625 | * one, just assume it is and set all the start class bits | |
3626 | * that could be some such above 255 code point's fold | |
3627 | * which will generate fals positives. As the code | |
3628 | * elsewhere that does compute the fold settles down, it | |
3629 | * can be extracted out and re-used here */ | |
3630 | for (i = 0; i < 256; i++){ | |
94dc5c2d | 3631 | if (HAS_NONLATIN1_FOLD_CLOSURE(i)) { |
d18bf9dc KW |
3632 | ANYOF_BITMAP_SET(data->start_class, i); |
3633 | } | |
3634 | } | |
3635 | } | |
8aa23a47 YO |
3636 | data->start_class->flags &= ~ANYOF_EOS; |
3637 | if (uc < 0x100) | |
3638 | data->start_class->flags &= ~ANYOF_UNICODE_ALL; | |
3639 | } | |
3640 | else if (flags & SCF_DO_STCLASS_OR) { | |
3641 | /* false positive possible if the class is case-folded */ | |
3642 | if (uc < 0x100) | |
3643 | ANYOF_BITMAP_SET(data->start_class, uc); | |
3644 | else | |
3645 | data->start_class->flags |= ANYOF_UNICODE_ALL; | |
3646 | data->start_class->flags &= ~ANYOF_EOS; | |
3647 | cl_and(data->start_class, and_withp); | |
3648 | } | |
3649 | flags &= ~SCF_DO_STCLASS; | |
3650 | } | |
3651 | else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */ | |
3652 | I32 l = STR_LEN(scan); | |
3653 | UV uc = *((U8*)STRING(scan)); | |
3654 | ||
3655 | /* Search for fixed substrings supports EXACT only. */ | |
3656 | if (flags & SCF_DO_SUBSTR) { | |
3657 | assert(data); | |
304ee84b | 3658 | SCAN_COMMIT(pRExC_state, data, minlenp); |
8aa23a47 YO |
3659 | } |
3660 | if (UTF) { | |
3661 | const U8 * const s = (U8 *)STRING(scan); | |
4b88fb76 | 3662 | uc = utf8_to_uvchr_buf(s, s + l, NULL); |
8aa23a47 | 3663 | l = utf8_length(s, s + l); |
8aa23a47 | 3664 | } |
7dff6b46 | 3665 | if (has_exactf_sharp_s) { |
f758bddf | 3666 | RExC_seen |= REG_SEEN_EXACTF_SHARP_S; |
bb914485 | 3667 | } |
9d071ca8 | 3668 | min += l - min_subtract; |
f646642f KW |
3669 | if (min < 0) { |
3670 | min = 0; | |
3671 | } | |
9d071ca8 | 3672 | delta += min_subtract; |
f646642f | 3673 | if (flags & SCF_DO_SUBSTR) { |
9d071ca8 | 3674 | data->pos_min += l - min_subtract; |
f646642f KW |
3675 | if (data->pos_min < 0) { |
3676 | data->pos_min = 0; | |
3677 | } | |
9d071ca8 KW |
3678 | data->pos_delta += min_subtract; |
3679 | if (min_subtract) { | |
d2197104 KW |
3680 | data->longest = &(data->longest_float); |
3681 | } | |
f646642f | 3682 | } |
8aa23a47 YO |
3683 | if (flags & SCF_DO_STCLASS_AND) { |
3684 | /* Check whether it is compatible with what we know already! */ | |
3685 | int compat = 1; | |
8aa23a47 | 3686 | if (uc >= 0x100 || |
54251c2e KW |
3687 | (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE)) |
3688 | && !ANYOF_BITMAP_TEST(data->start_class, uc) | |
3689 | && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc]))) | |
3690 | { | |
8aa23a47 | 3691 | compat = 0; |
54251c2e | 3692 | } |
8aa23a47 YO |
3693 | ANYOF_CLASS_ZERO(data->start_class); |
3694 | ANYOF_BITMAP_ZERO(data->start_class); | |
3695 | if (compat) { | |
3696 | ANYOF_BITMAP_SET(data->start_class, uc); | |
653099ff | 3697 | data->start_class->flags &= ~ANYOF_EOS; |
39065660 | 3698 | data->start_class->flags |= ANYOF_LOC_NONBITMAP_FOLD; |
970c8436 | 3699 | if (OP(scan) == EXACTFL) { |
af302e7f KW |
3700 | /* XXX This set is probably no longer necessary, and |
3701 | * probably wrong as LOCALE now is on in the initial | |
3702 | * state */ | |
8aa23a47 | 3703 | data->start_class->flags |= ANYOF_LOCALE; |
970c8436 KW |
3704 | } |
3705 | else { | |
3706 | ||
54251c2e KW |
3707 | /* Also set the other member of the fold pair. In case |
3708 | * that unicode semantics is called for at runtime, use | |
3709 | * the full latin1 fold. (Can't do this for locale, | |
a0c4c608 | 3710 | * because not known until runtime) */ |
54251c2e | 3711 | ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]); |
e22b340a | 3712 | |
a0c4c608 KW |
3713 | /* All other (EXACTFL handled above) folds except under |
3714 | * /iaa that include s, S, and sharp_s also may include | |
3715 | * the others */ | |
e22b340a KW |
3716 | if (OP(scan) != EXACTFA) { |
3717 | if (uc == 's' || uc == 'S') { | |
3718 | ANYOF_BITMAP_SET(data->start_class, | |
3719 | LATIN_SMALL_LETTER_SHARP_S); | |
3720 | } | |
3721 | else if (uc == LATIN_SMALL_LETTER_SHARP_S) { | |
3722 | ANYOF_BITMAP_SET(data->start_class, 's'); | |
3723 | ANYOF_BITMAP_SET(data->start_class, 'S'); | |
3724 | } | |
3725 | } | |
970c8436 | 3726 | } |
653099ff | 3727 | } |
d18bf9dc KW |
3728 | else if (uc >= 0x100) { |
3729 | int i; | |
3730 | for (i = 0; i < 256; i++){ | |
3731 | if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) { | |
3732 | ANYOF_BITMAP_SET(data->start_class, i); | |
3733 | } | |
3734 | } | |
3735 | } | |
8aa23a47 YO |
3736 | } |
3737 | else if (flags & SCF_DO_STCLASS_OR) { | |
39065660 | 3738 | if (data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) { |
8aa23a47 YO |
3739 | /* false positive possible if the class is case-folded. |
3740 | Assume that the locale settings are the same... */ | |
970c8436 | 3741 | if (uc < 0x100) { |
1aa99e6b | 3742 | ANYOF_BITMAP_SET(data->start_class, uc); |
970c8436 KW |
3743 | if (OP(scan) != EXACTFL) { |
3744 | ||
3745 | /* And set the other member of the fold pair, but | |
3746 | * can't do that in locale because not known until | |
3747 | * run-time */ | |
3748 | ANYOF_BITMAP_SET(data->start_class, | |
54251c2e | 3749 | PL_fold_latin1[uc]); |
e22b340a KW |
3750 | |
3751 | /* All folds except under /iaa that include s, S, | |
3752 | * and sharp_s also may include the others */ | |
3753 | if (OP(scan) != EXACTFA) { | |
3754 | if (uc == 's' || uc == 'S') { | |
3755 | ANYOF_BITMAP_SET(data->start_class, | |
3756 | LATIN_SMALL_LETTER_SHARP_S); | |
3757 | } | |
3758 | else if (uc == LATIN_SMALL_LETTER_SHARP_S) { | |
3759 | ANYOF_BITMAP_SET(data->start_class, 's'); | |
3760 | ANYOF_BITMAP_SET(data->start_class, 'S'); | |
3761 | } | |
3762 | } | |
970c8436 KW |
3763 | } |
3764 | } | |
653099ff GS |
3765 | data->start_class->flags &= ~ANYOF_EOS; |
3766 | } | |
8aa23a47 | 3767 | cl_and(data->start_class, and_withp); |
653099ff | 3768 | } |
8aa23a47 YO |
3769 | flags &= ~SCF_DO_STCLASS; |
3770 | } | |
e52fc539 | 3771 | else if (REGNODE_VARIES(OP(scan))) { |
8aa23a47 YO |
3772 | I32 mincount, maxcount, minnext, deltanext, fl = 0; |
3773 | I32 f = flags, pos_before = 0; | |
3774 | regnode * const oscan = scan; | |
3775 | struct regnode_charclass_class this_class; | |
3776 | struct regnode_charclass_class *oclass = NULL; | |
3777 | I32 next_is_eval = 0; | |
3778 | ||
3779 | switch (PL_regkind[OP(scan)]) { | |
3780 | case WHILEM: /* End of (?:...)* . */ | |
3781 | scan = NEXTOPER(scan); | |
3782 | goto finish; | |
3783 | case PLUS: | |
3784 | if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) { | |
3785 | next = NEXTOPER(scan); | |
3786 | if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) { | |
3787 | mincount = 1; | |
3788 | maxcount = REG_INFTY; | |
3789 | next = regnext(scan); | |
3790 | scan = NEXTOPER(scan); | |
3791 | goto do_curly; | |
3792 | } | |
3793 | } | |
3794 | if (flags & SCF_DO_SUBSTR) | |
3795 | data->pos_min++; | |
3796 | min++; | |
3797 | /* Fall through. */ | |
3798 | case STAR: | |
3799 | if (flags & SCF_DO_STCLASS) { | |
3800 | mincount = 0; | |
3801 | maxcount = REG_INFTY; | |
3802 | next = regnext(scan); | |
3803 | scan = NEXTOPER(scan); | |
3804 | goto do_curly; | |
3805 | } | |
3806 | is_inf = is_inf_internal = 1; | |
3807 | scan = regnext(scan); | |
c277df42 | 3808 | if (flags & SCF_DO_SUBSTR) { |
304ee84b | 3809 | SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */ |
8aa23a47 | 3810 | data->longest = &(data->longest_float); |
c277df42 | 3811 | } |
8aa23a47 YO |
3812 | goto optimize_curly_tail; |
3813 | case CURLY: | |
3814 | if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM) | |
3815 | && (scan->flags == stopparen)) | |
3816 | { | |
3817 | mincount = 1; | |
3818 | maxcount = 1; | |
3819 | } else { | |
3820 | mincount = ARG1(scan); | |
3821 | maxcount = ARG2(scan); | |
653099ff | 3822 | } |
8aa23a47 YO |
3823 | next = regnext(scan); |
3824 | if (OP(scan) == CURLYX) { | |
3825 | I32 lp = (data ? *(data->last_closep) : 0); | |
3826 | scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX); | |
653099ff | 3827 | } |
8aa23a47 YO |
3828 | scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS; |
3829 | next_is_eval = (OP(scan) == EVAL); | |
3830 | do_curly: | |
3831 | if (flags & SCF_DO_SUBSTR) { | |
304ee84b | 3832 | if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */ |
8aa23a47 | 3833 | pos_before = data->pos_min; |
b45f050a | 3834 | } |
8aa23a47 YO |
3835 | if (data) { |
3836 | fl = data->flags; | |
3837 | data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL); | |
3838 | if (is_inf) | |
3839 | data->flags |= SF_IS_INF; | |
3840 | } | |
3841 | if (flags & SCF_DO_STCLASS) { | |
e755fd73 | 3842 | cl_init(pRExC_state, &this_class); |
8aa23a47 YO |
3843 | oclass = data->start_class; |
3844 | data->start_class = &this_class; | |
3845 | f |= SCF_DO_STCLASS_AND; | |
3846 | f &= ~SCF_DO_STCLASS_OR; | |
3847 | } | |
779bcb7d NC |
3848 | /* Exclude from super-linear cache processing any {n,m} |
3849 | regops for which the combination of input pos and regex | |
3850 | pos is not enough information to determine if a match | |
3851 | will be possible. | |
3852 | ||
3853 | For example, in the regex /foo(bar\s*){4,8}baz/ with the | |
3854 | regex pos at the \s*, the prospects for a match depend not | |
3855 | only on the input position but also on how many (bar\s*) | |
3856 | repeats into the {4,8} we are. */ | |
3857 | if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY)) | |
8aa23a47 | 3858 | f &= ~SCF_WHILEM_VISITED_POS; |
b45f050a | 3859 | |
8aa23a47 YO |
3860 | /* This will finish on WHILEM, setting scan, or on NULL: */ |
3861 | minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, | |
3862 | last, data, stopparen, recursed, NULL, | |
3863 | (mincount == 0 | |
3864 | ? (f & ~SCF_DO_SUBSTR) : f),depth+1); | |
b515a41d | 3865 | |
8aa23a47 YO |
3866 | if (flags & SCF_DO_STCLASS) |
3867 | data->start_class = oclass; | |
3868 | if (mincount == 0 || minnext == 0) { | |
3869 | if (flags & SCF_DO_STCLASS_OR) { | |
3fffb88a | 3870 | cl_or(pRExC_state, data->start_class, &this_class); |
8aa23a47 YO |
3871 | } |
3872 | else if (flags & SCF_DO_STCLASS_AND) { | |
3873 | /* Switch to OR mode: cache the old value of | |
3874 | * data->start_class */ | |
3875 | INIT_AND_WITHP; | |
3876 | StructCopy(data->start_class, and_withp, | |
3877 | struct regnode_charclass_class); | |
3878 | flags &= ~SCF_DO_STCLASS_AND; | |
3879 | StructCopy(&this_class, data->start_class, | |
3880 | struct regnode_charclass_class); | |
3881 | flags |= SCF_DO_STCLASS_OR; | |
3882 | data->start_class->flags |= ANYOF_EOS; | |
3883 | } | |
3884 | } else { /* Non-zero len */ | |
3885 | if (flags & SCF_DO_STCLASS_OR) { | |
3fffb88a | 3886 | cl_or(pRExC_state, data->start_class, &this_class); |
8aa23a47 YO |
3887 | cl_and(data->start_class, and_withp); |
3888 | } | |
3889 | else if (flags & SCF_DO_STCLASS_AND) | |
3890 | cl_and(data->start_class, &this_class); | |
3891 | flags &= ~SCF_DO_STCLASS; | |
3892 | } | |
3893 | if (!scan) /* It was not CURLYX, but CURLY. */ | |
3894 | scan = next; | |
3895 | if ( /* ? quantifier ok, except for (?{ ... }) */ | |
3896 | (next_is_eval || !(mincount == 0 && maxcount == 1)) | |
3897 | && (minnext == 0) && (deltanext == 0) | |
3898 | && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR)) | |
668c081a | 3899 | && maxcount <= REG_INFTY/3) /* Complement check for big count */ |
8aa23a47 | 3900 | { |
668c081a NC |
3901 | ckWARNreg(RExC_parse, |
3902 | "Quantifier unexpected on zero-length expression"); | |
8aa23a47 YO |
3903 | } |
3904 | ||
3905 | min += minnext * mincount; | |
3906 | is_inf_internal |= ((maxcount == REG_INFTY | |
3907 | && (minnext + deltanext) > 0) | |
3908 | || deltanext == I32_MAX); | |
3909 | is_inf |= is_inf_internal; | |
3910 | delta += (minnext + deltanext) * maxcount - minnext * mincount; | |
3911 | ||
3912 | /* Try powerful optimization CURLYX => CURLYN. */ | |
3913 | if ( OP(oscan) == CURLYX && data | |
3914 | && data->flags & SF_IN_PAR | |
3915 | && !(data->flags & SF_HAS_EVAL) | |
3916 | && !deltanext && minnext == 1 ) { | |
3917 | /* Try to optimize to CURLYN. */ | |
3918 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; | |
3919 | regnode * const nxt1 = nxt; | |
497b47a8 | 3920 | #ifdef DEBUGGING |
8aa23a47 | 3921 | regnode *nxt2; |
497b47a8 | 3922 | #endif |
c277df42 | 3923 | |
8aa23a47 YO |
3924 | /* Skip open. */ |
3925 | nxt = regnext(nxt); | |
e52fc539 | 3926 | if (!REGNODE_SIMPLE(OP(nxt)) |
8aa23a47 YO |
3927 | && !(PL_regkind[OP(nxt)] == EXACT |
3928 | && STR_LEN(nxt) == 1)) | |
3929 | goto nogo; | |
497b47a8 | 3930 | #ifdef DEBUGGING |
8aa23a47 | 3931 | nxt2 = nxt; |
497b47a8 | 3932 | #endif |
8aa23a47 YO |
3933 | nxt = regnext(nxt); |
3934 | if (OP(nxt) != CLOSE) | |
3935 | goto nogo; | |
3936 | if (RExC_open_parens) { | |
3937 | RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/ | |
3938 | RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/ | |
3939 | } | |
3940 | /* Now we know that nxt2 is the only contents: */ | |
3941 | oscan->flags = (U8)ARG(nxt); | |
3942 | OP(oscan) = CURLYN; | |
3943 | OP(nxt1) = NOTHING; /* was OPEN. */ | |
40d049e4 | 3944 | |
c277df42 | 3945 | #ifdef DEBUGGING |
8aa23a47 | 3946 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
fda99bee KW |
3947 | NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */ |
3948 | NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */ | |
8aa23a47 YO |
3949 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ |
3950 | OP(nxt + 1) = OPTIMIZED; /* was count. */ | |
fda99bee | 3951 | NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */ |
b81d288d | 3952 | #endif |
8aa23a47 YO |
3953 | } |
3954 | nogo: | |
3955 | ||
3956 | /* Try optimization CURLYX => CURLYM. */ | |
3957 | if ( OP(oscan) == CURLYX && data | |
3958 | && !(data->flags & SF_HAS_PAR) | |
3959 | && !(data->flags & SF_HAS_EVAL) | |
3960 | && !deltanext /* atom is fixed width */ | |
3961 | && minnext != 0 /* CURLYM can't handle zero width */ | |
7dff6b46 | 3962 | && ! (RExC_seen & REG_SEEN_EXACTF_SHARP_S) /* Nor \xDF */ |
8aa23a47 YO |
3963 | ) { |
3964 | /* XXXX How to optimize if data == 0? */ | |
3965 | /* Optimize to a simpler form. */ | |
3966 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */ | |
3967 | regnode *nxt2; | |
3968 | ||
3969 | OP(oscan) = CURLYM; | |
3970 | while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/ | |
3971 | && (OP(nxt2) != WHILEM)) | |
3972 | nxt = nxt2; | |
3973 | OP(nxt2) = SUCCEED; /* Whas WHILEM */ | |
3974 | /* Need to optimize away parenths. */ | |
b3c0965f | 3975 | if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) { |
8aa23a47 YO |
3976 | /* Set the parenth number. */ |
3977 | regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/ | |
3978 | ||
8aa23a47 YO |
3979 | oscan->flags = (U8)ARG(nxt); |
3980 | if (RExC_open_parens) { | |
3981 | RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/ | |
3982 | RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/ | |
40d049e4 | 3983 | } |
8aa23a47 YO |
3984 | OP(nxt1) = OPTIMIZED; /* was OPEN. */ |
3985 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ | |
40d049e4 | 3986 | |
c277df42 | 3987 | #ifdef DEBUGGING |
8aa23a47 YO |
3988 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
3989 | OP(nxt + 1) = OPTIMIZED; /* was count. */ | |
486ec47a PA |
3990 | NEXT_OFF(nxt1 + 1) = 0; /* just for consistency. */ |
3991 | NEXT_OFF(nxt + 1) = 0; /* just for consistency. */ | |
b81d288d | 3992 | #endif |
c277df42 | 3993 | #if 0 |
8aa23a47 YO |
3994 | while ( nxt1 && (OP(nxt1) != WHILEM)) { |
3995 | regnode *nnxt = regnext(nxt1); | |
8aa23a47 YO |
3996 | if (nnxt == nxt) { |
3997 | if (reg_off_by_arg[OP(nxt1)]) | |
3998 | ARG_SET(nxt1, nxt2 - nxt1); | |
3999 | else if (nxt2 - nxt1 < U16_MAX) | |
4000 | NEXT_OFF(nxt1) = nxt2 - nxt1; | |
4001 | else | |
4002 | OP(nxt) = NOTHING; /* Cannot beautify */ | |
c277df42 | 4003 | } |
8aa23a47 | 4004 | nxt1 = nnxt; |
c277df42 | 4005 | } |
5d1c421c | 4006 | #endif |
8aa23a47 YO |
4007 | /* Optimize again: */ |
4008 | study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt, | |
4009 | NULL, stopparen, recursed, NULL, 0,depth+1); | |
4010 | } | |
4011 | else | |
4012 | oscan->flags = 0; | |
4013 | } | |
4014 | else if ((OP(oscan) == CURLYX) | |
4015 | && (flags & SCF_WHILEM_VISITED_POS) | |
4016 | /* See the comment on a similar expression above. | |
3b753521 | 4017 | However, this time it's not a subexpression |
8aa23a47 YO |
4018 | we care about, but the expression itself. */ |
4019 | && (maxcount == REG_INFTY) | |
4020 | && data && ++data->whilem_c < 16) { | |
4021 | /* This stays as CURLYX, we can put the count/of pair. */ | |
4022 | /* Find WHILEM (as in regexec.c) */ | |
4023 | regnode *nxt = oscan + NEXT_OFF(oscan); | |
4024 | ||
4025 | if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */ | |
4026 | nxt += ARG(nxt); | |
4027 | PREVOPER(nxt)->flags = (U8)(data->whilem_c | |
4028 | | (RExC_whilem_seen << 4)); /* On WHILEM */ | |
4029 | } | |
4030 | if (data && fl & (SF_HAS_PAR|SF_IN_PAR)) | |
4031 | pars++; | |
4032 | if (flags & SCF_DO_SUBSTR) { | |
4033 | SV *last_str = NULL; | |
4034 | int counted = mincount != 0; | |
a0ed51b3 | 4035 | |
8aa23a47 YO |
4036 | if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */ |
4037 | #if defined(SPARC64_GCC_WORKAROUND) | |
4038 | I32 b = 0; | |
4039 | STRLEN l = 0; | |
4040 | const char *s = NULL; | |
4041 | I32 old = 0; | |
b515a41d | 4042 | |
8aa23a47 YO |
4043 | if (pos_before >= data->last_start_min) |
4044 | b = pos_before; | |
4045 | else | |
4046 | b = data->last_start_min; | |
b515a41d | 4047 | |
8aa23a47 YO |
4048 | l = 0; |
4049 | s = SvPV_const(data->last_found, l); | |
4050 | old = b - data->last_start_min; | |
4051 | ||
4052 | #else | |
4053 | I32 b = pos_before >= data->last_start_min | |
4054 | ? pos_before : data->last_start_min; | |
4055 | STRLEN l; | |
4056 | const char * const s = SvPV_const(data->last_found, l); | |
4057 | I32 old = b - data->last_start_min; | |
4058 | #endif | |
4059 | ||
4060 | if (UTF) | |
4061 | old = utf8_hop((U8*)s, old) - (U8*)s; | |
8aa23a47 YO |
4062 | l -= old; |
4063 | /* Get the added string: */ | |
740cce10 | 4064 | last_str = newSVpvn_utf8(s + old, l, UTF); |
8aa23a47 YO |
4065 | if (deltanext == 0 && pos_before == b) { |
4066 | /* What was added is a constant string */ | |
4067 | if (mincount > 1) { | |
4068 | SvGROW(last_str, (mincount * l) + 1); | |
4069 | repeatcpy(SvPVX(last_str) + l, | |
4070 | SvPVX_const(last_str), l, mincount - 1); | |
4071 | SvCUR_set(last_str, SvCUR(last_str) * mincount); | |
4072 | /* Add additional parts. */ | |
4073 | SvCUR_set(data->last_found, | |
4074 | SvCUR(data->last_found) - l); | |
4075 | sv_catsv(data->last_found, last_str); | |
4076 | { | |
4077 | SV * sv = data->last_found; | |
4078 | MAGIC *mg = | |
4079 | SvUTF8(sv) && SvMAGICAL(sv) ? | |
4080 | mg_find(sv, PERL_MAGIC_utf8) : NULL; | |
4081 | if (mg && mg->mg_len >= 0) | |
bd94e887 | 4082 | mg->mg_len += CHR_SVLEN(last_str) - l; |
b515a41d | 4083 | } |
8aa23a47 | 4084 | data->last_end += l * (mincount - 1); |
b515a41d | 4085 | } |
8aa23a47 YO |
4086 | } else { |
4087 | /* start offset must point into the last copy */ | |
4088 | data->last_start_min += minnext * (mincount - 1); | |
4089 | data->last_start_max += is_inf ? I32_MAX | |
4090 | : (maxcount - 1) * (minnext + data->pos_delta); | |
4091 | } | |
c277df42 | 4092 | } |
8aa23a47 YO |
4093 | /* It is counted once already... */ |
4094 | data->pos_min += minnext * (mincount - counted); | |
4095 | data->pos_delta += - counted * deltanext + | |
4096 | (minnext + deltanext) * maxcount - minnext * mincount; | |
4097 | if (mincount != maxcount) { | |
4098 | /* Cannot extend fixed substrings found inside | |
4099 | the group. */ | |
304ee84b | 4100 | SCAN_COMMIT(pRExC_state,data,minlenp); |
8aa23a47 YO |
4101 | if (mincount && last_str) { |
4102 | SV * const sv = data->last_found; | |
4103 | MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ? | |
4104 | mg_find(sv, PERL_MAGIC_utf8) : NULL; | |
4105 | ||
4106 | if (mg) | |
4107 | mg->mg_len = -1; | |
4108 | sv_setsv(sv, last_str); | |
4109 | data->last_end = data->pos_min; | |
4110 | data->last_start_min = | |
4111 | data->pos_min - CHR_SVLEN(last_str); | |
4112 | data->last_start_max = is_inf | |
4113 | ? I32_MAX | |
4114 | : data->pos_min + data->pos_delta | |
4115 | - CHR_SVLEN(last_str); | |
4116 | } | |
4117 | data->longest = &(data->longest_float); | |
4118 | } | |
4119 | SvREFCNT_dec(last_str); | |
c277df42 | 4120 | } |
8aa23a47 YO |
4121 | if (data && (fl & SF_HAS_EVAL)) |
4122 | data->flags |= SF_HAS_EVAL; | |
4123 | optimize_curly_tail: | |
4124 | if (OP(oscan) != CURLYX) { | |
4125 | while (PL_regkind[OP(next = regnext(oscan))] == NOTHING | |
4126 | && NEXT_OFF(next)) | |
4127 | NEXT_OFF(oscan) += NEXT_OFF(next); | |
4128 | } | |
4129 | continue; | |
f56b6394 | 4130 | default: /* REF, ANYOFV, and CLUMP only? */ |
8aa23a47 | 4131 | if (flags & SCF_DO_SUBSTR) { |
304ee84b | 4132 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ |
8aa23a47 YO |
4133 | data->longest = &(data->longest_float); |
4134 | } | |
4135 | is_inf = is_inf_internal = 1; | |
4136 | if (flags & SCF_DO_STCLASS_OR) | |
3fffb88a | 4137 | cl_anything(pRExC_state, data->start_class); |
8aa23a47 YO |
4138 | flags &= ~SCF_DO_STCLASS; |
4139 | break; | |
c277df42 | 4140 | } |
8aa23a47 | 4141 | } |
e1d1eefb YO |
4142 | else if (OP(scan) == LNBREAK) { |
4143 | if (flags & SCF_DO_STCLASS) { | |
4144 | int value = 0; | |
4145 | data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */ | |
4146 | if (flags & SCF_DO_STCLASS_AND) { | |
4147 | for (value = 0; value < 256; value++) | |
e64b1bd1 | 4148 | if (!is_VERTWS_cp(value)) |
b9a59e08 KW |
4149 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
4150 | } | |
4151 | else { | |
e1d1eefb | 4152 | for (value = 0; value < 256; value++) |
e64b1bd1 | 4153 | if (is_VERTWS_cp(value)) |
b9a59e08 KW |
4154 | ANYOF_BITMAP_SET(data->start_class, value); |
4155 | } | |
e1d1eefb YO |
4156 | if (flags & SCF_DO_STCLASS_OR) |
4157 | cl_and(data->start_class, and_withp); | |
4158 | flags &= ~SCF_DO_STCLASS; | |
4159 | } | |
4160 | min += 1; | |
f9a79580 | 4161 | delta += 1; |
e1d1eefb YO |
4162 | if (flags & SCF_DO_SUBSTR) { |
4163 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ | |
4164 | data->pos_min += 1; | |
f9a79580 | 4165 | data->pos_delta += 1; |
e1d1eefb YO |
4166 | data->longest = &(data->longest_float); |
4167 | } | |
e1d1eefb | 4168 | } |
e52fc539 | 4169 | else if (REGNODE_SIMPLE(OP(scan))) { |
8aa23a47 | 4170 | int value = 0; |
653099ff | 4171 | |
8aa23a47 | 4172 | if (flags & SCF_DO_SUBSTR) { |
304ee84b | 4173 | SCAN_COMMIT(pRExC_state,data,minlenp); |
8aa23a47 YO |
4174 | data->pos_min++; |
4175 | } | |
4176 | min++; | |
4177 | if (flags & SCF_DO_STCLASS) { | |
4178 | data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */ | |
b515a41d | 4179 | |
8aa23a47 YO |
4180 | /* Some of the logic below assumes that switching |
4181 | locale on will only add false positives. */ | |
4182 | switch (PL_regkind[OP(scan)]) { | |
4183 | case SANY: | |
4184 | default: | |
4185 | do_default: | |
4186 | /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */ | |
4187 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ | |
3fffb88a | 4188 | cl_anything(pRExC_state, data->start_class); |
8aa23a47 YO |
4189 | break; |
4190 | case REG_ANY: | |
4191 | if (OP(scan) == SANY) | |
4192 | goto do_default; | |
4193 | if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */ | |
4194 | value = (ANYOF_BITMAP_TEST(data->start_class,'\n') | |
3a15e693 | 4195 | || ANYOF_CLASS_TEST_ANY_SET(data->start_class)); |
3fffb88a | 4196 | cl_anything(pRExC_state, data->start_class); |
653099ff | 4197 | } |
8aa23a47 YO |
4198 | if (flags & SCF_DO_STCLASS_AND || !value) |
4199 | ANYOF_BITMAP_CLEAR(data->start_class,'\n'); | |
4200 | break; | |
4201 | case ANYOF: | |
4202 | if (flags & SCF_DO_STCLASS_AND) | |
4203 | cl_and(data->start_class, | |
4204 | (struct regnode_charclass_class*)scan); | |
653099ff | 4205 | else |
3fffb88a | 4206 | cl_or(pRExC_state, data->start_class, |
8aa23a47 YO |
4207 | (struct regnode_charclass_class*)scan); |
4208 | break; | |
4209 | case ALNUM: | |
4210 | if (flags & SCF_DO_STCLASS_AND) { | |
4211 | if (!(data->start_class->flags & ANYOF_LOCALE)) { | |
4212 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM); | |
980866de | 4213 | if (OP(scan) == ALNUMU) { |
a12cf05f KW |
4214 | for (value = 0; value < 256; value++) { |
4215 | if (!isWORDCHAR_L1(value)) { | |
4216 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
4217 | } | |
4218 | } | |
4219 | } else { | |
4220 | for (value = 0; value < 256; value++) { | |
4221 | if (!isALNUM(value)) { | |
4222 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
4223 | } | |
4224 | } | |
4225 | } | |
8aa23a47 | 4226 | } |
653099ff | 4227 | } |
8aa23a47 YO |
4228 | else { |
4229 | if (data->start_class->flags & ANYOF_LOCALE) | |
4230 | ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM); | |
af302e7f KW |
4231 | |
4232 | /* Even if under locale, set the bits for non-locale | |
4233 | * in case it isn't a true locale-node. This will | |
4234 | * create false positives if it truly is locale */ | |
4235 | if (OP(scan) == ALNUMU) { | |
a12cf05f KW |
4236 | for (value = 0; value < 256; value++) { |
4237 | if (isWORDCHAR_L1(value)) { | |
4238 | ANYOF_BITMAP_SET(data->start_class, value); | |
4239 | } | |
4240 | } | |
4241 | } else { | |
4242 | for (value = 0; value < 256; value++) { | |
4243 | if (isALNUM(value)) { | |
4244 | ANYOF_BITMAP_SET(data->start_class, value); | |
4245 | } | |
4246 | } | |
4247 | } | |
8aa23a47 YO |
4248 | } |
4249 | break; | |
8aa23a47 YO |
4250 | case NALNUM: |
4251 | if (flags & SCF_DO_STCLASS_AND) { | |
4252 | if (!(data->start_class->flags & ANYOF_LOCALE)) { | |
4253 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM); | |
980866de | 4254 | if (OP(scan) == NALNUMU) { |
a12cf05f KW |
4255 | for (value = 0; value < 256; value++) { |
4256 | if (isWORDCHAR_L1(value)) { | |
4257 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
4258 | } | |
4259 | } | |
4260 | } else { | |
4261 | for (value = 0; value < 256; value++) { | |
4262 | if (isALNUM(value)) { | |
4263 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
4264 | } | |
4265 | } | |
4266 | } | |
653099ff GS |
4267 | } |
4268 | } | |
8aa23a47 YO |
4269 | else { |
4270 | if (data->start_class->flags & ANYOF_LOCALE) | |
4271 | ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM); | |
af302e7f | 4272 | |
75950e1c KW |
4273 | /* Even if under locale, set the bits for non-locale in |
4274 | * case it isn't a true locale-node. This will create | |
4275 | * false positives if it truly is locale */ | |
4276 | if (OP(scan) == NALNUMU) { | |
4277 | for (value = 0; value < 256; value++) { | |
4278 | if (! isWORDCHAR_L1(value)) { | |
4279 | ANYOF_BITMAP_SET(data->start_class, value); | |
4280 | } | |
e9a9c1bc | 4281 | } |
75950e1c KW |
4282 | } else { |
4283 | for (value = 0; value < 256; value++) { | |
4284 | if (! isALNUM(value)) { | |
4285 | ANYOF_BITMAP_SET(data->start_class, value); | |
4286 | } | |
4287 | } | |
4288 | } | |
653099ff | 4289 | } |
8aa23a47 | 4290 | break; |
8aa23a47 YO |
4291 | case SPACE: |
4292 | if (flags & SCF_DO_STCLASS_AND) { | |
4293 | if (!(data->start_class->flags & ANYOF_LOCALE)) { | |
4294 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE); | |
980866de | 4295 | if (OP(scan) == SPACEU) { |
a12cf05f KW |
4296 | for (value = 0; value < 256; value++) { |
4297 | if (!isSPACE_L1(value)) { | |
4298 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
4299 | } | |
4300 | } | |
4301 | } else { | |
4302 | for (value = 0; value < 256; value++) { | |
4303 | if (!isSPACE(value)) { | |
4304 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
4305 | } | |
4306 | } | |
4307 | } | |
653099ff GS |
4308 | } |
4309 | } | |
8aa23a47 | 4310 | else { |
a12cf05f | 4311 | if (data->start_class->flags & ANYOF_LOCALE) { |
8aa23a47 | 4312 | ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE); |
a12cf05f | 4313 | } |
af302e7f | 4314 | if (OP(scan) == SPACEU) { |
a12cf05f KW |
4315 | for (value = 0; value < 256; value++) { |
4316 | if (isSPACE_L1(value)) { | |
4317 | ANYOF_BITMAP_SET(data->start_class, value); | |
4318 | } | |
4319 | } | |
4320 | } else { | |
4321 | for (value = 0; value < 256; value++) { | |
4322 | if (isSPACE(value)) { | |
4323 | ANYOF_BITMAP_SET(data->start_class, value); | |
4324 | } | |
4325 | } | |
8aa23a47 | 4326 | } |
653099ff | 4327 | } |
8aa23a47 | 4328 | break; |
8aa23a47 YO |
4329 | case NSPACE: |
4330 | if (flags & SCF_DO_STCLASS_AND) { | |
4331 | if (!(data->start_class->flags & ANYOF_LOCALE)) { | |
4332 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE); | |
980866de | 4333 | if (OP(scan) == NSPACEU) { |
a12cf05f KW |
4334 | for (value = 0; value < 256; value++) { |
4335 | if (isSPACE_L1(value)) { | |
4336 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
4337 | } | |
4338 | } | |
4339 | } else { | |
4340 | for (value = 0; value < 256; value++) { | |
4341 | if (isSPACE(value)) { | |
4342 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
4343 | } | |
4344 | } | |
4345 | } | |
653099ff | 4346 | } |
8aa23a47 YO |
4347 | } |
4348 | else { | |
4349 | if (data->start_class->flags & ANYOF_LOCALE) | |
4350 | ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE); | |
af302e7f | 4351 | if (OP(scan) == NSPACEU) { |
a12cf05f KW |
4352 | for (value = 0; value < 256; value++) { |
4353 | if (!isSPACE_L1(value)) { | |
4354 | ANYOF_BITMAP_SET(data->start_class, value); | |
4355 | } | |
4356 | } | |
4357 | } | |
4358 | else { | |
4359 | for (value = 0; value < 256; value++) { | |
4360 | if (!isSPACE(value)) { | |
4361 | ANYOF_BITMAP_SET(data->start_class, value); | |
4362 | } | |
4363 | } | |
4364 | } | |
653099ff | 4365 | } |
8aa23a47 | 4366 | break; |
8aa23a47 YO |
4367 | case DIGIT: |
4368 | if (flags & SCF_DO_STCLASS_AND) { | |
bcc0256f | 4369 | if (!(data->start_class->flags & ANYOF_LOCALE)) { |
bf3c5c06 KW |
4370 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT); |
4371 | for (value = 0; value < 256; value++) | |
4372 | if (!isDIGIT(value)) | |
4373 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
bcc0256f | 4374 | } |
8aa23a47 YO |
4375 | } |
4376 | else { | |
4377 | if (data->start_class->flags & ANYOF_LOCALE) | |
4378 | ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT); | |
75950e1c KW |
4379 | for (value = 0; value < 256; value++) |
4380 | if (isDIGIT(value)) | |
4381 | ANYOF_BITMAP_SET(data->start_class, value); | |
8aa23a47 YO |
4382 | } |
4383 | break; | |
4384 | case NDIGIT: | |
4385 | if (flags & SCF_DO_STCLASS_AND) { | |
bcc0256f | 4386 | if (!(data->start_class->flags & ANYOF_LOCALE)) |
bf3c5c06 | 4387 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT); |
8aa23a47 YO |
4388 | for (value = 0; value < 256; value++) |
4389 | if (isDIGIT(value)) | |
4390 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
4391 | } | |
4392 | else { | |
4393 | if (data->start_class->flags & ANYOF_LOCALE) | |
4394 | ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT); | |
75950e1c KW |
4395 | for (value = 0; value < 256; value++) |
4396 | if (!isDIGIT(value)) | |
4397 | ANYOF_BITMAP_SET(data->start_class, value); | |
653099ff | 4398 | } |
8aa23a47 | 4399 | break; |
e1d1eefb YO |
4400 | CASE_SYNST_FNC(VERTWS); |
4401 | CASE_SYNST_FNC(HORIZWS); | |
686b73d4 | 4402 | |
8aa23a47 YO |
4403 | } |
4404 | if (flags & SCF_DO_STCLASS_OR) | |
4405 | cl_and(data->start_class, and_withp); | |
4406 | flags &= ~SCF_DO_STCLASS; | |
4407 | } | |
4408 | } | |
4409 | else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) { | |
4410 | data->flags |= (OP(scan) == MEOL | |
4411 | ? SF_BEFORE_MEOL | |
4412 | : SF_BEFORE_SEOL); | |
ac7af3f6 YO |
4413 | SCAN_COMMIT(pRExC_state, data, minlenp); |
4414 | ||
8aa23a47 YO |
4415 | } |
4416 | else if ( PL_regkind[OP(scan)] == BRANCHJ | |
4417 | /* Lookbehind, or need to calculate parens/evals/stclass: */ | |
4418 | && (scan->flags || data || (flags & SCF_DO_STCLASS)) | |
4419 | && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) { | |
3b6759a6 YO |
4420 | if ( OP(scan) == UNLESSM && |
4421 | scan->flags == 0 && | |
4422 | OP(NEXTOPER(NEXTOPER(scan))) == NOTHING && | |
4423 | OP(regnext(NEXTOPER(NEXTOPER(scan)))) == SUCCEED | |
4424 | ) { | |
4425 | regnode *opt; | |
4426 | regnode *upto= regnext(scan); | |
4427 | DEBUG_PARSE_r({ | |
4428 | SV * const mysv_val=sv_newmortal(); | |
4429 | DEBUG_STUDYDATA("OPFAIL",data,depth); | |
4430 | ||
4431 | /*DEBUG_PARSE_MSG("opfail");*/ | |
4432 | regprop(RExC_rx, mysv_val, upto); | |
4433 | PerlIO_printf(Perl_debug_log, "~ replace with OPFAIL pointed at %s (%"IVdf") offset %"IVdf"\n", | |
4434 | SvPV_nolen_const(mysv_val), | |
4435 | (IV)REG_NODE_NUM(upto), | |
4436 | (IV)(upto - scan) | |
4437 | ); | |
4438 | }); | |
4439 | OP(scan) = OPFAIL; | |
4440 | NEXT_OFF(scan) = upto - scan; | |
4441 | for (opt= scan + 1; opt < upto ; opt++) | |
4442 | OP(opt) = OPTIMIZED; | |
4443 | scan= upto; | |
4444 | continue; | |
4445 | } | |
8aa23a47 YO |
4446 | if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY |
4447 | || OP(scan) == UNLESSM ) | |
4448 | { | |
4449 | /* Negative Lookahead/lookbehind | |
4450 | In this case we can't do fixed string optimisation. | |
4451 | */ | |
1de06328 | 4452 | |
8aa23a47 YO |
4453 | I32 deltanext, minnext, fake = 0; |
4454 | regnode *nscan; | |
4455 | struct regnode_charclass_class intrnl; | |
4456 | int f = 0; | |
1de06328 | 4457 | |
8aa23a47 YO |
4458 | data_fake.flags = 0; |
4459 | if (data) { | |
4460 | data_fake.whilem_c = data->whilem_c; | |
4461 | data_fake.last_closep = data->last_closep; | |
c277df42 | 4462 | } |
8aa23a47 YO |
4463 | else |
4464 | data_fake.last_closep = &fake; | |
58e23c8d | 4465 | data_fake.pos_delta = delta; |
8aa23a47 YO |
4466 | if ( flags & SCF_DO_STCLASS && !scan->flags |
4467 | && OP(scan) == IFMATCH ) { /* Lookahead */ | |
e755fd73 | 4468 | cl_init(pRExC_state, &intrnl); |
8aa23a47 YO |
4469 | data_fake.start_class = &intrnl; |
4470 | f |= SCF_DO_STCLASS_AND; | |
4471 | } | |
4472 | if (flags & SCF_WHILEM_VISITED_POS) | |
4473 | f |= SCF_WHILEM_VISITED_POS; | |
4474 | next = regnext(scan); | |
4475 | nscan = NEXTOPER(NEXTOPER(scan)); | |
4476 | minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext, | |
4477 | last, &data_fake, stopparen, recursed, NULL, f, depth+1); | |
4478 | if (scan->flags) { | |
4479 | if (deltanext) { | |
58e23c8d | 4480 | FAIL("Variable length lookbehind not implemented"); |
8aa23a47 YO |
4481 | } |
4482 | else if (minnext > (I32)U8_MAX) { | |
58e23c8d | 4483 | FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX); |
8aa23a47 YO |
4484 | } |
4485 | scan->flags = (U8)minnext; | |
4486 | } | |
4487 | if (data) { | |
4488 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) | |
4489 | pars++; | |
4490 | if (data_fake.flags & SF_HAS_EVAL) | |
4491 | data->flags |= SF_HAS_EVAL; | |
4492 | data->whilem_c = data_fake.whilem_c; | |
4493 | } | |
4494 | if (f & SCF_DO_STCLASS_AND) { | |
906cdd2b HS |
4495 | if (flags & SCF_DO_STCLASS_OR) { |
4496 | /* OR before, AND after: ideally we would recurse with | |
4497 | * data_fake to get the AND applied by study of the | |
4498 | * remainder of the pattern, and then derecurse; | |
4499 | * *** HACK *** for now just treat as "no information". | |
4500 | * See [perl #56690]. | |
4501 | */ | |
e755fd73 | 4502 | cl_init(pRExC_state, data->start_class); |
906cdd2b HS |
4503 | } else { |
4504 | /* AND before and after: combine and continue */ | |
4505 | const int was = (data->start_class->flags & ANYOF_EOS); | |
4506 | ||
4507 | cl_and(data->start_class, &intrnl); | |
4508 | if (was) | |
4509 | data->start_class->flags |= ANYOF_EOS; | |
4510 | } | |
8aa23a47 | 4511 | } |
cb434fcc | 4512 | } |
8aa23a47 YO |
4513 | #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY |
4514 | else { | |
4515 | /* Positive Lookahead/lookbehind | |
4516 | In this case we can do fixed string optimisation, | |
4517 | but we must be careful about it. Note in the case of | |
4518 | lookbehind the positions will be offset by the minimum | |
4519 | length of the pattern, something we won't know about | |
4520 | until after the recurse. | |
4521 | */ | |
4522 | I32 deltanext, fake = 0; | |
4523 | regnode *nscan; | |
4524 | struct regnode_charclass_class intrnl; | |
4525 | int f = 0; | |
4526 | /* We use SAVEFREEPV so that when the full compile | |
4527 | is finished perl will clean up the allocated | |
3b753521 | 4528 | minlens when it's all done. This way we don't |
8aa23a47 YO |
4529 | have to worry about freeing them when we know |
4530 | they wont be used, which would be a pain. | |
4531 | */ | |
4532 | I32 *minnextp; | |
4533 | Newx( minnextp, 1, I32 ); | |
4534 | SAVEFREEPV(minnextp); | |
4535 | ||
4536 | if (data) { | |
4537 | StructCopy(data, &data_fake, scan_data_t); | |
4538 | if ((flags & SCF_DO_SUBSTR) && data->last_found) { | |
4539 | f |= SCF_DO_SUBSTR; | |
4540 | if (scan->flags) | |
304ee84b | 4541 | SCAN_COMMIT(pRExC_state, &data_fake,minlenp); |
8aa23a47 YO |
4542 | data_fake.last_found=newSVsv(data->last_found); |
4543 | } | |
4544 | } | |
4545 | else | |
4546 | data_fake.last_closep = &fake; | |
4547 | data_fake.flags = 0; | |
58e23c8d | 4548 | data_fake.pos_delta = delta; |
8aa23a47 YO |
4549 | if (is_inf) |
4550 | data_fake.flags |= SF_IS_INF; | |
4551 | if ( flags & SCF_DO_STCLASS && !scan->flags | |
4552 | && OP(scan) == IFMATCH ) { /* Lookahead */ | |
e755fd73 | 4553 | cl_init(pRExC_state, &intrnl); |
8aa23a47 YO |
4554 | data_fake.start_class = &intrnl; |
4555 | f |= SCF_DO_STCLASS_AND; | |
4556 | } | |
4557 | if (flags & SCF_WHILEM_VISITED_POS) | |
4558 | f |= SCF_WHILEM_VISITED_POS; | |
4559 | next = regnext(scan); | |
4560 | nscan = NEXTOPER(NEXTOPER(scan)); | |
4561 | ||
4562 | *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext, | |
4563 | last, &data_fake, stopparen, recursed, NULL, f,depth+1); | |
4564 | if (scan->flags) { | |
4565 | if (deltanext) { | |
58e23c8d | 4566 | FAIL("Variable length lookbehind not implemented"); |
8aa23a47 YO |
4567 | } |
4568 | else if (*minnextp > (I32)U8_MAX) { | |
58e23c8d | 4569 | FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX); |
8aa23a47 YO |
4570 | } |
4571 | scan->flags = (U8)*minnextp; | |
4572 | } | |
4573 | ||
4574 | *minnextp += min; | |
4575 | ||
4576 | if (f & SCF_DO_STCLASS_AND) { | |
4577 | const int was = (data->start_class->flags & ANYOF_EOS); | |
4578 | ||
4579 | cl_and(data->start_class, &intrnl); | |
4580 | if (was) | |
4581 | data->start_class->flags |= ANYOF_EOS; | |
4582 | } | |
4583 | if (data) { | |
4584 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) | |
4585 | pars++; | |
4586 | if (data_fake.flags & SF_HAS_EVAL) | |
4587 | data->flags |= SF_HAS_EVAL; | |
4588 | data->whilem_c = data_fake.whilem_c; | |
4589 | if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) { | |
4590 | if (RExC_rx->minlen<*minnextp) | |
4591 | RExC_rx->minlen=*minnextp; | |
304ee84b | 4592 | SCAN_COMMIT(pRExC_state, &data_fake, minnextp); |
8aa23a47 YO |
4593 | SvREFCNT_dec(data_fake.last_found); |
4594 | ||
4595 | if ( data_fake.minlen_fixed != minlenp ) | |
4596 | { | |
4597 | data->offset_fixed= data_fake.offset_fixed; | |
4598 | data->minlen_fixed= data_fake.minlen_fixed; | |
4599 | data->lookbehind_fixed+= scan->flags; | |
4600 | } | |
4601 | if ( data_fake.minlen_float != minlenp ) | |
4602 | { | |
4603 | data->minlen_float= data_fake.minlen_float; | |
4604 | data->offset_float_min=data_fake.offset_float_min; | |
4605 | data->offset_float_max=data_fake.offset_float_max; | |
4606 | data->lookbehind_float+= scan->flags; | |
4607 | } | |
4608 | } | |
4609 | } | |
40d049e4 | 4610 | } |
8aa23a47 YO |
4611 | #endif |
4612 | } | |
4613 | else if (OP(scan) == OPEN) { | |
4614 | if (stopparen != (I32)ARG(scan)) | |
4615 | pars++; | |
4616 | } | |
4617 | else if (OP(scan) == CLOSE) { | |
4618 | if (stopparen == (I32)ARG(scan)) { | |
4619 | break; | |
4620 | } | |
4621 | if ((I32)ARG(scan) == is_par) { | |
4622 | next = regnext(scan); | |
b515a41d | 4623 | |
8aa23a47 YO |
4624 | if ( next && (OP(next) != WHILEM) && next < last) |
4625 | is_par = 0; /* Disable optimization */ | |
40d049e4 | 4626 | } |
8aa23a47 YO |
4627 | if (data) |
4628 | *(data->last_closep) = ARG(scan); | |
4629 | } | |
4630 | else if (OP(scan) == EVAL) { | |
c277df42 IZ |
4631 | if (data) |
4632 | data->flags |= SF_HAS_EVAL; | |
8aa23a47 YO |
4633 | } |
4634 | else if ( PL_regkind[OP(scan)] == ENDLIKE ) { | |
4635 | if (flags & SCF_DO_SUBSTR) { | |
304ee84b | 4636 | SCAN_COMMIT(pRExC_state,data,minlenp); |
8aa23a47 | 4637 | flags &= ~SCF_DO_SUBSTR; |
40d049e4 | 4638 | } |
8aa23a47 YO |
4639 | if (data && OP(scan)==ACCEPT) { |
4640 | data->flags |= SCF_SEEN_ACCEPT; | |
4641 | if (stopmin > min) | |
4642 | stopmin = min; | |
e2e6a0f1 | 4643 | } |
8aa23a47 YO |
4644 | } |
4645 | else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */ | |
4646 | { | |
0f5d15d6 | 4647 | if (flags & SCF_DO_SUBSTR) { |
304ee84b | 4648 | SCAN_COMMIT(pRExC_state,data,minlenp); |
0f5d15d6 IZ |
4649 | data->longest = &(data->longest_float); |
4650 | } | |
4651 | is_inf = is_inf_internal = 1; | |
653099ff | 4652 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ |
3fffb88a | 4653 | cl_anything(pRExC_state, data->start_class); |
96776eda | 4654 | flags &= ~SCF_DO_STCLASS; |
8aa23a47 | 4655 | } |
58e23c8d | 4656 | else if (OP(scan) == GPOS) { |
bbe252da | 4657 | if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) && |
58e23c8d YO |
4658 | !(delta || is_inf || (data && data->pos_delta))) |
4659 | { | |
bbe252da YO |
4660 | if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR)) |
4661 | RExC_rx->extflags |= RXf_ANCH_GPOS; | |
58e23c8d YO |
4662 | if (RExC_rx->gofs < (U32)min) |
4663 | RExC_rx->gofs = min; | |
4664 | } else { | |
bbe252da | 4665 | RExC_rx->extflags |= RXf_GPOS_FLOAT; |
58e23c8d YO |
4666 | RExC_rx->gofs = 0; |
4667 | } | |
4668 | } | |
786e8c11 | 4669 | #ifdef TRIE_STUDY_OPT |
40d049e4 | 4670 | #ifdef FULL_TRIE_STUDY |
8aa23a47 YO |
4671 | else if (PL_regkind[OP(scan)] == TRIE) { |
4672 | /* NOTE - There is similar code to this block above for handling | |
4673 | BRANCH nodes on the initial study. If you change stuff here | |
4674 | check there too. */ | |
4675 | regnode *trie_node= scan; | |
4676 | regnode *tail= regnext(scan); | |
f8fc2ecf | 4677 | reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ]; |
8aa23a47 YO |
4678 | I32 max1 = 0, min1 = I32_MAX; |
4679 | struct regnode_charclass_class accum; | |
4680 | ||
4681 | if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */ | |
304ee84b | 4682 | SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */ |
8aa23a47 | 4683 | if (flags & SCF_DO_STCLASS) |
e755fd73 | 4684 | cl_init_zero(pRExC_state, &accum); |
8aa23a47 YO |
4685 | |
4686 | if (!trie->jump) { | |
4687 | min1= trie->minlen; | |
4688 | max1= trie->maxlen; | |
4689 | } else { | |
4690 | const regnode *nextbranch= NULL; | |
4691 | U32 word; | |
4692 | ||
4693 | for ( word=1 ; word <= trie->wordcount ; word++) | |
4694 | { | |
4695 | I32 deltanext=0, minnext=0, f = 0, fake; | |
4696 | struct regnode_charclass_class this_class; | |
4697 | ||
4698 | data_fake.flags = 0; | |
4699 | if (data) { | |
4700 | data_fake.whilem_c = data->whilem_c; | |
4701 | data_fake.last_closep = data->last_closep; | |
4702 | } | |
4703 | else | |
4704 | data_fake.last_closep = &fake; | |
58e23c8d | 4705 | data_fake.pos_delta = delta; |
8aa23a47 | 4706 | if (flags & SCF_DO_STCLASS) { |
e755fd73 | 4707 | cl_init(pRExC_state, &this_class); |
8aa23a47 YO |
4708 | data_fake.start_class = &this_class; |
4709 | f = SCF_DO_STCLASS_AND; | |
4710 | } | |
4711 | if (flags & SCF_WHILEM_VISITED_POS) | |
4712 | f |= SCF_WHILEM_VISITED_POS; | |
4713 | ||
4714 | if (trie->jump[word]) { | |
4715 | if (!nextbranch) | |
4716 | nextbranch = trie_node + trie->jump[0]; | |
4717 | scan= trie_node + trie->jump[word]; | |
4718 | /* We go from the jump point to the branch that follows | |
4719 | it. Note this means we need the vestigal unused branches | |
4720 | even though they arent otherwise used. | |
4721 | */ | |
4722 | minnext = study_chunk(pRExC_state, &scan, minlenp, | |
4723 | &deltanext, (regnode *)nextbranch, &data_fake, | |
4724 | stopparen, recursed, NULL, f,depth+1); | |
4725 | } | |
4726 | if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH) | |
4727 | nextbranch= regnext((regnode*)nextbranch); | |
4728 | ||
4729 | if (min1 > (I32)(minnext + trie->minlen)) | |
4730 | min1 = minnext + trie->minlen; | |
4731 | if (max1 < (I32)(minnext + deltanext + trie->maxlen)) | |
4732 | max1 = minnext + deltanext + trie->maxlen; | |
4733 | if (deltanext == I32_MAX) | |
4734 | is_inf = is_inf_internal = 1; | |
4735 | ||
4736 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) | |
4737 | pars++; | |
4738 | if (data_fake.flags & SCF_SEEN_ACCEPT) { | |
4739 | if ( stopmin > min + min1) | |
4740 | stopmin = min + min1; | |
4741 | flags &= ~SCF_DO_SUBSTR; | |
4742 | if (data) | |
4743 | data->flags |= SCF_SEEN_ACCEPT; | |
4744 | } | |
4745 | if (data) { | |
4746 | if (data_fake.flags & SF_HAS_EVAL) | |
4747 | data->flags |= SF_HAS_EVAL; | |
4748 | data->whilem_c = data_fake.whilem_c; | |
4749 | } | |
4750 | if (flags & SCF_DO_STCLASS) | |
3fffb88a | 4751 | cl_or(pRExC_state, &accum, &this_class); |
8aa23a47 YO |
4752 | } |
4753 | } | |
4754 | if (flags & SCF_DO_SUBSTR) { | |
4755 | data->pos_min += min1; | |
4756 | data->pos_delta += max1 - min1; | |
4757 | if (max1 != min1 || is_inf) | |
4758 | data->longest = &(data->longest_float); | |
4759 | } | |
4760 | min += min1; | |
4761 | delta += max1 - min1; | |
4762 | if (flags & SCF_DO_STCLASS_OR) { | |
3fffb88a | 4763 | cl_or(pRExC_state, data->start_class, &accum); |
8aa23a47 YO |
4764 | if (min1) { |
4765 | cl_and(data->start_class, and_withp); | |
4766 | flags &= ~SCF_DO_STCLASS; | |
4767 | } | |
4768 | } | |
4769 | else if (flags & SCF_DO_STCLASS_AND) { | |
4770 | if (min1) { | |
4771 | cl_and(data->start_class, &accum); | |
4772 | flags &= ~SCF_DO_STCLASS; | |
4773 | } | |
4774 | else { | |
4775 | /* Switch to OR mode: cache the old value of | |
4776 | * data->start_class */ | |
4777 | INIT_AND_WITHP; | |
4778 | StructCopy(data->start_class, and_withp, | |
4779 | struct regnode_charclass_class); | |
4780 | flags &= ~SCF_DO_STCLASS_AND; | |
4781 | StructCopy(&accum, data->start_class, | |
4782 | struct regnode_charclass_class); | |
4783 | flags |= SCF_DO_STCLASS_OR; | |
4784 | data->start_class->flags |= ANYOF_EOS; | |
4785 | } | |
4786 | } | |
4787 | scan= tail; | |
4788 | continue; | |
4789 | } | |
786e8c11 | 4790 | #else |
8aa23a47 | 4791 | else if (PL_regkind[OP(scan)] == TRIE) { |
f8fc2ecf | 4792 | reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ]; |
8aa23a47 YO |
4793 | U8*bang=NULL; |
4794 | ||
4795 | min += trie->minlen; | |
4796 | delta += (trie->maxlen - trie->minlen); | |
4797 | flags &= ~SCF_DO_STCLASS; /* xxx */ | |
4798 | if (flags & SCF_DO_SUBSTR) { | |
304ee84b | 4799 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ |
8aa23a47 YO |
4800 | data->pos_min += trie->minlen; |
4801 | data->pos_delta += (trie->maxlen - trie->minlen); | |
4802 | if (trie->maxlen != trie->minlen) | |
4803 | data->longest = &(data->longest_float); | |
4804 | } | |
4805 | if (trie->jump) /* no more substrings -- for now /grr*/ | |
4806 | flags &= ~SCF_DO_SUBSTR; | |
b515a41d | 4807 | } |
8aa23a47 | 4808 | #endif /* old or new */ |
686b73d4 | 4809 | #endif /* TRIE_STUDY_OPT */ |
e1d1eefb | 4810 | |
8aa23a47 YO |
4811 | /* Else: zero-length, ignore. */ |
4812 | scan = regnext(scan); | |
4813 | } | |
4814 | if (frame) { | |
4815 | last = frame->last; | |
4816 | scan = frame->next; | |
4817 | stopparen = frame->stop; | |
4818 | frame = frame->prev; | |
4819 | goto fake_study_recurse; | |
c277df42 IZ |
4820 | } |
4821 | ||
4822 | finish: | |
8aa23a47 | 4823 | assert(!frame); |
304ee84b | 4824 | DEBUG_STUDYDATA("pre-fin:",data,depth); |
8aa23a47 | 4825 | |
c277df42 | 4826 | *scanp = scan; |
aca2d497 | 4827 | *deltap = is_inf_internal ? I32_MAX : delta; |
b81d288d | 4828 | if (flags & SCF_DO_SUBSTR && is_inf) |
c277df42 | 4829 | data->pos_delta = I32_MAX - data->pos_min; |
786e8c11 | 4830 | if (is_par > (I32)U8_MAX) |
c277df42 IZ |
4831 | is_par = 0; |
4832 | if (is_par && pars==1 && data) { | |
4833 | data->flags |= SF_IN_PAR; | |
4834 | data->flags &= ~SF_HAS_PAR; | |
a0ed51b3 LW |
4835 | } |
4836 | else if (pars && data) { | |
c277df42 IZ |
4837 | data->flags |= SF_HAS_PAR; |
4838 | data->flags &= ~SF_IN_PAR; | |
4839 | } | |
653099ff | 4840 | if (flags & SCF_DO_STCLASS_OR) |
40d049e4 | 4841 | cl_and(data->start_class, and_withp); |
786e8c11 YO |
4842 | if (flags & SCF_TRIE_RESTUDY) |
4843 | data->flags |= SCF_TRIE_RESTUDY; | |
1de06328 | 4844 | |
304ee84b | 4845 | DEBUG_STUDYDATA("post-fin:",data,depth); |
1de06328 | 4846 | |
e2e6a0f1 | 4847 | return min < stopmin ? min : stopmin; |
c277df42 IZ |
4848 | } |
4849 | ||
2eccd3b2 NC |
4850 | STATIC U32 |
4851 | S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s) | |
c277df42 | 4852 | { |
4a4e7719 NC |
4853 | U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0; |
4854 | ||
7918f24d NC |
4855 | PERL_ARGS_ASSERT_ADD_DATA; |
4856 | ||
4a4e7719 NC |
4857 | Renewc(RExC_rxi->data, |
4858 | sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1), | |
4859 | char, struct reg_data); | |
4860 | if(count) | |
f8fc2ecf | 4861 | Renew(RExC_rxi->data->what, count + n, U8); |
4a4e7719 | 4862 | else |
f8fc2ecf | 4863 | Newx(RExC_rxi->data->what, n, U8); |
4a4e7719 NC |
4864 | RExC_rxi->data->count = count + n; |
4865 | Copy(s, RExC_rxi->data->what + count, n, U8); | |
4866 | return count; | |
c277df42 IZ |
4867 | } |
4868 | ||
f8149455 | 4869 | /*XXX: todo make this not included in a non debugging perl */ |
76234dfb | 4870 | #ifndef PERL_IN_XSUB_RE |
d88dccdf | 4871 | void |
864dbfa3 | 4872 | Perl_reginitcolors(pTHX) |
d88dccdf | 4873 | { |
97aff369 | 4874 | dVAR; |
1df70142 | 4875 | const char * const s = PerlEnv_getenv("PERL_RE_COLORS"); |
d88dccdf | 4876 | if (s) { |
1df70142 AL |
4877 | char *t = savepv(s); |
4878 | int i = 0; | |
4879 | PL_colors[0] = t; | |
d88dccdf | 4880 | while (++i < 6) { |
1df70142 AL |
4881 | t = strchr(t, '\t'); |
4882 | if (t) { | |
4883 | *t = '\0'; | |
4884 | PL_colors[i] = ++t; | |
d88dccdf IZ |
4885 | } |
4886 | else | |
1df70142 | 4887 | PL_colors[i] = t = (char *)""; |
d88dccdf IZ |
4888 | } |
4889 | } else { | |
1df70142 | 4890 | int i = 0; |
b81d288d | 4891 | while (i < 6) |
06b5626a | 4892 | PL_colors[i++] = (char *)""; |
d88dccdf IZ |
4893 | } |
4894 | PL_colorset = 1; | |
4895 | } | |
76234dfb | 4896 | #endif |
8615cb43 | 4897 | |
07be1b83 | 4898 | |
786e8c11 YO |
4899 | #ifdef TRIE_STUDY_OPT |
4900 | #define CHECK_RESTUDY_GOTO \ | |
4901 | if ( \ | |
4902 | (data.flags & SCF_TRIE_RESTUDY) \ | |
4903 | && ! restudied++ \ | |
4904 | ) goto reStudy | |
4905 | #else | |
4906 | #define CHECK_RESTUDY_GOTO | |
4907 | #endif | |
f9f4320a | 4908 | |
a687059c | 4909 | /* |
3a21f536 | 4910 | * pregcomp - compile a regular expression into internal code |
a687059c | 4911 | * |
3a21f536 DM |
4912 | * Decides which engine's compiler to call based on the hint currently in |
4913 | * scope | |
a687059c | 4914 | */ |
b9b4dddf | 4915 | |
6d5c990f | 4916 | #ifndef PERL_IN_XSUB_RE |
f8b2cf8a | 4917 | |
3c13cae6 | 4918 | /* return the currently in-scope regex engine (or the default if none) */ |
f8b2cf8a | 4919 | |
3c13cae6 | 4920 | regexp_engine const * |
f8b2cf8a DM |
4921 | Perl_current_re_engine(pTHX) |
4922 | { | |
4923 | dVAR; | |
4924 | ||
4925 | if (IN_PERL_COMPILETIME) { | |
4926 | HV * const table = GvHV(PL_hintgv); | |
4927 | SV **ptr; | |
4928 | ||
4929 | if (!table) | |
3c13cae6 | 4930 | return &PL_core_reg_engine; |
f8b2cf8a DM |
4931 | ptr = hv_fetchs(table, "regcomp", FALSE); |
4932 | if ( !(ptr && SvIOK(*ptr) && SvIV(*ptr))) | |
3c13cae6 | 4933 | return &PL_core_reg_engine; |
f8b2cf8a DM |
4934 | return INT2PTR(regexp_engine*,SvIV(*ptr)); |
4935 | } | |
4936 | else { | |
4937 | SV *ptr; | |
4938 | if (!PL_curcop->cop_hints_hash) | |
3c13cae6 | 4939 | return &PL_core_reg_engine; |
f8b2cf8a DM |
4940 | ptr = cop_hints_fetch_pvs(PL_curcop, "regcomp", 0); |
4941 | if ( !(ptr && SvIOK(ptr) && SvIV(ptr))) | |
3c13cae6 | 4942 | return &PL_core_reg_engine; |
f8b2cf8a DM |
4943 | return INT2PTR(regexp_engine*,SvIV(ptr)); |
4944 | } | |
4945 | } | |
4946 | ||
4947 | ||
3ab4a224 | 4948 | REGEXP * |
1593ad57 | 4949 | Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags) |
a687059c | 4950 | { |
97aff369 | 4951 | dVAR; |
3c13cae6 DM |
4952 | regexp_engine const *eng = current_re_engine(); |
4953 | GET_RE_DEBUG_FLAGS_DECL; | |
7918f24d NC |
4954 | |
4955 | PERL_ARGS_ASSERT_PREGCOMP; | |
4956 | ||
f8b2cf8a | 4957 | /* Dispatch a request to compile a regexp to correct regexp engine. */ |
3c13cae6 DM |
4958 | DEBUG_COMPILE_r({ |
4959 | PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n", | |
4960 | PTR2UV(eng)); | |
4961 | }); | |
4962 | return CALLREGCOMP_ENG(eng, pattern, flags); | |
2a5d9b1d | 4963 | } |
6d5c990f | 4964 | #endif |
2a5d9b1d | 4965 | |
40b6423c DM |
4966 | /* public(ish) entry point for the perl core's own regex compiling code. |
4967 | * It's actually a wrapper for Perl_re_op_compile that only takes an SV | |
4968 | * pattern rather than a list of OPs, and uses the internal engine rather | |
4969 | * than the current one */ | |
74529a43 DM |
4970 | |
4971 | REGEXP * | |
37acfcba | 4972 | Perl_re_compile(pTHX_ SV * const pattern, U32 rx_flags) |
74529a43 | 4973 | { |
9f141731 | 4974 | SV *pat = pattern; /* defeat constness! */ |
74529a43 | 4975 | PERL_ARGS_ASSERT_RE_COMPILE; |
40b6423c DM |
4976 | return Perl_re_op_compile(aTHX_ &pat, 1, NULL, |
4977 | #ifdef PERL_IN_XSUB_RE | |
4978 | &my_reg_engine, | |
4979 | #else | |
4980 | &PL_core_reg_engine, | |
4981 | #endif | |
4982 | NULL, NULL, rx_flags, 0); | |
74529a43 DM |
4983 | } |
4984 | ||
d24ca0c5 DM |
4985 | /* see if there are any run-time code blocks in the pattern. |
4986 | * False positives are allowed */ | |
4987 | ||
4988 | static bool | |
4989 | S_has_runtime_code(pTHX_ RExC_state_t * const pRExC_state, OP *expr, | |
4990 | U32 pm_flags, char *pat, STRLEN plen) | |
4991 | { | |
4992 | int n = 0; | |
4993 | STRLEN s; | |
4994 | ||
4995 | /* avoid infinitely recursing when we recompile the pattern parcelled up | |
4996 | * as qr'...'. A single constant qr// string can't have have any | |
4997 | * run-time component in it, and thus, no runtime code. (A non-qr | |
4998 | * string, however, can, e.g. $x =~ '(?{})') */ | |
4999 | if ((pm_flags & PMf_IS_QR) && expr && expr->op_type == OP_CONST) | |
5000 | return 0; | |
5001 | ||
5002 | for (s = 0; s < plen; s++) { | |
5003 | if (n < pRExC_state->num_code_blocks | |
5004 | && s == pRExC_state->code_blocks[n].start) | |
5005 | { | |
5006 | s = pRExC_state->code_blocks[n].end; | |
5007 | n++; | |
5008 | continue; | |
5009 | } | |
5010 | /* TODO ideally should handle [..], (#..), /#.../x to reduce false | |
5011 | * positives here */ | |
5012 | if (pat[s] == '(' && pat[s+1] == '?' && | |
5013 | (pat[s+2] == '{' || (pat[s+2] == '?' && pat[s+3] == '{')) | |
5014 | ) | |
5015 | return 1; | |
5016 | } | |
5017 | return 0; | |
5018 | } | |
5019 | ||
5020 | /* Handle run-time code blocks. We will already have compiled any direct | |
5021 | * or indirect literal code blocks. Now, take the pattern 'pat' and make a | |
5022 | * copy of it, but with any literal code blocks blanked out and | |
5023 | * appropriate chars escaped; then feed it into | |
5024 | * | |
5025 | * eval "qr'modified_pattern'" | |
5026 | * | |
5027 | * For example, | |
5028 | * | |
5029 | * a\bc(?{"this was literal"})def'ghi\\jkl(?{"this is runtime"})mno | |
5030 | * | |
5031 | * becomes | |
5032 | * | |
5033 | * qr'a\\bc def\'ghi\\\\jkl(?{"this is runtime"})mno' | |
5034 | * | |
5035 | * After eval_sv()-ing that, grab any new code blocks from the returned qr | |
5036 | * and merge them with any code blocks of the original regexp. | |
5037 | * | |
5038 | * If the pat is non-UTF8, while the evalled qr is UTF8, don't merge; | |
5039 | * instead, just save the qr and return FALSE; this tells our caller that | |
5040 | * the original pattern needs upgrading to utf8. | |
5041 | */ | |
5042 | ||
fbdb18e8 | 5043 | static bool |
d24ca0c5 DM |
5044 | S_compile_runtime_code(pTHX_ RExC_state_t * const pRExC_state, |
5045 | char *pat, STRLEN plen) | |
5046 | { | |
5047 | SV *qr; | |
5048 | ||
5049 | GET_RE_DEBUG_FLAGS_DECL; | |
5050 | ||
5051 | if (pRExC_state->runtime_code_qr) { | |
5052 | /* this is the second time we've been called; this should | |
5053 | * only happen if the main pattern got upgraded to utf8 | |
5054 | * during compilation; re-use the qr we compiled first time | |
5055 | * round (which should be utf8 too) | |
5056 | */ | |
5057 | qr = pRExC_state->runtime_code_qr; | |
5058 | pRExC_state->runtime_code_qr = NULL; | |
5059 | assert(RExC_utf8 && SvUTF8(qr)); | |
5060 | } | |
5061 | else { | |
5062 | int n = 0; | |
5063 | STRLEN s; | |
5064 | char *p, *newpat; | |
c8d84f8c | 5065 | int newlen = plen + 6; /* allow for "qr''x\0" extra chars */ |
d24ca0c5 DM |
5066 | SV *sv, *qr_ref; |
5067 | dSP; | |
5068 | ||
5069 | /* determine how many extra chars we need for ' and \ escaping */ | |
5070 | for (s = 0; s < plen; s++) { | |
5071 | if (pat[s] == '\'' || pat[s] == '\\') | |
5072 | newlen++; | |
5073 | } | |
5074 | ||
5075 | Newx(newpat, newlen, char); | |
5076 | p = newpat; | |
5077 | *p++ = 'q'; *p++ = 'r'; *p++ = '\''; | |
5078 | ||
5079 | for (s = 0; s < plen; s++) { | |
5080 | if (n < pRExC_state->num_code_blocks | |
5081 | && s == pRExC_state->code_blocks[n].start) | |
5082 | { | |
5083 | /* blank out literal code block */ | |
5084 | assert(pat[s] == '('); | |
5085 | while (s <= pRExC_state->code_blocks[n].end) { | |
5086 | *p++ = ' '; | |
5087 | s++; | |
5088 | } | |
5089 | s--; | |
5090 | n++; | |
5091 | continue; | |
5092 | } | |
5093 | if (pat[s] == '\'' || pat[s] == '\\') | |
5094 | *p++ = '\\'; | |
5095 | *p++ = pat[s]; | |
5096 | } | |
5097 | *p++ = '\''; | |
5098 | if (pRExC_state->pm_flags & RXf_PMf_EXTENDED) | |
5099 | *p++ = 'x'; | |
5100 | *p++ = '\0'; | |
5101 | DEBUG_COMPILE_r({ | |
5102 | PerlIO_printf(Perl_debug_log, | |
5103 | "%sre-parsing pattern for runtime code:%s %s\n", | |
5104 | PL_colors[4],PL_colors[5],newpat); | |
5105 | }); | |
5106 | ||
5107 | sv = newSVpvn_flags(newpat, p-newpat-1, RExC_utf8 ? SVf_UTF8 : 0); | |
5108 | Safefree(newpat); | |
5109 | ||
5110 | ENTER; | |
5111 | SAVETMPS; | |
5112 | save_re_context(); | |
5113 | PUSHSTACKi(PERLSI_REQUIRE); | |
5114 | /* this causes the toker to collapse \\ into \ when parsing | |
5115 | * qr''; normally only q'' does this. It also alters hints | |
5116 | * handling */ | |
5117 | PL_reg_state.re_reparsing = TRUE; | |
5118 | eval_sv(sv, G_SCALAR); | |
5119 | SvREFCNT_dec(sv); | |
5120 | SPAGAIN; | |
5121 | qr_ref = POPs; | |
5122 | PUTBACK; | |
5123 | if (SvTRUE(ERRSV)) | |
5124 | Perl_croak(aTHX_ "%s", SvPVx_nolen_const(ERRSV)); | |
5125 | assert(SvROK(qr_ref)); | |
5126 | qr = SvRV(qr_ref); | |
5127 | assert(SvTYPE(qr) == SVt_REGEXP && RX_ENGINE((REGEXP*)qr)->op_comp); | |
5128 | /* the leaving below frees the tmp qr_ref. | |
5129 | * Give qr a life of its own */ | |
5130 | SvREFCNT_inc(qr); | |
5131 | POPSTACK; | |
5132 | FREETMPS; | |
5133 | LEAVE; | |
5134 | ||
5135 | } | |
5136 | ||
5137 | if (!RExC_utf8 && SvUTF8(qr)) { | |
5138 | /* first time through; the pattern got upgraded; save the | |
5139 | * qr for the next time through */ | |
5140 | assert(!pRExC_state->runtime_code_qr); | |
5141 | pRExC_state->runtime_code_qr = qr; | |
5142 | return 0; | |
5143 | } | |
5144 | ||
5145 | ||
5146 | /* extract any code blocks within the returned qr// */ | |
5147 | ||
5148 | ||
5149 | /* merge the main (r1) and run-time (r2) code blocks into one */ | |
5150 | { | |
5151 | RXi_GET_DECL(((struct regexp*)SvANY(qr)), r2); | |
5152 | struct reg_code_block *new_block, *dst; | |
5153 | RExC_state_t * const r1 = pRExC_state; /* convenient alias */ | |
5154 | int i1 = 0, i2 = 0; | |
5155 | ||
5156 | if (!r2->num_code_blocks) /* we guessed wrong */ | |
5157 | return 1; | |
5158 | ||
5159 | Newx(new_block, | |
5160 | r1->num_code_blocks + r2->num_code_blocks, | |
5161 | struct reg_code_block); | |
5162 | dst = new_block; | |
5163 | ||
5164 | while ( i1 < r1->num_code_blocks | |
5165 | || i2 < r2->num_code_blocks) | |
5166 | { | |
5167 | struct reg_code_block *src; | |
5168 | bool is_qr = 0; | |
5169 | ||
5170 | if (i1 == r1->num_code_blocks) { | |
5171 | src = &r2->code_blocks[i2++]; | |
5172 | is_qr = 1; | |
5173 | } | |
5174 | else if (i2 == r2->num_code_blocks) | |
5175 | src = &r1->code_blocks[i1++]; | |
5176 | else if ( r1->code_blocks[i1].start | |
5177 | < r2->code_blocks[i2].start) | |
5178 | { | |
5179 | src = &r1->code_blocks[i1++]; | |
5180 | assert(src->end < r2->code_blocks[i2].start); | |
5181 | } | |
5182 | else { | |
5183 | assert( r1->code_blocks[i1].start | |
5184 | > r2->code_blocks[i2].start); | |
5185 | src = &r2->code_blocks[i2++]; | |
5186 | is_qr = 1; | |
5187 | assert(src->end < r1->code_blocks[i1].start); | |
5188 | } | |
5189 | ||
5190 | assert(pat[src->start] == '('); | |
5191 | assert(pat[src->end] == ')'); | |
5192 | dst->start = src->start; | |
5193 | dst->end = src->end; | |
5194 | dst->block = src->block; | |
5195 | dst->src_regex = is_qr ? (REGEXP*) SvREFCNT_inc( (SV*) qr) | |
5196 | : src->src_regex; | |
5197 | dst++; | |
5198 | } | |
5199 | r1->num_code_blocks += r2->num_code_blocks; | |
5200 | Safefree(r1->code_blocks); | |
5201 | r1->code_blocks = new_block; | |
5202 | } | |
5203 | ||
5204 | SvREFCNT_dec(qr); | |
5205 | return 1; | |
5206 | } | |
5207 | ||
68e2671b | 5208 | |
7d103382 KW |
5209 | STATIC bool |
5210 | S_setup_longest(pTHX_ RExC_state_t *pRExC_state, SV* sv_longest, SV** rx_utf8, SV** rx_substr, I32* rx_end_shift, I32 lookbehind, I32 offset, I32 *minlen, STRLEN longest_length, bool eol, bool meol) | |
5211 | { | |
5212 | /* This is the common code for setting up the floating and fixed length | |
5213 | * string data extracted from Perlre_op_compile() below. Returns a boolean | |
5214 | * as to whether succeeded or not */ | |
5215 | ||
5216 | I32 t,ml; | |
5217 | ||
5218 | if (! (longest_length | |
5219 | || (eol /* Can't have SEOL and MULTI */ | |
5220 | && (! meol || (RExC_flags & RXf_PMf_MULTILINE))) | |
5221 | ) | |
5222 | /* See comments for join_exact for why REG_SEEN_EXACTF_SHARP_S */ | |
5223 | || (RExC_seen & REG_SEEN_EXACTF_SHARP_S)) | |
5224 | { | |
5225 | return FALSE; | |
5226 | } | |
5227 | ||
5228 | /* copy the information about the longest from the reg_scan_data | |
5229 | over to the program. */ | |
5230 | if (SvUTF8(sv_longest)) { | |
5231 | *rx_utf8 = sv_longest; | |
5232 | *rx_substr = NULL; | |
5233 | } else { | |
5234 | *rx_substr = sv_longest; | |
5235 | *rx_utf8 = NULL; | |
5236 | } | |
5237 | /* end_shift is how many chars that must be matched that | |
5238 | follow this item. We calculate it ahead of time as once the | |
5239 | lookbehind offset is added in we lose the ability to correctly | |
5240 | calculate it.*/ | |
5241 | ml = minlen ? *(minlen) : (I32)longest_length; | |
5242 | *rx_end_shift = ml - offset | |
5243 | - longest_length + (SvTAIL(sv_longest) != 0) | |
5244 | + lookbehind; | |
5245 | ||
5246 | t = (eol/* Can't have SEOL and MULTI */ | |
5247 | && (! meol || (RExC_flags & RXf_PMf_MULTILINE))); | |
5248 | fbm_compile(sv_longest, t ? FBMcf_TAIL : 0); | |
5249 | ||
5250 | return TRUE; | |
5251 | } | |
5252 | ||
3a21f536 | 5253 | /* |
ab442d34 | 5254 | * Perl_re_op_compile - the perl internal RE engine's function to compile a |
74529a43 | 5255 | * regular expression into internal code. |
28dc7a3e DM |
5256 | * The pattern may be passed either as: |
5257 | * a list of SVs (patternp plus pat_count) | |
5258 | * a list of OPs (expr) | |
9f141731 | 5259 | * If both are passed, the SV list is used, but the OP list indicates |
ab442d34 | 5260 | * which SVs are actually pre-compiled code blocks |
9f141731 | 5261 | * |
ab442d34 | 5262 | * The SVs in the list have magic and qr overloading applied to them (and |
9f141731 DM |
5263 | * the list may be modified in-place with replacement SVs in the latter |
5264 | * case). | |
5265 | * | |
ab442d34 | 5266 | * If the pattern hasn't changed from old_re, then old_re will be |
9f141731 DM |
5267 | * returned. |
5268 | * | |
3c13cae6 DM |
5269 | * eng is the current engine. If that engine has an op_comp method, then |
5270 | * handle directly (i.e. we assume that op_comp was us); otherwise, just | |
5271 | * do the initial concatenation of arguments and pass on to the external | |
ab442d34 | 5272 | * engine. |
9f141731 | 5273 | * |
ab442d34 DM |
5274 | * If is_bare_re is not null, set it to a boolean indicating whether the |
5275 | * arg list reduced (after overloading) to a single bare regex which has | |
5276 | * been returned (i.e. /$qr/). | |
3a21f536 | 5277 | * |
514a91f1 DM |
5278 | * orig_rx_flags contains RXf_* flags. See perlreapi.pod for more details. |
5279 | * | |
a5ae69f0 DM |
5280 | * pm_flags contains the PMf_* flags, typically based on those from the |
5281 | * pm_flags field of the related PMOP. Currently we're only interested in | |
5282 | * PMf_HAS_CV, PMf_IS_QR, PMf_USE_RE_EVAL. | |
514a91f1 | 5283 | * |
3a21f536 DM |
5284 | * We can't allocate space until we know how big the compiled form will be, |
5285 | * but we can't compile it (and thus know how big it is) until we've got a | |
5286 | * place to put the code. So we cheat: we compile it twice, once with code | |
5287 | * generation turned off and size counting turned on, and once "for real". | |
5288 | * This also means that we don't allocate space until we are sure that the | |
5289 | * thing really will compile successfully, and we never have to move the | |
5290 | * code and thus invalidate pointers into it. (Note that it has to be in | |
5291 | * one piece because free() must be able to free it all.) [NB: not true in perl] | |
5292 | * | |
5293 | * Beware that the optimization-preparation code in here knows about some | |
5294 | * of the structure of the compiled regexp. [I'll say.] | |
5295 | */ | |
5296 | ||
3ab4a224 | 5297 | REGEXP * |
9f141731 | 5298 | Perl_re_op_compile(pTHX_ SV ** const patternp, int pat_count, |
6ae44cd2 | 5299 | OP *expr, const regexp_engine* eng, REGEXP *VOL old_re, |
76ac488f | 5300 | bool *is_bare_re, U32 orig_rx_flags, U32 pm_flags) |
2a5d9b1d RGS |
5301 | { |
5302 | dVAR; | |
288b8c02 NC |
5303 | REGEXP *rx; |
5304 | struct regexp *r; | |
eb578fdb | 5305 | regexp_internal *ri; |
3ab4a224 | 5306 | STRLEN plen; |
21f84aaf | 5307 | char * VOL exp; |
5d51ce98 | 5308 | char* xend; |
c277df42 | 5309 | regnode *scan; |
a0d0e21e | 5310 | I32 flags; |
a0d0e21e | 5311 | I32 minlen = 0; |
37acfcba | 5312 | U32 rx_flags; |
e174e89a | 5313 | SV * VOL pat; |
e7f38d0f YO |
5314 | |
5315 | /* these are all flags - maybe they should be turned | |
5316 | * into a single int with different bit masks */ | |
5317 | I32 sawlookahead = 0; | |
a0d0e21e LW |
5318 | I32 sawplus = 0; |
5319 | I32 sawopen = 0; | |
29b09c41 | 5320 | bool used_setjump = FALSE; |
37acfcba | 5321 | regex_charset initial_charset = get_regex_charset(orig_rx_flags); |
83dd4485 | 5322 | bool code_is_utf8 = 0; |
23f4026a | 5323 | bool VOL recompile = 0; |
d24ca0c5 | 5324 | bool runtime_code = 0; |
bbd61b5f KW |
5325 | U8 jump_ret = 0; |
5326 | dJMPENV; | |
2c2d71f5 | 5327 | scan_data_t data; |
830247a4 | 5328 | RExC_state_t RExC_state; |
be8e71aa | 5329 | RExC_state_t * const pRExC_state = &RExC_state; |
07be1b83 | 5330 | #ifdef TRIE_STUDY_OPT |
5d51ce98 | 5331 | int restudied; |
07be1b83 YO |
5332 | RExC_state_t copyRExC_state; |
5333 | #endif | |
2a5d9b1d | 5334 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d | 5335 | |
3c13cae6 DM |
5336 | PERL_ARGS_ASSERT_RE_OP_COMPILE; |
5337 | ||
6d5c990f | 5338 | DEBUG_r(if (!PL_colorset) reginitcolors()); |
a0d0e21e | 5339 | |
370b8f2f | 5340 | #ifndef PERL_IN_XSUB_RE |
a3e1f3a6 KW |
5341 | /* Initialize these here instead of as-needed, as is quick and avoids |
5342 | * having to test them each time otherwise */ | |
5343 | if (! PL_AboveLatin1) { | |
5344 | PL_AboveLatin1 = _new_invlist_C_array(AboveLatin1_invlist); | |
5345 | PL_ASCII = _new_invlist_C_array(ASCII_invlist); | |
5346 | PL_Latin1 = _new_invlist_C_array(Latin1_invlist); | |
3f427fd9 KW |
5347 | |
5348 | PL_L1PosixAlnum = _new_invlist_C_array(L1PosixAlnum_invlist); | |
5349 | PL_PosixAlnum = _new_invlist_C_array(PosixAlnum_invlist); | |
5350 | ||
5351 | PL_L1PosixAlpha = _new_invlist_C_array(L1PosixAlpha_invlist); | |
5352 | PL_PosixAlpha = _new_invlist_C_array(PosixAlpha_invlist); | |
5353 | ||
5354 | PL_PosixBlank = _new_invlist_C_array(PosixBlank_invlist); | |
5355 | PL_XPosixBlank = _new_invlist_C_array(XPosixBlank_invlist); | |
5356 | ||
dab0c3e7 KW |
5357 | PL_L1Cased = _new_invlist_C_array(L1Cased_invlist); |
5358 | ||
3f427fd9 KW |
5359 | PL_PosixCntrl = _new_invlist_C_array(PosixCntrl_invlist); |
5360 | PL_XPosixCntrl = _new_invlist_C_array(XPosixCntrl_invlist); | |
5361 | ||
5362 | PL_PosixDigit = _new_invlist_C_array(PosixDigit_invlist); | |
5363 | ||
5364 | PL_L1PosixGraph = _new_invlist_C_array(L1PosixGraph_invlist); | |
5365 | PL_PosixGraph = _new_invlist_C_array(PosixGraph_invlist); | |
5366 | ||
3f427fd9 KW |
5367 | PL_L1PosixLower = _new_invlist_C_array(L1PosixLower_invlist); |
5368 | PL_PosixLower = _new_invlist_C_array(PosixLower_invlist); | |
5369 | ||
5370 | PL_L1PosixPrint = _new_invlist_C_array(L1PosixPrint_invlist); | |
5371 | PL_PosixPrint = _new_invlist_C_array(PosixPrint_invlist); | |
5372 | ||
5373 | PL_L1PosixPunct = _new_invlist_C_array(L1PosixPunct_invlist); | |
5374 | PL_PosixPunct = _new_invlist_C_array(PosixPunct_invlist); | |
5375 | ||
5376 | PL_PerlSpace = _new_invlist_C_array(PerlSpace_invlist); | |
5377 | PL_XPerlSpace = _new_invlist_C_array(XPerlSpace_invlist); | |
5378 | ||
5379 | PL_PosixSpace = _new_invlist_C_array(PosixSpace_invlist); | |
5380 | PL_XPosixSpace = _new_invlist_C_array(XPosixSpace_invlist); | |
5381 | ||
5382 | PL_L1PosixUpper = _new_invlist_C_array(L1PosixUpper_invlist); | |
5383 | PL_PosixUpper = _new_invlist_C_array(PosixUpper_invlist); | |
5384 | ||
5385 | PL_VertSpace = _new_invlist_C_array(VertSpace_invlist); | |
5386 | ||
5387 | PL_PosixWord = _new_invlist_C_array(PosixWord_invlist); | |
5388 | PL_L1PosixWord = _new_invlist_C_array(L1PosixWord_invlist); | |
5389 | ||
5390 | PL_PosixXDigit = _new_invlist_C_array(PosixXDigit_invlist); | |
5391 | PL_XPosixXDigit = _new_invlist_C_array(XPosixXDigit_invlist); | |
a3e1f3a6 | 5392 | } |
370b8f2f | 5393 | #endif |
a3e1f3a6 | 5394 | |
b1603ef8 DM |
5395 | pRExC_state->code_blocks = NULL; |
5396 | pRExC_state->num_code_blocks = 0; | |
9f141731 DM |
5397 | |
5398 | if (is_bare_re) | |
76ac488f | 5399 | *is_bare_re = FALSE; |
9f141731 | 5400 | |
b1603ef8 DM |
5401 | if (expr && (expr->op_type == OP_LIST || |
5402 | (expr->op_type == OP_NULL && expr->op_targ == OP_LIST))) { | |
5403 | ||
83dd4485 DM |
5404 | /* is the source UTF8, and how many code blocks are there? */ |
5405 | OP *o; | |
5406 | int ncode = 0; | |
5407 | ||
5408 | for (o = cLISTOPx(expr)->op_first; o; o = o->op_sibling) { | |
5409 | if (o->op_type == OP_CONST && SvUTF8(cSVOPo_sv)) | |
5410 | code_is_utf8 = 1; | |
5411 | else if (o->op_type == OP_NULL && (o->op_flags & OPf_SPECIAL)) | |
5412 | /* count of DO blocks */ | |
5413 | ncode++; | |
5414 | } | |
83dd4485 | 5415 | if (ncode) { |
b1603ef8 | 5416 | pRExC_state->num_code_blocks = ncode; |
3d2bd50a | 5417 | Newx(pRExC_state->code_blocks, ncode, struct reg_code_block); |
83dd4485 DM |
5418 | } |
5419 | } | |
5420 | ||
9f141731 DM |
5421 | if (pat_count) { |
5422 | /* handle a list of SVs */ | |
5423 | ||
5424 | SV **svp; | |
5425 | ||
5426 | /* apply magic and RE overloading to each arg */ | |
5427 | for (svp = patternp; svp < patternp + pat_count; svp++) { | |
5428 | SV *rx = *svp; | |
5429 | SvGETMAGIC(rx); | |
5430 | if (SvROK(rx) && SvAMAGIC(rx)) { | |
5431 | SV *sv = AMG_CALLunary(rx, regexp_amg); | |
5432 | if (sv) { | |
5433 | if (SvROK(sv)) | |
5434 | sv = SvRV(sv); | |
5435 | if (SvTYPE(sv) != SVt_REGEXP) | |
5436 | Perl_croak(aTHX_ "Overloaded qr did not return a REGEXP"); | |
5437 | *svp = sv; | |
5438 | } | |
5439 | } | |
5440 | } | |
5441 | ||
5442 | if (pat_count > 1) { | |
346d3070 DM |
5443 | /* concat multiple args and find any code block indexes */ |
5444 | ||
5445 | OP *o = NULL; | |
5446 | int n = 0; | |
947535e3 | 5447 | bool utf8 = 0; |
e03b874a | 5448 | STRLEN orig_patlen = 0; |
346d3070 DM |
5449 | |
5450 | if (pRExC_state->num_code_blocks) { | |
5451 | o = cLISTOPx(expr)->op_first; | |
5452 | assert(o->op_type == OP_PUSHMARK); | |
5453 | o = o->op_sibling; | |
5454 | } | |
b1603ef8 | 5455 | |
9f141731 DM |
5456 | pat = newSVpvn("", 0); |
5457 | SAVEFREESV(pat); | |
947535e3 DM |
5458 | |
5459 | /* determine if the pattern is going to be utf8 (needed | |
5460 | * in advance to align code block indices correctly). | |
5461 | * XXX This could fail to be detected for an arg with | |
5462 | * overloading but not concat overloading; but the main effect | |
5463 | * in this obscure case is to need a 'use re eval' for a | |
5464 | * literal code block */ | |
5465 | for (svp = patternp; svp < patternp + pat_count; svp++) { | |
5466 | if (SvUTF8(*svp)) | |
5467 | utf8 = 1; | |
5468 | } | |
5469 | if (utf8) | |
5470 | SvUTF8_on(pat); | |
5471 | ||
9f141731 DM |
5472 | for (svp = patternp; svp < patternp + pat_count; svp++) { |
5473 | SV *sv, *msv = *svp; | |
b30fcab9 | 5474 | SV *rx; |
346d3070 DM |
5475 | bool code = 0; |
5476 | if (o) { | |
5477 | if (o->op_type == OP_NULL && (o->op_flags & OPf_SPECIAL)) { | |
b30fcab9 DM |
5478 | assert(n < pRExC_state->num_code_blocks); |
5479 | pRExC_state->code_blocks[n].start = SvCUR(pat); | |
5480 | pRExC_state->code_blocks[n].block = o; | |
5481 | pRExC_state->code_blocks[n].src_regex = NULL; | |
346d3070 | 5482 | n++; |
346d3070 DM |
5483 | code = 1; |
5484 | o = o->op_sibling; /* skip CONST */ | |
5485 | assert(o); | |
5486 | } | |
5487 | o = o->op_sibling;; | |
5488 | } | |
5489 | ||
e03b874a DM |
5490 | if ((SvAMAGIC(pat) || SvAMAGIC(msv)) && |
5491 | (sv = amagic_call(pat, msv, concat_amg, AMGf_assign))) | |
5492 | { | |
5493 | sv_setsv(pat, sv); | |
5494 | /* overloading involved: all bets are off over literal | |
5495 | * code. Pretend we haven't seen it */ | |
5496 | pRExC_state->num_code_blocks -= n; | |
5497 | n = 0; | |
5498 | rx = NULL; | |
5499 | ||
5500 | } | |
5501 | else { | |
5502 | while (SvAMAGIC(msv) | |
5503 | && (sv = AMG_CALLunary(msv, string_amg)) | |
00bf6ed2 DM |
5504 | && sv != msv |
5505 | && !( SvROK(msv) | |
5506 | && SvROK(sv) | |
5507 | && SvRV(msv) == SvRV(sv)) | |
5508 | ) { | |
e03b874a DM |
5509 | msv = sv; |
5510 | SvGETMAGIC(msv); | |
5511 | } | |
5512 | if (SvROK(msv) && SvTYPE(SvRV(msv)) == SVt_REGEXP) | |
5513 | msv = SvRV(msv); | |
5514 | orig_patlen = SvCUR(pat); | |
5515 | sv_catsv_nomg(pat, msv); | |
5516 | rx = msv; | |
5517 | if (code) | |
5518 | pRExC_state->code_blocks[n-1].end = SvCUR(pat)-1; | |
5519 | } | |
5520 | ||
b30fcab9 | 5521 | /* extract any code blocks within any embedded qr//'s */ |
e03b874a | 5522 | if (rx && SvTYPE(rx) == SVt_REGEXP |
3c13cae6 | 5523 | && RX_ENGINE((REGEXP*)rx)->op_comp) |
b30fcab9 DM |
5524 | { |
5525 | ||
5526 | RXi_GET_DECL(((struct regexp*)SvANY(rx)), ri); | |
5527 | if (ri->num_code_blocks) { | |
5528 | int i; | |
629cd4f3 DM |
5529 | /* the presence of an embedded qr// with code means |
5530 | * we should always recompile: the text of the | |
5531 | * qr// may not have changed, but it may be a | |
5532 | * different closure than last time */ | |
5533 | recompile = 1; | |
b30fcab9 DM |
5534 | Renew(pRExC_state->code_blocks, |
5535 | pRExC_state->num_code_blocks + ri->num_code_blocks, | |
5536 | struct reg_code_block); | |
5537 | pRExC_state->num_code_blocks += ri->num_code_blocks; | |
5538 | for (i=0; i < ri->num_code_blocks; i++) { | |
5539 | struct reg_code_block *src, *dst; | |
e03b874a | 5540 | STRLEN offset = orig_patlen |
b30fcab9 DM |
5541 | + ((struct regexp *)SvANY(rx))->pre_prefix; |
5542 | assert(n < pRExC_state->num_code_blocks); | |
5543 | src = &ri->code_blocks[i]; | |
5544 | dst = &pRExC_state->code_blocks[n]; | |
5545 | dst->start = src->start + offset; | |
5546 | dst->end = src->end + offset; | |
5547 | dst->block = src->block; | |
5548 | dst->src_regex = (REGEXP*) SvREFCNT_inc( (SV*) | |
5549 | src->src_regex | |
5550 | ? src->src_regex | |
5551 | : (REGEXP*)rx); | |
5552 | n++; | |
5553 | } | |
5554 | } | |
5555 | } | |
9f141731 DM |
5556 | } |
5557 | SvSETMAGIC(pat); | |
5558 | } | |
e03b874a DM |
5559 | else { |
5560 | SV *sv; | |
9f141731 | 5561 | pat = *patternp; |
e03b874a DM |
5562 | while (SvAMAGIC(pat) |
5563 | && (sv = AMG_CALLunary(pat, string_amg)) | |
5564 | && sv != pat) | |
5565 | { | |
5566 | pat = sv; | |
5567 | SvGETMAGIC(pat); | |
5568 | } | |
5569 | } | |
9f141731 DM |
5570 | |
5571 | /* handle bare regex: foo =~ $re */ | |
5572 | { | |
5573 | SV *re = pat; | |
5574 | if (SvROK(re)) | |
5575 | re = SvRV(re); | |
5576 | if (SvTYPE(re) == SVt_REGEXP) { | |
5577 | if (is_bare_re) | |
76ac488f | 5578 | *is_bare_re = TRUE; |
9f141731 | 5579 | SvREFCNT_inc(re); |
3d2bd50a | 5580 | Safefree(pRExC_state->code_blocks); |
9f141731 DM |
5581 | return (REGEXP*)re; |
5582 | } | |
5583 | } | |
5584 | } | |
5585 | else { | |
5586 | /* not a list of SVs, so must be a list of OPs */ | |
5587 | assert(expr); | |
68e2671b | 5588 | if (expr->op_type == OP_LIST) { |
1eacd84c DM |
5589 | int i = -1; |
5590 | bool is_code = 0; | |
5591 | OP *o; | |
5592 | ||
68e2671b DM |
5593 | pat = newSVpvn("", 0); |
5594 | SAVEFREESV(pat); | |
83dd4485 | 5595 | if (code_is_utf8) |
68e2671b | 5596 | SvUTF8_on(pat); |
1eacd84c DM |
5597 | |
5598 | /* given a list of CONSTs and DO blocks in expr, append all | |
5599 | * the CONSTs to pat, and record the start and end of each | |
5600 | * code block in code_blocks[] (each DO{} op is followed by an | |
5601 | * OP_CONST containing the corresponding literal '(?{...}) | |
5602 | * text) | |
5603 | */ | |
5604 | for (o = cLISTOPx(expr)->op_first; o; o = o->op_sibling) { | |
5605 | if (o->op_type == OP_CONST) { | |
5606 | sv_catsv(pat, cSVOPo_sv); | |
5607 | if (is_code) { | |
5608 | pRExC_state->code_blocks[i].end = SvCUR(pat)-1; | |
5609 | is_code = 0; | |
5610 | } | |
5611 | } | |
5612 | else if (o->op_type == OP_NULL && (o->op_flags & OPf_SPECIAL)) { | |
5613 | assert(i+1 < pRExC_state->num_code_blocks); | |
5614 | pRExC_state->code_blocks[++i].start = SvCUR(pat); | |
5615 | pRExC_state->code_blocks[i].block = o; | |
b30fcab9 | 5616 | pRExC_state->code_blocks[i].src_regex = NULL; |
1eacd84c DM |
5617 | is_code = 1; |
5618 | } | |
5619 | } | |
68e2671b DM |
5620 | } |
5621 | else { | |
5622 | assert(expr->op_type == OP_CONST); | |
5623 | pat = cSVOPx_sv(expr); | |
74529a43 | 5624 | } |
74529a43 | 5625 | } |
74529a43 | 5626 | |
9f141731 | 5627 | exp = SvPV_nomg(pat, plen); |
1b34bc43 | 5628 | |
3c13cae6 | 5629 | if (!eng->op_comp) { |
9f141731 DM |
5630 | if ((SvUTF8(pat) && IN_BYTES) |
5631 | || SvGMAGICAL(pat) || SvAMAGIC(pat)) | |
5632 | { | |
5633 | /* make a temporary copy; either to convert to bytes, | |
5634 | * or to avoid repeating get-magic / overloaded stringify */ | |
5635 | pat = newSVpvn_flags(exp, plen, SVs_TEMP | | |
5636 | (IN_BYTES ? 0 : SvUTF8(pat))); | |
5637 | } | |
3d2bd50a | 5638 | Safefree(pRExC_state->code_blocks); |
37acfcba | 5639 | return CALLREGCOMP_ENG(eng, pat, orig_rx_flags); |
1b34bc43 | 5640 | } |
9f141731 | 5641 | |
9f141731 DM |
5642 | /* ignore the utf8ness if the pattern is 0 length */ |
5643 | RExC_utf8 = RExC_orig_utf8 = (plen == 0 || IN_BYTES) ? 0 : SvUTF8(pat); | |
e40e74fe | 5644 | RExC_uni_semantics = 0; |
4624b182 | 5645 | RExC_contains_locale = 0; |
d24ca0c5 | 5646 | pRExC_state->runtime_code_qr = NULL; |
7b597bb8 | 5647 | |
d6bd454d | 5648 | /****************** LONG JUMP TARGET HERE***********************/ |
bbd61b5f KW |
5649 | /* Longjmp back to here if have to switch in midstream to utf8 */ |
5650 | if (! RExC_orig_utf8) { | |
5651 | JMPENV_PUSH(jump_ret); | |
29b09c41 | 5652 | used_setjump = TRUE; |
bbd61b5f KW |
5653 | } |
5654 | ||
5d51ce98 | 5655 | if (jump_ret == 0) { /* First time through */ |
29b09c41 | 5656 | xend = exp + plen; |
29b09c41 | 5657 | |
5d51ce98 KW |
5658 | DEBUG_COMPILE_r({ |
5659 | SV *dsv= sv_newmortal(); | |
5660 | RE_PV_QUOTED_DECL(s, RExC_utf8, | |
5661 | dsv, exp, plen, 60); | |
5662 | PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n", | |
5663 | PL_colors[4],PL_colors[5],s); | |
5664 | }); | |
5665 | } | |
5666 | else { /* longjumped back */ | |
2bd8e0da DM |
5667 | U8 *src, *dst; |
5668 | int n=0; | |
5669 | STRLEN s = 0, d = 0; | |
5670 | bool do_end = 0; | |
bbd61b5f | 5671 | |
5d51ce98 KW |
5672 | /* If the cause for the longjmp was other than changing to utf8, pop |
5673 | * our own setjmp, and longjmp to the correct handler */ | |
bbd61b5f KW |
5674 | if (jump_ret != UTF8_LONGJMP) { |
5675 | JMPENV_POP; | |
5676 | JMPENV_JUMP(jump_ret); | |
5677 | } | |
5678 | ||
595598ee KW |
5679 | GET_RE_DEBUG_FLAGS; |
5680 | ||
bbd61b5f KW |
5681 | /* It's possible to write a regexp in ascii that represents Unicode |
5682 | codepoints outside of the byte range, such as via \x{100}. If we | |
5683 | detect such a sequence we have to convert the entire pattern to utf8 | |
5684 | and then recompile, as our sizing calculation will have been based | |
5685 | on 1 byte == 1 character, but we will need to use utf8 to encode | |
5686 | at least some part of the pattern, and therefore must convert the whole | |
5687 | thing. | |
5688 | -- dmq */ | |
5689 | DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, | |
5690 | "UTF8 mismatch! Converting to utf8 for resizing and compile\n")); | |
68e2671b | 5691 | |
2bd8e0da DM |
5692 | /* upgrade pattern to UTF8, and if there are code blocks, |
5693 | * recalculate the indices. | |
5694 | * This is essentially an unrolled Perl_bytes_to_utf8() */ | |
5695 | ||
5696 | src = (U8*)SvPV_nomg(pat, plen); | |
5697 | Newx(dst, plen * 2 + 1, U8); | |
5698 | ||
5699 | while (s < plen) { | |
5700 | const UV uv = NATIVE_TO_ASCII(src[s]); | |
5701 | if (UNI_IS_INVARIANT(uv)) | |
5702 | dst[d] = (U8)UTF_TO_NATIVE(uv); | |
5703 | else { | |
5704 | dst[d++] = (U8)UTF8_EIGHT_BIT_HI(uv); | |
5705 | dst[d] = (U8)UTF8_EIGHT_BIT_LO(uv); | |
5706 | } | |
5707 | if (n < pRExC_state->num_code_blocks) { | |
5708 | if (!do_end && pRExC_state->code_blocks[n].start == s) { | |
5709 | pRExC_state->code_blocks[n].start = d; | |
5710 | assert(dst[d] == '('); | |
5711 | do_end = 1; | |
5712 | } | |
5713 | else if (do_end && pRExC_state->code_blocks[n].end == s) { | |
5714 | pRExC_state->code_blocks[n].end = d; | |
5715 | assert(dst[d] == ')'); | |
5716 | do_end = 0; | |
5717 | n++; | |
5718 | } | |
5719 | } | |
5720 | s++; | |
5721 | d++; | |
68e2671b | 5722 | } |
2bd8e0da DM |
5723 | dst[d] = '\0'; |
5724 | plen = d; | |
5725 | exp = (char*) dst; | |
5726 | xend = exp + plen; | |
5727 | SAVEFREEPV(exp); | |
68e2671b | 5728 | RExC_orig_utf8 = RExC_utf8 = 1; |
3b16d10d | 5729 | } |
6ae44cd2 | 5730 | |
3b16d10d | 5731 | /* return old regex if pattern hasn't changed */ |
6ae44cd2 | 5732 | |
3b16d10d | 5733 | if ( old_re |
629cd4f3 | 5734 | && !recompile |
3b16d10d DM |
5735 | && !!RX_UTF8(old_re) == !!RExC_utf8 |
5736 | && RX_PRECOMP(old_re) | |
5737 | && RX_PRELEN(old_re) == plen | |
5738 | && memEQ(RX_PRECOMP(old_re), exp, plen)) | |
5739 | { | |
629cd4f3 | 5740 | /* with runtime code, always recompile */ |
d24ca0c5 DM |
5741 | runtime_code = S_has_runtime_code(aTHX_ pRExC_state, expr, pm_flags, |
5742 | exp, plen); | |
5743 | if (!runtime_code) { | |
629cd4f3 DM |
5744 | if (used_setjump) { |
5745 | JMPENV_POP; | |
5746 | } | |
5747 | Safefree(pRExC_state->code_blocks); | |
5748 | return old_re; | |
3b16d10d | 5749 | } |
bbd61b5f | 5750 | } |
d24ca0c5 DM |
5751 | else if ((pm_flags & PMf_USE_RE_EVAL) |
5752 | /* this second condition covers the non-regex literal case, | |
5753 | * i.e. $foo =~ '(?{})'. */ | |
5754 | || ( !PL_reg_state.re_reparsing && IN_PERL_COMPILETIME | |
5755 | && (PL_hints & HINT_RE_EVAL)) | |
5756 | ) | |
5757 | runtime_code = S_has_runtime_code(aTHX_ pRExC_state, expr, pm_flags, | |
5758 | exp, plen); | |
bbd61b5f | 5759 | |
5d51ce98 KW |
5760 | #ifdef TRIE_STUDY_OPT |
5761 | restudied = 0; | |
5762 | #endif | |
5763 | ||
37acfcba | 5764 | rx_flags = orig_rx_flags; |
a62b1201 | 5765 | |
4624b182 KW |
5766 | if (initial_charset == REGEX_LOCALE_CHARSET) { |
5767 | RExC_contains_locale = 1; | |
5768 | } | |
5769 | else if (RExC_utf8 && initial_charset == REGEX_DEPENDS_CHARSET) { | |
5770 | ||
5771 | /* Set to use unicode semantics if the pattern is in utf8 and has the | |
5772 | * 'depends' charset specified, as it means unicode when utf8 */ | |
37acfcba | 5773 | set_regex_charset(&rx_flags, REGEX_UNICODE_CHARSET); |
29b09c41 KW |
5774 | } |
5775 | ||
02daf0ab | 5776 | RExC_precomp = exp; |
37acfcba | 5777 | RExC_flags = rx_flags; |
514a91f1 | 5778 | RExC_pm_flags = pm_flags; |
d24ca0c5 DM |
5779 | |
5780 | if (runtime_code) { | |
5781 | if (PL_tainting && PL_tainted) | |
5782 | Perl_croak(aTHX_ "Eval-group in insecure regular expression"); | |
5783 | ||
5784 | if (!S_compile_runtime_code(aTHX_ pRExC_state, exp, plen)) { | |
5785 | /* whoops, we have a non-utf8 pattern, whilst run-time code | |
5786 | * got compiled as utf8. Try again with a utf8 pattern */ | |
5787 | JMPENV_JUMP(UTF8_LONGJMP); | |
5788 | } | |
5789 | } | |
5790 | assert(!pRExC_state->runtime_code_qr); | |
5791 | ||
830247a4 | 5792 | RExC_sawback = 0; |
bbce6d69 | 5793 | |
830247a4 | 5794 | RExC_seen = 0; |
b57e4118 | 5795 | RExC_in_lookbehind = 0; |
830247a4 | 5796 | RExC_seen_zerolen = *exp == '^' ? -1 : 0; |
830247a4 | 5797 | RExC_extralen = 0; |
e2a7e165 | 5798 | RExC_override_recoding = 0; |
c277df42 | 5799 | |
bbce6d69 | 5800 | /* First pass: determine size, legality. */ |
830247a4 | 5801 | RExC_parse = exp; |
fac92740 | 5802 | RExC_start = exp; |
830247a4 IZ |
5803 | RExC_end = xend; |
5804 | RExC_naughty = 0; | |
5805 | RExC_npar = 1; | |
e2e6a0f1 | 5806 | RExC_nestroot = 0; |
830247a4 IZ |
5807 | RExC_size = 0L; |
5808 | RExC_emit = &PL_regdummy; | |
5809 | RExC_whilem_seen = 0; | |
40d049e4 YO |
5810 | RExC_open_parens = NULL; |
5811 | RExC_close_parens = NULL; | |
5812 | RExC_opend = NULL; | |
81714fb9 | 5813 | RExC_paren_names = NULL; |
1f1031fe YO |
5814 | #ifdef DEBUGGING |
5815 | RExC_paren_name_list = NULL; | |
5816 | #endif | |
40d049e4 YO |
5817 | RExC_recurse = NULL; |
5818 | RExC_recurse_count = 0; | |
b1603ef8 | 5819 | pRExC_state->code_index = 0; |
81714fb9 | 5820 | |
85ddcde9 JH |
5821 | #if 0 /* REGC() is (currently) a NOP at the first pass. |
5822 | * Clever compilers notice this and complain. --jhi */ | |
830247a4 | 5823 | REGC((U8)REG_MAGIC, (char*)RExC_emit); |
85ddcde9 | 5824 | #endif |
5a415bbc KW |
5825 | DEBUG_PARSE_r( |
5826 | PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"); | |
5827 | RExC_lastnum=0; | |
5828 | RExC_lastparse=NULL; | |
5829 | ); | |
3dab1dad | 5830 | if (reg(pRExC_state, 0, &flags,1) == NULL) { |
c445ea15 | 5831 | RExC_precomp = NULL; |
3d2bd50a | 5832 | Safefree(pRExC_state->code_blocks); |
a0d0e21e LW |
5833 | return(NULL); |
5834 | } | |
bbd61b5f | 5835 | |
29b09c41 KW |
5836 | /* Here, finished first pass. Get rid of any added setjmp */ |
5837 | if (used_setjump) { | |
bbd61b5f | 5838 | JMPENV_POP; |
02daf0ab | 5839 | } |
e40e74fe | 5840 | |
07be1b83 | 5841 | DEBUG_PARSE_r({ |
81714fb9 YO |
5842 | PerlIO_printf(Perl_debug_log, |
5843 | "Required size %"IVdf" nodes\n" | |
5844 | "Starting second pass (creation)\n", | |
5845 | (IV)RExC_size); | |
07be1b83 YO |
5846 | RExC_lastnum=0; |
5847 | RExC_lastparse=NULL; | |
5848 | }); | |
e40e74fe KW |
5849 | |
5850 | /* The first pass could have found things that force Unicode semantics */ | |
5851 | if ((RExC_utf8 || RExC_uni_semantics) | |
37acfcba | 5852 | && get_regex_charset(rx_flags) == REGEX_DEPENDS_CHARSET) |
e40e74fe | 5853 | { |
37acfcba | 5854 | set_regex_charset(&rx_flags, REGEX_UNICODE_CHARSET); |
e40e74fe KW |
5855 | } |
5856 | ||
c277df42 IZ |
5857 | /* Small enough for pointer-storage convention? |
5858 | If extralen==0, this means that we will not need long jumps. */ | |
830247a4 IZ |
5859 | if (RExC_size >= 0x10000L && RExC_extralen) |
5860 | RExC_size += RExC_extralen; | |
c277df42 | 5861 | else |
830247a4 IZ |
5862 | RExC_extralen = 0; |
5863 | if (RExC_whilem_seen > 15) | |
5864 | RExC_whilem_seen = 15; | |
a0d0e21e | 5865 | |
f9f4320a YO |
5866 | /* Allocate space and zero-initialize. Note, the two step process |
5867 | of zeroing when in debug mode, thus anything assigned has to | |
5868 | happen after that */ | |
d2f13c59 | 5869 | rx = (REGEXP*) newSV_type(SVt_REGEXP); |
288b8c02 | 5870 | r = (struct regexp*)SvANY(rx); |
f8fc2ecf YO |
5871 | Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), |
5872 | char, regexp_internal); | |
5873 | if ( r == NULL || ri == NULL ) | |
b45f050a | 5874 | FAIL("Regexp out of space"); |
0f79a09d GS |
5875 | #ifdef DEBUGGING |
5876 | /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */ | |
f8fc2ecf | 5877 | Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char); |
58e23c8d | 5878 | #else |
f8fc2ecf YO |
5879 | /* bulk initialize base fields with 0. */ |
5880 | Zero(ri, sizeof(regexp_internal), char); | |
0f79a09d | 5881 | #endif |
58e23c8d YO |
5882 | |
5883 | /* non-zero initialization begins here */ | |
f8fc2ecf | 5884 | RXi_SET( r, ri ); |
3c13cae6 | 5885 | r->engine= eng; |
37acfcba | 5886 | r->extflags = rx_flags; |
514a91f1 | 5887 | if (pm_flags & PMf_IS_QR) { |
3d2bd50a DM |
5888 | ri->code_blocks = pRExC_state->code_blocks; |
5889 | ri->num_code_blocks = pRExC_state->num_code_blocks; | |
5890 | } | |
5891 | else | |
5892 | SAVEFREEPV(pRExC_state->code_blocks); | |
5893 | ||
bcdf7404 | 5894 | { |
f7819f85 | 5895 | bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY); |
a62b1201 | 5896 | bool has_charset = (get_regex_charset(r->extflags) != REGEX_DEPENDS_CHARSET); |
c5ea2ffa KW |
5897 | |
5898 | /* The caret is output if there are any defaults: if not all the STD | |
5899 | * flags are set, or if no character set specifier is needed */ | |
5900 | bool has_default = | |
5901 | (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD) | |
5902 | || ! has_charset); | |
bcdf7404 | 5903 | bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT); |
14f3b9f2 NC |
5904 | U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) |
5905 | >> RXf_PMf_STD_PMMOD_SHIFT); | |
bcdf7404 YO |
5906 | const char *fptr = STD_PAT_MODS; /*"msix"*/ |
5907 | char *p; | |
fb85c044 | 5908 | /* Allocate for the worst case, which is all the std flags are turned |
c5ea2ffa KW |
5909 | * on. If more precision is desired, we could do a population count of |
5910 | * the flags set. This could be done with a small lookup table, or by | |
5911 | * shifting, masking and adding, or even, when available, assembly | |
5912 | * language for a machine-language population count. | |
5913 | * We never output a minus, as all those are defaults, so are | |
5914 | * covered by the caret */ | |
fb85c044 | 5915 | const STRLEN wraplen = plen + has_p + has_runon |
c5ea2ffa | 5916 | + has_default /* If needs a caret */ |
a62b1201 KW |
5917 | |
5918 | /* If needs a character set specifier */ | |
5919 | + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0) | |
bcdf7404 YO |
5920 | + (sizeof(STD_PAT_MODS) - 1) |
5921 | + (sizeof("(?:)") - 1); | |
5922 | ||
c5ea2ffa | 5923 | p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */ |
f7c278bf | 5924 | SvPOK_on(rx); |
9f141731 DM |
5925 | if (RExC_utf8) |
5926 | SvFLAGS(rx) |= SVf_UTF8; | |
bcdf7404 | 5927 | *p++='('; *p++='?'; |
9de15fec KW |
5928 | |
5929 | /* If a default, cover it using the caret */ | |
c5ea2ffa | 5930 | if (has_default) { |
85508812 | 5931 | *p++= DEFAULT_PAT_MOD; |
fb85c044 | 5932 | } |
c5ea2ffa | 5933 | if (has_charset) { |
a62b1201 KW |
5934 | STRLEN len; |
5935 | const char* const name = get_regex_charset_name(r->extflags, &len); | |
5936 | Copy(name, p, len, char); | |
5937 | p += len; | |
9de15fec | 5938 | } |
f7819f85 A |
5939 | if (has_p) |
5940 | *p++ = KEEPCOPY_PAT_MOD; /*'p'*/ | |
bcdf7404 | 5941 | { |
bcdf7404 | 5942 | char ch; |
bcdf7404 YO |
5943 | while((ch = *fptr++)) { |
5944 | if(reganch & 1) | |
5945 | *p++ = ch; | |
bcdf7404 YO |
5946 | reganch >>= 1; |
5947 | } | |
bcdf7404 YO |
5948 | } |
5949 | ||
28d8d7f4 | 5950 | *p++ = ':'; |
bb661a58 | 5951 | Copy(RExC_precomp, p, plen, char); |
efd26800 NC |
5952 | assert ((RX_WRAPPED(rx) - p) < 16); |
5953 | r->pre_prefix = p - RX_WRAPPED(rx); | |
bb661a58 | 5954 | p += plen; |
bcdf7404 | 5955 | if (has_runon) |
28d8d7f4 YO |
5956 | *p++ = '\n'; |
5957 | *p++ = ')'; | |
5958 | *p = 0; | |
fb85c044 | 5959 | SvCUR_set(rx, p - SvPVX_const(rx)); |
bcdf7404 YO |
5960 | } |
5961 | ||
bbe252da | 5962 | r->intflags = 0; |
830247a4 | 5963 | r->nparens = RExC_npar - 1; /* set early to validate backrefs */ |
81714fb9 | 5964 | |
6bda09f9 | 5965 | if (RExC_seen & REG_SEEN_RECURSE) { |
40d049e4 YO |
5966 | Newxz(RExC_open_parens, RExC_npar,regnode *); |
5967 | SAVEFREEPV(RExC_open_parens); | |
5968 | Newxz(RExC_close_parens,RExC_npar,regnode *); | |
5969 | SAVEFREEPV(RExC_close_parens); | |
6bda09f9 YO |
5970 | } |
5971 | ||
5972 | /* Useful during FAIL. */ | |
7122b237 YO |
5973 | #ifdef RE_TRACK_PATTERN_OFFSETS |
5974 | Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */ | |
a3621e74 | 5975 | DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log, |
2af232bd | 5976 | "%s %"UVuf" bytes for offset annotations.\n", |
7122b237 | 5977 | ri->u.offsets ? "Got" : "Couldn't get", |
392fbf5d | 5978 | (UV)((2*RExC_size+1) * sizeof(U32)))); |
7122b237 YO |
5979 | #endif |
5980 | SetProgLen(ri,RExC_size); | |
288b8c02 | 5981 | RExC_rx_sv = rx; |
830247a4 | 5982 | RExC_rx = r; |
f8fc2ecf | 5983 | RExC_rxi = ri; |
bbce6d69 | 5984 | |
5985 | /* Second pass: emit code. */ | |
37acfcba | 5986 | RExC_flags = rx_flags; /* don't let top level (?i) bleed */ |
514a91f1 | 5987 | RExC_pm_flags = pm_flags; |
830247a4 IZ |
5988 | RExC_parse = exp; |
5989 | RExC_end = xend; | |
5990 | RExC_naughty = 0; | |
5991 | RExC_npar = 1; | |
f8fc2ecf YO |
5992 | RExC_emit_start = ri->program; |
5993 | RExC_emit = ri->program; | |
3b57cd43 | 5994 | RExC_emit_bound = ri->program + RExC_size + 1; |
68e2671b | 5995 | pRExC_state->code_index = 0; |
3b57cd43 | 5996 | |
830247a4 | 5997 | REGC((U8)REG_MAGIC, (char*) RExC_emit++); |
80757612 | 5998 | if (reg(pRExC_state, 0, &flags,1) == NULL) { |
288b8c02 | 5999 | ReREFCNT_dec(rx); |
a0d0e21e | 6000 | return(NULL); |
80757612 | 6001 | } |
07be1b83 YO |
6002 | /* XXXX To minimize changes to RE engine we always allocate |
6003 | 3-units-long substrs field. */ | |
6004 | Newx(r->substrs, 1, struct reg_substr_data); | |
40d049e4 YO |
6005 | if (RExC_recurse_count) { |
6006 | Newxz(RExC_recurse,RExC_recurse_count,regnode *); | |
6007 | SAVEFREEPV(RExC_recurse); | |
6008 | } | |
a0d0e21e | 6009 | |
07be1b83 | 6010 | reStudy: |
e7f38d0f | 6011 | r->minlen = minlen = sawlookahead = sawplus = sawopen = 0; |
07be1b83 | 6012 | Zero(r->substrs, 1, struct reg_substr_data); |
a3621e74 | 6013 | |
07be1b83 | 6014 | #ifdef TRIE_STUDY_OPT |
0934c9d9 SH |
6015 | if (!restudied) { |
6016 | StructCopy(&zero_scan_data, &data, scan_data_t); | |
6017 | copyRExC_state = RExC_state; | |
6018 | } else { | |
5d458dd8 | 6019 | U32 seen=RExC_seen; |
07be1b83 | 6020 | DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n")); |
5d458dd8 YO |
6021 | |
6022 | RExC_state = copyRExC_state; | |
6023 | if (seen & REG_TOP_LEVEL_BRANCHES) | |
6024 | RExC_seen |= REG_TOP_LEVEL_BRANCHES; | |
6025 | else | |
6026 | RExC_seen &= ~REG_TOP_LEVEL_BRANCHES; | |
1de06328 | 6027 | if (data.last_found) { |
07be1b83 | 6028 | SvREFCNT_dec(data.longest_fixed); |
07be1b83 | 6029 | SvREFCNT_dec(data.longest_float); |
07be1b83 | 6030 | SvREFCNT_dec(data.last_found); |
1de06328 | 6031 | } |
40d049e4 | 6032 | StructCopy(&zero_scan_data, &data, scan_data_t); |
07be1b83 | 6033 | } |
40d049e4 YO |
6034 | #else |
6035 | StructCopy(&zero_scan_data, &data, scan_data_t); | |
07be1b83 | 6036 | #endif |
fc8cd66c | 6037 | |
a0d0e21e | 6038 | /* Dig out information for optimizations. */ |
f7819f85 | 6039 | r->extflags = RExC_flags; /* was pm_op */ |
c737faaf YO |
6040 | /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */ |
6041 | ||
a0ed51b3 | 6042 | if (UTF) |
8f6ae13c | 6043 | SvUTF8_on(rx); /* Unicode in it? */ |
f8fc2ecf | 6044 | ri->regstclass = NULL; |
830247a4 | 6045 | if (RExC_naughty >= 10) /* Probably an expensive pattern. */ |
bbe252da | 6046 | r->intflags |= PREGf_NAUGHTY; |
f8fc2ecf | 6047 | scan = ri->program + 1; /* First BRANCH. */ |
2779dcf1 | 6048 | |
1de06328 YO |
6049 | /* testing for BRANCH here tells us whether there is "must appear" |
6050 | data in the pattern. If there is then we can use it for optimisations */ | |
eaf3ca90 | 6051 | if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */ |
c277df42 | 6052 | I32 fake; |
c5254dd6 | 6053 | STRLEN longest_float_length, longest_fixed_length; |
07be1b83 | 6054 | struct regnode_charclass_class ch_class; /* pointed to by data */ |
653099ff | 6055 | int stclass_flag; |
07be1b83 | 6056 | I32 last_close = 0; /* pointed to by data */ |
5339e136 YO |
6057 | regnode *first= scan; |
6058 | regnode *first_next= regnext(first); | |
639081d6 YO |
6059 | /* |
6060 | * Skip introductions and multiplicators >= 1 | |
6061 | * so that we can extract the 'meat' of the pattern that must | |
6062 | * match in the large if() sequence following. | |
6063 | * NOTE that EXACT is NOT covered here, as it is normally | |
6064 | * picked up by the optimiser separately. | |
6065 | * | |
6066 | * This is unfortunate as the optimiser isnt handling lookahead | |
6067 | * properly currently. | |
6068 | * | |
6069 | */ | |
a0d0e21e | 6070 | while ((OP(first) == OPEN && (sawopen = 1)) || |
653099ff | 6071 | /* An OR of *one* alternative - should not happen now. */ |
5339e136 | 6072 | (OP(first) == BRANCH && OP(first_next) != BRANCH) || |
07be1b83 | 6073 | /* for now we can't handle lookbehind IFMATCH*/ |
e7f38d0f | 6074 | (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) || |
a0d0e21e LW |
6075 | (OP(first) == PLUS) || |
6076 | (OP(first) == MINMOD) || | |
653099ff | 6077 | /* An {n,m} with n>0 */ |
5339e136 YO |
6078 | (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) || |
6079 | (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END )) | |
07be1b83 | 6080 | { |
639081d6 YO |
6081 | /* |
6082 | * the only op that could be a regnode is PLUS, all the rest | |
6083 | * will be regnode_1 or regnode_2. | |
6084 | * | |
6085 | */ | |
a0d0e21e LW |
6086 | if (OP(first) == PLUS) |
6087 | sawplus = 1; | |
6088 | else | |
3dab1dad | 6089 | first += regarglen[OP(first)]; |
686b73d4 | 6090 | |
639081d6 | 6091 | first = NEXTOPER(first); |
5339e136 | 6092 | first_next= regnext(first); |
a687059c LW |
6093 | } |
6094 | ||
a0d0e21e LW |
6095 | /* Starting-point info. */ |
6096 | again: | |
786e8c11 | 6097 | DEBUG_PEEP("first:",first,0); |
07be1b83 | 6098 | /* Ignore EXACT as we deal with it later. */ |
3dab1dad | 6099 | if (PL_regkind[OP(first)] == EXACT) { |
1aa99e6b | 6100 | if (OP(first) == EXACT) |
6f207bd3 | 6101 | NOOP; /* Empty, get anchored substr later. */ |
e5fbd0ff | 6102 | else |
f8fc2ecf | 6103 | ri->regstclass = first; |
b3c9acc1 | 6104 | } |
686b73d4 | 6105 | #ifdef TRIE_STCLASS |
786e8c11 | 6106 | else if (PL_regkind[OP(first)] == TRIE && |
f8fc2ecf | 6107 | ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0) |
07be1b83 | 6108 | { |
786e8c11 | 6109 | regnode *trie_op; |
07be1b83 | 6110 | /* this can happen only on restudy */ |
786e8c11 | 6111 | if ( OP(first) == TRIE ) { |
c944940b | 6112 | struct regnode_1 *trieop = (struct regnode_1 *) |
446bd890 | 6113 | PerlMemShared_calloc(1, sizeof(struct regnode_1)); |
786e8c11 YO |
6114 | StructCopy(first,trieop,struct regnode_1); |
6115 | trie_op=(regnode *)trieop; | |
6116 | } else { | |
c944940b | 6117 | struct regnode_charclass *trieop = (struct regnode_charclass *) |
446bd890 | 6118 | PerlMemShared_calloc(1, sizeof(struct regnode_charclass)); |
786e8c11 YO |
6119 | StructCopy(first,trieop,struct regnode_charclass); |
6120 | trie_op=(regnode *)trieop; | |
6121 | } | |
1de06328 | 6122 | OP(trie_op)+=2; |
786e8c11 | 6123 | make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0); |
f8fc2ecf | 6124 | ri->regstclass = trie_op; |
07be1b83 | 6125 | } |
686b73d4 | 6126 | #endif |
e52fc539 | 6127 | else if (REGNODE_SIMPLE(OP(first))) |
f8fc2ecf | 6128 | ri->regstclass = first; |
3dab1dad YO |
6129 | else if (PL_regkind[OP(first)] == BOUND || |
6130 | PL_regkind[OP(first)] == NBOUND) | |
f8fc2ecf | 6131 | ri->regstclass = first; |
3dab1dad | 6132 | else if (PL_regkind[OP(first)] == BOL) { |
bbe252da YO |
6133 | r->extflags |= (OP(first) == MBOL |
6134 | ? RXf_ANCH_MBOL | |
cad2e5aa | 6135 | : (OP(first) == SBOL |
bbe252da YO |
6136 | ? RXf_ANCH_SBOL |
6137 | : RXf_ANCH_BOL)); | |
a0d0e21e | 6138 | first = NEXTOPER(first); |
774d564b | 6139 | goto again; |
6140 | } | |
6141 | else if (OP(first) == GPOS) { | |
bbe252da | 6142 | r->extflags |= RXf_ANCH_GPOS; |
774d564b | 6143 | first = NEXTOPER(first); |
6144 | goto again; | |
a0d0e21e | 6145 | } |
cf2a2b69 YO |
6146 | else if ((!sawopen || !RExC_sawback) && |
6147 | (OP(first) == STAR && | |
3dab1dad | 6148 | PL_regkind[OP(NEXTOPER(first))] == REG_ANY) && |
471f5387 | 6149 | !(r->extflags & RXf_ANCH) && !pRExC_state->num_code_blocks) |
a0d0e21e LW |
6150 | { |
6151 | /* turn .* into ^.* with an implied $*=1 */ | |
1df70142 AL |
6152 | const int type = |
6153 | (OP(NEXTOPER(first)) == REG_ANY) | |
bbe252da YO |
6154 | ? RXf_ANCH_MBOL |
6155 | : RXf_ANCH_SBOL; | |
6156 | r->extflags |= type; | |
6157 | r->intflags |= PREGf_IMPLICIT; | |
a0d0e21e | 6158 | first = NEXTOPER(first); |
774d564b | 6159 | goto again; |
a0d0e21e | 6160 | } |
e7f38d0f | 6161 | if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback) |
471f5387 | 6162 | && !pRExC_state->num_code_blocks) /* May examine pos and $& */ |
cad2e5aa | 6163 | /* x+ must match at the 1st pos of run of x's */ |
bbe252da | 6164 | r->intflags |= PREGf_SKIP; |
a0d0e21e | 6165 | |
c277df42 | 6166 | /* Scan is after the zeroth branch, first is atomic matcher. */ |
be8e71aa | 6167 | #ifdef TRIE_STUDY_OPT |
81714fb9 | 6168 | DEBUG_PARSE_r( |
be8e71aa YO |
6169 | if (!restudied) |
6170 | PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n", | |
6171 | (IV)(first - scan + 1)) | |
6172 | ); | |
6173 | #else | |
81714fb9 | 6174 | DEBUG_PARSE_r( |
be8e71aa YO |
6175 | PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n", |
6176 | (IV)(first - scan + 1)) | |
6177 | ); | |
6178 | #endif | |
6179 | ||
6180 | ||
a0d0e21e LW |
6181 | /* |
6182 | * If there's something expensive in the r.e., find the | |
6183 | * longest literal string that must appear and make it the | |
6184 | * regmust. Resolve ties in favor of later strings, since | |
6185 | * the regstart check works with the beginning of the r.e. | |
6186 | * and avoiding duplication strengthens checking. Not a | |
6187 | * strong reason, but sufficient in the absence of others. | |
6188 | * [Now we resolve ties in favor of the earlier string if | |
c277df42 | 6189 | * it happens that c_offset_min has been invalidated, since the |
a0d0e21e LW |
6190 | * earlier string may buy us something the later one won't.] |
6191 | */ | |
686b73d4 | 6192 | |
396482e1 GA |
6193 | data.longest_fixed = newSVpvs(""); |
6194 | data.longest_float = newSVpvs(""); | |
6195 | data.last_found = newSVpvs(""); | |
c277df42 IZ |
6196 | data.longest = &(data.longest_fixed); |
6197 | first = scan; | |
f8fc2ecf | 6198 | if (!ri->regstclass) { |
e755fd73 | 6199 | cl_init(pRExC_state, &ch_class); |
653099ff GS |
6200 | data.start_class = &ch_class; |
6201 | stclass_flag = SCF_DO_STCLASS_AND; | |
6202 | } else /* XXXX Check for BOUND? */ | |
6203 | stclass_flag = 0; | |
cb434fcc | 6204 | data.last_closep = &last_close; |
de8c5301 | 6205 | |
1de06328 | 6206 | minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */ |
40d049e4 YO |
6207 | &data, -1, NULL, NULL, |
6208 | SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0); | |
07be1b83 | 6209 | |
686b73d4 | 6210 | |
786e8c11 YO |
6211 | CHECK_RESTUDY_GOTO; |
6212 | ||
6213 | ||
830247a4 | 6214 | if ( RExC_npar == 1 && data.longest == &(data.longest_fixed) |
b81d288d | 6215 | && data.last_start_min == 0 && data.last_end > 0 |
830247a4 | 6216 | && !RExC_seen_zerolen |
2bf803e2 | 6217 | && !(RExC_seen & REG_SEEN_VERBARG) |
bbe252da YO |
6218 | && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS))) |
6219 | r->extflags |= RXf_CHECK_ALL; | |
304ee84b | 6220 | scan_commit(pRExC_state, &data,&minlen,0); |
c277df42 IZ |
6221 | SvREFCNT_dec(data.last_found); |
6222 | ||
a0ed51b3 | 6223 | longest_float_length = CHR_SVLEN(data.longest_float); |
cf93c79d | 6224 | |
7d103382 KW |
6225 | if (! ((SvCUR(data.longest_fixed) /* ok to leave SvCUR */ |
6226 | && data.offset_fixed == data.offset_float_min | |
6227 | && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))) | |
6228 | && S_setup_longest (aTHX_ pRExC_state, | |
6229 | data.longest_float, | |
6230 | &(r->float_utf8), | |
6231 | &(r->float_substr), | |
6232 | &(r->float_end_shift), | |
6233 | data.lookbehind_float, | |
6234 | data.offset_float_min, | |
6235 | data.minlen_float, | |
6236 | longest_float_length, | |
6237 | data.flags & SF_FL_BEFORE_EOL, | |
6238 | data.flags & SF_FL_BEFORE_MEOL)) | |
6239 | { | |
1de06328 | 6240 | r->float_min_offset = data.offset_float_min - data.lookbehind_float; |
c277df42 | 6241 | r->float_max_offset = data.offset_float_max; |
1182767e | 6242 | if (data.offset_float_max < I32_MAX) /* Don't offset infinity */ |
1de06328 | 6243 | r->float_max_offset -= data.lookbehind_float; |
a0ed51b3 LW |
6244 | } |
6245 | else { | |
c445ea15 | 6246 | r->float_substr = r->float_utf8 = NULL; |
c277df42 | 6247 | SvREFCNT_dec(data.longest_float); |
c5254dd6 | 6248 | longest_float_length = 0; |
a0d0e21e | 6249 | } |
c277df42 | 6250 | |
a0ed51b3 | 6251 | longest_fixed_length = CHR_SVLEN(data.longest_fixed); |
a0c4c608 | 6252 | |
7d103382 KW |
6253 | if (S_setup_longest (aTHX_ pRExC_state, |
6254 | data.longest_fixed, | |
6255 | &(r->anchored_utf8), | |
6256 | &(r->anchored_substr), | |
6257 | &(r->anchored_end_shift), | |
6258 | data.lookbehind_fixed, | |
6259 | data.offset_fixed, | |
6260 | data.minlen_fixed, | |
6261 | longest_fixed_length, | |
6262 | data.flags & SF_FIX_BEFORE_EOL, | |
6263 | data.flags & SF_FIX_BEFORE_MEOL)) | |
1de06328 | 6264 | { |
1de06328 | 6265 | r->anchored_offset = data.offset_fixed - data.lookbehind_fixed; |
a0ed51b3 LW |
6266 | } |
6267 | else { | |
c445ea15 | 6268 | r->anchored_substr = r->anchored_utf8 = NULL; |
c277df42 | 6269 | SvREFCNT_dec(data.longest_fixed); |
c5254dd6 | 6270 | longest_fixed_length = 0; |
a0d0e21e | 6271 | } |
7d103382 | 6272 | |
f8fc2ecf YO |
6273 | if (ri->regstclass |
6274 | && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY)) | |
6275 | ri->regstclass = NULL; | |
f4244008 | 6276 | |
33b8afdf JH |
6277 | if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset) |
6278 | && stclass_flag | |
653099ff | 6279 | && !(data.start_class->flags & ANYOF_EOS) |
eb160463 GS |
6280 | && !cl_is_anything(data.start_class)) |
6281 | { | |
2eccd3b2 | 6282 | const U32 n = add_data(pRExC_state, 1, "f"); |
c613755a | 6283 | data.start_class->flags |= ANYOF_IS_SYNTHETIC; |
653099ff | 6284 | |
f8fc2ecf | 6285 | Newx(RExC_rxi->data->data[n], 1, |
653099ff GS |
6286 | struct regnode_charclass_class); |
6287 | StructCopy(data.start_class, | |
f8fc2ecf | 6288 | (struct regnode_charclass_class*)RExC_rxi->data->data[n], |
653099ff | 6289 | struct regnode_charclass_class); |
f8fc2ecf | 6290 | ri->regstclass = (regnode*)RExC_rxi->data->data[n]; |
bbe252da | 6291 | r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */ |
a3621e74 | 6292 | DEBUG_COMPILE_r({ SV *sv = sv_newmortal(); |
32fc9b6a | 6293 | regprop(r, sv, (regnode*)data.start_class); |
9c5ffd7c | 6294 | PerlIO_printf(Perl_debug_log, |
a0288114 | 6295 | "synthetic stclass \"%s\".\n", |
3f7c398e | 6296 | SvPVX_const(sv));}); |
653099ff | 6297 | } |
c277df42 IZ |
6298 | |
6299 | /* A temporary algorithm prefers floated substr to fixed one to dig more info. */ | |
c5254dd6 | 6300 | if (longest_fixed_length > longest_float_length) { |
1de06328 | 6301 | r->check_end_shift = r->anchored_end_shift; |
c277df42 | 6302 | r->check_substr = r->anchored_substr; |
33b8afdf | 6303 | r->check_utf8 = r->anchored_utf8; |
c277df42 | 6304 | r->check_offset_min = r->check_offset_max = r->anchored_offset; |
bbe252da YO |
6305 | if (r->extflags & RXf_ANCH_SINGLE) |
6306 | r->extflags |= RXf_NOSCAN; | |
a0ed51b3 LW |
6307 | } |
6308 | else { | |
1de06328 | 6309 | r->check_end_shift = r->float_end_shift; |
c277df42 | 6310 | r->check_substr = r->float_substr; |
33b8afdf | 6311 | r->check_utf8 = r->float_utf8; |
1de06328 YO |
6312 | r->check_offset_min = r->float_min_offset; |
6313 | r->check_offset_max = r->float_max_offset; | |
a0d0e21e | 6314 | } |
30382c73 IZ |
6315 | /* XXXX Currently intuiting is not compatible with ANCH_GPOS. |
6316 | This should be changed ASAP! */ | |
bbe252da YO |
6317 | if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) { |
6318 | r->extflags |= RXf_USE_INTUIT; | |
33b8afdf | 6319 | if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8)) |
bbe252da | 6320 | r->extflags |= RXf_INTUIT_TAIL; |
cad2e5aa | 6321 | } |
1de06328 YO |
6322 | /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere) |
6323 | if ( (STRLEN)minlen < longest_float_length ) | |
6324 | minlen= longest_float_length; | |
6325 | if ( (STRLEN)minlen < longest_fixed_length ) | |
6326 | minlen= longest_fixed_length; | |
6327 | */ | |
a0ed51b3 LW |
6328 | } |
6329 | else { | |
c277df42 IZ |
6330 | /* Several toplevels. Best we can is to set minlen. */ |
6331 | I32 fake; | |
653099ff | 6332 | struct regnode_charclass_class ch_class; |
cb434fcc | 6333 | I32 last_close = 0; |
686b73d4 | 6334 | |
5d458dd8 | 6335 | DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n")); |
07be1b83 | 6336 | |
f8fc2ecf | 6337 | scan = ri->program + 1; |
e755fd73 | 6338 | cl_init(pRExC_state, &ch_class); |
653099ff | 6339 | data.start_class = &ch_class; |
cb434fcc | 6340 | data.last_closep = &last_close; |
07be1b83 | 6341 | |
de8c5301 | 6342 | |
1de06328 | 6343 | minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size, |
40d049e4 | 6344 | &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0); |
de8c5301 | 6345 | |
786e8c11 | 6346 | CHECK_RESTUDY_GOTO; |
07be1b83 | 6347 | |
33b8afdf | 6348 | r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8 |
c445ea15 | 6349 | = r->float_substr = r->float_utf8 = NULL; |
f4244008 | 6350 | |
653099ff | 6351 | if (!(data.start_class->flags & ANYOF_EOS) |
eb160463 GS |
6352 | && !cl_is_anything(data.start_class)) |
6353 | { | |
2eccd3b2 | 6354 | const U32 n = add_data(pRExC_state, 1, "f"); |
c613755a | 6355 | data.start_class->flags |= ANYOF_IS_SYNTHETIC; |
653099ff | 6356 | |
f8fc2ecf | 6357 | Newx(RExC_rxi->data->data[n], 1, |
653099ff GS |
6358 | struct regnode_charclass_class); |
6359 | StructCopy(data.start_class, | |
f8fc2ecf | 6360 | (struct regnode_charclass_class*)RExC_rxi->data->data[n], |
653099ff | 6361 | struct regnode_charclass_class); |
f8fc2ecf | 6362 | ri->regstclass = (regnode*)RExC_rxi->data->data[n]; |
bbe252da | 6363 | r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */ |
a3621e74 | 6364 | DEBUG_COMPILE_r({ SV* sv = sv_newmortal(); |
32fc9b6a | 6365 | regprop(r, sv, (regnode*)data.start_class); |
9c5ffd7c | 6366 | PerlIO_printf(Perl_debug_log, |
a0288114 | 6367 | "synthetic stclass \"%s\".\n", |
3f7c398e | 6368 | SvPVX_const(sv));}); |
653099ff | 6369 | } |
a0d0e21e LW |
6370 | } |
6371 | ||
1de06328 YO |
6372 | /* Guard against an embedded (?=) or (?<=) with a longer minlen than |
6373 | the "real" pattern. */ | |
cf9788e3 RGS |
6374 | DEBUG_OPTIMISE_r({ |
6375 | PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n", | |
70685ca0 | 6376 | (IV)minlen, (IV)r->minlen); |
cf9788e3 | 6377 | }); |
de8c5301 | 6378 | r->minlenret = minlen; |
1de06328 YO |
6379 | if (r->minlen < minlen) |
6380 | r->minlen = minlen; | |
6381 | ||
b81d288d | 6382 | if (RExC_seen & REG_SEEN_GPOS) |
bbe252da | 6383 | r->extflags |= RXf_GPOS_SEEN; |
830247a4 | 6384 | if (RExC_seen & REG_SEEN_LOOKBEHIND) |
bbe252da | 6385 | r->extflags |= RXf_LOOKBEHIND_SEEN; |
471f5387 | 6386 | if (pRExC_state->num_code_blocks) |
bbe252da | 6387 | r->extflags |= RXf_EVAL_SEEN; |
f33976b4 | 6388 | if (RExC_seen & REG_SEEN_CANY) |
bbe252da | 6389 | r->extflags |= RXf_CANY_SEEN; |
e2e6a0f1 | 6390 | if (RExC_seen & REG_SEEN_VERBARG) |
bbe252da | 6391 | r->intflags |= PREGf_VERBARG_SEEN; |
5d458dd8 | 6392 | if (RExC_seen & REG_SEEN_CUTGROUP) |
bbe252da | 6393 | r->intflags |= PREGf_CUTGROUP_SEEN; |
732caac7 DM |
6394 | if (pm_flags & PMf_USE_RE_EVAL) |
6395 | r->intflags |= PREGf_USE_RE_EVAL; | |
81714fb9 | 6396 | if (RExC_paren_names) |
85fbaab2 | 6397 | RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names)); |
81714fb9 | 6398 | else |
5daac39c | 6399 | RXp_PAREN_NAMES(r) = NULL; |
0ac6acae | 6400 | |
7bd1e614 | 6401 | #ifdef STUPID_PATTERN_CHECKS |
5509d87a | 6402 | if (RX_PRELEN(rx) == 0) |
640f820d | 6403 | r->extflags |= RXf_NULL; |
5255171e | 6404 | if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3)) |
0ac6acae | 6405 | r->extflags |= RXf_WHITE; |
5509d87a | 6406 | else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^') |
e357fc67 | 6407 | r->extflags |= RXf_START_ONLY; |
f1b875a0 | 6408 | #else |
5255171e | 6409 | { |
7bd1e614 | 6410 | regnode *first = ri->program + 1; |
39aa8307 | 6411 | U8 fop = OP(first); |
f6d9469c DM |
6412 | |
6413 | if (PL_regkind[fop] == NOTHING && OP(NEXTOPER(first)) == END) | |
640f820d | 6414 | r->extflags |= RXf_NULL; |
f6d9469c | 6415 | else if (PL_regkind[fop] == BOL && OP(NEXTOPER(first)) == END) |
7bd1e614 | 6416 | r->extflags |= RXf_START_ONLY; |
f6d9469c DM |
6417 | else if (fop == PLUS && OP(NEXTOPER(first)) == SPACE |
6418 | && OP(regnext(first)) == END) | |
7bd1e614 YO |
6419 | r->extflags |= RXf_WHITE; |
6420 | } | |
f1b875a0 | 6421 | #endif |
1f1031fe YO |
6422 | #ifdef DEBUGGING |
6423 | if (RExC_paren_names) { | |
af534a04 | 6424 | ri->name_list_idx = add_data( pRExC_state, 1, "a" ); |
1f1031fe YO |
6425 | ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list); |
6426 | } else | |
1f1031fe | 6427 | #endif |
cde0cee5 | 6428 | ri->name_list_idx = 0; |
1f1031fe | 6429 | |
40d049e4 YO |
6430 | if (RExC_recurse_count) { |
6431 | for ( ; RExC_recurse_count ; RExC_recurse_count-- ) { | |
6432 | const regnode *scan = RExC_recurse[RExC_recurse_count-1]; | |
6433 | ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan ); | |
6434 | } | |
6435 | } | |
f0ab9afb | 6436 | Newxz(r->offs, RExC_npar, regexp_paren_pair); |
c74340f9 YO |
6437 | /* assume we don't need to swap parens around before we match */ |
6438 | ||
be8e71aa YO |
6439 | DEBUG_DUMP_r({ |
6440 | PerlIO_printf(Perl_debug_log,"Final program:\n"); | |
3dab1dad YO |
6441 | regdump(r); |
6442 | }); | |
7122b237 YO |
6443 | #ifdef RE_TRACK_PATTERN_OFFSETS |
6444 | DEBUG_OFFSETS_r(if (ri->u.offsets) { | |
6445 | const U32 len = ri->u.offsets[0]; | |
8e9a8a48 YO |
6446 | U32 i; |
6447 | GET_RE_DEBUG_FLAGS_DECL; | |
7122b237 | 6448 | PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]); |
8e9a8a48 | 6449 | for (i = 1; i <= len; i++) { |
7122b237 | 6450 | if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2]) |
8e9a8a48 | 6451 | PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ", |
7122b237 | 6452 | (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]); |
8e9a8a48 YO |
6453 | } |
6454 | PerlIO_printf(Perl_debug_log, "\n"); | |
6455 | }); | |
7122b237 | 6456 | #endif |
288b8c02 | 6457 | return rx; |
a687059c LW |
6458 | } |
6459 | ||
93b32b6d | 6460 | |
81714fb9 | 6461 | SV* |
192b9cd1 AB |
6462 | Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value, |
6463 | const U32 flags) | |
6464 | { | |
7918f24d NC |
6465 | PERL_ARGS_ASSERT_REG_NAMED_BUFF; |
6466 | ||
192b9cd1 AB |
6467 | PERL_UNUSED_ARG(value); |
6468 | ||
f1b875a0 | 6469 | if (flags & RXapif_FETCH) { |
192b9cd1 | 6470 | return reg_named_buff_fetch(rx, key, flags); |
f1b875a0 | 6471 | } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) { |
6ad8f254 | 6472 | Perl_croak_no_modify(aTHX); |
192b9cd1 | 6473 | return NULL; |
f1b875a0 | 6474 | } else if (flags & RXapif_EXISTS) { |
192b9cd1 AB |
6475 | return reg_named_buff_exists(rx, key, flags) |
6476 | ? &PL_sv_yes | |
6477 | : &PL_sv_no; | |
f1b875a0 | 6478 | } else if (flags & RXapif_REGNAMES) { |
192b9cd1 | 6479 | return reg_named_buff_all(rx, flags); |
f1b875a0 | 6480 | } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) { |
192b9cd1 AB |
6481 | return reg_named_buff_scalar(rx, flags); |
6482 | } else { | |
6483 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags); | |
6484 | return NULL; | |
6485 | } | |
6486 | } | |
6487 | ||
6488 | SV* | |
6489 | Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey, | |
6490 | const U32 flags) | |
6491 | { | |
7918f24d | 6492 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER; |
192b9cd1 AB |
6493 | PERL_UNUSED_ARG(lastkey); |
6494 | ||
f1b875a0 | 6495 | if (flags & RXapif_FIRSTKEY) |
192b9cd1 | 6496 | return reg_named_buff_firstkey(rx, flags); |
f1b875a0 | 6497 | else if (flags & RXapif_NEXTKEY) |
192b9cd1 AB |
6498 | return reg_named_buff_nextkey(rx, flags); |
6499 | else { | |
6500 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags); | |
6501 | return NULL; | |
6502 | } | |
6503 | } | |
6504 | ||
6505 | SV* | |
288b8c02 NC |
6506 | Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv, |
6507 | const U32 flags) | |
81714fb9 | 6508 | { |
44a2ac75 YO |
6509 | AV *retarray = NULL; |
6510 | SV *ret; | |
288b8c02 | 6511 | struct regexp *const rx = (struct regexp *)SvANY(r); |
7918f24d NC |
6512 | |
6513 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH; | |
6514 | ||
f1b875a0 | 6515 | if (flags & RXapif_ALL) |
44a2ac75 | 6516 | retarray=newAV(); |
93b32b6d | 6517 | |
5daac39c NC |
6518 | if (rx && RXp_PAREN_NAMES(rx)) { |
6519 | HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 ); | |
93b32b6d YO |
6520 | if (he_str) { |
6521 | IV i; | |
6522 | SV* sv_dat=HeVAL(he_str); | |
6523 | I32 *nums=(I32*)SvPVX(sv_dat); | |
6524 | for ( i=0; i<SvIVX(sv_dat); i++ ) { | |
192b9cd1 AB |
6525 | if ((I32)(rx->nparens) >= nums[i] |
6526 | && rx->offs[nums[i]].start != -1 | |
6527 | && rx->offs[nums[i]].end != -1) | |
93b32b6d | 6528 | { |
49d7dfbc | 6529 | ret = newSVpvs(""); |
288b8c02 | 6530 | CALLREG_NUMBUF_FETCH(r,nums[i],ret); |
93b32b6d YO |
6531 | if (!retarray) |
6532 | return ret; | |
6533 | } else { | |
7402016d AB |
6534 | if (retarray) |
6535 | ret = newSVsv(&PL_sv_undef); | |
93b32b6d | 6536 | } |
ec83ea38 | 6537 | if (retarray) |
93b32b6d | 6538 | av_push(retarray, ret); |
81714fb9 | 6539 | } |
93b32b6d | 6540 | if (retarray) |
ad64d0ec | 6541 | return newRV_noinc(MUTABLE_SV(retarray)); |
192b9cd1 AB |
6542 | } |
6543 | } | |
6544 | return NULL; | |
6545 | } | |
6546 | ||
6547 | bool | |
288b8c02 | 6548 | Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key, |
192b9cd1 AB |
6549 | const U32 flags) |
6550 | { | |
288b8c02 | 6551 | struct regexp *const rx = (struct regexp *)SvANY(r); |
7918f24d NC |
6552 | |
6553 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS; | |
6554 | ||
5daac39c | 6555 | if (rx && RXp_PAREN_NAMES(rx)) { |
f1b875a0 | 6556 | if (flags & RXapif_ALL) { |
5daac39c | 6557 | return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0); |
192b9cd1 | 6558 | } else { |
288b8c02 | 6559 | SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags); |
6499cc01 RGS |
6560 | if (sv) { |
6561 | SvREFCNT_dec(sv); | |
192b9cd1 AB |
6562 | return TRUE; |
6563 | } else { | |
6564 | return FALSE; | |
6565 | } | |
6566 | } | |
6567 | } else { | |
6568 | return FALSE; | |
6569 | } | |
6570 | } | |
6571 | ||
6572 | SV* | |
288b8c02 | 6573 | Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags) |
192b9cd1 | 6574 | { |
288b8c02 | 6575 | struct regexp *const rx = (struct regexp *)SvANY(r); |
7918f24d NC |
6576 | |
6577 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY; | |
6578 | ||
5daac39c NC |
6579 | if ( rx && RXp_PAREN_NAMES(rx) ) { |
6580 | (void)hv_iterinit(RXp_PAREN_NAMES(rx)); | |
192b9cd1 | 6581 | |
288b8c02 | 6582 | return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY); |
1e1d4b91 JJ |
6583 | } else { |
6584 | return FALSE; | |
6585 | } | |
192b9cd1 AB |
6586 | } |
6587 | ||
6588 | SV* | |
288b8c02 | 6589 | Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags) |
192b9cd1 | 6590 | { |
288b8c02 | 6591 | struct regexp *const rx = (struct regexp *)SvANY(r); |
250257bb | 6592 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
6593 | |
6594 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY; | |
6595 | ||
5daac39c NC |
6596 | if (rx && RXp_PAREN_NAMES(rx)) { |
6597 | HV *hv = RXp_PAREN_NAMES(rx); | |
192b9cd1 AB |
6598 | HE *temphe; |
6599 | while ( (temphe = hv_iternext_flags(hv,0)) ) { | |
6600 | IV i; | |
6601 | IV parno = 0; | |
6602 | SV* sv_dat = HeVAL(temphe); | |
6603 | I32 *nums = (I32*)SvPVX(sv_dat); | |
6604 | for ( i = 0; i < SvIVX(sv_dat); i++ ) { | |
250257bb | 6605 | if ((I32)(rx->lastparen) >= nums[i] && |
192b9cd1 AB |
6606 | rx->offs[nums[i]].start != -1 && |
6607 | rx->offs[nums[i]].end != -1) | |
6608 | { | |
6609 | parno = nums[i]; | |
6610 | break; | |
6611 | } | |
6612 | } | |
f1b875a0 | 6613 | if (parno || flags & RXapif_ALL) { |
a663657d | 6614 | return newSVhek(HeKEY_hek(temphe)); |
192b9cd1 | 6615 | } |
81714fb9 YO |
6616 | } |
6617 | } | |
44a2ac75 YO |
6618 | return NULL; |
6619 | } | |
6620 | ||
192b9cd1 | 6621 | SV* |
288b8c02 | 6622 | Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags) |
192b9cd1 AB |
6623 | { |
6624 | SV *ret; | |
6625 | AV *av; | |
6626 | I32 length; | |
288b8c02 | 6627 | struct regexp *const rx = (struct regexp *)SvANY(r); |
192b9cd1 | 6628 | |
7918f24d NC |
6629 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR; |
6630 | ||
5daac39c | 6631 | if (rx && RXp_PAREN_NAMES(rx)) { |
f1b875a0 | 6632 | if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) { |
5daac39c | 6633 | return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx))); |
f1b875a0 | 6634 | } else if (flags & RXapif_ONE) { |
288b8c02 | 6635 | ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES)); |
502c6561 | 6636 | av = MUTABLE_AV(SvRV(ret)); |
192b9cd1 | 6637 | length = av_len(av); |
ec83ea38 | 6638 | SvREFCNT_dec(ret); |
192b9cd1 AB |
6639 | return newSViv(length + 1); |
6640 | } else { | |
6641 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags); | |
6642 | return NULL; | |
6643 | } | |
6644 | } | |
6645 | return &PL_sv_undef; | |
6646 | } | |
6647 | ||
6648 | SV* | |
288b8c02 | 6649 | Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags) |
192b9cd1 | 6650 | { |
288b8c02 | 6651 | struct regexp *const rx = (struct regexp *)SvANY(r); |
192b9cd1 AB |
6652 | AV *av = newAV(); |
6653 | ||
7918f24d NC |
6654 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL; |
6655 | ||
5daac39c NC |
6656 | if (rx && RXp_PAREN_NAMES(rx)) { |
6657 | HV *hv= RXp_PAREN_NAMES(rx); | |
192b9cd1 AB |
6658 | HE *temphe; |
6659 | (void)hv_iterinit(hv); | |
6660 | while ( (temphe = hv_iternext_flags(hv,0)) ) { | |
6661 | IV i; | |
6662 | IV parno = 0; | |
6663 | SV* sv_dat = HeVAL(temphe); | |
6664 | I32 *nums = (I32*)SvPVX(sv_dat); | |
6665 | for ( i = 0; i < SvIVX(sv_dat); i++ ) { | |
250257bb | 6666 | if ((I32)(rx->lastparen) >= nums[i] && |
192b9cd1 AB |
6667 | rx->offs[nums[i]].start != -1 && |
6668 | rx->offs[nums[i]].end != -1) | |
6669 | { | |
6670 | parno = nums[i]; | |
6671 | break; | |
6672 | } | |
6673 | } | |
f1b875a0 | 6674 | if (parno || flags & RXapif_ALL) { |
a663657d | 6675 | av_push(av, newSVhek(HeKEY_hek(temphe))); |
192b9cd1 AB |
6676 | } |
6677 | } | |
6678 | } | |
6679 | ||
ad64d0ec | 6680 | return newRV_noinc(MUTABLE_SV(av)); |
192b9cd1 AB |
6681 | } |
6682 | ||
49d7dfbc | 6683 | void |
288b8c02 NC |
6684 | Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren, |
6685 | SV * const sv) | |
44a2ac75 | 6686 | { |
288b8c02 | 6687 | struct regexp *const rx = (struct regexp *)SvANY(r); |
44a2ac75 | 6688 | char *s = NULL; |
a9d504c3 | 6689 | I32 i = 0; |
44a2ac75 | 6690 | I32 s1, t1; |
2c7b5d76 | 6691 | I32 n = paren; |
7918f24d NC |
6692 | |
6693 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH; | |
44a2ac75 | 6694 | |
2c7b5d76 DM |
6695 | if ( ( n == RX_BUFF_IDX_CARET_PREMATCH |
6696 | || n == RX_BUFF_IDX_CARET_FULLMATCH | |
6697 | || n == RX_BUFF_IDX_CARET_POSTMATCH | |
6698 | ) | |
6699 | && !(rx->extflags & RXf_PMf_KEEPCOPY) | |
6700 | ) | |
6701 | goto ret_undef; | |
6702 | ||
6703 | if (!rx->subbeg) | |
6704 | goto ret_undef; | |
6705 | ||
6706 | if (n == RX_BUFF_IDX_CARET_FULLMATCH) | |
6707 | /* no need to distinguish between them any more */ | |
6708 | n = RX_BUFF_IDX_FULLMATCH; | |
6709 | ||
6710 | if ((n == RX_BUFF_IDX_PREMATCH || n == RX_BUFF_IDX_CARET_PREMATCH) | |
6711 | && rx->offs[0].start != -1) | |
6712 | { | |
6713 | /* $`, ${^PREMATCH} */ | |
f0ab9afb | 6714 | i = rx->offs[0].start; |
cde0cee5 | 6715 | s = rx->subbeg; |
44a2ac75 YO |
6716 | } |
6717 | else | |
2c7b5d76 DM |
6718 | if ((n == RX_BUFF_IDX_POSTMATCH || n == RX_BUFF_IDX_CARET_POSTMATCH) |
6719 | && rx->offs[0].end != -1) | |
6720 | { | |
6721 | /* $', ${^POSTMATCH} */ | |
6502e081 DM |
6722 | s = rx->subbeg - rx->suboffset + rx->offs[0].end; |
6723 | i = rx->sublen + rx->suboffset - rx->offs[0].end; | |
44a2ac75 YO |
6724 | } |
6725 | else | |
2c7b5d76 DM |
6726 | if ( 0 <= n && n <= (I32)rx->nparens && |
6727 | (s1 = rx->offs[n].start) != -1 && | |
6728 | (t1 = rx->offs[n].end) != -1) | |
44a2ac75 | 6729 | { |
2c7b5d76 | 6730 | /* $&, ${^MATCH}, $1 ... */ |
44a2ac75 | 6731 | i = t1 - s1; |
6502e081 | 6732 | s = rx->subbeg + s1 - rx->suboffset; |
cde0cee5 | 6733 | } else { |
2c7b5d76 | 6734 | goto ret_undef; |
cde0cee5 | 6735 | } |
2c7b5d76 DM |
6736 | |
6737 | assert(s >= rx->subbeg); | |
cde0cee5 YO |
6738 | assert(rx->sublen >= (s - rx->subbeg) + i ); |
6739 | if (i >= 0) { | |
6740 | const int oldtainted = PL_tainted; | |
6741 | TAINT_NOT; | |
6742 | sv_setpvn(sv, s, i); | |
6743 | PL_tainted = oldtainted; | |
6744 | if ( (rx->extflags & RXf_CANY_SEEN) | |
07bc277f | 6745 | ? (RXp_MATCH_UTF8(rx) |
cde0cee5 | 6746 | && (!i || is_utf8_string((U8*)s, i))) |
07bc277f | 6747 | : (RXp_MATCH_UTF8(rx)) ) |
cde0cee5 YO |
6748 | { |
6749 | SvUTF8_on(sv); | |
6750 | } | |
6751 | else | |
6752 | SvUTF8_off(sv); | |
6753 | if (PL_tainting) { | |
07bc277f | 6754 | if (RXp_MATCH_TAINTED(rx)) { |
cde0cee5 YO |
6755 | if (SvTYPE(sv) >= SVt_PVMG) { |
6756 | MAGIC* const mg = SvMAGIC(sv); | |
6757 | MAGIC* mgt; | |
6758 | PL_tainted = 1; | |
6759 | SvMAGIC_set(sv, mg->mg_moremagic); | |
6760 | SvTAINT(sv); | |
6761 | if ((mgt = SvMAGIC(sv))) { | |
6762 | mg->mg_moremagic = mgt; | |
6763 | SvMAGIC_set(sv, mg); | |
44a2ac75 | 6764 | } |
cde0cee5 YO |
6765 | } else { |
6766 | PL_tainted = 1; | |
6767 | SvTAINT(sv); | |
6768 | } | |
6769 | } else | |
6770 | SvTAINTED_off(sv); | |
44a2ac75 | 6771 | } |
81714fb9 | 6772 | } else { |
2c7b5d76 | 6773 | ret_undef: |
44a2ac75 | 6774 | sv_setsv(sv,&PL_sv_undef); |
49d7dfbc | 6775 | return; |
81714fb9 YO |
6776 | } |
6777 | } | |
93b32b6d | 6778 | |
2fdbfb4d AB |
6779 | void |
6780 | Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren, | |
6781 | SV const * const value) | |
6782 | { | |
7918f24d NC |
6783 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE; |
6784 | ||
2fdbfb4d AB |
6785 | PERL_UNUSED_ARG(rx); |
6786 | PERL_UNUSED_ARG(paren); | |
6787 | PERL_UNUSED_ARG(value); | |
6788 | ||
6789 | if (!PL_localizing) | |
6ad8f254 | 6790 | Perl_croak_no_modify(aTHX); |
2fdbfb4d AB |
6791 | } |
6792 | ||
6793 | I32 | |
288b8c02 | 6794 | Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv, |
2fdbfb4d AB |
6795 | const I32 paren) |
6796 | { | |
288b8c02 | 6797 | struct regexp *const rx = (struct regexp *)SvANY(r); |
2fdbfb4d AB |
6798 | I32 i; |
6799 | I32 s1, t1; | |
6800 | ||
7918f24d NC |
6801 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH; |
6802 | ||
2fdbfb4d | 6803 | /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */ |
2c7b5d76 DM |
6804 | switch (paren) { |
6805 | case RX_BUFF_IDX_CARET_PREMATCH: /* ${^PREMATCH} */ | |
6806 | if (!(rx->extflags & RXf_PMf_KEEPCOPY)) | |
6807 | goto warn_undef; | |
6808 | /*FALLTHROUGH*/ | |
6809 | ||
6810 | case RX_BUFF_IDX_PREMATCH: /* $` */ | |
2fdbfb4d AB |
6811 | if (rx->offs[0].start != -1) { |
6812 | i = rx->offs[0].start; | |
6813 | if (i > 0) { | |
6814 | s1 = 0; | |
6815 | t1 = i; | |
6816 | goto getlen; | |
6817 | } | |
6818 | } | |
6819 | return 0; | |
2c7b5d76 DM |
6820 | |
6821 | case RX_BUFF_IDX_CARET_POSTMATCH: /* ${^POSTMATCH} */ | |
6822 | if (!(rx->extflags & RXf_PMf_KEEPCOPY)) | |
6823 | goto warn_undef; | |
6824 | case RX_BUFF_IDX_POSTMATCH: /* $' */ | |
2fdbfb4d AB |
6825 | if (rx->offs[0].end != -1) { |
6826 | i = rx->sublen - rx->offs[0].end; | |
6827 | if (i > 0) { | |
6828 | s1 = rx->offs[0].end; | |
6829 | t1 = rx->sublen; | |
6830 | goto getlen; | |
6831 | } | |
6832 | } | |
6833 | return 0; | |
2c7b5d76 DM |
6834 | |
6835 | case RX_BUFF_IDX_CARET_FULLMATCH: /* ${^MATCH} */ | |
6836 | if (!(rx->extflags & RXf_PMf_KEEPCOPY)) | |
6837 | goto warn_undef; | |
2c7b5d76 DM |
6838 | /*FALLTHROUGH*/ |
6839 | ||
192b9cd1 AB |
6840 | /* $& / ${^MATCH}, $1, $2, ... */ |
6841 | default: | |
2fdbfb4d AB |
6842 | if (paren <= (I32)rx->nparens && |
6843 | (s1 = rx->offs[paren].start) != -1 && | |
6844 | (t1 = rx->offs[paren].end) != -1) | |
6845 | { | |
6846 | i = t1 - s1; | |
6847 | goto getlen; | |
6848 | } else { | |
2c7b5d76 | 6849 | warn_undef: |
2fdbfb4d | 6850 | if (ckWARN(WARN_UNINITIALIZED)) |
ad64d0ec | 6851 | report_uninit((const SV *)sv); |
2fdbfb4d AB |
6852 | return 0; |
6853 | } | |
6854 | } | |
6855 | getlen: | |
07bc277f | 6856 | if (i > 0 && RXp_MATCH_UTF8(rx)) { |
6502e081 | 6857 | const char * const s = rx->subbeg - rx->suboffset + s1; |
2fdbfb4d AB |
6858 | const U8 *ep; |
6859 | STRLEN el; | |
6860 | ||
6861 | i = t1 - s1; | |
6862 | if (is_utf8_string_loclen((U8*)s, i, &ep, &el)) | |
6863 | i = el; | |
6864 | } | |
6865 | return i; | |
6866 | } | |
6867 | ||
fe578d7f | 6868 | SV* |
49d7dfbc | 6869 | Perl_reg_qr_package(pTHX_ REGEXP * const rx) |
fe578d7f | 6870 | { |
7918f24d | 6871 | PERL_ARGS_ASSERT_REG_QR_PACKAGE; |
fe578d7f | 6872 | PERL_UNUSED_ARG(rx); |
0fc92fc6 YO |
6873 | if (0) |
6874 | return NULL; | |
6875 | else | |
6876 | return newSVpvs("Regexp"); | |
fe578d7f | 6877 | } |
0a4db386 | 6878 | |
894be9b7 | 6879 | /* Scans the name of a named buffer from the pattern. |
0a4db386 YO |
6880 | * If flags is REG_RSN_RETURN_NULL returns null. |
6881 | * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name | |
6882 | * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding | |
6883 | * to the parsed name as looked up in the RExC_paren_names hash. | |
6884 | * If there is an error throws a vFAIL().. type exception. | |
894be9b7 | 6885 | */ |
0a4db386 YO |
6886 | |
6887 | #define REG_RSN_RETURN_NULL 0 | |
6888 | #define REG_RSN_RETURN_NAME 1 | |
6889 | #define REG_RSN_RETURN_DATA 2 | |
6890 | ||
894be9b7 | 6891 | STATIC SV* |
7918f24d NC |
6892 | S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) |
6893 | { | |
894be9b7 | 6894 | char *name_start = RExC_parse; |
1f1031fe | 6895 | |
7918f24d NC |
6896 | PERL_ARGS_ASSERT_REG_SCAN_NAME; |
6897 | ||
1f1031fe YO |
6898 | if (isIDFIRST_lazy_if(RExC_parse, UTF)) { |
6899 | /* skip IDFIRST by using do...while */ | |
6900 | if (UTF) | |
6901 | do { | |
6902 | RExC_parse += UTF8SKIP(RExC_parse); | |
6903 | } while (isALNUM_utf8((U8*)RExC_parse)); | |
6904 | else | |
6905 | do { | |
6906 | RExC_parse++; | |
6907 | } while (isALNUM(*RExC_parse)); | |
1f4f6bf1 YO |
6908 | } else { |
6909 | RExC_parse++; /* so the <- from the vFAIL is after the offending character */ | |
6910 | vFAIL("Group name must start with a non-digit word character"); | |
894be9b7 | 6911 | } |
0a4db386 | 6912 | if ( flags ) { |
59cd0e26 NC |
6913 | SV* sv_name |
6914 | = newSVpvn_flags(name_start, (int)(RExC_parse - name_start), | |
6915 | SVs_TEMP | (UTF ? SVf_UTF8 : 0)); | |
0a4db386 YO |
6916 | if ( flags == REG_RSN_RETURN_NAME) |
6917 | return sv_name; | |
6918 | else if (flags==REG_RSN_RETURN_DATA) { | |
6919 | HE *he_str = NULL; | |
6920 | SV *sv_dat = NULL; | |
6921 | if ( ! sv_name ) /* should not happen*/ | |
6922 | Perl_croak(aTHX_ "panic: no svname in reg_scan_name"); | |
6923 | if (RExC_paren_names) | |
6924 | he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 ); | |
6925 | if ( he_str ) | |
6926 | sv_dat = HeVAL(he_str); | |
6927 | if ( ! sv_dat ) | |
6928 | vFAIL("Reference to nonexistent named group"); | |
6929 | return sv_dat; | |
6930 | } | |
6931 | else { | |
5637ef5b NC |
6932 | Perl_croak(aTHX_ "panic: bad flag %lx in reg_scan_name", |
6933 | (unsigned long) flags); | |
0a4db386 | 6934 | } |
118e2215 | 6935 | assert(0); /* NOT REACHED */ |
894be9b7 | 6936 | } |
0a4db386 | 6937 | return NULL; |
894be9b7 YO |
6938 | } |
6939 | ||
3dab1dad YO |
6940 | #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \ |
6941 | int rem=(int)(RExC_end - RExC_parse); \ | |
6942 | int cut; \ | |
6943 | int num; \ | |
6944 | int iscut=0; \ | |
6945 | if (rem>10) { \ | |
6946 | rem=10; \ | |
6947 | iscut=1; \ | |
6948 | } \ | |
6949 | cut=10-rem; \ | |
6950 | if (RExC_lastparse!=RExC_parse) \ | |
6951 | PerlIO_printf(Perl_debug_log," >%.*s%-*s", \ | |
6952 | rem, RExC_parse, \ | |
6953 | cut + 4, \ | |
6954 | iscut ? "..." : "<" \ | |
6955 | ); \ | |
6956 | else \ | |
6957 | PerlIO_printf(Perl_debug_log,"%16s",""); \ | |
6958 | \ | |
6959 | if (SIZE_ONLY) \ | |
3b57cd43 | 6960 | num = RExC_size + 1; \ |
3dab1dad YO |
6961 | else \ |
6962 | num=REG_NODE_NUM(RExC_emit); \ | |
6963 | if (RExC_lastnum!=num) \ | |
0a4db386 | 6964 | PerlIO_printf(Perl_debug_log,"|%4d",num); \ |
3dab1dad | 6965 | else \ |
0a4db386 | 6966 | PerlIO_printf(Perl_debug_log,"|%4s",""); \ |
be8e71aa YO |
6967 | PerlIO_printf(Perl_debug_log,"|%*s%-4s", \ |
6968 | (int)((depth*2)), "", \ | |
3dab1dad YO |
6969 | (funcname) \ |
6970 | ); \ | |
6971 | RExC_lastnum=num; \ | |
6972 | RExC_lastparse=RExC_parse; \ | |
6973 | }) | |
6974 | ||
07be1b83 YO |
6975 | |
6976 | ||
3dab1dad YO |
6977 | #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \ |
6978 | DEBUG_PARSE_MSG((funcname)); \ | |
6979 | PerlIO_printf(Perl_debug_log,"%4s","\n"); \ | |
6980 | }) | |
6bda09f9 YO |
6981 | #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \ |
6982 | DEBUG_PARSE_MSG((funcname)); \ | |
6983 | PerlIO_printf(Perl_debug_log,fmt "\n",args); \ | |
6984 | }) | |
d764b54e KW |
6985 | |
6986 | /* This section of code defines the inversion list object and its methods. The | |
6987 | * interfaces are highly subject to change, so as much as possible is static to | |
fa2d2a23 KW |
6988 | * this file. An inversion list is here implemented as a malloc'd C UV array |
6989 | * with some added info that is placed as UVs at the beginning in a header | |
6990 | * portion. An inversion list for Unicode is an array of code points, sorted | |
6991 | * by ordinal number. The zeroth element is the first code point in the list. | |
6992 | * The 1th element is the first element beyond that not in the list. In other | |
6993 | * words, the first range is | |
6994 | * invlist[0]..(invlist[1]-1) | |
dbe7a391 KW |
6995 | * The other ranges follow. Thus every element whose index is divisible by two |
6996 | * marks the beginning of a range that is in the list, and every element not | |
fa2d2a23 KW |
6997 | * divisible by two marks the beginning of a range not in the list. A single |
6998 | * element inversion list that contains the single code point N generally | |
6999 | * consists of two elements | |
7000 | * invlist[0] == N | |
7001 | * invlist[1] == N+1 | |
7002 | * (The exception is when N is the highest representable value on the | |
7003 | * machine, in which case the list containing just it would be a single | |
7004 | * element, itself. By extension, if the last range in the list extends to | |
7005 | * infinity, then the first element of that range will be in the inversion list | |
7006 | * at a position that is divisible by two, and is the final element in the | |
7007 | * list.) | |
f1b67122 KW |
7008 | * Taking the complement (inverting) an inversion list is quite simple, if the |
7009 | * first element is 0, remove it; otherwise add a 0 element at the beginning. | |
98e1e01b KW |
7010 | * This implementation reserves an element at the beginning of each inversion |
7011 | * list to contain 0 when the list contains 0, and contains 1 otherwise. The | |
7012 | * actual beginning of the list is either that element if 0, or the next one if | |
7013 | * 1. | |
f1b67122 | 7014 | * |
fa2d2a23 KW |
7015 | * More about inversion lists can be found in "Unicode Demystified" |
7016 | * Chapter 13 by Richard Gillam, published by Addison-Wesley. | |
97b14ce7 | 7017 | * More will be coming when functionality is added later. |
d764b54e | 7018 | * |
fa2d2a23 KW |
7019 | * The inversion list data structure is currently implemented as an SV pointing |
7020 | * to an array of UVs that the SV thinks are bytes. This allows us to have an | |
7021 | * array of UV whose memory management is automatically handled by the existing | |
7022 | * facilities for SV's. | |
62672576 | 7023 | * |
d764b54e KW |
7024 | * Some of the methods should always be private to the implementation, and some |
7025 | * should eventually be made public */ | |
7026 | ||
9e7f4f43 KW |
7027 | /* The header definitions are in F<inline_invlist.c> */ |
7028 | ||
97b14ce7 KW |
7029 | #define TO_INTERNAL_SIZE(x) ((x + HEADER_LENGTH) * sizeof(UV)) |
7030 | #define FROM_INTERNAL_SIZE(x) ((x / sizeof(UV)) - HEADER_LENGTH) | |
7031 | ||
d764b54e | 7032 | #define INVLIST_INITIAL_LEN 10 |
d764b54e KW |
7033 | |
7034 | PERL_STATIC_INLINE UV* | |
f1b67122 KW |
7035 | S__invlist_array_init(pTHX_ SV* const invlist, const bool will_have_0) |
7036 | { | |
7037 | /* Returns a pointer to the first element in the inversion list's array. | |
7038 | * This is called upon initialization of an inversion list. Where the | |
7039 | * array begins depends on whether the list has the code point U+0000 | |
7040 | * in it or not. The other parameter tells it whether the code that | |
7041 | * follows this call is about to put a 0 in the inversion list or not. | |
7042 | * The first element is either the element with 0, if 0, or the next one, | |
7043 | * if 1 */ | |
7044 | ||
7045 | UV* zero = get_invlist_zero_addr(invlist); | |
7046 | ||
7047 | PERL_ARGS_ASSERT__INVLIST_ARRAY_INIT; | |
7048 | ||
7049 | /* Must be empty */ | |
29f51c6b | 7050 | assert(! *_get_invlist_len_addr(invlist)); |
f1b67122 KW |
7051 | |
7052 | /* 1^1 = 0; 1^0 = 1 */ | |
7053 | *zero = 1 ^ will_have_0; | |
7054 | return zero + *zero; | |
7055 | } | |
7056 | ||
7057 | PERL_STATIC_INLINE UV* | |
a25abddc | 7058 | S_invlist_array(pTHX_ SV* const invlist) |
d764b54e KW |
7059 | { |
7060 | /* Returns the pointer to the inversion list's array. Every time the | |
7061 | * length changes, this needs to be called in case malloc or realloc moved | |
7062 | * it */ | |
7063 | ||
d764b54e KW |
7064 | PERL_ARGS_ASSERT_INVLIST_ARRAY; |
7065 | ||
dbe7a391 KW |
7066 | /* Must not be empty. If these fail, you probably didn't check for <len> |
7067 | * being non-zero before trying to get the array */ | |
29f51c6b | 7068 | assert(*_get_invlist_len_addr(invlist)); |
f1b67122 KW |
7069 | assert(*get_invlist_zero_addr(invlist) == 0 |
7070 | || *get_invlist_zero_addr(invlist) == 1); | |
7071 | ||
7072 | /* The array begins either at the element reserved for zero if the | |
7073 | * list contains 0 (that element will be set to 0), or otherwise the next | |
7074 | * element (in which case the reserved element will be set to 1). */ | |
7075 | return (UV *) (get_invlist_zero_addr(invlist) | |
7076 | + *get_invlist_zero_addr(invlist)); | |
d764b54e KW |
7077 | } |
7078 | ||
c56a880b KW |
7079 | PERL_STATIC_INLINE void |
7080 | S_invlist_set_len(pTHX_ SV* const invlist, const UV len) | |
7081 | { | |
7082 | /* Sets the current number of elements stored in the inversion list */ | |
7083 | ||
7084 | PERL_ARGS_ASSERT_INVLIST_SET_LEN; | |
7085 | ||
29f51c6b | 7086 | *_get_invlist_len_addr(invlist) = len; |
f1b67122 | 7087 | |
32f89ef6 KW |
7088 | assert(len <= SvLEN(invlist)); |
7089 | ||
f1b67122 KW |
7090 | SvCUR_set(invlist, TO_INTERNAL_SIZE(len)); |
7091 | /* If the list contains U+0000, that element is part of the header, | |
7092 | * and should not be counted as part of the array. It will contain | |
7093 | * 0 in that case, and 1 otherwise. So we could flop 0=>1, 1=>0 and | |
7094 | * subtract: | |
7095 | * SvCUR_set(invlist, | |
7096 | * TO_INTERNAL_SIZE(len | |
7097 | * - (*get_invlist_zero_addr(inv_list) ^ 1))); | |
7098 | * But, this is only valid if len is not 0. The consequences of not doing | |
9479a769 KW |
7099 | * this is that the memory allocation code may think that 1 more UV is |
7100 | * being used than actually is, and so might do an unnecessary grow. That | |
7101 | * seems worth not bothering to make this the precise amount. | |
25e94a65 KW |
7102 | * |
7103 | * Note that when inverting, SvCUR shouldn't change */ | |
c56a880b KW |
7104 | } |
7105 | ||
15896d2f KW |
7106 | PERL_STATIC_INLINE IV* |
7107 | S_get_invlist_previous_index_addr(pTHX_ SV* invlist) | |
7108 | { | |
7109 | /* Return the address of the UV that is reserved to hold the cached index | |
7110 | * */ | |
7111 | ||
7112 | PERL_ARGS_ASSERT_GET_INVLIST_PREVIOUS_INDEX_ADDR; | |
7113 | ||
7114 | return (IV *) (SvPVX(invlist) + (INVLIST_PREVIOUS_INDEX_OFFSET * sizeof (UV))); | |
7115 | } | |
7116 | ||
7117 | PERL_STATIC_INLINE IV | |
7118 | S_invlist_previous_index(pTHX_ SV* const invlist) | |
7119 | { | |
7120 | /* Returns cached index of previous search */ | |
7121 | ||
7122 | PERL_ARGS_ASSERT_INVLIST_PREVIOUS_INDEX; | |
7123 | ||
7124 | return *get_invlist_previous_index_addr(invlist); | |
7125 | } | |
7126 | ||
7127 | PERL_STATIC_INLINE void | |
7128 | S_invlist_set_previous_index(pTHX_ SV* const invlist, const IV index) | |
7129 | { | |
7130 | /* Caches <index> for later retrieval */ | |
7131 | ||
7132 | PERL_ARGS_ASSERT_INVLIST_SET_PREVIOUS_INDEX; | |
7133 | ||
7134 | assert(index == 0 || index < (int) _invlist_len(invlist)); | |
7135 | ||
7136 | *get_invlist_previous_index_addr(invlist) = index; | |
7137 | } | |
7138 | ||
d764b54e | 7139 | PERL_STATIC_INLINE UV |
a25abddc | 7140 | S_invlist_max(pTHX_ SV* const invlist) |
d764b54e KW |
7141 | { |
7142 | /* Returns the maximum number of elements storable in the inversion list's | |
7143 | * array, without having to realloc() */ | |
7144 | ||
d764b54e KW |
7145 | PERL_ARGS_ASSERT_INVLIST_MAX; |
7146 | ||
005b65ed | 7147 | return FROM_INTERNAL_SIZE(SvLEN(invlist)); |
d764b54e KW |
7148 | } |
7149 | ||
f1b67122 KW |
7150 | PERL_STATIC_INLINE UV* |
7151 | S_get_invlist_zero_addr(pTHX_ SV* invlist) | |
7152 | { | |
7153 | /* Return the address of the UV that is reserved to hold 0 if the inversion | |
7154 | * list contains 0. This has to be the last element of the heading, as the | |
7155 | * list proper starts with either it if 0, or the next element if not. | |
7156 | * (But we force it to contain either 0 or 1) */ | |
7157 | ||
7158 | PERL_ARGS_ASSERT_GET_INVLIST_ZERO_ADDR; | |
7159 | ||
7160 | return (UV *) (SvPVX(invlist) + (INVLIST_ZERO_OFFSET * sizeof (UV))); | |
7161 | } | |
d764b54e | 7162 | |
8d69a883 | 7163 | #ifndef PERL_IN_XSUB_RE |
a25abddc | 7164 | SV* |
d764b54e KW |
7165 | Perl__new_invlist(pTHX_ IV initial_size) |
7166 | { | |
7167 | ||
7168 | /* Return a pointer to a newly constructed inversion list, with enough | |
7169 | * space to store 'initial_size' elements. If that number is negative, a | |
7170 | * system default is used instead */ | |
7171 | ||
97b14ce7 KW |
7172 | SV* new_list; |
7173 | ||
d764b54e KW |
7174 | if (initial_size < 0) { |
7175 | initial_size = INVLIST_INITIAL_LEN; | |
7176 | } | |
7177 | ||
7178 | /* Allocate the initial space */ | |
97b14ce7 KW |
7179 | new_list = newSV(TO_INTERNAL_SIZE(initial_size)); |
7180 | invlist_set_len(new_list, 0); | |
7181 | ||
f3dc70d1 KW |
7182 | /* Force iterinit() to be used to get iteration to work */ |
7183 | *get_invlist_iter_addr(new_list) = UV_MAX; | |
7184 | ||
f1b67122 KW |
7185 | /* This should force a segfault if a method doesn't initialize this |
7186 | * properly */ | |
7187 | *get_invlist_zero_addr(new_list) = UV_MAX; | |
7188 | ||
15896d2f | 7189 | *get_invlist_previous_index_addr(new_list) = 0; |
f59ff194 | 7190 | *get_invlist_version_id_addr(new_list) = INVLIST_VERSION_ID; |
15896d2f | 7191 | #if HEADER_LENGTH != 5 |
f59ff194 KW |
7192 | # error Need to regenerate VERSION_ID by running perl -E 'say int(rand 2**31-1)', and then changing the #if to the new length |
7193 | #endif | |
7194 | ||
97b14ce7 | 7195 | return new_list; |
d764b54e | 7196 | } |
8d69a883 | 7197 | #endif |
d764b54e | 7198 | |
f59ff194 KW |
7199 | STATIC SV* |
7200 | S__new_invlist_C_array(pTHX_ UV* list) | |
7201 | { | |
7202 | /* Return a pointer to a newly constructed inversion list, initialized to | |
7203 | * point to <list>, which has to be in the exact correct inversion list | |
7204 | * form, including internal fields. Thus this is a dangerous routine that | |
7205 | * should not be used in the wrong hands */ | |
7206 | ||
7207 | SV* invlist = newSV_type(SVt_PV); | |
7208 | ||
7209 | PERL_ARGS_ASSERT__NEW_INVLIST_C_ARRAY; | |
7210 | ||
7211 | SvPV_set(invlist, (char *) list); | |
7212 | SvLEN_set(invlist, 0); /* Means we own the contents, and the system | |
7213 | shouldn't touch it */ | |
29f51c6b | 7214 | SvCUR_set(invlist, TO_INTERNAL_SIZE(_invlist_len(invlist))); |
f59ff194 KW |
7215 | |
7216 | if (*get_invlist_version_id_addr(invlist) != INVLIST_VERSION_ID) { | |
7217 | Perl_croak(aTHX_ "panic: Incorrect version for previously generated inversion list"); | |
7218 | } | |
7219 | ||
7220 | return invlist; | |
7221 | } | |
7222 | ||
d764b54e | 7223 | STATIC void |
a25abddc | 7224 | S_invlist_extend(pTHX_ SV* const invlist, const UV new_max) |
d764b54e | 7225 | { |
62672576 | 7226 | /* Grow the maximum size of an inversion list */ |
d764b54e KW |
7227 | |
7228 | PERL_ARGS_ASSERT_INVLIST_EXTEND; | |
7229 | ||
005b65ed | 7230 | SvGROW((SV *)invlist, TO_INTERNAL_SIZE(new_max)); |
d764b54e KW |
7231 | } |
7232 | ||
7233 | PERL_STATIC_INLINE void | |
a25abddc | 7234 | S_invlist_trim(pTHX_ SV* const invlist) |
d764b54e KW |
7235 | { |
7236 | PERL_ARGS_ASSERT_INVLIST_TRIM; | |
7237 | ||
7238 | /* Change the length of the inversion list to how many entries it currently | |
7239 | * has */ | |
7240 | ||
62672576 | 7241 | SvPV_shrink_to_cur((SV *) invlist); |
d764b54e KW |
7242 | } |
7243 | ||
8dc9348a KW |
7244 | #define _invlist_union_complement_2nd(a, b, output) _invlist_union_maybe_complement_2nd(a, b, TRUE, output) |
7245 | ||
9d501133 KW |
7246 | STATIC void |
7247 | S__append_range_to_invlist(pTHX_ SV* const invlist, const UV start, const UV end) | |
d764b54e KW |
7248 | { |
7249 | /* Subject to change or removal. Append the range from 'start' to 'end' at | |
7250 | * the end of the inversion list. The range must be above any existing | |
7251 | * ones. */ | |
7252 | ||
f1b67122 | 7253 | UV* array; |
d764b54e | 7254 | UV max = invlist_max(invlist); |
29f51c6b | 7255 | UV len = _invlist_len(invlist); |
d764b54e KW |
7256 | |
7257 | PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST; | |
7258 | ||
f1b67122 KW |
7259 | if (len == 0) { /* Empty lists must be initialized */ |
7260 | array = _invlist_array_init(invlist, start == 0); | |
7261 | } | |
7262 | else { | |
d764b54e KW |
7263 | /* Here, the existing list is non-empty. The current max entry in the |
7264 | * list is generally the first value not in the set, except when the | |
7265 | * set extends to the end of permissible values, in which case it is | |
7266 | * the first entry in that final set, and so this call is an attempt to | |
7267 | * append out-of-order */ | |
7268 | ||
7269 | UV final_element = len - 1; | |
f1b67122 | 7270 | array = invlist_array(invlist); |
d764b54e | 7271 | if (array[final_element] > start |
4096c37b | 7272 | || ELEMENT_RANGE_MATCHES_INVLIST(final_element)) |
d764b54e | 7273 | { |
5637ef5b NC |
7274 | Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list, final=%"UVuf", start=%"UVuf", match=%c", |
7275 | array[final_element], start, | |
7276 | ELEMENT_RANGE_MATCHES_INVLIST(final_element) ? 't' : 'f'); | |
d764b54e KW |
7277 | } |
7278 | ||
7279 | /* Here, it is a legal append. If the new range begins with the first | |
7280 | * value not in the set, it is extending the set, so the new first | |
7281 | * value not in the set is one greater than the newly extended range. | |
7282 | * */ | |
7283 | if (array[final_element] == start) { | |
7284 | if (end != UV_MAX) { | |
7285 | array[final_element] = end + 1; | |
7286 | } | |
7287 | else { | |
7288 | /* But if the end is the maximum representable on the machine, | |
dbe7a391 | 7289 | * just let the range that this would extend to have no end */ |
d764b54e KW |
7290 | invlist_set_len(invlist, len - 1); |
7291 | } | |
7292 | return; | |
7293 | } | |
7294 | } | |
7295 | ||
7296 | /* Here the new range doesn't extend any existing set. Add it */ | |
7297 | ||
7298 | len += 2; /* Includes an element each for the start and end of range */ | |
7299 | ||
7300 | /* If overflows the existing space, extend, which may cause the array to be | |
7301 | * moved */ | |
7302 | if (max < len) { | |
7303 | invlist_extend(invlist, len); | |
f1b67122 KW |
7304 | invlist_set_len(invlist, len); /* Have to set len here to avoid assert |
7305 | failure in invlist_array() */ | |
d764b54e KW |
7306 | array = invlist_array(invlist); |
7307 | } | |
f1b67122 KW |
7308 | else { |
7309 | invlist_set_len(invlist, len); | |
7310 | } | |
d764b54e KW |
7311 | |
7312 | /* The next item on the list starts the range, the one after that is | |
7313 | * one past the new range. */ | |
7314 | array[len - 2] = start; | |
7315 | if (end != UV_MAX) { | |
7316 | array[len - 1] = end + 1; | |
7317 | } | |
7318 | else { | |
7319 | /* But if the end is the maximum representable on the machine, just let | |
7320 | * the range have no end */ | |
7321 | invlist_set_len(invlist, len - 1); | |
7322 | } | |
7323 | } | |
7324 | ||
ce057c8e KW |
7325 | #ifndef PERL_IN_XSUB_RE |
7326 | ||
7327 | IV | |
7328 | Perl__invlist_search(pTHX_ SV* const invlist, const UV cp) | |
d5e82ecc KW |
7329 | { |
7330 | /* Searches the inversion list for the entry that contains the input code | |
7331 | * point <cp>. If <cp> is not in the list, -1 is returned. Otherwise, the | |
7332 | * return value is the index into the list's array of the range that | |
7333 | * contains <cp> */ | |
7334 | ||
7335 | IV low = 0; | |
15896d2f | 7336 | IV mid; |
29f51c6b | 7337 | IV high = _invlist_len(invlist); |
dcb44827 KW |
7338 | const IV highest_element = high - 1; |
7339 | const UV* array; | |
d5e82ecc | 7340 | |
ce057c8e | 7341 | PERL_ARGS_ASSERT__INVLIST_SEARCH; |
d5e82ecc | 7342 | |
dcb44827 KW |
7343 | /* If list is empty, return failure. */ |
7344 | if (high == 0) { | |
d5e82ecc KW |
7345 | return -1; |
7346 | } | |
7347 | ||
dcb44827 KW |
7348 | /* If the code point is before the first element, return failure. (We |
7349 | * can't combine this with the test above, because we can't get the array | |
7350 | * unless we know the list is non-empty) */ | |
7351 | array = invlist_array(invlist); | |
15896d2f KW |
7352 | |
7353 | mid = invlist_previous_index(invlist); | |
7354 | assert(mid >=0 && mid <= highest_element); | |
7355 | ||
7356 | /* <mid> contains the cache of the result of the previous call to this | |
7357 | * function (0 the first time). See if this call is for the same result, | |
7358 | * or if it is for mid-1. This is under the theory that calls to this | |
7359 | * function will often be for related code points that are near each other. | |
7360 | * And benchmarks show that caching gives better results. We also test | |
7361 | * here if the code point is within the bounds of the list. These tests | |
7362 | * replace others that would have had to be made anyway to make sure that | |
7363 | * the array bounds were not exceeded, and give us extra information at the | |
7364 | * same time */ | |
7365 | if (cp >= array[mid]) { | |
7366 | if (cp >= array[highest_element]) { | |
7367 | return highest_element; | |
7368 | } | |
7369 | ||
7370 | /* Here, array[mid] <= cp < array[highest_element]. This means that | |
7371 | * the final element is not the answer, so can exclude it; it also | |
7372 | * means that <mid> is not the final element, so can refer to 'mid + 1' | |
7373 | * safely */ | |
7374 | if (cp < array[mid + 1]) { | |
7375 | return mid; | |
7376 | } | |
7377 | high--; | |
7378 | low = mid + 1; | |
7379 | } | |
7380 | else { /* cp < aray[mid] */ | |
7381 | if (cp < array[0]) { /* Fail if outside the array */ | |
7382 | return -1; | |
7383 | } | |
7384 | high = mid; | |
7385 | if (cp >= array[mid - 1]) { | |
7386 | goto found_entry; | |
7387 | } | |
dcb44827 KW |
7388 | } |
7389 | ||
d5e82ecc KW |
7390 | /* Binary search. What we are looking for is <i> such that |
7391 | * array[i] <= cp < array[i+1] | |
fc4f706d KW |
7392 | * The loop below converges on the i+1. Note that there may not be an |
7393 | * (i+1)th element in the array, and things work nonetheless */ | |
d5e82ecc | 7394 | while (low < high) { |
15896d2f | 7395 | mid = (low + high) / 2; |
fc4f706d KW |
7396 | assert(mid <= highest_element); |
7397 | if (array[mid] <= cp) { /* cp >= array[mid] */ | |
d5e82ecc KW |
7398 | low = mid + 1; |
7399 | ||
7400 | /* We could do this extra test to exit the loop early. | |
7401 | if (cp < array[low]) { | |
7402 | return mid; | |
7403 | } | |
7404 | */ | |
7405 | } | |
7406 | else { /* cp < array[mid] */ | |
7407 | high = mid; | |
7408 | } | |
7409 | } | |
7410 | ||
15896d2f KW |
7411 | found_entry: |
7412 | high--; | |
7413 | invlist_set_previous_index(invlist, high); | |
7414 | return high; | |
d5e82ecc KW |
7415 | } |
7416 | ||
86f766ab | 7417 | void |
b6a0ff33 KW |
7418 | Perl__invlist_populate_swatch(pTHX_ SV* const invlist, const UV start, const UV end, U8* swatch) |
7419 | { | |
7420 | /* populates a swatch of a swash the same way swatch_get() does in utf8.c, | |
7421 | * but is used when the swash has an inversion list. This makes this much | |
7422 | * faster, as it uses a binary search instead of a linear one. This is | |
7423 | * intimately tied to that function, and perhaps should be in utf8.c, | |
7424 | * except it is intimately tied to inversion lists as well. It assumes | |
7425 | * that <swatch> is all 0's on input */ | |
7426 | ||
7427 | UV current = start; | |
29f51c6b | 7428 | const IV len = _invlist_len(invlist); |
b6a0ff33 KW |
7429 | IV i; |
7430 | const UV * array; | |
7431 | ||
7432 | PERL_ARGS_ASSERT__INVLIST_POPULATE_SWATCH; | |
7433 | ||
7434 | if (len == 0) { /* Empty inversion list */ | |
7435 | return; | |
7436 | } | |
7437 | ||
7438 | array = invlist_array(invlist); | |
7439 | ||
7440 | /* Find which element it is */ | |
ce057c8e | 7441 | i = _invlist_search(invlist, start); |
b6a0ff33 KW |
7442 | |
7443 | /* We populate from <start> to <end> */ | |
7444 | while (current < end) { | |
7445 | UV upper; | |
7446 | ||
7447 | /* The inversion list gives the results for every possible code point | |
7448 | * after the first one in the list. Only those ranges whose index is | |
7449 | * even are ones that the inversion list matches. For the odd ones, | |
7450 | * and if the initial code point is not in the list, we have to skip | |
7451 | * forward to the next element */ | |
7452 | if (i == -1 || ! ELEMENT_RANGE_MATCHES_INVLIST(i)) { | |
7453 | i++; | |
7454 | if (i >= len) { /* Finished if beyond the end of the array */ | |
7455 | return; | |
7456 | } | |
7457 | current = array[i]; | |
7458 | if (current >= end) { /* Finished if beyond the end of what we | |
7459 | are populating */ | |
01c5845a KW |
7460 | if (LIKELY(end < UV_MAX)) { |
7461 | return; | |
7462 | } | |
7463 | ||
7464 | /* We get here when the upper bound is the maximum | |
7465 | * representable on the machine, and we are looking for just | |
7466 | * that code point. Have to special case it */ | |
7467 | i = len; | |
7468 | goto join_end_of_list; | |
b6a0ff33 KW |
7469 | } |
7470 | } | |
7471 | assert(current >= start); | |
7472 | ||
7473 | /* The current range ends one below the next one, except don't go past | |
7474 | * <end> */ | |
7475 | i++; | |
7476 | upper = (i < len && array[i] < end) ? array[i] : end; | |
7477 | ||
7478 | /* Here we are in a range that matches. Populate a bit in the 3-bit U8 | |
7479 | * for each code point in it */ | |
7480 | for (; current < upper; current++) { | |
7481 | const STRLEN offset = (STRLEN)(current - start); | |
7482 | swatch[offset >> 3] |= 1 << (offset & 7); | |
7483 | } | |
7484 | ||
01c5845a KW |
7485 | join_end_of_list: |
7486 | ||
b6a0ff33 KW |
7487 | /* Quit if at the end of the list */ |
7488 | if (i >= len) { | |
7489 | ||
7490 | /* But first, have to deal with the highest possible code point on | |
7491 | * the platform. The previous code assumes that <end> is one | |
7492 | * beyond where we want to populate, but that is impossible at the | |
7493 | * platform's infinity, so have to handle it specially */ | |
7494 | if (UNLIKELY(end == UV_MAX && ELEMENT_RANGE_MATCHES_INVLIST(len-1))) | |
7495 | { | |
7496 | const STRLEN offset = (STRLEN)(end - start); | |
7497 | swatch[offset >> 3] |= 1 << (offset & 7); | |
7498 | } | |
7499 | return; | |
7500 | } | |
7501 | ||
7502 | /* Advance to the next range, which will be for code points not in the | |
7503 | * inversion list */ | |
7504 | current = array[i]; | |
7505 | } | |
7506 | ||
7507 | return; | |
7508 | } | |
7509 | ||
7510 | void | |
164173a2 | 7511 | Perl__invlist_union_maybe_complement_2nd(pTHX_ SV* const a, SV* const b, bool complement_b, SV** output) |
d764b54e | 7512 | { |
4065ba03 KW |
7513 | /* Take the union of two inversion lists and point <output> to it. *output |
7514 | * should be defined upon input, and if it points to one of the two lists, | |
f45adb79 KW |
7515 | * the reference count to that list will be decremented. The first list, |
7516 | * <a>, may be NULL, in which case a copy of the second list is returned. | |
164173a2 KW |
7517 | * If <complement_b> is TRUE, the union is taken of the complement |
7518 | * (inversion) of <b> instead of b itself. | |
f45adb79 | 7519 | * |
d764b54e KW |
7520 | * The basis for this comes from "Unicode Demystified" Chapter 13 by |
7521 | * Richard Gillam, published by Addison-Wesley, and explained at some | |
7522 | * length there. The preface says to incorporate its examples into your | |
7523 | * code at your own risk. | |
7524 | * | |
7525 | * The algorithm is like a merge sort. | |
7526 | * | |
7527 | * XXX A potential performance improvement is to keep track as we go along | |
7528 | * if only one of the inputs contributes to the result, meaning the other | |
7529 | * is a subset of that one. In that case, we can skip the final copy and | |
a2995b7f KW |
7530 | * return the larger of the input lists, but then outside code might need |
7531 | * to keep track of whether to free the input list or not */ | |
d764b54e | 7532 | |
f1b67122 KW |
7533 | UV* array_a; /* a's array */ |
7534 | UV* array_b; | |
7535 | UV len_a; /* length of a's array */ | |
7536 | UV len_b; | |
d764b54e | 7537 | |
a25abddc | 7538 | SV* u; /* the resulting union */ |
d764b54e KW |
7539 | UV* array_u; |
7540 | UV len_u; | |
7541 | ||
7542 | UV i_a = 0; /* current index into a's array */ | |
7543 | UV i_b = 0; | |
7544 | UV i_u = 0; | |
7545 | ||
7546 | /* running count, as explained in the algorithm source book; items are | |
7547 | * stopped accumulating and are output when the count changes to/from 0. | |
7548 | * The count is incremented when we start a range that's in the set, and | |
7549 | * decremented when we start a range that's not in the set. So its range | |
7550 | * is 0 to 2. Only when the count is zero is something not in the set. | |
7551 | */ | |
7552 | UV count = 0; | |
7553 | ||
164173a2 | 7554 | PERL_ARGS_ASSERT__INVLIST_UNION_MAYBE_COMPLEMENT_2ND; |
60825692 | 7555 | assert(a != b); |
d764b54e | 7556 | |
f1b67122 | 7557 | /* If either one is empty, the union is the other one */ |
29f51c6b | 7558 | if (a == NULL || ((len_a = _invlist_len(a)) == 0)) { |
4065ba03 | 7559 | if (*output == a) { |
f45adb79 KW |
7560 | if (a != NULL) { |
7561 | SvREFCNT_dec(a); | |
7562 | } | |
f1b67122 | 7563 | } |
4065ba03 | 7564 | if (*output != b) { |
f1b67122 | 7565 | *output = invlist_clone(b); |
164173a2 KW |
7566 | if (complement_b) { |
7567 | _invlist_invert(*output); | |
7568 | } | |
dbe7a391 | 7569 | } /* else *output already = b; */ |
f1b67122 KW |
7570 | return; |
7571 | } | |
29f51c6b | 7572 | else if ((len_b = _invlist_len(b)) == 0) { |
4065ba03 | 7573 | if (*output == b) { |
f1b67122 KW |
7574 | SvREFCNT_dec(b); |
7575 | } | |
164173a2 KW |
7576 | |
7577 | /* The complement of an empty list is a list that has everything in it, | |
7578 | * so the union with <a> includes everything too */ | |
7579 | if (complement_b) { | |
7580 | if (a == *output) { | |
7581 | SvREFCNT_dec(a); | |
7582 | } | |
7583 | *output = _new_invlist(1); | |
7584 | _append_range_to_invlist(*output, 0, UV_MAX); | |
7585 | } | |
7586 | else if (*output != a) { | |
7587 | *output = invlist_clone(a); | |
7588 | } | |
7589 | /* else *output already = a; */ | |
f1b67122 KW |
7590 | return; |
7591 | } | |
7592 | ||
7593 | /* Here both lists exist and are non-empty */ | |
7594 | array_a = invlist_array(a); | |
7595 | array_b = invlist_array(b); | |
7596 | ||
164173a2 KW |
7597 | /* If are to take the union of 'a' with the complement of b, set it |
7598 | * up so are looking at b's complement. */ | |
7599 | if (complement_b) { | |
7600 | ||
7601 | /* To complement, we invert: if the first element is 0, remove it. To | |
7602 | * do this, we just pretend the array starts one later, and clear the | |
7603 | * flag as we don't have to do anything else later */ | |
7604 | if (array_b[0] == 0) { | |
7605 | array_b++; | |
7606 | len_b--; | |
7607 | complement_b = FALSE; | |
7608 | } | |
7609 | else { | |
7610 | ||
7611 | /* But if the first element is not zero, we unshift a 0 before the | |
7612 | * array. The data structure reserves a space for that 0 (which | |
7613 | * should be a '1' right now), so physical shifting is unneeded, | |
7614 | * but temporarily change that element to 0. Before exiting the | |
7615 | * routine, we must restore the element to '1' */ | |
7616 | array_b--; | |
7617 | len_b++; | |
7618 | array_b[0] = 0; | |
7619 | } | |
7620 | } | |
7621 | ||
d764b54e KW |
7622 | /* Size the union for the worst case: that the sets are completely |
7623 | * disjoint */ | |
7624 | u = _new_invlist(len_a + len_b); | |
f1b67122 KW |
7625 | |
7626 | /* Will contain U+0000 if either component does */ | |
7627 | array_u = _invlist_array_init(u, (len_a > 0 && array_a[0] == 0) | |
7628 | || (len_b > 0 && array_b[0] == 0)); | |
d764b54e KW |
7629 | |
7630 | /* Go through each list item by item, stopping when exhausted one of | |
7631 | * them */ | |
7632 | while (i_a < len_a && i_b < len_b) { | |
7633 | UV cp; /* The element to potentially add to the union's array */ | |
7634 | bool cp_in_set; /* is it in the the input list's set or not */ | |
7635 | ||
7636 | /* We need to take one or the other of the two inputs for the union. | |
7637 | * Since we are merging two sorted lists, we take the smaller of the | |
7638 | * next items. In case of a tie, we take the one that is in its set | |
7639 | * first. If we took one not in the set first, it would decrement the | |
7640 | * count, possibly to 0 which would cause it to be output as ending the | |
7641 | * range, and the next time through we would take the same number, and | |
7642 | * output it again as beginning the next range. By doing it the | |
7643 | * opposite way, there is no possibility that the count will be | |
7644 | * momentarily decremented to 0, and thus the two adjoining ranges will | |
7645 | * be seamlessly merged. (In a tie and both are in the set or both not | |
7646 | * in the set, it doesn't matter which we take first.) */ | |
7647 | if (array_a[i_a] < array_b[i_b] | |
4096c37b KW |
7648 | || (array_a[i_a] == array_b[i_b] |
7649 | && ELEMENT_RANGE_MATCHES_INVLIST(i_a))) | |
d764b54e | 7650 | { |
4096c37b | 7651 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_a); |
d764b54e KW |
7652 | cp= array_a[i_a++]; |
7653 | } | |
7654 | else { | |
4096c37b | 7655 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_b); |
d764b54e KW |
7656 | cp= array_b[i_b++]; |
7657 | } | |
7658 | ||
7659 | /* Here, have chosen which of the two inputs to look at. Only output | |
7660 | * if the running count changes to/from 0, which marks the | |
7661 | * beginning/end of a range in that's in the set */ | |
7662 | if (cp_in_set) { | |
7663 | if (count == 0) { | |
7664 | array_u[i_u++] = cp; | |
7665 | } | |
7666 | count++; | |
7667 | } | |
7668 | else { | |
7669 | count--; | |
7670 | if (count == 0) { | |
7671 | array_u[i_u++] = cp; | |
7672 | } | |
7673 | } | |
7674 | } | |
7675 | ||
7676 | /* Here, we are finished going through at least one of the lists, which | |
7677 | * means there is something remaining in at most one. We check if the list | |
7678 | * that hasn't been exhausted is positioned such that we are in the middle | |
bac5f0ae KW |
7679 | * of a range in its set or not. (i_a and i_b point to the element beyond |
7680 | * the one we care about.) If in the set, we decrement 'count'; if 0, there | |
7681 | * is potentially more to output. | |
d764b54e KW |
7682 | * There are four cases: |
7683 | * 1) Both weren't in their sets, count is 0, and remains 0. What's left | |
7684 | * in the union is entirely from the non-exhausted set. | |
7685 | * 2) Both were in their sets, count is 2. Nothing further should | |
7686 | * be output, as everything that remains will be in the exhausted | |
7687 | * list's set, hence in the union; decrementing to 1 but not 0 insures | |
7688 | * that | |
7689 | * 3) the exhausted was in its set, non-exhausted isn't, count is 1. | |
7690 | * Nothing further should be output because the union includes | |
bac5f0ae | 7691 | * everything from the exhausted set. Not decrementing ensures that. |
d764b54e KW |
7692 | * 4) the exhausted wasn't in its set, non-exhausted is, count is 1; |
7693 | * decrementing to 0 insures that we look at the remainder of the | |
7694 | * non-exhausted set */ | |
4096c37b KW |
7695 | if ((i_a != len_a && PREV_RANGE_MATCHES_INVLIST(i_a)) |
7696 | || (i_b != len_b && PREV_RANGE_MATCHES_INVLIST(i_b))) | |
d764b54e KW |
7697 | { |
7698 | count--; | |
7699 | } | |
7700 | ||
7701 | /* The final length is what we've output so far, plus what else is about to | |
7702 | * be output. (If 'count' is non-zero, then the input list we exhausted | |
7703 | * has everything remaining up to the machine's limit in its set, and hence | |
7704 | * in the union, so there will be no further output. */ | |
7705 | len_u = i_u; | |
7706 | if (count == 0) { | |
7707 | /* At most one of the subexpressions will be non-zero */ | |
7708 | len_u += (len_a - i_a) + (len_b - i_b); | |
7709 | } | |
7710 | ||
7711 | /* Set result to final length, which can change the pointer to array_u, so | |
7712 | * re-find it */ | |
29f51c6b | 7713 | if (len_u != _invlist_len(u)) { |
d764b54e KW |
7714 | invlist_set_len(u, len_u); |
7715 | invlist_trim(u); | |
7716 | array_u = invlist_array(u); | |
7717 | } | |
7718 | ||
7719 | /* When 'count' is 0, the list that was exhausted (if one was shorter than | |
7720 | * the other) ended with everything above it not in its set. That means | |
7721 | * that the remaining part of the union is precisely the same as the | |
7722 | * non-exhausted list, so can just copy it unchanged. (If both list were | |
7723 | * exhausted at the same time, then the operations below will be both 0.) | |
7724 | */ | |
7725 | if (count == 0) { | |
7726 | IV copy_count; /* At most one will have a non-zero copy count */ | |
7727 | if ((copy_count = len_a - i_a) > 0) { | |
7728 | Copy(array_a + i_a, array_u + i_u, copy_count, UV); | |
7729 | } | |
7730 | else if ((copy_count = len_b - i_b) > 0) { | |
7731 | Copy(array_b + i_b, array_u + i_u, copy_count, UV); | |
7732 | } | |
7733 | } | |
7734 | ||
a2995b7f | 7735 | /* We may be removing a reference to one of the inputs */ |
4065ba03 | 7736 | if (a == *output || b == *output) { |
a2995b7f KW |
7737 | SvREFCNT_dec(*output); |
7738 | } | |
7739 | ||
164173a2 KW |
7740 | /* If we've changed b, restore it */ |
7741 | if (complement_b) { | |
7742 | array_b[0] = 1; | |
7743 | } | |
7744 | ||
a2995b7f KW |
7745 | *output = u; |
7746 | return; | |
d764b54e KW |
7747 | } |
7748 | ||
86f766ab | 7749 | void |
52ae8f7e | 7750 | Perl__invlist_intersection_maybe_complement_2nd(pTHX_ SV* const a, SV* const b, bool complement_b, SV** i) |
d764b54e | 7751 | { |
4065ba03 KW |
7752 | /* Take the intersection of two inversion lists and point <i> to it. *i |
7753 | * should be defined upon input, and if it points to one of the two lists, | |
7754 | * the reference count to that list will be decremented. | |
52ae8f7e KW |
7755 | * If <complement_b> is TRUE, the result will be the intersection of <a> |
7756 | * and the complement (or inversion) of <b> instead of <b> directly. | |
7757 | * | |
a2995b7f KW |
7758 | * The basis for this comes from "Unicode Demystified" Chapter 13 by |
7759 | * Richard Gillam, published by Addison-Wesley, and explained at some | |
7760 | * length there. The preface says to incorporate its examples into your | |
7761 | * code at your own risk. In fact, it had bugs | |
d764b54e KW |
7762 | * |
7763 | * The algorithm is like a merge sort, and is essentially the same as the | |
7764 | * union above | |
7765 | */ | |
7766 | ||
f1b67122 KW |
7767 | UV* array_a; /* a's array */ |
7768 | UV* array_b; | |
7769 | UV len_a; /* length of a's array */ | |
7770 | UV len_b; | |
d764b54e | 7771 | |
a25abddc | 7772 | SV* r; /* the resulting intersection */ |
d764b54e KW |
7773 | UV* array_r; |
7774 | UV len_r; | |
7775 | ||
7776 | UV i_a = 0; /* current index into a's array */ | |
7777 | UV i_b = 0; | |
7778 | UV i_r = 0; | |
7779 | ||
7780 | /* running count, as explained in the algorithm source book; items are | |
7781 | * stopped accumulating and are output when the count changes to/from 2. | |
7782 | * The count is incremented when we start a range that's in the set, and | |
7783 | * decremented when we start a range that's not in the set. So its range | |
7784 | * is 0 to 2. Only when the count is 2 is something in the intersection. | |
7785 | */ | |
7786 | UV count = 0; | |
7787 | ||
52ae8f7e | 7788 | PERL_ARGS_ASSERT__INVLIST_INTERSECTION_MAYBE_COMPLEMENT_2ND; |
60825692 | 7789 | assert(a != b); |
d764b54e | 7790 | |
52ae8f7e | 7791 | /* Special case if either one is empty */ |
29f51c6b KW |
7792 | len_a = _invlist_len(a); |
7793 | if ((len_a == 0) || ((len_b = _invlist_len(b)) == 0)) { | |
f1b67122 | 7794 | |
52ae8f7e KW |
7795 | if (len_a != 0 && complement_b) { |
7796 | ||
7797 | /* Here, 'a' is not empty, therefore from the above 'if', 'b' must | |
7798 | * be empty. Here, also we are using 'b's complement, which hence | |
7799 | * must be every possible code point. Thus the intersection is | |
7800 | * simply 'a'. */ | |
7801 | if (*i != a) { | |
7802 | *i = invlist_clone(a); | |
7803 | ||
7804 | if (*i == b) { | |
7805 | SvREFCNT_dec(b); | |
7806 | } | |
7807 | } | |
7808 | /* else *i is already 'a' */ | |
7809 | return; | |
7810 | } | |
7811 | ||
7812 | /* Here, 'a' or 'b' is empty and not using the complement of 'b'. The | |
7813 | * intersection must be empty */ | |
4065ba03 | 7814 | if (*i == a) { |
f1b67122 KW |
7815 | SvREFCNT_dec(a); |
7816 | } | |
4065ba03 | 7817 | else if (*i == b) { |
f1b67122 KW |
7818 | SvREFCNT_dec(b); |
7819 | } | |
2ea86699 | 7820 | *i = _new_invlist(0); |
f1b67122 KW |
7821 | return; |
7822 | } | |
7823 | ||
7824 | /* Here both lists exist and are non-empty */ | |
7825 | array_a = invlist_array(a); | |
7826 | array_b = invlist_array(b); | |
7827 | ||
52ae8f7e KW |
7828 | /* If are to take the intersection of 'a' with the complement of b, set it |
7829 | * up so are looking at b's complement. */ | |
7830 | if (complement_b) { | |
7831 | ||
7832 | /* To complement, we invert: if the first element is 0, remove it. To | |
7833 | * do this, we just pretend the array starts one later, and clear the | |
7834 | * flag as we don't have to do anything else later */ | |
7835 | if (array_b[0] == 0) { | |
7836 | array_b++; | |
7837 | len_b--; | |
7838 | complement_b = FALSE; | |
7839 | } | |
7840 | else { | |
7841 | ||
7842 | /* But if the first element is not zero, we unshift a 0 before the | |
7843 | * array. The data structure reserves a space for that 0 (which | |
7844 | * should be a '1' right now), so physical shifting is unneeded, | |
7845 | * but temporarily change that element to 0. Before exiting the | |
7846 | * routine, we must restore the element to '1' */ | |
7847 | array_b--; | |
7848 | len_b++; | |
7849 | array_b[0] = 0; | |
7850 | } | |
7851 | } | |
7852 | ||
d764b54e KW |
7853 | /* Size the intersection for the worst case: that the intersection ends up |
7854 | * fragmenting everything to be completely disjoint */ | |
7855 | r= _new_invlist(len_a + len_b); | |
f1b67122 KW |
7856 | |
7857 | /* Will contain U+0000 iff both components do */ | |
7858 | array_r = _invlist_array_init(r, len_a > 0 && array_a[0] == 0 | |
7859 | && len_b > 0 && array_b[0] == 0); | |
d764b54e KW |
7860 | |
7861 | /* Go through each list item by item, stopping when exhausted one of | |
7862 | * them */ | |
7863 | while (i_a < len_a && i_b < len_b) { | |
7864 | UV cp; /* The element to potentially add to the intersection's | |
7865 | array */ | |
7866 | bool cp_in_set; /* Is it in the input list's set or not */ | |
7867 | ||
c4a30257 KW |
7868 | /* We need to take one or the other of the two inputs for the |
7869 | * intersection. Since we are merging two sorted lists, we take the | |
7870 | * smaller of the next items. In case of a tie, we take the one that | |
7871 | * is not in its set first (a difference from the union algorithm). If | |
7872 | * we took one in the set first, it would increment the count, possibly | |
7873 | * to 2 which would cause it to be output as starting a range in the | |
7874 | * intersection, and the next time through we would take that same | |
7875 | * number, and output it again as ending the set. By doing it the | |
7876 | * opposite of this, there is no possibility that the count will be | |
7877 | * momentarily incremented to 2. (In a tie and both are in the set or | |
7878 | * both not in the set, it doesn't matter which we take first.) */ | |
d764b54e | 7879 | if (array_a[i_a] < array_b[i_b] |
4096c37b KW |
7880 | || (array_a[i_a] == array_b[i_b] |
7881 | && ! ELEMENT_RANGE_MATCHES_INVLIST(i_a))) | |
d764b54e | 7882 | { |
4096c37b | 7883 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_a); |
d764b54e KW |
7884 | cp= array_a[i_a++]; |
7885 | } | |
7886 | else { | |
4096c37b | 7887 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_b); |
d764b54e KW |
7888 | cp= array_b[i_b++]; |
7889 | } | |
7890 | ||
7891 | /* Here, have chosen which of the two inputs to look at. Only output | |
7892 | * if the running count changes to/from 2, which marks the | |
7893 | * beginning/end of a range that's in the intersection */ | |
7894 | if (cp_in_set) { | |
7895 | count++; | |
7896 | if (count == 2) { | |
7897 | array_r[i_r++] = cp; | |
7898 | } | |
7899 | } | |
7900 | else { | |
7901 | if (count == 2) { | |
7902 | array_r[i_r++] = cp; | |
7903 | } | |
7904 | count--; | |
7905 | } | |
7906 | } | |
7907 | ||
c4a30257 KW |
7908 | /* Here, we are finished going through at least one of the lists, which |
7909 | * means there is something remaining in at most one. We check if the list | |
7910 | * that has been exhausted is positioned such that we are in the middle | |
7911 | * of a range in its set or not. (i_a and i_b point to elements 1 beyond | |
7912 | * the ones we care about.) There are four cases: | |
7913 | * 1) Both weren't in their sets, count is 0, and remains 0. There's | |
7914 | * nothing left in the intersection. | |
7915 | * 2) Both were in their sets, count is 2 and perhaps is incremented to | |
7916 | * above 2. What should be output is exactly that which is in the | |
7917 | * non-exhausted set, as everything it has is also in the intersection | |
7918 | * set, and everything it doesn't have can't be in the intersection | |
7919 | * 3) The exhausted was in its set, non-exhausted isn't, count is 1, and | |
7920 | * gets incremented to 2. Like the previous case, the intersection is | |
7921 | * everything that remains in the non-exhausted set. | |
7922 | * 4) the exhausted wasn't in its set, non-exhausted is, count is 1, and | |
7923 | * remains 1. And the intersection has nothing more. */ | |
4096c37b KW |
7924 | if ((i_a == len_a && PREV_RANGE_MATCHES_INVLIST(i_a)) |
7925 | || (i_b == len_b && PREV_RANGE_MATCHES_INVLIST(i_b))) | |
d764b54e | 7926 | { |
c4a30257 | 7927 | count++; |
d764b54e KW |
7928 | } |
7929 | ||
7930 | /* The final length is what we've output so far plus what else is in the | |
c4a30257 | 7931 | * intersection. At most one of the subexpressions below will be non-zero */ |
d764b54e | 7932 | len_r = i_r; |
c4a30257 | 7933 | if (count >= 2) { |
d764b54e KW |
7934 | len_r += (len_a - i_a) + (len_b - i_b); |
7935 | } | |
7936 | ||
7937 | /* Set result to final length, which can change the pointer to array_r, so | |
7938 | * re-find it */ | |
29f51c6b | 7939 | if (len_r != _invlist_len(r)) { |
d764b54e KW |
7940 | invlist_set_len(r, len_r); |
7941 | invlist_trim(r); | |
7942 | array_r = invlist_array(r); | |
7943 | } | |
7944 | ||
7945 | /* Finish outputting any remaining */ | |
c4a30257 | 7946 | if (count >= 2) { /* At most one will have a non-zero copy count */ |
d764b54e KW |
7947 | IV copy_count; |
7948 | if ((copy_count = len_a - i_a) > 0) { | |
7949 | Copy(array_a + i_a, array_r + i_r, copy_count, UV); | |
7950 | } | |
7951 | else if ((copy_count = len_b - i_b) > 0) { | |
7952 | Copy(array_b + i_b, array_r + i_r, copy_count, UV); | |
7953 | } | |
7954 | } | |
7955 | ||
a2995b7f | 7956 | /* We may be removing a reference to one of the inputs */ |
4065ba03 | 7957 | if (a == *i || b == *i) { |
a2995b7f KW |
7958 | SvREFCNT_dec(*i); |
7959 | } | |
7960 | ||
52ae8f7e KW |
7961 | /* If we've changed b, restore it */ |
7962 | if (complement_b) { | |
7963 | array_b[0] = 1; | |
7964 | } | |
7965 | ||
a2995b7f KW |
7966 | *i = r; |
7967 | return; | |
d764b54e KW |
7968 | } |
7969 | ||
9d501133 KW |
7970 | SV* |
7971 | Perl__add_range_to_invlist(pTHX_ SV* invlist, const UV start, const UV end) | |
d764b54e KW |
7972 | { |
7973 | /* Add the range from 'start' to 'end' inclusive to the inversion list's | |
7974 | * set. A pointer to the inversion list is returned. This may actually be | |
c52a3e71 KW |
7975 | * a new list, in which case the passed in one has been destroyed. The |
7976 | * passed in inversion list can be NULL, in which case a new one is created | |
7977 | * with just the one range in it */ | |
d764b54e | 7978 | |
a25abddc | 7979 | SV* range_invlist; |
c52a3e71 | 7980 | UV len; |
d764b54e | 7981 | |
c52a3e71 KW |
7982 | if (invlist == NULL) { |
7983 | invlist = _new_invlist(2); | |
7984 | len = 0; | |
7985 | } | |
7986 | else { | |
29f51c6b | 7987 | len = _invlist_len(invlist); |
c52a3e71 | 7988 | } |
d764b54e KW |
7989 | |
7990 | /* If comes after the final entry, can just append it to the end */ | |
7991 | if (len == 0 | |
7992 | || start >= invlist_array(invlist) | |
29f51c6b | 7993 | [_invlist_len(invlist) - 1]) |
d764b54e KW |
7994 | { |
7995 | _append_range_to_invlist(invlist, start, end); | |
7996 | return invlist; | |
7997 | } | |
7998 | ||
7999 | /* Here, can't just append things, create and return a new inversion list | |
8000 | * which is the union of this range and the existing inversion list */ | |
8001 | range_invlist = _new_invlist(2); | |
8002 | _append_range_to_invlist(range_invlist, start, end); | |
8003 | ||
37e85ffe | 8004 | _invlist_union(invlist, range_invlist, &invlist); |
d764b54e | 8005 | |
0a89af2f | 8006 | /* The temporary can be freed */ |
318c430e | 8007 | SvREFCNT_dec(range_invlist); |
d764b54e | 8008 | |
6d63a9fb | 8009 | return invlist; |
d764b54e KW |
8010 | } |
8011 | ||
9d501133 KW |
8012 | #endif |
8013 | ||
a25abddc KW |
8014 | PERL_STATIC_INLINE SV* |
8015 | S_add_cp_to_invlist(pTHX_ SV* invlist, const UV cp) { | |
9d501133 | 8016 | return _add_range_to_invlist(invlist, cp, cp); |
c229b64c KW |
8017 | } |
8018 | ||
3c234b35 | 8019 | #ifndef PERL_IN_XSUB_RE |
86f766ab KW |
8020 | void |
8021 | Perl__invlist_invert(pTHX_ SV* const invlist) | |
25e94a65 KW |
8022 | { |
8023 | /* Complement the input inversion list. This adds a 0 if the list didn't | |
8024 | * have a zero; removes it otherwise. As described above, the data | |
8025 | * structure is set up so that this is very efficient */ | |
8026 | ||
29f51c6b | 8027 | UV* len_pos = _get_invlist_len_addr(invlist); |
25e94a65 | 8028 | |
37e85ffe | 8029 | PERL_ARGS_ASSERT__INVLIST_INVERT; |
25e94a65 KW |
8030 | |
8031 | /* The inverse of matching nothing is matching everything */ | |
8032 | if (*len_pos == 0) { | |
8033 | _append_range_to_invlist(invlist, 0, UV_MAX); | |
8034 | return; | |
8035 | } | |
8036 | ||
8037 | /* The exclusive or complents 0 to 1; and 1 to 0. If the result is 1, the | |
8038 | * zero element was a 0, so it is being removed, so the length decrements | |
8039 | * by 1; and vice-versa. SvCUR is unaffected */ | |
8040 | if (*get_invlist_zero_addr(invlist) ^= 1) { | |
8041 | (*len_pos)--; | |
8042 | } | |
8043 | else { | |
8044 | (*len_pos)++; | |
8045 | } | |
8046 | } | |
89302fc2 KW |
8047 | |
8048 | void | |
8049 | Perl__invlist_invert_prop(pTHX_ SV* const invlist) | |
8050 | { | |
8051 | /* Complement the input inversion list (which must be a Unicode property, | |
8052 | * all of which don't match above the Unicode maximum code point.) And | |
8053 | * Perl has chosen to not have the inversion match above that either. This | |
8054 | * adds a 0x110000 if the list didn't end with it, and removes it if it did | |
8055 | */ | |
8056 | ||
8057 | UV len; | |
8058 | UV* array; | |
8059 | ||
8060 | PERL_ARGS_ASSERT__INVLIST_INVERT_PROP; | |
8061 | ||
8062 | _invlist_invert(invlist); | |
8063 | ||
29f51c6b | 8064 | len = _invlist_len(invlist); |
89302fc2 KW |
8065 | |
8066 | if (len != 0) { /* If empty do nothing */ | |
8067 | array = invlist_array(invlist); | |
8068 | if (array[len - 1] != PERL_UNICODE_MAX + 1) { | |
8069 | /* Add 0x110000. First, grow if necessary */ | |
8070 | len++; | |
8071 | if (invlist_max(invlist) < len) { | |
8072 | invlist_extend(invlist, len); | |
8073 | array = invlist_array(invlist); | |
8074 | } | |
8075 | invlist_set_len(invlist, len); | |
8076 | array[len - 1] = PERL_UNICODE_MAX + 1; | |
8077 | } | |
8078 | else { /* Remove the 0x110000 */ | |
8079 | invlist_set_len(invlist, len - 1); | |
8080 | } | |
8081 | } | |
8082 | ||
8083 | return; | |
8084 | } | |
3c234b35 | 8085 | #endif |
25e94a65 KW |
8086 | |
8087 | PERL_STATIC_INLINE SV* | |
8088 | S_invlist_clone(pTHX_ SV* const invlist) | |
8089 | { | |
8090 | ||
8091 | /* Return a new inversion list that is a copy of the input one, which is | |
8092 | * unchanged */ | |
8093 | ||
6c6c83ac KW |
8094 | /* Need to allocate extra space to accommodate Perl's addition of a |
8095 | * trailing NUL to SvPV's, since it thinks they are always strings */ | |
29f51c6b | 8096 | SV* new_invlist = _new_invlist(_invlist_len(invlist) + 1); |
6d47fb3d | 8097 | STRLEN length = SvCUR(invlist); |
25e94a65 KW |
8098 | |
8099 | PERL_ARGS_ASSERT_INVLIST_CLONE; | |
8100 | ||
6d47fb3d KW |
8101 | SvCUR_set(new_invlist, length); /* This isn't done automatically */ |
8102 | Copy(SvPVX(invlist), SvPVX(new_invlist), length, char); | |
8103 | ||
25e94a65 KW |
8104 | return new_invlist; |
8105 | } | |
8106 | ||
f3dc70d1 KW |
8107 | PERL_STATIC_INLINE UV* |
8108 | S_get_invlist_iter_addr(pTHX_ SV* invlist) | |
8109 | { | |
8110 | /* Return the address of the UV that contains the current iteration | |
8111 | * position */ | |
8112 | ||
8113 | PERL_ARGS_ASSERT_GET_INVLIST_ITER_ADDR; | |
8114 | ||
8115 | return (UV *) (SvPVX(invlist) + (INVLIST_ITER_OFFSET * sizeof (UV))); | |
8116 | } | |
8117 | ||
f59ff194 KW |
8118 | PERL_STATIC_INLINE UV* |
8119 | S_get_invlist_version_id_addr(pTHX_ SV* invlist) | |
8120 | { | |
8121 | /* Return the address of the UV that contains the version id. */ | |
8122 | ||
8123 | PERL_ARGS_ASSERT_GET_INVLIST_VERSION_ID_ADDR; | |
8124 | ||
8125 | return (UV *) (SvPVX(invlist) + (INVLIST_VERSION_ID_OFFSET * sizeof (UV))); | |
8126 | } | |
8127 | ||
f3dc70d1 KW |
8128 | PERL_STATIC_INLINE void |
8129 | S_invlist_iterinit(pTHX_ SV* invlist) /* Initialize iterator for invlist */ | |
8130 | { | |
8131 | PERL_ARGS_ASSERT_INVLIST_ITERINIT; | |
8132 | ||
8133 | *get_invlist_iter_addr(invlist) = 0; | |
8134 | } | |
8135 | ||
8136 | STATIC bool | |
8137 | S_invlist_iternext(pTHX_ SV* invlist, UV* start, UV* end) | |
8138 | { | |
dbe7a391 KW |
8139 | /* An C<invlist_iterinit> call on <invlist> must be used to set this up. |
8140 | * This call sets in <*start> and <*end>, the next range in <invlist>. | |
8141 | * Returns <TRUE> if successful and the next call will return the next | |
8142 | * range; <FALSE> if was already at the end of the list. If the latter, | |
8143 | * <*start> and <*end> are unchanged, and the next call to this function | |
8144 | * will start over at the beginning of the list */ | |
8145 | ||
f3dc70d1 | 8146 | UV* pos = get_invlist_iter_addr(invlist); |
29f51c6b | 8147 | UV len = _invlist_len(invlist); |
f3dc70d1 KW |
8148 | UV *array; |
8149 | ||
8150 | PERL_ARGS_ASSERT_INVLIST_ITERNEXT; | |
8151 | ||
8152 | if (*pos >= len) { | |
8153 | *pos = UV_MAX; /* Force iternit() to be required next time */ | |
8154 | return FALSE; | |
8155 | } | |
8156 | ||
8157 | array = invlist_array(invlist); | |
8158 | ||
8159 | *start = array[(*pos)++]; | |
8160 | ||
8161 | if (*pos >= len) { | |
8162 | *end = UV_MAX; | |
8163 | } | |
8164 | else { | |
8165 | *end = array[(*pos)++] - 1; | |
8166 | } | |
8167 | ||
8168 | return TRUE; | |
8169 | } | |
8170 | ||
4f3e8b0f KW |
8171 | PERL_STATIC_INLINE UV |
8172 | S_invlist_highest(pTHX_ SV* const invlist) | |
8173 | { | |
8174 | /* Returns the highest code point that matches an inversion list. This API | |
8175 | * has an ambiguity, as it returns 0 under either the highest is actually | |
8176 | * 0, or if the list is empty. If this distinction matters to you, check | |
8177 | * for emptiness before calling this function */ | |
8178 | ||
29f51c6b | 8179 | UV len = _invlist_len(invlist); |
4f3e8b0f KW |
8180 | UV *array; |
8181 | ||
8182 | PERL_ARGS_ASSERT_INVLIST_HIGHEST; | |
8183 | ||
8184 | if (len == 0) { | |
8185 | return 0; | |
8186 | } | |
8187 | ||
8188 | array = invlist_array(invlist); | |
8189 | ||
8190 | /* The last element in the array in the inversion list always starts a | |
8191 | * range that goes to infinity. That range may be for code points that are | |
8192 | * matched in the inversion list, or it may be for ones that aren't | |
8193 | * matched. In the latter case, the highest code point in the set is one | |
8194 | * less than the beginning of this range; otherwise it is the final element | |
8195 | * of this range: infinity */ | |
8196 | return (ELEMENT_RANGE_MATCHES_INVLIST(len - 1)) | |
8197 | ? UV_MAX | |
8198 | : array[len - 1] - 1; | |
8199 | } | |
8200 | ||
b2b97e77 KW |
8201 | #ifndef PERL_IN_XSUB_RE |
8202 | SV * | |
8203 | Perl__invlist_contents(pTHX_ SV* const invlist) | |
8204 | { | |
8205 | /* Get the contents of an inversion list into a string SV so that they can | |
8206 | * be printed out. It uses the format traditionally done for debug tracing | |
8207 | */ | |
8208 | ||
8209 | UV start, end; | |
8210 | SV* output = newSVpvs("\n"); | |
8211 | ||
8212 | PERL_ARGS_ASSERT__INVLIST_CONTENTS; | |
8213 | ||
8214 | invlist_iterinit(invlist); | |
8215 | while (invlist_iternext(invlist, &start, &end)) { | |
8216 | if (end == UV_MAX) { | |
8217 | Perl_sv_catpvf(aTHX_ output, "%04"UVXf"\tINFINITY\n", start); | |
8218 | } | |
8219 | else if (end != start) { | |
8220 | Perl_sv_catpvf(aTHX_ output, "%04"UVXf"\t%04"UVXf"\n", | |
8221 | start, end); | |
8222 | } | |
8223 | else { | |
8224 | Perl_sv_catpvf(aTHX_ output, "%04"UVXf"\n", start); | |
8225 | } | |
8226 | } | |
8227 | ||
8228 | return output; | |
8229 | } | |
8230 | #endif | |
8231 | ||
768318b8 KW |
8232 | #if 0 |
8233 | void | |
8234 | S_invlist_dump(pTHX_ SV* const invlist, const char * const header) | |
8235 | { | |
8236 | /* Dumps out the ranges in an inversion list. The string 'header' | |
8237 | * if present is output on a line before the first range */ | |
8238 | ||
8239 | UV start, end; | |
8240 | ||
8241 | if (header && strlen(header)) { | |
8242 | PerlIO_printf(Perl_debug_log, "%s\n", header); | |
8243 | } | |
8244 | invlist_iterinit(invlist); | |
8245 | while (invlist_iternext(invlist, &start, &end)) { | |
8246 | if (end == UV_MAX) { | |
8247 | PerlIO_printf(Perl_debug_log, "0x%04"UVXf" .. INFINITY\n", start); | |
8248 | } | |
8249 | else { | |
8250 | PerlIO_printf(Perl_debug_log, "0x%04"UVXf" .. 0x%04"UVXf"\n", start, end); | |
8251 | } | |
8252 | } | |
8253 | } | |
8254 | #endif | |
8255 | ||
76a181d9 KW |
8256 | #if 0 |
8257 | bool | |
8258 | S__invlistEQ(pTHX_ SV* const a, SV* const b, bool complement_b) | |
8259 | { | |
8260 | /* Return a boolean as to if the two passed in inversion lists are | |
8261 | * identical. The final argument, if TRUE, says to take the complement of | |
8262 | * the second inversion list before doing the comparison */ | |
8263 | ||
8264 | UV* array_a = invlist_array(a); | |
8265 | UV* array_b = invlist_array(b); | |
29f51c6b KW |
8266 | UV len_a = _invlist_len(a); |
8267 | UV len_b = _invlist_len(b); | |
76a181d9 KW |
8268 | |
8269 | UV i = 0; /* current index into the arrays */ | |
8270 | bool retval = TRUE; /* Assume are identical until proven otherwise */ | |
8271 | ||
8272 | PERL_ARGS_ASSERT__INVLISTEQ; | |
8273 | ||
8274 | /* If are to compare 'a' with the complement of b, set it | |
8275 | * up so are looking at b's complement. */ | |
8276 | if (complement_b) { | |
8277 | ||
8278 | /* The complement of nothing is everything, so <a> would have to have | |
8279 | * just one element, starting at zero (ending at infinity) */ | |
8280 | if (len_b == 0) { | |
8281 | return (len_a == 1 && array_a[0] == 0); | |
8282 | } | |
8283 | else if (array_b[0] == 0) { | |
8284 | ||
8285 | /* Otherwise, to complement, we invert. Here, the first element is | |
8286 | * 0, just remove it. To do this, we just pretend the array starts | |
8287 | * one later, and clear the flag as we don't have to do anything | |
8288 | * else later */ | |
8289 | ||
8290 | array_b++; | |
8291 | len_b--; | |
8292 | complement_b = FALSE; | |
8293 | } | |
8294 | else { | |
8295 | ||
8296 | /* But if the first element is not zero, we unshift a 0 before the | |
8297 | * array. The data structure reserves a space for that 0 (which | |
8298 | * should be a '1' right now), so physical shifting is unneeded, | |
8299 | * but temporarily change that element to 0. Before exiting the | |
8300 | * routine, we must restore the element to '1' */ | |
8301 | array_b--; | |
8302 | len_b++; | |
8303 | array_b[0] = 0; | |
8304 | } | |
8305 | } | |
8306 | ||
8307 | /* Make sure that the lengths are the same, as well as the final element | |
8308 | * before looping through the remainder. (Thus we test the length, final, | |
8309 | * and first elements right off the bat) */ | |
8310 | if (len_a != len_b || array_a[len_a-1] != array_b[len_a-1]) { | |
8311 | retval = FALSE; | |
8312 | } | |
8313 | else for (i = 0; i < len_a - 1; i++) { | |
8314 | if (array_a[i] != array_b[i]) { | |
8315 | retval = FALSE; | |
8316 | break; | |
8317 | } | |
8318 | } | |
8319 | ||
8320 | if (complement_b) { | |
8321 | array_b[0] = 1; | |
8322 | } | |
8323 | return retval; | |
8324 | } | |
8325 | #endif | |
8326 | ||
97b14ce7 | 8327 | #undef HEADER_LENGTH |
060b7a35 | 8328 | #undef INVLIST_INITIAL_LENGTH |
005b65ed KW |
8329 | #undef TO_INTERNAL_SIZE |
8330 | #undef FROM_INTERNAL_SIZE | |
f1b67122 KW |
8331 | #undef INVLIST_LEN_OFFSET |
8332 | #undef INVLIST_ZERO_OFFSET | |
f3dc70d1 | 8333 | #undef INVLIST_ITER_OFFSET |
f59ff194 | 8334 | #undef INVLIST_VERSION_ID |
060b7a35 | 8335 | |
d764b54e KW |
8336 | /* End of inversion list object */ |
8337 | ||
a687059c LW |
8338 | /* |
8339 | - reg - regular expression, i.e. main body or parenthesized thing | |
8340 | * | |
8341 | * Caller must absorb opening parenthesis. | |
8342 | * | |
8343 | * Combining parenthesis handling with the base level of regular expression | |
8344 | * is a trifle forced, but the need to tie the tails of the branches to what | |
8345 | * follows makes it hard to avoid. | |
8346 | */ | |
07be1b83 YO |
8347 | #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1) |
8348 | #ifdef DEBUGGING | |
8349 | #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1) | |
8350 | #else | |
8351 | #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1) | |
8352 | #endif | |
3dab1dad | 8353 | |
76e3520e | 8354 | STATIC regnode * |
3dab1dad | 8355 | S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth) |
c277df42 | 8356 | /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */ |
a687059c | 8357 | { |
27da23d5 | 8358 | dVAR; |
eb578fdb KW |
8359 | regnode *ret; /* Will be the head of the group. */ |
8360 | regnode *br; | |
8361 | regnode *lastbr; | |
8362 | regnode *ender = NULL; | |
8363 | I32 parno = 0; | |
cbbf8932 | 8364 | I32 flags; |
f7819f85 | 8365 | U32 oregflags = RExC_flags; |
6136c704 AL |
8366 | bool have_branch = 0; |
8367 | bool is_open = 0; | |
594d7033 YO |
8368 | I32 freeze_paren = 0; |
8369 | I32 after_freeze = 0; | |
9d1d55b5 JP |
8370 | |
8371 | /* for (?g), (?gc), and (?o) warnings; warning | |
8372 | about (?c) will warn about (?g) -- japhy */ | |
8373 | ||
6136c704 AL |
8374 | #define WASTED_O 0x01 |
8375 | #define WASTED_G 0x02 | |
8376 | #define WASTED_C 0x04 | |
8377 | #define WASTED_GC (0x02|0x04) | |
cbbf8932 | 8378 | I32 wastedflags = 0x00; |
9d1d55b5 | 8379 | |
fac92740 | 8380 | char * parse_start = RExC_parse; /* MJD */ |
a28509cc | 8381 | char * const oregcomp_parse = RExC_parse; |
a0d0e21e | 8382 | |
3dab1dad | 8383 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
8384 | |
8385 | PERL_ARGS_ASSERT_REG; | |
3dab1dad YO |
8386 | DEBUG_PARSE("reg "); |
8387 | ||
821b33a5 | 8388 | *flagp = 0; /* Tentatively. */ |
a0d0e21e | 8389 | |
9d1d55b5 | 8390 | |
a0d0e21e LW |
8391 | /* Make an OPEN node, if parenthesized. */ |
8392 | if (paren) { | |
e2e6a0f1 YO |
8393 | if ( *RExC_parse == '*') { /* (*VERB:ARG) */ |
8394 | char *start_verb = RExC_parse; | |
8395 | STRLEN verb_len = 0; | |
8396 | char *start_arg = NULL; | |
8397 | unsigned char op = 0; | |
8398 | int argok = 1; | |
8399 | int internal_argval = 0; /* internal_argval is only useful if !argok */ | |
8400 | while ( *RExC_parse && *RExC_parse != ')' ) { | |
8401 | if ( *RExC_parse == ':' ) { | |
8402 | start_arg = RExC_parse + 1; | |
8403 | break; | |
8404 | } | |
8405 | RExC_parse++; | |
8406 | } | |
8407 | ++start_verb; | |
8408 | verb_len = RExC_parse - start_verb; | |
8409 | if ( start_arg ) { | |
8410 | RExC_parse++; | |
8411 | while ( *RExC_parse && *RExC_parse != ')' ) | |
8412 | RExC_parse++; | |
8413 | if ( *RExC_parse != ')' ) | |
8414 | vFAIL("Unterminated verb pattern argument"); | |
8415 | if ( RExC_parse == start_arg ) | |
8416 | start_arg = NULL; | |
8417 | } else { | |
8418 | if ( *RExC_parse != ')' ) | |
8419 | vFAIL("Unterminated verb pattern"); | |
8420 | } | |
5d458dd8 | 8421 | |
e2e6a0f1 YO |
8422 | switch ( *start_verb ) { |
8423 | case 'A': /* (*ACCEPT) */ | |
568a785a | 8424 | if ( memEQs(start_verb,verb_len,"ACCEPT") ) { |
e2e6a0f1 YO |
8425 | op = ACCEPT; |
8426 | internal_argval = RExC_nestroot; | |
8427 | } | |
8428 | break; | |
8429 | case 'C': /* (*COMMIT) */ | |
568a785a | 8430 | if ( memEQs(start_verb,verb_len,"COMMIT") ) |
e2e6a0f1 | 8431 | op = COMMIT; |
e2e6a0f1 YO |
8432 | break; |
8433 | case 'F': /* (*FAIL) */ | |
568a785a | 8434 | if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) { |
e2e6a0f1 YO |
8435 | op = OPFAIL; |
8436 | argok = 0; | |
8437 | } | |
8438 | break; | |
5d458dd8 YO |
8439 | case ':': /* (*:NAME) */ |
8440 | case 'M': /* (*MARK:NAME) */ | |
568a785a | 8441 | if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) { |
e2e6a0f1 | 8442 | op = MARKPOINT; |
5d458dd8 YO |
8443 | argok = -1; |
8444 | } | |
8445 | break; | |
8446 | case 'P': /* (*PRUNE) */ | |
568a785a | 8447 | if ( memEQs(start_verb,verb_len,"PRUNE") ) |
5d458dd8 | 8448 | op = PRUNE; |
e2e6a0f1 | 8449 | break; |
5d458dd8 | 8450 | case 'S': /* (*SKIP) */ |
568a785a | 8451 | if ( memEQs(start_verb,verb_len,"SKIP") ) |
5d458dd8 YO |
8452 | op = SKIP; |
8453 | break; | |
8454 | case 'T': /* (*THEN) */ | |
8455 | /* [19:06] <TimToady> :: is then */ | |
568a785a | 8456 | if ( memEQs(start_verb,verb_len,"THEN") ) { |
5d458dd8 YO |
8457 | op = CUTGROUP; |
8458 | RExC_seen |= REG_SEEN_CUTGROUP; | |
8459 | } | |
e2e6a0f1 YO |
8460 | break; |
8461 | } | |
8462 | if ( ! op ) { | |
8463 | RExC_parse++; | |
8464 | vFAIL3("Unknown verb pattern '%.*s'", | |
8465 | verb_len, start_verb); | |
8466 | } | |
8467 | if ( argok ) { | |
8468 | if ( start_arg && internal_argval ) { | |
8469 | vFAIL3("Verb pattern '%.*s' may not have an argument", | |
8470 | verb_len, start_verb); | |
8471 | } else if ( argok < 0 && !start_arg ) { | |
8472 | vFAIL3("Verb pattern '%.*s' has a mandatory argument", | |
8473 | verb_len, start_verb); | |
8474 | } else { | |
8475 | ret = reganode(pRExC_state, op, internal_argval); | |
8476 | if ( ! internal_argval && ! SIZE_ONLY ) { | |
8477 | if (start_arg) { | |
8478 | SV *sv = newSVpvn( start_arg, RExC_parse - start_arg); | |
8479 | ARG(ret) = add_data( pRExC_state, 1, "S" ); | |
f8fc2ecf | 8480 | RExC_rxi->data->data[ARG(ret)]=(void*)sv; |
e2e6a0f1 YO |
8481 | ret->flags = 0; |
8482 | } else { | |
8483 | ret->flags = 1; | |
8484 | } | |
8485 | } | |
8486 | } | |
8487 | if (!internal_argval) | |
8488 | RExC_seen |= REG_SEEN_VERBARG; | |
8489 | } else if ( start_arg ) { | |
8490 | vFAIL3("Verb pattern '%.*s' may not have an argument", | |
8491 | verb_len, start_verb); | |
8492 | } else { | |
8493 | ret = reg_node(pRExC_state, op); | |
8494 | } | |
8495 | nextchar(pRExC_state); | |
8496 | return ret; | |
8497 | } else | |
fac92740 | 8498 | if (*RExC_parse == '?') { /* (?...) */ |
6136c704 | 8499 | bool is_logical = 0; |
a28509cc | 8500 | const char * const seqstart = RExC_parse; |
fb85c044 | 8501 | bool has_use_defaults = FALSE; |
ca9dfc88 | 8502 | |
830247a4 IZ |
8503 | RExC_parse++; |
8504 | paren = *RExC_parse++; | |
c277df42 | 8505 | ret = NULL; /* For look-ahead/behind. */ |
a0d0e21e | 8506 | switch (paren) { |
894be9b7 | 8507 | |
1f1031fe YO |
8508 | case 'P': /* (?P...) variants for those used to PCRE/Python */ |
8509 | paren = *RExC_parse++; | |
8510 | if ( paren == '<') /* (?P<...>) named capture */ | |
8511 | goto named_capture; | |
8512 | else if (paren == '>') { /* (?P>name) named recursion */ | |
8513 | goto named_recursion; | |
8514 | } | |
8515 | else if (paren == '=') { /* (?P=...) named backref */ | |
8516 | /* this pretty much dupes the code for \k<NAME> in regatom(), if | |
8517 | you change this make sure you change that */ | |
8518 | char* name_start = RExC_parse; | |
8519 | U32 num = 0; | |
8520 | SV *sv_dat = reg_scan_name(pRExC_state, | |
8521 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
8522 | if (RExC_parse == name_start || *RExC_parse != ')') | |
8523 | vFAIL2("Sequence %.3s... not terminated",parse_start); | |
8524 | ||
8525 | if (!SIZE_ONLY) { | |
8526 | num = add_data( pRExC_state, 1, "S" ); | |
8527 | RExC_rxi->data->data[num]=(void*)sv_dat; | |
5a5094bd | 8528 | SvREFCNT_inc_simple_void(sv_dat); |
1f1031fe YO |
8529 | } |
8530 | RExC_sawback = 1; | |
4444fd9f KW |
8531 | ret = reganode(pRExC_state, |
8532 | ((! FOLD) | |
8533 | ? NREF | |
a725e29c | 8534 | : (ASCII_FOLD_RESTRICTED) |
2f7f8cb1 KW |
8535 | ? NREFFA |
8536 | : (AT_LEAST_UNI_SEMANTICS) | |
8537 | ? NREFFU | |
8538 | : (LOC) | |
8539 | ? NREFFL | |
8540 | : NREFF), | |
4444fd9f | 8541 | num); |
1f1031fe YO |
8542 | *flagp |= HASWIDTH; |
8543 | ||
8544 | Set_Node_Offset(ret, parse_start+1); | |
8545 | Set_Node_Cur_Length(ret); /* MJD */ | |
8546 | ||
8547 | nextchar(pRExC_state); | |
8548 | return ret; | |
8549 | } | |
57b84237 YO |
8550 | RExC_parse++; |
8551 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
8552 | /*NOTREACHED*/ | |
8553 | case '<': /* (?<...) */ | |
b81d288d | 8554 | if (*RExC_parse == '!') |
c277df42 | 8555 | paren = ','; |
0a4db386 | 8556 | else if (*RExC_parse != '=') |
1f1031fe | 8557 | named_capture: |
0a4db386 | 8558 | { /* (?<...>) */ |
81714fb9 | 8559 | char *name_start; |
894be9b7 | 8560 | SV *svname; |
81714fb9 YO |
8561 | paren= '>'; |
8562 | case '\'': /* (?'...') */ | |
8563 | name_start= RExC_parse; | |
0a4db386 YO |
8564 | svname = reg_scan_name(pRExC_state, |
8565 | SIZE_ONLY ? /* reverse test from the others */ | |
8566 | REG_RSN_RETURN_NAME : | |
8567 | REG_RSN_RETURN_NULL); | |
57b84237 YO |
8568 | if (RExC_parse == name_start) { |
8569 | RExC_parse++; | |
8570 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
8571 | /*NOTREACHED*/ | |
8572 | } | |
81714fb9 YO |
8573 | if (*RExC_parse != paren) |
8574 | vFAIL2("Sequence (?%c... not terminated", | |
8575 | paren=='>' ? '<' : paren); | |
8576 | if (SIZE_ONLY) { | |
e62cc96a YO |
8577 | HE *he_str; |
8578 | SV *sv_dat = NULL; | |
486ec47a | 8579 | if (!svname) /* shouldn't happen */ |
894be9b7 YO |
8580 | Perl_croak(aTHX_ |
8581 | "panic: reg_scan_name returned NULL"); | |
81714fb9 YO |
8582 | if (!RExC_paren_names) { |
8583 | RExC_paren_names= newHV(); | |
ad64d0ec | 8584 | sv_2mortal(MUTABLE_SV(RExC_paren_names)); |
1f1031fe YO |
8585 | #ifdef DEBUGGING |
8586 | RExC_paren_name_list= newAV(); | |
ad64d0ec | 8587 | sv_2mortal(MUTABLE_SV(RExC_paren_name_list)); |
1f1031fe | 8588 | #endif |
81714fb9 YO |
8589 | } |
8590 | he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 ); | |
e62cc96a | 8591 | if ( he_str ) |
81714fb9 | 8592 | sv_dat = HeVAL(he_str); |
e62cc96a | 8593 | if ( ! sv_dat ) { |
81714fb9 | 8594 | /* croak baby croak */ |
e62cc96a YO |
8595 | Perl_croak(aTHX_ |
8596 | "panic: paren_name hash element allocation failed"); | |
8597 | } else if ( SvPOK(sv_dat) ) { | |
76a476f9 YO |
8598 | /* (?|...) can mean we have dupes so scan to check |
8599 | its already been stored. Maybe a flag indicating | |
8600 | we are inside such a construct would be useful, | |
8601 | but the arrays are likely to be quite small, so | |
8602 | for now we punt -- dmq */ | |
8603 | IV count = SvIV(sv_dat); | |
8604 | I32 *pv = (I32*)SvPVX(sv_dat); | |
8605 | IV i; | |
8606 | for ( i = 0 ; i < count ; i++ ) { | |
8607 | if ( pv[i] == RExC_npar ) { | |
8608 | count = 0; | |
8609 | break; | |
8610 | } | |
8611 | } | |
8612 | if ( count ) { | |
8613 | pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1); | |
8614 | SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32)); | |
8615 | pv[count] = RExC_npar; | |
3a92e6ae | 8616 | SvIV_set(sv_dat, SvIVX(sv_dat) + 1); |
76a476f9 | 8617 | } |
81714fb9 YO |
8618 | } else { |
8619 | (void)SvUPGRADE(sv_dat,SVt_PVNV); | |
8620 | sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32)); | |
8621 | SvIOK_on(sv_dat); | |
3ec35e0f | 8622 | SvIV_set(sv_dat, 1); |
e62cc96a | 8623 | } |
1f1031fe | 8624 | #ifdef DEBUGGING |
17a3c617 | 8625 | /* Yes this does cause a memory leak in debugging Perls */ |
1f1031fe YO |
8626 | if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname))) |
8627 | SvREFCNT_dec(svname); | |
8628 | #endif | |
e62cc96a | 8629 | |
81714fb9 YO |
8630 | /*sv_dump(sv_dat);*/ |
8631 | } | |
8632 | nextchar(pRExC_state); | |
8633 | paren = 1; | |
8634 | goto capturing_parens; | |
8635 | } | |
8636 | RExC_seen |= REG_SEEN_LOOKBEHIND; | |
b57e4118 | 8637 | RExC_in_lookbehind++; |
830247a4 | 8638 | RExC_parse++; |
fac92740 | 8639 | case '=': /* (?=...) */ |
89c6a13e | 8640 | RExC_seen_zerolen++; |
5c3fa2e7 | 8641 | break; |
fac92740 | 8642 | case '!': /* (?!...) */ |
830247a4 | 8643 | RExC_seen_zerolen++; |
e2e6a0f1 YO |
8644 | if (*RExC_parse == ')') { |
8645 | ret=reg_node(pRExC_state, OPFAIL); | |
8646 | nextchar(pRExC_state); | |
8647 | return ret; | |
8648 | } | |
594d7033 YO |
8649 | break; |
8650 | case '|': /* (?|...) */ | |
8651 | /* branch reset, behave like a (?:...) except that | |
8652 | buffers in alternations share the same numbers */ | |
8653 | paren = ':'; | |
8654 | after_freeze = freeze_paren = RExC_npar; | |
8655 | break; | |
fac92740 MJD |
8656 | case ':': /* (?:...) */ |
8657 | case '>': /* (?>...) */ | |
a0d0e21e | 8658 | break; |
fac92740 MJD |
8659 | case '$': /* (?$...) */ |
8660 | case '@': /* (?@...) */ | |
8615cb43 | 8661 | vFAIL2("Sequence (?%c...) not implemented", (int)paren); |
a0d0e21e | 8662 | break; |
fac92740 | 8663 | case '#': /* (?#...) */ |
830247a4 IZ |
8664 | while (*RExC_parse && *RExC_parse != ')') |
8665 | RExC_parse++; | |
8666 | if (*RExC_parse != ')') | |
c277df42 | 8667 | FAIL("Sequence (?#... not terminated"); |
830247a4 | 8668 | nextchar(pRExC_state); |
a0d0e21e LW |
8669 | *flagp = TRYAGAIN; |
8670 | return NULL; | |
894be9b7 YO |
8671 | case '0' : /* (?0) */ |
8672 | case 'R' : /* (?R) */ | |
8673 | if (*RExC_parse != ')') | |
6bda09f9 | 8674 | FAIL("Sequence (?R) not terminated"); |
1a147d38 | 8675 | ret = reg_node(pRExC_state, GOSTART); |
a3b492c3 | 8676 | *flagp |= POSTPONED; |
7f69552c YO |
8677 | nextchar(pRExC_state); |
8678 | return ret; | |
8679 | /*notreached*/ | |
894be9b7 YO |
8680 | { /* named and numeric backreferences */ |
8681 | I32 num; | |
894be9b7 YO |
8682 | case '&': /* (?&NAME) */ |
8683 | parse_start = RExC_parse - 1; | |
1f1031fe | 8684 | named_recursion: |
894be9b7 | 8685 | { |
0a4db386 YO |
8686 | SV *sv_dat = reg_scan_name(pRExC_state, |
8687 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
8688 | num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0; | |
894be9b7 YO |
8689 | } |
8690 | goto gen_recurse_regop; | |
118e2215 | 8691 | assert(0); /* NOT REACHED */ |
542fa716 YO |
8692 | case '+': |
8693 | if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) { | |
8694 | RExC_parse++; | |
8695 | vFAIL("Illegal pattern"); | |
8696 | } | |
8697 | goto parse_recursion; | |
8698 | /* NOT REACHED*/ | |
8699 | case '-': /* (?-1) */ | |
8700 | if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) { | |
8701 | RExC_parse--; /* rewind to let it be handled later */ | |
8702 | goto parse_flags; | |
8703 | } | |
8704 | /*FALLTHROUGH */ | |
6bda09f9 YO |
8705 | case '1': case '2': case '3': case '4': /* (?1) */ |
8706 | case '5': case '6': case '7': case '8': case '9': | |
8707 | RExC_parse--; | |
542fa716 | 8708 | parse_recursion: |
894be9b7 YO |
8709 | num = atoi(RExC_parse); |
8710 | parse_start = RExC_parse - 1; /* MJD */ | |
542fa716 YO |
8711 | if (*RExC_parse == '-') |
8712 | RExC_parse++; | |
6bda09f9 YO |
8713 | while (isDIGIT(*RExC_parse)) |
8714 | RExC_parse++; | |
8715 | if (*RExC_parse!=')') | |
8716 | vFAIL("Expecting close bracket"); | |
686b73d4 | 8717 | |
894be9b7 | 8718 | gen_recurse_regop: |
542fa716 YO |
8719 | if ( paren == '-' ) { |
8720 | /* | |
8721 | Diagram of capture buffer numbering. | |
8722 | Top line is the normal capture buffer numbers | |
3b753521 | 8723 | Bottom line is the negative indexing as from |
542fa716 YO |
8724 | the X (the (?-2)) |
8725 | ||
8726 | + 1 2 3 4 5 X 6 7 | |
8727 | /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/ | |
8728 | - 5 4 3 2 1 X x x | |
8729 | ||
8730 | */ | |
8731 | num = RExC_npar + num; | |
8732 | if (num < 1) { | |
8733 | RExC_parse++; | |
8734 | vFAIL("Reference to nonexistent group"); | |
8735 | } | |
8736 | } else if ( paren == '+' ) { | |
8737 | num = RExC_npar + num - 1; | |
8738 | } | |
8739 | ||
1a147d38 | 8740 | ret = reganode(pRExC_state, GOSUB, num); |
6bda09f9 YO |
8741 | if (!SIZE_ONLY) { |
8742 | if (num > (I32)RExC_rx->nparens) { | |
8743 | RExC_parse++; | |
8744 | vFAIL("Reference to nonexistent group"); | |
8745 | } | |
40d049e4 | 8746 | ARG2L_SET( ret, RExC_recurse_count++); |
6bda09f9 | 8747 | RExC_emit++; |
226de585 | 8748 | DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log, |
acff02b8 | 8749 | "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret))); |
894be9b7 | 8750 | } else { |
6bda09f9 | 8751 | RExC_size++; |
6bda09f9 | 8752 | } |
0a4db386 | 8753 | RExC_seen |= REG_SEEN_RECURSE; |
6bda09f9 | 8754 | Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */ |
58663417 RGS |
8755 | Set_Node_Offset(ret, parse_start); /* MJD */ |
8756 | ||
a3b492c3 | 8757 | *flagp |= POSTPONED; |
6bda09f9 YO |
8758 | nextchar(pRExC_state); |
8759 | return ret; | |
894be9b7 | 8760 | } /* named and numeric backreferences */ |
118e2215 | 8761 | assert(0); /* NOT REACHED */ |
894be9b7 | 8762 | |
fac92740 | 8763 | case '?': /* (??...) */ |
6136c704 | 8764 | is_logical = 1; |
57b84237 YO |
8765 | if (*RExC_parse != '{') { |
8766 | RExC_parse++; | |
8767 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
8768 | /*NOTREACHED*/ | |
8769 | } | |
a3b492c3 | 8770 | *flagp |= POSTPONED; |
830247a4 | 8771 | paren = *RExC_parse++; |
0f5d15d6 | 8772 | /* FALL THROUGH */ |
fac92740 | 8773 | case '{': /* (?{...}) */ |
c277df42 | 8774 | { |
2eccd3b2 | 8775 | U32 n = 0; |
d24ca0c5 | 8776 | struct reg_code_block *cb; |
c277df42 | 8777 | |
830247a4 | 8778 | RExC_seen_zerolen++; |
68e2671b | 8779 | |
d24ca0c5 DM |
8780 | if ( !pRExC_state->num_code_blocks |
8781 | || pRExC_state->code_index >= pRExC_state->num_code_blocks | |
8782 | || pRExC_state->code_blocks[pRExC_state->code_index].start | |
8783 | != (STRLEN)((RExC_parse -3 - (is_logical ? 1 : 0)) | |
68e2671b DM |
8784 | - RExC_start) |
8785 | ) { | |
d24ca0c5 DM |
8786 | if (RExC_pm_flags & PMf_USE_RE_EVAL) |
8787 | FAIL("panic: Sequence (?{...}): no code block found\n"); | |
8788 | FAIL("Eval-group not allowed at runtime, use re 'eval'"); | |
c277df42 | 8789 | } |
d24ca0c5 DM |
8790 | /* this is a pre-compiled code block (?{...}) */ |
8791 | cb = &pRExC_state->code_blocks[pRExC_state->code_index]; | |
8792 | RExC_parse = RExC_start + cb->end; | |
5f616ea7 | 8793 | if (!SIZE_ONLY) { |
d24ca0c5 DM |
8794 | OP *o = cb->block; |
8795 | if (cb->src_regex) { | |
8796 | n = add_data(pRExC_state, 2, "rl"); | |
8797 | RExC_rxi->data->data[n] = | |
8798 | (void*)SvREFCNT_inc((SV*)cb->src_regex); | |
2e2e3f36 | 8799 | RExC_rxi->data->data[n+1] = (void*)o; |
68e2671b | 8800 | } |
d24ca0c5 DM |
8801 | else { |
8802 | n = add_data(pRExC_state, 1, | |
8803 | (RExC_pm_flags & PMf_HAS_CV) ? "L" : "l"); | |
2e2e3f36 | 8804 | RExC_rxi->data->data[n] = (void*)o; |
68e2671b | 8805 | } |
c277df42 | 8806 | } |
d24ca0c5 | 8807 | pRExC_state->code_index++; |
830247a4 | 8808 | nextchar(pRExC_state); |
68e2671b | 8809 | |
6136c704 | 8810 | if (is_logical) { |
ec841a27 | 8811 | regnode *eval; |
830247a4 | 8812 | ret = reg_node(pRExC_state, LOGICAL); |
ec841a27 DM |
8813 | eval = reganode(pRExC_state, EVAL, n); |
8814 | if (!SIZE_ONLY) { | |
0f5d15d6 | 8815 | ret->flags = 2; |
ec841a27 | 8816 | /* for later propagation into (??{}) return value */ |
db703679 | 8817 | eval->flags = (U8) (RExC_flags & RXf_PMf_COMPILETIME); |
ec841a27 DM |
8818 | } |
8819 | REGTAIL(pRExC_state, ret, eval); | |
fac92740 | 8820 | /* deal with the length of this later - MJD */ |
0f5d15d6 IZ |
8821 | return ret; |
8822 | } | |
ccb2c380 MP |
8823 | ret = reganode(pRExC_state, EVAL, n); |
8824 | Set_Node_Length(ret, RExC_parse - parse_start + 1); | |
8825 | Set_Node_Offset(ret, parse_start); | |
8826 | return ret; | |
c277df42 | 8827 | } |
fac92740 | 8828 | case '(': /* (?(?{...})...) and (?(?=...)...) */ |
c277df42 | 8829 | { |
0a4db386 | 8830 | int is_define= 0; |
fac92740 | 8831 | if (RExC_parse[0] == '?') { /* (?(?...)) */ |
b81d288d AB |
8832 | if (RExC_parse[1] == '=' || RExC_parse[1] == '!' |
8833 | || RExC_parse[1] == '<' | |
830247a4 | 8834 | || RExC_parse[1] == '{') { /* Lookahead or eval. */ |
c277df42 | 8835 | I32 flag; |
686b73d4 | 8836 | |
830247a4 | 8837 | ret = reg_node(pRExC_state, LOGICAL); |
0f5d15d6 IZ |
8838 | if (!SIZE_ONLY) |
8839 | ret->flags = 1; | |
3dab1dad | 8840 | REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1)); |
c277df42 | 8841 | goto insert_if; |
b81d288d | 8842 | } |
a0ed51b3 | 8843 | } |
0a4db386 YO |
8844 | else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */ |
8845 | || RExC_parse[0] == '\'' ) /* (?('NAME')...) */ | |
8846 | { | |
8847 | char ch = RExC_parse[0] == '<' ? '>' : '\''; | |
8848 | char *name_start= RExC_parse++; | |
2eccd3b2 | 8849 | U32 num = 0; |
0a4db386 YO |
8850 | SV *sv_dat=reg_scan_name(pRExC_state, |
8851 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
8852 | if (RExC_parse == name_start || *RExC_parse != ch) | |
8853 | vFAIL2("Sequence (?(%c... not terminated", | |
8854 | (ch == '>' ? '<' : ch)); | |
8855 | RExC_parse++; | |
8856 | if (!SIZE_ONLY) { | |
8857 | num = add_data( pRExC_state, 1, "S" ); | |
f8fc2ecf | 8858 | RExC_rxi->data->data[num]=(void*)sv_dat; |
5a5094bd | 8859 | SvREFCNT_inc_simple_void(sv_dat); |
0a4db386 YO |
8860 | } |
8861 | ret = reganode(pRExC_state,NGROUPP,num); | |
8862 | goto insert_if_check_paren; | |
8863 | } | |
8864 | else if (RExC_parse[0] == 'D' && | |
8865 | RExC_parse[1] == 'E' && | |
8866 | RExC_parse[2] == 'F' && | |
8867 | RExC_parse[3] == 'I' && | |
8868 | RExC_parse[4] == 'N' && | |
8869 | RExC_parse[5] == 'E') | |
8870 | { | |
8871 | ret = reganode(pRExC_state,DEFINEP,0); | |
8872 | RExC_parse +=6 ; | |
8873 | is_define = 1; | |
8874 | goto insert_if_check_paren; | |
8875 | } | |
8876 | else if (RExC_parse[0] == 'R') { | |
8877 | RExC_parse++; | |
8878 | parno = 0; | |
8879 | if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) { | |
8880 | parno = atoi(RExC_parse++); | |
8881 | while (isDIGIT(*RExC_parse)) | |
8882 | RExC_parse++; | |
8883 | } else if (RExC_parse[0] == '&') { | |
8884 | SV *sv_dat; | |
8885 | RExC_parse++; | |
8886 | sv_dat = reg_scan_name(pRExC_state, | |
8887 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
8888 | parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0; | |
8889 | } | |
1a147d38 | 8890 | ret = reganode(pRExC_state,INSUBP,parno); |
0a4db386 YO |
8891 | goto insert_if_check_paren; |
8892 | } | |
830247a4 | 8893 | else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) { |
fac92740 | 8894 | /* (?(1)...) */ |
6136c704 | 8895 | char c; |
830247a4 | 8896 | parno = atoi(RExC_parse++); |
c277df42 | 8897 | |
830247a4 IZ |
8898 | while (isDIGIT(*RExC_parse)) |
8899 | RExC_parse++; | |
fac92740 | 8900 | ret = reganode(pRExC_state, GROUPP, parno); |
2af232bd | 8901 | |
0a4db386 | 8902 | insert_if_check_paren: |
830247a4 | 8903 | if ((c = *nextchar(pRExC_state)) != ')') |
b45f050a | 8904 | vFAIL("Switch condition not recognized"); |
c277df42 | 8905 | insert_if: |
3dab1dad YO |
8906 | REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0)); |
8907 | br = regbranch(pRExC_state, &flags, 1,depth+1); | |
c277df42 | 8908 | if (br == NULL) |
830247a4 | 8909 | br = reganode(pRExC_state, LONGJMP, 0); |
c277df42 | 8910 | else |
3dab1dad | 8911 | REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0)); |
830247a4 | 8912 | c = *nextchar(pRExC_state); |
d1b80229 IZ |
8913 | if (flags&HASWIDTH) |
8914 | *flagp |= HASWIDTH; | |
c277df42 | 8915 | if (c == '|') { |
0a4db386 YO |
8916 | if (is_define) |
8917 | vFAIL("(?(DEFINE)....) does not allow branches"); | |
830247a4 | 8918 | lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */ |
3dab1dad YO |
8919 | regbranch(pRExC_state, &flags, 1,depth+1); |
8920 | REGTAIL(pRExC_state, ret, lastbr); | |
d1b80229 IZ |
8921 | if (flags&HASWIDTH) |
8922 | *flagp |= HASWIDTH; | |
830247a4 | 8923 | c = *nextchar(pRExC_state); |
a0ed51b3 LW |
8924 | } |
8925 | else | |
c277df42 IZ |
8926 | lastbr = NULL; |
8927 | if (c != ')') | |
8615cb43 | 8928 | vFAIL("Switch (?(condition)... contains too many branches"); |
830247a4 | 8929 | ender = reg_node(pRExC_state, TAIL); |
3dab1dad | 8930 | REGTAIL(pRExC_state, br, ender); |
c277df42 | 8931 | if (lastbr) { |
3dab1dad YO |
8932 | REGTAIL(pRExC_state, lastbr, ender); |
8933 | REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); | |
a0ed51b3 LW |
8934 | } |
8935 | else | |
3dab1dad | 8936 | REGTAIL(pRExC_state, ret, ender); |
3b57cd43 YO |
8937 | RExC_size++; /* XXX WHY do we need this?!! |
8938 | For large programs it seems to be required | |
8939 | but I can't figure out why. -- dmq*/ | |
c277df42 | 8940 | return ret; |
a0ed51b3 LW |
8941 | } |
8942 | else { | |
830247a4 | 8943 | vFAIL2("Unknown switch condition (?(%.2s", RExC_parse); |
c277df42 IZ |
8944 | } |
8945 | } | |
1b1626e4 | 8946 | case 0: |
830247a4 | 8947 | RExC_parse--; /* for vFAIL to print correctly */ |
8615cb43 | 8948 | vFAIL("Sequence (? incomplete"); |
1b1626e4 | 8949 | break; |
85508812 KW |
8950 | case DEFAULT_PAT_MOD: /* Use default flags with the exceptions |
8951 | that follow */ | |
fb85c044 KW |
8952 | has_use_defaults = TRUE; |
8953 | STD_PMMOD_FLAGS_CLEAR(&RExC_flags); | |
e40e74fe KW |
8954 | set_regex_charset(&RExC_flags, (RExC_utf8 || RExC_uni_semantics) |
8955 | ? REGEX_UNICODE_CHARSET | |
8956 | : REGEX_DEPENDS_CHARSET); | |
fb85c044 | 8957 | goto parse_flags; |
a0d0e21e | 8958 | default: |
cde0cee5 YO |
8959 | --RExC_parse; |
8960 | parse_flags: /* (?i) */ | |
8961 | { | |
8962 | U32 posflags = 0, negflags = 0; | |
8963 | U32 *flagsp = &posflags; | |
f6a766d5 | 8964 | char has_charset_modifier = '\0'; |
96f54887 KW |
8965 | regex_charset cs = get_regex_charset(RExC_flags); |
8966 | if (cs == REGEX_DEPENDS_CHARSET | |
8967 | && (RExC_utf8 || RExC_uni_semantics)) | |
8968 | { | |
8969 | cs = REGEX_UNICODE_CHARSET; | |
8970 | } | |
cde0cee5 YO |
8971 | |
8972 | while (*RExC_parse) { | |
8973 | /* && strchr("iogcmsx", *RExC_parse) */ | |
9d1d55b5 JP |
8974 | /* (?g), (?gc) and (?o) are useless here |
8975 | and must be globally applied -- japhy */ | |
cde0cee5 YO |
8976 | switch (*RExC_parse) { |
8977 | CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp); | |
9de15fec | 8978 | case LOCALE_PAT_MOD: |
f6a766d5 KW |
8979 | if (has_charset_modifier) { |
8980 | goto excess_modifier; | |
8981 | } | |
8982 | else if (flagsp == &negflags) { | |
9442e3b8 | 8983 | goto neg_modifier; |
9de15fec | 8984 | } |
a62b1201 | 8985 | cs = REGEX_LOCALE_CHARSET; |
f6a766d5 | 8986 | has_charset_modifier = LOCALE_PAT_MOD; |
4624b182 | 8987 | RExC_contains_locale = 1; |
9de15fec KW |
8988 | break; |
8989 | case UNICODE_PAT_MOD: | |
f6a766d5 KW |
8990 | if (has_charset_modifier) { |
8991 | goto excess_modifier; | |
8992 | } | |
8993 | else if (flagsp == &negflags) { | |
9442e3b8 | 8994 | goto neg_modifier; |
9de15fec | 8995 | } |
a62b1201 | 8996 | cs = REGEX_UNICODE_CHARSET; |
f6a766d5 | 8997 | has_charset_modifier = UNICODE_PAT_MOD; |
9de15fec | 8998 | break; |
cfaf538b | 8999 | case ASCII_RESTRICT_PAT_MOD: |
f6a766d5 | 9000 | if (flagsp == &negflags) { |
9442e3b8 | 9001 | goto neg_modifier; |
cfaf538b | 9002 | } |
f6a766d5 KW |
9003 | if (has_charset_modifier) { |
9004 | if (cs != REGEX_ASCII_RESTRICTED_CHARSET) { | |
9005 | goto excess_modifier; | |
9006 | } | |
2f7f8cb1 | 9007 | /* Doubled modifier implies more restricted */ |
f6a766d5 KW |
9008 | cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET; |
9009 | } | |
2f7f8cb1 KW |
9010 | else { |
9011 | cs = REGEX_ASCII_RESTRICTED_CHARSET; | |
9012 | } | |
f6a766d5 | 9013 | has_charset_modifier = ASCII_RESTRICT_PAT_MOD; |
cfaf538b | 9014 | break; |
50e91148 | 9015 | case DEPENDS_PAT_MOD: |
9442e3b8 | 9016 | if (has_use_defaults) { |
9de15fec | 9017 | goto fail_modifiers; |
f6a766d5 | 9018 | } |
9442e3b8 KW |
9019 | else if (flagsp == &negflags) { |
9020 | goto neg_modifier; | |
9021 | } | |
f6a766d5 KW |
9022 | else if (has_charset_modifier) { |
9023 | goto excess_modifier; | |
9de15fec | 9024 | } |
7b98bc43 KW |
9025 | |
9026 | /* The dual charset means unicode semantics if the | |
9027 | * pattern (or target, not known until runtime) are | |
e40e74fe KW |
9028 | * utf8, or something in the pattern indicates unicode |
9029 | * semantics */ | |
9030 | cs = (RExC_utf8 || RExC_uni_semantics) | |
a62b1201 KW |
9031 | ? REGEX_UNICODE_CHARSET |
9032 | : REGEX_DEPENDS_CHARSET; | |
f6a766d5 | 9033 | has_charset_modifier = DEPENDS_PAT_MOD; |
9de15fec | 9034 | break; |
f6a766d5 KW |
9035 | excess_modifier: |
9036 | RExC_parse++; | |
9037 | if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) { | |
0c96c706 | 9038 | vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD); |
f6a766d5 KW |
9039 | } |
9040 | else if (has_charset_modifier == *(RExC_parse - 1)) { | |
0c96c706 | 9041 | vFAIL2("Regexp modifier \"%c\" may not appear twice", *(RExC_parse - 1)); |
f6a766d5 KW |
9042 | } |
9043 | else { | |
0c96c706 | 9044 | vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1)); |
f6a766d5 KW |
9045 | } |
9046 | /*NOTREACHED*/ | |
9442e3b8 KW |
9047 | neg_modifier: |
9048 | RExC_parse++; | |
9049 | vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"", *(RExC_parse - 1)); | |
9050 | /*NOTREACHED*/ | |
f7819f85 A |
9051 | case ONCE_PAT_MOD: /* 'o' */ |
9052 | case GLOBAL_PAT_MOD: /* 'g' */ | |
9d1d55b5 | 9053 | if (SIZE_ONLY && ckWARN(WARN_REGEXP)) { |
6136c704 | 9054 | const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G; |
9d1d55b5 JP |
9055 | if (! (wastedflags & wflagbit) ) { |
9056 | wastedflags |= wflagbit; | |
9057 | vWARN5( | |
9058 | RExC_parse + 1, | |
9059 | "Useless (%s%c) - %suse /%c modifier", | |
9060 | flagsp == &negflags ? "?-" : "?", | |
9061 | *RExC_parse, | |
9062 | flagsp == &negflags ? "don't " : "", | |
9063 | *RExC_parse | |
9064 | ); | |
9065 | } | |
9066 | } | |
cde0cee5 YO |
9067 | break; |
9068 | ||
f7819f85 | 9069 | case CONTINUE_PAT_MOD: /* 'c' */ |
9d1d55b5 | 9070 | if (SIZE_ONLY && ckWARN(WARN_REGEXP)) { |
6136c704 AL |
9071 | if (! (wastedflags & WASTED_C) ) { |
9072 | wastedflags |= WASTED_GC; | |
9d1d55b5 JP |
9073 | vWARN3( |
9074 | RExC_parse + 1, | |
9075 | "Useless (%sc) - %suse /gc modifier", | |
9076 | flagsp == &negflags ? "?-" : "?", | |
9077 | flagsp == &negflags ? "don't " : "" | |
9078 | ); | |
9079 | } | |
9080 | } | |
cde0cee5 | 9081 | break; |
f7819f85 | 9082 | case KEEPCOPY_PAT_MOD: /* 'p' */ |
cde0cee5 | 9083 | if (flagsp == &negflags) { |
668c081a NC |
9084 | if (SIZE_ONLY) |
9085 | ckWARNreg(RExC_parse + 1,"Useless use of (?-p)"); | |
cde0cee5 YO |
9086 | } else { |
9087 | *flagsp |= RXf_PMf_KEEPCOPY; | |
9088 | } | |
9089 | break; | |
9090 | case '-': | |
3b753521 | 9091 | /* A flag is a default iff it is following a minus, so |
fb85c044 KW |
9092 | * if there is a minus, it means will be trying to |
9093 | * re-specify a default which is an error */ | |
9094 | if (has_use_defaults || flagsp == &negflags) { | |
9de15fec | 9095 | fail_modifiers: |
57b84237 YO |
9096 | RExC_parse++; |
9097 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
9098 | /*NOTREACHED*/ | |
9099 | } | |
cde0cee5 YO |
9100 | flagsp = &negflags; |
9101 | wastedflags = 0; /* reset so (?g-c) warns twice */ | |
9102 | break; | |
9103 | case ':': | |
9104 | paren = ':'; | |
9105 | /*FALLTHROUGH*/ | |
9106 | case ')': | |
9107 | RExC_flags |= posflags; | |
9108 | RExC_flags &= ~negflags; | |
a62b1201 | 9109 | set_regex_charset(&RExC_flags, cs); |
f7819f85 A |
9110 | if (paren != ':') { |
9111 | oregflags |= posflags; | |
9112 | oregflags &= ~negflags; | |
a62b1201 | 9113 | set_regex_charset(&oregflags, cs); |
f7819f85 | 9114 | } |
cde0cee5 YO |
9115 | nextchar(pRExC_state); |
9116 | if (paren != ':') { | |
9117 | *flagp = TRYAGAIN; | |
9118 | return NULL; | |
9119 | } else { | |
9120 | ret = NULL; | |
9121 | goto parse_rest; | |
9122 | } | |
9123 | /*NOTREACHED*/ | |
9124 | default: | |
cde0cee5 YO |
9125 | RExC_parse++; |
9126 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
9127 | /*NOTREACHED*/ | |
9128 | } | |
830247a4 | 9129 | ++RExC_parse; |
48c036b1 | 9130 | } |
cde0cee5 | 9131 | }} /* one for the default block, one for the switch */ |
a0d0e21e | 9132 | } |
fac92740 | 9133 | else { /* (...) */ |
81714fb9 | 9134 | capturing_parens: |
830247a4 IZ |
9135 | parno = RExC_npar; |
9136 | RExC_npar++; | |
e2e6a0f1 | 9137 | |
830247a4 | 9138 | ret = reganode(pRExC_state, OPEN, parno); |
e2e6a0f1 YO |
9139 | if (!SIZE_ONLY ){ |
9140 | if (!RExC_nestroot) | |
9141 | RExC_nestroot = parno; | |
c009da3d YO |
9142 | if (RExC_seen & REG_SEEN_RECURSE |
9143 | && !RExC_open_parens[parno-1]) | |
9144 | { | |
e2e6a0f1 | 9145 | DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log, |
40d049e4 YO |
9146 | "Setting open paren #%"IVdf" to %d\n", |
9147 | (IV)parno, REG_NODE_NUM(ret))); | |
e2e6a0f1 YO |
9148 | RExC_open_parens[parno-1]= ret; |
9149 | } | |
6bda09f9 | 9150 | } |
fac92740 MJD |
9151 | Set_Node_Length(ret, 1); /* MJD */ |
9152 | Set_Node_Offset(ret, RExC_parse); /* MJD */ | |
6136c704 | 9153 | is_open = 1; |
a0d0e21e | 9154 | } |
a0ed51b3 | 9155 | } |
fac92740 | 9156 | else /* ! paren */ |
a0d0e21e | 9157 | ret = NULL; |
cde0cee5 YO |
9158 | |
9159 | parse_rest: | |
a0d0e21e | 9160 | /* Pick up the branches, linking them together. */ |
fac92740 | 9161 | parse_start = RExC_parse; /* MJD */ |
3dab1dad | 9162 | br = regbranch(pRExC_state, &flags, 1,depth+1); |
ee91d26e | 9163 | |
fac92740 | 9164 | /* branch_len = (paren != 0); */ |
2af232bd | 9165 | |
a0d0e21e LW |
9166 | if (br == NULL) |
9167 | return(NULL); | |
830247a4 IZ |
9168 | if (*RExC_parse == '|') { |
9169 | if (!SIZE_ONLY && RExC_extralen) { | |
6bda09f9 | 9170 | reginsert(pRExC_state, BRANCHJ, br, depth+1); |
a0ed51b3 | 9171 | } |
fac92740 | 9172 | else { /* MJD */ |
6bda09f9 | 9173 | reginsert(pRExC_state, BRANCH, br, depth+1); |
fac92740 MJD |
9174 | Set_Node_Length(br, paren != 0); |
9175 | Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start); | |
9176 | } | |
c277df42 IZ |
9177 | have_branch = 1; |
9178 | if (SIZE_ONLY) | |
830247a4 | 9179 | RExC_extralen += 1; /* For BRANCHJ-BRANCH. */ |
a0ed51b3 LW |
9180 | } |
9181 | else if (paren == ':') { | |
c277df42 IZ |
9182 | *flagp |= flags&SIMPLE; |
9183 | } | |
6136c704 | 9184 | if (is_open) { /* Starts with OPEN. */ |
3dab1dad | 9185 | REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */ |
a0ed51b3 LW |
9186 | } |
9187 | else if (paren != '?') /* Not Conditional */ | |
a0d0e21e | 9188 | ret = br; |
8ae10a67 | 9189 | *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED); |
c277df42 | 9190 | lastbr = br; |
830247a4 IZ |
9191 | while (*RExC_parse == '|') { |
9192 | if (!SIZE_ONLY && RExC_extralen) { | |
9193 | ender = reganode(pRExC_state, LONGJMP,0); | |
3dab1dad | 9194 | REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */ |
c277df42 IZ |
9195 | } |
9196 | if (SIZE_ONLY) | |
830247a4 IZ |
9197 | RExC_extralen += 2; /* Account for LONGJMP. */ |
9198 | nextchar(pRExC_state); | |
594d7033 YO |
9199 | if (freeze_paren) { |
9200 | if (RExC_npar > after_freeze) | |
9201 | after_freeze = RExC_npar; | |
9202 | RExC_npar = freeze_paren; | |
9203 | } | |
3dab1dad | 9204 | br = regbranch(pRExC_state, &flags, 0, depth+1); |
2af232bd | 9205 | |
a687059c | 9206 | if (br == NULL) |
a0d0e21e | 9207 | return(NULL); |
3dab1dad | 9208 | REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */ |
c277df42 | 9209 | lastbr = br; |
8ae10a67 | 9210 | *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED); |
a0d0e21e LW |
9211 | } |
9212 | ||
c277df42 IZ |
9213 | if (have_branch || paren != ':') { |
9214 | /* Make a closing node, and hook it on the end. */ | |
9215 | switch (paren) { | |
9216 | case ':': | |
830247a4 | 9217 | ender = reg_node(pRExC_state, TAIL); |
c277df42 IZ |
9218 | break; |
9219 | case 1: | |
830247a4 | 9220 | ender = reganode(pRExC_state, CLOSE, parno); |
40d049e4 YO |
9221 | if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) { |
9222 | DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log, | |
9223 | "Setting close paren #%"IVdf" to %d\n", | |
9224 | (IV)parno, REG_NODE_NUM(ender))); | |
9225 | RExC_close_parens[parno-1]= ender; | |
e2e6a0f1 YO |
9226 | if (RExC_nestroot == parno) |
9227 | RExC_nestroot = 0; | |
40d049e4 | 9228 | } |
fac92740 MJD |
9229 | Set_Node_Offset(ender,RExC_parse+1); /* MJD */ |
9230 | Set_Node_Length(ender,1); /* MJD */ | |
c277df42 IZ |
9231 | break; |
9232 | case '<': | |
c277df42 IZ |
9233 | case ',': |
9234 | case '=': | |
9235 | case '!': | |
c277df42 | 9236 | *flagp &= ~HASWIDTH; |
821b33a5 IZ |
9237 | /* FALL THROUGH */ |
9238 | case '>': | |
830247a4 | 9239 | ender = reg_node(pRExC_state, SUCCEED); |
c277df42 IZ |
9240 | break; |
9241 | case 0: | |
830247a4 | 9242 | ender = reg_node(pRExC_state, END); |
40d049e4 YO |
9243 | if (!SIZE_ONLY) { |
9244 | assert(!RExC_opend); /* there can only be one! */ | |
9245 | RExC_opend = ender; | |
9246 | } | |
c277df42 IZ |
9247 | break; |
9248 | } | |
3b6759a6 YO |
9249 | DEBUG_PARSE_r(if (!SIZE_ONLY) { |
9250 | SV * const mysv_val1=sv_newmortal(); | |
9251 | SV * const mysv_val2=sv_newmortal(); | |
9252 | DEBUG_PARSE_MSG("lsbr"); | |
9253 | regprop(RExC_rx, mysv_val1, lastbr); | |
9254 | regprop(RExC_rx, mysv_val2, ender); | |
9255 | PerlIO_printf(Perl_debug_log, "~ tying lastbr %s (%"IVdf") to ender %s (%"IVdf") offset %"IVdf"\n", | |
9256 | SvPV_nolen_const(mysv_val1), | |
9257 | (IV)REG_NODE_NUM(lastbr), | |
9258 | SvPV_nolen_const(mysv_val2), | |
9259 | (IV)REG_NODE_NUM(ender), | |
9260 | (IV)(ender - lastbr) | |
9261 | ); | |
9262 | }); | |
eaf3ca90 | 9263 | REGTAIL(pRExC_state, lastbr, ender); |
a0d0e21e | 9264 | |
9674d46a | 9265 | if (have_branch && !SIZE_ONLY) { |
3b6759a6 | 9266 | char is_nothing= 1; |
eaf3ca90 YO |
9267 | if (depth==1) |
9268 | RExC_seen |= REG_TOP_LEVEL_BRANCHES; | |
9269 | ||
c277df42 | 9270 | /* Hook the tails of the branches to the closing node. */ |
9674d46a AL |
9271 | for (br = ret; br; br = regnext(br)) { |
9272 | const U8 op = PL_regkind[OP(br)]; | |
9273 | if (op == BRANCH) { | |
07be1b83 | 9274 | REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender); |
3b6759a6 YO |
9275 | if (OP(NEXTOPER(br)) != NOTHING || regnext(NEXTOPER(br)) != ender) |
9276 | is_nothing= 0; | |
9674d46a AL |
9277 | } |
9278 | else if (op == BRANCHJ) { | |
07be1b83 | 9279 | REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender); |
3b6759a6 YO |
9280 | /* for now we always disable this optimisation * / |
9281 | if (OP(NEXTOPER(NEXTOPER(br))) != NOTHING || regnext(NEXTOPER(NEXTOPER(br))) != ender) | |
9282 | */ | |
9283 | is_nothing= 0; | |
9674d46a | 9284 | } |
c277df42 | 9285 | } |
3b6759a6 YO |
9286 | if (is_nothing) { |
9287 | br= PL_regkind[OP(ret)] != BRANCH ? regnext(ret) : ret; | |
9288 | DEBUG_PARSE_r(if (!SIZE_ONLY) { | |
9289 | SV * const mysv_val1=sv_newmortal(); | |
9290 | SV * const mysv_val2=sv_newmortal(); | |
9291 | DEBUG_PARSE_MSG("NADA"); | |
9292 | regprop(RExC_rx, mysv_val1, ret); | |
9293 | regprop(RExC_rx, mysv_val2, ender); | |
9294 | PerlIO_printf(Perl_debug_log, "~ converting ret %s (%"IVdf") to ender %s (%"IVdf") offset %"IVdf"\n", | |
9295 | SvPV_nolen_const(mysv_val1), | |
9296 | (IV)REG_NODE_NUM(ret), | |
9297 | SvPV_nolen_const(mysv_val2), | |
9298 | (IV)REG_NODE_NUM(ender), | |
9299 | (IV)(ender - ret) | |
9300 | ); | |
9301 | }); | |
9302 | OP(br)= NOTHING; | |
9303 | if (OP(ender) == TAIL) { | |
9304 | NEXT_OFF(br)= 0; | |
9305 | RExC_emit= br + 1; | |
9306 | } else { | |
9307 | regnode *opt; | |
9308 | for ( opt= br + 1; opt < ender ; opt++ ) | |
9309 | OP(opt)= OPTIMIZED; | |
9310 | NEXT_OFF(br)= ender - br; | |
9311 | } | |
9312 | } | |
c277df42 | 9313 | } |
a0d0e21e | 9314 | } |
c277df42 IZ |
9315 | |
9316 | { | |
e1ec3a88 AL |
9317 | const char *p; |
9318 | static const char parens[] = "=!<,>"; | |
c277df42 IZ |
9319 | |
9320 | if (paren && (p = strchr(parens, paren))) { | |
eb160463 | 9321 | U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH; |
c277df42 IZ |
9322 | int flag = (p - parens) > 1; |
9323 | ||
9324 | if (paren == '>') | |
9325 | node = SUSPEND, flag = 0; | |
6bda09f9 | 9326 | reginsert(pRExC_state, node,ret, depth+1); |
45948336 EP |
9327 | Set_Node_Cur_Length(ret); |
9328 | Set_Node_Offset(ret, parse_start + 1); | |
c277df42 | 9329 | ret->flags = flag; |
07be1b83 | 9330 | REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL)); |
c277df42 | 9331 | } |
a0d0e21e LW |
9332 | } |
9333 | ||
9334 | /* Check for proper termination. */ | |
ce3e6498 | 9335 | if (paren) { |
e2509266 | 9336 | RExC_flags = oregflags; |
830247a4 IZ |
9337 | if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') { |
9338 | RExC_parse = oregcomp_parse; | |
380a0633 | 9339 | vFAIL("Unmatched ("); |
ce3e6498 | 9340 | } |
a0ed51b3 | 9341 | } |
830247a4 IZ |
9342 | else if (!paren && RExC_parse < RExC_end) { |
9343 | if (*RExC_parse == ')') { | |
9344 | RExC_parse++; | |
380a0633 | 9345 | vFAIL("Unmatched )"); |
a0ed51b3 LW |
9346 | } |
9347 | else | |
b45f050a | 9348 | FAIL("Junk on end of regexp"); /* "Can't happen". */ |
118e2215 | 9349 | assert(0); /* NOTREACHED */ |
a0d0e21e | 9350 | } |
b57e4118 KW |
9351 | |
9352 | if (RExC_in_lookbehind) { | |
9353 | RExC_in_lookbehind--; | |
9354 | } | |
fd4be6f0 | 9355 | if (after_freeze > RExC_npar) |
594d7033 | 9356 | RExC_npar = after_freeze; |
a0d0e21e | 9357 | return(ret); |
a687059c LW |
9358 | } |
9359 | ||
9360 | /* | |
9361 | - regbranch - one alternative of an | operator | |
9362 | * | |
9363 | * Implements the concatenation operator. | |
9364 | */ | |
76e3520e | 9365 | STATIC regnode * |
3dab1dad | 9366 | S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth) |
a687059c | 9367 | { |
97aff369 | 9368 | dVAR; |
eb578fdb KW |
9369 | regnode *ret; |
9370 | regnode *chain = NULL; | |
9371 | regnode *latest; | |
c277df42 | 9372 | I32 flags = 0, c = 0; |
3dab1dad | 9373 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
9374 | |
9375 | PERL_ARGS_ASSERT_REGBRANCH; | |
9376 | ||
3dab1dad | 9377 | DEBUG_PARSE("brnc"); |
02daf0ab | 9378 | |
b81d288d | 9379 | if (first) |
c277df42 IZ |
9380 | ret = NULL; |
9381 | else { | |
b81d288d | 9382 | if (!SIZE_ONLY && RExC_extralen) |
830247a4 | 9383 | ret = reganode(pRExC_state, BRANCHJ,0); |
fac92740 | 9384 | else { |
830247a4 | 9385 | ret = reg_node(pRExC_state, BRANCH); |
fac92740 MJD |
9386 | Set_Node_Length(ret, 1); |
9387 | } | |
c277df42 | 9388 | } |
686b73d4 | 9389 | |
b81d288d | 9390 | if (!first && SIZE_ONLY) |
830247a4 | 9391 | RExC_extralen += 1; /* BRANCHJ */ |
b81d288d | 9392 | |
c277df42 | 9393 | *flagp = WORST; /* Tentatively. */ |
a0d0e21e | 9394 | |
830247a4 IZ |
9395 | RExC_parse--; |
9396 | nextchar(pRExC_state); | |
9397 | while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') { | |
a0d0e21e | 9398 | flags &= ~TRYAGAIN; |
3dab1dad | 9399 | latest = regpiece(pRExC_state, &flags,depth+1); |
a0d0e21e LW |
9400 | if (latest == NULL) { |
9401 | if (flags & TRYAGAIN) | |
9402 | continue; | |
9403 | return(NULL); | |
a0ed51b3 LW |
9404 | } |
9405 | else if (ret == NULL) | |
c277df42 | 9406 | ret = latest; |
8ae10a67 | 9407 | *flagp |= flags&(HASWIDTH|POSTPONED); |
c277df42 | 9408 | if (chain == NULL) /* First piece. */ |
a0d0e21e LW |
9409 | *flagp |= flags&SPSTART; |
9410 | else { | |
830247a4 | 9411 | RExC_naughty++; |
3dab1dad | 9412 | REGTAIL(pRExC_state, chain, latest); |
a687059c | 9413 | } |
a0d0e21e | 9414 | chain = latest; |
c277df42 IZ |
9415 | c++; |
9416 | } | |
9417 | if (chain == NULL) { /* Loop ran zero times. */ | |
830247a4 | 9418 | chain = reg_node(pRExC_state, NOTHING); |
c277df42 IZ |
9419 | if (ret == NULL) |
9420 | ret = chain; | |
9421 | } | |
9422 | if (c == 1) { | |
9423 | *flagp |= flags&SIMPLE; | |
a0d0e21e | 9424 | } |
a687059c | 9425 | |
d4c19fe8 | 9426 | return ret; |
a687059c LW |
9427 | } |
9428 | ||
9429 | /* | |
9430 | - regpiece - something followed by possible [*+?] | |
9431 | * | |
9432 | * Note that the branching code sequences used for ? and the general cases | |
9433 | * of * and + are somewhat optimized: they use the same NOTHING node as | |
9434 | * both the endmarker for their branch list and the body of the last branch. | |
9435 | * It might seem that this node could be dispensed with entirely, but the | |
9436 | * endmarker role is not redundant. | |
9437 | */ | |
76e3520e | 9438 | STATIC regnode * |
3dab1dad | 9439 | S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth) |
a687059c | 9440 | { |
97aff369 | 9441 | dVAR; |
eb578fdb KW |
9442 | regnode *ret; |
9443 | char op; | |
9444 | char *next; | |
a0d0e21e | 9445 | I32 flags; |
1df70142 | 9446 | const char * const origparse = RExC_parse; |
a0d0e21e | 9447 | I32 min; |
c277df42 | 9448 | I32 max = REG_INFTY; |
f19a8d85 | 9449 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 9450 | char *parse_start; |
f19a8d85 | 9451 | #endif |
10edeb5d | 9452 | const char *maxpos = NULL; |
31c15ce5 KW |
9453 | |
9454 | /* Save the original in case we change the emitted regop to a FAIL. */ | |
9455 | regnode * const orig_emit = RExC_emit; | |
9456 | ||
3dab1dad | 9457 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
9458 | |
9459 | PERL_ARGS_ASSERT_REGPIECE; | |
9460 | ||
3dab1dad | 9461 | DEBUG_PARSE("piec"); |
a0d0e21e | 9462 | |
3dab1dad | 9463 | ret = regatom(pRExC_state, &flags,depth+1); |
a0d0e21e LW |
9464 | if (ret == NULL) { |
9465 | if (flags & TRYAGAIN) | |
9466 | *flagp |= TRYAGAIN; | |
9467 | return(NULL); | |
9468 | } | |
9469 | ||
830247a4 | 9470 | op = *RExC_parse; |
a0d0e21e | 9471 | |
830247a4 | 9472 | if (op == '{' && regcurly(RExC_parse)) { |
10edeb5d | 9473 | maxpos = NULL; |
f19a8d85 | 9474 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 9475 | parse_start = RExC_parse; /* MJD */ |
f19a8d85 | 9476 | #endif |
830247a4 | 9477 | next = RExC_parse + 1; |
a0d0e21e LW |
9478 | while (isDIGIT(*next) || *next == ',') { |
9479 | if (*next == ',') { | |
9480 | if (maxpos) | |
9481 | break; | |
9482 | else | |
9483 | maxpos = next; | |
a687059c | 9484 | } |
a0d0e21e LW |
9485 | next++; |
9486 | } | |
9487 | if (*next == '}') { /* got one */ | |
9488 | if (!maxpos) | |
9489 | maxpos = next; | |
830247a4 IZ |
9490 | RExC_parse++; |
9491 | min = atoi(RExC_parse); | |
a0d0e21e LW |
9492 | if (*maxpos == ',') |
9493 | maxpos++; | |
9494 | else | |
830247a4 | 9495 | maxpos = RExC_parse; |
a0d0e21e LW |
9496 | max = atoi(maxpos); |
9497 | if (!max && *maxpos != '0') | |
c277df42 IZ |
9498 | max = REG_INFTY; /* meaning "infinity" */ |
9499 | else if (max >= REG_INFTY) | |
8615cb43 | 9500 | vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1); |
830247a4 IZ |
9501 | RExC_parse = next; |
9502 | nextchar(pRExC_state); | |
31c15ce5 KW |
9503 | if (max < min) { /* If can't match, warn and optimize to fail |
9504 | unconditionally */ | |
9505 | if (SIZE_ONLY) { | |
9506 | ckWARNreg(RExC_parse, "Quantifier {n,m} with n > m can't match"); | |
9507 | ||
9508 | /* We can't back off the size because we have to reserve | |
9509 | * enough space for all the things we are about to throw | |
9510 | * away, but we can shrink it by the ammount we are about | |
9511 | * to re-use here */ | |
9512 | RExC_size = PREVOPER(RExC_size) - regarglen[(U8)OPFAIL]; | |
9513 | } | |
9514 | else { | |
9515 | RExC_emit = orig_emit; | |
9516 | } | |
9517 | ret = reg_node(pRExC_state, OPFAIL); | |
9518 | return ret; | |
9519 | } | |
a0d0e21e LW |
9520 | |
9521 | do_curly: | |
9522 | if ((flags&SIMPLE)) { | |
830247a4 | 9523 | RExC_naughty += 2 + RExC_naughty / 2; |
6bda09f9 | 9524 | reginsert(pRExC_state, CURLY, ret, depth+1); |
fac92740 MJD |
9525 | Set_Node_Offset(ret, parse_start+1); /* MJD */ |
9526 | Set_Node_Cur_Length(ret); | |
a0d0e21e LW |
9527 | } |
9528 | else { | |
3dab1dad | 9529 | regnode * const w = reg_node(pRExC_state, WHILEM); |
2c2d71f5 JH |
9530 | |
9531 | w->flags = 0; | |
3dab1dad | 9532 | REGTAIL(pRExC_state, ret, w); |
830247a4 | 9533 | if (!SIZE_ONLY && RExC_extralen) { |
6bda09f9 YO |
9534 | reginsert(pRExC_state, LONGJMP,ret, depth+1); |
9535 | reginsert(pRExC_state, NOTHING,ret, depth+1); | |
c277df42 IZ |
9536 | NEXT_OFF(ret) = 3; /* Go over LONGJMP. */ |
9537 | } | |
6bda09f9 | 9538 | reginsert(pRExC_state, CURLYX,ret, depth+1); |
fac92740 MJD |
9539 | /* MJD hk */ |
9540 | Set_Node_Offset(ret, parse_start+1); | |
2af232bd | 9541 | Set_Node_Length(ret, |
fac92740 | 9542 | op == '{' ? (RExC_parse - parse_start) : 1); |
2af232bd | 9543 | |
830247a4 | 9544 | if (!SIZE_ONLY && RExC_extralen) |
c277df42 | 9545 | NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */ |
3dab1dad | 9546 | REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING)); |
c277df42 | 9547 | if (SIZE_ONLY) |
830247a4 IZ |
9548 | RExC_whilem_seen++, RExC_extralen += 3; |
9549 | RExC_naughty += 4 + RExC_naughty; /* compound interest */ | |
a0d0e21e | 9550 | } |
c277df42 | 9551 | ret->flags = 0; |
a0d0e21e LW |
9552 | |
9553 | if (min > 0) | |
821b33a5 IZ |
9554 | *flagp = WORST; |
9555 | if (max > 0) | |
9556 | *flagp |= HASWIDTH; | |
c277df42 | 9557 | if (!SIZE_ONLY) { |
eb160463 GS |
9558 | ARG1_SET(ret, (U16)min); |
9559 | ARG2_SET(ret, (U16)max); | |
a687059c | 9560 | } |
a687059c | 9561 | |
a0d0e21e | 9562 | goto nest_check; |
a687059c | 9563 | } |
a0d0e21e | 9564 | } |
a687059c | 9565 | |
a0d0e21e LW |
9566 | if (!ISMULT1(op)) { |
9567 | *flagp = flags; | |
a687059c | 9568 | return(ret); |
a0d0e21e | 9569 | } |
bb20fd44 | 9570 | |
c277df42 | 9571 | #if 0 /* Now runtime fix should be reliable. */ |
b45f050a JF |
9572 | |
9573 | /* if this is reinstated, don't forget to put this back into perldiag: | |
9574 | ||
9575 | =item Regexp *+ operand could be empty at {#} in regex m/%s/ | |
9576 | ||
9577 | (F) The part of the regexp subject to either the * or + quantifier | |
9578 | could match an empty string. The {#} shows in the regular | |
9579 | expression about where the problem was discovered. | |
9580 | ||
9581 | */ | |
9582 | ||
bb20fd44 | 9583 | if (!(flags&HASWIDTH) && op != '?') |
b45f050a | 9584 | vFAIL("Regexp *+ operand could be empty"); |
b81d288d | 9585 | #endif |
bb20fd44 | 9586 | |
f19a8d85 | 9587 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 9588 | parse_start = RExC_parse; |
f19a8d85 | 9589 | #endif |
830247a4 | 9590 | nextchar(pRExC_state); |
a0d0e21e | 9591 | |
821b33a5 | 9592 | *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH); |
a0d0e21e LW |
9593 | |
9594 | if (op == '*' && (flags&SIMPLE)) { | |
6bda09f9 | 9595 | reginsert(pRExC_state, STAR, ret, depth+1); |
c277df42 | 9596 | ret->flags = 0; |
830247a4 | 9597 | RExC_naughty += 4; |
a0d0e21e LW |
9598 | } |
9599 | else if (op == '*') { | |
9600 | min = 0; | |
9601 | goto do_curly; | |
a0ed51b3 LW |
9602 | } |
9603 | else if (op == '+' && (flags&SIMPLE)) { | |
6bda09f9 | 9604 | reginsert(pRExC_state, PLUS, ret, depth+1); |
c277df42 | 9605 | ret->flags = 0; |
830247a4 | 9606 | RExC_naughty += 3; |
a0d0e21e LW |
9607 | } |
9608 | else if (op == '+') { | |
9609 | min = 1; | |
9610 | goto do_curly; | |
a0ed51b3 LW |
9611 | } |
9612 | else if (op == '?') { | |
a0d0e21e LW |
9613 | min = 0; max = 1; |
9614 | goto do_curly; | |
9615 | } | |
9616 | nest_check: | |
668c081a NC |
9617 | if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) { |
9618 | ckWARN3reg(RExC_parse, | |
9619 | "%.*s matches null string many times", | |
9620 | (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0), | |
9621 | origparse); | |
a0d0e21e LW |
9622 | } |
9623 | ||
b9b4dddf | 9624 | if (RExC_parse < RExC_end && *RExC_parse == '?') { |
830247a4 | 9625 | nextchar(pRExC_state); |
6bda09f9 | 9626 | reginsert(pRExC_state, MINMOD, ret, depth+1); |
3dab1dad | 9627 | REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE); |
a0d0e21e | 9628 | } |
b9b4dddf YO |
9629 | #ifndef REG_ALLOW_MINMOD_SUSPEND |
9630 | else | |
9631 | #endif | |
9632 | if (RExC_parse < RExC_end && *RExC_parse == '+') { | |
9633 | regnode *ender; | |
9634 | nextchar(pRExC_state); | |
9635 | ender = reg_node(pRExC_state, SUCCEED); | |
9636 | REGTAIL(pRExC_state, ret, ender); | |
9637 | reginsert(pRExC_state, SUSPEND, ret, depth+1); | |
9638 | ret->flags = 0; | |
9639 | ender = reg_node(pRExC_state, TAIL); | |
9640 | REGTAIL(pRExC_state, ret, ender); | |
9641 | /*ret= ender;*/ | |
9642 | } | |
9643 | ||
9644 | if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) { | |
830247a4 | 9645 | RExC_parse++; |
b45f050a JF |
9646 | vFAIL("Nested quantifiers"); |
9647 | } | |
a0d0e21e LW |
9648 | |
9649 | return(ret); | |
a687059c LW |
9650 | } |
9651 | ||
6012a526 KW |
9652 | STATIC bool |
9653 | S_grok_bslash_N(pTHX_ RExC_state_t *pRExC_state, regnode** node_p, UV *valuep, I32 *flagp, U32 depth, bool in_char_class) | |
9654 | { | |
fc8cd66c | 9655 | |
6012a526 | 9656 | /* This is expected to be called by a parser routine that has recognized '\N' |
b9e8c997 KW |
9657 | and needs to handle the rest. RExC_parse is expected to point at the first |
9658 | char following the N at the time of the call. On successful return, | |
9659 | RExC_parse has been updated to point to just after the sequence identified | |
6012a526 | 9660 | by this routine, and <*flagp> has been updated. |
ff3f963a | 9661 | |
6012a526 | 9662 | The \N may be inside (indicated by the boolean <in_char_class>) or outside a |
ff3f963a KW |
9663 | character class. |
9664 | ||
9665 | \N may begin either a named sequence, or if outside a character class, mean | |
9666 | to match a non-newline. For non single-quoted regexes, the tokenizer has | |
6012a526 | 9667 | attempted to decide which, and in the case of a named sequence, converted it |
ff3f963a KW |
9668 | into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...}, |
9669 | where c1... are the characters in the sequence. For single-quoted regexes, | |
9670 | the tokenizer passes the \N sequence through unchanged; this code will not | |
b9e8c997 KW |
9671 | attempt to determine this nor expand those, instead raising a syntax error. |
9672 | The net effect is that if the beginning of the passed-in pattern isn't '{U+' | |
9673 | or there is no '}', it signals that this \N occurrence means to match a | |
9674 | non-newline. | |
9675 | ||
ff3f963a KW |
9676 | Only the \N{U+...} form should occur in a character class, for the same |
9677 | reason that '.' inside a character class means to just match a period: it | |
9678 | just doesn't make sense. | |
afefe6bf | 9679 | |
b9e8c997 KW |
9680 | The function raises an error (via vFAIL), and doesn't return for various |
9681 | syntax errors. Otherwise it returns TRUE and sets <node_p> or <valuep> on | |
9682 | success; it returns FALSE otherwise. | |
9683 | ||
9684 | If <valuep> is non-null, it means the caller can accept an input sequence | |
9685 | consisting of a just a single code point; <*valuep> is set to that value | |
9686 | if the input is such. | |
9687 | ||
b9c48b5b KW |
9688 | If <node_p> is non-null it signifies that the caller can accept any other |
9689 | legal sequence (i.e., one that isn't just a single code point). <*node_p> | |
9690 | is set as follows: | |
b9e8c997 KW |
9691 | 1) \N means not-a-NL: points to a newly created REG_ANY node; |
9692 | 2) \N{}: points to a new NOTHING node; | |
9693 | 3) otherwise: points to a new EXACT node containing the resolved | |
9694 | string. | |
b9c48b5b KW |
9695 | Note that FALSE is returned for single code point sequences if <valuep> is |
9696 | null. | |
fc8cd66c | 9697 | */ |
b9e8c997 | 9698 | |
c3c41406 | 9699 | char * endbrace; /* '}' following the name */ |
c3c41406 | 9700 | char* p; |
b9e8c997 KW |
9701 | char *endchar; /* Points to '.' or '}' ending cur char in the input |
9702 | stream */ | |
9703 | bool has_multiple_chars; /* true if the input stream contains a sequence of | |
9704 | more than one character */ | |
ff3f963a KW |
9705 | |
9706 | GET_RE_DEBUG_FLAGS_DECL; | |
7918f24d | 9707 | |
59f4a9a3 | 9708 | PERL_ARGS_ASSERT_GROK_BSLASH_N; |
ff3f963a KW |
9709 | |
9710 | GET_RE_DEBUG_FLAGS; | |
c3c41406 | 9711 | |
b9c48b5b | 9712 | assert(cBOOL(node_p) ^ cBOOL(valuep)); /* Exactly one should be set */ |
b9e8c997 | 9713 | |
c3c41406 KW |
9714 | /* The [^\n] meaning of \N ignores spaces and comments under the /x |
9715 | * modifier. The other meaning does not */ | |
9716 | p = (RExC_flags & RXf_PMf_EXTENDED) | |
9717 | ? regwhite( pRExC_state, RExC_parse ) | |
9718 | : RExC_parse; | |
b9e8c997 | 9719 | |
ff3f963a | 9720 | /* Disambiguate between \N meaning a named character versus \N meaning |
c3c41406 KW |
9721 | * [^\n]. The former is assumed when it can't be the latter. */ |
9722 | if (*p != '{' || regcurly(p)) { | |
9723 | RExC_parse = p; | |
b9e8c997 | 9724 | if (! node_p) { |
afefe6bf | 9725 | /* no bare \N in a charclass */ |
b9e8c997 KW |
9726 | if (in_char_class) { |
9727 | vFAIL("\\N in a character class must be a named character: \\N{...}"); | |
9728 | } | |
9729 | return FALSE; | |
9730 | } | |
afefe6bf | 9731 | nextchar(pRExC_state); |
b9e8c997 | 9732 | *node_p = reg_node(pRExC_state, REG_ANY); |
afefe6bf RGS |
9733 | *flagp |= HASWIDTH|SIMPLE; |
9734 | RExC_naughty++; | |
9735 | RExC_parse--; | |
b9e8c997 KW |
9736 | Set_Node_Length(*node_p, 1); /* MJD */ |
9737 | return TRUE; | |
fc8cd66c | 9738 | } |
a4893424 | 9739 | |
4fa352c8 | 9740 | /* Here, we have decided it should be a named character or sequence */ |
c3c41406 KW |
9741 | |
9742 | /* The test above made sure that the next real character is a '{', but | |
9743 | * under the /x modifier, it could be separated by space (or a comment and | |
9744 | * \n) and this is not allowed (for consistency with \x{...} and the | |
9745 | * tokenizer handling of \N{NAME}). */ | |
9746 | if (*RExC_parse != '{') { | |
9747 | vFAIL("Missing braces on \\N{}"); | |
9748 | } | |
9749 | ||
ff3f963a | 9750 | RExC_parse++; /* Skip past the '{' */ |
c3c41406 KW |
9751 | |
9752 | if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */ | |
9753 | || ! (endbrace == RExC_parse /* nothing between the {} */ | |
9754 | || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */ | |
9755 | && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */ | |
9756 | { | |
9757 | if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */ | |
9758 | vFAIL("\\N{NAME} must be resolved by the lexer"); | |
9759 | } | |
9760 | ||
ff3f963a | 9761 | if (endbrace == RExC_parse) { /* empty: \N{} */ |
b9e8c997 KW |
9762 | bool ret = TRUE; |
9763 | if (node_p) { | |
9764 | *node_p = reg_node(pRExC_state,NOTHING); | |
a4893424 | 9765 | } |
b9e8c997 KW |
9766 | else if (in_char_class) { |
9767 | if (SIZE_ONLY && in_char_class) { | |
9768 | ckWARNreg(RExC_parse, | |
9769 | "Ignoring zero length \\N{} in character class" | |
9770 | ); | |
9771 | } | |
9772 | ret = FALSE; | |
ff3f963a | 9773 | } |
b9e8c997 KW |
9774 | else { |
9775 | return FALSE; | |
9776 | } | |
9777 | nextchar(pRExC_state); | |
9778 | return ret; | |
fc8cd66c | 9779 | } |
ff3f963a | 9780 | |
86e88272 | 9781 | RExC_uni_semantics = 1; /* Unicode named chars imply Unicode semantics */ |
ff3f963a KW |
9782 | RExC_parse += 2; /* Skip past the 'U+' */ |
9783 | ||
b9e8c997 KW |
9784 | endchar = RExC_parse + strcspn(RExC_parse, ".}"); |
9785 | ||
9786 | /* Code points are separated by dots. If none, there is only one code | |
9787 | * point, and is terminated by the brace */ | |
9788 | has_multiple_chars = (endchar < endbrace); | |
9789 | ||
b9c48b5b | 9790 | if (valuep && (! has_multiple_chars || in_char_class)) { |
b9e8c997 KW |
9791 | /* We only pay attention to the first char of |
9792 | multichar strings being returned in char classes. I kinda wonder | |
ff3f963a KW |
9793 | if this makes sense as it does change the behaviour |
9794 | from earlier versions, OTOH that behaviour was broken | |
9795 | as well. XXX Solution is to recharacterize as | |
9796 | [rest-of-class]|multi1|multi2... */ | |
9797 | ||
b9e8c997 KW |
9798 | STRLEN length_of_hex = (STRLEN)(endchar - RExC_parse); |
9799 | I32 grok_hex_flags = PERL_SCAN_ALLOW_UNDERSCORES | |
ff3f963a KW |
9800 | | PERL_SCAN_DISALLOW_PREFIX |
9801 | | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0); | |
ff3f963a | 9802 | |
b9e8c997 | 9803 | *valuep = grok_hex(RExC_parse, &length_of_hex, &grok_hex_flags, NULL); |
ff3f963a KW |
9804 | |
9805 | /* The tokenizer should have guaranteed validity, but it's possible to | |
9806 | * bypass it by using single quoting, so check */ | |
c3c41406 KW |
9807 | if (length_of_hex == 0 |
9808 | || length_of_hex != (STRLEN)(endchar - RExC_parse) ) | |
9809 | { | |
9810 | RExC_parse += length_of_hex; /* Includes all the valid */ | |
9811 | RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */ | |
9812 | ? UTF8SKIP(RExC_parse) | |
9813 | : 1; | |
9814 | /* Guard against malformed utf8 */ | |
b9e8c997 KW |
9815 | if (RExC_parse >= endchar) { |
9816 | RExC_parse = endchar; | |
9817 | } | |
c3c41406 | 9818 | vFAIL("Invalid hexadecimal number in \\N{U+...}"); |
b9e8c997 | 9819 | } |
ff3f963a | 9820 | |
b9e8c997 KW |
9821 | if (in_char_class && has_multiple_chars) { |
9822 | ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class"); | |
9823 | } | |
9824 | RExC_parse = endbrace + 1; | |
9825 | } | |
b9c48b5b | 9826 | else if (! node_p || ! has_multiple_chars) { |
ff3f963a | 9827 | |
b9e8c997 KW |
9828 | /* Here, the input is legal, but not according to the caller's |
9829 | * options. We fail without advancing the parse, so that the | |
9830 | * caller can try again */ | |
9831 | RExC_parse = p; | |
9832 | return FALSE; | |
fc8cd66c | 9833 | } |
b9e8c997 | 9834 | else { |
e2a7e165 KW |
9835 | |
9836 | /* What is done here is to convert this to a sub-pattern of the form | |
9837 | * (?:\x{char1}\x{char2}...) | |
9838 | * and then call reg recursively. That way, it retains its atomicness, | |
9839 | * while not having to worry about special handling that some code | |
9840 | * points may have. toke.c has converted the original Unicode values | |
9841 | * to native, so that we can just pass on the hex values unchanged. We | |
9842 | * do have to set a flag to keep recoding from happening in the | |
9843 | * recursion */ | |
9844 | ||
9845 | SV * substitute_parse = newSVpvn_flags("?:", 2, SVf_UTF8|SVs_TEMP); | |
9846 | STRLEN len; | |
e2a7e165 | 9847 | char *orig_end = RExC_end; |
70ce1368 | 9848 | I32 flags; |
e2a7e165 KW |
9849 | |
9850 | while (RExC_parse < endbrace) { | |
ff3f963a | 9851 | |
e2a7e165 KW |
9852 | /* Convert to notation the rest of the code understands */ |
9853 | sv_catpv(substitute_parse, "\\x{"); | |
9854 | sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse); | |
9855 | sv_catpv(substitute_parse, "}"); | |
ff3f963a KW |
9856 | |
9857 | /* Point to the beginning of the next character in the sequence. */ | |
9858 | RExC_parse = endchar + 1; | |
b9e8c997 | 9859 | endchar = RExC_parse + strcspn(RExC_parse, ".}"); |
ff3f963a | 9860 | } |
e2a7e165 | 9861 | sv_catpv(substitute_parse, ")"); |
ff3f963a | 9862 | |
e2a7e165 | 9863 | RExC_parse = SvPV(substitute_parse, len); |
ff3f963a | 9864 | |
e2a7e165 KW |
9865 | /* Don't allow empty number */ |
9866 | if (len < 8) { | |
9867 | vFAIL("Invalid hexadecimal number in \\N{U+...}"); | |
ff3f963a | 9868 | } |
e2a7e165 | 9869 | RExC_end = RExC_parse + len; |
ff3f963a | 9870 | |
e2a7e165 KW |
9871 | /* The values are Unicode, and therefore not subject to recoding */ |
9872 | RExC_override_recoding = 1; | |
9873 | ||
70ce1368 KW |
9874 | *node_p = reg(pRExC_state, 1, &flags, depth+1); |
9875 | *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED); | |
e2a7e165 KW |
9876 | |
9877 | RExC_parse = endbrace; | |
9878 | RExC_end = orig_end; | |
9879 | RExC_override_recoding = 0; | |
ff3f963a | 9880 | |
b9e8c997 | 9881 | nextchar(pRExC_state); |
ff3f963a KW |
9882 | } |
9883 | ||
b9e8c997 | 9884 | return TRUE; |
fc8cd66c YO |
9885 | } |
9886 | ||
9887 | ||
9e08bc66 TS |
9888 | /* |
9889 | * reg_recode | |
9890 | * | |
9891 | * It returns the code point in utf8 for the value in *encp. | |
9892 | * value: a code value in the source encoding | |
9893 | * encp: a pointer to an Encode object | |
9894 | * | |
9895 | * If the result from Encode is not a single character, | |
9896 | * it returns U+FFFD (Replacement character) and sets *encp to NULL. | |
9897 | */ | |
9898 | STATIC UV | |
9899 | S_reg_recode(pTHX_ const char value, SV **encp) | |
9900 | { | |
9901 | STRLEN numlen = 1; | |
59cd0e26 | 9902 | SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP); |
c86f7df5 | 9903 | const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv); |
9e08bc66 TS |
9904 | const STRLEN newlen = SvCUR(sv); |
9905 | UV uv = UNICODE_REPLACEMENT; | |
9906 | ||
7918f24d NC |
9907 | PERL_ARGS_ASSERT_REG_RECODE; |
9908 | ||
9e08bc66 TS |
9909 | if (newlen) |
9910 | uv = SvUTF8(sv) | |
9911 | ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT) | |
9912 | : *(U8*)s; | |
9913 | ||
9914 | if (!newlen || numlen != newlen) { | |
9915 | uv = UNICODE_REPLACEMENT; | |
c86f7df5 | 9916 | *encp = NULL; |
9e08bc66 TS |
9917 | } |
9918 | return uv; | |
9919 | } | |
9920 | ||
86dc2bf5 KW |
9921 | PERL_STATIC_INLINE U8 |
9922 | S_compute_EXACTish(pTHX_ RExC_state_t *pRExC_state) | |
9923 | { | |
9924 | U8 op; | |
9925 | ||
9926 | PERL_ARGS_ASSERT_COMPUTE_EXACTISH; | |
9927 | ||
9928 | if (! FOLD) { | |
9929 | return EXACT; | |
9930 | } | |
9931 | ||
9932 | op = get_regex_charset(RExC_flags); | |
9933 | if (op >= REGEX_ASCII_RESTRICTED_CHARSET) { | |
9934 | op--; /* /a is same as /u, and map /aa's offset to what /a's would have | |
9935 | been, so there is no hole */ | |
9936 | } | |
9937 | ||
9938 | return op + EXACTF; | |
9939 | } | |
9940 | ||
5f820f89 | 9941 | PERL_STATIC_INLINE void |
34b39fc9 | 9942 | S_alloc_maybe_populate_EXACT(pTHX_ RExC_state_t *pRExC_state, regnode *node, I32* flagp, STRLEN len, UV code_point) |
5f820f89 | 9943 | { |
34b39fc9 KW |
9944 | /* This knows the details about sizing an EXACTish node, setting flags for |
9945 | * it (by setting <*flagp>, and potentially populating it with a single | |
9946 | * character. | |
9947 | * | |
9948 | * If <len> is non-zero, this function assumes that the node has already | |
9949 | * been populated, and just does the sizing. In this case <code_point> | |
9950 | * should be the final code point that has already been placed into the | |
9951 | * node. This value will be ignored except that under some circumstances | |
9952 | * <*flagp> is set based on it. | |
9953 | * | |
9954 | * If <len is zero, the function assumes that the node is to contain only | |
9955 | * the single character given by <code_point> and calculates what <len> | |
9956 | * should be. In pass 1, it sizes the node appropriately. In pass 2, it | |
9957 | * additionally will populate the node's STRING with <code_point>, if <len> | |
9958 | * is 0. In both cases <*flagp> is appropriately set | |
5f820f89 KW |
9959 | * |
9960 | * It knows that under FOLD, UTF characters and the Latin Sharp S must be | |
9961 | * folded (the latter only when the rules indicate it can match 'ss') */ | |
9962 | ||
9963 | bool len_passed_in = cBOOL(len != 0); | |
9964 | U8 character[UTF8_MAXBYTES_CASE+1]; | |
9965 | ||
9966 | PERL_ARGS_ASSERT_ALLOC_MAYBE_POPULATE_EXACT; | |
9967 | ||
9968 | if (! len_passed_in) { | |
9969 | if (UTF) { | |
9970 | if (FOLD) { | |
9971 | to_uni_fold(NATIVE_TO_UNI(code_point), character, &len); | |
9972 | } | |
9973 | else { | |
9974 | uvchr_to_utf8( character, code_point); | |
9975 | len = UTF8SKIP(character); | |
9976 | } | |
9977 | } | |
9978 | else if (! FOLD | |
9979 | || code_point != LATIN_SMALL_LETTER_SHARP_S | |
a725e29c | 9980 | || ASCII_FOLD_RESTRICTED |
5f820f89 KW |
9981 | || ! AT_LEAST_UNI_SEMANTICS) |
9982 | { | |
9983 | *character = (U8) code_point; | |
9984 | len = 1; | |
9985 | } | |
9986 | else { | |
9987 | *character = 's'; | |
9988 | *(character + 1) = 's'; | |
9989 | len = 2; | |
9990 | } | |
9991 | } | |
9992 | ||
9993 | if (SIZE_ONLY) { | |
9994 | RExC_size += STR_SZ(len); | |
9995 | } | |
9996 | else { | |
9997 | RExC_emit += STR_SZ(len); | |
9998 | STR_LEN(node) = len; | |
9999 | if (! len_passed_in) { | |
10000 | Copy((char *) character, STRING(node), len, char); | |
10001 | } | |
10002 | } | |
34b39fc9 KW |
10003 | |
10004 | *flagp |= HASWIDTH; | |
10005 | if (len == 1 && UNI_IS_INVARIANT(code_point)) | |
10006 | *flagp |= SIMPLE; | |
5f820f89 | 10007 | } |
fc8cd66c | 10008 | |
a687059c LW |
10009 | /* |
10010 | - regatom - the lowest level | |
ee9b8eae YO |
10011 | |
10012 | Try to identify anything special at the start of the pattern. If there | |
10013 | is, then handle it as required. This may involve generating a single regop, | |
10014 | such as for an assertion; or it may involve recursing, such as to | |
10015 | handle a () structure. | |
10016 | ||
10017 | If the string doesn't start with something special then we gobble up | |
10018 | as much literal text as we can. | |
10019 | ||
10020 | Once we have been able to handle whatever type of thing started the | |
10021 | sequence, we return. | |
10022 | ||
10023 | Note: we have to be careful with escapes, as they can be both literal | |
639c2774 KW |
10024 | and special, and in the case of \10 and friends, context determines which. |
10025 | ||
10026 | A summary of the code structure is: | |
10027 | ||
10028 | switch (first_byte) { | |
10029 | cases for each special: | |
10030 | handle this special; | |
10031 | break; | |
10032 | case '\\': | |
10033 | switch (2nd byte) { | |
10034 | cases for each unambiguous special: | |
10035 | handle this special; | |
10036 | break; | |
10037 | cases for each ambigous special/literal: | |
10038 | disambiguate; | |
10039 | if (special) handle here | |
10040 | else goto defchar; | |
10041 | default: // unambiguously literal: | |
10042 | goto defchar; | |
10043 | } | |
10044 | default: // is a literal char | |
10045 | // FALL THROUGH | |
10046 | defchar: | |
10047 | create EXACTish node for literal; | |
10048 | while (more input and node isn't full) { | |
10049 | switch (input_byte) { | |
10050 | cases for each special; | |
10051 | make sure parse pointer is set so that the next call to | |
10052 | regatom will see this special first | |
10053 | goto loopdone; // EXACTish node terminated by prev. char | |
10054 | default: | |
10055 | append char to EXACTISH node; | |
10056 | } | |
10057 | get next input byte; | |
10058 | } | |
10059 | loopdone: | |
10060 | } | |
10061 | return the generated node; | |
10062 | ||
10063 | Specifically there are two separate switches for handling | |
ee9b8eae YO |
10064 | escape sequences, with the one for handling literal escapes requiring |
10065 | a dummy entry for all of the special escapes that are actually handled | |
10066 | by the other. | |
10067 | */ | |
10068 | ||
76e3520e | 10069 | STATIC regnode * |
3dab1dad | 10070 | S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth) |
a687059c | 10071 | { |
97aff369 | 10072 | dVAR; |
b9e8c997 | 10073 | regnode *ret = NULL; |
a0d0e21e | 10074 | I32 flags; |
45948336 | 10075 | char *parse_start = RExC_parse; |
980866de | 10076 | U8 op; |
3dab1dad YO |
10077 | GET_RE_DEBUG_FLAGS_DECL; |
10078 | DEBUG_PARSE("atom"); | |
a0d0e21e LW |
10079 | *flagp = WORST; /* Tentatively. */ |
10080 | ||
7918f24d | 10081 | PERL_ARGS_ASSERT_REGATOM; |
ee9b8eae | 10082 | |
a0d0e21e | 10083 | tryagain: |
f9a79580 | 10084 | switch ((U8)*RExC_parse) { |
a0d0e21e | 10085 | case '^': |
830247a4 IZ |
10086 | RExC_seen_zerolen++; |
10087 | nextchar(pRExC_state); | |
bbe252da | 10088 | if (RExC_flags & RXf_PMf_MULTILINE) |
830247a4 | 10089 | ret = reg_node(pRExC_state, MBOL); |
bbe252da | 10090 | else if (RExC_flags & RXf_PMf_SINGLELINE) |
830247a4 | 10091 | ret = reg_node(pRExC_state, SBOL); |
a0d0e21e | 10092 | else |
830247a4 | 10093 | ret = reg_node(pRExC_state, BOL); |
fac92740 | 10094 | Set_Node_Length(ret, 1); /* MJD */ |
a0d0e21e LW |
10095 | break; |
10096 | case '$': | |
830247a4 | 10097 | nextchar(pRExC_state); |
b81d288d | 10098 | if (*RExC_parse) |
830247a4 | 10099 | RExC_seen_zerolen++; |
bbe252da | 10100 | if (RExC_flags & RXf_PMf_MULTILINE) |
830247a4 | 10101 | ret = reg_node(pRExC_state, MEOL); |
bbe252da | 10102 | else if (RExC_flags & RXf_PMf_SINGLELINE) |
830247a4 | 10103 | ret = reg_node(pRExC_state, SEOL); |
a0d0e21e | 10104 | else |
830247a4 | 10105 | ret = reg_node(pRExC_state, EOL); |
fac92740 | 10106 | Set_Node_Length(ret, 1); /* MJD */ |
a0d0e21e LW |
10107 | break; |
10108 | case '.': | |
830247a4 | 10109 | nextchar(pRExC_state); |
bbe252da | 10110 | if (RExC_flags & RXf_PMf_SINGLELINE) |
ffc61ed2 JH |
10111 | ret = reg_node(pRExC_state, SANY); |
10112 | else | |
10113 | ret = reg_node(pRExC_state, REG_ANY); | |
10114 | *flagp |= HASWIDTH|SIMPLE; | |
830247a4 | 10115 | RExC_naughty++; |
fac92740 | 10116 | Set_Node_Length(ret, 1); /* MJD */ |
a0d0e21e LW |
10117 | break; |
10118 | case '[': | |
b45f050a | 10119 | { |
3dab1dad | 10120 | char * const oregcomp_parse = ++RExC_parse; |
34b39fc9 | 10121 | ret = regclass(pRExC_state, flagp,depth+1); |
830247a4 IZ |
10122 | if (*RExC_parse != ']') { |
10123 | RExC_parse = oregcomp_parse; | |
b45f050a JF |
10124 | vFAIL("Unmatched ["); |
10125 | } | |
830247a4 | 10126 | nextchar(pRExC_state); |
fac92740 | 10127 | Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */ |
a0d0e21e | 10128 | break; |
b45f050a | 10129 | } |
a0d0e21e | 10130 | case '(': |
830247a4 | 10131 | nextchar(pRExC_state); |
3dab1dad | 10132 | ret = reg(pRExC_state, 1, &flags,depth+1); |
a0d0e21e | 10133 | if (ret == NULL) { |
bf93d4cc | 10134 | if (flags & TRYAGAIN) { |
830247a4 | 10135 | if (RExC_parse == RExC_end) { |
bf93d4cc GS |
10136 | /* Make parent create an empty node if needed. */ |
10137 | *flagp |= TRYAGAIN; | |
10138 | return(NULL); | |
10139 | } | |
a0d0e21e | 10140 | goto tryagain; |
bf93d4cc | 10141 | } |
a0d0e21e LW |
10142 | return(NULL); |
10143 | } | |
a3b492c3 | 10144 | *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED); |
a0d0e21e LW |
10145 | break; |
10146 | case '|': | |
10147 | case ')': | |
10148 | if (flags & TRYAGAIN) { | |
10149 | *flagp |= TRYAGAIN; | |
10150 | return NULL; | |
10151 | } | |
b45f050a | 10152 | vFAIL("Internal urp"); |
a0d0e21e LW |
10153 | /* Supposed to be caught earlier. */ |
10154 | break; | |
10155 | case '?': | |
10156 | case '+': | |
10157 | case '*': | |
830247a4 | 10158 | RExC_parse++; |
b45f050a | 10159 | vFAIL("Quantifier follows nothing"); |
a0d0e21e LW |
10160 | break; |
10161 | case '\\': | |
ee9b8eae YO |
10162 | /* Special Escapes |
10163 | ||
10164 | This switch handles escape sequences that resolve to some kind | |
10165 | of special regop and not to literal text. Escape sequnces that | |
10166 | resolve to literal text are handled below in the switch marked | |
10167 | "Literal Escapes". | |
10168 | ||
10169 | Every entry in this switch *must* have a corresponding entry | |
10170 | in the literal escape switch. However, the opposite is not | |
10171 | required, as the default for this switch is to jump to the | |
10172 | literal text handling code. | |
10173 | */ | |
a0a388a1 | 10174 | switch ((U8)*++RExC_parse) { |
ee9b8eae | 10175 | /* Special Escapes */ |
a0d0e21e | 10176 | case 'A': |
830247a4 IZ |
10177 | RExC_seen_zerolen++; |
10178 | ret = reg_node(pRExC_state, SBOL); | |
a0d0e21e | 10179 | *flagp |= SIMPLE; |
ee9b8eae | 10180 | goto finish_meta_pat; |
a0d0e21e | 10181 | case 'G': |
830247a4 IZ |
10182 | ret = reg_node(pRExC_state, GPOS); |
10183 | RExC_seen |= REG_SEEN_GPOS; | |
a0d0e21e | 10184 | *flagp |= SIMPLE; |
ee9b8eae YO |
10185 | goto finish_meta_pat; |
10186 | case 'K': | |
10187 | RExC_seen_zerolen++; | |
10188 | ret = reg_node(pRExC_state, KEEPS); | |
10189 | *flagp |= SIMPLE; | |
37923168 RGS |
10190 | /* XXX:dmq : disabling in-place substitution seems to |
10191 | * be necessary here to avoid cases of memory corruption, as | |
10192 | * with: C<$_="x" x 80; s/x\K/y/> -- rgs | |
10193 | */ | |
10194 | RExC_seen |= REG_SEEN_LOOKBEHIND; | |
ee9b8eae | 10195 | goto finish_meta_pat; |
a0d0e21e | 10196 | case 'Z': |
830247a4 | 10197 | ret = reg_node(pRExC_state, SEOL); |
a0d0e21e | 10198 | *flagp |= SIMPLE; |
a1917ab9 | 10199 | RExC_seen_zerolen++; /* Do not optimize RE away */ |
ee9b8eae | 10200 | goto finish_meta_pat; |
b85d18e9 | 10201 | case 'z': |
830247a4 | 10202 | ret = reg_node(pRExC_state, EOS); |
b85d18e9 | 10203 | *flagp |= SIMPLE; |
830247a4 | 10204 | RExC_seen_zerolen++; /* Do not optimize RE away */ |
ee9b8eae | 10205 | goto finish_meta_pat; |
4a2d328f | 10206 | case 'C': |
f33976b4 DB |
10207 | ret = reg_node(pRExC_state, CANY); |
10208 | RExC_seen |= REG_SEEN_CANY; | |
a0ed51b3 | 10209 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 10210 | goto finish_meta_pat; |
a0ed51b3 | 10211 | case 'X': |
830247a4 | 10212 | ret = reg_node(pRExC_state, CLUMP); |
a0ed51b3 | 10213 | *flagp |= HASWIDTH; |
ee9b8eae | 10214 | goto finish_meta_pat; |
a0d0e21e | 10215 | case 'w': |
693fefec KW |
10216 | op = ALNUM + get_regex_charset(RExC_flags); |
10217 | if (op > ALNUMA) { /* /aa is same as /a */ | |
10218 | op = ALNUMA; | |
a12cf05f | 10219 | } |
980866de | 10220 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 10221 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 10222 | goto finish_meta_pat; |
a0d0e21e | 10223 | case 'W': |
693fefec KW |
10224 | op = NALNUM + get_regex_charset(RExC_flags); |
10225 | if (op > NALNUMA) { /* /aa is same as /a */ | |
10226 | op = NALNUMA; | |
a12cf05f | 10227 | } |
980866de | 10228 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 10229 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 10230 | goto finish_meta_pat; |
a0d0e21e | 10231 | case 'b': |
830247a4 IZ |
10232 | RExC_seen_zerolen++; |
10233 | RExC_seen |= REG_SEEN_LOOKBEHIND; | |
693fefec KW |
10234 | op = BOUND + get_regex_charset(RExC_flags); |
10235 | if (op > BOUNDA) { /* /aa is same as /a */ | |
10236 | op = BOUNDA; | |
a12cf05f | 10237 | } |
63ac0dad | 10238 | ret = reg_node(pRExC_state, op); |
b988e673 | 10239 | FLAGS(ret) = get_regex_charset(RExC_flags); |
a0d0e21e | 10240 | *flagp |= SIMPLE; |
ee9b8eae | 10241 | goto finish_meta_pat; |
a0d0e21e | 10242 | case 'B': |
830247a4 IZ |
10243 | RExC_seen_zerolen++; |
10244 | RExC_seen |= REG_SEEN_LOOKBEHIND; | |
693fefec KW |
10245 | op = NBOUND + get_regex_charset(RExC_flags); |
10246 | if (op > NBOUNDA) { /* /aa is same as /a */ | |
10247 | op = NBOUNDA; | |
a12cf05f | 10248 | } |
63ac0dad | 10249 | ret = reg_node(pRExC_state, op); |
b988e673 | 10250 | FLAGS(ret) = get_regex_charset(RExC_flags); |
a0d0e21e | 10251 | *flagp |= SIMPLE; |
ee9b8eae | 10252 | goto finish_meta_pat; |
a0d0e21e | 10253 | case 's': |
693fefec KW |
10254 | op = SPACE + get_regex_charset(RExC_flags); |
10255 | if (op > SPACEA) { /* /aa is same as /a */ | |
10256 | op = SPACEA; | |
a12cf05f | 10257 | } |
980866de | 10258 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 10259 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 10260 | goto finish_meta_pat; |
a0d0e21e | 10261 | case 'S': |
693fefec KW |
10262 | op = NSPACE + get_regex_charset(RExC_flags); |
10263 | if (op > NSPACEA) { /* /aa is same as /a */ | |
10264 | op = NSPACEA; | |
6ab9ea91 | 10265 | } |
56ae17b4 | 10266 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 10267 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 10268 | goto finish_meta_pat; |
a0d0e21e | 10269 | case 'D': |
693fefec KW |
10270 | op = NDIGIT; |
10271 | goto join_D_and_d; | |
10272 | case 'd': | |
10273 | op = DIGIT; | |
10274 | join_D_and_d: | |
10275 | { | |
10276 | U8 offset = get_regex_charset(RExC_flags); | |
10277 | if (offset == REGEX_UNICODE_CHARSET) { | |
10278 | offset = REGEX_DEPENDS_CHARSET; | |
10279 | } | |
10280 | else if (offset == REGEX_ASCII_MORE_RESTRICTED_CHARSET) { | |
10281 | offset = REGEX_ASCII_RESTRICTED_CHARSET; | |
10282 | } | |
10283 | op += offset; | |
6ab9ea91 | 10284 | } |
56ae17b4 | 10285 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 10286 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 10287 | goto finish_meta_pat; |
e1d1eefb YO |
10288 | case 'R': |
10289 | ret = reg_node(pRExC_state, LNBREAK); | |
10290 | *flagp |= HASWIDTH|SIMPLE; | |
10291 | goto finish_meta_pat; | |
10292 | case 'h': | |
10293 | ret = reg_node(pRExC_state, HORIZWS); | |
10294 | *flagp |= HASWIDTH|SIMPLE; | |
10295 | goto finish_meta_pat; | |
10296 | case 'H': | |
10297 | ret = reg_node(pRExC_state, NHORIZWS); | |
10298 | *flagp |= HASWIDTH|SIMPLE; | |
10299 | goto finish_meta_pat; | |
ee9b8eae | 10300 | case 'v': |
e1d1eefb YO |
10301 | ret = reg_node(pRExC_state, VERTWS); |
10302 | *flagp |= HASWIDTH|SIMPLE; | |
ee9b8eae YO |
10303 | goto finish_meta_pat; |
10304 | case 'V': | |
e1d1eefb YO |
10305 | ret = reg_node(pRExC_state, NVERTWS); |
10306 | *flagp |= HASWIDTH|SIMPLE; | |
ee9b8eae | 10307 | finish_meta_pat: |
830247a4 | 10308 | nextchar(pRExC_state); |
fac92740 | 10309 | Set_Node_Length(ret, 2); /* MJD */ |
ee9b8eae | 10310 | break; |
a14b48bc LW |
10311 | case 'p': |
10312 | case 'P': | |
686b73d4 | 10313 | { |
3dab1dad | 10314 | char* const oldregxend = RExC_end; |
d008bc60 | 10315 | #ifdef DEBUGGING |
ccb2c380 | 10316 | char* parse_start = RExC_parse - 2; |
d008bc60 | 10317 | #endif |
a14b48bc | 10318 | |
830247a4 | 10319 | if (RExC_parse[1] == '{') { |
3568d838 | 10320 | /* a lovely hack--pretend we saw [\pX] instead */ |
830247a4 IZ |
10321 | RExC_end = strchr(RExC_parse, '}'); |
10322 | if (!RExC_end) { | |
3dab1dad | 10323 | const U8 c = (U8)*RExC_parse; |
830247a4 IZ |
10324 | RExC_parse += 2; |
10325 | RExC_end = oldregxend; | |
0da60cf5 | 10326 | vFAIL2("Missing right brace on \\%c{}", c); |
b45f050a | 10327 | } |
830247a4 | 10328 | RExC_end++; |
a14b48bc | 10329 | } |
af6f566e | 10330 | else { |
830247a4 | 10331 | RExC_end = RExC_parse + 2; |
af6f566e HS |
10332 | if (RExC_end > oldregxend) |
10333 | RExC_end = oldregxend; | |
10334 | } | |
830247a4 | 10335 | RExC_parse--; |
a14b48bc | 10336 | |
34b39fc9 | 10337 | ret = regclass(pRExC_state, flagp,depth+1); |
a14b48bc | 10338 | |
830247a4 IZ |
10339 | RExC_end = oldregxend; |
10340 | RExC_parse--; | |
ccb2c380 MP |
10341 | |
10342 | Set_Node_Offset(ret, parse_start + 2); | |
10343 | Set_Node_Cur_Length(ret); | |
830247a4 | 10344 | nextchar(pRExC_state); |
a14b48bc LW |
10345 | } |
10346 | break; | |
fc8cd66c | 10347 | case 'N': |
b9c48b5b KW |
10348 | /* Handle \N and \N{NAME} with multiple code points here and not |
10349 | * below because it can be multicharacter. join_exact() will join | |
10350 | * them up later on. Also this makes sure that things like | |
10351 | * /\N{BLAH}+/ and \N{BLAH} being multi char Just Happen. dmq. | |
10352 | * The options to the grok function call causes it to fail if the | |
10353 | * sequence is just a single code point. We then go treat it as | |
10354 | * just another character in the current EXACT node, and hence it | |
10355 | * gets uniform treatment with all the other characters. The | |
10356 | * special treatment for quantifiers is not needed for such single | |
10357 | * character sequences */ | |
fc8cd66c | 10358 | ++RExC_parse; |
b9c48b5b KW |
10359 | if (! grok_bslash_N(pRExC_state, &ret, NULL, flagp, depth, FALSE)) { |
10360 | RExC_parse--; | |
10361 | goto defchar; | |
10362 | } | |
fc8cd66c | 10363 | break; |
0a4db386 | 10364 | case 'k': /* Handle \k<NAME> and \k'NAME' */ |
1f1031fe | 10365 | parse_named_seq: |
81714fb9 YO |
10366 | { |
10367 | char ch= RExC_parse[1]; | |
1f1031fe YO |
10368 | if (ch != '<' && ch != '\'' && ch != '{') { |
10369 | RExC_parse++; | |
10370 | vFAIL2("Sequence %.2s... not terminated",parse_start); | |
81714fb9 | 10371 | } else { |
1f1031fe YO |
10372 | /* this pretty much dupes the code for (?P=...) in reg(), if |
10373 | you change this make sure you change that */ | |
81714fb9 | 10374 | char* name_start = (RExC_parse += 2); |
2eccd3b2 | 10375 | U32 num = 0; |
0a4db386 YO |
10376 | SV *sv_dat = reg_scan_name(pRExC_state, |
10377 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
1f1031fe | 10378 | ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\''; |
81714fb9 | 10379 | if (RExC_parse == name_start || *RExC_parse != ch) |
1f1031fe YO |
10380 | vFAIL2("Sequence %.3s... not terminated",parse_start); |
10381 | ||
10382 | if (!SIZE_ONLY) { | |
10383 | num = add_data( pRExC_state, 1, "S" ); | |
10384 | RExC_rxi->data->data[num]=(void*)sv_dat; | |
5a5094bd | 10385 | SvREFCNT_inc_simple_void(sv_dat); |
1f1031fe YO |
10386 | } |
10387 | ||
81714fb9 YO |
10388 | RExC_sawback = 1; |
10389 | ret = reganode(pRExC_state, | |
4444fd9f KW |
10390 | ((! FOLD) |
10391 | ? NREF | |
a725e29c | 10392 | : (ASCII_FOLD_RESTRICTED) |
2f7f8cb1 KW |
10393 | ? NREFFA |
10394 | : (AT_LEAST_UNI_SEMANTICS) | |
10395 | ? NREFFU | |
10396 | : (LOC) | |
10397 | ? NREFFL | |
10398 | : NREFF), | |
4444fd9f | 10399 | num); |
81714fb9 | 10400 | *flagp |= HASWIDTH; |
1f1031fe | 10401 | |
81714fb9 YO |
10402 | /* override incorrect value set in reganode MJD */ |
10403 | Set_Node_Offset(ret, parse_start+1); | |
10404 | Set_Node_Cur_Length(ret); /* MJD */ | |
10405 | nextchar(pRExC_state); | |
1f1031fe | 10406 | |
81714fb9 YO |
10407 | } |
10408 | break; | |
1f1031fe | 10409 | } |
2bf803e2 | 10410 | case 'g': |
a0d0e21e LW |
10411 | case '1': case '2': case '3': case '4': |
10412 | case '5': case '6': case '7': case '8': case '9': | |
10413 | { | |
c74340f9 | 10414 | I32 num; |
2bf803e2 YO |
10415 | bool isg = *RExC_parse == 'g'; |
10416 | bool isrel = 0; | |
10417 | bool hasbrace = 0; | |
10418 | if (isg) { | |
c74340f9 | 10419 | RExC_parse++; |
2bf803e2 YO |
10420 | if (*RExC_parse == '{') { |
10421 | RExC_parse++; | |
10422 | hasbrace = 1; | |
10423 | } | |
10424 | if (*RExC_parse == '-') { | |
10425 | RExC_parse++; | |
10426 | isrel = 1; | |
10427 | } | |
1f1031fe YO |
10428 | if (hasbrace && !isDIGIT(*RExC_parse)) { |
10429 | if (isrel) RExC_parse--; | |
10430 | RExC_parse -= 2; | |
10431 | goto parse_named_seq; | |
10432 | } } | |
c74340f9 | 10433 | num = atoi(RExC_parse); |
b72d83b2 RGS |
10434 | if (isg && num == 0) |
10435 | vFAIL("Reference to invalid group 0"); | |
c74340f9 | 10436 | if (isrel) { |
5624f11d | 10437 | num = RExC_npar - num; |
c74340f9 YO |
10438 | if (num < 1) |
10439 | vFAIL("Reference to nonexistent or unclosed group"); | |
10440 | } | |
2bf803e2 | 10441 | if (!isg && num > 9 && num >= RExC_npar) |
639c2774 | 10442 | /* Probably a character specified in octal, e.g. \35 */ |
a0d0e21e LW |
10443 | goto defchar; |
10444 | else { | |
3dab1dad | 10445 | char * const parse_start = RExC_parse - 1; /* MJD */ |
830247a4 IZ |
10446 | while (isDIGIT(*RExC_parse)) |
10447 | RExC_parse++; | |
1f1031fe YO |
10448 | if (parse_start == RExC_parse - 1) |
10449 | vFAIL("Unterminated \\g... pattern"); | |
2bf803e2 YO |
10450 | if (hasbrace) { |
10451 | if (*RExC_parse != '}') | |
10452 | vFAIL("Unterminated \\g{...} pattern"); | |
10453 | RExC_parse++; | |
10454 | } | |
c74340f9 YO |
10455 | if (!SIZE_ONLY) { |
10456 | if (num > (I32)RExC_rx->nparens) | |
10457 | vFAIL("Reference to nonexistent group"); | |
c74340f9 | 10458 | } |
830247a4 | 10459 | RExC_sawback = 1; |
eb160463 | 10460 | ret = reganode(pRExC_state, |
4444fd9f KW |
10461 | ((! FOLD) |
10462 | ? REF | |
a725e29c | 10463 | : (ASCII_FOLD_RESTRICTED) |
2f7f8cb1 KW |
10464 | ? REFFA |
10465 | : (AT_LEAST_UNI_SEMANTICS) | |
10466 | ? REFFU | |
10467 | : (LOC) | |
10468 | ? REFFL | |
10469 | : REFF), | |
4444fd9f | 10470 | num); |
a0d0e21e | 10471 | *flagp |= HASWIDTH; |
2af232bd | 10472 | |
fac92740 | 10473 | /* override incorrect value set in reganode MJD */ |
2af232bd | 10474 | Set_Node_Offset(ret, parse_start+1); |
fac92740 | 10475 | Set_Node_Cur_Length(ret); /* MJD */ |
830247a4 IZ |
10476 | RExC_parse--; |
10477 | nextchar(pRExC_state); | |
a0d0e21e LW |
10478 | } |
10479 | } | |
10480 | break; | |
10481 | case '\0': | |
830247a4 | 10482 | if (RExC_parse >= RExC_end) |
b45f050a | 10483 | FAIL("Trailing \\"); |
a0d0e21e LW |
10484 | /* FALL THROUGH */ |
10485 | default: | |
a0288114 | 10486 | /* Do not generate "unrecognized" warnings here, we fall |
c9f97d15 | 10487 | back into the quick-grab loop below */ |
45948336 | 10488 | parse_start--; |
a0d0e21e LW |
10489 | goto defchar; |
10490 | } | |
10491 | break; | |
4633a7c4 LW |
10492 | |
10493 | case '#': | |
bbe252da | 10494 | if (RExC_flags & RXf_PMf_EXTENDED) { |
bcdf7404 | 10495 | if ( reg_skipcomment( pRExC_state ) ) |
4633a7c4 LW |
10496 | goto tryagain; |
10497 | } | |
10498 | /* FALL THROUGH */ | |
10499 | ||
f9a79580 | 10500 | default: |
561784a5 KW |
10501 | |
10502 | parse_start = RExC_parse - 1; | |
10503 | ||
10504 | RExC_parse++; | |
10505 | ||
10506 | defchar: { | |
eb578fdb | 10507 | STRLEN len = 0; |
b9c48b5b | 10508 | UV ender; |
eb578fdb | 10509 | char *p; |
3dab1dad | 10510 | char *s; |
3d26503b | 10511 | #define MAX_NODE_STRING_SIZE 127 |
26faadbd KW |
10512 | char foldbuf[MAX_NODE_STRING_SIZE+UTF8_MAXBYTES_CASE]; |
10513 | char *s0; | |
10514 | U8 upper_parse = MAX_NODE_STRING_SIZE; | |
80aecb99 | 10515 | STRLEN foldlen; |
d82f9944 | 10516 | U8 node_type; |
271751d1 | 10517 | bool next_is_quantifier; |
d9fc6391 | 10518 | char * oldp = NULL; |
f06dbbb7 | 10519 | |
58ae7d3f | 10520 | ender = 0; |
86dc2bf5 | 10521 | node_type = compute_EXACTish(pRExC_state); |
d82f9944 | 10522 | ret = reg_node(pRExC_state, node_type); |
4924e08c KW |
10523 | |
10524 | /* In pass1, folded, we use a temporary buffer instead of the | |
10525 | * actual node, as the node doesn't exist yet */ | |
10526 | s = (SIZE_ONLY && FOLD) ? foldbuf : STRING(ret); | |
3f410cf6 | 10527 | |
26faadbd KW |
10528 | s0 = s; |
10529 | ||
10530 | reparse: | |
10531 | ||
3f410cf6 KW |
10532 | /* XXX The node can hold up to 255 bytes, yet this only goes to |
10533 | * 127. I (khw) do not know why. Keeping it somewhat less than | |
10534 | * 255 allows us to not have to worry about overflow due to | |
10535 | * converting to utf8 and fold expansion, but that value is | |
10536 | * 255-UTF8_MAXBYTES_CASE. join_exact() may join adjacent nodes | |
10537 | * split up by this limit into a single one using the real max of | |
10538 | * 255. Even at 127, this breaks under rare circumstances. If | |
10539 | * folding, we do not want to split a node at a character that is a | |
10540 | * non-final in a multi-char fold, as an input string could just | |
10541 | * happen to want to match across the node boundary. The join | |
10542 | * would solve that problem if the join actually happens. But a | |
10543 | * series of more than two nodes in a row each of 127 would cause | |
10544 | * the first join to succeed to get to 254, but then there wouldn't | |
10545 | * be room for the next one, which could at be one of those split | |
10546 | * multi-char folds. I don't know of any fool-proof solution. One | |
10547 | * could back off to end with only a code point that isn't such a | |
10548 | * non-final, but it is possible for there not to be any in the | |
10549 | * entire node. */ | |
26faadbd KW |
10550 | for (p = RExC_parse - 1; |
10551 | len < upper_parse && p < RExC_end; | |
3f410cf6 | 10552 | len++) |
a0d0e21e | 10553 | { |
26faadbd | 10554 | oldp = p; |
5b5a24f7 | 10555 | |
bbe252da | 10556 | if (RExC_flags & RXf_PMf_EXTENDED) |
bcdf7404 | 10557 | p = regwhite( pRExC_state, p ); |
f9a79580 | 10558 | switch ((U8)*p) { |
a0d0e21e LW |
10559 | case '^': |
10560 | case '$': | |
10561 | case '.': | |
10562 | case '[': | |
10563 | case '(': | |
10564 | case ')': | |
10565 | case '|': | |
10566 | goto loopdone; | |
10567 | case '\\': | |
ee9b8eae YO |
10568 | /* Literal Escapes Switch |
10569 | ||
10570 | This switch is meant to handle escape sequences that | |
10571 | resolve to a literal character. | |
10572 | ||
10573 | Every escape sequence that represents something | |
10574 | else, like an assertion or a char class, is handled | |
10575 | in the switch marked 'Special Escapes' above in this | |
10576 | routine, but also has an entry here as anything that | |
10577 | isn't explicitly mentioned here will be treated as | |
10578 | an unescaped equivalent literal. | |
10579 | */ | |
10580 | ||
a0a388a1 | 10581 | switch ((U8)*++p) { |
ee9b8eae YO |
10582 | /* These are all the special escapes. */ |
10583 | case 'A': /* Start assertion */ | |
10584 | case 'b': case 'B': /* Word-boundary assertion*/ | |
10585 | case 'C': /* Single char !DANGEROUS! */ | |
10586 | case 'd': case 'D': /* digit class */ | |
10587 | case 'g': case 'G': /* generic-backref, pos assertion */ | |
e1d1eefb | 10588 | case 'h': case 'H': /* HORIZWS */ |
ee9b8eae | 10589 | case 'k': case 'K': /* named backref, keep marker */ |
38a44b82 | 10590 | case 'p': case 'P': /* Unicode property */ |
e1d1eefb | 10591 | case 'R': /* LNBREAK */ |
ee9b8eae | 10592 | case 's': case 'S': /* space class */ |
e1d1eefb | 10593 | case 'v': case 'V': /* VERTWS */ |
ee9b8eae YO |
10594 | case 'w': case 'W': /* word class */ |
10595 | case 'X': /* eXtended Unicode "combining character sequence" */ | |
10596 | case 'z': case 'Z': /* End of line/string assertion */ | |
a0d0e21e LW |
10597 | --p; |
10598 | goto loopdone; | |
ee9b8eae YO |
10599 | |
10600 | /* Anything after here is an escape that resolves to a | |
10601 | literal. (Except digits, which may or may not) | |
10602 | */ | |
a0d0e21e LW |
10603 | case 'n': |
10604 | ender = '\n'; | |
10605 | p++; | |
a687059c | 10606 | break; |
b9c48b5b KW |
10607 | case 'N': /* Handle a single-code point named character. */ |
10608 | /* The options cause it to fail if a multiple code | |
10609 | * point sequence. Handle those in the switch() above | |
10610 | * */ | |
10611 | RExC_parse = p + 1; | |
10612 | if (! grok_bslash_N(pRExC_state, NULL, &ender, | |
10613 | flagp, depth, FALSE)) | |
10614 | { | |
10615 | RExC_parse = p = oldp; | |
10616 | goto loopdone; | |
10617 | } | |
10618 | p = RExC_parse; | |
78412e20 KW |
10619 | if (ender > 0xff) { |
10620 | REQUIRE_UTF8; | |
10621 | } | |
b9c48b5b | 10622 | break; |
a0d0e21e LW |
10623 | case 'r': |
10624 | ender = '\r'; | |
10625 | p++; | |
a687059c | 10626 | break; |
a0d0e21e LW |
10627 | case 't': |
10628 | ender = '\t'; | |
10629 | p++; | |
a687059c | 10630 | break; |
a0d0e21e LW |
10631 | case 'f': |
10632 | ender = '\f'; | |
10633 | p++; | |
a687059c | 10634 | break; |
a0d0e21e | 10635 | case 'e': |
c7f1f016 | 10636 | ender = ASCII_TO_NATIVE('\033'); |
a0d0e21e | 10637 | p++; |
a687059c | 10638 | break; |
a0d0e21e | 10639 | case 'a': |
c7f1f016 | 10640 | ender = ASCII_TO_NATIVE('\007'); |
a0d0e21e | 10641 | p++; |
a687059c | 10642 | break; |
f0a2b745 KW |
10643 | case 'o': |
10644 | { | |
10645 | STRLEN brace_len = len; | |
00c0cb6d | 10646 | UV result; |
454155d9 KW |
10647 | const char* error_msg; |
10648 | ||
10649 | bool valid = grok_bslash_o(p, | |
10650 | &result, | |
10651 | &brace_len, | |
10652 | &error_msg, | |
10653 | 1); | |
10654 | p += brace_len; | |
10655 | if (! valid) { | |
10656 | RExC_parse = p; /* going to die anyway; point | |
10657 | to exact spot of failure */ | |
f0a2b745 KW |
10658 | vFAIL(error_msg); |
10659 | } | |
00c0cb6d DG |
10660 | else |
10661 | { | |
10662 | ender = result; | |
10663 | } | |
f0a2b745 KW |
10664 | if (PL_encoding && ender < 0x100) { |
10665 | goto recode_encoding; | |
10666 | } | |
10667 | if (ender > 0xff) { | |
62fed28b | 10668 | REQUIRE_UTF8; |
f0a2b745 KW |
10669 | } |
10670 | break; | |
10671 | } | |
a0d0e21e | 10672 | case 'x': |
a0481293 KW |
10673 | { |
10674 | STRLEN brace_len = len; | |
10675 | UV result; | |
10676 | const char* error_msg; | |
686b73d4 | 10677 | |
a0481293 KW |
10678 | bool valid = grok_bslash_x(p, |
10679 | &result, | |
10680 | &brace_len, | |
10681 | &error_msg, | |
10682 | 1); | |
10683 | p += brace_len; | |
10684 | if (! valid) { | |
10685 | RExC_parse = p; /* going to die anyway; point | |
10686 | to exact spot of failure */ | |
10687 | vFAIL(error_msg); | |
b45f050a | 10688 | } |
de5f0749 | 10689 | else { |
a0481293 | 10690 | ender = result; |
a0ed51b3 | 10691 | } |
a0481293 KW |
10692 | if (PL_encoding && ender < 0x100) { |
10693 | goto recode_encoding; | |
10694 | } | |
10695 | if (ender > 0xff) { | |
10696 | REQUIRE_UTF8; | |
10697 | } | |
10698 | break; | |
a0ed51b3 | 10699 | } |
a0d0e21e LW |
10700 | case 'c': |
10701 | p++; | |
17a3df4c | 10702 | ender = grok_bslash_c(*p++, UTF, SIZE_ONLY); |
a687059c | 10703 | break; |
a0d0e21e | 10704 | case '0': case '1': case '2': case '3':case '4': |
726ee55d | 10705 | case '5': case '6': case '7': |
a0d0e21e | 10706 | if (*p == '0' || |
ca67da41 | 10707 | (isDIGIT(p[1]) && atoi(p) >= RExC_npar)) |
c99e91e9 KW |
10708 | { |
10709 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT; | |
1df70142 | 10710 | STRLEN numlen = 3; |
53305cf1 | 10711 | ender = grok_oct(p, &numlen, &flags, NULL); |
fa1639c5 | 10712 | if (ender > 0xff) { |
62fed28b | 10713 | REQUIRE_UTF8; |
609122bd | 10714 | } |
a0d0e21e LW |
10715 | p += numlen; |
10716 | } | |
10717 | else { | |
10718 | --p; | |
10719 | goto loopdone; | |
a687059c | 10720 | } |
9e08bc66 TS |
10721 | if (PL_encoding && ender < 0x100) |
10722 | goto recode_encoding; | |
10723 | break; | |
10724 | recode_encoding: | |
e2a7e165 | 10725 | if (! RExC_override_recoding) { |
9e08bc66 TS |
10726 | SV* enc = PL_encoding; |
10727 | ender = reg_recode((const char)(U8)ender, &enc); | |
668c081a NC |
10728 | if (!enc && SIZE_ONLY) |
10729 | ckWARNreg(p, "Invalid escape in the specified encoding"); | |
62fed28b | 10730 | REQUIRE_UTF8; |
9e08bc66 | 10731 | } |
a687059c | 10732 | break; |
a0d0e21e | 10733 | case '\0': |
830247a4 | 10734 | if (p >= RExC_end) |
b45f050a | 10735 | FAIL("Trailing \\"); |
a687059c | 10736 | /* FALL THROUGH */ |
a0d0e21e | 10737 | default: |
726ee55d | 10738 | if (!SIZE_ONLY&& isALNUMC(*p)) { |
2a53d331 | 10739 | ckWARN2reg(p + 1, "Unrecognized escape \\%.1s passed through", p); |
216bfc0a | 10740 | } |
a0ed51b3 | 10741 | goto normal_default; |
a0d0e21e LW |
10742 | } |
10743 | break; | |
2a53d331 KW |
10744 | case '{': |
10745 | /* Currently we don't warn when the lbrace is at the start | |
10746 | * of a construct. This catches it in the middle of a | |
10747 | * literal string, or when its the first thing after | |
10748 | * something like "\b" */ | |
10749 | if (! SIZE_ONLY | |
10750 | && (len || (p > RExC_start && isALPHA_A(*(p -1))))) | |
10751 | { | |
10752 | ckWARNregdep(p + 1, "Unescaped left brace in regex is deprecated, passed through"); | |
10753 | } | |
10754 | /*FALLTHROUGH*/ | |
a687059c | 10755 | default: |
a0ed51b3 | 10756 | normal_default: |
fd400ab9 | 10757 | if (UTF8_IS_START(*p) && UTF) { |
1df70142 | 10758 | STRLEN numlen; |
5e12f4fb | 10759 | ender = utf8n_to_uvchr((U8*)p, RExC_end - p, |
9f7f3913 | 10760 | &numlen, UTF8_ALLOW_DEFAULT); |
a0ed51b3 LW |
10761 | p += numlen; |
10762 | } | |
10763 | else | |
5b67c30a | 10764 | ender = (U8) *p++; |
a0d0e21e | 10765 | break; |
7e2509c1 KW |
10766 | } /* End of switch on the literal */ |
10767 | ||
26faadbd KW |
10768 | /* Here, have looked at the literal character and <ender> |
10769 | * contains its ordinal, <p> points to the character after it | |
10770 | */ | |
10771 | ||
bcdf7404 YO |
10772 | if ( RExC_flags & RXf_PMf_EXTENDED) |
10773 | p = regwhite( pRExC_state, p ); | |
271751d1 KW |
10774 | |
10775 | /* If the next thing is a quantifier, it applies to this | |
10776 | * character only, which means that this character has to be in | |
10777 | * its own node and can't just be appended to the string in an | |
10778 | * existing node, so if there are already other characters in | |
10779 | * the node, close the node with just them, and set up to do | |
10780 | * this character again next time through, when it will be the | |
10781 | * only thing in its new node */ | |
10782 | if ((next_is_quantifier = (p < RExC_end && ISMULT2(p))) && len) | |
10783 | { | |
10784 | p = oldp; | |
10785 | goto loopdone; | |
10786 | } | |
10787 | ||
4924e08c | 10788 | if (FOLD) { |
aed19a82 KW |
10789 | if (UTF |
10790 | /* See comments for join_exact() as to why we fold | |
10791 | * this non-UTF at compile time */ | |
10792 | || (node_type == EXACTFU | |
10793 | && ender == LATIN_SMALL_LETTER_SHARP_S)) | |
10794 | { | |
10795 | ||
4924e08c | 10796 | |
5c4c47ec KW |
10797 | /* Prime the casefolded buffer. Locale rules, which |
10798 | * apply only to code points < 256, aren't known until | |
10799 | * execution, so for them, just output the original | |
10800 | * character using utf8. If we start to fold non-UTF | |
10801 | * patterns, be sure to update join_exact() */ | |
10802 | if (LOC && ender < 256) { | |
10803 | if (UNI_IS_INVARIANT(ender)) { | |
10804 | *s = (U8) ender; | |
10805 | foldlen = 1; | |
10806 | } else { | |
10807 | *s = UTF8_TWO_BYTE_HI(ender); | |
10808 | *(s + 1) = UTF8_TWO_BYTE_LO(ender); | |
10809 | foldlen = 2; | |
10810 | } | |
10811 | } | |
10812 | else { | |
10813 | ender = _to_uni_fold_flags(ender, (U8 *) s, &foldlen, | |
10814 | FOLD_FLAGS_FULL | |
10815 | | ((LOC) ? FOLD_FLAGS_LOCALE | |
10816 | : (ASCII_FOLD_RESTRICTED) | |
10817 | ? FOLD_FLAGS_NOMIX_ASCII | |
10818 | : 0) | |
10819 | ); | |
10820 | } | |
4924e08c KW |
10821 | s += foldlen; |
10822 | ||
10823 | /* The loop increments <len> each time, as all but this | |
10824 | * path (and the one just below for UTF) through it add | |
10825 | * a single byte to the EXACTish node. But this one | |
10826 | * has changed len to be the correct final value, so | |
10827 | * subtract one to cancel out the increment that | |
10828 | * follows */ | |
10829 | len += foldlen - 1; | |
5c4c47ec KW |
10830 | } |
10831 | else { | |
26faadbd | 10832 | *(s++) = ender; |
5c4c47ec | 10833 | } |
4924e08c KW |
10834 | } |
10835 | else if (UTF) { | |
5c4c47ec KW |
10836 | const STRLEN unilen = reguni(pRExC_state, ender, s); |
10837 | if (unilen > 0) { | |
10838 | s += unilen; | |
10839 | len += unilen; | |
10840 | } | |
89bc6737 | 10841 | |
4924e08c | 10842 | /* See comment just above for - 1 */ |
5c4c47ec | 10843 | len--; |
a0ed51b3 | 10844 | } |
d669c36c | 10845 | else { |
eb160463 | 10846 | REGC((char)ender, s++); |
89bc6737 KW |
10847 | } |
10848 | ||
10849 | if (next_is_quantifier) { | |
10850 | ||
10851 | /* Here, the next input is a quantifier, and to get here, | |
10852 | * the current character is the only one in the node. | |
10853 | * Also, here <len> doesn't include the final byte for this | |
10854 | * character */ | |
10855 | len++; | |
10856 | goto loopdone; | |
d669c36c | 10857 | } |
87b8b349 KW |
10858 | |
10859 | } /* End of loop through literal characters */ | |
10860 | ||
26faadbd KW |
10861 | /* Here we have either exhausted the input or ran out of room in |
10862 | * the node. (If we encountered a character that can't be in the | |
10863 | * node, transfer is made directly to <loopdone>, and so we | |
10864 | * wouldn't have fallen off the end of the loop.) In the latter | |
10865 | * case, we artificially have to split the node into two, because | |
10866 | * we just don't have enough space to hold everything. This | |
10867 | * creates a problem if the final character participates in a | |
10868 | * multi-character fold in the non-final position, as a match that | |
10869 | * should have occurred won't, due to the way nodes are matched, | |
10870 | * and our artificial boundary. So back off until we find a non- | |
10871 | * problematic character -- one that isn't at the beginning or | |
10872 | * middle of such a fold. (Either it doesn't participate in any | |
10873 | * folds, or appears only in the final position of all the folds it | |
10874 | * does participate in.) A better solution with far fewer false | |
10875 | * positives, and that would fill the nodes more completely, would | |
10876 | * be to actually have available all the multi-character folds to | |
10877 | * test against, and to back-off only far enough to be sure that | |
10878 | * this node isn't ending with a partial one. <upper_parse> is set | |
10879 | * further below (if we need to reparse the node) to include just | |
10880 | * up through that final non-problematic character that this code | |
10881 | * identifies, so when it is set to less than the full node, we can | |
10882 | * skip the rest of this */ | |
10883 | if (FOLD && p < RExC_end && upper_parse == MAX_NODE_STRING_SIZE) { | |
10884 | ||
10885 | const STRLEN full_len = len; | |
10886 | ||
10887 | assert(len >= MAX_NODE_STRING_SIZE); | |
10888 | ||
10889 | /* Here, <s> points to the final byte of the final character. | |
10890 | * Look backwards through the string until find a non- | |
10891 | * problematic character */ | |
10892 | ||
10893 | if (! UTF) { | |
10894 | ||
10895 | /* These two have no multi-char folds to non-UTF characters | |
10896 | */ | |
10897 | if (ASCII_FOLD_RESTRICTED || LOC) { | |
10898 | goto loopdone; | |
10899 | } | |
10900 | ||
10901 | while (--s >= s0 && IS_NON_FINAL_FOLD(*s)) { } | |
10902 | len = s - s0 + 1; | |
10903 | } | |
10904 | else { | |
10905 | if (! PL_NonL1NonFinalFold) { | |
10906 | PL_NonL1NonFinalFold = _new_invlist_C_array( | |
10907 | NonL1_Perl_Non_Final_Folds_invlist); | |
10908 | } | |
10909 | ||
10910 | /* Point to the first byte of the final character */ | |
10911 | s = (char *) utf8_hop((U8 *) s, -1); | |
10912 | ||
10913 | while (s >= s0) { /* Search backwards until find | |
10914 | non-problematic char */ | |
10915 | if (UTF8_IS_INVARIANT(*s)) { | |
10916 | ||
10917 | /* There are no ascii characters that participate | |
10918 | * in multi-char folds under /aa. In EBCDIC, the | |
10919 | * non-ascii invariants are all control characters, | |
10920 | * so don't ever participate in any folds. */ | |
10921 | if (ASCII_FOLD_RESTRICTED | |
10922 | || ! IS_NON_FINAL_FOLD(*s)) | |
10923 | { | |
10924 | break; | |
10925 | } | |
10926 | } | |
10927 | else if (UTF8_IS_DOWNGRADEABLE_START(*s)) { | |
10928 | ||
10929 | /* No Latin1 characters participate in multi-char | |
10930 | * folds under /l */ | |
10931 | if (LOC | |
10932 | || ! IS_NON_FINAL_FOLD(TWO_BYTE_UTF8_TO_UNI( | |
10933 | *s, *(s+1)))) | |
10934 | { | |
10935 | break; | |
10936 | } | |
10937 | } | |
10938 | else if (! _invlist_contains_cp( | |
10939 | PL_NonL1NonFinalFold, | |
10940 | valid_utf8_to_uvchr((U8 *) s, NULL))) | |
10941 | { | |
10942 | break; | |
10943 | } | |
10944 | ||
10945 | /* Here, the current character is problematic in that | |
10946 | * it does occur in the non-final position of some | |
10947 | * fold, so try the character before it, but have to | |
10948 | * special case the very first byte in the string, so | |
10949 | * we don't read outside the string */ | |
10950 | s = (s == s0) ? s -1 : (char *) utf8_hop((U8 *) s, -1); | |
10951 | } /* End of loop backwards through the string */ | |
10952 | ||
10953 | /* If there were only problematic characters in the string, | |
10954 | * <s> will point to before s0, in which case the length | |
10955 | * should be 0, otherwise include the length of the | |
10956 | * non-problematic character just found */ | |
10957 | len = (s < s0) ? 0 : s - s0 + UTF8SKIP(s); | |
10958 | } | |
10959 | ||
10960 | /* Here, have found the final character, if any, that is | |
10961 | * non-problematic as far as ending the node without splitting | |
10962 | * it across a potential multi-char fold. <len> contains the | |
10963 | * number of bytes in the node up-to and including that | |
10964 | * character, or is 0 if there is no such character, meaning | |
10965 | * the whole node contains only problematic characters. In | |
10966 | * this case, give up and just take the node as-is. We can't | |
10967 | * do any better */ | |
10968 | if (len == 0) { | |
10969 | len = full_len; | |
10970 | } else { | |
10971 | ||
10972 | /* Here, the node does contain some characters that aren't | |
10973 | * problematic. If one such is the final character in the | |
10974 | * node, we are done */ | |
10975 | if (len == full_len) { | |
10976 | goto loopdone; | |
10977 | } | |
10978 | else if (len + ((UTF) ? UTF8SKIP(s) : 1) == full_len) { | |
10979 | ||
10980 | /* If the final character is problematic, but the | |
10981 | * penultimate is not, back-off that last character to | |
10982 | * later start a new node with it */ | |
10983 | p = oldp; | |
10984 | goto loopdone; | |
10985 | } | |
10986 | ||
10987 | /* Here, the final non-problematic character is earlier | |
10988 | * in the input than the penultimate character. What we do | |
10989 | * is reparse from the beginning, going up only as far as | |
10990 | * this final ok one, thus guaranteeing that the node ends | |
10991 | * in an acceptable character. The reason we reparse is | |
10992 | * that we know how far in the character is, but we don't | |
10993 | * know how to correlate its position with the input parse. | |
10994 | * An alternate implementation would be to build that | |
10995 | * correlation as we go along during the original parse, | |
10996 | * but that would entail extra work for every node, whereas | |
10997 | * this code gets executed only when the string is too | |
10998 | * large for the node, and the final two characters are | |
10999 | * problematic, an infrequent occurrence. Yet another | |
11000 | * possible strategy would be to save the tail of the | |
11001 | * string, and the next time regatom is called, initialize | |
11002 | * with that. The problem with this is that unless you | |
11003 | * back off one more character, you won't be guaranteed | |
11004 | * regatom will get called again, unless regbranch, | |
11005 | * regpiece ... are also changed. If you do back off that | |
11006 | * extra character, so that there is input guaranteed to | |
11007 | * force calling regatom, you can't handle the case where | |
11008 | * just the first character in the node is acceptable. I | |
11009 | * (khw) decided to try this method which doesn't have that | |
11010 | * pitfall; if performance issues are found, we can do a | |
11011 | * combination of the current approach plus that one */ | |
11012 | upper_parse = len; | |
11013 | len = 0; | |
11014 | s = s0; | |
11015 | goto reparse; | |
11016 | } | |
11017 | } /* End of verifying node ends with an appropriate char */ | |
11018 | ||
7e2509c1 KW |
11019 | loopdone: /* Jumped to when encounters something that shouldn't be in |
11020 | the node */ | |
34b39fc9 | 11021 | |
0ccf8511 KW |
11022 | /* I (khw) don't know if you can get here with zero length, but the |
11023 | * old code handled this situation by creating a zero-length EXACT | |
11024 | * node. Might as well be NOTHING instead */ | |
11025 | if (len == 0) { | |
11026 | OP(ret) = NOTHING; | |
11027 | } | |
11028 | else{ | |
34b39fc9 | 11029 | alloc_maybe_populate_EXACT(pRExC_state, ret, flagp, len, ender); |
0ccf8511 | 11030 | } |
34b39fc9 | 11031 | |
830247a4 | 11032 | RExC_parse = p - 1; |
fac92740 | 11033 | Set_Node_Cur_Length(ret); /* MJD */ |
830247a4 | 11034 | nextchar(pRExC_state); |
793db0cb JH |
11035 | { |
11036 | /* len is STRLEN which is unsigned, need to copy to signed */ | |
11037 | IV iv = len; | |
11038 | if (iv < 0) | |
11039 | vFAIL("Internal disaster"); | |
11040 | } | |
686b73d4 | 11041 | |
87b8b349 | 11042 | } /* End of label 'defchar:' */ |
a0d0e21e | 11043 | break; |
87b8b349 | 11044 | } /* End of giant switch on input character */ |
a687059c | 11045 | |
a0d0e21e | 11046 | return(ret); |
a687059c LW |
11047 | } |
11048 | ||
873ef191 | 11049 | STATIC char * |
bcdf7404 | 11050 | S_regwhite( RExC_state_t *pRExC_state, char *p ) |
5b5a24f7 | 11051 | { |
bcdf7404 | 11052 | const char *e = RExC_end; |
7918f24d NC |
11053 | |
11054 | PERL_ARGS_ASSERT_REGWHITE; | |
11055 | ||
5b5a24f7 CS |
11056 | while (p < e) { |
11057 | if (isSPACE(*p)) | |
11058 | ++p; | |
11059 | else if (*p == '#') { | |
bcdf7404 | 11060 | bool ended = 0; |
5b5a24f7 | 11061 | do { |
bcdf7404 YO |
11062 | if (*p++ == '\n') { |
11063 | ended = 1; | |
11064 | break; | |
11065 | } | |
11066 | } while (p < e); | |
11067 | if (!ended) | |
11068 | RExC_seen |= REG_SEEN_RUN_ON_COMMENT; | |
5b5a24f7 CS |
11069 | } |
11070 | else | |
11071 | break; | |
11072 | } | |
11073 | return p; | |
11074 | } | |
11075 | ||
b8c5462f JH |
11076 | /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]]. |
11077 | Character classes ([:foo:]) can also be negated ([:^foo:]). | |
11078 | Returns a named class id (ANYOF_XXX) if successful, -1 otherwise. | |
11079 | Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed, | |
beeb77fc | 11080 | but trigger failures because they are currently unimplemented. */ |
9a86a77b JH |
11081 | |
11082 | #define POSIXCC_DONE(c) ((c) == ':') | |
11083 | #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.') | |
11084 | #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c)) | |
11085 | ||
b8c5462f | 11086 | STATIC I32 |
830247a4 | 11087 | S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value) |
620e46c5 | 11088 | { |
97aff369 | 11089 | dVAR; |
936ed897 | 11090 | I32 namedclass = OOB_NAMEDCLASS; |
620e46c5 | 11091 | |
7918f24d NC |
11092 | PERL_ARGS_ASSERT_REGPPOSIXCC; |
11093 | ||
830247a4 | 11094 | if (value == '[' && RExC_parse + 1 < RExC_end && |
620e46c5 | 11095 | /* I smell either [: or [= or [. -- POSIX has been here, right? */ |
9a86a77b | 11096 | POSIXCC(UCHARAT(RExC_parse))) { |
1df70142 | 11097 | const char c = UCHARAT(RExC_parse); |
097eb12c | 11098 | char* const s = RExC_parse++; |
686b73d4 | 11099 | |
9a86a77b | 11100 | while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c) |
830247a4 IZ |
11101 | RExC_parse++; |
11102 | if (RExC_parse == RExC_end) | |
620e46c5 | 11103 | /* Grandfather lone [:, [=, [. */ |
830247a4 | 11104 | RExC_parse = s; |
620e46c5 | 11105 | else { |
3dab1dad | 11106 | const char* const t = RExC_parse++; /* skip over the c */ |
80916619 NC |
11107 | assert(*t == c); |
11108 | ||
9a86a77b | 11109 | if (UCHARAT(RExC_parse) == ']') { |
3dab1dad | 11110 | const char *posixcc = s + 1; |
830247a4 | 11111 | RExC_parse++; /* skip over the ending ] */ |
3dab1dad | 11112 | |
b8c5462f | 11113 | if (*s == ':') { |
1df70142 AL |
11114 | const I32 complement = *posixcc == '^' ? *posixcc++ : 0; |
11115 | const I32 skip = t - posixcc; | |
80916619 NC |
11116 | |
11117 | /* Initially switch on the length of the name. */ | |
11118 | switch (skip) { | |
11119 | case 4: | |
3dab1dad | 11120 | if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */ |
575a8a29 | 11121 | namedclass = ANYOF_ALNUM; |
cc4319de | 11122 | break; |
80916619 NC |
11123 | case 5: |
11124 | /* Names all of length 5. */ | |
11125 | /* alnum alpha ascii blank cntrl digit graph lower | |
11126 | print punct space upper */ | |
11127 | /* Offset 4 gives the best switch position. */ | |
11128 | switch (posixcc[4]) { | |
11129 | case 'a': | |
3dab1dad | 11130 | if (memEQ(posixcc, "alph", 4)) /* alpha */ |
575a8a29 | 11131 | namedclass = ANYOF_ALPHA; |
80916619 NC |
11132 | break; |
11133 | case 'e': | |
3dab1dad | 11134 | if (memEQ(posixcc, "spac", 4)) /* space */ |
575a8a29 | 11135 | namedclass = ANYOF_PSXSPC; |
80916619 NC |
11136 | break; |
11137 | case 'h': | |
3dab1dad | 11138 | if (memEQ(posixcc, "grap", 4)) /* graph */ |
575a8a29 | 11139 | namedclass = ANYOF_GRAPH; |
80916619 NC |
11140 | break; |
11141 | case 'i': | |
3dab1dad | 11142 | if (memEQ(posixcc, "asci", 4)) /* ascii */ |
575a8a29 | 11143 | namedclass = ANYOF_ASCII; |
80916619 NC |
11144 | break; |
11145 | case 'k': | |
3dab1dad | 11146 | if (memEQ(posixcc, "blan", 4)) /* blank */ |
575a8a29 | 11147 | namedclass = ANYOF_BLANK; |
80916619 NC |
11148 | break; |
11149 | case 'l': | |
3dab1dad | 11150 | if (memEQ(posixcc, "cntr", 4)) /* cntrl */ |
575a8a29 | 11151 | namedclass = ANYOF_CNTRL; |
80916619 NC |
11152 | break; |
11153 | case 'm': | |
3dab1dad | 11154 | if (memEQ(posixcc, "alnu", 4)) /* alnum */ |
575a8a29 | 11155 | namedclass = ANYOF_ALNUMC; |
80916619 NC |
11156 | break; |
11157 | case 'r': | |
3dab1dad | 11158 | if (memEQ(posixcc, "lowe", 4)) /* lower */ |
575a8a29 | 11159 | namedclass = ANYOF_LOWER; |
3dab1dad | 11160 | else if (memEQ(posixcc, "uppe", 4)) /* upper */ |
575a8a29 | 11161 | namedclass = ANYOF_UPPER; |
80916619 NC |
11162 | break; |
11163 | case 't': | |
3dab1dad | 11164 | if (memEQ(posixcc, "digi", 4)) /* digit */ |
575a8a29 | 11165 | namedclass = ANYOF_DIGIT; |
3dab1dad | 11166 | else if (memEQ(posixcc, "prin", 4)) /* print */ |
575a8a29 | 11167 | namedclass = ANYOF_PRINT; |
3dab1dad | 11168 | else if (memEQ(posixcc, "punc", 4)) /* punct */ |
575a8a29 | 11169 | namedclass = ANYOF_PUNCT; |
80916619 | 11170 | break; |
b8c5462f JH |
11171 | } |
11172 | break; | |
80916619 | 11173 | case 6: |
3dab1dad | 11174 | if (memEQ(posixcc, "xdigit", 6)) |
575a8a29 | 11175 | namedclass = ANYOF_XDIGIT; |
b8c5462f JH |
11176 | break; |
11177 | } | |
80916619 NC |
11178 | |
11179 | if (namedclass == OOB_NAMEDCLASS) | |
b45f050a JF |
11180 | Simple_vFAIL3("POSIX class [:%.*s:] unknown", |
11181 | t - s - 1, s + 1); | |
575a8a29 KW |
11182 | |
11183 | /* The #defines are structured so each complement is +1 to | |
11184 | * the normal one */ | |
11185 | if (complement) { | |
11186 | namedclass++; | |
11187 | } | |
80916619 NC |
11188 | assert (posixcc[skip] == ':'); |
11189 | assert (posixcc[skip+1] == ']'); | |
b45f050a | 11190 | } else if (!SIZE_ONLY) { |
b8c5462f | 11191 | /* [[=foo=]] and [[.foo.]] are still future. */ |
b45f050a | 11192 | |
830247a4 | 11193 | /* adjust RExC_parse so the warning shows after |
b45f050a | 11194 | the class closes */ |
9a86a77b | 11195 | while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']') |
830247a4 | 11196 | RExC_parse++; |
b45f050a JF |
11197 | Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c); |
11198 | } | |
b8c5462f JH |
11199 | } else { |
11200 | /* Maternal grandfather: | |
11201 | * "[:" ending in ":" but not in ":]" */ | |
830247a4 | 11202 | RExC_parse = s; |
767d463e | 11203 | } |
620e46c5 JH |
11204 | } |
11205 | } | |
11206 | ||
b8c5462f JH |
11207 | return namedclass; |
11208 | } | |
11209 | ||
11210 | STATIC void | |
830247a4 | 11211 | S_checkposixcc(pTHX_ RExC_state_t *pRExC_state) |
b8c5462f | 11212 | { |
97aff369 | 11213 | dVAR; |
7918f24d NC |
11214 | |
11215 | PERL_ARGS_ASSERT_CHECKPOSIXCC; | |
11216 | ||
3dab1dad | 11217 | if (POSIXCC(UCHARAT(RExC_parse))) { |
1df70142 AL |
11218 | const char *s = RExC_parse; |
11219 | const char c = *s++; | |
b8c5462f | 11220 | |
3dab1dad | 11221 | while (isALNUM(*s)) |
b8c5462f JH |
11222 | s++; |
11223 | if (*s && c == *s && s[1] == ']') { | |
668c081a NC |
11224 | ckWARN3reg(s+2, |
11225 | "POSIX syntax [%c %c] belongs inside character classes", | |
11226 | c, c); | |
b45f050a JF |
11227 | |
11228 | /* [[=foo=]] and [[.foo.]] are still future. */ | |
9a86a77b | 11229 | if (POSIXCC_NOTYET(c)) { |
830247a4 | 11230 | /* adjust RExC_parse so the error shows after |
b45f050a | 11231 | the class closes */ |
9a86a77b | 11232 | while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']') |
3dab1dad | 11233 | NOOP; |
b45f050a JF |
11234 | Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c); |
11235 | } | |
b8c5462f JH |
11236 | } |
11237 | } | |
620e46c5 JH |
11238 | } |
11239 | ||
ea317ccb KW |
11240 | /* Generate the code to add a full posix character <class> to the bracketed |
11241 | * character class given by <node>. (<node> is needed only under locale rules) | |
11242 | * destlist is the inversion list for non-locale rules that this class is | |
11243 | * to be added to | |
11244 | * sourcelist is the ASCII-range inversion list to add under /a rules | |
11245 | * Xsourcelist is the full Unicode range list to use otherwise. */ | |
11246 | #define DO_POSIX(node, class, destlist, sourcelist, Xsourcelist) \ | |
11247 | if (LOC) { \ | |
11248 | SV* scratch_list = NULL; \ | |
11249 | \ | |
11250 | /* Set this class in the node for runtime matching */ \ | |
11251 | ANYOF_CLASS_SET(node, class); \ | |
11252 | \ | |
11253 | /* For above Latin1 code points, we use the full Unicode range */ \ | |
11254 | _invlist_intersection(PL_AboveLatin1, \ | |
11255 | Xsourcelist, \ | |
11256 | &scratch_list); \ | |
11257 | /* And set the output to it, adding instead if there already is an \ | |
11258 | * output. Checking if <destlist> is NULL first saves an extra \ | |
11259 | * clone. Its reference count will be decremented at the next \ | |
11260 | * union, etc, or if this is the only instance, at the end of the \ | |
11261 | * routine */ \ | |
11262 | if (! destlist) { \ | |
11263 | destlist = scratch_list; \ | |
11264 | } \ | |
11265 | else { \ | |
11266 | _invlist_union(destlist, scratch_list, &destlist); \ | |
11267 | SvREFCNT_dec(scratch_list); \ | |
11268 | } \ | |
11269 | } \ | |
11270 | else { \ | |
11271 | /* For non-locale, just add it to any existing list */ \ | |
11272 | _invlist_union(destlist, \ | |
11273 | (AT_LEAST_ASCII_RESTRICTED) \ | |
11274 | ? sourcelist \ | |
11275 | : Xsourcelist, \ | |
11276 | &destlist); \ | |
11277 | } | |
11278 | ||
11279 | /* Like DO_POSIX, but matches the complement of <sourcelist> and <Xsourcelist>. | |
11280 | */ | |
11281 | #define DO_N_POSIX(node, class, destlist, sourcelist, Xsourcelist) \ | |
11282 | if (LOC) { \ | |
11283 | SV* scratch_list = NULL; \ | |
11284 | ANYOF_CLASS_SET(node, class); \ | |
11285 | _invlist_subtract(PL_AboveLatin1, Xsourcelist, &scratch_list); \ | |
11286 | if (! destlist) { \ | |
11287 | destlist = scratch_list; \ | |
11288 | } \ | |
11289 | else { \ | |
11290 | _invlist_union(destlist, scratch_list, &destlist); \ | |
11291 | SvREFCNT_dec(scratch_list); \ | |
11292 | } \ | |
11293 | } \ | |
11294 | else { \ | |
11295 | _invlist_union_complement_2nd(destlist, \ | |
11296 | (AT_LEAST_ASCII_RESTRICTED) \ | |
11297 | ? sourcelist \ | |
11298 | : Xsourcelist, \ | |
11299 | &destlist); \ | |
11300 | /* Under /d, everything in the upper half of the Latin1 range \ | |
11301 | * matches this complement */ \ | |
11302 | if (DEPENDS_SEMANTICS) { \ | |
11303 | ANYOF_FLAGS(node) |= ANYOF_NON_UTF8_LATIN1_ALL; \ | |
11304 | } \ | |
11305 | } | |
11306 | ||
11307 | /* Generate the code to add a posix character <class> to the bracketed | |
11308 | * character class given by <node>. (<node> is needed only under locale rules) | |
11309 | * destlist is the inversion list for non-locale rules that this class is | |
11310 | * to be added to | |
11311 | * sourcelist is the ASCII-range inversion list to add under /a rules | |
11312 | * l1_sourcelist is the Latin1 range list to use otherwise. | |
11313 | * Xpropertyname is the name to add to <run_time_list> of the property to | |
11314 | * specify the code points above Latin1 that will have to be | |
11315 | * determined at run-time | |
11316 | * run_time_list is a SV* that contains text names of properties that are to | |
11317 | * be computed at run time. This concatenates <Xpropertyname> | |
18890cbd | 11318 | * to it, appropriately |
ea317ccb KW |
11319 | * This is essentially DO_POSIX, but we know only the Latin1 values at compile |
11320 | * time */ | |
11321 | #define DO_POSIX_LATIN1_ONLY_KNOWN(node, class, destlist, sourcelist, \ | |
11322 | l1_sourcelist, Xpropertyname, run_time_list) \ | |
4f6289a3 RB |
11323 | /* First, resolve whether to use the ASCII-only list or the L1 \ |
11324 | * list */ \ | |
11325 | DO_POSIX_LATIN1_ONLY_KNOWN_L1_RESOLVED(node, class, destlist, \ | |
11326 | ((AT_LEAST_ASCII_RESTRICTED) ? sourcelist : l1_sourcelist),\ | |
11327 | Xpropertyname, run_time_list) | |
11328 | ||
11329 | #define DO_POSIX_LATIN1_ONLY_KNOWN_L1_RESOLVED(node, class, destlist, sourcelist, \ | |
11330 | Xpropertyname, run_time_list) \ | |
ea317ccb KW |
11331 | /* If not /a matching, there are going to be code points we will have \ |
11332 | * to defer to runtime to look-up */ \ | |
11333 | if (! AT_LEAST_ASCII_RESTRICTED) { \ | |
11334 | Perl_sv_catpvf(aTHX_ run_time_list, "+utf8::%s\n", Xpropertyname); \ | |
11335 | } \ | |
11336 | if (LOC) { \ | |
11337 | ANYOF_CLASS_SET(node, class); \ | |
11338 | } \ | |
11339 | else { \ | |
4f6289a3 | 11340 | _invlist_union(destlist, sourcelist, &destlist); \ |
ea317ccb KW |
11341 | } |
11342 | ||
11343 | /* Like DO_POSIX_LATIN1_ONLY_KNOWN, but for the complement. A combination of | |
5073ffbd KW |
11344 | * this and DO_N_POSIX. Sets <matches_above_unicode> only if it can; unchanged |
11345 | * otherwise */ | |
ea317ccb | 11346 | #define DO_N_POSIX_LATIN1_ONLY_KNOWN(node, class, destlist, sourcelist, \ |
5073ffbd | 11347 | l1_sourcelist, Xpropertyname, run_time_list, matches_above_unicode) \ |
ea317ccb KW |
11348 | if (AT_LEAST_ASCII_RESTRICTED) { \ |
11349 | _invlist_union_complement_2nd(destlist, sourcelist, &destlist); \ | |
11350 | } \ | |
11351 | else { \ | |
11352 | Perl_sv_catpvf(aTHX_ run_time_list, "!utf8::%s\n", Xpropertyname); \ | |
5073ffbd | 11353 | matches_above_unicode = TRUE; \ |
ea317ccb | 11354 | if (LOC) { \ |
5073ffbd | 11355 | ANYOF_CLASS_SET(node, namedclass); \ |
ea317ccb KW |
11356 | } \ |
11357 | else { \ | |
11358 | SV* scratch_list = NULL; \ | |
11359 | _invlist_subtract(PL_Latin1, l1_sourcelist, &scratch_list); \ | |
11360 | if (! destlist) { \ | |
11361 | destlist = scratch_list; \ | |
11362 | } \ | |
11363 | else { \ | |
11364 | _invlist_union(destlist, scratch_list, &destlist); \ | |
11365 | SvREFCNT_dec(scratch_list); \ | |
11366 | } \ | |
11367 | if (DEPENDS_SEMANTICS) { \ | |
11368 | ANYOF_FLAGS(node) |= ANYOF_NON_UTF8_LATIN1_ALL; \ | |
11369 | } \ | |
11370 | } \ | |
11371 | } | |
a12cf05f | 11372 | |
c8453963 KW |
11373 | STATIC void |
11374 | S_add_alternate(pTHX_ AV** alternate_ptr, U8* string, STRLEN len) | |
11375 | { | |
11376 | /* Adds input 'string' with length 'len' to the ANYOF node's unicode | |
11377 | * alternate list, pointed to by 'alternate_ptr'. This is an array of | |
11378 | * the multi-character folds of characters in the node */ | |
11379 | SV *sv; | |
11380 | ||
11381 | PERL_ARGS_ASSERT_ADD_ALTERNATE; | |
11382 | ||
11383 | if (! *alternate_ptr) { | |
11384 | *alternate_ptr = newAV(); | |
11385 | } | |
11386 | sv = newSVpvn_utf8((char*)string, len, TRUE); | |
11387 | av_push(*alternate_ptr, sv); | |
11388 | return; | |
11389 | } | |
11390 | ||
66c69124 KW |
11391 | /* The names of properties whose definitions are not known at compile time are |
11392 | * stored in this SV, after a constant heading. So if the length has been | |
11393 | * changed since initialization, then there is a run-time definition. */ | |
11394 | #define HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION (SvCUR(listsv) != initial_listsv_len) | |
11395 | ||
0658cdde KW |
11396 | /* This converts the named class defined in regcomp.h to its equivalent class |
11397 | * number defined in handy.h. */ | |
11398 | #define namedclass_to_classnum(class) ((class) / 2) | |
11399 | ||
7f6f358c YO |
11400 | /* |
11401 | parse a class specification and produce either an ANYOF node that | |
ddad5e0b | 11402 | matches the pattern or perhaps will be optimized into an EXACTish node |
679d1424 KW |
11403 | instead. The node contains a bit map for the first 256 characters, with the |
11404 | corresponding bit set if that character is in the list. For characters | |
11405 | above 255, a range list is used */ | |
89836f1f | 11406 | |
76e3520e | 11407 | STATIC regnode * |
34b39fc9 | 11408 | S_regclass(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth) |
a687059c | 11409 | { |
97aff369 | 11410 | dVAR; |
eb578fdb KW |
11411 | UV nextvalue; |
11412 | UV prevvalue = OOB_UNICODE; | |
11413 | IV range = 0; | |
4276fd36 | 11414 | UV value = 0; |
eb578fdb | 11415 | regnode *ret; |
ba210ebe | 11416 | STRLEN numlen; |
8b27d3db | 11417 | IV namedclass = OOB_NAMEDCLASS; |
cbbf8932 | 11418 | char *rangebegin = NULL; |
936ed897 | 11419 | bool need_class = 0; |
827f5bb8 | 11420 | bool allow_full_fold = TRUE; /* Assume wants multi-char folding */ |
c445ea15 | 11421 | SV *listsv = NULL; |
137165a6 KW |
11422 | STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more |
11423 | than just initialized. */ | |
570fcd08 KW |
11424 | SV* properties = NULL; /* Code points that match \p{} \P{} */ |
11425 | SV* posixes = NULL; /* Code points that match classes like, [:word:], | |
11426 | extended beyond the Latin1 range */ | |
dc415b4a KW |
11427 | UV element_count = 0; /* Number of distinct elements in the class. |
11428 | Optimizations may be possible if this is tiny */ | |
ffc61ed2 | 11429 | UV n; |
53742956 | 11430 | |
dc415b4a | 11431 | /* Unicode properties are stored in a swash; this holds the current one |
88d45d28 KW |
11432 | * being parsed. If this swash is the only above-latin1 component of the |
11433 | * character class, an optimization is to pass it directly on to the | |
11434 | * execution engine. Otherwise, it is set to NULL to indicate that there | |
11435 | * are other things in the class that have to be dealt with at execution | |
11436 | * time */ | |
dc415b4a KW |
11437 | SV* swash = NULL; /* Code points that match \p{} \P{} */ |
11438 | ||
11439 | /* Set if a component of this character class is user-defined; just passed | |
11440 | * on to the engine */ | |
7335417f | 11441 | bool has_user_defined_property = FALSE; |
dc415b4a | 11442 | |
68823f48 KW |
11443 | /* inversion list of code points this node matches only when the target |
11444 | * string is in UTF-8. (Because is under /d) */ | |
11445 | SV* depends_list = NULL; | |
11446 | ||
cfbb2758 KW |
11447 | /* inversion list of code points this node matches. For much of the |
11448 | * function, it includes only those that match regardless of the utf8ness | |
11449 | * of the target string */ | |
11450 | SV* cp_list = NULL; | |
11451 | ||
53742956 | 11452 | /* List of multi-character folds that are matched by this node */ |
cbbf8932 | 11453 | AV* unicode_alternate = NULL; |
1b2d223b | 11454 | #ifdef EBCDIC |
8f850557 KW |
11455 | /* In a range, counts how many 0-2 of the ends of it came from literals, |
11456 | * not escapes. Thus we can tell if 'A' was input vs \x{C1} */ | |
1b2d223b JH |
11457 | UV literal_endpoint = 0; |
11458 | #endif | |
520939ba | 11459 | bool invert = FALSE; /* Is this class to be complemented */ |
ffc61ed2 | 11460 | |
5073ffbd KW |
11461 | /* Is there any thing like \W or [:^digit:] that matches above the legal |
11462 | * Unicode range? */ | |
11463 | bool runtime_posix_matches_above_Unicode = FALSE; | |
11464 | ||
3dab1dad | 11465 | regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in |
7f6f358c | 11466 | case we need to change the emitted regop to an EXACT. */ |
07be1b83 | 11467 | const char * orig_parse = RExC_parse; |
6197c735 | 11468 | const I32 orig_size = RExC_size; |
72f13be8 | 11469 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
11470 | |
11471 | PERL_ARGS_ASSERT_REGCLASS; | |
76e84362 SH |
11472 | #ifndef DEBUGGING |
11473 | PERL_UNUSED_ARG(depth); | |
11474 | #endif | |
72f13be8 | 11475 | |
3dab1dad | 11476 | DEBUG_PARSE("clas"); |
7f6f358c YO |
11477 | |
11478 | /* Assume we are going to generate an ANYOF node. */ | |
ffc61ed2 JH |
11479 | ret = reganode(pRExC_state, ANYOF, 0); |
11480 | ||
56ca34ca KW |
11481 | |
11482 | if (!SIZE_ONLY) { | |
ffc61ed2 | 11483 | ANYOF_FLAGS(ret) = 0; |
56ca34ca | 11484 | } |
ffc61ed2 | 11485 | |
9a86a77b | 11486 | if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */ |
ffc61ed2 JH |
11487 | RExC_naughty++; |
11488 | RExC_parse++; | |
520939ba | 11489 | invert = TRUE; |
827f5bb8 KW |
11490 | |
11491 | /* We have decided to not allow multi-char folds in inverted character | |
ac455f4c KW |
11492 | * classes, due to the confusion that can happen, especially with |
11493 | * classes that are designed for a non-Unicode world: You have the | |
11494 | * peculiar case that: | |
827f5bb8 KW |
11495 | "s s" =~ /^[^\xDF]+$/i => Y |
11496 | "ss" =~ /^[^\xDF]+$/i => N | |
11497 | * | |
11498 | * See [perl #89750] */ | |
11499 | allow_full_fold = FALSE; | |
ffc61ed2 | 11500 | } |
a0d0e21e | 11501 | |
73060fc4 | 11502 | if (SIZE_ONLY) { |
830247a4 | 11503 | RExC_size += ANYOF_SKIP; |
73060fc4 JH |
11504 | listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */ |
11505 | } | |
936ed897 | 11506 | else { |
830247a4 | 11507 | RExC_emit += ANYOF_SKIP; |
3a15e693 | 11508 | if (LOC) { |
936ed897 | 11509 | ANYOF_FLAGS(ret) |= ANYOF_LOCALE; |
3a15e693 | 11510 | } |
396482e1 | 11511 | listsv = newSVpvs("# comment\n"); |
137165a6 | 11512 | initial_listsv_len = SvCUR(listsv); |
a0d0e21e | 11513 | } |
b8c5462f | 11514 | |
9a86a77b JH |
11515 | nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0; |
11516 | ||
b938889d | 11517 | if (!SIZE_ONLY && POSIXCC(nextvalue)) |
830247a4 | 11518 | checkposixcc(pRExC_state); |
b8c5462f | 11519 | |
f064b6ad HS |
11520 | /* allow 1st char to be ] (allowing it to be - is dealt with later) */ |
11521 | if (UCHARAT(RExC_parse) == ']') | |
11522 | goto charclassloop; | |
ffc61ed2 | 11523 | |
fc8cd66c | 11524 | parseit: |
9a86a77b | 11525 | while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') { |
ffc61ed2 JH |
11526 | |
11527 | charclassloop: | |
11528 | ||
11529 | namedclass = OOB_NAMEDCLASS; /* initialize as illegal */ | |
11530 | ||
dc415b4a | 11531 | if (!range) { |
830247a4 | 11532 | rangebegin = RExC_parse; |
dc415b4a KW |
11533 | element_count++; |
11534 | } | |
ffc61ed2 | 11535 | if (UTF) { |
5e12f4fb | 11536 | value = utf8n_to_uvchr((U8*)RExC_parse, |
3568d838 | 11537 | RExC_end - RExC_parse, |
9f7f3913 | 11538 | &numlen, UTF8_ALLOW_DEFAULT); |
ffc61ed2 JH |
11539 | RExC_parse += numlen; |
11540 | } | |
11541 | else | |
11542 | value = UCHARAT(RExC_parse++); | |
7f6f358c | 11543 | |
9a86a77b JH |
11544 | nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0; |
11545 | if (value == '[' && POSIXCC(nextvalue)) | |
830247a4 | 11546 | namedclass = regpposixcc(pRExC_state, value); |
620e46c5 | 11547 | else if (value == '\\') { |
ffc61ed2 | 11548 | if (UTF) { |
5e12f4fb | 11549 | value = utf8n_to_uvchr((U8*)RExC_parse, |
ffc61ed2 | 11550 | RExC_end - RExC_parse, |
9f7f3913 | 11551 | &numlen, UTF8_ALLOW_DEFAULT); |
ffc61ed2 JH |
11552 | RExC_parse += numlen; |
11553 | } | |
11554 | else | |
11555 | value = UCHARAT(RExC_parse++); | |
470c3474 | 11556 | /* Some compilers cannot handle switching on 64-bit integer |
ffc61ed2 | 11557 | * values, therefore value cannot be an UV. Yes, this will |
e2962f66 JH |
11558 | * be a problem later if we want switch on Unicode. |
11559 | * A similar issue a little bit later when switching on | |
11560 | * namedclass. --jhi */ | |
ffc61ed2 | 11561 | switch ((I32)value) { |
b8c5462f JH |
11562 | case 'w': namedclass = ANYOF_ALNUM; break; |
11563 | case 'W': namedclass = ANYOF_NALNUM; break; | |
11564 | case 's': namedclass = ANYOF_SPACE; break; | |
11565 | case 'S': namedclass = ANYOF_NSPACE; break; | |
11566 | case 'd': namedclass = ANYOF_DIGIT; break; | |
11567 | case 'D': namedclass = ANYOF_NDIGIT; break; | |
e1d1eefb YO |
11568 | case 'v': namedclass = ANYOF_VERTWS; break; |
11569 | case 'V': namedclass = ANYOF_NVERTWS; break; | |
11570 | case 'h': namedclass = ANYOF_HORIZWS; break; | |
11571 | case 'H': namedclass = ANYOF_NHORIZWS; break; | |
fc8cd66c YO |
11572 | case 'N': /* Handle \N{NAME} in class */ |
11573 | { | |
11574 | /* We only pay attention to the first char of | |
11575 | multichar strings being returned. I kinda wonder | |
11576 | if this makes sense as it does change the behaviour | |
11577 | from earlier versions, OTOH that behaviour was broken | |
11578 | as well. */ | |
34b39fc9 | 11579 | if (! grok_bslash_N(pRExC_state, NULL, &value, flagp, depth, |
b9e8c997 KW |
11580 | TRUE /* => charclass */)) |
11581 | { | |
fc8cd66c YO |
11582 | goto parseit; |
11583 | } | |
fc8cd66c YO |
11584 | } |
11585 | break; | |
ffc61ed2 JH |
11586 | case 'p': |
11587 | case 'P': | |
3dab1dad YO |
11588 | { |
11589 | char *e; | |
5d3d13d1 KW |
11590 | |
11591 | /* This routine will handle any undefined properties */ | |
11592 | U8 swash_init_flags = _CORE_SWASH_INIT_RETURN_IF_UNDEF; | |
83199d38 | 11593 | |
af6f566e | 11594 | if (RExC_parse >= RExC_end) |
2a4859cd | 11595 | vFAIL2("Empty \\%c{}", (U8)value); |
ffc61ed2 | 11596 | if (*RExC_parse == '{') { |
1df70142 | 11597 | const U8 c = (U8)value; |
ffc61ed2 JH |
11598 | e = strchr(RExC_parse++, '}'); |
11599 | if (!e) | |
0da60cf5 | 11600 | vFAIL2("Missing right brace on \\%c{}", c); |
ab13f0c7 JH |
11601 | while (isSPACE(UCHARAT(RExC_parse))) |
11602 | RExC_parse++; | |
11603 | if (e == RExC_parse) | |
0da60cf5 | 11604 | vFAIL2("Empty \\%c{}", c); |
ffc61ed2 | 11605 | n = e - RExC_parse; |
ab13f0c7 JH |
11606 | while (isSPACE(UCHARAT(RExC_parse + n - 1))) |
11607 | n--; | |
ffc61ed2 JH |
11608 | } |
11609 | else { | |
11610 | e = RExC_parse; | |
11611 | n = 1; | |
11612 | } | |
ee410026 | 11613 | if (!SIZE_ONLY) { |
dc415b4a KW |
11614 | SV* invlist; |
11615 | char* name; | |
5073ffbd | 11616 | |
ab13f0c7 JH |
11617 | if (UCHARAT(RExC_parse) == '^') { |
11618 | RExC_parse++; | |
11619 | n--; | |
11620 | value = value == 'p' ? 'P' : 'p'; /* toggle */ | |
11621 | while (isSPACE(UCHARAT(RExC_parse))) { | |
11622 | RExC_parse++; | |
11623 | n--; | |
11624 | } | |
11625 | } | |
dc415b4a KW |
11626 | /* Try to get the definition of the property into |
11627 | * <invlist>. If /i is in effect, the effective property | |
11628 | * will have its name be <__NAME_i>. The design is | |
11629 | * discussed in commit | |
11630 | * 2f833f5208e26b208886e51e09e2c072b5eabb46 */ | |
11631 | Newx(name, n + sizeof("_i__\n"), char); | |
11632 | ||
11633 | sprintf(name, "%s%.*s%s\n", | |
11634 | (FOLD) ? "__" : "", | |
11635 | (int)n, | |
11636 | RExC_parse, | |
11637 | (FOLD) ? "_i" : "" | |
11638 | ); | |
11639 | ||
11640 | /* Look up the property name, and get its swash and | |
11641 | * inversion list, if the property is found */ | |
dc415b4a KW |
11642 | if (swash) { |
11643 | SvREFCNT_dec(swash); | |
11644 | } | |
11645 | swash = _core_swash_init("utf8", name, &PL_sv_undef, | |
11646 | 1, /* binary */ | |
11647 | 0, /* not tr/// */ | |
83199d38 KW |
11648 | NULL, /* No inversion list */ |
11649 | &swash_init_flags | |
dc415b4a | 11650 | ); |
872dd7e0 | 11651 | if (! swash || ! (invlist = _get_swash_invlist(swash))) { |
dc415b4a KW |
11652 | if (swash) { |
11653 | SvREFCNT_dec(swash); | |
11654 | swash = NULL; | |
11655 | } | |
11656 | ||
11657 | /* Here didn't find it. It could be a user-defined | |
11658 | * property that will be available at run-time. Add it | |
11659 | * to the list to look up then */ | |
11660 | Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s\n", | |
11661 | (value == 'p' ? '+' : '!'), | |
11662 | name); | |
7335417f | 11663 | has_user_defined_property = TRUE; |
dc415b4a KW |
11664 | |
11665 | /* We don't know yet, so have to assume that the | |
11666 | * property could match something in the Latin1 range, | |
18890cbd KW |
11667 | * hence something that isn't utf8. Note that this |
11668 | * would cause things in <depends_list> to match | |
11669 | * inappropriately, except that any \p{}, including | |
11670 | * this one forces Unicode semantics, which means there | |
11671 | * is <no depends_list> */ | |
dc415b4a KW |
11672 | ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP_NON_UTF8; |
11673 | } | |
11674 | else { | |
11675 | ||
11676 | /* Here, did get the swash and its inversion list. If | |
11677 | * the swash is from a user-defined property, then this | |
11678 | * whole character class should be regarded as such */ | |
afb790dd | 11679 | has_user_defined_property = |
83199d38 KW |
11680 | (swash_init_flags |
11681 | & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY); | |
dc415b4a KW |
11682 | |
11683 | /* Invert if asking for the complement */ | |
11684 | if (value == 'P') { | |
26cdf4c4 KW |
11685 | _invlist_union_complement_2nd(properties, |
11686 | invlist, | |
11687 | &properties); | |
2f833f52 | 11688 | |
dc415b4a KW |
11689 | /* The swash can't be used as-is, because we've |
11690 | * inverted things; delay removing it to here after | |
11691 | * have copied its invlist above */ | |
11692 | SvREFCNT_dec(swash); | |
11693 | swash = NULL; | |
11694 | } | |
11695 | else { | |
112b0fc6 | 11696 | _invlist_union(properties, invlist, &properties); |
dc415b4a KW |
11697 | } |
11698 | } | |
11699 | Safefree(name); | |
ffc61ed2 JH |
11700 | } |
11701 | RExC_parse = e + 1; | |
f81125e2 | 11702 | namedclass = ANYOF_MAX; /* no official name, but it's named */ |
e40e74fe KW |
11703 | |
11704 | /* \p means they want Unicode semantics */ | |
11705 | RExC_uni_semantics = 1; | |
3dab1dad | 11706 | } |
f81125e2 | 11707 | break; |
b8c5462f JH |
11708 | case 'n': value = '\n'; break; |
11709 | case 'r': value = '\r'; break; | |
11710 | case 't': value = '\t'; break; | |
11711 | case 'f': value = '\f'; break; | |
11712 | case 'b': value = '\b'; break; | |
c7f1f016 NIS |
11713 | case 'e': value = ASCII_TO_NATIVE('\033');break; |
11714 | case 'a': value = ASCII_TO_NATIVE('\007');break; | |
f0a2b745 KW |
11715 | case 'o': |
11716 | RExC_parse--; /* function expects to be pointed at the 'o' */ | |
454155d9 KW |
11717 | { |
11718 | const char* error_msg; | |
11719 | bool valid = grok_bslash_o(RExC_parse, | |
f0a2b745 KW |
11720 | &value, |
11721 | &numlen, | |
454155d9 KW |
11722 | &error_msg, |
11723 | SIZE_ONLY); | |
11724 | RExC_parse += numlen; | |
11725 | if (! valid) { | |
11726 | vFAIL(error_msg); | |
11727 | } | |
f0a2b745 | 11728 | } |
f0a2b745 KW |
11729 | if (PL_encoding && value < 0x100) { |
11730 | goto recode_encoding; | |
11731 | } | |
11732 | break; | |
b8c5462f | 11733 | case 'x': |
a0481293 KW |
11734 | RExC_parse--; /* function expects to be pointed at the 'x' */ |
11735 | { | |
11736 | const char* error_msg; | |
11737 | bool valid = grok_bslash_x(RExC_parse, | |
11738 | &value, | |
11739 | &numlen, | |
11740 | &error_msg, | |
11741 | 1); | |
ffc61ed2 | 11742 | RExC_parse += numlen; |
a0481293 KW |
11743 | if (! valid) { |
11744 | vFAIL(error_msg); | |
11745 | } | |
ffc61ed2 | 11746 | } |
9e08bc66 TS |
11747 | if (PL_encoding && value < 0x100) |
11748 | goto recode_encoding; | |
b8c5462f JH |
11749 | break; |
11750 | case 'c': | |
17a3df4c | 11751 | value = grok_bslash_c(*RExC_parse++, UTF, SIZE_ONLY); |
b8c5462f JH |
11752 | break; |
11753 | case '0': case '1': case '2': case '3': case '4': | |
c99e91e9 | 11754 | case '5': case '6': case '7': |
9e08bc66 | 11755 | { |
c99e91e9 KW |
11756 | /* Take 1-3 octal digits */ |
11757 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT; | |
9e08bc66 TS |
11758 | numlen = 3; |
11759 | value = grok_oct(--RExC_parse, &numlen, &flags, NULL); | |
11760 | RExC_parse += numlen; | |
11761 | if (PL_encoding && value < 0x100) | |
11762 | goto recode_encoding; | |
11763 | break; | |
11764 | } | |
11765 | recode_encoding: | |
e2a7e165 | 11766 | if (! RExC_override_recoding) { |
9e08bc66 TS |
11767 | SV* enc = PL_encoding; |
11768 | value = reg_recode((const char)(U8)value, &enc); | |
668c081a NC |
11769 | if (!enc && SIZE_ONLY) |
11770 | ckWARNreg(RExC_parse, | |
11771 | "Invalid escape in the specified encoding"); | |
9e08bc66 TS |
11772 | break; |
11773 | } | |
1028017a | 11774 | default: |
c99e91e9 KW |
11775 | /* Allow \_ to not give an error */ |
11776 | if (!SIZE_ONLY && isALNUM(value) && value != '_') { | |
668c081a NC |
11777 | ckWARN2reg(RExC_parse, |
11778 | "Unrecognized escape \\%c in character class passed through", | |
11779 | (int)value); | |
c99e91e9 | 11780 | } |
1028017a | 11781 | break; |
b8c5462f | 11782 | } |
ffc61ed2 | 11783 | } /* end of \blah */ |
1b2d223b JH |
11784 | #ifdef EBCDIC |
11785 | else | |
11786 | literal_endpoint++; | |
11787 | #endif | |
ffc61ed2 | 11788 | |
8b27d3db KW |
11789 | /* What matches in a locale is not known until runtime. This |
11790 | * includes what the Posix classes (like \w, [:space:]) match. | |
11791 | * Room must be reserved (one time per class) to store such | |
11792 | * classes, either if Perl is compiled so that locale nodes always | |
11793 | * should have this space, or if there is such class info to be | |
11794 | * stored. The space will contain a bit for each named class that | |
11795 | * is to be matched against. This isn't needed for \p{} and | |
11796 | * pseudo-classes, as they are not affected by locale, and hence | |
11797 | * are dealt with separately */ | |
11798 | if (LOC | |
11799 | && ! need_class | |
11800 | && (ANYOF_LOCALE == ANYOF_CLASS | |
11801 | || (namedclass > OOB_NAMEDCLASS && namedclass < ANYOF_MAX))) | |
11802 | { | |
2c63ecad KW |
11803 | need_class = 1; |
11804 | if (SIZE_ONLY) { | |
dd58aee1 | 11805 | RExC_size += ANYOF_CLASS_SKIP - ANYOF_SKIP; |
2c63ecad KW |
11806 | } |
11807 | else { | |
dd58aee1 | 11808 | RExC_emit += ANYOF_CLASS_SKIP - ANYOF_SKIP; |
2c63ecad KW |
11809 | ANYOF_CLASS_ZERO(ret); |
11810 | } | |
9051cfd9 | 11811 | ANYOF_FLAGS(ret) |= ANYOF_CLASS; |
2c63ecad | 11812 | } |
ffc61ed2 | 11813 | |
8b27d3db KW |
11814 | if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */ |
11815 | ||
d5788240 | 11816 | /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a |
1d791ab2 KW |
11817 | * literal, as is the character that began the false range, i.e. |
11818 | * the 'a' in the examples */ | |
ffc61ed2 | 11819 | if (range) { |
73b437c8 | 11820 | if (!SIZE_ONLY) { |
668c081a NC |
11821 | const int w = |
11822 | RExC_parse >= rangebegin ? | |
11823 | RExC_parse - rangebegin : 0; | |
11824 | ckWARN4reg(RExC_parse, | |
b45f050a | 11825 | "False [] range \"%*.*s\"", |
097eb12c | 11826 | w, w, rangebegin); |
cfbb2758 KW |
11827 | cp_list = add_cp_to_invlist(cp_list, '-'); |
11828 | cp_list = add_cp_to_invlist(cp_list, prevvalue); | |
b8c5462f | 11829 | } |
ffc61ed2 JH |
11830 | |
11831 | range = 0; /* this was not a true range */ | |
3172e3fd | 11832 | element_count += 2; /* So counts for three values */ |
73b437c8 | 11833 | } |
ffc61ed2 | 11834 | |
c9f92f3e | 11835 | if (! SIZE_ONLY) { |
e2962f66 | 11836 | switch ((I32)namedclass) { |
ea317ccb KW |
11837 | |
11838 | case ANYOF_ALNUMC: /* C's alnum, in contrast to \w */ | |
570fcd08 | 11839 | DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
ea317ccb KW |
11840 | PL_PosixAlnum, PL_L1PosixAlnum, "XPosixAlnum", listsv); |
11841 | break; | |
11842 | case ANYOF_NALNUMC: | |
570fcd08 | 11843 | DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
5073ffbd KW |
11844 | PL_PosixAlnum, PL_L1PosixAlnum, "XPosixAlnum", listsv, |
11845 | runtime_posix_matches_above_Unicode); | |
ea317ccb KW |
11846 | break; |
11847 | case ANYOF_ALPHA: | |
570fcd08 | 11848 | DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
ea317ccb KW |
11849 | PL_PosixAlpha, PL_L1PosixAlpha, "XPosixAlpha", listsv); |
11850 | break; | |
11851 | case ANYOF_NALPHA: | |
570fcd08 | 11852 | DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
5073ffbd KW |
11853 | PL_PosixAlpha, PL_L1PosixAlpha, "XPosixAlpha", listsv, |
11854 | runtime_posix_matches_above_Unicode); | |
ea317ccb | 11855 | break; |
73b437c8 | 11856 | case ANYOF_ASCII: |
ea317ccb KW |
11857 | if (LOC) { |
11858 | ANYOF_CLASS_SET(ret, namedclass); | |
73b437c8 | 11859 | } |
ea317ccb | 11860 | else { |
570fcd08 | 11861 | _invlist_union(posixes, PL_ASCII, &posixes); |
ea317ccb | 11862 | } |
73b437c8 JH |
11863 | break; |
11864 | case ANYOF_NASCII: | |
ea317ccb KW |
11865 | if (LOC) { |
11866 | ANYOF_CLASS_SET(ret, namedclass); | |
73b437c8 | 11867 | } |
ea317ccb | 11868 | else { |
570fcd08 KW |
11869 | _invlist_union_complement_2nd(posixes, |
11870 | PL_ASCII, &posixes); | |
ea317ccb KW |
11871 | if (DEPENDS_SEMANTICS) { |
11872 | ANYOF_FLAGS(ret) |= ANYOF_NON_UTF8_LATIN1_ALL; | |
11873 | } | |
11874 | } | |
11875 | break; | |
11876 | case ANYOF_BLANK: | |
570fcd08 | 11877 | DO_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
11878 | PL_PosixBlank, PL_XPosixBlank); |
11879 | break; | |
11880 | case ANYOF_NBLANK: | |
570fcd08 | 11881 | DO_N_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
11882 | PL_PosixBlank, PL_XPosixBlank); |
11883 | break; | |
11884 | case ANYOF_CNTRL: | |
570fcd08 | 11885 | DO_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
11886 | PL_PosixCntrl, PL_XPosixCntrl); |
11887 | break; | |
11888 | case ANYOF_NCNTRL: | |
570fcd08 | 11889 | DO_N_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
11890 | PL_PosixCntrl, PL_XPosixCntrl); |
11891 | break; | |
ffc61ed2 | 11892 | case ANYOF_DIGIT: |
4f6289a3 RB |
11893 | /* There are no digits in the Latin1 range outside of |
11894 | * ASCII, so call the macro that doesn't have to resolve | |
11895 | * them */ | |
570fcd08 | 11896 | DO_POSIX_LATIN1_ONLY_KNOWN_L1_RESOLVED(ret, namedclass, posixes, |
4f6289a3 | 11897 | PL_PosixDigit, "XPosixDigit", listsv); |
ea317ccb KW |
11898 | break; |
11899 | case ANYOF_NDIGIT: | |
570fcd08 | 11900 | DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
5073ffbd KW |
11901 | PL_PosixDigit, PL_PosixDigit, "XPosixDigit", listsv, |
11902 | runtime_posix_matches_above_Unicode); | |
ea317ccb KW |
11903 | break; |
11904 | case ANYOF_GRAPH: | |
570fcd08 | 11905 | DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
ea317ccb KW |
11906 | PL_PosixGraph, PL_L1PosixGraph, "XPosixGraph", listsv); |
11907 | break; | |
11908 | case ANYOF_NGRAPH: | |
570fcd08 | 11909 | DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
5073ffbd KW |
11910 | PL_PosixGraph, PL_L1PosixGraph, "XPosixGraph", listsv, |
11911 | runtime_posix_matches_above_Unicode); | |
ea317ccb KW |
11912 | break; |
11913 | case ANYOF_HORIZWS: | |
cfbb2758 | 11914 | /* For these, we use the cp_list, as /d doesn't make a |
ea317ccb KW |
11915 | * difference in what these match. There would be problems |
11916 | * if these characters had folds other than themselves, as | |
cfbb2758 | 11917 | * cp_list is subject to folding. It turns out that \h |
24caacbc | 11918 | * is just a synonym for XPosixBlank */ |
cfbb2758 | 11919 | _invlist_union(cp_list, PL_XPosixBlank, &cp_list); |
ea317ccb KW |
11920 | break; |
11921 | case ANYOF_NHORIZWS: | |
cfbb2758 KW |
11922 | _invlist_union_complement_2nd(cp_list, |
11923 | PL_XPosixBlank, &cp_list); | |
ea317ccb KW |
11924 | break; |
11925 | case ANYOF_LOWER: | |
11926 | case ANYOF_NLOWER: | |
11927 | { /* These require special handling, as they differ under | |
dab0c3e7 KW |
11928 | folding, matching Cased there (which in the ASCII range |
11929 | is the same as Alpha */ | |
ea317ccb KW |
11930 | |
11931 | SV* ascii_source; | |
11932 | SV* l1_source; | |
11933 | const char *Xname; | |
11934 | ||
11935 | if (FOLD && ! LOC) { | |
11936 | ascii_source = PL_PosixAlpha; | |
dab0c3e7 KW |
11937 | l1_source = PL_L1Cased; |
11938 | Xname = "Cased"; | |
ea317ccb | 11939 | } |
ffc61ed2 | 11940 | else { |
ea317ccb KW |
11941 | ascii_source = PL_PosixLower; |
11942 | l1_source = PL_L1PosixLower; | |
11943 | Xname = "XPosixLower"; | |
11944 | } | |
11945 | if (namedclass == ANYOF_LOWER) { | |
570fcd08 | 11946 | DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
ea317ccb KW |
11947 | ascii_source, l1_source, Xname, listsv); |
11948 | } | |
11949 | else { | |
11950 | DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, | |
5073ffbd KW |
11951 | posixes, ascii_source, l1_source, Xname, listsv, |
11952 | runtime_posix_matches_above_Unicode); | |
ffc61ed2 | 11953 | } |
ffc61ed2 | 11954 | break; |
ea317ccb KW |
11955 | } |
11956 | case ANYOF_PRINT: | |
570fcd08 | 11957 | DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
ea317ccb KW |
11958 | PL_PosixPrint, PL_L1PosixPrint, "XPosixPrint", listsv); |
11959 | break; | |
11960 | case ANYOF_NPRINT: | |
570fcd08 | 11961 | DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
5073ffbd KW |
11962 | PL_PosixPrint, PL_L1PosixPrint, "XPosixPrint", listsv, |
11963 | runtime_posix_matches_above_Unicode); | |
ea317ccb KW |
11964 | break; |
11965 | case ANYOF_PUNCT: | |
570fcd08 | 11966 | DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
ea317ccb KW |
11967 | PL_PosixPunct, PL_L1PosixPunct, "XPosixPunct", listsv); |
11968 | break; | |
11969 | case ANYOF_NPUNCT: | |
570fcd08 | 11970 | DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
5073ffbd KW |
11971 | PL_PosixPunct, PL_L1PosixPunct, "XPosixPunct", listsv, |
11972 | runtime_posix_matches_above_Unicode); | |
ea317ccb KW |
11973 | break; |
11974 | case ANYOF_PSXSPC: | |
570fcd08 | 11975 | DO_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
11976 | PL_PosixSpace, PL_XPosixSpace); |
11977 | break; | |
11978 | case ANYOF_NPSXSPC: | |
570fcd08 | 11979 | DO_N_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
11980 | PL_PosixSpace, PL_XPosixSpace); |
11981 | break; | |
11982 | case ANYOF_SPACE: | |
570fcd08 | 11983 | DO_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
11984 | PL_PerlSpace, PL_XPerlSpace); |
11985 | break; | |
11986 | case ANYOF_NSPACE: | |
570fcd08 | 11987 | DO_N_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
11988 | PL_PerlSpace, PL_XPerlSpace); |
11989 | break; | |
11990 | case ANYOF_UPPER: /* Same as LOWER, above */ | |
11991 | case ANYOF_NUPPER: | |
11992 | { | |
11993 | SV* ascii_source; | |
11994 | SV* l1_source; | |
11995 | const char *Xname; | |
11996 | ||
11997 | if (FOLD && ! LOC) { | |
11998 | ascii_source = PL_PosixAlpha; | |
dab0c3e7 KW |
11999 | l1_source = PL_L1Cased; |
12000 | Xname = "Cased"; | |
ea317ccb | 12001 | } |
ffc61ed2 | 12002 | else { |
ea317ccb KW |
12003 | ascii_source = PL_PosixUpper; |
12004 | l1_source = PL_L1PosixUpper; | |
12005 | Xname = "XPosixUpper"; | |
ffc61ed2 | 12006 | } |
ea317ccb | 12007 | if (namedclass == ANYOF_UPPER) { |
570fcd08 | 12008 | DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
ea317ccb | 12009 | ascii_source, l1_source, Xname, listsv); |
cfaf538b | 12010 | } |
ea317ccb KW |
12011 | else { |
12012 | DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, | |
5073ffbd KW |
12013 | posixes, ascii_source, l1_source, Xname, listsv, |
12014 | runtime_posix_matches_above_Unicode); | |
ea317ccb KW |
12015 | } |
12016 | break; | |
12017 | } | |
12018 | case ANYOF_ALNUM: /* Really is 'Word' */ | |
570fcd08 | 12019 | DO_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
ea317ccb KW |
12020 | PL_PosixWord, PL_L1PosixWord, "XPosixWord", listsv); |
12021 | break; | |
12022 | case ANYOF_NALNUM: | |
570fcd08 | 12023 | DO_N_POSIX_LATIN1_ONLY_KNOWN(ret, namedclass, posixes, |
5073ffbd KW |
12024 | PL_PosixWord, PL_L1PosixWord, "XPosixWord", listsv, |
12025 | runtime_posix_matches_above_Unicode); | |
ea317ccb KW |
12026 | break; |
12027 | case ANYOF_VERTWS: | |
cfbb2758 | 12028 | /* For these, we use the cp_list, as /d doesn't make a |
ea317ccb KW |
12029 | * difference in what these match. There would be problems |
12030 | * if these characters had folds other than themselves, as | |
cfbb2758 KW |
12031 | * cp_list is subject to folding */ |
12032 | _invlist_union(cp_list, PL_VertSpace, &cp_list); | |
ea317ccb KW |
12033 | break; |
12034 | case ANYOF_NVERTWS: | |
cfbb2758 KW |
12035 | _invlist_union_complement_2nd(cp_list, |
12036 | PL_VertSpace, &cp_list); | |
ea317ccb KW |
12037 | break; |
12038 | case ANYOF_XDIGIT: | |
570fcd08 | 12039 | DO_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
12040 | PL_PosixXDigit, PL_XPosixXDigit); |
12041 | break; | |
12042 | case ANYOF_NXDIGIT: | |
570fcd08 | 12043 | DO_N_POSIX(ret, namedclass, posixes, |
ea317ccb KW |
12044 | PL_PosixXDigit, PL_XPosixXDigit); |
12045 | break; | |
f81125e2 JP |
12046 | case ANYOF_MAX: |
12047 | /* this is to handle \p and \P */ | |
12048 | break; | |
73b437c8 | 12049 | default: |
b45f050a | 12050 | vFAIL("Invalid [::] class"); |
73b437c8 | 12051 | break; |
b8c5462f | 12052 | } |
ce1c68b2 | 12053 | |
18890cbd | 12054 | continue; /* Go get next character */ |
a0d0e21e | 12055 | } |
ffc61ed2 JH |
12056 | } /* end of namedclass \blah */ |
12057 | ||
a0d0e21e | 12058 | if (range) { |
7cc714e6 | 12059 | if (prevvalue > value) /* b-a */ { |
d4c19fe8 AL |
12060 | const int w = RExC_parse - rangebegin; |
12061 | Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin); | |
3568d838 | 12062 | range = 0; /* not a valid range */ |
73b437c8 | 12063 | } |
a0d0e21e LW |
12064 | } |
12065 | else { | |
18890cbd | 12066 | prevvalue = value; /* save the beginning of the potential range */ |
646253b5 KW |
12067 | if (RExC_parse+1 < RExC_end |
12068 | && *RExC_parse == '-' | |
12069 | && RExC_parse[1] != ']') | |
12070 | { | |
830247a4 | 12071 | RExC_parse++; |
ffc61ed2 JH |
12072 | |
12073 | /* a bad range like \w-, [:word:]- ? */ | |
12074 | if (namedclass > OOB_NAMEDCLASS) { | |
afd78fd5 | 12075 | if (ckWARN(WARN_REGEXP)) { |
d4c19fe8 | 12076 | const int w = |
afd78fd5 JH |
12077 | RExC_parse >= rangebegin ? |
12078 | RExC_parse - rangebegin : 0; | |
830247a4 | 12079 | vWARN4(RExC_parse, |
b45f050a | 12080 | "False [] range \"%*.*s\"", |
097eb12c | 12081 | w, w, rangebegin); |
afd78fd5 | 12082 | } |
bd771b08 | 12083 | if (!SIZE_ONLY) { |
cfbb2758 | 12084 | cp_list = add_cp_to_invlist(cp_list, '-'); |
bd771b08 | 12085 | } |
2111a35c | 12086 | element_count++; |
73b437c8 | 12087 | } else |
ffc61ed2 JH |
12088 | range = 1; /* yeah, it's a range! */ |
12089 | continue; /* but do it the next time */ | |
a0d0e21e | 12090 | } |
a687059c | 12091 | } |
ffc61ed2 | 12092 | |
18890cbd KW |
12093 | /* Here, <prevvalue> is the beginning of the range, if any; or <value> |
12094 | * if not */ | |
12095 | ||
046c4055 KW |
12096 | /* non-Latin1 code point implies unicode semantics. Must be set in |
12097 | * pass1 so is there for the whole of pass 2 */ | |
56ca34ca KW |
12098 | if (value > 255) { |
12099 | RExC_uni_semantics = 1; | |
12100 | } | |
12101 | ||
18890cbd | 12102 | /* Ready to process either the single value, or the completed range */ |
ae5c130c | 12103 | if (!SIZE_ONLY) { |
68823f48 | 12104 | #ifndef EBCDIC |
cfbb2758 | 12105 | cp_list = _add_range_to_invlist(cp_list, prevvalue, value); |
68823f48 KW |
12106 | #else |
12107 | UV* this_range = _new_invlist(1); | |
12108 | _append_range_to_invlist(this_range, prevvalue, value); | |
12109 | ||
12110 | /* In EBCDIC, the ranges 'A-Z' and 'a-z' are each not contiguous. | |
12111 | * If this range was specified using something like 'i-j', we want | |
12112 | * to include only the 'i' and the 'j', and not anything in | |
12113 | * between, so exclude non-ASCII, non-alphabetics from it. | |
12114 | * However, if the range was specified with something like | |
12115 | * [\x89-\x91] or [\x89-j], all code points within it should be | |
12116 | * included. literal_endpoint==2 means both ends of the range used | |
12117 | * a literal character, not \x{foo} */ | |
12118 | if (literal_endpoint == 2 | |
12119 | && (prevvalue >= 'a' && value <= 'z') | |
12120 | || (prevvalue >= 'A' && value <= 'Z')) | |
12121 | { | |
12122 | _invlist_intersection(this_range, PL_ASCII, &this_range, ); | |
12123 | _invlist_intersection(this_range, PL_Alpha, &this_range, ); | |
68823f48 | 12124 | } |
cfbb2758 | 12125 | _invlist_union(cp_list, this_range, &cp_list); |
68823f48 | 12126 | literal_endpoint = 0; |
1b2d223b | 12127 | #endif |
8ada0baa | 12128 | } |
ffc61ed2 JH |
12129 | |
12130 | range = 0; /* this range (if it was one) is done now */ | |
18890cbd | 12131 | } /* End of loop through all the text within the brackets */ |
ffc61ed2 | 12132 | |
c9f92f3e KW |
12133 | /* If the character class contains only a single element, it may be |
12134 | * optimizable into another node type which is smaller and runs faster. | |
12135 | * Check if this is the case for this class */ | |
12136 | if (element_count == 1) { | |
12137 | U8 op = END; | |
0658cdde | 12138 | U8 arg = 0; |
3172e3fd | 12139 | |
c9f92f3e KW |
12140 | if (namedclass > OOB_NAMEDCLASS) { /* this is a named class, like \w or |
12141 | [:digit:] or \p{foo} */ | |
3172e3fd | 12142 | |
c9f92f3e KW |
12143 | /* Certain named classes have equivalents that can appear outside a |
12144 | * character class, e.g. \w, \H. We use these instead of a | |
12145 | * character class. */ | |
3172e3fd | 12146 | switch ((I32)namedclass) { |
c9f92f3e KW |
12147 | U8 offset; |
12148 | ||
12149 | /* The first group is for node types that depend on the charset | |
12150 | * modifier to the regex. We first calculate the base node | |
12151 | * type, and if it should be inverted */ | |
12152 | ||
3172e3fd KW |
12153 | case ANYOF_NALNUM: |
12154 | invert = ! invert; | |
12155 | /* FALLTHROUGH */ | |
12156 | case ANYOF_ALNUM: | |
12157 | op = ALNUM; | |
c9f92f3e KW |
12158 | goto join_charset_classes; |
12159 | ||
3172e3fd KW |
12160 | case ANYOF_NSPACE: |
12161 | invert = ! invert; | |
12162 | /* FALLTHROUGH */ | |
12163 | case ANYOF_SPACE: | |
12164 | op = SPACE; | |
c9f92f3e KW |
12165 | goto join_charset_classes; |
12166 | ||
3172e3fd KW |
12167 | case ANYOF_NDIGIT: |
12168 | invert = ! invert; | |
12169 | /* FALLTHROUGH */ | |
12170 | case ANYOF_DIGIT: | |
12171 | op = DIGIT; | |
12172 | ||
c9f92f3e KW |
12173 | join_charset_classes: |
12174 | ||
12175 | /* Now that we have the base node type, we take advantage | |
12176 | * of the enum ordering of the charset modifiers to get the | |
12177 | * exact node type, For example the base SPACE also has | |
12178 | * SPACEL, SPACEU, and SPACEA */ | |
12179 | ||
12180 | offset = get_regex_charset(RExC_flags); | |
12181 | ||
12182 | /* /aa is the same as /a for these */ | |
12183 | if (offset == REGEX_ASCII_MORE_RESTRICTED_CHARSET) { | |
12184 | offset = REGEX_ASCII_RESTRICTED_CHARSET; | |
12185 | } | |
12186 | else if (op == DIGIT && offset == REGEX_UNICODE_CHARSET) { | |
12187 | offset = REGEX_DEPENDS_CHARSET; /* There is no DIGITU */ | |
3172e3fd | 12188 | } |
3172e3fd | 12189 | |
c9f92f3e | 12190 | op += offset; |
3172e3fd | 12191 | |
c9f92f3e KW |
12192 | /* The number of varieties of each of these is the same, |
12193 | * hence, so is the delta between the normal and | |
12194 | * complemented nodes */ | |
12195 | if (invert) { | |
12196 | op += NALNUM - ALNUM; | |
12197 | } | |
34b39fc9 | 12198 | *flagp |= HASWIDTH|SIMPLE; |
c9f92f3e KW |
12199 | break; |
12200 | ||
12201 | /* The second group doesn't depend of the charset modifiers. | |
12202 | * We just have normal and complemented */ | |
3172e3fd KW |
12203 | case ANYOF_NHORIZWS: |
12204 | invert = ! invert; | |
12205 | /* FALLTHROUGH */ | |
12206 | case ANYOF_HORIZWS: | |
fa0174d2 | 12207 | is_horizws: |
c9f92f3e | 12208 | op = (invert) ? NHORIZWS : HORIZWS; |
34b39fc9 | 12209 | *flagp |= HASWIDTH|SIMPLE; |
3172e3fd | 12210 | break; |
c9f92f3e | 12211 | |
3172e3fd KW |
12212 | case ANYOF_NVERTWS: |
12213 | invert = ! invert; | |
12214 | /* FALLTHROUGH */ | |
12215 | case ANYOF_VERTWS: | |
c9f92f3e | 12216 | op = (invert) ? NVERTWS : VERTWS; |
34b39fc9 | 12217 | *flagp |= HASWIDTH|SIMPLE; |
3172e3fd | 12218 | break; |
3172e3fd | 12219 | |
0658cdde KW |
12220 | case ANYOF_MAX: |
12221 | break; | |
c9f92f3e | 12222 | |
fa0174d2 KW |
12223 | case ANYOF_NBLANK: |
12224 | invert = ! invert; | |
12225 | /* FALLTHROUGH */ | |
12226 | case ANYOF_BLANK: | |
12227 | if (AT_LEAST_UNI_SEMANTICS && ! AT_LEAST_ASCII_RESTRICTED) { | |
12228 | goto is_horizws; | |
12229 | } | |
12230 | /* FALLTHROUGH */ | |
0658cdde KW |
12231 | default: |
12232 | /* A generic posix class. All the /a ones can be handled | |
12233 | * by the POSIXA opcode. And all are closed under folding | |
12234 | * in the ASCII range, so FOLD doesn't matter */ | |
12235 | if (AT_LEAST_ASCII_RESTRICTED | |
12236 | || (! LOC && namedclass == ANYOF_ASCII)) | |
12237 | { | |
12238 | /* The odd numbered ones are the complements of the | |
12239 | * next-lower even number one */ | |
12240 | if (namedclass % 2 == 1) { | |
12241 | invert = ! invert; | |
12242 | namedclass--; | |
12243 | } | |
12244 | arg = namedclass_to_classnum(namedclass); | |
12245 | op = (invert) ? NPOSIXA : POSIXA; | |
12246 | } | |
12247 | break; | |
3172e3fd KW |
12248 | } |
12249 | } | |
3e6d27d5 KW |
12250 | else if (value == prevvalue) { |
12251 | ||
12252 | /* Here, the class consists of just a single code point */ | |
12253 | ||
12254 | if (invert) { | |
12255 | if (! LOC && value == '\n') { | |
12256 | op = REG_ANY; /* Optimize [^\n] */ | |
34b39fc9 KW |
12257 | *flagp |= HASWIDTH|SIMPLE; |
12258 | RExC_naughty++; | |
3e6d27d5 KW |
12259 | } |
12260 | } | |
12261 | else if (value < 256 || UTF) { | |
12262 | ||
12263 | /* Optimize a single value into an EXACTish node, but not if it | |
12264 | * would require converting the pattern to UTF-8. */ | |
12265 | op = compute_EXACTish(pRExC_state); | |
5a279aa0 | 12266 | } |
3e6d27d5 KW |
12267 | } /* Otherwise is a range */ |
12268 | else if (! LOC) { /* locale could vary these */ | |
12269 | if (prevvalue == '0') { | |
12270 | if (value == '9') { | |
12271 | op = (invert) ? NDIGITA : DIGITA; | |
34b39fc9 | 12272 | *flagp |= HASWIDTH|SIMPLE; |
3e6d27d5 | 12273 | } |
c9f92f3e | 12274 | } |
3a64b515 | 12275 | } |
3172e3fd | 12276 | |
c9f92f3e KW |
12277 | /* Here, we have changed <op> away from its initial value iff we found |
12278 | * an optimization */ | |
12279 | if (op != END) { | |
12280 | ||
18890cbd KW |
12281 | /* Throw away this ANYOF regnode, and emit the calculated one, |
12282 | * which should correspond to the beginning, not current, state of | |
12283 | * the parse */ | |
12284 | const char * cur_parse = RExC_parse; | |
12285 | RExC_parse = (char *)orig_parse; | |
6197c735 KW |
12286 | if ( SIZE_ONLY) { |
12287 | if (! LOC) { | |
12288 | ||
12289 | /* To get locale nodes to not use the full ANYOF size would | |
12290 | * require moving the code above that writes the portions | |
12291 | * of it that aren't in other nodes to after this point. | |
12292 | * e.g. ANYOF_CLASS_SET */ | |
12293 | RExC_size = orig_size; | |
12294 | } | |
12295 | } | |
12296 | else { | |
12297 | RExC_emit = (regnode *)orig_emit; | |
12298 | } | |
3172e3fd | 12299 | |
18890cbd KW |
12300 | ret = reg_node(pRExC_state, op); |
12301 | ||
0658cdde KW |
12302 | if (PL_regkind[op] == POSIXD) { |
12303 | if (! SIZE_ONLY) { | |
12304 | FLAGS(ret) = arg; | |
12305 | } | |
34b39fc9 | 12306 | *flagp |= HASWIDTH|SIMPLE; |
0658cdde KW |
12307 | } |
12308 | else if (PL_regkind[op] == EXACT) { | |
34b39fc9 | 12309 | alloc_maybe_populate_EXACT(pRExC_state, ret, flagp, 0, value); |
3e6d27d5 KW |
12310 | } |
12311 | ||
18890cbd KW |
12312 | RExC_parse = (char *) cur_parse; |
12313 | ||
12314 | SvREFCNT_dec(listsv); | |
12315 | return ret; | |
c9f92f3e | 12316 | } |
3172e3fd KW |
12317 | } |
12318 | ||
7f6f358c YO |
12319 | if (SIZE_ONLY) |
12320 | return ret; | |
6012a526 | 12321 | /****** !SIZE_ONLY (Pass 2) AFTER HERE *********/ |
7f6f358c | 12322 | |
68823f48 KW |
12323 | /* If folding, we calculate all characters that could fold to or from the |
12324 | * ones already on the list */ | |
cfbb2758 | 12325 | if (FOLD && cp_list) { |
0d527bf8 | 12326 | UV start, end; /* End points of code point ranges */ |
56ca34ca | 12327 | |
4065ba03 | 12328 | SV* fold_intersection = NULL; |
93e5bb1c | 12329 | |
68823f48 KW |
12330 | /* In the Latin1 range, the characters that can be folded-to or -from |
12331 | * are precisely the alphabetic characters. If the highest code point | |
12332 | * is within Latin1, we can use the compiled-in list, and not have to | |
4f3e8b0f KW |
12333 | * go out to disk. */ |
12334 | if (invlist_highest(cp_list) < 256) { | |
cfbb2758 | 12335 | _invlist_intersection(PL_L1PosixAlpha, cp_list, &fold_intersection); |
68823f48 KW |
12336 | } |
12337 | else { | |
12338 | ||
ea220d70 KW |
12339 | /* Here, there are non-Latin1 code points, so we will have to go |
12340 | * fetch the list of all the characters that participate in folds | |
12341 | */ | |
8f850557 | 12342 | if (! PL_utf8_foldable) { |
ea220d70 KW |
12343 | SV* swash = swash_init("utf8", "_Perl_Any_Folds", |
12344 | &PL_sv_undef, 1, 0); | |
3fdfee00 | 12345 | PL_utf8_foldable = _get_swash_invlist(swash); |
8f850557 KW |
12346 | SvREFCNT_dec(swash); |
12347 | } | |
68823f48 | 12348 | |
8f850557 KW |
12349 | /* This is a hash that for a particular fold gives all characters |
12350 | * that are involved in it */ | |
12351 | if (! PL_utf8_foldclosures) { | |
12352 | ||
12353 | /* If we were unable to find any folds, then we likely won't be | |
12354 | * able to find the closures. So just create an empty list. | |
12355 | * Folding will effectively be restricted to the non-Unicode | |
12356 | * rules hard-coded into Perl. (This case happens legitimately | |
12357 | * during compilation of Perl itself before the Unicode tables | |
12358 | * are generated) */ | |
29f51c6b | 12359 | if (_invlist_len(PL_utf8_foldable) == 0) { |
8f850557 KW |
12360 | PL_utf8_foldclosures = newHV(); |
12361 | } | |
12362 | else { | |
12363 | /* If the folds haven't been read in, call a fold function | |
12364 | * to force that */ | |
12365 | if (! PL_utf8_tofold) { | |
12366 | U8 dummy[UTF8_MAXBYTES+1]; | |
12367 | STRLEN dummy_len; | |
12368 | ||
61dad979 KW |
12369 | /* This string is just a short named one above \xff */ |
12370 | to_utf8_fold((U8*) HYPHEN_UTF8, dummy, &dummy_len); | |
8f850557 KW |
12371 | assert(PL_utf8_tofold); /* Verify that worked */ |
12372 | } | |
12373 | PL_utf8_foldclosures = | |
12374 | _swash_inversion_hash(PL_utf8_tofold); | |
12375 | } | |
12376 | } | |
93e5bb1c | 12377 | |
8f850557 KW |
12378 | /* Only the characters in this class that participate in folds need |
12379 | * be checked. Get the intersection of this class and all the | |
12380 | * possible characters that are foldable. This can quickly narrow | |
12381 | * down a large class */ | |
cfbb2758 | 12382 | _invlist_intersection(PL_utf8_foldable, cp_list, |
8f850557 | 12383 | &fold_intersection); |
68823f48 | 12384 | } |
93e5bb1c KW |
12385 | |
12386 | /* Now look at the foldable characters in this class individually */ | |
0d527bf8 KW |
12387 | invlist_iterinit(fold_intersection); |
12388 | while (invlist_iternext(fold_intersection, &start, &end)) { | |
93e5bb1c KW |
12389 | UV j; |
12390 | ||
68823f48 KW |
12391 | /* Locale folding for Latin1 characters is deferred until runtime */ |
12392 | if (LOC && start < 256) { | |
12393 | start = 256; | |
12394 | } | |
12395 | ||
93e5bb1c KW |
12396 | /* Look at every character in the range */ |
12397 | for (j = start; j <= end; j++) { | |
12398 | ||
93e5bb1c KW |
12399 | U8 foldbuf[UTF8_MAXBYTES_CASE+1]; |
12400 | STRLEN foldlen; | |
68823f48 KW |
12401 | UV f; |
12402 | ||
12403 | if (j < 256) { | |
12404 | ||
12405 | /* We have the latin1 folding rules hard-coded here so that | |
12406 | * an innocent-looking character class, like /[ks]/i won't | |
12407 | * have to go out to disk to find the possible matches. | |
12408 | * XXX It would be better to generate these via regen, in | |
12409 | * case a new version of the Unicode standard adds new | |
12410 | * mappings, though that is not really likely, and may be | |
12411 | * caught by the default: case of the switch below. */ | |
12412 | ||
12413 | if (PL_fold_latin1[j] != j) { | |
12414 | ||
12415 | /* ASCII is always matched; non-ASCII is matched only | |
12416 | * under Unicode rules */ | |
12417 | if (isASCII(j) || AT_LEAST_UNI_SEMANTICS) { | |
cfbb2758 KW |
12418 | cp_list = |
12419 | add_cp_to_invlist(cp_list, PL_fold_latin1[j]); | |
68823f48 KW |
12420 | } |
12421 | else { | |
12422 | depends_list = | |
18890cbd | 12423 | add_cp_to_invlist(depends_list, PL_fold_latin1[j]); |
68823f48 KW |
12424 | } |
12425 | } | |
12426 | ||
12427 | if (HAS_NONLATIN1_FOLD_CLOSURE(j) | |
a725e29c | 12428 | && (! isASCII(j) || ! ASCII_FOLD_RESTRICTED)) |
68823f48 KW |
12429 | { |
12430 | /* Certain Latin1 characters have matches outside | |
12431 | * Latin1, or are multi-character. To get here, 'j' is | |
12432 | * one of those characters. None of these matches is | |
12433 | * valid for ASCII characters under /aa, which is why | |
12434 | * the 'if' just above excludes those. The matches | |
12435 | * fall into three categories: | |
12436 | * 1) They are singly folded-to or -from an above 255 | |
12437 | * character, e.g., LATIN SMALL LETTER Y WITH | |
12438 | * DIAERESIS and LATIN CAPITAL LETTER Y WITH | |
12439 | * DIAERESIS; | |
12440 | * 2) They are part of a multi-char fold with another | |
12441 | * latin1 character; only LATIN SMALL LETTER | |
12442 | * SHARP S => "ss" fits this; | |
12443 | * 3) They are part of a multi-char fold with a | |
12444 | * character outside of Latin1, such as various | |
12445 | * ligatures. | |
12446 | * We aren't dealing fully with multi-char folds, except | |
12447 | * we do deal with the pattern containing a character | |
12448 | * that has a multi-char fold (not so much the inverse). | |
12449 | * For types 1) and 3), the matches only happen when the | |
12450 | * target string is utf8; that's not true for 2), and we | |
12451 | * set a flag for it. | |
12452 | * | |
12453 | * The code below adds the single fold closures for 'j' | |
12454 | * to the inversion list. */ | |
12455 | switch (j) { | |
12456 | case 'k': | |
12457 | case 'K': | |
cfbb2758 | 12458 | cp_list = |
8d64d87f | 12459 | add_cp_to_invlist(cp_list, KELVIN_SIGN); |
68823f48 KW |
12460 | break; |
12461 | case 's': | |
12462 | case 'S': | |
8d64d87f KW |
12463 | cp_list = add_cp_to_invlist(cp_list, |
12464 | LATIN_SMALL_LETTER_LONG_S); | |
68823f48 KW |
12465 | break; |
12466 | case MICRO_SIGN: | |
cfbb2758 | 12467 | cp_list = add_cp_to_invlist(cp_list, |
68823f48 | 12468 | GREEK_CAPITAL_LETTER_MU); |
7c23e637 KW |
12469 | cp_list = add_cp_to_invlist(cp_list, |
12470 | GREEK_SMALL_LETTER_MU); | |
68823f48 KW |
12471 | break; |
12472 | case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE: | |
12473 | case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE: | |
cfbb2758 | 12474 | cp_list = |
8d64d87f | 12475 | add_cp_to_invlist(cp_list, ANGSTROM_SIGN); |
68823f48 KW |
12476 | break; |
12477 | case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: | |
cfbb2758 | 12478 | cp_list = add_cp_to_invlist(cp_list, |
68823f48 KW |
12479 | LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS); |
12480 | break; | |
12481 | case LATIN_SMALL_LETTER_SHARP_S: | |
cfbb2758 | 12482 | cp_list = add_cp_to_invlist(cp_list, |
68823f48 KW |
12483 | LATIN_CAPITAL_LETTER_SHARP_S); |
12484 | ||
12485 | /* Under /a, /d, and /u, this can match the two | |
12486 | * chars "ss" */ | |
a725e29c | 12487 | if (! ASCII_FOLD_RESTRICTED) { |
68823f48 KW |
12488 | add_alternate(&unicode_alternate, |
12489 | (U8 *) "ss", 2); | |
12490 | ||
12491 | /* And under /u or /a, it can match even if | |
12492 | * the target is not utf8 */ | |
12493 | if (AT_LEAST_UNI_SEMANTICS) { | |
12494 | ANYOF_FLAGS(ret) |= | |
12495 | ANYOF_NONBITMAP_NON_UTF8; | |
12496 | } | |
12497 | } | |
12498 | break; | |
12499 | case 'F': case 'f': | |
12500 | case 'I': case 'i': | |
12501 | case 'L': case 'l': | |
12502 | case 'T': case 't': | |
12503 | case 'A': case 'a': | |
12504 | case 'H': case 'h': | |
12505 | case 'J': case 'j': | |
12506 | case 'N': case 'n': | |
12507 | case 'W': case 'w': | |
12508 | case 'Y': case 'y': | |
12509 | /* These all are targets of multi-character | |
12510 | * folds from code points that require UTF8 to | |
12511 | * express, so they can't match unless the | |
12512 | * target string is in UTF-8, so no action here | |
12513 | * is necessary, as regexec.c properly handles | |
12514 | * the general case for UTF-8 matching */ | |
12515 | break; | |
12516 | default: | |
12517 | /* Use deprecated warning to increase the | |
12518 | * chances of this being output */ | |
12519 | ckWARN2regdep(RExC_parse, "Perl folding rules are not up-to-date for 0x%"UVXf"; please use the perlbug utility to report;", j); | |
12520 | break; | |
12521 | } | |
12522 | } | |
12523 | continue; | |
12524 | } | |
12525 | ||
12526 | /* Here is an above Latin1 character. We don't have the rules | |
12527 | * hard-coded for it. First, get its fold */ | |
12528 | f = _to_uni_fold_flags(j, foldbuf, &foldlen, | |
c2df36c4 KW |
12529 | ((allow_full_fold) ? FOLD_FLAGS_FULL : 0) |
12530 | | ((LOC) | |
12531 | ? FOLD_FLAGS_LOCALE | |
a725e29c | 12532 | : (ASCII_FOLD_RESTRICTED) |
c2df36c4 KW |
12533 | ? FOLD_FLAGS_NOMIX_ASCII |
12534 | : 0)); | |
93e5bb1c KW |
12535 | |
12536 | if (foldlen > (STRLEN)UNISKIP(f)) { | |
12537 | ||
dbe7a391 KW |
12538 | /* Any multicharacter foldings (disallowed in lookbehind |
12539 | * patterns) require the following transform: [ABCDEF] -> | |
12540 | * (?:[ABCabcDEFd]|pq|rst) where E folds into "pq" and F | |
12541 | * folds into "rst", all other characters fold to single | |
12542 | * characters. We save away these multicharacter foldings, | |
12543 | * to be later saved as part of the additional "s" data. */ | |
93e5bb1c KW |
12544 | if (! RExC_in_lookbehind) { |
12545 | U8* loc = foldbuf; | |
12546 | U8* e = foldbuf + foldlen; | |
12547 | ||
dbe7a391 KW |
12548 | /* If any of the folded characters of this are in the |
12549 | * Latin1 range, tell the regex engine that this can | |
68823f48 | 12550 | * match a non-utf8 target string. */ |
8f850557 | 12551 | while (loc < e) { |
8f850557 KW |
12552 | if (UTF8_IS_INVARIANT(*loc) |
12553 | || UTF8_IS_DOWNGRADEABLE_START(*loc)) | |
12554 | { | |
8f850557 KW |
12555 | ANYOF_FLAGS(ret) |
12556 | |= ANYOF_NONBITMAP_NON_UTF8; | |
12557 | break; | |
12558 | } | |
12559 | loc += UTF8SKIP(loc); | |
12560 | } | |
17580e7a | 12561 | |
93e5bb1c | 12562 | add_alternate(&unicode_alternate, foldbuf, foldlen); |
93e5bb1c KW |
12563 | } |
12564 | } | |
68823f48 KW |
12565 | else { |
12566 | /* Single character fold of above Latin1. Add everything | |
12567 | * in its fold closure to the list that this node should | |
12568 | * match */ | |
93e5bb1c KW |
12569 | SV** listp; |
12570 | ||
dbe7a391 KW |
12571 | /* The fold closures data structure is a hash with the keys |
12572 | * being every character that is folded to, like 'k', and | |
12573 | * the values each an array of everything that folds to its | |
12574 | * key. e.g. [ 'k', 'K', KELVIN_SIGN ] */ | |
93e5bb1c KW |
12575 | if ((listp = hv_fetch(PL_utf8_foldclosures, |
12576 | (char *) foldbuf, foldlen, FALSE))) | |
12577 | { | |
12578 | AV* list = (AV*) *listp; | |
12579 | IV k; | |
12580 | for (k = 0; k <= av_len(list); k++) { | |
12581 | SV** c_p = av_fetch(list, k, FALSE); | |
12582 | UV c; | |
12583 | if (c_p == NULL) { | |
12584 | Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); | |
12585 | } | |
12586 | c = SvUV(*c_p); | |
12587 | ||
dbe7a391 KW |
12588 | /* /aa doesn't allow folds between ASCII and non-; |
12589 | * /l doesn't allow them between above and below | |
12590 | * 256 */ | |
a725e29c | 12591 | if ((ASCII_FOLD_RESTRICTED |
18890cbd | 12592 | && (isASCII(c) != isASCII(j))) |
8f850557 | 12593 | || (LOC && ((c < 256) != (j < 256)))) |
93e5bb1c KW |
12594 | { |
12595 | continue; | |
12596 | } | |
56ca34ca | 12597 | |
68823f48 KW |
12598 | /* Folds involving non-ascii Latin1 characters |
12599 | * under /d are added to a separate list */ | |
8f850557 KW |
12600 | if (isASCII(c) || c > 255 || AT_LEAST_UNI_SEMANTICS) |
12601 | { | |
cfbb2758 | 12602 | cp_list = add_cp_to_invlist(cp_list, c); |
68823f48 KW |
12603 | } |
12604 | else { | |
18890cbd | 12605 | depends_list = add_cp_to_invlist(depends_list, c); |
56ca34ca KW |
12606 | } |
12607 | } | |
12608 | } | |
12609 | } | |
8f850557 | 12610 | } |
93e5bb1c | 12611 | } |
318c430e | 12612 | SvREFCNT_dec(fold_intersection); |
56ca34ca KW |
12613 | } |
12614 | ||
570fcd08 KW |
12615 | /* And combine the result (if any) with any inversion list from posix |
12616 | * classes. The lists are kept separate up to now because we don't want to | |
18890cbd KW |
12617 | * fold the classes (folding of those is automatically handled by the swash |
12618 | * fetching code) */ | |
570fcd08 | 12619 | if (posixes) { |
5863c149 | 12620 | if (! DEPENDS_SEMANTICS) { |
cfbb2758 | 12621 | if (cp_list) { |
570fcd08 KW |
12622 | _invlist_union(cp_list, posixes, &cp_list); |
12623 | SvREFCNT_dec(posixes); | |
8f850557 KW |
12624 | } |
12625 | else { | |
570fcd08 | 12626 | cp_list = posixes; |
8f850557 | 12627 | } |
68823f48 KW |
12628 | } |
12629 | else { | |
26cdf4c4 KW |
12630 | /* Under /d, we put into a separate list the Latin1 things that |
12631 | * match only when the target string is utf8 */ | |
68823f48 | 12632 | SV* nonascii_but_latin1_properties = NULL; |
570fcd08 | 12633 | _invlist_intersection(posixes, PL_Latin1, |
8f850557 KW |
12634 | &nonascii_but_latin1_properties); |
12635 | _invlist_subtract(nonascii_but_latin1_properties, PL_ASCII, | |
12636 | &nonascii_but_latin1_properties); | |
570fcd08 KW |
12637 | _invlist_subtract(posixes, nonascii_but_latin1_properties, |
12638 | &posixes); | |
cfbb2758 | 12639 | if (cp_list) { |
570fcd08 KW |
12640 | _invlist_union(cp_list, posixes, &cp_list); |
12641 | SvREFCNT_dec(posixes); | |
68823f48 KW |
12642 | } |
12643 | else { | |
570fcd08 | 12644 | cp_list = posixes; |
68823f48 KW |
12645 | } |
12646 | ||
12647 | if (depends_list) { | |
12648 | _invlist_union(depends_list, nonascii_but_latin1_properties, | |
12649 | &depends_list); | |
12650 | SvREFCNT_dec(nonascii_but_latin1_properties); | |
12651 | } | |
12652 | else { | |
12653 | depends_list = nonascii_but_latin1_properties; | |
12654 | } | |
12655 | } | |
dc415b4a KW |
12656 | } |
12657 | ||
570fcd08 | 12658 | /* And combine the result (if any) with any inversion list from properties. |
5073ffbd KW |
12659 | * The lists are kept separate up to now so that we can distinguish the two |
12660 | * in regards to matching above-Unicode. A run-time warning is generated | |
12661 | * if a Unicode property is matched against a non-Unicode code point. But, | |
12662 | * we allow user-defined properties to match anything, without any warning, | |
12663 | * and we also suppress the warning if there is a portion of the character | |
12664 | * class that isn't a Unicode property, and which matches above Unicode, \W | |
12665 | * or [\x{110000}] for example. | |
570fcd08 KW |
12666 | * (Note that in this case, unlike the Posix one above, there is no |
12667 | * <depends_list>, because having a Unicode property forces Unicode | |
12668 | * semantics */ | |
12669 | if (properties) { | |
5073ffbd | 12670 | bool warn_super = ! has_user_defined_property; |
570fcd08 | 12671 | if (cp_list) { |
5073ffbd KW |
12672 | |
12673 | /* If it matters to the final outcome, see if a non-property | |
12674 | * component of the class matches above Unicode. If so, the | |
12675 | * warning gets suppressed. This is true even if just a single | |
12676 | * such code point is specified, as though not strictly correct if | |
12677 | * another such code point is matched against, the fact that they | |
12678 | * are using above-Unicode code points indicates they should know | |
12679 | * the issues involved */ | |
12680 | if (warn_super) { | |
12681 | bool non_prop_matches_above_Unicode = | |
12682 | runtime_posix_matches_above_Unicode | |
12683 | | (invlist_highest(cp_list) > PERL_UNICODE_MAX); | |
12684 | if (invert) { | |
12685 | non_prop_matches_above_Unicode = | |
12686 | ! non_prop_matches_above_Unicode; | |
12687 | } | |
12688 | warn_super = ! non_prop_matches_above_Unicode; | |
12689 | } | |
12690 | ||
12691 | _invlist_union(properties, cp_list, &cp_list); | |
570fcd08 KW |
12692 | SvREFCNT_dec(properties); |
12693 | } | |
12694 | else { | |
12695 | cp_list = properties; | |
12696 | } | |
5073ffbd KW |
12697 | |
12698 | if (warn_super) { | |
12699 | ANYOF_FLAGS(ret) |= ANYOF_WARN_SUPER; | |
12700 | } | |
570fcd08 KW |
12701 | } |
12702 | ||
ea364ff5 KW |
12703 | /* Here, we have calculated what code points should be in the character |
12704 | * class. | |
12705 | * | |
12706 | * Now we can see about various optimizations. Fold calculation (which we | |
12707 | * did above) needs to take place before inversion. Otherwise /[^k]/i | |
12708 | * would invert to include K, which under /i would match k, which it | |
ce712dff KW |
12709 | * shouldn't. Therefore we can't invert folded locale now, as it won't be |
12710 | * folded until runtime */ | |
ea364ff5 | 12711 | |
ce712dff KW |
12712 | /* Optimize inverted simple patterns (e.g. [^a-z]) when everything is known |
12713 | * at compile time. Besides not inverting folded locale now, we can't invert | |
12714 | * if there are things such as \w, which aren't known until runtime */ | |
520939ba | 12715 | if (invert |
ce712dff | 12716 | && ! (LOC && (FOLD || (ANYOF_FLAGS(ret) & ANYOF_CLASS))) |
ea364ff5 KW |
12717 | && ! depends_list |
12718 | && ! unicode_alternate | |
66c69124 | 12719 | && ! HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION) |
ea364ff5 KW |
12720 | { |
12721 | _invlist_invert(cp_list); | |
12722 | ||
12723 | /* Any swash can't be used as-is, because we've inverted things */ | |
12724 | if (swash) { | |
12725 | SvREFCNT_dec(swash); | |
12726 | swash = NULL; | |
12727 | } | |
12728 | ||
12729 | /* Clear the invert flag since have just done it here */ | |
520939ba | 12730 | invert = FALSE; |
ea364ff5 KW |
12731 | } |
12732 | ||
a0a06948 KW |
12733 | /* If we didn't do folding, it's because some information isn't available |
12734 | * until runtime; set the run-time fold flag for these. (We don't have to | |
12735 | * worry about properties folding, as that is taken care of by the swash | |
12736 | * fetching) */ | |
12737 | if (FOLD && (LOC || unicode_alternate)) | |
12738 | { | |
12739 | ANYOF_FLAGS(ret) |= ANYOF_LOC_NONBITMAP_FOLD; | |
12740 | } | |
12741 | ||
12742 | /* Some character classes are equivalent to other nodes. Such nodes take | |
12743 | * up less room and generally fewer operations to execute than ANYOF nodes. | |
12744 | * Above, we checked for and optimized into some such equivalents for | |
12745 | * certain common classes that are easy to test. Getting to this point in | |
12746 | * the code means that the class didn't get optimized there. Since this | |
12747 | * code is only executed in Pass 2, it is too late to save space--it has | |
12748 | * been allocated in Pass 1, and currently isn't given back. But turning | |
12749 | * things into an EXACTish node can allow the optimizer to join it to any | |
12750 | * adjacent such nodes. And if the class is equivalent to things like /./, | |
12751 | * expensive run-time swashes can be avoided. Now that we have more | |
12752 | * complete information, we can find things necessarily missed by the | |
12753 | * earlier code. I (khw) am not sure how much to look for here. It would | |
12754 | * be easy, but perhaps too slow, to check any candidates against all the | |
12755 | * node types they could possibly match using _invlistEQ(). */ | |
12756 | ||
12757 | if (cp_list | |
12758 | && ! unicode_alternate | |
12759 | && ! invert | |
12760 | && ! depends_list | |
12761 | && ! (ANYOF_FLAGS(ret) & ANYOF_CLASS) | |
12762 | && ! HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION) | |
12763 | { | |
12764 | UV start, end; | |
12765 | U8 op = END; /* The optimzation node-type */ | |
12766 | const char * cur_parse= RExC_parse; | |
12767 | ||
12768 | invlist_iterinit(cp_list); | |
12769 | if (! invlist_iternext(cp_list, &start, &end)) { | |
12770 | ||
12771 | /* Here, the list is empty. This happens, for example, when a | |
12772 | * Unicode property is the only thing in the character class, and | |
12773 | * it doesn't match anything. (perluniprops.pod notes such | |
12774 | * properties) */ | |
12775 | op = OPFAIL; | |
34b39fc9 | 12776 | *flagp |= HASWIDTH|SIMPLE; |
a0a06948 KW |
12777 | } |
12778 | else if (start == end) { /* The range is a single code point */ | |
12779 | if (! invlist_iternext(cp_list, &start, &end) | |
12780 | ||
12781 | /* Don't do this optimization if it would require changing | |
12782 | * the pattern to UTF-8 */ | |
12783 | && (start < 256 || UTF)) | |
12784 | { | |
12785 | /* Here, the list contains a single code point. Can optimize | |
12786 | * into an EXACT node */ | |
12787 | ||
12788 | value = start; | |
12789 | ||
12790 | if (! FOLD) { | |
12791 | op = EXACT; | |
12792 | } | |
12793 | else if (LOC) { | |
12794 | ||
12795 | /* A locale node under folding with one code point can be | |
12796 | * an EXACTFL, as its fold won't be calculated until | |
12797 | * runtime */ | |
12798 | op = EXACTFL; | |
12799 | } | |
12800 | else { | |
12801 | ||
12802 | /* Here, we are generally folding, but there is only one | |
12803 | * code point to match. If we have to, we use an EXACT | |
12804 | * node, but it would be better for joining with adjacent | |
12805 | * nodes in the optimization pass if we used the same | |
12806 | * EXACTFish node that any such are likely to be. We can | |
12807 | * do this iff the code point doesn't participate in any | |
12808 | * folds. For example, an EXACTF of a colon is the same as | |
12809 | * an EXACT one, since nothing folds to or from a colon. | |
12810 | * In the Latin1 range, being an alpha means that the | |
12811 | * character participates in a fold (except for the | |
12812 | * feminine and masculine ordinals, which I (khw) don't | |
12813 | * think are worrying about optimizing for). */ | |
12814 | if (value < 256) { | |
12815 | if (isALPHA_L1(value)) { | |
12816 | op = EXACT; | |
12817 | } | |
12818 | } | |
12819 | else { | |
12820 | if (! PL_utf8_foldable) { | |
12821 | SV* swash = swash_init("utf8", "_Perl_Any_Folds", | |
12822 | &PL_sv_undef, 1, 0); | |
12823 | PL_utf8_foldable = _get_swash_invlist(swash); | |
12824 | SvREFCNT_dec(swash); | |
12825 | } | |
12826 | if (_invlist_contains_cp(PL_utf8_foldable, value)) { | |
12827 | op = EXACT; | |
12828 | } | |
12829 | } | |
12830 | ||
12831 | /* If we haven't found the node type, above, it means we | |
12832 | * can use the prevailing one */ | |
12833 | if (op == END) { | |
12834 | op = compute_EXACTish(pRExC_state); | |
12835 | } | |
12836 | } | |
12837 | } | |
12838 | } | |
12839 | else if (start == 0) { | |
12840 | if (end == UV_MAX) { | |
12841 | op = SANY; | |
34b39fc9 KW |
12842 | *flagp |= HASWIDTH|SIMPLE; |
12843 | RExC_naughty++; | |
a0a06948 KW |
12844 | } |
12845 | else if (end == '\n' - 1 | |
12846 | && invlist_iternext(cp_list, &start, &end) | |
12847 | && start == '\n' + 1 && end == UV_MAX) | |
12848 | { | |
12849 | op = REG_ANY; | |
34b39fc9 KW |
12850 | *flagp |= HASWIDTH|SIMPLE; |
12851 | RExC_naughty++; | |
a0a06948 KW |
12852 | } |
12853 | } | |
12854 | ||
12855 | if (op != END) { | |
12856 | RExC_parse = (char *)orig_parse; | |
12857 | RExC_emit = (regnode *)orig_emit; | |
12858 | ||
12859 | ret = reg_node(pRExC_state, op); | |
12860 | ||
12861 | RExC_parse = (char *)cur_parse; | |
12862 | ||
12863 | if (PL_regkind[op] == EXACT) { | |
34b39fc9 | 12864 | alloc_maybe_populate_EXACT(pRExC_state, ret, flagp, 0, value); |
a0a06948 KW |
12865 | } |
12866 | ||
12867 | SvREFCNT_dec(listsv); | |
12868 | return ret; | |
12869 | } | |
12870 | } | |
12871 | ||
cfbb2758 | 12872 | /* Here, <cp_list> contains all the code points we can determine at |
68823f48 | 12873 | * compile time that match under all conditions. Go through it, and |
e4e94b48 | 12874 | * for things that belong in the bitmap, put them there, and delete from |
a0a06948 KW |
12875 | * <cp_list>. While we are at it, see if everything above 255 is in the |
12876 | * list, and if so, set a flag to speed up execution */ | |
94d80c53 | 12877 | ANYOF_BITMAP_ZERO(ret); |
cfbb2758 | 12878 | if (cp_list) { |
e4e94b48 | 12879 | |
e4e94b48 KW |
12880 | /* This gets set if we actually need to modify things */ |
12881 | bool change_invlist = FALSE; | |
12882 | ||
12883 | UV start, end; | |
12884 | ||
cfbb2758 KW |
12885 | /* Start looking through <cp_list> */ |
12886 | invlist_iterinit(cp_list); | |
12887 | while (invlist_iternext(cp_list, &start, &end)) { | |
e4e94b48 KW |
12888 | UV high; |
12889 | int i; | |
12890 | ||
a0a06948 KW |
12891 | if (end == UV_MAX && start <= 256) { |
12892 | ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; | |
12893 | } | |
12894 | ||
e4e94b48 | 12895 | /* Quit if are above what we should change */ |
68823f48 | 12896 | if (start > 255) { |
e4e94b48 KW |
12897 | break; |
12898 | } | |
12899 | ||
12900 | change_invlist = TRUE; | |
12901 | ||
12902 | /* Set all the bits in the range, up to the max that we are doing */ | |
68823f48 | 12903 | high = (end < 255) ? end : 255; |
e4e94b48 KW |
12904 | for (i = start; i <= (int) high; i++) { |
12905 | if (! ANYOF_BITMAP_TEST(ret, i)) { | |
12906 | ANYOF_BITMAP_SET(ret, i); | |
e4e94b48 KW |
12907 | prevvalue = value; |
12908 | value = i; | |
12909 | } | |
12910 | } | |
12911 | } | |
12912 | ||
a3e1f3a6 | 12913 | /* Done with loop; remove any code points that are in the bitmap from |
cfbb2758 | 12914 | * <cp_list> */ |
e4e94b48 | 12915 | if (change_invlist) { |
cfbb2758 | 12916 | _invlist_subtract(cp_list, PL_Latin1, &cp_list); |
e4e94b48 KW |
12917 | } |
12918 | ||
12919 | /* If have completely emptied it, remove it completely */ | |
29f51c6b | 12920 | if (_invlist_len(cp_list) == 0) { |
cfbb2758 KW |
12921 | SvREFCNT_dec(cp_list); |
12922 | cp_list = NULL; | |
e4e94b48 KW |
12923 | } |
12924 | } | |
dc415b4a | 12925 | |
520939ba KW |
12926 | if (invert) { |
12927 | ANYOF_FLAGS(ret) |= ANYOF_INVERT; | |
12928 | } | |
12929 | ||
a0a06948 KW |
12930 | /* Here, the bitmap has been populated with all the Latin1 code points that |
12931 | * always match. Can now add to the overall list those that match only | |
12932 | * when the target string is UTF-8 (<depends_list>). */ | |
68823f48 | 12933 | if (depends_list) { |
cfbb2758 KW |
12934 | if (cp_list) { |
12935 | _invlist_union(cp_list, depends_list, &cp_list); | |
68823f48 KW |
12936 | SvREFCNT_dec(depends_list); |
12937 | } | |
12938 | else { | |
cfbb2758 | 12939 | cp_list = depends_list; |
68823f48 KW |
12940 | } |
12941 | } | |
12942 | ||
dc415b4a KW |
12943 | /* If there is a swash and more than one element, we can't use the swash in |
12944 | * the optimization below. */ | |
12945 | if (swash && element_count > 1) { | |
12946 | SvREFCNT_dec(swash); | |
12947 | swash = NULL; | |
12948 | } | |
a0a06948 | 12949 | |
cfbb2758 | 12950 | if (! cp_list |
66c69124 | 12951 | && ! HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION |
c16787fd KW |
12952 | && ! unicode_alternate) |
12953 | { | |
137165a6 KW |
12954 | ARG_SET(ret, ANYOF_NONBITMAP_EMPTY); |
12955 | SvREFCNT_dec(listsv); | |
12956 | SvREFCNT_dec(unicode_alternate); | |
12957 | } | |
12958 | else { | |
0bd1039c KW |
12959 | /* av[0] stores the character class description in its textual form: |
12960 | * used later (regexec.c:Perl_regclass_swash()) to initialize the | |
12961 | * appropriate swash, and is also useful for dumping the regnode. | |
12962 | * av[1] if NULL, is a placeholder to later contain the swash computed | |
12963 | * from av[0]. But if no further computation need be done, the | |
12964 | * swash is stored there now. | |
12965 | * av[2] stores the multicharacter foldings, used later in | |
12966 | * regexec.c:S_reginclass(). | |
cfbb2758 | 12967 | * av[3] stores the cp_list inversion list for use in addition or |
26cdf4c4 | 12968 | * instead of av[0]; used only if av[1] is NULL |
0bd1039c | 12969 | * av[4] is set if any component of the class is from a user-defined |
26cdf4c4 | 12970 | * property; used only if av[1] is NULL */ |
097eb12c | 12971 | AV * const av = newAV(); |
ffc61ed2 | 12972 | SV *rv; |
0bd1039c | 12973 | |
66c69124 KW |
12974 | av_store(av, 0, (HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION) |
12975 | ? listsv | |
12976 | : &PL_sv_undef); | |
88d45d28 KW |
12977 | if (swash) { |
12978 | av_store(av, 1, swash); | |
cfbb2758 | 12979 | SvREFCNT_dec(cp_list); |
88d45d28 KW |
12980 | } |
12981 | else { | |
12982 | av_store(av, 1, NULL); | |
cfbb2758 KW |
12983 | if (cp_list) { |
12984 | av_store(av, 3, cp_list); | |
dc415b4a | 12985 | av_store(av, 4, newSVuv(has_user_defined_property)); |
c16787fd | 12986 | } |
88d45d28 | 12987 | } |
827f5bb8 KW |
12988 | |
12989 | /* Store any computed multi-char folds only if we are allowing | |
12990 | * them */ | |
12991 | if (allow_full_fold) { | |
7b4a7e58 KW |
12992 | av_store(av, 2, MUTABLE_SV(unicode_alternate)); |
12993 | if (unicode_alternate) { /* This node is variable length */ | |
12994 | OP(ret) = ANYOFV; | |
12995 | } | |
827f5bb8 KW |
12996 | } |
12997 | else { | |
12998 | av_store(av, 2, NULL); | |
12999 | } | |
ad64d0ec | 13000 | rv = newRV_noinc(MUTABLE_SV(av)); |
19860706 | 13001 | n = add_data(pRExC_state, 1, "s"); |
f8fc2ecf | 13002 | RExC_rxi->data->data[n] = (void*)rv; |
ffc61ed2 | 13003 | ARG_SET(ret, n); |
a0ed51b3 | 13004 | } |
34b39fc9 KW |
13005 | |
13006 | *flagp |= HASWIDTH|SIMPLE; | |
a0ed51b3 LW |
13007 | return ret; |
13008 | } | |
66c69124 | 13009 | #undef HAS_NONLOCALE_RUNTIME_PROPERTY_DEFINITION |
89836f1f | 13010 | |
a0ed51b3 | 13011 | |
bcdf7404 YO |
13012 | /* reg_skipcomment() |
13013 | ||
13014 | Absorbs an /x style # comments from the input stream. | |
13015 | Returns true if there is more text remaining in the stream. | |
13016 | Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment | |
13017 | terminates the pattern without including a newline. | |
13018 | ||
13019 | Note its the callers responsibility to ensure that we are | |
13020 | actually in /x mode | |
13021 | ||
13022 | */ | |
13023 | ||
13024 | STATIC bool | |
13025 | S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state) | |
13026 | { | |
13027 | bool ended = 0; | |
7918f24d NC |
13028 | |
13029 | PERL_ARGS_ASSERT_REG_SKIPCOMMENT; | |
13030 | ||
bcdf7404 YO |
13031 | while (RExC_parse < RExC_end) |
13032 | if (*RExC_parse++ == '\n') { | |
13033 | ended = 1; | |
13034 | break; | |
13035 | } | |
13036 | if (!ended) { | |
13037 | /* we ran off the end of the pattern without ending | |
13038 | the comment, so we have to add an \n when wrapping */ | |
13039 | RExC_seen |= REG_SEEN_RUN_ON_COMMENT; | |
13040 | return 0; | |
13041 | } else | |
13042 | return 1; | |
13043 | } | |
13044 | ||
13045 | /* nextchar() | |
13046 | ||
3b753521 | 13047 | Advances the parse position, and optionally absorbs |
bcdf7404 YO |
13048 | "whitespace" from the inputstream. |
13049 | ||
13050 | Without /x "whitespace" means (?#...) style comments only, | |
13051 | with /x this means (?#...) and # comments and whitespace proper. | |
13052 | ||
13053 | Returns the RExC_parse point from BEFORE the scan occurs. | |
13054 | ||
13055 | This is the /x friendly way of saying RExC_parse++. | |
13056 | */ | |
13057 | ||
76e3520e | 13058 | STATIC char* |
830247a4 | 13059 | S_nextchar(pTHX_ RExC_state_t *pRExC_state) |
a0d0e21e | 13060 | { |
097eb12c | 13061 | char* const retval = RExC_parse++; |
a0d0e21e | 13062 | |
7918f24d NC |
13063 | PERL_ARGS_ASSERT_NEXTCHAR; |
13064 | ||
4633a7c4 | 13065 | for (;;) { |
d224c965 KW |
13066 | if (RExC_end - RExC_parse >= 3 |
13067 | && *RExC_parse == '(' | |
13068 | && RExC_parse[1] == '?' | |
13069 | && RExC_parse[2] == '#') | |
13070 | { | |
e994fd66 AE |
13071 | while (*RExC_parse != ')') { |
13072 | if (RExC_parse == RExC_end) | |
13073 | FAIL("Sequence (?#... not terminated"); | |
830247a4 | 13074 | RExC_parse++; |
e994fd66 | 13075 | } |
830247a4 | 13076 | RExC_parse++; |
4633a7c4 LW |
13077 | continue; |
13078 | } | |
bbe252da | 13079 | if (RExC_flags & RXf_PMf_EXTENDED) { |
830247a4 IZ |
13080 | if (isSPACE(*RExC_parse)) { |
13081 | RExC_parse++; | |
748a9306 LW |
13082 | continue; |
13083 | } | |
830247a4 | 13084 | else if (*RExC_parse == '#') { |
bcdf7404 YO |
13085 | if ( reg_skipcomment( pRExC_state ) ) |
13086 | continue; | |
748a9306 | 13087 | } |
748a9306 | 13088 | } |
4633a7c4 | 13089 | return retval; |
a0d0e21e | 13090 | } |
a687059c LW |
13091 | } |
13092 | ||
13093 | /* | |
c277df42 | 13094 | - reg_node - emit a node |
a0d0e21e | 13095 | */ |
76e3520e | 13096 | STATIC regnode * /* Location. */ |
830247a4 | 13097 | S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op) |
a687059c | 13098 | { |
97aff369 | 13099 | dVAR; |
eb578fdb | 13100 | regnode *ptr; |
504618e9 | 13101 | regnode * const ret = RExC_emit; |
07be1b83 | 13102 | GET_RE_DEBUG_FLAGS_DECL; |
a687059c | 13103 | |
7918f24d NC |
13104 | PERL_ARGS_ASSERT_REG_NODE; |
13105 | ||
c277df42 | 13106 | if (SIZE_ONLY) { |
830247a4 IZ |
13107 | SIZE_ALIGN(RExC_size); |
13108 | RExC_size += 1; | |
a0d0e21e LW |
13109 | return(ret); |
13110 | } | |
3b57cd43 | 13111 | if (RExC_emit >= RExC_emit_bound) |
5637ef5b NC |
13112 | Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d, %p>=%p", |
13113 | op, RExC_emit, RExC_emit_bound); | |
3b57cd43 | 13114 | |
c277df42 | 13115 | NODE_ALIGN_FILL(ret); |
a0d0e21e | 13116 | ptr = ret; |
c277df42 | 13117 | FILL_ADVANCE_NODE(ptr, op); |
7122b237 | 13118 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 13119 | if (RExC_offsets) { /* MJD */ |
07be1b83 | 13120 | MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n", |
fac92740 | 13121 | "reg_node", __LINE__, |
13d6edb4 | 13122 | PL_reg_name[op], |
07be1b83 YO |
13123 | (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] |
13124 | ? "Overwriting end of array!\n" : "OK", | |
13125 | (UV)(RExC_emit - RExC_emit_start), | |
13126 | (UV)(RExC_parse - RExC_start), | |
13127 | (UV)RExC_offsets[0])); | |
ccb2c380 | 13128 | Set_Node_Offset(RExC_emit, RExC_parse + (op == END)); |
fac92740 | 13129 | } |
7122b237 | 13130 | #endif |
830247a4 | 13131 | RExC_emit = ptr; |
a0d0e21e | 13132 | return(ret); |
a687059c LW |
13133 | } |
13134 | ||
13135 | /* | |
a0d0e21e LW |
13136 | - reganode - emit a node with an argument |
13137 | */ | |
76e3520e | 13138 | STATIC regnode * /* Location. */ |
830247a4 | 13139 | S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg) |
fe14fcc3 | 13140 | { |
97aff369 | 13141 | dVAR; |
eb578fdb | 13142 | regnode *ptr; |
504618e9 | 13143 | regnode * const ret = RExC_emit; |
07be1b83 | 13144 | GET_RE_DEBUG_FLAGS_DECL; |
fe14fcc3 | 13145 | |
7918f24d NC |
13146 | PERL_ARGS_ASSERT_REGANODE; |
13147 | ||
c277df42 | 13148 | if (SIZE_ONLY) { |
830247a4 IZ |
13149 | SIZE_ALIGN(RExC_size); |
13150 | RExC_size += 2; | |
6bda09f9 YO |
13151 | /* |
13152 | We can't do this: | |
13153 | ||
13154 | assert(2==regarglen[op]+1); | |
686b73d4 | 13155 | |
6bda09f9 YO |
13156 | Anything larger than this has to allocate the extra amount. |
13157 | If we changed this to be: | |
13158 | ||
13159 | RExC_size += (1 + regarglen[op]); | |
13160 | ||
13161 | then it wouldn't matter. Its not clear what side effect | |
13162 | might come from that so its not done so far. | |
13163 | -- dmq | |
13164 | */ | |
a0d0e21e LW |
13165 | return(ret); |
13166 | } | |
3b57cd43 | 13167 | if (RExC_emit >= RExC_emit_bound) |
5637ef5b NC |
13168 | Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d, %p>=%p", |
13169 | op, RExC_emit, RExC_emit_bound); | |
3b57cd43 | 13170 | |
c277df42 | 13171 | NODE_ALIGN_FILL(ret); |
a0d0e21e | 13172 | ptr = ret; |
c277df42 | 13173 | FILL_ADVANCE_NODE_ARG(ptr, op, arg); |
7122b237 | 13174 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 13175 | if (RExC_offsets) { /* MJD */ |
07be1b83 | 13176 | MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n", |
fac92740 | 13177 | "reganode", |
ccb2c380 | 13178 | __LINE__, |
13d6edb4 | 13179 | PL_reg_name[op], |
07be1b83 | 13180 | (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ? |
fac92740 | 13181 | "Overwriting end of array!\n" : "OK", |
07be1b83 YO |
13182 | (UV)(RExC_emit - RExC_emit_start), |
13183 | (UV)(RExC_parse - RExC_start), | |
13184 | (UV)RExC_offsets[0])); | |
ccb2c380 | 13185 | Set_Cur_Node_Offset; |
fac92740 | 13186 | } |
7122b237 | 13187 | #endif |
830247a4 | 13188 | RExC_emit = ptr; |
a0d0e21e | 13189 | return(ret); |
fe14fcc3 LW |
13190 | } |
13191 | ||
13192 | /* | |
cd439c50 | 13193 | - reguni - emit (if appropriate) a Unicode character |
a0ed51b3 | 13194 | */ |
71207a34 AL |
13195 | STATIC STRLEN |
13196 | S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s) | |
a0ed51b3 | 13197 | { |
97aff369 | 13198 | dVAR; |
7918f24d NC |
13199 | |
13200 | PERL_ARGS_ASSERT_REGUNI; | |
13201 | ||
71207a34 | 13202 | return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s); |
a0ed51b3 LW |
13203 | } |
13204 | ||
13205 | /* | |
a0d0e21e LW |
13206 | - reginsert - insert an operator in front of already-emitted operand |
13207 | * | |
13208 | * Means relocating the operand. | |
13209 | */ | |
76e3520e | 13210 | STATIC void |
6bda09f9 | 13211 | S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth) |
a687059c | 13212 | { |
97aff369 | 13213 | dVAR; |
eb578fdb KW |
13214 | regnode *src; |
13215 | regnode *dst; | |
13216 | regnode *place; | |
504618e9 | 13217 | const int offset = regarglen[(U8)op]; |
6bda09f9 | 13218 | const int size = NODE_STEP_REGNODE + offset; |
07be1b83 | 13219 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
13220 | |
13221 | PERL_ARGS_ASSERT_REGINSERT; | |
def51078 | 13222 | PERL_UNUSED_ARG(depth); |
22c35a8c | 13223 | /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */ |
13d6edb4 | 13224 | DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]); |
c277df42 | 13225 | if (SIZE_ONLY) { |
6bda09f9 | 13226 | RExC_size += size; |
a0d0e21e LW |
13227 | return; |
13228 | } | |
a687059c | 13229 | |
830247a4 | 13230 | src = RExC_emit; |
6bda09f9 | 13231 | RExC_emit += size; |
830247a4 | 13232 | dst = RExC_emit; |
40d049e4 | 13233 | if (RExC_open_parens) { |
6bda09f9 | 13234 | int paren; |
3b57cd43 | 13235 | /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/ |
6bda09f9 | 13236 | for ( paren=0 ; paren < RExC_npar ; paren++ ) { |
40d049e4 | 13237 | if ( RExC_open_parens[paren] >= opnd ) { |
3b57cd43 | 13238 | /*DEBUG_PARSE_FMT("open"," - %d",size);*/ |
40d049e4 YO |
13239 | RExC_open_parens[paren] += size; |
13240 | } else { | |
3b57cd43 | 13241 | /*DEBUG_PARSE_FMT("open"," - %s","ok");*/ |
40d049e4 YO |
13242 | } |
13243 | if ( RExC_close_parens[paren] >= opnd ) { | |
3b57cd43 | 13244 | /*DEBUG_PARSE_FMT("close"," - %d",size);*/ |
40d049e4 YO |
13245 | RExC_close_parens[paren] += size; |
13246 | } else { | |
3b57cd43 | 13247 | /*DEBUG_PARSE_FMT("close"," - %s","ok");*/ |
40d049e4 YO |
13248 | } |
13249 | } | |
6bda09f9 | 13250 | } |
40d049e4 | 13251 | |
fac92740 | 13252 | while (src > opnd) { |
c277df42 | 13253 | StructCopy(--src, --dst, regnode); |
7122b237 | 13254 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 13255 | if (RExC_offsets) { /* MJD 20010112 */ |
07be1b83 | 13256 | MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n", |
fac92740 | 13257 | "reg_insert", |
ccb2c380 | 13258 | __LINE__, |
13d6edb4 | 13259 | PL_reg_name[op], |
07be1b83 YO |
13260 | (UV)(dst - RExC_emit_start) > RExC_offsets[0] |
13261 | ? "Overwriting end of array!\n" : "OK", | |
13262 | (UV)(src - RExC_emit_start), | |
13263 | (UV)(dst - RExC_emit_start), | |
13264 | (UV)RExC_offsets[0])); | |
ccb2c380 MP |
13265 | Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src)); |
13266 | Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src)); | |
fac92740 | 13267 | } |
7122b237 | 13268 | #endif |
fac92740 MJD |
13269 | } |
13270 | ||
a0d0e21e LW |
13271 | |
13272 | place = opnd; /* Op node, where operand used to be. */ | |
7122b237 | 13273 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 13274 | if (RExC_offsets) { /* MJD */ |
07be1b83 | 13275 | MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n", |
fac92740 | 13276 | "reginsert", |
ccb2c380 | 13277 | __LINE__, |
13d6edb4 | 13278 | PL_reg_name[op], |
07be1b83 | 13279 | (UV)(place - RExC_emit_start) > RExC_offsets[0] |
fac92740 | 13280 | ? "Overwriting end of array!\n" : "OK", |
07be1b83 YO |
13281 | (UV)(place - RExC_emit_start), |
13282 | (UV)(RExC_parse - RExC_start), | |
786e8c11 | 13283 | (UV)RExC_offsets[0])); |
ccb2c380 | 13284 | Set_Node_Offset(place, RExC_parse); |
45948336 | 13285 | Set_Node_Length(place, 1); |
fac92740 | 13286 | } |
7122b237 | 13287 | #endif |
c277df42 IZ |
13288 | src = NEXTOPER(place); |
13289 | FILL_ADVANCE_NODE(place, op); | |
13290 | Zero(src, offset, regnode); | |
a687059c LW |
13291 | } |
13292 | ||
13293 | /* | |
c277df42 | 13294 | - regtail - set the next-pointer at the end of a node chain of p to val. |
3dab1dad | 13295 | - SEE ALSO: regtail_study |
a0d0e21e | 13296 | */ |
097eb12c | 13297 | /* TODO: All three parms should be const */ |
76e3520e | 13298 | STATIC void |
3dab1dad | 13299 | S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth) |
a687059c | 13300 | { |
97aff369 | 13301 | dVAR; |
eb578fdb | 13302 | regnode *scan; |
72f13be8 | 13303 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
13304 | |
13305 | PERL_ARGS_ASSERT_REGTAIL; | |
f9049ba1 SP |
13306 | #ifndef DEBUGGING |
13307 | PERL_UNUSED_ARG(depth); | |
13308 | #endif | |
a0d0e21e | 13309 | |
c277df42 | 13310 | if (SIZE_ONLY) |
a0d0e21e LW |
13311 | return; |
13312 | ||
13313 | /* Find last node. */ | |
13314 | scan = p; | |
13315 | for (;;) { | |
504618e9 | 13316 | regnode * const temp = regnext(scan); |
3dab1dad YO |
13317 | DEBUG_PARSE_r({ |
13318 | SV * const mysv=sv_newmortal(); | |
13319 | DEBUG_PARSE_MSG((scan==p ? "tail" : "")); | |
13320 | regprop(RExC_rx, mysv, scan); | |
eaf3ca90 YO |
13321 | PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n", |
13322 | SvPV_nolen_const(mysv), REG_NODE_NUM(scan), | |
13323 | (temp == NULL ? "->" : ""), | |
13d6edb4 | 13324 | (temp == NULL ? PL_reg_name[OP(val)] : "") |
eaf3ca90 | 13325 | ); |
3dab1dad YO |
13326 | }); |
13327 | if (temp == NULL) | |
13328 | break; | |
13329 | scan = temp; | |
13330 | } | |
13331 | ||
13332 | if (reg_off_by_arg[OP(scan)]) { | |
13333 | ARG_SET(scan, val - scan); | |
13334 | } | |
13335 | else { | |
13336 | NEXT_OFF(scan) = val - scan; | |
13337 | } | |
13338 | } | |
13339 | ||
07be1b83 | 13340 | #ifdef DEBUGGING |
3dab1dad YO |
13341 | /* |
13342 | - regtail_study - set the next-pointer at the end of a node chain of p to val. | |
13343 | - Look for optimizable sequences at the same time. | |
13344 | - currently only looks for EXACT chains. | |
07be1b83 | 13345 | |
486ec47a | 13346 | This is experimental code. The idea is to use this routine to perform |
07be1b83 YO |
13347 | in place optimizations on branches and groups as they are constructed, |
13348 | with the long term intention of removing optimization from study_chunk so | |
13349 | that it is purely analytical. | |
13350 | ||
13351 | Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used | |
13352 | to control which is which. | |
13353 | ||
3dab1dad YO |
13354 | */ |
13355 | /* TODO: All four parms should be const */ | |
07be1b83 | 13356 | |
3dab1dad YO |
13357 | STATIC U8 |
13358 | S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth) | |
13359 | { | |
13360 | dVAR; | |
eb578fdb | 13361 | regnode *scan; |
07be1b83 YO |
13362 | U8 exact = PSEUDO; |
13363 | #ifdef EXPERIMENTAL_INPLACESCAN | |
13364 | I32 min = 0; | |
13365 | #endif | |
3dab1dad YO |
13366 | GET_RE_DEBUG_FLAGS_DECL; |
13367 | ||
7918f24d NC |
13368 | PERL_ARGS_ASSERT_REGTAIL_STUDY; |
13369 | ||
07be1b83 | 13370 | |
3dab1dad YO |
13371 | if (SIZE_ONLY) |
13372 | return exact; | |
13373 | ||
13374 | /* Find last node. */ | |
13375 | ||
13376 | scan = p; | |
13377 | for (;;) { | |
13378 | regnode * const temp = regnext(scan); | |
07be1b83 | 13379 | #ifdef EXPERIMENTAL_INPLACESCAN |
f758bddf KW |
13380 | if (PL_regkind[OP(scan)] == EXACT) { |
13381 | bool has_exactf_sharp_s; /* Unexamined in this routine */ | |
13382 | if (join_exact(pRExC_state,scan,&min, &has_exactf_sharp_s, 1,val,depth+1)) | |
07be1b83 | 13383 | return EXACT; |
f758bddf | 13384 | } |
07be1b83 | 13385 | #endif |
3dab1dad YO |
13386 | if ( exact ) { |
13387 | switch (OP(scan)) { | |
13388 | case EXACT: | |
13389 | case EXACTF: | |
2f7f8cb1 | 13390 | case EXACTFA: |
2c2b7f86 | 13391 | case EXACTFU: |
3c760661 | 13392 | case EXACTFU_SS: |
fab2782b | 13393 | case EXACTFU_TRICKYFOLD: |
3dab1dad YO |
13394 | case EXACTFL: |
13395 | if( exact == PSEUDO ) | |
13396 | exact= OP(scan); | |
07be1b83 YO |
13397 | else if ( exact != OP(scan) ) |
13398 | exact= 0; | |
3dab1dad YO |
13399 | case NOTHING: |
13400 | break; | |
13401 | default: | |
13402 | exact= 0; | |
13403 | } | |
13404 | } | |
13405 | DEBUG_PARSE_r({ | |
13406 | SV * const mysv=sv_newmortal(); | |
13407 | DEBUG_PARSE_MSG((scan==p ? "tsdy" : "")); | |
13408 | regprop(RExC_rx, mysv, scan); | |
eaf3ca90 | 13409 | PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n", |
3dab1dad | 13410 | SvPV_nolen_const(mysv), |
eaf3ca90 | 13411 | REG_NODE_NUM(scan), |
13d6edb4 | 13412 | PL_reg_name[exact]); |
3dab1dad | 13413 | }); |
a0d0e21e LW |
13414 | if (temp == NULL) |
13415 | break; | |
13416 | scan = temp; | |
13417 | } | |
07be1b83 YO |
13418 | DEBUG_PARSE_r({ |
13419 | SV * const mysv_val=sv_newmortal(); | |
13420 | DEBUG_PARSE_MSG(""); | |
13421 | regprop(RExC_rx, mysv_val, val); | |
70685ca0 JH |
13422 | PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n", |
13423 | SvPV_nolen_const(mysv_val), | |
13424 | (IV)REG_NODE_NUM(val), | |
13425 | (IV)(val - scan) | |
07be1b83 YO |
13426 | ); |
13427 | }); | |
c277df42 IZ |
13428 | if (reg_off_by_arg[OP(scan)]) { |
13429 | ARG_SET(scan, val - scan); | |
a0ed51b3 LW |
13430 | } |
13431 | else { | |
c277df42 IZ |
13432 | NEXT_OFF(scan) = val - scan; |
13433 | } | |
3dab1dad YO |
13434 | |
13435 | return exact; | |
a687059c | 13436 | } |
07be1b83 | 13437 | #endif |
a687059c LW |
13438 | |
13439 | /* | |
fd181c75 | 13440 | - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form |
a687059c | 13441 | */ |
f7819f85 | 13442 | #ifdef DEBUGGING |
c33269f7 | 13443 | static void |
7918f24d NC |
13444 | S_regdump_extflags(pTHX_ const char *lead, const U32 flags) |
13445 | { | |
f7819f85 A |
13446 | int bit; |
13447 | int set=0; | |
a62b1201 | 13448 | regex_charset cs; |
7918f24d | 13449 | |
f7819f85 A |
13450 | for (bit=0; bit<32; bit++) { |
13451 | if (flags & (1<<bit)) { | |
a62b1201 KW |
13452 | if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */ |
13453 | continue; | |
13454 | } | |
f7819f85 A |
13455 | if (!set++ && lead) |
13456 | PerlIO_printf(Perl_debug_log, "%s",lead); | |
13457 | PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]); | |
13458 | } | |
13459 | } | |
a62b1201 KW |
13460 | if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) { |
13461 | if (!set++ && lead) { | |
13462 | PerlIO_printf(Perl_debug_log, "%s",lead); | |
13463 | } | |
13464 | switch (cs) { | |
13465 | case REGEX_UNICODE_CHARSET: | |
13466 | PerlIO_printf(Perl_debug_log, "UNICODE"); | |
13467 | break; | |
13468 | case REGEX_LOCALE_CHARSET: | |
13469 | PerlIO_printf(Perl_debug_log, "LOCALE"); | |
13470 | break; | |
cfaf538b KW |
13471 | case REGEX_ASCII_RESTRICTED_CHARSET: |
13472 | PerlIO_printf(Perl_debug_log, "ASCII-RESTRICTED"); | |
13473 | break; | |
2f7f8cb1 KW |
13474 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
13475 | PerlIO_printf(Perl_debug_log, "ASCII-MORE_RESTRICTED"); | |
13476 | break; | |
a62b1201 KW |
13477 | default: |
13478 | PerlIO_printf(Perl_debug_log, "UNKNOWN CHARACTER SET"); | |
13479 | break; | |
13480 | } | |
13481 | } | |
f7819f85 A |
13482 | if (lead) { |
13483 | if (set) | |
13484 | PerlIO_printf(Perl_debug_log, "\n"); | |
13485 | else | |
13486 | PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead); | |
13487 | } | |
13488 | } | |
13489 | #endif | |
13490 | ||
a687059c | 13491 | void |
097eb12c | 13492 | Perl_regdump(pTHX_ const regexp *r) |
a687059c | 13493 | { |
35ff7856 | 13494 | #ifdef DEBUGGING |
97aff369 | 13495 | dVAR; |
c445ea15 | 13496 | SV * const sv = sv_newmortal(); |
ab3bbdeb | 13497 | SV *dsv= sv_newmortal(); |
f8fc2ecf | 13498 | RXi_GET_DECL(r,ri); |
f7819f85 | 13499 | GET_RE_DEBUG_FLAGS_DECL; |
a687059c | 13500 | |
7918f24d NC |
13501 | PERL_ARGS_ASSERT_REGDUMP; |
13502 | ||
f8fc2ecf | 13503 | (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0); |
a0d0e21e LW |
13504 | |
13505 | /* Header fields of interest. */ | |
ab3bbdeb YO |
13506 | if (r->anchored_substr) { |
13507 | RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr), | |
13508 | RE_SV_DUMPLEN(r->anchored_substr), 30); | |
7b0972df | 13509 | PerlIO_printf(Perl_debug_log, |
ab3bbdeb YO |
13510 | "anchored %s%s at %"IVdf" ", |
13511 | s, RE_SV_TAIL(r->anchored_substr), | |
7b0972df | 13512 | (IV)r->anchored_offset); |
ab3bbdeb YO |
13513 | } else if (r->anchored_utf8) { |
13514 | RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8), | |
13515 | RE_SV_DUMPLEN(r->anchored_utf8), 30); | |
33b8afdf | 13516 | PerlIO_printf(Perl_debug_log, |
ab3bbdeb YO |
13517 | "anchored utf8 %s%s at %"IVdf" ", |
13518 | s, RE_SV_TAIL(r->anchored_utf8), | |
33b8afdf | 13519 | (IV)r->anchored_offset); |
ab3bbdeb YO |
13520 | } |
13521 | if (r->float_substr) { | |
13522 | RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr), | |
13523 | RE_SV_DUMPLEN(r->float_substr), 30); | |
7b0972df | 13524 | PerlIO_printf(Perl_debug_log, |
ab3bbdeb YO |
13525 | "floating %s%s at %"IVdf"..%"UVuf" ", |
13526 | s, RE_SV_TAIL(r->float_substr), | |
7b0972df | 13527 | (IV)r->float_min_offset, (UV)r->float_max_offset); |
ab3bbdeb YO |
13528 | } else if (r->float_utf8) { |
13529 | RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8), | |
13530 | RE_SV_DUMPLEN(r->float_utf8), 30); | |
33b8afdf | 13531 | PerlIO_printf(Perl_debug_log, |
ab3bbdeb YO |
13532 | "floating utf8 %s%s at %"IVdf"..%"UVuf" ", |
13533 | s, RE_SV_TAIL(r->float_utf8), | |
33b8afdf | 13534 | (IV)r->float_min_offset, (UV)r->float_max_offset); |
ab3bbdeb | 13535 | } |
33b8afdf | 13536 | if (r->check_substr || r->check_utf8) |
b81d288d | 13537 | PerlIO_printf(Perl_debug_log, |
10edeb5d JH |
13538 | (const char *) |
13539 | (r->check_substr == r->float_substr | |
13540 | && r->check_utf8 == r->float_utf8 | |
13541 | ? "(checking floating" : "(checking anchored")); | |
bbe252da | 13542 | if (r->extflags & RXf_NOSCAN) |
c277df42 | 13543 | PerlIO_printf(Perl_debug_log, " noscan"); |
bbe252da | 13544 | if (r->extflags & RXf_CHECK_ALL) |
c277df42 | 13545 | PerlIO_printf(Perl_debug_log, " isall"); |
33b8afdf | 13546 | if (r->check_substr || r->check_utf8) |
c277df42 IZ |
13547 | PerlIO_printf(Perl_debug_log, ") "); |
13548 | ||
f8fc2ecf YO |
13549 | if (ri->regstclass) { |
13550 | regprop(r, sv, ri->regstclass); | |
1de06328 | 13551 | PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv)); |
46fc3d4c | 13552 | } |
bbe252da | 13553 | if (r->extflags & RXf_ANCH) { |
774d564b | 13554 | PerlIO_printf(Perl_debug_log, "anchored"); |
bbe252da | 13555 | if (r->extflags & RXf_ANCH_BOL) |
774d564b | 13556 | PerlIO_printf(Perl_debug_log, "(BOL)"); |
bbe252da | 13557 | if (r->extflags & RXf_ANCH_MBOL) |
c277df42 | 13558 | PerlIO_printf(Perl_debug_log, "(MBOL)"); |
bbe252da | 13559 | if (r->extflags & RXf_ANCH_SBOL) |
cad2e5aa | 13560 | PerlIO_printf(Perl_debug_log, "(SBOL)"); |
bbe252da | 13561 | if (r->extflags & RXf_ANCH_GPOS) |
774d564b | 13562 | PerlIO_printf(Perl_debug_log, "(GPOS)"); |
13563 | PerlIO_putc(Perl_debug_log, ' '); | |
13564 | } | |
bbe252da | 13565 | if (r->extflags & RXf_GPOS_SEEN) |
70685ca0 | 13566 | PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs); |
bbe252da | 13567 | if (r->intflags & PREGf_SKIP) |
760ac839 | 13568 | PerlIO_printf(Perl_debug_log, "plus "); |
bbe252da | 13569 | if (r->intflags & PREGf_IMPLICIT) |
760ac839 | 13570 | PerlIO_printf(Perl_debug_log, "implicit "); |
70685ca0 | 13571 | PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen); |
bbe252da | 13572 | if (r->extflags & RXf_EVAL_SEEN) |
ce862d02 | 13573 | PerlIO_printf(Perl_debug_log, "with eval "); |
760ac839 | 13574 | PerlIO_printf(Perl_debug_log, "\n"); |
f7819f85 | 13575 | DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags)); |
65e66c80 | 13576 | #else |
7918f24d | 13577 | PERL_ARGS_ASSERT_REGDUMP; |
96a5add6 | 13578 | PERL_UNUSED_CONTEXT; |
65e66c80 | 13579 | PERL_UNUSED_ARG(r); |
17c3b450 | 13580 | #endif /* DEBUGGING */ |
a687059c LW |
13581 | } |
13582 | ||
13583 | /* | |
a0d0e21e LW |
13584 | - regprop - printable representation of opcode |
13585 | */ | |
3339dfd8 YO |
13586 | #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \ |
13587 | STMT_START { \ | |
13588 | if (do_sep) { \ | |
13589 | Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \ | |
13590 | if (flags & ANYOF_INVERT) \ | |
13591 | /*make sure the invert info is in each */ \ | |
13592 | sv_catpvs(sv, "^"); \ | |
13593 | do_sep = 0; \ | |
13594 | } \ | |
13595 | } STMT_END | |
13596 | ||
46fc3d4c | 13597 | void |
32fc9b6a | 13598 | Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o) |
a687059c | 13599 | { |
35ff7856 | 13600 | #ifdef DEBUGGING |
97aff369 | 13601 | dVAR; |
eb578fdb | 13602 | int k; |
89e1cabd KW |
13603 | |
13604 | /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */ | |
13605 | static const char * const anyofs[] = { | |
13606 | "\\w", | |
13607 | "\\W", | |
13608 | "\\s", | |
13609 | "\\S", | |
13610 | "\\d", | |
13611 | "\\D", | |
13612 | "[:alnum:]", | |
13613 | "[:^alnum:]", | |
13614 | "[:alpha:]", | |
13615 | "[:^alpha:]", | |
13616 | "[:ascii:]", | |
13617 | "[:^ascii:]", | |
13618 | "[:cntrl:]", | |
13619 | "[:^cntrl:]", | |
13620 | "[:graph:]", | |
13621 | "[:^graph:]", | |
13622 | "[:lower:]", | |
13623 | "[:^lower:]", | |
13624 | "[:print:]", | |
13625 | "[:^print:]", | |
13626 | "[:punct:]", | |
13627 | "[:^punct:]", | |
13628 | "[:upper:]", | |
13629 | "[:^upper:]", | |
13630 | "[:xdigit:]", | |
13631 | "[:^xdigit:]", | |
13632 | "[:space:]", | |
13633 | "[:^space:]", | |
13634 | "[:blank:]", | |
13635 | "[:^blank:]" | |
13636 | }; | |
f8fc2ecf | 13637 | RXi_GET_DECL(prog,progi); |
1de06328 | 13638 | GET_RE_DEBUG_FLAGS_DECL; |
f8fc2ecf | 13639 | |
7918f24d | 13640 | PERL_ARGS_ASSERT_REGPROP; |
a0d0e21e | 13641 | |
76f68e9b | 13642 | sv_setpvs(sv, ""); |
8aa23a47 | 13643 | |
03363afd | 13644 | if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */ |
830247a4 IZ |
13645 | /* It would be nice to FAIL() here, but this may be called from |
13646 | regexec.c, and it would be hard to supply pRExC_state. */ | |
a5ca303d | 13647 | Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX); |
13d6edb4 | 13648 | sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */ |
9b155405 | 13649 | |
3dab1dad | 13650 | k = PL_regkind[OP(o)]; |
9b155405 | 13651 | |
2a782b5b | 13652 | if (k == EXACT) { |
f92a2122 | 13653 | sv_catpvs(sv, " "); |
ab3bbdeb YO |
13654 | /* Using is_utf8_string() (via PERL_PV_UNI_DETECT) |
13655 | * is a crude hack but it may be the best for now since | |
13656 | * we have no flag "this EXACTish node was UTF-8" | |
13657 | * --jhi */ | |
f92a2122 NC |
13658 | pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1], |
13659 | PERL_PV_ESCAPE_UNI_DETECT | | |
c89df6cf | 13660 | PERL_PV_ESCAPE_NONASCII | |
f92a2122 NC |
13661 | PERL_PV_PRETTY_ELLIPSES | |
13662 | PERL_PV_PRETTY_LTGT | | |
13663 | PERL_PV_PRETTY_NOCLEAR | |
13664 | ); | |
bb263b4e | 13665 | } else if (k == TRIE) { |
3dab1dad | 13666 | /* print the details of the trie in dumpuntil instead, as |
f8fc2ecf | 13667 | * progi->data isn't available here */ |
1de06328 | 13668 | const char op = OP(o); |
647f639f | 13669 | const U32 n = ARG(o); |
1de06328 | 13670 | const reg_ac_data * const ac = IS_TRIE_AC(op) ? |
f8fc2ecf | 13671 | (reg_ac_data *)progi->data->data[n] : |
1de06328 | 13672 | NULL; |
3251b653 NC |
13673 | const reg_trie_data * const trie |
13674 | = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie]; | |
1de06328 | 13675 | |
13d6edb4 | 13676 | Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]); |
1de06328 YO |
13677 | DEBUG_TRIE_COMPILE_r( |
13678 | Perl_sv_catpvf(aTHX_ sv, | |
13679 | "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">", | |
13680 | (UV)trie->startstate, | |
1e2e3d02 | 13681 | (IV)trie->statecount-1, /* -1 because of the unused 0 element */ |
1de06328 YO |
13682 | (UV)trie->wordcount, |
13683 | (UV)trie->minlen, | |
13684 | (UV)trie->maxlen, | |
13685 | (UV)TRIE_CHARCOUNT(trie), | |
13686 | (UV)trie->uniquecharcount | |
13687 | ) | |
13688 | ); | |
13689 | if ( IS_ANYOF_TRIE(op) || trie->bitmap ) { | |
13690 | int i; | |
13691 | int rangestart = -1; | |
f46cb337 | 13692 | U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie); |
f3a2811a | 13693 | sv_catpvs(sv, "["); |
1de06328 YO |
13694 | for (i = 0; i <= 256; i++) { |
13695 | if (i < 256 && BITMAP_TEST(bitmap,i)) { | |
13696 | if (rangestart == -1) | |
13697 | rangestart = i; | |
13698 | } else if (rangestart != -1) { | |
13699 | if (i <= rangestart + 3) | |
13700 | for (; rangestart < i; rangestart++) | |
13701 | put_byte(sv, rangestart); | |
13702 | else { | |
13703 | put_byte(sv, rangestart); | |
13704 | sv_catpvs(sv, "-"); | |
13705 | put_byte(sv, i - 1); | |
13706 | } | |
13707 | rangestart = -1; | |
13708 | } | |
13709 | } | |
f3a2811a | 13710 | sv_catpvs(sv, "]"); |
1de06328 YO |
13711 | } |
13712 | ||
a3621e74 | 13713 | } else if (k == CURLY) { |
cb434fcc | 13714 | if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX) |
cea2e8a9 GS |
13715 | Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */ |
13716 | Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o)); | |
a0d0e21e | 13717 | } |
2c2d71f5 JH |
13718 | else if (k == WHILEM && o->flags) /* Ordinal/of */ |
13719 | Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4); | |
1f1031fe | 13720 | else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) { |
894356b3 | 13721 | Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */ |
5daac39c | 13722 | if ( RXp_PAREN_NAMES(prog) ) { |
9d6ecd7a | 13723 | if ( k != REF || (OP(o) < NREF)) { |
502c6561 | 13724 | AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]); |
ee9b8eae YO |
13725 | SV **name= av_fetch(list, ARG(o), 0 ); |
13726 | if (name) | |
13727 | Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name)); | |
13728 | } | |
13729 | else { | |
502c6561 | 13730 | AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]); |
ad64d0ec | 13731 | SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]); |
ee9b8eae YO |
13732 | I32 *nums=(I32*)SvPVX(sv_dat); |
13733 | SV **name= av_fetch(list, nums[0], 0 ); | |
13734 | I32 n; | |
13735 | if (name) { | |
13736 | for ( n=0; n<SvIVX(sv_dat); n++ ) { | |
13737 | Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf, | |
13738 | (n ? "," : ""), (IV)nums[n]); | |
13739 | } | |
13740 | Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name)); | |
1f1031fe | 13741 | } |
1f1031fe | 13742 | } |
ee9b8eae | 13743 | } |
1f1031fe | 13744 | } else if (k == GOSUB) |
6bda09f9 | 13745 | Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */ |
e2e6a0f1 YO |
13746 | else if (k == VERB) { |
13747 | if (!o->flags) | |
13748 | Perl_sv_catpvf(aTHX_ sv, ":%"SVf, | |
ad64d0ec | 13749 | SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ])))); |
e2e6a0f1 | 13750 | } else if (k == LOGICAL) |
04ebc1ab | 13751 | Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */ |
653099ff GS |
13752 | else if (k == ANYOF) { |
13753 | int i, rangestart = -1; | |
2d03de9c | 13754 | const U8 flags = ANYOF_FLAGS(o); |
24d786f4 | 13755 | int do_sep = 0; |
0bd48802 | 13756 | |
653099ff | 13757 | |
19860706 | 13758 | if (flags & ANYOF_LOCALE) |
396482e1 | 13759 | sv_catpvs(sv, "{loc}"); |
39065660 | 13760 | if (flags & ANYOF_LOC_NONBITMAP_FOLD) |
396482e1 | 13761 | sv_catpvs(sv, "{i}"); |
653099ff | 13762 | Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]); |
19860706 | 13763 | if (flags & ANYOF_INVERT) |
396482e1 | 13764 | sv_catpvs(sv, "^"); |
686b73d4 | 13765 | |
3339dfd8 | 13766 | /* output what the standard cp 0-255 bitmap matches */ |
ffc61ed2 JH |
13767 | for (i = 0; i <= 256; i++) { |
13768 | if (i < 256 && ANYOF_BITMAP_TEST(o,i)) { | |
13769 | if (rangestart == -1) | |
13770 | rangestart = i; | |
13771 | } else if (rangestart != -1) { | |
13772 | if (i <= rangestart + 3) | |
13773 | for (; rangestart < i; rangestart++) | |
653099ff | 13774 | put_byte(sv, rangestart); |
ffc61ed2 JH |
13775 | else { |
13776 | put_byte(sv, rangestart); | |
396482e1 | 13777 | sv_catpvs(sv, "-"); |
ffc61ed2 | 13778 | put_byte(sv, i - 1); |
653099ff | 13779 | } |
24d786f4 | 13780 | do_sep = 1; |
ffc61ed2 | 13781 | rangestart = -1; |
653099ff | 13782 | } |
847a199f | 13783 | } |
3339dfd8 YO |
13784 | |
13785 | EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags); | |
3a15e693 KW |
13786 | /* output any special charclass tests (used entirely under use locale) */ |
13787 | if (ANYOF_CLASS_TEST_ANY_SET(o)) | |
bb7a0f54 | 13788 | for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++) |
24d786f4 | 13789 | if (ANYOF_CLASS_TEST(o,i)) { |
ffc61ed2 | 13790 | sv_catpv(sv, anyofs[i]); |
24d786f4 YO |
13791 | do_sep = 1; |
13792 | } | |
13793 | ||
3339dfd8 YO |
13794 | EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags); |
13795 | ||
11454c59 KW |
13796 | if (flags & ANYOF_NON_UTF8_LATIN1_ALL) { |
13797 | sv_catpvs(sv, "{non-utf8-latin1-all}"); | |
13798 | } | |
13799 | ||
3339dfd8 | 13800 | /* output information about the unicode matching */ |
ef87b810 | 13801 | if (flags & ANYOF_UNICODE_ALL) |
396482e1 | 13802 | sv_catpvs(sv, "{unicode_all}"); |
137165a6 | 13803 | else if (ANYOF_NONBITMAP(o)) |
ef87b810 | 13804 | sv_catpvs(sv, "{unicode}"); |
f5ecd18d | 13805 | if (flags & ANYOF_NONBITMAP_NON_UTF8) |
ef87b810 | 13806 | sv_catpvs(sv, "{outside bitmap}"); |
ffc61ed2 | 13807 | |
1aa9930e | 13808 | if (ANYOF_NONBITMAP(o)) { |
dbe7a391 | 13809 | SV *lv; /* Set if there is something outside the bit map */ |
32fc9b6a | 13810 | SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0); |
f1114c30 KW |
13811 | bool byte_output = FALSE; /* If something in the bitmap has been |
13812 | output */ | |
686b73d4 | 13813 | |
c16787fd | 13814 | if (lv && lv != &PL_sv_undef) { |
ffc61ed2 | 13815 | if (sw) { |
89ebb4a3 | 13816 | U8 s[UTF8_MAXBYTES_CASE+1]; |
24d786f4 | 13817 | |
dbe7a391 | 13818 | for (i = 0; i <= 256; i++) { /* Look at chars in bitmap */ |
1df70142 | 13819 | uvchr_to_utf8(s, i); |
686b73d4 | 13820 | |
dcf8909a KW |
13821 | if (i < 256 |
13822 | && ! ANYOF_BITMAP_TEST(o, i) /* Don't duplicate | |
13823 | things already | |
13824 | output as part | |
13825 | of the bitmap */ | |
13826 | && swash_fetch(sw, s, TRUE)) | |
13827 | { | |
ffc61ed2 JH |
13828 | if (rangestart == -1) |
13829 | rangestart = i; | |
13830 | } else if (rangestart != -1) { | |
f1114c30 | 13831 | byte_output = TRUE; |
ffc61ed2 JH |
13832 | if (i <= rangestart + 3) |
13833 | for (; rangestart < i; rangestart++) { | |
7128c099 | 13834 | put_byte(sv, rangestart); |
ffc61ed2 JH |
13835 | } |
13836 | else { | |
7128c099 | 13837 | put_byte(sv, rangestart); |
396482e1 | 13838 | sv_catpvs(sv, "-"); |
7128c099 | 13839 | put_byte(sv, i-1); |
ffc61ed2 | 13840 | } |
e87973a9 | 13841 | rangestart = -1; |
19860706 | 13842 | } |
e87973a9 | 13843 | } |
19860706 | 13844 | } |
fde631ed | 13845 | |
ffc61ed2 | 13846 | { |
2e0de35c | 13847 | char *s = savesvpv(lv); |
c445ea15 | 13848 | char * const origs = s; |
686b73d4 | 13849 | |
3dab1dad YO |
13850 | while (*s && *s != '\n') |
13851 | s++; | |
686b73d4 | 13852 | |
ffc61ed2 | 13853 | if (*s == '\n') { |
2d03de9c | 13854 | const char * const t = ++s; |
686b73d4 | 13855 | |
f1114c30 KW |
13856 | if (byte_output) { |
13857 | sv_catpvs(sv, " "); | |
13858 | } | |
13859 | ||
ffc61ed2 | 13860 | while (*s) { |
c574ffb9 KW |
13861 | if (*s == '\n') { |
13862 | ||
13863 | /* Truncate very long output */ | |
13864 | if (s - origs > 256) { | |
13865 | Perl_sv_catpvf(aTHX_ sv, | |
13866 | "%.*s...", | |
13867 | (int) (s - origs - 1), | |
13868 | t); | |
13869 | goto out_dump; | |
13870 | } | |
ffc61ed2 | 13871 | *s = ' '; |
1a9c8476 KW |
13872 | } |
13873 | else if (*s == '\t') { | |
13874 | *s = '-'; | |
13875 | } | |
ffc61ed2 JH |
13876 | s++; |
13877 | } | |
13878 | if (s[-1] == ' ') | |
13879 | s[-1] = 0; | |
686b73d4 | 13880 | |
ffc61ed2 | 13881 | sv_catpv(sv, t); |
fde631ed | 13882 | } |
686b73d4 | 13883 | |
c574ffb9 KW |
13884 | out_dump: |
13885 | ||
ffc61ed2 | 13886 | Safefree(origs); |
fde631ed | 13887 | } |
c16787fd | 13888 | SvREFCNT_dec(lv); |
fde631ed | 13889 | } |
653099ff | 13890 | } |
ffc61ed2 | 13891 | |
653099ff GS |
13892 | Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]); |
13893 | } | |
0658cdde KW |
13894 | else if (k == POSIXD) { |
13895 | U8 index = FLAGS(o) * 2; | |
13896 | if (index > (sizeof(anyofs) / sizeof(anyofs[0]))) { | |
13897 | Perl_sv_catpvf(aTHX_ sv, "[illegal type=%d])", index); | |
13898 | } | |
13899 | else { | |
13900 | sv_catpv(sv, anyofs[index]); | |
13901 | } | |
13902 | } | |
9b155405 | 13903 | else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH)) |
07be1b83 | 13904 | Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags)); |
65e66c80 | 13905 | #else |
96a5add6 | 13906 | PERL_UNUSED_CONTEXT; |
65e66c80 SP |
13907 | PERL_UNUSED_ARG(sv); |
13908 | PERL_UNUSED_ARG(o); | |
f9049ba1 | 13909 | PERL_UNUSED_ARG(prog); |
17c3b450 | 13910 | #endif /* DEBUGGING */ |
35ff7856 | 13911 | } |
a687059c | 13912 | |
cad2e5aa | 13913 | SV * |
288b8c02 | 13914 | Perl_re_intuit_string(pTHX_ REGEXP * const r) |
cad2e5aa | 13915 | { /* Assume that RE_INTUIT is set */ |
97aff369 | 13916 | dVAR; |
288b8c02 | 13917 | struct regexp *const prog = (struct regexp *)SvANY(r); |
a3621e74 | 13918 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
13919 | |
13920 | PERL_ARGS_ASSERT_RE_INTUIT_STRING; | |
96a5add6 AL |
13921 | PERL_UNUSED_CONTEXT; |
13922 | ||
a3621e74 | 13923 | DEBUG_COMPILE_r( |
cfd0369c | 13924 | { |
2d03de9c | 13925 | const char * const s = SvPV_nolen_const(prog->check_substr |
cfd0369c | 13926 | ? prog->check_substr : prog->check_utf8); |
cad2e5aa JH |
13927 | |
13928 | if (!PL_colorset) reginitcolors(); | |
13929 | PerlIO_printf(Perl_debug_log, | |
a0288114 | 13930 | "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n", |
33b8afdf JH |
13931 | PL_colors[4], |
13932 | prog->check_substr ? "" : "utf8 ", | |
13933 | PL_colors[5],PL_colors[0], | |
cad2e5aa JH |
13934 | s, |
13935 | PL_colors[1], | |
13936 | (strlen(s) > 60 ? "..." : "")); | |
13937 | } ); | |
13938 | ||
33b8afdf | 13939 | return prog->check_substr ? prog->check_substr : prog->check_utf8; |
cad2e5aa JH |
13940 | } |
13941 | ||
84da74a7 | 13942 | /* |
f8149455 | 13943 | pregfree() |
84da74a7 | 13944 | |
f8149455 YO |
13945 | handles refcounting and freeing the perl core regexp structure. When |
13946 | it is necessary to actually free the structure the first thing it | |
3b753521 | 13947 | does is call the 'free' method of the regexp_engine associated to |
f8149455 YO |
13948 | the regexp, allowing the handling of the void *pprivate; member |
13949 | first. (This routine is not overridable by extensions, which is why | |
13950 | the extensions free is called first.) | |
13951 | ||
13952 | See regdupe and regdupe_internal if you change anything here. | |
84da74a7 | 13953 | */ |
f8149455 | 13954 | #ifndef PERL_IN_XSUB_RE |
2b69d0c2 | 13955 | void |
84679df5 | 13956 | Perl_pregfree(pTHX_ REGEXP *r) |
a687059c | 13957 | { |
288b8c02 NC |
13958 | SvREFCNT_dec(r); |
13959 | } | |
13960 | ||
13961 | void | |
13962 | Perl_pregfree2(pTHX_ REGEXP *rx) | |
13963 | { | |
27da23d5 | 13964 | dVAR; |
288b8c02 | 13965 | struct regexp *const r = (struct regexp *)SvANY(rx); |
fc32ee4a | 13966 | GET_RE_DEBUG_FLAGS_DECL; |
a3621e74 | 13967 | |
7918f24d NC |
13968 | PERL_ARGS_ASSERT_PREGFREE2; |
13969 | ||
28d8d7f4 YO |
13970 | if (r->mother_re) { |
13971 | ReREFCNT_dec(r->mother_re); | |
13972 | } else { | |
288b8c02 | 13973 | CALLREGFREE_PVT(rx); /* free the private data */ |
ef8d46e8 | 13974 | SvREFCNT_dec(RXp_PAREN_NAMES(r)); |
28d8d7f4 YO |
13975 | } |
13976 | if (r->substrs) { | |
ef8d46e8 VP |
13977 | SvREFCNT_dec(r->anchored_substr); |
13978 | SvREFCNT_dec(r->anchored_utf8); | |
13979 | SvREFCNT_dec(r->float_substr); | |
13980 | SvREFCNT_dec(r->float_utf8); | |
28d8d7f4 YO |
13981 | Safefree(r->substrs); |
13982 | } | |
288b8c02 | 13983 | RX_MATCH_COPY_FREE(rx); |
f8c7b90f | 13984 | #ifdef PERL_OLD_COPY_ON_WRITE |
ef8d46e8 | 13985 | SvREFCNT_dec(r->saved_copy); |
ed252734 | 13986 | #endif |
f0ab9afb | 13987 | Safefree(r->offs); |
d63c20f2 | 13988 | SvREFCNT_dec(r->qr_anoncv); |
f8149455 | 13989 | } |
28d8d7f4 YO |
13990 | |
13991 | /* reg_temp_copy() | |
13992 | ||
13993 | This is a hacky workaround to the structural issue of match results | |
13994 | being stored in the regexp structure which is in turn stored in | |
13995 | PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern | |
13996 | could be PL_curpm in multiple contexts, and could require multiple | |
13997 | result sets being associated with the pattern simultaneously, such | |
13998 | as when doing a recursive match with (??{$qr}) | |
13999 | ||
14000 | The solution is to make a lightweight copy of the regexp structure | |
14001 | when a qr// is returned from the code executed by (??{$qr}) this | |
486ec47a | 14002 | lightweight copy doesn't actually own any of its data except for |
28d8d7f4 YO |
14003 | the starp/end and the actual regexp structure itself. |
14004 | ||
14005 | */ | |
14006 | ||
14007 | ||
84679df5 | 14008 | REGEXP * |
f0826785 | 14009 | Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx) |
7918f24d | 14010 | { |
f0826785 | 14011 | struct regexp *ret; |
288b8c02 | 14012 | struct regexp *const r = (struct regexp *)SvANY(rx); |
7918f24d NC |
14013 | |
14014 | PERL_ARGS_ASSERT_REG_TEMP_COPY; | |
14015 | ||
f0826785 BM |
14016 | if (!ret_x) |
14017 | ret_x = (REGEXP*) newSV_type(SVt_REGEXP); | |
14018 | ret = (struct regexp *)SvANY(ret_x); | |
14019 | ||
288b8c02 | 14020 | (void)ReREFCNT_inc(rx); |
f7c278bf NC |
14021 | /* We can take advantage of the existing "copied buffer" mechanism in SVs |
14022 | by pointing directly at the buffer, but flagging that the allocated | |
14023 | space in the copy is zero. As we've just done a struct copy, it's now | |
14024 | a case of zero-ing that, rather than copying the current length. */ | |
14025 | SvPV_set(ret_x, RX_WRAPPED(rx)); | |
8f6ae13c | 14026 | SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8); |
b6f60916 NC |
14027 | memcpy(&(ret->xpv_cur), &(r->xpv_cur), |
14028 | sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur)); | |
f7c278bf | 14029 | SvLEN_set(ret_x, 0); |
b9ad13ac | 14030 | SvSTASH_set(ret_x, NULL); |
703c388d | 14031 | SvMAGIC_set(ret_x, NULL); |
77465632 FP |
14032 | if (r->offs) { |
14033 | const I32 npar = r->nparens+1; | |
14034 | Newx(ret->offs, npar, regexp_paren_pair); | |
14035 | Copy(r->offs, ret->offs, npar, regexp_paren_pair); | |
14036 | } | |
28d8d7f4 | 14037 | if (r->substrs) { |
28d8d7f4 | 14038 | Newx(ret->substrs, 1, struct reg_substr_data); |
6ab65676 NC |
14039 | StructCopy(r->substrs, ret->substrs, struct reg_substr_data); |
14040 | ||
14041 | SvREFCNT_inc_void(ret->anchored_substr); | |
14042 | SvREFCNT_inc_void(ret->anchored_utf8); | |
14043 | SvREFCNT_inc_void(ret->float_substr); | |
14044 | SvREFCNT_inc_void(ret->float_utf8); | |
14045 | ||
14046 | /* check_substr and check_utf8, if non-NULL, point to either their | |
14047 | anchored or float namesakes, and don't hold a second reference. */ | |
486913e4 | 14048 | } |
288b8c02 | 14049 | RX_MATCH_COPIED_off(ret_x); |
28d8d7f4 | 14050 | #ifdef PERL_OLD_COPY_ON_WRITE |
b89b0c6f | 14051 | ret->saved_copy = NULL; |
28d8d7f4 | 14052 | #endif |
288b8c02 | 14053 | ret->mother_re = rx; |
d63c20f2 | 14054 | SvREFCNT_inc_void(ret->qr_anoncv); |
28d8d7f4 | 14055 | |
288b8c02 | 14056 | return ret_x; |
28d8d7f4 | 14057 | } |
f8149455 YO |
14058 | #endif |
14059 | ||
14060 | /* regfree_internal() | |
14061 | ||
14062 | Free the private data in a regexp. This is overloadable by | |
14063 | extensions. Perl takes care of the regexp structure in pregfree(), | |
3b753521 | 14064 | this covers the *pprivate pointer which technically perl doesn't |
f8149455 YO |
14065 | know about, however of course we have to handle the |
14066 | regexp_internal structure when no extension is in use. | |
14067 | ||
14068 | Note this is called before freeing anything in the regexp | |
14069 | structure. | |
14070 | */ | |
14071 | ||
14072 | void | |
288b8c02 | 14073 | Perl_regfree_internal(pTHX_ REGEXP * const rx) |
f8149455 YO |
14074 | { |
14075 | dVAR; | |
288b8c02 | 14076 | struct regexp *const r = (struct regexp *)SvANY(rx); |
f8149455 YO |
14077 | RXi_GET_DECL(r,ri); |
14078 | GET_RE_DEBUG_FLAGS_DECL; | |
7918f24d NC |
14079 | |
14080 | PERL_ARGS_ASSERT_REGFREE_INTERNAL; | |
14081 | ||
f8149455 YO |
14082 | DEBUG_COMPILE_r({ |
14083 | if (!PL_colorset) | |
14084 | reginitcolors(); | |
14085 | { | |
14086 | SV *dsv= sv_newmortal(); | |
3c8556c3 | 14087 | RE_PV_QUOTED_DECL(s, RX_UTF8(rx), |
5509d87a | 14088 | dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60); |
f8149455 YO |
14089 | PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n", |
14090 | PL_colors[4],PL_colors[5],s); | |
14091 | } | |
14092 | }); | |
7122b237 YO |
14093 | #ifdef RE_TRACK_PATTERN_OFFSETS |
14094 | if (ri->u.offsets) | |
14095 | Safefree(ri->u.offsets); /* 20010421 MJD */ | |
14096 | #endif | |
b30fcab9 DM |
14097 | if (ri->code_blocks) { |
14098 | int n; | |
14099 | for (n = 0; n < ri->num_code_blocks; n++) | |
14100 | SvREFCNT_dec(ri->code_blocks[n].src_regex); | |
3d2bd50a | 14101 | Safefree(ri->code_blocks); |
b30fcab9 | 14102 | } |
3d2bd50a | 14103 | |
f8fc2ecf YO |
14104 | if (ri->data) { |
14105 | int n = ri->data->count; | |
dfad63ad | 14106 | |
c277df42 | 14107 | while (--n >= 0) { |
261faec3 | 14108 | /* If you add a ->what type here, update the comment in regcomp.h */ |
f8fc2ecf | 14109 | switch (ri->data->what[n]) { |
af534a04 | 14110 | case 'a': |
b30fcab9 | 14111 | case 'r': |
c277df42 | 14112 | case 's': |
81714fb9 | 14113 | case 'S': |
55eed653 | 14114 | case 'u': |
ad64d0ec | 14115 | SvREFCNT_dec(MUTABLE_SV(ri->data->data[n])); |
c277df42 | 14116 | break; |
653099ff | 14117 | case 'f': |
f8fc2ecf | 14118 | Safefree(ri->data->data[n]); |
653099ff | 14119 | break; |
68e2671b | 14120 | case 'l': |
d63c20f2 | 14121 | case 'L': |
9e55ce06 | 14122 | break; |
07be1b83 | 14123 | case 'T': |
be8e71aa YO |
14124 | { /* Aho Corasick add-on structure for a trie node. |
14125 | Used in stclass optimization only */ | |
07be1b83 | 14126 | U32 refcount; |
f8fc2ecf | 14127 | reg_ac_data *aho=(reg_ac_data*)ri->data->data[n]; |
07be1b83 YO |
14128 | OP_REFCNT_LOCK; |
14129 | refcount = --aho->refcount; | |
14130 | OP_REFCNT_UNLOCK; | |
14131 | if ( !refcount ) { | |
446bd890 NC |
14132 | PerlMemShared_free(aho->states); |
14133 | PerlMemShared_free(aho->fail); | |
446bd890 NC |
14134 | /* do this last!!!! */ |
14135 | PerlMemShared_free(ri->data->data[n]); | |
14136 | PerlMemShared_free(ri->regstclass); | |
07be1b83 YO |
14137 | } |
14138 | } | |
14139 | break; | |
a3621e74 | 14140 | case 't': |
07be1b83 | 14141 | { |
be8e71aa | 14142 | /* trie structure. */ |
07be1b83 | 14143 | U32 refcount; |
f8fc2ecf | 14144 | reg_trie_data *trie=(reg_trie_data*)ri->data->data[n]; |
07be1b83 YO |
14145 | OP_REFCNT_LOCK; |
14146 | refcount = --trie->refcount; | |
14147 | OP_REFCNT_UNLOCK; | |
14148 | if ( !refcount ) { | |
446bd890 | 14149 | PerlMemShared_free(trie->charmap); |
446bd890 NC |
14150 | PerlMemShared_free(trie->states); |
14151 | PerlMemShared_free(trie->trans); | |
07be1b83 | 14152 | if (trie->bitmap) |
446bd890 | 14153 | PerlMemShared_free(trie->bitmap); |
786e8c11 | 14154 | if (trie->jump) |
446bd890 | 14155 | PerlMemShared_free(trie->jump); |
2e64971a | 14156 | PerlMemShared_free(trie->wordinfo); |
446bd890 NC |
14157 | /* do this last!!!! */ |
14158 | PerlMemShared_free(ri->data->data[n]); | |
a3621e74 | 14159 | } |
07be1b83 YO |
14160 | } |
14161 | break; | |
c277df42 | 14162 | default: |
f8fc2ecf | 14163 | Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]); |
c277df42 IZ |
14164 | } |
14165 | } | |
f8fc2ecf YO |
14166 | Safefree(ri->data->what); |
14167 | Safefree(ri->data); | |
a0d0e21e | 14168 | } |
28d8d7f4 | 14169 | |
f8fc2ecf | 14170 | Safefree(ri); |
a687059c | 14171 | } |
c277df42 | 14172 | |
a09252eb NC |
14173 | #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t)) |
14174 | #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t)) | |
84da74a7 YO |
14175 | #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL) |
14176 | ||
14177 | /* | |
32cd70f6 | 14178 | re_dup - duplicate a regexp. |
84da74a7 | 14179 | |
8233f606 DM |
14180 | This routine is expected to clone a given regexp structure. It is only |
14181 | compiled under USE_ITHREADS. | |
32cd70f6 | 14182 | |
f8149455 YO |
14183 | After all of the core data stored in struct regexp is duplicated |
14184 | the regexp_engine.dupe method is used to copy any private data | |
14185 | stored in the *pprivate pointer. This allows extensions to handle | |
14186 | any duplication it needs to do. | |
14187 | ||
14188 | See pregfree() and regfree_internal() if you change anything here. | |
84da74a7 | 14189 | */ |
a3c0e9ca | 14190 | #if defined(USE_ITHREADS) |
f8149455 | 14191 | #ifndef PERL_IN_XSUB_RE |
288b8c02 NC |
14192 | void |
14193 | Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param) | |
84da74a7 | 14194 | { |
84da74a7 | 14195 | dVAR; |
a86a1ca7 | 14196 | I32 npar; |
288b8c02 NC |
14197 | const struct regexp *r = (const struct regexp *)SvANY(sstr); |
14198 | struct regexp *ret = (struct regexp *)SvANY(dstr); | |
f8149455 | 14199 | |
7918f24d NC |
14200 | PERL_ARGS_ASSERT_RE_DUP_GUTS; |
14201 | ||
84da74a7 | 14202 | npar = r->nparens+1; |
f0ab9afb NC |
14203 | Newx(ret->offs, npar, regexp_paren_pair); |
14204 | Copy(r->offs, ret->offs, npar, regexp_paren_pair); | |
6057429f | 14205 | if(ret->swap) { |
28d8d7f4 | 14206 | /* no need to copy these */ |
f0ab9afb | 14207 | Newx(ret->swap, npar, regexp_paren_pair); |
28d8d7f4 | 14208 | } |
84da74a7 | 14209 | |
6057429f | 14210 | if (ret->substrs) { |
32cd70f6 NC |
14211 | /* Do it this way to avoid reading from *r after the StructCopy(). |
14212 | That way, if any of the sv_dup_inc()s dislodge *r from the L1 | |
14213 | cache, it doesn't matter. */ | |
66b1de87 NC |
14214 | const bool anchored = r->check_substr |
14215 | ? r->check_substr == r->anchored_substr | |
14216 | : r->check_utf8 == r->anchored_utf8; | |
785a26d5 | 14217 | Newx(ret->substrs, 1, struct reg_substr_data); |
a86a1ca7 NC |
14218 | StructCopy(r->substrs, ret->substrs, struct reg_substr_data); |
14219 | ||
32cd70f6 NC |
14220 | ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param); |
14221 | ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param); | |
14222 | ret->float_substr = sv_dup_inc(ret->float_substr, param); | |
14223 | ret->float_utf8 = sv_dup_inc(ret->float_utf8, param); | |
a86a1ca7 | 14224 | |
32cd70f6 NC |
14225 | /* check_substr and check_utf8, if non-NULL, point to either their |
14226 | anchored or float namesakes, and don't hold a second reference. */ | |
14227 | ||
14228 | if (ret->check_substr) { | |
14229 | if (anchored) { | |
14230 | assert(r->check_utf8 == r->anchored_utf8); | |
14231 | ret->check_substr = ret->anchored_substr; | |
14232 | ret->check_utf8 = ret->anchored_utf8; | |
14233 | } else { | |
14234 | assert(r->check_substr == r->float_substr); | |
14235 | assert(r->check_utf8 == r->float_utf8); | |
14236 | ret->check_substr = ret->float_substr; | |
14237 | ret->check_utf8 = ret->float_utf8; | |
14238 | } | |
66b1de87 NC |
14239 | } else if (ret->check_utf8) { |
14240 | if (anchored) { | |
14241 | ret->check_utf8 = ret->anchored_utf8; | |
14242 | } else { | |
14243 | ret->check_utf8 = ret->float_utf8; | |
14244 | } | |
32cd70f6 | 14245 | } |
6057429f | 14246 | } |
f8149455 | 14247 | |
5daac39c | 14248 | RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param); |
d63c20f2 | 14249 | ret->qr_anoncv = MUTABLE_CV(sv_dup_inc((const SV *)ret->qr_anoncv, param)); |
bcdf7404 | 14250 | |
6057429f | 14251 | if (ret->pprivate) |
288b8c02 | 14252 | RXi_SET(ret,CALLREGDUPE_PVT(dstr,param)); |
f8149455 | 14253 | |
288b8c02 | 14254 | if (RX_MATCH_COPIED(dstr)) |
6057429f | 14255 | ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen); |
f8149455 YO |
14256 | else |
14257 | ret->subbeg = NULL; | |
14258 | #ifdef PERL_OLD_COPY_ON_WRITE | |
14259 | ret->saved_copy = NULL; | |
14260 | #endif | |
6057429f | 14261 | |
c2123ae3 NC |
14262 | if (ret->mother_re) { |
14263 | if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) { | |
14264 | /* Our storage points directly to our mother regexp, but that's | |
14265 | 1: a buffer in a different thread | |
14266 | 2: something we no longer hold a reference on | |
14267 | so we need to copy it locally. */ | |
d5aafdca FC |
14268 | /* Note we need to use SvCUR(), rather than |
14269 | SvLEN(), on our mother_re, because it, in | |
c2123ae3 NC |
14270 | turn, may well be pointing to its own mother_re. */ |
14271 | SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re), | |
14272 | SvCUR(ret->mother_re)+1)); | |
14273 | SvLEN_set(dstr, SvCUR(ret->mother_re)+1); | |
14274 | } | |
14275 | ret->mother_re = NULL; | |
14276 | } | |
6057429f | 14277 | ret->gofs = 0; |
f8149455 YO |
14278 | } |
14279 | #endif /* PERL_IN_XSUB_RE */ | |
14280 | ||
14281 | /* | |
14282 | regdupe_internal() | |
14283 | ||
14284 | This is the internal complement to regdupe() which is used to copy | |
14285 | the structure pointed to by the *pprivate pointer in the regexp. | |
14286 | This is the core version of the extension overridable cloning hook. | |
14287 | The regexp structure being duplicated will be copied by perl prior | |
14288 | to this and will be provided as the regexp *r argument, however | |
14289 | with the /old/ structures pprivate pointer value. Thus this routine | |
14290 | may override any copying normally done by perl. | |
14291 | ||
14292 | It returns a pointer to the new regexp_internal structure. | |
14293 | */ | |
14294 | ||
14295 | void * | |
288b8c02 | 14296 | Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param) |
f8149455 YO |
14297 | { |
14298 | dVAR; | |
288b8c02 | 14299 | struct regexp *const r = (struct regexp *)SvANY(rx); |
f8149455 | 14300 | regexp_internal *reti; |
0780bc72 | 14301 | int len; |
f8149455 | 14302 | RXi_GET_DECL(r,ri); |
7918f24d NC |
14303 | |
14304 | PERL_ARGS_ASSERT_REGDUPE_INTERNAL; | |
f8149455 | 14305 | |
7122b237 | 14306 | len = ProgLen(ri); |
f8149455 | 14307 | |
45cf4570 | 14308 | Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal); |
f8149455 | 14309 | Copy(ri->program, reti->program, len+1, regnode); |
3d2bd50a DM |
14310 | |
14311 | reti->num_code_blocks = ri->num_code_blocks; | |
14312 | if (ri->code_blocks) { | |
b30fcab9 | 14313 | int n; |
3d2bd50a DM |
14314 | Newxc(reti->code_blocks, ri->num_code_blocks, struct reg_code_block, |
14315 | struct reg_code_block); | |
14316 | Copy(ri->code_blocks, reti->code_blocks, ri->num_code_blocks, | |
14317 | struct reg_code_block); | |
b30fcab9 DM |
14318 | for (n = 0; n < ri->num_code_blocks; n++) |
14319 | reti->code_blocks[n].src_regex = (REGEXP*) | |
14320 | sv_dup_inc((SV*)(ri->code_blocks[n].src_regex), param); | |
3d2bd50a DM |
14321 | } |
14322 | else | |
14323 | reti->code_blocks = NULL; | |
f8149455 | 14324 | |
f8fc2ecf | 14325 | reti->regstclass = NULL; |
bcdf7404 | 14326 | |
f8fc2ecf | 14327 | if (ri->data) { |
84da74a7 | 14328 | struct reg_data *d; |
f8fc2ecf | 14329 | const int count = ri->data->count; |
84da74a7 YO |
14330 | int i; |
14331 | ||
14332 | Newxc(d, sizeof(struct reg_data) + count*sizeof(void *), | |
14333 | char, struct reg_data); | |
14334 | Newx(d->what, count, U8); | |
14335 | ||
14336 | d->count = count; | |
14337 | for (i = 0; i < count; i++) { | |
f8fc2ecf | 14338 | d->what[i] = ri->data->what[i]; |
84da74a7 | 14339 | switch (d->what[i]) { |
d24ca0c5 | 14340 | /* see also regcomp.h and regfree_internal() */ |
af534a04 | 14341 | case 'a': /* actually an AV, but the dup function is identical. */ |
b30fcab9 | 14342 | case 'r': |
84da74a7 | 14343 | case 's': |
81714fb9 | 14344 | case 'S': |
55eed653 | 14345 | case 'u': /* actually an HV, but the dup function is identical. */ |
ad64d0ec | 14346 | d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param); |
84da74a7 | 14347 | break; |
84da74a7 YO |
14348 | case 'f': |
14349 | /* This is cheating. */ | |
14350 | Newx(d->data[i], 1, struct regnode_charclass_class); | |
f8fc2ecf | 14351 | StructCopy(ri->data->data[i], d->data[i], |
84da74a7 | 14352 | struct regnode_charclass_class); |
f8fc2ecf | 14353 | reti->regstclass = (regnode*)d->data[i]; |
84da74a7 | 14354 | break; |
23eab42c NC |
14355 | case 'T': |
14356 | /* Trie stclasses are readonly and can thus be shared | |
14357 | * without duplication. We free the stclass in pregfree | |
14358 | * when the corresponding reg_ac_data struct is freed. | |
14359 | */ | |
14360 | reti->regstclass= ri->regstclass; | |
14361 | /* Fall through */ | |
84da74a7 | 14362 | case 't': |
84da74a7 | 14363 | OP_REFCNT_LOCK; |
0536c0a7 | 14364 | ((reg_trie_data*)ri->data->data[i])->refcount++; |
84da74a7 | 14365 | OP_REFCNT_UNLOCK; |
0536c0a7 | 14366 | /* Fall through */ |
68e2671b | 14367 | case 'l': |
d63c20f2 | 14368 | case 'L': |
0536c0a7 | 14369 | d->data[i] = ri->data->data[i]; |
84da74a7 | 14370 | break; |
84da74a7 | 14371 | default: |
f8fc2ecf | 14372 | Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]); |
84da74a7 YO |
14373 | } |
14374 | } | |
14375 | ||
f8fc2ecf | 14376 | reti->data = d; |
84da74a7 YO |
14377 | } |
14378 | else | |
f8fc2ecf | 14379 | reti->data = NULL; |
84da74a7 | 14380 | |
cde0cee5 YO |
14381 | reti->name_list_idx = ri->name_list_idx; |
14382 | ||
7122b237 YO |
14383 | #ifdef RE_TRACK_PATTERN_OFFSETS |
14384 | if (ri->u.offsets) { | |
14385 | Newx(reti->u.offsets, 2*len+1, U32); | |
14386 | Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32); | |
14387 | } | |
14388 | #else | |
14389 | SetProgLen(reti,len); | |
14390 | #endif | |
14391 | ||
f8149455 | 14392 | return (void*)reti; |
84da74a7 | 14393 | } |
f8149455 YO |
14394 | |
14395 | #endif /* USE_ITHREADS */ | |
84da74a7 | 14396 | |
f8149455 | 14397 | #ifndef PERL_IN_XSUB_RE |
bcdf7404 | 14398 | |
c277df42 IZ |
14399 | /* |
14400 | - regnext - dig the "next" pointer out of a node | |
c277df42 IZ |
14401 | */ |
14402 | regnode * | |
864dbfa3 | 14403 | Perl_regnext(pTHX_ register regnode *p) |
c277df42 | 14404 | { |
97aff369 | 14405 | dVAR; |
eb578fdb | 14406 | I32 offset; |
c277df42 | 14407 | |
f8fc2ecf | 14408 | if (!p) |
c277df42 IZ |
14409 | return(NULL); |
14410 | ||
35db910f KW |
14411 | if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */ |
14412 | Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX); | |
14413 | } | |
14414 | ||
c277df42 IZ |
14415 | offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p)); |
14416 | if (offset == 0) | |
14417 | return(NULL); | |
14418 | ||
c277df42 | 14419 | return(p+offset); |
c277df42 | 14420 | } |
76234dfb | 14421 | #endif |
c277df42 | 14422 | |
686b73d4 | 14423 | STATIC void |
cea2e8a9 | 14424 | S_re_croak2(pTHX_ const char* pat1,const char* pat2,...) |
c277df42 IZ |
14425 | { |
14426 | va_list args; | |
14427 | STRLEN l1 = strlen(pat1); | |
14428 | STRLEN l2 = strlen(pat2); | |
14429 | char buf[512]; | |
06bf62c7 | 14430 | SV *msv; |
73d840c0 | 14431 | const char *message; |
c277df42 | 14432 | |
7918f24d NC |
14433 | PERL_ARGS_ASSERT_RE_CROAK2; |
14434 | ||
c277df42 IZ |
14435 | if (l1 > 510) |
14436 | l1 = 510; | |
14437 | if (l1 + l2 > 510) | |
14438 | l2 = 510 - l1; | |
14439 | Copy(pat1, buf, l1 , char); | |
14440 | Copy(pat2, buf + l1, l2 , char); | |
3b818b81 GS |
14441 | buf[l1 + l2] = '\n'; |
14442 | buf[l1 + l2 + 1] = '\0'; | |
8736538c AS |
14443 | #ifdef I_STDARG |
14444 | /* ANSI variant takes additional second argument */ | |
c277df42 | 14445 | va_start(args, pat2); |
8736538c AS |
14446 | #else |
14447 | va_start(args); | |
14448 | #endif | |
5a844595 | 14449 | msv = vmess(buf, &args); |
c277df42 | 14450 | va_end(args); |
cfd0369c | 14451 | message = SvPV_const(msv,l1); |
c277df42 IZ |
14452 | if (l1 > 512) |
14453 | l1 = 512; | |
14454 | Copy(message, buf, l1 , char); | |
197cf9b9 | 14455 | buf[l1-1] = '\0'; /* Overwrite \n */ |
cea2e8a9 | 14456 | Perl_croak(aTHX_ "%s", buf); |
c277df42 | 14457 | } |
a0ed51b3 LW |
14458 | |
14459 | /* XXX Here's a total kludge. But we need to re-enter for swash routines. */ | |
14460 | ||
76234dfb | 14461 | #ifndef PERL_IN_XSUB_RE |
a0ed51b3 | 14462 | void |
864dbfa3 | 14463 | Perl_save_re_context(pTHX) |
b81d288d | 14464 | { |
97aff369 | 14465 | dVAR; |
1ade1aa1 NC |
14466 | |
14467 | struct re_save_state *state; | |
14468 | ||
14469 | SAVEVPTR(PL_curcop); | |
14470 | SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1); | |
14471 | ||
14472 | state = (struct re_save_state *)(PL_savestack + PL_savestack_ix); | |
14473 | PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE; | |
c6bf6a65 | 14474 | SSPUSHUV(SAVEt_RE_STATE); |
1ade1aa1 | 14475 | |
46ab3289 | 14476 | Copy(&PL_reg_state, state, 1, struct re_save_state); |
1ade1aa1 | 14477 | |
c445ea15 | 14478 | PL_reg_oldsaved = NULL; |
a5db57d6 | 14479 | PL_reg_oldsavedlen = 0; |
6502e081 DM |
14480 | PL_reg_oldsavedoffset = 0; |
14481 | PL_reg_oldsavedcoffset = 0; | |
a5db57d6 | 14482 | PL_reg_maxiter = 0; |
a5db57d6 | 14483 | PL_reg_leftiter = 0; |
c445ea15 | 14484 | PL_reg_poscache = NULL; |
a5db57d6 | 14485 | PL_reg_poscache_size = 0; |
1ade1aa1 NC |
14486 | #ifdef PERL_OLD_COPY_ON_WRITE |
14487 | PL_nrs = NULL; | |
14488 | #endif | |
ada6e8a9 | 14489 | |
c445ea15 AL |
14490 | /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */ |
14491 | if (PL_curpm) { | |
14492 | const REGEXP * const rx = PM_GETRE(PL_curpm); | |
14493 | if (rx) { | |
1df70142 | 14494 | U32 i; |
07bc277f | 14495 | for (i = 1; i <= RX_NPARENS(rx); i++) { |
1df70142 | 14496 | char digits[TYPE_CHARS(long)]; |
d9fad198 | 14497 | const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i); |
49f27e4b NC |
14498 | GV *const *const gvp |
14499 | = (GV**)hv_fetch(PL_defstash, digits, len, 0); | |
14500 | ||
b37c2d43 AL |
14501 | if (gvp) { |
14502 | GV * const gv = *gvp; | |
14503 | if (SvTYPE(gv) == SVt_PVGV && GvSV(gv)) | |
14504 | save_scalar(gv); | |
49f27e4b | 14505 | } |
ada6e8a9 AMS |
14506 | } |
14507 | } | |
14508 | } | |
a0ed51b3 | 14509 | } |
76234dfb | 14510 | #endif |
51371543 | 14511 | |
51371543 | 14512 | static void |
acfe0abc | 14513 | clear_re(pTHX_ void *r) |
51371543 | 14514 | { |
97aff369 | 14515 | dVAR; |
84679df5 | 14516 | ReREFCNT_dec((REGEXP *)r); |
51371543 | 14517 | } |
ffbc6a93 | 14518 | |
a28509cc AL |
14519 | #ifdef DEBUGGING |
14520 | ||
14521 | STATIC void | |
14522 | S_put_byte(pTHX_ SV *sv, int c) | |
14523 | { | |
7918f24d NC |
14524 | PERL_ARGS_ASSERT_PUT_BYTE; |
14525 | ||
7fddd944 NC |
14526 | /* Our definition of isPRINT() ignores locales, so only bytes that are |
14527 | not part of UTF-8 are considered printable. I assume that the same | |
14528 | holds for UTF-EBCDIC. | |
14529 | Also, code point 255 is not printable in either (it's E0 in EBCDIC, | |
14530 | which Wikipedia says: | |
14531 | ||
14532 | EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all | |
14533 | ones (binary 1111 1111, hexadecimal FF). It is similar, but not | |
14534 | identical, to the ASCII delete (DEL) or rubout control character. | |
14535 | ) So the old condition can be simplified to !isPRINT(c) */ | |
9ce2357e KW |
14536 | if (!isPRINT(c)) { |
14537 | if (c < 256) { | |
14538 | Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c); | |
14539 | } | |
14540 | else { | |
14541 | Perl_sv_catpvf(aTHX_ sv, "\\x{%x}", c); | |
14542 | } | |
14543 | } | |
5e7aa789 | 14544 | else { |
88c9ea1e | 14545 | const char string = c; |
5e7aa789 NC |
14546 | if (c == '-' || c == ']' || c == '\\' || c == '^') |
14547 | sv_catpvs(sv, "\\"); | |
14548 | sv_catpvn(sv, &string, 1); | |
14549 | } | |
a28509cc AL |
14550 | } |
14551 | ||
786e8c11 | 14552 | |
3dab1dad YO |
14553 | #define CLEAR_OPTSTART \ |
14554 | if (optstart) STMT_START { \ | |
70685ca0 | 14555 | DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \ |
3dab1dad YO |
14556 | optstart=NULL; \ |
14557 | } STMT_END | |
14558 | ||
786e8c11 | 14559 | #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1); |
3dab1dad | 14560 | |
b5a2f8d8 NC |
14561 | STATIC const regnode * |
14562 | S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node, | |
786e8c11 YO |
14563 | const regnode *last, const regnode *plast, |
14564 | SV* sv, I32 indent, U32 depth) | |
a28509cc | 14565 | { |
97aff369 | 14566 | dVAR; |
eb578fdb KW |
14567 | U8 op = PSEUDO; /* Arbitrary non-END op. */ |
14568 | const regnode *next; | |
3dab1dad | 14569 | const regnode *optstart= NULL; |
1f1031fe | 14570 | |
f8fc2ecf | 14571 | RXi_GET_DECL(r,ri); |
3dab1dad | 14572 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
14573 | |
14574 | PERL_ARGS_ASSERT_DUMPUNTIL; | |
14575 | ||
786e8c11 YO |
14576 | #ifdef DEBUG_DUMPUNTIL |
14577 | PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start, | |
14578 | last ? last-start : 0,plast ? plast-start : 0); | |
14579 | #endif | |
14580 | ||
14581 | if (plast && plast < last) | |
14582 | last= plast; | |
14583 | ||
14584 | while (PL_regkind[op] != END && (!last || node < last)) { | |
a28509cc | 14585 | /* While that wasn't END last time... */ |
a28509cc AL |
14586 | NODE_ALIGN(node); |
14587 | op = OP(node); | |
de734bd5 | 14588 | if (op == CLOSE || op == WHILEM) |
786e8c11 | 14589 | indent--; |
b5a2f8d8 | 14590 | next = regnext((regnode *)node); |
1f1031fe | 14591 | |
a28509cc | 14592 | /* Where, what. */ |
8e11feef | 14593 | if (OP(node) == OPTIMIZED) { |
e68ec53f | 14594 | if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) |
8e11feef | 14595 | optstart = node; |
3dab1dad | 14596 | else |
8e11feef | 14597 | goto after_print; |
3dab1dad YO |
14598 | } else |
14599 | CLEAR_OPTSTART; | |
686b73d4 | 14600 | |
32fc9b6a | 14601 | regprop(r, sv, node); |
a28509cc | 14602 | PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start), |
786e8c11 | 14603 | (int)(2*indent + 1), "", SvPVX_const(sv)); |
1f1031fe YO |
14604 | |
14605 | if (OP(node) != OPTIMIZED) { | |
14606 | if (next == NULL) /* Next ptr. */ | |
14607 | PerlIO_printf(Perl_debug_log, " (0)"); | |
14608 | else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH ) | |
14609 | PerlIO_printf(Perl_debug_log, " (FAIL)"); | |
14610 | else | |
14611 | PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start)); | |
14612 | (void)PerlIO_putc(Perl_debug_log, '\n'); | |
14613 | } | |
14614 | ||
a28509cc AL |
14615 | after_print: |
14616 | if (PL_regkind[(U8)op] == BRANCHJ) { | |
be8e71aa YO |
14617 | assert(next); |
14618 | { | |
eb578fdb KW |
14619 | const regnode *nnode = (OP(next) == LONGJMP |
14620 | ? regnext((regnode *)next) | |
14621 | : next); | |
be8e71aa YO |
14622 | if (last && nnode > last) |
14623 | nnode = last; | |
786e8c11 | 14624 | DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode); |
be8e71aa | 14625 | } |
a28509cc AL |
14626 | } |
14627 | else if (PL_regkind[(U8)op] == BRANCH) { | |
be8e71aa | 14628 | assert(next); |
786e8c11 | 14629 | DUMPUNTIL(NEXTOPER(node), next); |
a28509cc AL |
14630 | } |
14631 | else if ( PL_regkind[(U8)op] == TRIE ) { | |
7f69552c | 14632 | const regnode *this_trie = node; |
1de06328 | 14633 | const char op = OP(node); |
647f639f | 14634 | const U32 n = ARG(node); |
1de06328 | 14635 | const reg_ac_data * const ac = op>=AHOCORASICK ? |
f8fc2ecf | 14636 | (reg_ac_data *)ri->data->data[n] : |
1de06328 | 14637 | NULL; |
3251b653 NC |
14638 | const reg_trie_data * const trie = |
14639 | (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie]; | |
2b8b4781 | 14640 | #ifdef DEBUGGING |
502c6561 | 14641 | AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]); |
2b8b4781 | 14642 | #endif |
786e8c11 | 14643 | const regnode *nextbranch= NULL; |
a28509cc | 14644 | I32 word_idx; |
76f68e9b | 14645 | sv_setpvs(sv, ""); |
786e8c11 | 14646 | for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) { |
2b8b4781 | 14647 | SV ** const elem_ptr = av_fetch(trie_words,word_idx,0); |
686b73d4 | 14648 | |
786e8c11 YO |
14649 | PerlIO_printf(Perl_debug_log, "%*s%s ", |
14650 | (int)(2*(indent+3)), "", | |
14651 | elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60, | |
ab3bbdeb YO |
14652 | PL_colors[0], PL_colors[1], |
14653 | (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) | | |
95b611b0 | 14654 | PERL_PV_PRETTY_ELLIPSES | |
7f69552c | 14655 | PERL_PV_PRETTY_LTGT |
786e8c11 YO |
14656 | ) |
14657 | : "???" | |
14658 | ); | |
14659 | if (trie->jump) { | |
40d049e4 | 14660 | U16 dist= trie->jump[word_idx+1]; |
70685ca0 JH |
14661 | PerlIO_printf(Perl_debug_log, "(%"UVuf")\n", |
14662 | (UV)((dist ? this_trie + dist : next) - start)); | |
786e8c11 YO |
14663 | if (dist) { |
14664 | if (!nextbranch) | |
24b23f37 | 14665 | nextbranch= this_trie + trie->jump[0]; |
7f69552c YO |
14666 | DUMPUNTIL(this_trie + dist, nextbranch); |
14667 | } | |
786e8c11 YO |
14668 | if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH) |
14669 | nextbranch= regnext((regnode *)nextbranch); | |
14670 | } else { | |
14671 | PerlIO_printf(Perl_debug_log, "\n"); | |
a28509cc | 14672 | } |
786e8c11 YO |
14673 | } |
14674 | if (last && next > last) | |
14675 | node= last; | |
14676 | else | |
14677 | node= next; | |
a28509cc | 14678 | } |
786e8c11 YO |
14679 | else if ( op == CURLY ) { /* "next" might be very big: optimizer */ |
14680 | DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, | |
14681 | NEXTOPER(node) + EXTRA_STEP_2ARGS + 1); | |
a28509cc AL |
14682 | } |
14683 | else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) { | |
be8e71aa | 14684 | assert(next); |
786e8c11 | 14685 | DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next); |
a28509cc AL |
14686 | } |
14687 | else if ( op == PLUS || op == STAR) { | |
786e8c11 | 14688 | DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1); |
a28509cc | 14689 | } |
f56b6394 | 14690 | else if (PL_regkind[(U8)op] == ANYOF) { |
a28509cc | 14691 | /* arglen 1 + class block */ |
4a3ee7a8 | 14692 | node += 1 + ((ANYOF_FLAGS(node) & ANYOF_CLASS) |
a28509cc AL |
14693 | ? ANYOF_CLASS_SKIP : ANYOF_SKIP); |
14694 | node = NEXTOPER(node); | |
14695 | } | |
14696 | else if (PL_regkind[(U8)op] == EXACT) { | |
14697 | /* Literal string, where present. */ | |
14698 | node += NODE_SZ_STR(node) - 1; | |
14699 | node = NEXTOPER(node); | |
14700 | } | |
14701 | else { | |
14702 | node = NEXTOPER(node); | |
14703 | node += regarglen[(U8)op]; | |
14704 | } | |
14705 | if (op == CURLYX || op == OPEN) | |
786e8c11 | 14706 | indent++; |
a28509cc | 14707 | } |
3dab1dad | 14708 | CLEAR_OPTSTART; |
786e8c11 | 14709 | #ifdef DEBUG_DUMPUNTIL |
70685ca0 | 14710 | PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent); |
786e8c11 | 14711 | #endif |
1de06328 | 14712 | return node; |
a28509cc AL |
14713 | } |
14714 | ||
14715 | #endif /* DEBUGGING */ | |
14716 | ||
241d1a3b NC |
14717 | /* |
14718 | * Local variables: | |
14719 | * c-indentation-style: bsd | |
14720 | * c-basic-offset: 4 | |
14d04a33 | 14721 | * indent-tabs-mode: nil |
241d1a3b NC |
14722 | * End: |
14723 | * | |
14d04a33 | 14724 | * ex: set ts=8 sts=4 sw=4 et: |
37442d52 | 14725 | */ |