]>
Commit | Line | Data |
---|---|---|
2b83a2a4 RM |
1 | /* Extended regular expression matching and search library, |
2 | version 0.12. | |
51702635 | 3 | (Implements POSIX draft P1003.2/D11.2, except for some of the |
2b83a2a4 | 4 | internationalization features.) |
e4c785c8 | 5 | Copyright (C) 1993-1999, 2000, 2001 Free Software Foundation, Inc. |
2b83a2a4 | 6 | |
c84142e8 UD |
7 | The GNU C Library is free software; you can redistribute it and/or |
8 | modify it under the terms of the GNU Library General Public License as | |
9 | published by the Free Software Foundation; either version 2 of the | |
10 | License, or (at your option) any later version. | |
2b83a2a4 | 11 | |
c84142e8 UD |
12 | The GNU C Library is distributed in the hope that it will be useful, |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | Library General Public License for more details. | |
2b83a2a4 | 16 | |
c84142e8 UD |
17 | You should have received a copy of the GNU Library General Public |
18 | License along with the GNU C Library; see the file COPYING.LIB. If not, | |
19 | write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
20 | Boston, MA 02111-1307, USA. */ | |
2b83a2a4 RM |
21 | |
22 | /* AIX requires this to be the first thing in the file. */ | |
86187531 | 23 | #if defined _AIX && !defined REGEX_MALLOC |
2b83a2a4 RM |
24 | #pragma alloca |
25 | #endif | |
26 | ||
80b55d32 | 27 | #undef _GNU_SOURCE |
2b83a2a4 RM |
28 | #define _GNU_SOURCE |
29 | ||
30 | #ifdef HAVE_CONFIG_H | |
86187531 | 31 | # include <config.h> |
2b83a2a4 RM |
32 | #endif |
33 | ||
07b51ba5 UD |
34 | #ifndef PARAMS |
35 | # if defined __GNUC__ || (defined __STDC__ && __STDC__) | |
36 | # define PARAMS(args) args | |
37 | # else | |
38 | # define PARAMS(args) () | |
39 | # endif /* GCC. */ | |
40 | #endif /* Not PARAMS. */ | |
41 | ||
86187531 UD |
42 | #if defined STDC_HEADERS && !defined emacs |
43 | # include <stddef.h> | |
4cca6b86 | 44 | #else |
2b83a2a4 | 45 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ |
86187531 | 46 | # include <sys/types.h> |
4cca6b86 | 47 | #endif |
2b83a2a4 | 48 | |
409dfcea | 49 | #define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC) |
7ce241a0 | 50 | |
51702635 UD |
51 | /* For platform which support the ISO C amendement 1 functionality we |
52 | support user defined character classes. */ | |
409dfcea | 53 | #if defined _LIBC || WIDE_CHAR_SUPPORT |
7ba4fcfc | 54 | /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ |
51702635 | 55 | # include <wchar.h> |
7ba4fcfc | 56 | # include <wctype.h> |
a9ddb793 | 57 | #endif |
2ad4fab2 | 58 | |
e4c785c8 UD |
59 | /* This is for multi byte string support. */ |
60 | #ifdef MBS_SUPPORT | |
61 | # define CHAR_TYPE wchar_t | |
62 | # define US_CHAR_TYPE wchar_t/* unsigned character type */ | |
63 | # define COMPILED_BUFFER_VAR wc_buffer | |
64 | # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */ | |
441f7d1e | 65 | # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_TYPE)+1) |
672fd41b UD |
66 | # define PUT_CHAR(c) \ |
67 | do { \ | |
28d2fb9a | 68 | if (MB_CUR_MAX == 1) \ |
672fd41b UD |
69 | putchar (c); \ |
70 | else \ | |
71 | printf ("%C", (wint_t) c); /* Should we use wide stream?? */ \ | |
72 | } while (0) | |
e4c785c8 UD |
73 | # define TRUE 1 |
74 | # define FALSE 0 | |
75 | #else | |
76 | # define CHAR_TYPE char | |
77 | # define US_CHAR_TYPE unsigned char /* unsigned character type */ | |
78 | # define COMPILED_BUFFER_VAR bufp->buffer | |
79 | # define OFFSET_ADDRESS_SIZE 2 | |
80 | # define PUT_CHAR(c) putchar (c) | |
81 | #endif /* MBS_SUPPORT */ | |
82 | ||
a9ddb793 | 83 | #ifdef _LIBC |
2ad4fab2 UD |
84 | /* We have to keep the namespace clean. */ |
85 | # define regfree(preg) __regfree (preg) | |
86 | # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) | |
87 | # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) | |
88 | # define regerror(errcode, preg, errbuf, errbuf_size) \ | |
89 | __regerror(errcode, preg, errbuf, errbuf_size) | |
90 | # define re_set_registers(bu, re, nu, st, en) \ | |
91 | __re_set_registers (bu, re, nu, st, en) | |
92 | # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ | |
93 | __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
94 | # define re_match(bufp, string, size, pos, regs) \ | |
95 | __re_match (bufp, string, size, pos, regs) | |
96 | # define re_search(bufp, string, size, startpos, range, regs) \ | |
97 | __re_search (bufp, string, size, startpos, range, regs) | |
98 | # define re_compile_pattern(pattern, length, bufp) \ | |
99 | __re_compile_pattern (pattern, length, bufp) | |
100 | # define re_set_syntax(syntax) __re_set_syntax (syntax) | |
101 | # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ | |
102 | __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) | |
103 | # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) | |
104 | ||
a63a3c2c UD |
105 | # define btowc __btowc |
106 | ||
107 | /* We are also using some library internals. */ | |
108 | # include <locale/localeinfo.h> | |
3216711f | 109 | # include <locale/elem-hash.h> |
a63a3c2c | 110 | # include <langinfo.h> |
e4c785c8 | 111 | # include <locale/coll-lookup.h> |
51702635 UD |
112 | #endif |
113 | ||
2b83a2a4 | 114 | /* This is for other GNU distributions with internationalized messages. */ |
86187531 | 115 | #if HAVE_LIBINTL_H || defined _LIBC |
2b83a2a4 | 116 | # include <libintl.h> |
0bb258e3 UD |
117 | # ifdef _LIBC |
118 | # undef gettext | |
119 | # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES) | |
120 | # endif | |
2b83a2a4 RM |
121 | #else |
122 | # define gettext(msgid) (msgid) | |
123 | #endif | |
124 | ||
91c7b85d RM |
125 | #ifndef gettext_noop |
126 | /* This define is so xgettext can find the internationalizable | |
127 | strings. */ | |
86187531 | 128 | # define gettext_noop(String) String |
91c7b85d RM |
129 | #endif |
130 | ||
2b83a2a4 RM |
131 | /* The `emacs' switch turns on certain matching commands |
132 | that make sense only in Emacs. */ | |
133 | #ifdef emacs | |
134 | ||
86187531 UD |
135 | # include "lisp.h" |
136 | # include "buffer.h" | |
137 | # include "syntax.h" | |
2b83a2a4 RM |
138 | |
139 | #else /* not emacs */ | |
140 | ||
141 | /* If we are not linking with Emacs proper, | |
142 | we can't use the relocating allocator | |
143 | even if config.h says that we can. */ | |
86187531 | 144 | # undef REL_ALLOC |
2b83a2a4 | 145 | |
86187531 UD |
146 | # if defined STDC_HEADERS || defined _LIBC |
147 | # include <stdlib.h> | |
148 | # else | |
2b83a2a4 RM |
149 | char *malloc (); |
150 | char *realloc (); | |
86187531 | 151 | # endif |
2b83a2a4 | 152 | |
5bf62f2d RM |
153 | /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. |
154 | If nothing else has been done, use the method below. */ | |
86187531 UD |
155 | # ifdef INHIBIT_STRING_HEADER |
156 | # if !(defined HAVE_BZERO && defined HAVE_BCOPY) | |
157 | # if !defined bzero && !defined bcopy | |
158 | # undef INHIBIT_STRING_HEADER | |
159 | # endif | |
160 | # endif | |
161 | # endif | |
5bf62f2d RM |
162 | |
163 | /* This is the normal way of making sure we have a bcopy and a bzero. | |
164 | This is used in most programs--a few other programs avoid this | |
165 | by defining INHIBIT_STRING_HEADER. */ | |
86187531 UD |
166 | # ifndef INHIBIT_STRING_HEADER |
167 | # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC | |
168 | # include <string.h> | |
2ad4fab2 UD |
169 | # ifndef bzero |
170 | # ifndef _LIBC | |
171 | # define bzero(s, n) (memset (s, '\0', n), (s)) | |
172 | # else | |
173 | # define bzero(s, n) __bzero (s, n) | |
174 | # endif | |
86187531 UD |
175 | # endif |
176 | # else | |
177 | # include <strings.h> | |
178 | # ifndef memcmp | |
179 | # define memcmp(s1, s2, n) bcmp (s1, s2, n) | |
180 | # endif | |
181 | # ifndef memcpy | |
182 | # define memcpy(d, s, n) (bcopy (s, d, n), (d)) | |
183 | # endif | |
184 | # endif | |
185 | # endif | |
2b83a2a4 RM |
186 | |
187 | /* Define the syntax stuff for \<, \>, etc. */ | |
188 | ||
189 | /* This must be nonzero for the wordchar and notwordchar pattern | |
190 | commands in re_match_2. */ | |
86187531 UD |
191 | # ifndef Sword |
192 | # define Sword 1 | |
193 | # endif | |
2b83a2a4 | 194 | |
86187531 UD |
195 | # ifdef SWITCH_ENUM_BUG |
196 | # define SWITCH_ENUM_CAST(x) ((int)(x)) | |
197 | # else | |
198 | # define SWITCH_ENUM_CAST(x) (x) | |
199 | # endif | |
2b83a2a4 | 200 | |
2b83a2a4 | 201 | #endif /* not emacs */ |
bdd5fccd UD |
202 | |
203 | #if defined _LIBC || HAVE_LIMITS_H | |
204 | # include <limits.h> | |
205 | #endif | |
206 | ||
207 | #ifndef MB_LEN_MAX | |
208 | # define MB_LEN_MAX 1 | |
209 | #endif | |
2b83a2a4 RM |
210 | \f |
211 | /* Get the interface, including the syntax bits. */ | |
5c2a0669 | 212 | #include <regex.h> |
2b83a2a4 RM |
213 | |
214 | /* isalpha etc. are used for the character classes. */ | |
215 | #include <ctype.h> | |
216 | ||
217 | /* Jim Meyering writes: | |
218 | ||
219 | "... Some ctype macros are valid only for character codes that | |
220 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
221 | using /bin/cc or gcc but without giving an ansi option). So, all | |
222 | ctype uses should be through macros like ISPRINT... If | |
223 | STDC_HEADERS is defined, then autoconf has verified that the ctype | |
224 | macros don't need to be guarded with references to isascii. ... | |
225 | Defining isascii to 1 should let any compiler worth its salt | |
06698672 UD |
226 | eliminate the && through constant folding." |
227 | Solaris defines some of these symbols so we must undefine them first. */ | |
2b83a2a4 | 228 | |
06698672 | 229 | #undef ISASCII |
86187531 UD |
230 | #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) |
231 | # define ISASCII(c) 1 | |
2b83a2a4 | 232 | #else |
86187531 | 233 | # define ISASCII(c) isascii(c) |
2b83a2a4 RM |
234 | #endif |
235 | ||
236 | #ifdef isblank | |
86187531 | 237 | # define ISBLANK(c) (ISASCII (c) && isblank (c)) |
2b83a2a4 | 238 | #else |
86187531 | 239 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
2b83a2a4 RM |
240 | #endif |
241 | #ifdef isgraph | |
86187531 | 242 | # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) |
2b83a2a4 | 243 | #else |
86187531 | 244 | # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) |
2b83a2a4 RM |
245 | #endif |
246 | ||
06698672 | 247 | #undef ISPRINT |
2b83a2a4 RM |
248 | #define ISPRINT(c) (ISASCII (c) && isprint (c)) |
249 | #define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
250 | #define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
251 | #define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
252 | #define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
253 | #define ISLOWER(c) (ISASCII (c) && islower (c)) | |
254 | #define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
255 | #define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
256 | #define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
257 | #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
258 | ||
4caef86c UD |
259 | #ifdef _tolower |
260 | # define TOLOWER(c) _tolower(c) | |
261 | #else | |
262 | # define TOLOWER(c) tolower(c) | |
263 | #endif | |
264 | ||
2b83a2a4 | 265 | #ifndef NULL |
86187531 | 266 | # define NULL (void *)0 |
2b83a2a4 RM |
267 | #endif |
268 | ||
269 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
270 | since ours (we hope) works properly with all combinations of | |
271 | machines, compilers, `char' and `unsigned char' argument types. | |
272 | (Per Bothner suggested the basic approach.) */ | |
273 | #undef SIGN_EXTEND_CHAR | |
274 | #if __STDC__ | |
86187531 | 275 | # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
2b83a2a4 RM |
276 | #else /* not __STDC__ */ |
277 | /* As in Harbison and Steele. */ | |
86187531 | 278 | # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) |
2b83a2a4 RM |
279 | #endif |
280 | \f | |
2864e767 UD |
281 | #ifndef emacs |
282 | /* How many characters in the character set. */ | |
283 | # define CHAR_SET_SIZE 256 | |
284 | ||
285 | # ifdef SYNTAX_TABLE | |
286 | ||
287 | extern char *re_syntax_table; | |
288 | ||
289 | # else /* not SYNTAX_TABLE */ | |
290 | ||
291 | static char re_syntax_table[CHAR_SET_SIZE]; | |
292 | ||
2d0aea11 UD |
293 | static void init_syntax_once PARAMS ((void)); |
294 | ||
2864e767 UD |
295 | static void |
296 | init_syntax_once () | |
297 | { | |
298 | register int c; | |
299 | static int done = 0; | |
300 | ||
301 | if (done) | |
302 | return; | |
303 | bzero (re_syntax_table, sizeof re_syntax_table); | |
304 | ||
305 | for (c = 0; c < CHAR_SET_SIZE; ++c) | |
306 | if (ISALNUM (c)) | |
307 | re_syntax_table[c] = Sword; | |
308 | ||
309 | re_syntax_table['_'] = Sword; | |
310 | ||
311 | done = 1; | |
312 | } | |
313 | ||
314 | # endif /* not SYNTAX_TABLE */ | |
315 | ||
7186e974 | 316 | # define SYNTAX(c) re_syntax_table[(unsigned char) (c)] |
2864e767 UD |
317 | |
318 | #endif /* emacs */ | |
319 | \f | |
2b83a2a4 RM |
320 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we |
321 | use `alloca' instead of `malloc'. This is because using malloc in | |
322 | re_search* or re_match* could cause memory leaks when C-g is used in | |
323 | Emacs; also, malloc is slower and causes storage fragmentation. On | |
91c7b85d RM |
324 | the other hand, malloc is more portable, and easier to debug. |
325 | ||
2b83a2a4 RM |
326 | Because we sometimes use alloca, some routines have to be macros, |
327 | not functions -- `alloca'-allocated space disappears at the end of the | |
328 | function it is called in. */ | |
329 | ||
330 | #ifdef REGEX_MALLOC | |
331 | ||
86187531 UD |
332 | # define REGEX_ALLOCATE malloc |
333 | # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) | |
334 | # define REGEX_FREE free | |
2b83a2a4 RM |
335 | |
336 | #else /* not REGEX_MALLOC */ | |
337 | ||
338 | /* Emacs already defines alloca, sometimes. */ | |
86187531 | 339 | # ifndef alloca |
2b83a2a4 RM |
340 | |
341 | /* Make alloca work the best possible way. */ | |
86187531 UD |
342 | # ifdef __GNUC__ |
343 | # define alloca __builtin_alloca | |
344 | # else /* not __GNUC__ */ | |
345 | # if HAVE_ALLOCA_H | |
346 | # include <alloca.h> | |
347 | # endif /* HAVE_ALLOCA_H */ | |
348 | # endif /* not __GNUC__ */ | |
2b83a2a4 | 349 | |
86187531 | 350 | # endif /* not alloca */ |
2b83a2a4 | 351 | |
86187531 | 352 | # define REGEX_ALLOCATE alloca |
2b83a2a4 RM |
353 | |
354 | /* Assumes a `char *destination' variable. */ | |
86187531 | 355 | # define REGEX_REALLOCATE(source, osize, nsize) \ |
2b83a2a4 | 356 | (destination = (char *) alloca (nsize), \ |
86187531 | 357 | memcpy (destination, source, osize)) |
2b83a2a4 RM |
358 | |
359 | /* No need to do anything to free, after alloca. */ | |
86187531 | 360 | # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
2b83a2a4 RM |
361 | |
362 | #endif /* not REGEX_MALLOC */ | |
363 | ||
364 | /* Define how to allocate the failure stack. */ | |
365 | ||
86187531 | 366 | #if defined REL_ALLOC && defined REGEX_MALLOC |
ff48a63c | 367 | |
86187531 | 368 | # define REGEX_ALLOCATE_STACK(size) \ |
2b83a2a4 | 369 | r_alloc (&failure_stack_ptr, (size)) |
86187531 | 370 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
2b83a2a4 | 371 | r_re_alloc (&failure_stack_ptr, (nsize)) |
86187531 | 372 | # define REGEX_FREE_STACK(ptr) \ |
2b83a2a4 RM |
373 | r_alloc_free (&failure_stack_ptr) |
374 | ||
ff48a63c | 375 | #else /* not using relocating allocator */ |
2b83a2a4 | 376 | |
86187531 | 377 | # ifdef REGEX_MALLOC |
2b83a2a4 | 378 | |
86187531 UD |
379 | # define REGEX_ALLOCATE_STACK malloc |
380 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) | |
381 | # define REGEX_FREE_STACK free | |
2b83a2a4 | 382 | |
86187531 | 383 | # else /* not REGEX_MALLOC */ |
2b83a2a4 | 384 | |
86187531 | 385 | # define REGEX_ALLOCATE_STACK alloca |
2b83a2a4 | 386 | |
86187531 | 387 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
2b83a2a4 RM |
388 | REGEX_REALLOCATE (source, osize, nsize) |
389 | /* No need to explicitly free anything. */ | |
86187531 | 390 | # define REGEX_FREE_STACK(arg) |
2b83a2a4 | 391 | |
86187531 | 392 | # endif /* not REGEX_MALLOC */ |
ff48a63c | 393 | #endif /* not using relocating allocator */ |
2b83a2a4 RM |
394 | |
395 | ||
396 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
397 | `string1' or just past its end. This works if PTR is NULL, which is | |
398 | a good thing. */ | |
399 | #define FIRST_STRING_P(ptr) \ | |
400 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) | |
401 | ||
402 | /* (Re)Allocate N items of type T using malloc, or fail. */ | |
403 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) | |
404 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) | |
405 | #define RETALLOC_IF(addr, n, t) \ | |
406 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) | |
407 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) | |
408 | ||
409 | #define BYTEWIDTH 8 /* In bits. */ | |
410 | ||
411 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) | |
412 | ||
413 | #undef MAX | |
414 | #undef MIN | |
415 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
416 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
417 | ||
418 | typedef char boolean; | |
419 | #define false 0 | |
420 | #define true 1 | |
421 | ||
07b51ba5 UD |
422 | static int re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp, |
423 | const char *string1, int size1, | |
424 | const char *string2, int size2, | |
425 | int pos, | |
426 | struct re_registers *regs, | |
427 | int stop)); | |
2b83a2a4 RM |
428 | \f |
429 | /* These are the command codes that appear in compiled regular | |
430 | expressions. Some opcodes are followed by argument bytes. A | |
431 | command code can specify any interpretation whatsoever for its | |
432 | arguments. Zero bytes may appear in the compiled regular expression. */ | |
433 | ||
434 | typedef enum | |
435 | { | |
436 | no_op = 0, | |
437 | ||
438 | /* Succeed right away--no more backtracking. */ | |
439 | succeed, | |
440 | ||
441 | /* Followed by one byte giving n, then by n literal bytes. */ | |
442 | exactn, | |
443 | ||
e4c785c8 UD |
444 | #ifdef MBS_SUPPORT |
445 | /* Same as exactn, but contains binary data. */ | |
446 | exactn_bin, | |
447 | #endif | |
448 | ||
2b83a2a4 RM |
449 | /* Matches any (more or less) character. */ |
450 | anychar, | |
451 | ||
452 | /* Matches any one char belonging to specified set. First | |
453 | following byte is number of bitmap bytes. Then come bytes | |
454 | for a bitmap saying which chars are in. Bits in each byte | |
455 | are ordered low-bit-first. A character is in the set if its | |
456 | bit is 1. A character too large to have a bit in the map is | |
457 | automatically not in the set. */ | |
e4c785c8 UD |
458 | /* ifdef MBS_SUPPORT, following element is length of character |
459 | classes, length of collating symbols, length of equivalence | |
460 | classes, length of character ranges, and length of characters. | |
461 | Next, character class element, collating symbols elements, | |
462 | equivalence class elements, range elements, and character | |
463 | elements follow. | |
464 | See regex_compile function. */ | |
2b83a2a4 RM |
465 | charset, |
466 | ||
467 | /* Same parameters as charset, but match any character that is | |
468 | not one of those specified. */ | |
469 | charset_not, | |
470 | ||
471 | /* Start remembering the text that is matched, for storing in a | |
472 | register. Followed by one byte with the register number, in | |
473 | the range 0 to one less than the pattern buffer's re_nsub | |
474 | field. Then followed by one byte with the number of groups | |
475 | inner to this one. (This last has to be part of the | |
476 | start_memory only because we need it in the on_failure_jump | |
477 | of re_match_2.) */ | |
478 | start_memory, | |
479 | ||
480 | /* Stop remembering the text that is matched and store it in a | |
481 | memory register. Followed by one byte with the register | |
482 | number, in the range 0 to one less than `re_nsub' in the | |
483 | pattern buffer, and one byte with the number of inner groups, | |
484 | just like `start_memory'. (We need the number of inner | |
485 | groups here because we don't have any easy way of finding the | |
486 | corresponding start_memory when we're at a stop_memory.) */ | |
487 | stop_memory, | |
488 | ||
489 | /* Match a duplicate of something remembered. Followed by one | |
490 | byte containing the register number. */ | |
491 | duplicate, | |
492 | ||
493 | /* Fail unless at beginning of line. */ | |
494 | begline, | |
495 | ||
496 | /* Fail unless at end of line. */ | |
497 | endline, | |
498 | ||
499 | /* Succeeds if at beginning of buffer (if emacs) or at beginning | |
500 | of string to be matched (if not). */ | |
501 | begbuf, | |
502 | ||
503 | /* Analogously, for end of buffer/string. */ | |
504 | endbuf, | |
91c7b85d | 505 | |
2b83a2a4 | 506 | /* Followed by two byte relative address to which to jump. */ |
91c7b85d | 507 | jump, |
2b83a2a4 RM |
508 | |
509 | /* Same as jump, but marks the end of an alternative. */ | |
510 | jump_past_alt, | |
511 | ||
512 | /* Followed by two-byte relative address of place to resume at | |
513 | in case of failure. */ | |
e4c785c8 | 514 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
2b83a2a4 | 515 | on_failure_jump, |
91c7b85d | 516 | |
2b83a2a4 RM |
517 | /* Like on_failure_jump, but pushes a placeholder instead of the |
518 | current string position when executed. */ | |
519 | on_failure_keep_string_jump, | |
91c7b85d | 520 | |
2b83a2a4 RM |
521 | /* Throw away latest failure point and then jump to following |
522 | two-byte relative address. */ | |
e4c785c8 | 523 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
2b83a2a4 RM |
524 | pop_failure_jump, |
525 | ||
526 | /* Change to pop_failure_jump if know won't have to backtrack to | |
527 | match; otherwise change to jump. This is used to jump | |
528 | back to the beginning of a repeat. If what follows this jump | |
529 | clearly won't match what the repeat does, such that we can be | |
530 | sure that there is no use backtracking out of repetitions | |
531 | already matched, then we change it to a pop_failure_jump. | |
532 | Followed by two-byte address. */ | |
e4c785c8 | 533 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
2b83a2a4 RM |
534 | maybe_pop_jump, |
535 | ||
536 | /* Jump to following two-byte address, and push a dummy failure | |
537 | point. This failure point will be thrown away if an attempt | |
538 | is made to use it for a failure. A `+' construct makes this | |
539 | before the first repeat. Also used as an intermediary kind | |
540 | of jump when compiling an alternative. */ | |
e4c785c8 | 541 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
2b83a2a4 RM |
542 | dummy_failure_jump, |
543 | ||
544 | /* Push a dummy failure point and continue. Used at the end of | |
545 | alternatives. */ | |
546 | push_dummy_failure, | |
547 | ||
548 | /* Followed by two-byte relative address and two-byte number n. | |
549 | After matching N times, jump to the address upon failure. */ | |
e4c785c8 | 550 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
2b83a2a4 RM |
551 | succeed_n, |
552 | ||
553 | /* Followed by two-byte relative address, and two-byte number n. | |
554 | Jump to the address N times, then fail. */ | |
e4c785c8 | 555 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
2b83a2a4 RM |
556 | jump_n, |
557 | ||
558 | /* Set the following two-byte relative address to the | |
559 | subsequent two-byte number. The address *includes* the two | |
560 | bytes of number. */ | |
e4c785c8 | 561 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
2b83a2a4 RM |
562 | set_number_at, |
563 | ||
564 | wordchar, /* Matches any word-constituent character. */ | |
565 | notwordchar, /* Matches any char that is not a word-constituent. */ | |
566 | ||
567 | wordbeg, /* Succeeds if at word beginning. */ | |
568 | wordend, /* Succeeds if at word end. */ | |
569 | ||
570 | wordbound, /* Succeeds if at a word boundary. */ | |
571 | notwordbound /* Succeeds if not at a word boundary. */ | |
572 | ||
573 | #ifdef emacs | |
574 | ,before_dot, /* Succeeds if before point. */ | |
575 | at_dot, /* Succeeds if at point. */ | |
576 | after_dot, /* Succeeds if after point. */ | |
577 | ||
578 | /* Matches any character whose syntax is specified. Followed by | |
579 | a byte which contains a syntax code, e.g., Sword. */ | |
580 | syntaxspec, | |
581 | ||
582 | /* Matches any character whose syntax is not that specified. */ | |
583 | notsyntaxspec | |
584 | #endif /* emacs */ | |
585 | } re_opcode_t; | |
586 | \f | |
587 | /* Common operations on the compiled pattern. */ | |
588 | ||
589 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
e4c785c8 | 590 | /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ |
2b83a2a4 | 591 | |
e4c785c8 UD |
592 | #ifdef MBS_SUPPORT |
593 | # define STORE_NUMBER(destination, number) \ | |
594 | do { \ | |
595 | *(destination) = (US_CHAR_TYPE)(number); \ | |
596 | } while (0) | |
597 | #else | |
598 | # define STORE_NUMBER(destination, number) \ | |
2b83a2a4 RM |
599 | do { \ |
600 | (destination)[0] = (number) & 0377; \ | |
601 | (destination)[1] = (number) >> 8; \ | |
602 | } while (0) | |
e4c785c8 | 603 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
604 | |
605 | /* Same as STORE_NUMBER, except increment DESTINATION to | |
606 | the byte after where the number is stored. Therefore, DESTINATION | |
607 | must be an lvalue. */ | |
e4c785c8 | 608 | /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ |
2b83a2a4 RM |
609 | |
610 | #define STORE_NUMBER_AND_INCR(destination, number) \ | |
611 | do { \ | |
612 | STORE_NUMBER (destination, number); \ | |
e4c785c8 | 613 | (destination) += OFFSET_ADDRESS_SIZE; \ |
2b83a2a4 RM |
614 | } while (0) |
615 | ||
616 | /* Put into DESTINATION a number stored in two contiguous bytes starting | |
617 | at SOURCE. */ | |
e4c785c8 | 618 | /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ |
2b83a2a4 | 619 | |
e4c785c8 UD |
620 | #ifdef MBS_SUPPORT |
621 | # define EXTRACT_NUMBER(destination, source) \ | |
622 | do { \ | |
623 | (destination) = *(source); \ | |
624 | } while (0) | |
625 | #else | |
626 | # define EXTRACT_NUMBER(destination, source) \ | |
2b83a2a4 RM |
627 | do { \ |
628 | (destination) = *(source) & 0377; \ | |
629 | (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
630 | } while (0) | |
e4c785c8 | 631 | #endif |
2b83a2a4 RM |
632 | |
633 | #ifdef DEBUG | |
e4c785c8 | 634 | static void extract_number _RE_ARGS ((int *dest, US_CHAR_TYPE *source)); |
2b83a2a4 RM |
635 | static void |
636 | extract_number (dest, source) | |
637 | int *dest; | |
e4c785c8 | 638 | US_CHAR_TYPE *source; |
2b83a2a4 | 639 | { |
e4c785c8 UD |
640 | #ifdef MBS_SUPPORT |
641 | *dest = *source; | |
642 | #else | |
91c7b85d | 643 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
2b83a2a4 RM |
644 | *dest = *source & 0377; |
645 | *dest += temp << 8; | |
e4c785c8 | 646 | #endif |
2b83a2a4 RM |
647 | } |
648 | ||
86187531 UD |
649 | # ifndef EXTRACT_MACROS /* To debug the macros. */ |
650 | # undef EXTRACT_NUMBER | |
651 | # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
652 | # endif /* not EXTRACT_MACROS */ | |
2b83a2a4 RM |
653 | |
654 | #endif /* DEBUG */ | |
655 | ||
656 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
657 | SOURCE must be an lvalue. */ | |
658 | ||
659 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
660 | do { \ | |
661 | EXTRACT_NUMBER (destination, source); \ | |
e4c785c8 | 662 | (source) += OFFSET_ADDRESS_SIZE; \ |
2b83a2a4 RM |
663 | } while (0) |
664 | ||
665 | #ifdef DEBUG | |
4cca6b86 | 666 | static void extract_number_and_incr _RE_ARGS ((int *destination, |
e4c785c8 | 667 | US_CHAR_TYPE **source)); |
2b83a2a4 RM |
668 | static void |
669 | extract_number_and_incr (destination, source) | |
670 | int *destination; | |
e4c785c8 | 671 | US_CHAR_TYPE **source; |
91c7b85d | 672 | { |
2b83a2a4 | 673 | extract_number (destination, *source); |
e4c785c8 | 674 | *source += OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
675 | } |
676 | ||
86187531 UD |
677 | # ifndef EXTRACT_MACROS |
678 | # undef EXTRACT_NUMBER_AND_INCR | |
679 | # define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
2b83a2a4 | 680 | extract_number_and_incr (&dest, &src) |
86187531 | 681 | # endif /* not EXTRACT_MACROS */ |
2b83a2a4 RM |
682 | |
683 | #endif /* DEBUG */ | |
684 | \f | |
685 | /* If DEBUG is defined, Regex prints many voluminous messages about what | |
686 | it is doing (if the variable `debug' is nonzero). If linked with the | |
687 | main program in `iregex.c', you can enter patterns and strings | |
688 | interactively. And if linked with the main program in `main.c' and | |
689 | the other test files, you can run the already-written tests. */ | |
690 | ||
691 | #ifdef DEBUG | |
692 | ||
693 | /* We use standard I/O for debugging. */ | |
86187531 | 694 | # include <stdio.h> |
2b83a2a4 RM |
695 | |
696 | /* It is useful to test things that ``must'' be true when debugging. */ | |
86187531 | 697 | # include <assert.h> |
2b83a2a4 | 698 | |
c4563d2d | 699 | static int debug; |
2b83a2a4 | 700 | |
86187531 UD |
701 | # define DEBUG_STATEMENT(e) e |
702 | # define DEBUG_PRINT1(x) if (debug) printf (x) | |
703 | # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) | |
704 | # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) | |
705 | # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) | |
706 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ | |
2b83a2a4 | 707 | if (debug) print_partial_compiled_pattern (s, e) |
86187531 | 708 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
2b83a2a4 RM |
709 | if (debug) print_double_string (w, s1, sz1, s2, sz2) |
710 | ||
711 | ||
712 | /* Print the fastmap in human-readable form. */ | |
713 | ||
714 | void | |
715 | print_fastmap (fastmap) | |
716 | char *fastmap; | |
717 | { | |
718 | unsigned was_a_range = 0; | |
91c7b85d RM |
719 | unsigned i = 0; |
720 | ||
2b83a2a4 RM |
721 | while (i < (1 << BYTEWIDTH)) |
722 | { | |
723 | if (fastmap[i++]) | |
724 | { | |
725 | was_a_range = 0; | |
726 | putchar (i - 1); | |
727 | while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
728 | { | |
729 | was_a_range = 1; | |
730 | i++; | |
731 | } | |
732 | if (was_a_range) | |
733 | { | |
734 | printf ("-"); | |
735 | putchar (i - 1); | |
736 | } | |
737 | } | |
738 | } | |
91c7b85d | 739 | putchar ('\n'); |
2b83a2a4 RM |
740 | } |
741 | ||
742 | ||
743 | /* Print a compiled pattern string in human-readable form, starting at | |
744 | the START pointer into it and ending just before the pointer END. */ | |
745 | ||
746 | void | |
747 | print_partial_compiled_pattern (start, end) | |
e4c785c8 UD |
748 | US_CHAR_TYPE *start; |
749 | US_CHAR_TYPE *end; | |
2b83a2a4 RM |
750 | { |
751 | int mcnt, mcnt2; | |
e4c785c8 UD |
752 | US_CHAR_TYPE *p1; |
753 | US_CHAR_TYPE *p = start; | |
754 | US_CHAR_TYPE *pend = end; | |
2b83a2a4 RM |
755 | |
756 | if (start == NULL) | |
757 | { | |
758 | printf ("(null)\n"); | |
759 | return; | |
760 | } | |
91c7b85d | 761 | |
2b83a2a4 RM |
762 | /* Loop over pattern commands. */ |
763 | while (p < pend) | |
764 | { | |
c7e41631 | 765 | #ifdef _LIBC |
672fd41b | 766 | printf ("%td:\t", p - start); |
c7e41631 UD |
767 | #else |
768 | printf ("%ld:\t", (long int) (p - start)); | |
769 | #endif | |
2b83a2a4 RM |
770 | |
771 | switch ((re_opcode_t) *p++) | |
772 | { | |
773 | case no_op: | |
774 | printf ("/no_op"); | |
775 | break; | |
776 | ||
777 | case exactn: | |
778 | mcnt = *p++; | |
779 | printf ("/exactn/%d", mcnt); | |
780 | do | |
781 | { | |
782 | putchar ('/'); | |
e4c785c8 UD |
783 | PUT_CHAR (*p++); |
784 | } | |
785 | while (--mcnt); | |
786 | break; | |
787 | ||
788 | #ifdef MBS_SUPPORT | |
789 | case exactn_bin: | |
790 | mcnt = *p++; | |
791 | printf ("/exactn_bin/%d", mcnt); | |
792 | do | |
793 | { | |
672fd41b | 794 | printf("/%lx", (long int) *p++); |
2b83a2a4 RM |
795 | } |
796 | while (--mcnt); | |
797 | break; | |
e4c785c8 | 798 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
799 | |
800 | case start_memory: | |
801 | mcnt = *p++; | |
672fd41b | 802 | printf ("/start_memory/%d/%ld", mcnt, (long int) *p++); |
2b83a2a4 RM |
803 | break; |
804 | ||
805 | case stop_memory: | |
806 | mcnt = *p++; | |
672fd41b | 807 | printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++); |
2b83a2a4 RM |
808 | break; |
809 | ||
810 | case duplicate: | |
672fd41b | 811 | printf ("/duplicate/%ld", (long int) *p++); |
2b83a2a4 RM |
812 | break; |
813 | ||
814 | case anychar: | |
815 | printf ("/anychar"); | |
816 | break; | |
817 | ||
818 | case charset: | |
819 | case charset_not: | |
820 | { | |
e4c785c8 UD |
821 | #ifdef MBS_SUPPORT |
822 | int i, length; | |
823 | wchar_t *workp = p; | |
824 | printf ("/charset [%s", | |
825 | (re_opcode_t) *(workp - 1) == charset_not ? "^" : ""); | |
826 | p += 5; | |
827 | length = *workp++; /* the length of char_classes */ | |
828 | for (i=0 ; i<length ; i++) | |
672fd41b | 829 | printf("[:%lx:]", (long int) *p++); |
e4c785c8 UD |
830 | length = *workp++; /* the length of collating_symbol */ |
831 | for (i=0 ; i<length ;) | |
832 | { | |
833 | printf("[."); | |
834 | while(*p != 0) | |
835 | PUT_CHAR((i++,*p++)); | |
836 | i++,p++; | |
837 | printf(".]"); | |
838 | } | |
839 | length = *workp++; /* the length of equivalence_class */ | |
840 | for (i=0 ; i<length ;) | |
841 | { | |
842 | printf("[="); | |
843 | while(*p != 0) | |
844 | PUT_CHAR((i++,*p++)); | |
845 | i++,p++; | |
846 | printf("=]"); | |
847 | } | |
848 | length = *workp++; /* the length of char_range */ | |
849 | for (i=0 ; i<length ; i++) | |
850 | { | |
851 | wchar_t range_start = *p++; | |
852 | wchar_t range_end = *p++; | |
672fd41b UD |
853 | if (MB_CUR_MAX == 1) |
854 | printf("%c-%c", (char) range_start, (char) range_end); | |
855 | else | |
856 | printf("%C-%C", (wint_t) range_start, (wint_t) range_end); | |
e4c785c8 UD |
857 | } |
858 | length = *workp++; /* the length of char */ | |
859 | for (i=0 ; i<length ; i++) | |
672fd41b UD |
860 | if (MB_CUR_MAX == 1) |
861 | putchar (*p++); | |
862 | else | |
863 | printf("%C", (wint_t) *p++); | |
e4c785c8 UD |
864 | putchar (']'); |
865 | #else | |
2b83a2a4 RM |
866 | register int c, last = -100; |
867 | register int in_range = 0; | |
868 | ||
869 | printf ("/charset [%s", | |
870 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); | |
91c7b85d | 871 | |
2b83a2a4 RM |
872 | assert (p + *p < pend); |
873 | ||
874 | for (c = 0; c < 256; c++) | |
875 | if (c / 8 < *p | |
876 | && (p[1 + (c/8)] & (1 << (c % 8)))) | |
877 | { | |
878 | /* Are we starting a range? */ | |
879 | if (last + 1 == c && ! in_range) | |
880 | { | |
881 | putchar ('-'); | |
882 | in_range = 1; | |
883 | } | |
884 | /* Have we broken a range? */ | |
885 | else if (last + 1 != c && in_range) | |
886 | { | |
887 | putchar (last); | |
888 | in_range = 0; | |
889 | } | |
91c7b85d | 890 | |
2b83a2a4 RM |
891 | if (! in_range) |
892 | putchar (c); | |
893 | ||
894 | last = c; | |
895 | } | |
896 | ||
897 | if (in_range) | |
898 | putchar (last); | |
899 | ||
900 | putchar (']'); | |
901 | ||
902 | p += 1 + *p; | |
e4c785c8 | 903 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
904 | } |
905 | break; | |
906 | ||
907 | case begline: | |
908 | printf ("/begline"); | |
909 | break; | |
910 | ||
911 | case endline: | |
912 | printf ("/endline"); | |
913 | break; | |
914 | ||
915 | case on_failure_jump: | |
916 | extract_number_and_incr (&mcnt, &p); | |
c7e41631 | 917 | #ifdef _LIBC |
672fd41b | 918 | printf ("/on_failure_jump to %td", p + mcnt - start); |
c7e41631 UD |
919 | #else |
920 | printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start)); | |
921 | #endif | |
2b83a2a4 RM |
922 | break; |
923 | ||
924 | case on_failure_keep_string_jump: | |
925 | extract_number_and_incr (&mcnt, &p); | |
c7e41631 | 926 | #ifdef _LIBC |
672fd41b | 927 | printf ("/on_failure_keep_string_jump to %td", p + mcnt - start); |
c7e41631 UD |
928 | #else |
929 | printf ("/on_failure_keep_string_jump to %ld", | |
930 | (long int) (p + mcnt - start)); | |
931 | #endif | |
2b83a2a4 RM |
932 | break; |
933 | ||
934 | case dummy_failure_jump: | |
935 | extract_number_and_incr (&mcnt, &p); | |
c7e41631 | 936 | #ifdef _LIBC |
672fd41b | 937 | printf ("/dummy_failure_jump to %td", p + mcnt - start); |
c7e41631 UD |
938 | #else |
939 | printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start)); | |
940 | #endif | |
2b83a2a4 RM |
941 | break; |
942 | ||
943 | case push_dummy_failure: | |
944 | printf ("/push_dummy_failure"); | |
945 | break; | |
91c7b85d | 946 | |
2b83a2a4 RM |
947 | case maybe_pop_jump: |
948 | extract_number_and_incr (&mcnt, &p); | |
c7e41631 | 949 | #ifdef _LIBC |
672fd41b | 950 | printf ("/maybe_pop_jump to %td", p + mcnt - start); |
c7e41631 UD |
951 | #else |
952 | printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start)); | |
953 | #endif | |
2b83a2a4 RM |
954 | break; |
955 | ||
956 | case pop_failure_jump: | |
957 | extract_number_and_incr (&mcnt, &p); | |
c7e41631 | 958 | #ifdef _LIBC |
672fd41b | 959 | printf ("/pop_failure_jump to %td", p + mcnt - start); |
c7e41631 UD |
960 | #else |
961 | printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start)); | |
962 | #endif | |
91c7b85d RM |
963 | break; |
964 | ||
2b83a2a4 RM |
965 | case jump_past_alt: |
966 | extract_number_and_incr (&mcnt, &p); | |
c7e41631 | 967 | #ifdef _LIBC |
672fd41b | 968 | printf ("/jump_past_alt to %td", p + mcnt - start); |
c7e41631 UD |
969 | #else |
970 | printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start)); | |
971 | #endif | |
91c7b85d RM |
972 | break; |
973 | ||
2b83a2a4 RM |
974 | case jump: |
975 | extract_number_and_incr (&mcnt, &p); | |
c7e41631 | 976 | #ifdef _LIBC |
672fd41b | 977 | printf ("/jump to %td", p + mcnt - start); |
c7e41631 UD |
978 | #else |
979 | printf ("/jump to %ld", (long int) (p + mcnt - start)); | |
980 | #endif | |
2b83a2a4 RM |
981 | break; |
982 | ||
91c7b85d | 983 | case succeed_n: |
2b83a2a4 | 984 | extract_number_and_incr (&mcnt, &p); |
5929563f | 985 | p1 = p + mcnt; |
2b83a2a4 | 986 | extract_number_and_incr (&mcnt2, &p); |
c7e41631 | 987 | #ifdef _LIBC |
672fd41b | 988 | printf ("/succeed_n to %td, %d times", p1 - start, mcnt2); |
c7e41631 UD |
989 | #else |
990 | printf ("/succeed_n to %ld, %d times", | |
991 | (long int) (p1 - start), mcnt2); | |
992 | #endif | |
2b83a2a4 | 993 | break; |
91c7b85d RM |
994 | |
995 | case jump_n: | |
2b83a2a4 | 996 | extract_number_and_incr (&mcnt, &p); |
5929563f | 997 | p1 = p + mcnt; |
2b83a2a4 | 998 | extract_number_and_incr (&mcnt2, &p); |
5929563f | 999 | printf ("/jump_n to %d, %d times", p1 - start, mcnt2); |
2b83a2a4 | 1000 | break; |
91c7b85d RM |
1001 | |
1002 | case set_number_at: | |
2b83a2a4 | 1003 | extract_number_and_incr (&mcnt, &p); |
5929563f | 1004 | p1 = p + mcnt; |
2b83a2a4 | 1005 | extract_number_and_incr (&mcnt2, &p); |
c7e41631 | 1006 | #ifdef _LIBC |
672fd41b | 1007 | printf ("/set_number_at location %td to %d", p1 - start, mcnt2); |
c7e41631 UD |
1008 | #else |
1009 | printf ("/set_number_at location %ld to %d", | |
1010 | (long int) (p1 - start), mcnt2); | |
1011 | #endif | |
2b83a2a4 | 1012 | break; |
91c7b85d | 1013 | |
2b83a2a4 RM |
1014 | case wordbound: |
1015 | printf ("/wordbound"); | |
1016 | break; | |
1017 | ||
1018 | case notwordbound: | |
1019 | printf ("/notwordbound"); | |
1020 | break; | |
1021 | ||
1022 | case wordbeg: | |
1023 | printf ("/wordbeg"); | |
1024 | break; | |
91c7b85d | 1025 | |
2b83a2a4 RM |
1026 | case wordend: |
1027 | printf ("/wordend"); | |
e4c785c8 | 1028 | break; |
91c7b85d | 1029 | |
86187531 | 1030 | # ifdef emacs |
2b83a2a4 RM |
1031 | case before_dot: |
1032 | printf ("/before_dot"); | |
1033 | break; | |
1034 | ||
1035 | case at_dot: | |
1036 | printf ("/at_dot"); | |
1037 | break; | |
1038 | ||
1039 | case after_dot: | |
1040 | printf ("/after_dot"); | |
1041 | break; | |
1042 | ||
1043 | case syntaxspec: | |
1044 | printf ("/syntaxspec"); | |
1045 | mcnt = *p++; | |
1046 | printf ("/%d", mcnt); | |
1047 | break; | |
91c7b85d | 1048 | |
2b83a2a4 RM |
1049 | case notsyntaxspec: |
1050 | printf ("/notsyntaxspec"); | |
1051 | mcnt = *p++; | |
1052 | printf ("/%d", mcnt); | |
1053 | break; | |
86187531 | 1054 | # endif /* emacs */ |
2b83a2a4 RM |
1055 | |
1056 | case wordchar: | |
1057 | printf ("/wordchar"); | |
1058 | break; | |
91c7b85d | 1059 | |
2b83a2a4 RM |
1060 | case notwordchar: |
1061 | printf ("/notwordchar"); | |
1062 | break; | |
1063 | ||
1064 | case begbuf: | |
1065 | printf ("/begbuf"); | |
1066 | break; | |
1067 | ||
1068 | case endbuf: | |
1069 | printf ("/endbuf"); | |
1070 | break; | |
1071 | ||
1072 | default: | |
672fd41b | 1073 | printf ("?%ld", (long int) *(p-1)); |
2b83a2a4 RM |
1074 | } |
1075 | ||
1076 | putchar ('\n'); | |
1077 | } | |
1078 | ||
c7e41631 | 1079 | #ifdef _LIBC |
672fd41b | 1080 | printf ("%td:\tend of pattern.\n", p - start); |
c7e41631 UD |
1081 | #else |
1082 | printf ("%ld:\tend of pattern.\n", (long int) (p - start)); | |
1083 | #endif | |
2b83a2a4 RM |
1084 | } |
1085 | ||
1086 | ||
1087 | void | |
1088 | print_compiled_pattern (bufp) | |
1089 | struct re_pattern_buffer *bufp; | |
1090 | { | |
e4c785c8 | 1091 | US_CHAR_TYPE *buffer = (US_CHAR_TYPE*) bufp->buffer; |
2b83a2a4 | 1092 | |
e4c785c8 UD |
1093 | print_partial_compiled_pattern (buffer, buffer |
1094 | + bufp->used / sizeof(US_CHAR_TYPE)); | |
5929563f UD |
1095 | printf ("%ld bytes used/%ld bytes allocated.\n", |
1096 | bufp->used, bufp->allocated); | |
2b83a2a4 RM |
1097 | |
1098 | if (bufp->fastmap_accurate && bufp->fastmap) | |
1099 | { | |
1100 | printf ("fastmap: "); | |
1101 | print_fastmap (bufp->fastmap); | |
1102 | } | |
1103 | ||
c7e41631 UD |
1104 | #ifdef _LIBC |
1105 | printf ("re_nsub: %Zd\t", bufp->re_nsub); | |
1106 | #else | |
1107 | printf ("re_nsub: %ld\t", (long int) bufp->re_nsub); | |
1108 | #endif | |
2b83a2a4 RM |
1109 | printf ("regs_alloc: %d\t", bufp->regs_allocated); |
1110 | printf ("can_be_null: %d\t", bufp->can_be_null); | |
1111 | printf ("newline_anchor: %d\n", bufp->newline_anchor); | |
1112 | printf ("no_sub: %d\t", bufp->no_sub); | |
1113 | printf ("not_bol: %d\t", bufp->not_bol); | |
1114 | printf ("not_eol: %d\t", bufp->not_eol); | |
5929563f | 1115 | printf ("syntax: %lx\n", bufp->syntax); |
2b83a2a4 RM |
1116 | /* Perhaps we should print the translate table? */ |
1117 | } | |
1118 | ||
1119 | ||
1120 | void | |
1121 | print_double_string (where, string1, size1, string2, size2) | |
e4c785c8 UD |
1122 | const CHAR_TYPE *where; |
1123 | const CHAR_TYPE *string1; | |
1124 | const CHAR_TYPE *string2; | |
2b83a2a4 RM |
1125 | int size1; |
1126 | int size2; | |
1127 | { | |
5929563f | 1128 | int this_char; |
91c7b85d | 1129 | |
2b83a2a4 RM |
1130 | if (where == NULL) |
1131 | printf ("(null)"); | |
1132 | else | |
1133 | { | |
1134 | if (FIRST_STRING_P (where)) | |
1135 | { | |
1136 | for (this_char = where - string1; this_char < size1; this_char++) | |
e4c785c8 | 1137 | PUT_CHAR (string1[this_char]); |
2b83a2a4 | 1138 | |
91c7b85d | 1139 | where = string2; |
2b83a2a4 RM |
1140 | } |
1141 | ||
1142 | for (this_char = where - string2; this_char < size2; this_char++) | |
e4c785c8 | 1143 | PUT_CHAR (string2[this_char]); |
2b83a2a4 RM |
1144 | } |
1145 | } | |
1146 | ||
4cca6b86 UD |
1147 | void |
1148 | printchar (c) | |
1149 | int c; | |
1150 | { | |
1151 | putc (c, stderr); | |
1152 | } | |
1153 | ||
2b83a2a4 RM |
1154 | #else /* not DEBUG */ |
1155 | ||
86187531 UD |
1156 | # undef assert |
1157 | # define assert(e) | |
2b83a2a4 | 1158 | |
86187531 UD |
1159 | # define DEBUG_STATEMENT(e) |
1160 | # define DEBUG_PRINT1(x) | |
1161 | # define DEBUG_PRINT2(x1, x2) | |
1162 | # define DEBUG_PRINT3(x1, x2, x3) | |
1163 | # define DEBUG_PRINT4(x1, x2, x3, x4) | |
1164 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
1165 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
2b83a2a4 RM |
1166 | |
1167 | #endif /* not DEBUG */ | |
1168 | \f | |
e4c785c8 UD |
1169 | #ifdef MBS_SUPPORT |
1170 | /* This convert a multibyte string to a wide character string. | |
1171 | And write their correspondances to offset_buffer(see below) | |
1172 | and write whether each wchar_t is binary data to is_binary. | |
1173 | This assume invalid multibyte sequences as binary data. | |
1174 | We assume offset_buffer and is_binary is already allocated | |
1175 | enough space. */ | |
d4620e04 AJ |
1176 | |
1177 | static size_t convert_mbs_to_wcs (CHAR_TYPE *dest, const unsigned char* src, | |
1178 | size_t len, int *offset_buffer, | |
770d454d | 1179 | char *is_binary); |
d4620e04 | 1180 | static size_t |
e4c785c8 UD |
1181 | convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary) |
1182 | CHAR_TYPE *dest; | |
1183 | const unsigned char* src; | |
1184 | size_t len; /* the length of multibyte string. */ | |
1185 | ||
1186 | /* It hold correspondances between src(char string) and | |
1187 | dest(wchar_t string) for optimization. | |
1188 | e.g. src = "xxxyzz" | |
1189 | dest = {'X', 'Y', 'Z'} | |
1190 | (each "xxx", "y" and "zz" represent one multibyte character | |
1191 | corresponding to 'X', 'Y' and 'Z'.) | |
1192 | offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")} | |
1193 | = {0, 3, 4, 6} | |
1194 | */ | |
1195 | int *offset_buffer; | |
770d454d | 1196 | char *is_binary; |
e4c785c8 UD |
1197 | { |
1198 | wchar_t *pdest = dest; | |
1199 | const unsigned char *psrc = src; | |
1200 | size_t wc_count = 0; | |
1201 | ||
1202 | if (MB_CUR_MAX == 1) | |
1203 | { /* We don't need conversion. */ | |
1204 | for ( ; wc_count < len ; ++wc_count) | |
1205 | { | |
1206 | *pdest++ = *psrc++; | |
1207 | is_binary[wc_count] = FALSE; | |
1208 | offset_buffer[wc_count] = wc_count; | |
1209 | } | |
1210 | offset_buffer[wc_count] = wc_count; | |
1211 | } | |
1212 | else | |
1213 | { | |
1214 | /* We need conversion. */ | |
1215 | mbstate_t mbs; | |
1216 | int consumed; | |
1217 | size_t mb_remain = len; | |
1218 | size_t mb_count = 0; | |
1219 | ||
1220 | /* Initialize the conversion state. */ | |
1221 | memset (&mbs, 0, sizeof (mbstate_t)); | |
1222 | ||
1223 | offset_buffer[0] = 0; | |
1224 | for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed, | |
1225 | psrc += consumed) | |
1226 | { | |
1227 | consumed = mbrtowc (pdest, psrc, mb_remain, &mbs); | |
1228 | ||
1229 | if (consumed <= 0) | |
1230 | /* failed to convert. maybe src contains binary data. | |
1231 | So we consume 1 byte manualy. */ | |
1232 | { | |
1233 | *pdest = *psrc; | |
1234 | consumed = 1; | |
1235 | is_binary[wc_count] = TRUE; | |
1236 | } | |
1237 | else | |
1238 | is_binary[wc_count] = FALSE; | |
1239 | /* In sjis encoding, we use yen sign as escape character in | |
1240 | place of reverse solidus. So we convert 0x5c(yen sign in | |
1241 | sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse | |
1242 | solidus in UCS2). */ | |
1243 | if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5) | |
1244 | *pdest = (wchar_t) *psrc; | |
1245 | ||
1246 | offset_buffer[wc_count + 1] = mb_count += consumed; | |
1247 | } | |
1248 | } | |
1249 | ||
1250 | return wc_count; | |
1251 | } | |
1252 | ||
1253 | #endif /* MBS_SUPPORT */ | |
1254 | ||
2b83a2a4 RM |
1255 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can |
1256 | also be assigned to arbitrarily: each pattern buffer stores its own | |
1257 | syntax, so it can be changed between regex compilations. */ | |
b9337b6a UD |
1258 | /* This has no initializer because initialized variables in Emacs |
1259 | become read-only after dumping. */ | |
2b83a2a4 RM |
1260 | reg_syntax_t re_syntax_options; |
1261 | ||
1262 | ||
1263 | /* Specify the precise syntax of regexps for compilation. This provides | |
1264 | for compatibility for various utilities which historically have | |
1265 | different, incompatible syntaxes. | |
1266 | ||
1267 | The argument SYNTAX is a bit mask comprised of the various bits | |
1268 | defined in regex.h. We return the old syntax. */ | |
1269 | ||
1270 | reg_syntax_t | |
1271 | re_set_syntax (syntax) | |
1272 | reg_syntax_t syntax; | |
1273 | { | |
1274 | reg_syntax_t ret = re_syntax_options; | |
91c7b85d | 1275 | |
2b83a2a4 | 1276 | re_syntax_options = syntax; |
51702635 UD |
1277 | #ifdef DEBUG |
1278 | if (syntax & RE_DEBUG) | |
1279 | debug = 1; | |
1280 | else if (debug) /* was on but now is not */ | |
1281 | debug = 0; | |
1282 | #endif /* DEBUG */ | |
2b83a2a4 RM |
1283 | return ret; |
1284 | } | |
2ad4fab2 UD |
1285 | #ifdef _LIBC |
1286 | weak_alias (__re_set_syntax, re_set_syntax) | |
1287 | #endif | |
2b83a2a4 RM |
1288 | \f |
1289 | /* This table gives an error message for each of the error codes listed | |
1290 | in regex.h. Obviously the order here has to be same as there. | |
1291 | POSIX doesn't require that we do anything for REG_NOERROR, | |
1292 | but why not be nice? */ | |
1293 | ||
c4563d2d | 1294 | static const char re_error_msgid[] = |
91c7b85d | 1295 | { |
c4563d2d UD |
1296 | #define REG_NOERROR_IDX 0 |
1297 | gettext_noop ("Success") /* REG_NOERROR */ | |
1298 | "\0" | |
1299 | #define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success") | |
1300 | gettext_noop ("No match") /* REG_NOMATCH */ | |
1301 | "\0" | |
1302 | #define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match") | |
1303 | gettext_noop ("Invalid regular expression") /* REG_BADPAT */ | |
1304 | "\0" | |
1305 | #define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression") | |
d0738b5d | 1306 | gettext_noop ("Invalid collation character") /* REG_ECOLLATE */ |
c4563d2d UD |
1307 | "\0" |
1308 | #define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character") | |
1309 | gettext_noop ("Invalid character class name") /* REG_ECTYPE */ | |
1310 | "\0" | |
1311 | #define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name") | |
1312 | gettext_noop ("Trailing backslash") /* REG_EESCAPE */ | |
1313 | "\0" | |
1314 | #define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash") | |
1315 | gettext_noop ("Invalid back reference") /* REG_ESUBREG */ | |
1316 | "\0" | |
1317 | #define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference") | |
1318 | gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */ | |
1319 | "\0" | |
1320 | #define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^") | |
1321 | gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */ | |
1322 | "\0" | |
1323 | #define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(") | |
1324 | gettext_noop ("Unmatched \\{") /* REG_EBRACE */ | |
1325 | "\0" | |
1326 | #define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{") | |
1327 | gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */ | |
1328 | "\0" | |
1329 | #define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}") | |
1330 | gettext_noop ("Invalid range end") /* REG_ERANGE */ | |
1331 | "\0" | |
1332 | #define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end") | |
1333 | gettext_noop ("Memory exhausted") /* REG_ESPACE */ | |
1334 | "\0" | |
1335 | #define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted") | |
1336 | gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */ | |
1337 | "\0" | |
1338 | #define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression") | |
1339 | gettext_noop ("Premature end of regular expression") /* REG_EEND */ | |
1340 | "\0" | |
1341 | #define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression") | |
1342 | gettext_noop ("Regular expression too big") /* REG_ESIZE */ | |
1343 | "\0" | |
1344 | #define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big") | |
d0738b5d | 1345 | gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */ |
2b83a2a4 | 1346 | }; |
c4563d2d UD |
1347 | |
1348 | static const size_t re_error_msgid_idx[] = | |
1349 | { | |
1350 | REG_NOERROR_IDX, | |
1351 | REG_NOMATCH_IDX, | |
1352 | REG_BADPAT_IDX, | |
1353 | REG_ECOLLATE_IDX, | |
1354 | REG_ECTYPE_IDX, | |
1355 | REG_EESCAPE_IDX, | |
1356 | REG_ESUBREG_IDX, | |
1357 | REG_EBRACK_IDX, | |
1358 | REG_EPAREN_IDX, | |
1359 | REG_EBRACE_IDX, | |
1360 | REG_BADBR_IDX, | |
1361 | REG_ERANGE_IDX, | |
1362 | REG_ESPACE_IDX, | |
1363 | REG_BADRPT_IDX, | |
1364 | REG_EEND_IDX, | |
1365 | REG_ESIZE_IDX, | |
1366 | REG_ERPAREN_IDX | |
1367 | }; | |
2b83a2a4 RM |
1368 | \f |
1369 | /* Avoiding alloca during matching, to placate r_alloc. */ | |
1370 | ||
1371 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the | |
1372 | searching and matching functions should not call alloca. On some | |
1373 | systems, alloca is implemented in terms of malloc, and if we're | |
1374 | using the relocating allocator routines, then malloc could cause a | |
1375 | relocation, which might (if the strings being searched are in the | |
1376 | ralloc heap) shift the data out from underneath the regexp | |
1377 | routines. | |
1378 | ||
91c7b85d | 1379 | Here's another reason to avoid allocation: Emacs |
2b83a2a4 RM |
1380 | processes input from X in a signal handler; processing X input may |
1381 | call malloc; if input arrives while a matching routine is calling | |
1382 | malloc, then we're scrod. But Emacs can't just block input while | |
1383 | calling matching routines; then we don't notice interrupts when | |
1384 | they come in. So, Emacs blocks input around all regexp calls | |
1385 | except the matching calls, which it leaves unprotected, in the | |
1386 | faith that they will not malloc. */ | |
1387 | ||
1388 | /* Normally, this is fine. */ | |
1389 | #define MATCH_MAY_ALLOCATE | |
1390 | ||
1391 | /* When using GNU C, we are not REALLY using the C alloca, no matter | |
1392 | what config.h may say. So don't take precautions for it. */ | |
1393 | #ifdef __GNUC__ | |
86187531 | 1394 | # undef C_ALLOCA |
2b83a2a4 RM |
1395 | #endif |
1396 | ||
1397 | /* The match routines may not allocate if (1) they would do it with malloc | |
1398 | and (2) it's not safe for them to use malloc. | |
1399 | Note that if REL_ALLOC is defined, matching would not use malloc for the | |
1400 | failure stack, but we would still use it for the register vectors; | |
1401 | so REL_ALLOC should not affect this. */ | |
86187531 UD |
1402 | #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs |
1403 | # undef MATCH_MAY_ALLOCATE | |
2b83a2a4 RM |
1404 | #endif |
1405 | ||
1406 | \f | |
1407 | /* Failure stack declarations and macros; both re_compile_fastmap and | |
1408 | re_match_2 use a failure stack. These have to be macros because of | |
1409 | REGEX_ALLOCATE_STACK. */ | |
91c7b85d | 1410 | |
2b83a2a4 RM |
1411 | |
1412 | /* Number of failure points for which to initially allocate space | |
1413 | when matching. If this number is exceeded, we allocate more | |
1414 | space, so it is not a hard limit. */ | |
1415 | #ifndef INIT_FAILURE_ALLOC | |
86187531 | 1416 | # define INIT_FAILURE_ALLOC 5 |
2b83a2a4 RM |
1417 | #endif |
1418 | ||
1419 | /* Roughly the maximum number of failure points on the stack. Would be | |
51702635 | 1420 | exactly that if always used MAX_FAILURE_ITEMS items each time we failed. |
2b83a2a4 RM |
1421 | This is a variable only so users of regex can assign to it; we never |
1422 | change it ourselves. */ | |
4cca6b86 UD |
1423 | |
1424 | #ifdef INT_IS_16BIT | |
1425 | ||
86187531 | 1426 | # if defined MATCH_MAY_ALLOCATE |
51702635 UD |
1427 | /* 4400 was enough to cause a crash on Alpha OSF/1, |
1428 | whose default stack limit is 2mb. */ | |
1429 | long int re_max_failures = 4000; | |
86187531 | 1430 | # else |
51702635 | 1431 | long int re_max_failures = 2000; |
86187531 | 1432 | # endif |
4cca6b86 UD |
1433 | |
1434 | union fail_stack_elt | |
1435 | { | |
e4c785c8 | 1436 | US_CHAR_TYPE *pointer; |
51702635 | 1437 | long int integer; |
4cca6b86 UD |
1438 | }; |
1439 | ||
1440 | typedef union fail_stack_elt fail_stack_elt_t; | |
1441 | ||
1442 | typedef struct | |
1443 | { | |
1444 | fail_stack_elt_t *stack; | |
51702635 UD |
1445 | unsigned long int size; |
1446 | unsigned long int avail; /* Offset of next open position. */ | |
4cca6b86 UD |
1447 | } fail_stack_type; |
1448 | ||
1449 | #else /* not INT_IS_16BIT */ | |
1450 | ||
86187531 | 1451 | # if defined MATCH_MAY_ALLOCATE |
5f0e6fc7 | 1452 | /* 4400 was enough to cause a crash on Alpha OSF/1, |
710f7bab | 1453 | whose default stack limit is 2mb. */ |
ca8d5a5f | 1454 | int re_max_failures = 4000; |
86187531 | 1455 | # else |
2b83a2a4 | 1456 | int re_max_failures = 2000; |
86187531 | 1457 | # endif |
2b83a2a4 RM |
1458 | |
1459 | union fail_stack_elt | |
1460 | { | |
e4c785c8 | 1461 | US_CHAR_TYPE *pointer; |
2b83a2a4 RM |
1462 | int integer; |
1463 | }; | |
1464 | ||
1465 | typedef union fail_stack_elt fail_stack_elt_t; | |
1466 | ||
1467 | typedef struct | |
1468 | { | |
1469 | fail_stack_elt_t *stack; | |
1470 | unsigned size; | |
1471 | unsigned avail; /* Offset of next open position. */ | |
1472 | } fail_stack_type; | |
1473 | ||
4cca6b86 UD |
1474 | #endif /* INT_IS_16BIT */ |
1475 | ||
2b83a2a4 RM |
1476 | #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) |
1477 | #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
1478 | #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
1479 | ||
1480 | ||
1481 | /* Define macros to initialize and free the failure stack. | |
1482 | Do `return -2' if the alloc fails. */ | |
1483 | ||
1484 | #ifdef MATCH_MAY_ALLOCATE | |
86187531 | 1485 | # define INIT_FAIL_STACK() \ |
2b83a2a4 RM |
1486 | do { \ |
1487 | fail_stack.stack = (fail_stack_elt_t *) \ | |
86187531 | 1488 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \ |
2b83a2a4 RM |
1489 | \ |
1490 | if (fail_stack.stack == NULL) \ | |
1491 | return -2; \ | |
1492 | \ | |
1493 | fail_stack.size = INIT_FAILURE_ALLOC; \ | |
1494 | fail_stack.avail = 0; \ | |
1495 | } while (0) | |
1496 | ||
86187531 | 1497 | # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) |
2b83a2a4 | 1498 | #else |
86187531 | 1499 | # define INIT_FAIL_STACK() \ |
2b83a2a4 RM |
1500 | do { \ |
1501 | fail_stack.avail = 0; \ | |
1502 | } while (0) | |
1503 | ||
86187531 | 1504 | # define RESET_FAIL_STACK() |
2b83a2a4 RM |
1505 | #endif |
1506 | ||
1507 | ||
1508 | /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. | |
1509 | ||
1510 | Return 1 if succeeds, and 0 if either ran out of memory | |
91c7b85d RM |
1511 | allocating space for it or it was already too large. |
1512 | ||
2b83a2a4 RM |
1513 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
1514 | ||
1515 | #define DOUBLE_FAIL_STACK(fail_stack) \ | |
cccda09f | 1516 | ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \ |
2b83a2a4 RM |
1517 | ? 0 \ |
1518 | : ((fail_stack).stack = (fail_stack_elt_t *) \ | |
1519 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ | |
1520 | (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
1521 | ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ | |
1522 | \ | |
1523 | (fail_stack).stack == NULL \ | |
1524 | ? 0 \ | |
1525 | : ((fail_stack).size <<= 1, \ | |
1526 | 1))) | |
1527 | ||
1528 | ||
91c7b85d | 1529 | /* Push pointer POINTER on FAIL_STACK. |
2b83a2a4 RM |
1530 | Return 1 if was able to do so and 0 if ran out of memory allocating |
1531 | space to do so. */ | |
1532 | #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ | |
1533 | ((FAIL_STACK_FULL () \ | |
1534 | && !DOUBLE_FAIL_STACK (FAIL_STACK)) \ | |
1535 | ? 0 \ | |
1536 | : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ | |
1537 | 1)) | |
1538 | ||
1539 | /* Push a pointer value onto the failure stack. | |
1540 | Assumes the variable `fail_stack'. Probably should only | |
1541 | be called from within `PUSH_FAILURE_POINT'. */ | |
1542 | #define PUSH_FAILURE_POINTER(item) \ | |
e4c785c8 | 1543 | fail_stack.stack[fail_stack.avail++].pointer = (US_CHAR_TYPE *) (item) |
2b83a2a4 RM |
1544 | |
1545 | /* This pushes an integer-valued item onto the failure stack. | |
1546 | Assumes the variable `fail_stack'. Probably should only | |
1547 | be called from within `PUSH_FAILURE_POINT'. */ | |
1548 | #define PUSH_FAILURE_INT(item) \ | |
1549 | fail_stack.stack[fail_stack.avail++].integer = (item) | |
1550 | ||
1551 | /* Push a fail_stack_elt_t value onto the failure stack. | |
1552 | Assumes the variable `fail_stack'. Probably should only | |
1553 | be called from within `PUSH_FAILURE_POINT'. */ | |
1554 | #define PUSH_FAILURE_ELT(item) \ | |
1555 | fail_stack.stack[fail_stack.avail++] = (item) | |
1556 | ||
1557 | /* These three POP... operations complement the three PUSH... operations. | |
1558 | All assume that `fail_stack' is nonempty. */ | |
1559 | #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
1560 | #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
1561 | #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
1562 | ||
1563 | /* Used to omit pushing failure point id's when we're not debugging. */ | |
1564 | #ifdef DEBUG | |
86187531 UD |
1565 | # define DEBUG_PUSH PUSH_FAILURE_INT |
1566 | # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () | |
2b83a2a4 | 1567 | #else |
86187531 UD |
1568 | # define DEBUG_PUSH(item) |
1569 | # define DEBUG_POP(item_addr) | |
2b83a2a4 RM |
1570 | #endif |
1571 | ||
1572 | ||
1573 | /* Push the information about the state we will need | |
91c7b85d RM |
1574 | if we ever fail back to it. |
1575 | ||
2b83a2a4 | 1576 | Requires variables fail_stack, regstart, regend, reg_info, and |
789b13c4 UD |
1577 | num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination' |
1578 | be declared. | |
91c7b85d | 1579 | |
2b83a2a4 RM |
1580 | Does `return FAILURE_CODE' if runs out of memory. */ |
1581 | ||
1582 | #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ | |
1583 | do { \ | |
1584 | char *destination; \ | |
1585 | /* Must be int, so when we don't save any registers, the arithmetic \ | |
1586 | of 0 + -1 isn't done as unsigned. */ \ | |
4cca6b86 UD |
1587 | /* Can't be int, since there is not a shred of a guarantee that int \ |
1588 | is wide enough to hold a value of something to which pointer can \ | |
1589 | be assigned */ \ | |
bca973bc | 1590 | active_reg_t this_reg; \ |
2b83a2a4 RM |
1591 | \ |
1592 | DEBUG_STATEMENT (failure_id++); \ | |
1593 | DEBUG_STATEMENT (nfailure_points_pushed++); \ | |
1594 | DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ | |
1595 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ | |
1596 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ | |
1597 | \ | |
bca973bc | 1598 | DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \ |
2b83a2a4 RM |
1599 | DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ |
1600 | \ | |
1601 | /* Ensure we have enough space allocated for what we will push. */ \ | |
1602 | while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
1603 | { \ | |
1604 | if (!DOUBLE_FAIL_STACK (fail_stack)) \ | |
1605 | return failure_code; \ | |
1606 | \ | |
1607 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ | |
1608 | (fail_stack).size); \ | |
1609 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ | |
1610 | } \ | |
1611 | \ | |
1612 | /* Push the info, starting with the registers. */ \ | |
1613 | DEBUG_PRINT1 ("\n"); \ | |
1614 | \ | |
3bbceb12 | 1615 | if (1) \ |
537257ae MB |
1616 | for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ |
1617 | this_reg++) \ | |
1618 | { \ | |
bca973bc | 1619 | DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \ |
537257ae | 1620 | DEBUG_STATEMENT (num_regs_pushed++); \ |
2b83a2a4 | 1621 | \ |
bca973bc | 1622 | DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ |
537257ae | 1623 | PUSH_FAILURE_POINTER (regstart[this_reg]); \ |
2b83a2a4 | 1624 | \ |
bca973bc | 1625 | DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ |
537257ae MB |
1626 | PUSH_FAILURE_POINTER (regend[this_reg]); \ |
1627 | \ | |
bca973bc UD |
1628 | DEBUG_PRINT2 (" info: %p\n ", \ |
1629 | reg_info[this_reg].word.pointer); \ | |
537257ae MB |
1630 | DEBUG_PRINT2 (" match_null=%d", \ |
1631 | REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ | |
1632 | DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ | |
1633 | DEBUG_PRINT2 (" matched_something=%d", \ | |
1634 | MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1635 | DEBUG_PRINT2 (" ever_matched=%d", \ | |
1636 | EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ | |
1637 | DEBUG_PRINT1 ("\n"); \ | |
1638 | PUSH_FAILURE_ELT (reg_info[this_reg].word); \ | |
1639 | } \ | |
2b83a2a4 | 1640 | \ |
bca973bc | 1641 | DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\ |
2b83a2a4 RM |
1642 | PUSH_FAILURE_INT (lowest_active_reg); \ |
1643 | \ | |
bca973bc | 1644 | DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\ |
2b83a2a4 RM |
1645 | PUSH_FAILURE_INT (highest_active_reg); \ |
1646 | \ | |
bca973bc | 1647 | DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \ |
2b83a2a4 RM |
1648 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ |
1649 | PUSH_FAILURE_POINTER (pattern_place); \ | |
1650 | \ | |
bca973bc | 1651 | DEBUG_PRINT2 (" Pushing string %p: `", string_place); \ |
2b83a2a4 RM |
1652 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ |
1653 | size2); \ | |
1654 | DEBUG_PRINT1 ("'\n"); \ | |
1655 | PUSH_FAILURE_POINTER (string_place); \ | |
1656 | \ | |
1657 | DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ | |
1658 | DEBUG_PUSH (failure_id); \ | |
1659 | } while (0) | |
1660 | ||
1661 | /* This is the number of items that are pushed and popped on the stack | |
1662 | for each register. */ | |
1663 | #define NUM_REG_ITEMS 3 | |
1664 | ||
1665 | /* Individual items aside from the registers. */ | |
1666 | #ifdef DEBUG | |
86187531 | 1667 | # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ |
2b83a2a4 | 1668 | #else |
86187531 | 1669 | # define NUM_NONREG_ITEMS 4 |
2b83a2a4 RM |
1670 | #endif |
1671 | ||
1672 | /* We push at most this many items on the stack. */ | |
a641835a RM |
1673 | /* We used to use (num_regs - 1), which is the number of registers |
1674 | this regexp will save; but that was changed to 5 | |
1675 | to avoid stack overflow for a regexp with lots of parens. */ | |
1676 | #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS) | |
2b83a2a4 RM |
1677 | |
1678 | /* We actually push this many items. */ | |
537257ae | 1679 | #define NUM_FAILURE_ITEMS \ |
3bbceb12 | 1680 | (((0 \ |
537257ae MB |
1681 | ? 0 : highest_active_reg - lowest_active_reg + 1) \ |
1682 | * NUM_REG_ITEMS) \ | |
1683 | + NUM_NONREG_ITEMS) | |
2b83a2a4 RM |
1684 | |
1685 | /* How many items can still be added to the stack without overflowing it. */ | |
1686 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
1687 | ||
1688 | ||
1689 | /* Pops what PUSH_FAIL_STACK pushes. | |
1690 | ||
1691 | We restore into the parameters, all of which should be lvalues: | |
1692 | STR -- the saved data position. | |
1693 | PAT -- the saved pattern position. | |
1694 | LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
1695 | REGSTART, REGEND -- arrays of string positions. | |
1696 | REG_INFO -- array of information about each subexpression. | |
91c7b85d | 1697 | |
2b83a2a4 RM |
1698 | Also assumes the variables `fail_stack' and (if debugging), `bufp', |
1699 | `pend', `string1', `size1', `string2', and `size2'. */ | |
2b83a2a4 RM |
1700 | #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ |
1701 | { \ | |
bca973bc UD |
1702 | DEBUG_STATEMENT (unsigned failure_id;) \ |
1703 | active_reg_t this_reg; \ | |
e4c785c8 | 1704 | const US_CHAR_TYPE *string_temp; \ |
2b83a2a4 RM |
1705 | \ |
1706 | assert (!FAIL_STACK_EMPTY ()); \ | |
1707 | \ | |
1708 | /* Remove failure points and point to how many regs pushed. */ \ | |
1709 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ | |
1710 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ | |
1711 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ | |
1712 | \ | |
1713 | assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
1714 | \ | |
1715 | DEBUG_POP (&failure_id); \ | |
1716 | DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ | |
1717 | \ | |
1718 | /* If the saved string location is NULL, it came from an \ | |
1719 | on_failure_keep_string_jump opcode, and we want to throw away the \ | |
1720 | saved NULL, thus retaining our current position in the string. */ \ | |
1721 | string_temp = POP_FAILURE_POINTER (); \ | |
1722 | if (string_temp != NULL) \ | |
e4c785c8 | 1723 | str = (const CHAR_TYPE *) string_temp; \ |
2b83a2a4 | 1724 | \ |
bca973bc | 1725 | DEBUG_PRINT2 (" Popping string %p: `", str); \ |
2b83a2a4 RM |
1726 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
1727 | DEBUG_PRINT1 ("'\n"); \ | |
1728 | \ | |
e4c785c8 | 1729 | pat = (US_CHAR_TYPE *) POP_FAILURE_POINTER (); \ |
bca973bc | 1730 | DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \ |
2b83a2a4 RM |
1731 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ |
1732 | \ | |
1733 | /* Restore register info. */ \ | |
4cca6b86 | 1734 | high_reg = (active_reg_t) POP_FAILURE_INT (); \ |
bca973bc | 1735 | DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \ |
2b83a2a4 | 1736 | \ |
4cca6b86 | 1737 | low_reg = (active_reg_t) POP_FAILURE_INT (); \ |
bca973bc | 1738 | DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \ |
2b83a2a4 | 1739 | \ |
3bbceb12 | 1740 | if (1) \ |
537257ae MB |
1741 | for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ |
1742 | { \ | |
bca973bc | 1743 | DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \ |
2b83a2a4 | 1744 | \ |
537257ae | 1745 | reg_info[this_reg].word = POP_FAILURE_ELT (); \ |
bca973bc UD |
1746 | DEBUG_PRINT2 (" info: %p\n", \ |
1747 | reg_info[this_reg].word.pointer); \ | |
2b83a2a4 | 1748 | \ |
e4c785c8 | 1749 | regend[this_reg] = (const CHAR_TYPE *) POP_FAILURE_POINTER (); \ |
bca973bc | 1750 | DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ |
2b83a2a4 | 1751 | \ |
e4c785c8 | 1752 | regstart[this_reg] = (const CHAR_TYPE *) POP_FAILURE_POINTER ();\ |
bca973bc | 1753 | DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ |
537257ae | 1754 | } \ |
57aefafe RM |
1755 | else \ |
1756 | { \ | |
1757 | for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \ | |
1758 | { \ | |
3bbceb12 | 1759 | reg_info[this_reg].word.integer = 0; \ |
57aefafe RM |
1760 | regend[this_reg] = 0; \ |
1761 | regstart[this_reg] = 0; \ | |
1762 | } \ | |
1763 | highest_active_reg = high_reg; \ | |
1764 | } \ | |
2b83a2a4 RM |
1765 | \ |
1766 | set_regs_matched_done = 0; \ | |
1767 | DEBUG_STATEMENT (nfailure_points_popped++); \ | |
1768 | } /* POP_FAILURE_POINT */ | |
1769 | ||
2b83a2a4 RM |
1770 | \f |
1771 | /* Structure for per-register (a.k.a. per-group) information. | |
1772 | Other register information, such as the | |
1773 | starting and ending positions (which are addresses), and the list of | |
1774 | inner groups (which is a bits list) are maintained in separate | |
91c7b85d RM |
1775 | variables. |
1776 | ||
2b83a2a4 RM |
1777 | We are making a (strictly speaking) nonportable assumption here: that |
1778 | the compiler will pack our bit fields into something that fits into | |
1779 | the type of `word', i.e., is something that fits into one item on the | |
1780 | failure stack. */ | |
1781 | ||
4cca6b86 UD |
1782 | |
1783 | /* Declarations and macros for re_match_2. */ | |
1784 | ||
2b83a2a4 RM |
1785 | typedef union |
1786 | { | |
1787 | fail_stack_elt_t word; | |
1788 | struct | |
1789 | { | |
1790 | /* This field is one if this group can match the empty string, | |
1791 | zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ | |
1792 | #define MATCH_NULL_UNSET_VALUE 3 | |
1793 | unsigned match_null_string_p : 2; | |
1794 | unsigned is_active : 1; | |
1795 | unsigned matched_something : 1; | |
1796 | unsigned ever_matched_something : 1; | |
1797 | } bits; | |
1798 | } register_info_type; | |
1799 | ||
1800 | #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
1801 | #define IS_ACTIVE(R) ((R).bits.is_active) | |
1802 | #define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
1803 | #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
1804 | ||
1805 | ||
1806 | /* Call this when have matched a real character; it sets `matched' flags | |
1807 | for the subexpressions which we are currently inside. Also records | |
1808 | that those subexprs have matched. */ | |
1809 | #define SET_REGS_MATCHED() \ | |
1810 | do \ | |
1811 | { \ | |
1812 | if (!set_regs_matched_done) \ | |
1813 | { \ | |
4cca6b86 | 1814 | active_reg_t r; \ |
2b83a2a4 RM |
1815 | set_regs_matched_done = 1; \ |
1816 | for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
1817 | { \ | |
1818 | MATCHED_SOMETHING (reg_info[r]) \ | |
1819 | = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
1820 | = 1; \ | |
1821 | } \ | |
1822 | } \ | |
1823 | } \ | |
1824 | while (0) | |
1825 | ||
1826 | /* Registers are set to a sentinel when they haven't yet matched. */ | |
e4c785c8 | 1827 | static CHAR_TYPE reg_unset_dummy; |
2b83a2a4 RM |
1828 | #define REG_UNSET_VALUE (®_unset_dummy) |
1829 | #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
1830 | \f | |
1831 | /* Subroutine declarations and macros for regex_compile. */ | |
1832 | ||
4cca6b86 UD |
1833 | static reg_errcode_t regex_compile _RE_ARGS ((const char *pattern, size_t size, |
1834 | reg_syntax_t syntax, | |
1835 | struct re_pattern_buffer *bufp)); | |
e4c785c8 UD |
1836 | static void store_op1 _RE_ARGS ((re_opcode_t op, US_CHAR_TYPE *loc, int arg)); |
1837 | static void store_op2 _RE_ARGS ((re_opcode_t op, US_CHAR_TYPE *loc, | |
4cca6b86 | 1838 | int arg1, int arg2)); |
e4c785c8 UD |
1839 | static void insert_op1 _RE_ARGS ((re_opcode_t op, US_CHAR_TYPE *loc, |
1840 | int arg, US_CHAR_TYPE *end)); | |
1841 | static void insert_op2 _RE_ARGS ((re_opcode_t op, US_CHAR_TYPE *loc, | |
1842 | int arg1, int arg2, US_CHAR_TYPE *end)); | |
1843 | static boolean at_begline_loc_p _RE_ARGS ((const CHAR_TYPE *pattern, | |
1844 | const CHAR_TYPE *p, | |
4cca6b86 | 1845 | reg_syntax_t syntax)); |
e4c785c8 UD |
1846 | static boolean at_endline_loc_p _RE_ARGS ((const CHAR_TYPE *p, |
1847 | const CHAR_TYPE *pend, | |
4cca6b86 | 1848 | reg_syntax_t syntax)); |
e4c785c8 UD |
1849 | #ifdef MBS_SUPPORT |
1850 | static reg_errcode_t compile_range _RE_ARGS ((CHAR_TYPE range_start, | |
1851 | const CHAR_TYPE **p_ptr, | |
1852 | const CHAR_TYPE *pend, | |
1853 | char *translate, | |
1854 | reg_syntax_t syntax, | |
1855 | US_CHAR_TYPE *b, | |
1856 | CHAR_TYPE *char_set)); | |
1857 | static void insert_space _RE_ARGS ((int num, CHAR_TYPE *loc, CHAR_TYPE *end)); | |
1858 | #else | |
ac8295d2 | 1859 | static reg_errcode_t compile_range _RE_ARGS ((unsigned int range_start, |
e4c785c8 UD |
1860 | const CHAR_TYPE **p_ptr, |
1861 | const CHAR_TYPE *pend, | |
4cca6b86 UD |
1862 | char *translate, |
1863 | reg_syntax_t syntax, | |
e4c785c8 UD |
1864 | US_CHAR_TYPE *b)); |
1865 | #endif /* MBS_SUPPORT */ | |
2b83a2a4 | 1866 | |
91c7b85d | 1867 | /* Fetch the next character in the uncompiled pattern---translating it |
2b83a2a4 RM |
1868 | if necessary. Also cast from a signed character in the constant |
1869 | string passed to us by the user to an unsigned char that we can use | |
1870 | as an array index (in, e.g., `translate'). */ | |
e4c785c8 UD |
1871 | /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, |
1872 | because it is impossible to allocate 4GB array for some encodings | |
1873 | which have 4 byte character_set like UCS4. */ | |
03a75825 | 1874 | #ifndef PATFETCH |
e4c785c8 UD |
1875 | # ifdef MBS_SUPPORT |
1876 | # define PATFETCH(c) \ | |
1877 | do {if (p == pend) return REG_EEND; \ | |
1878 | c = (US_CHAR_TYPE) *p++; \ | |
1879 | if (translate && (c <= 0xff)) c = (US_CHAR_TYPE) translate[c]; \ | |
1880 | } while (0) | |
1881 | # else | |
1882 | # define PATFETCH(c) \ | |
2b83a2a4 RM |
1883 | do {if (p == pend) return REG_EEND; \ |
1884 | c = (unsigned char) *p++; \ | |
03a75825 | 1885 | if (translate) c = (unsigned char) translate[c]; \ |
2b83a2a4 | 1886 | } while (0) |
e4c785c8 | 1887 | # endif /* MBS_SUPPORT */ |
03a75825 | 1888 | #endif |
2b83a2a4 RM |
1889 | |
1890 | /* Fetch the next character in the uncompiled pattern, with no | |
1891 | translation. */ | |
1892 | #define PATFETCH_RAW(c) \ | |
1893 | do {if (p == pend) return REG_EEND; \ | |
e4c785c8 | 1894 | c = (US_CHAR_TYPE) *p++; \ |
2b83a2a4 RM |
1895 | } while (0) |
1896 | ||
1897 | /* Go backwards one character in the pattern. */ | |
1898 | #define PATUNFETCH p-- | |
1899 | ||
1900 | ||
1901 | /* If `translate' is non-null, return translate[D], else just D. We | |
1902 | cast the subscript to translate because some data is declared as | |
1903 | `char *', to avoid warnings when a string constant is passed. But | |
1904 | when we use a character as a subscript we must make it unsigned. */ | |
e4c785c8 UD |
1905 | /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, |
1906 | because it is impossible to allocate 4GB array for some encodings | |
1907 | which have 4 byte character_set like UCS4. */ | |
03a75825 | 1908 | #ifndef TRANSLATE |
e4c785c8 UD |
1909 | # ifdef MBS_SUPPORT |
1910 | # define TRANSLATE(d) \ | |
2d0aea11 UD |
1911 | ((translate && ((US_CHAR_TYPE) (d)) <= 0xff) \ |
1912 | ? (char) translate[(unsigned char) (d)] : (d)) | |
e4c785c8 UD |
1913 | #else |
1914 | # define TRANSLATE(d) \ | |
03a75825 | 1915 | (translate ? (char) translate[(unsigned char) (d)] : (d)) |
e4c785c8 | 1916 | # endif /* MBS_SUPPORT */ |
03a75825 | 1917 | #endif |
2b83a2a4 RM |
1918 | |
1919 | ||
1920 | /* Macros for outputting the compiled pattern into `buffer'. */ | |
1921 | ||
1922 | /* If the buffer isn't allocated when it comes in, use this. */ | |
e4c785c8 | 1923 | #define INIT_BUF_SIZE (32 * sizeof(US_CHAR_TYPE)) |
2b83a2a4 RM |
1924 | |
1925 | /* Make sure we have at least N more bytes of space in buffer. */ | |
e4c785c8 UD |
1926 | #ifdef MBS_SUPPORT |
1927 | # define GET_BUFFER_SPACE(n) \ | |
672fd41b | 1928 | while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \ |
e4c785c8 UD |
1929 | + (n)*sizeof(CHAR_TYPE)) > bufp->allocated) \ |
1930 | EXTEND_BUFFER () | |
1931 | #else | |
1932 | # define GET_BUFFER_SPACE(n) \ | |
cccda09f | 1933 | while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \ |
2b83a2a4 | 1934 | EXTEND_BUFFER () |
e4c785c8 | 1935 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
1936 | |
1937 | /* Make sure we have one more byte of buffer space and then add C to it. */ | |
1938 | #define BUF_PUSH(c) \ | |
1939 | do { \ | |
1940 | GET_BUFFER_SPACE (1); \ | |
672fd41b | 1941 | *b++ = (US_CHAR_TYPE) (c); \ |
2b83a2a4 RM |
1942 | } while (0) |
1943 | ||
1944 | ||
1945 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
1946 | #define BUF_PUSH_2(c1, c2) \ | |
1947 | do { \ | |
1948 | GET_BUFFER_SPACE (2); \ | |
e4c785c8 UD |
1949 | *b++ = (US_CHAR_TYPE) (c1); \ |
1950 | *b++ = (US_CHAR_TYPE) (c2); \ | |
2b83a2a4 RM |
1951 | } while (0) |
1952 | ||
1953 | ||
1954 | /* As with BUF_PUSH_2, except for three bytes. */ | |
1955 | #define BUF_PUSH_3(c1, c2, c3) \ | |
1956 | do { \ | |
1957 | GET_BUFFER_SPACE (3); \ | |
e4c785c8 UD |
1958 | *b++ = (US_CHAR_TYPE) (c1); \ |
1959 | *b++ = (US_CHAR_TYPE) (c2); \ | |
1960 | *b++ = (US_CHAR_TYPE) (c3); \ | |
2b83a2a4 RM |
1961 | } while (0) |
1962 | ||
2b83a2a4 RM |
1963 | /* Store a jump with opcode OP at LOC to location TO. We store a |
1964 | relative address offset by the three bytes the jump itself occupies. */ | |
1965 | #define STORE_JUMP(op, loc, to) \ | |
e4c785c8 | 1966 | store_op1 (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE))) |
2b83a2a4 RM |
1967 | |
1968 | /* Likewise, for a two-argument jump. */ | |
1969 | #define STORE_JUMP2(op, loc, to, arg) \ | |
e4c785c8 | 1970 | store_op2 (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg) |
2b83a2a4 RM |
1971 | |
1972 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ | |
1973 | #define INSERT_JUMP(op, loc, to) \ | |
e4c785c8 | 1974 | insert_op1 (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b) |
2b83a2a4 RM |
1975 | |
1976 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ | |
1977 | #define INSERT_JUMP2(op, loc, to, arg) \ | |
e4c785c8 UD |
1978 | insert_op2 (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\ |
1979 | arg, b) | |
2b83a2a4 RM |
1980 | |
1981 | ||
1982 | /* This is not an arbitrary limit: the arguments which represent offsets | |
1983 | into the pattern are two bytes long. So if 2^16 bytes turns out to | |
1984 | be too small, many things would have to change. */ | |
4cca6b86 UD |
1985 | /* Any other compiler which, like MSC, has allocation limit below 2^16 |
1986 | bytes will have to use approach similar to what was done below for | |
1987 | MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up | |
1988 | reallocating to 0 bytes. Such thing is not going to work too well. | |
1989 | You have been warned!! */ | |
86187531 | 1990 | #if defined _MSC_VER && !defined WIN32 |
4cca6b86 UD |
1991 | /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. |
1992 | The REALLOC define eliminates a flurry of conversion warnings, | |
1993 | but is not required. */ | |
86187531 UD |
1994 | # define MAX_BUF_SIZE 65500L |
1995 | # define REALLOC(p,s) realloc ((p), (size_t) (s)) | |
4cca6b86 | 1996 | #else |
86187531 UD |
1997 | # define MAX_BUF_SIZE (1L << 16) |
1998 | # define REALLOC(p,s) realloc ((p), (s)) | |
4cca6b86 | 1999 | #endif |
2b83a2a4 RM |
2000 | |
2001 | /* Extend the buffer by twice its current size via realloc and | |
2002 | reset the pointers that pointed into the old block to point to the | |
2003 | correct places in the new one. If extending the buffer results in it | |
2004 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ | |
8ccd2cb1 | 2005 | #if __BOUNDED_POINTERS__ |
13550428 | 2006 | # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) |
8ccd2cb1 | 2007 | # define MOVE_BUFFER_POINTER(P) \ |
13550428 GM |
2008 | (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) |
2009 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ | |
2010 | else \ | |
2011 | { \ | |
2012 | SET_HIGH_BOUND (b); \ | |
2013 | SET_HIGH_BOUND (begalt); \ | |
2014 | if (fixup_alt_jump) \ | |
2015 | SET_HIGH_BOUND (fixup_alt_jump); \ | |
2016 | if (laststart) \ | |
2017 | SET_HIGH_BOUND (laststart); \ | |
2018 | if (pending_exact) \ | |
2019 | SET_HIGH_BOUND (pending_exact); \ | |
2020 | } | |
8ccd2cb1 GM |
2021 | #else |
2022 | # define MOVE_BUFFER_POINTER(P) (P) += incr | |
13550428 | 2023 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND |
8ccd2cb1 | 2024 | #endif |
e4c785c8 UD |
2025 | |
2026 | #ifdef MBS_SUPPORT | |
2027 | # define EXTEND_BUFFER() \ | |
2028 | do { \ | |
2029 | US_CHAR_TYPE *old_buffer = COMPILED_BUFFER_VAR; \ | |
2030 | int wchar_count; \ | |
2031 | if (bufp->allocated + sizeof(US_CHAR_TYPE) > MAX_BUF_SIZE) \ | |
2032 | return REG_ESIZE; \ | |
2033 | bufp->allocated <<= 1; \ | |
2034 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
2035 | bufp->allocated = MAX_BUF_SIZE; \ | |
2036 | /* How many characters the new buffer can have? */ \ | |
2037 | wchar_count = bufp->allocated / sizeof(US_CHAR_TYPE); \ | |
2038 | if (wchar_count == 0) wchar_count = 1; \ | |
2039 | /* Truncate the buffer to CHAR_TYPE align. */ \ | |
2040 | bufp->allocated = wchar_count * sizeof(US_CHAR_TYPE); \ | |
2041 | RETALLOC (COMPILED_BUFFER_VAR, wchar_count, US_CHAR_TYPE); \ | |
2042 | bufp->buffer = (char*)COMPILED_BUFFER_VAR; \ | |
2043 | if (COMPILED_BUFFER_VAR == NULL) \ | |
2044 | return REG_ESPACE; \ | |
2045 | /* If the buffer moved, move all the pointers into it. */ \ | |
2046 | if (old_buffer != COMPILED_BUFFER_VAR) \ | |
2047 | { \ | |
2048 | int incr = COMPILED_BUFFER_VAR - old_buffer; \ | |
2049 | MOVE_BUFFER_POINTER (b); \ | |
2050 | MOVE_BUFFER_POINTER (begalt); \ | |
2051 | if (fixup_alt_jump) \ | |
2052 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ | |
2053 | if (laststart) \ | |
2054 | MOVE_BUFFER_POINTER (laststart); \ | |
2055 | if (pending_exact) \ | |
2056 | MOVE_BUFFER_POINTER (pending_exact); \ | |
2057 | } \ | |
2058 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ | |
2059 | } while (0) | |
2060 | #else | |
2061 | # define EXTEND_BUFFER() \ | |
8ccd2cb1 | 2062 | do { \ |
e4c785c8 | 2063 | US_CHAR_TYPE *old_buffer = COMPILED_BUFFER_VAR; \ |
8ccd2cb1 | 2064 | if (bufp->allocated == MAX_BUF_SIZE) \ |
2b83a2a4 RM |
2065 | return REG_ESIZE; \ |
2066 | bufp->allocated <<= 1; \ | |
2067 | if (bufp->allocated > MAX_BUF_SIZE) \ | |
8ccd2cb1 | 2068 | bufp->allocated = MAX_BUF_SIZE; \ |
e4c785c8 UD |
2069 | bufp->buffer = (US_CHAR_TYPE *) REALLOC (COMPILED_BUFFER_VAR, \ |
2070 | bufp->allocated); \ | |
2071 | if (COMPILED_BUFFER_VAR == NULL) \ | |
2b83a2a4 RM |
2072 | return REG_ESPACE; \ |
2073 | /* If the buffer moved, move all the pointers into it. */ \ | |
e4c785c8 | 2074 | if (old_buffer != COMPILED_BUFFER_VAR) \ |
2b83a2a4 | 2075 | { \ |
e4c785c8 | 2076 | int incr = COMPILED_BUFFER_VAR - old_buffer; \ |
8ccd2cb1 GM |
2077 | MOVE_BUFFER_POINTER (b); \ |
2078 | MOVE_BUFFER_POINTER (begalt); \ | |
2079 | if (fixup_alt_jump) \ | |
2080 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ | |
2081 | if (laststart) \ | |
2082 | MOVE_BUFFER_POINTER (laststart); \ | |
2083 | if (pending_exact) \ | |
2084 | MOVE_BUFFER_POINTER (pending_exact); \ | |
2b83a2a4 | 2085 | } \ |
13550428 | 2086 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
2b83a2a4 | 2087 | } while (0) |
e4c785c8 | 2088 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
2089 | |
2090 | /* Since we have one byte reserved for the register number argument to | |
2091 | {start,stop}_memory, the maximum number of groups we can report | |
2092 | things about is what fits in that byte. */ | |
2093 | #define MAX_REGNUM 255 | |
2094 | ||
2095 | /* But patterns can have more than `MAX_REGNUM' registers. We just | |
2096 | ignore the excess. */ | |
2097 | typedef unsigned regnum_t; | |
2098 | ||
2099 | ||
2100 | /* Macros for the compile stack. */ | |
2101 | ||
2102 | /* Since offsets can go either forwards or backwards, this type needs to | |
2103 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ | |
4cca6b86 UD |
2104 | /* int may be not enough when sizeof(int) == 2. */ |
2105 | typedef long pattern_offset_t; | |
2b83a2a4 RM |
2106 | |
2107 | typedef struct | |
2108 | { | |
2109 | pattern_offset_t begalt_offset; | |
2110 | pattern_offset_t fixup_alt_jump; | |
2111 | pattern_offset_t inner_group_offset; | |
91c7b85d | 2112 | pattern_offset_t laststart_offset; |
2b83a2a4 RM |
2113 | regnum_t regnum; |
2114 | } compile_stack_elt_t; | |
2115 | ||
2116 | ||
2117 | typedef struct | |
2118 | { | |
2119 | compile_stack_elt_t *stack; | |
2120 | unsigned size; | |
2121 | unsigned avail; /* Offset of next open position. */ | |
2122 | } compile_stack_type; | |
2123 | ||
2124 | ||
2125 | #define INIT_COMPILE_STACK_SIZE 32 | |
2126 | ||
2127 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
2128 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
2129 | ||
2130 | /* The next available element. */ | |
2131 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) | |
2132 | ||
2133 | ||
2134 | /* Set the bit for character C in a list. */ | |
2135 | #define SET_LIST_BIT(c) \ | |
2136 | (b[((unsigned char) (c)) / BYTEWIDTH] \ | |
2137 | |= 1 << (((unsigned char) c) % BYTEWIDTH)) | |
2138 | ||
2139 | ||
2140 | /* Get the next unsigned number in the uncompiled pattern. */ | |
0a45b76c UD |
2141 | #define GET_UNSIGNED_NUMBER(num) \ |
2142 | { \ | |
2143 | while (p != pend) \ | |
2144 | { \ | |
2145 | PATFETCH (c); \ | |
2146 | if (c < '0' || c > '9') \ | |
2147 | break; \ | |
2148 | if (num <= RE_DUP_MAX) \ | |
2149 | { \ | |
2150 | if (num < 0) \ | |
2151 | num = 0; \ | |
2152 | num = num * 10 + c - '0'; \ | |
2153 | } \ | |
2154 | } \ | |
2155 | } | |
2b83a2a4 | 2156 | |
409dfcea | 2157 | #if defined _LIBC || WIDE_CHAR_SUPPORT |
51702635 UD |
2158 | /* The GNU C library provides support for user-defined character classes |
2159 | and the functions from ISO C amendement 1. */ | |
2160 | # ifdef CHARCLASS_NAME_MAX | |
2161 | # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX | |
2162 | # else | |
2163 | /* This shouldn't happen but some implementation might still have this | |
2164 | problem. Use a reasonable default value. */ | |
2165 | # define CHAR_CLASS_MAX_LENGTH 256 | |
2166 | # endif | |
2167 | ||
2ad4fab2 UD |
2168 | # ifdef _LIBC |
2169 | # define IS_CHAR_CLASS(string) __wctype (string) | |
2170 | # else | |
2171 | # define IS_CHAR_CLASS(string) wctype (string) | |
2172 | # endif | |
51702635 UD |
2173 | #else |
2174 | # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
2b83a2a4 | 2175 | |
51702635 | 2176 | # define IS_CHAR_CLASS(string) \ |
2b83a2a4 RM |
2177 | (STREQ (string, "alpha") || STREQ (string, "upper") \ |
2178 | || STREQ (string, "lower") || STREQ (string, "digit") \ | |
2179 | || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
2180 | || STREQ (string, "space") || STREQ (string, "print") \ | |
2181 | || STREQ (string, "punct") || STREQ (string, "graph") \ | |
2182 | || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
51702635 | 2183 | #endif |
2b83a2a4 RM |
2184 | \f |
2185 | #ifndef MATCH_MAY_ALLOCATE | |
2186 | ||
2187 | /* If we cannot allocate large objects within re_match_2_internal, | |
2188 | we make the fail stack and register vectors global. | |
2189 | The fail stack, we grow to the maximum size when a regexp | |
2190 | is compiled. | |
2191 | The register vectors, we adjust in size each time we | |
2192 | compile a regexp, according to the number of registers it needs. */ | |
2193 | ||
2194 | static fail_stack_type fail_stack; | |
2195 | ||
2196 | /* Size with which the following vectors are currently allocated. | |
2197 | That is so we can make them bigger as needed, | |
2198 | but never make them smaller. */ | |
2199 | static int regs_allocated_size; | |
2200 | ||
2201 | static const char ** regstart, ** regend; | |
2202 | static const char ** old_regstart, ** old_regend; | |
2203 | static const char **best_regstart, **best_regend; | |
91c7b85d | 2204 | static register_info_type *reg_info; |
2b83a2a4 RM |
2205 | static const char **reg_dummy; |
2206 | static register_info_type *reg_info_dummy; | |
2207 | ||
2208 | /* Make the register vectors big enough for NUM_REGS registers, | |
2209 | but don't make them smaller. */ | |
2210 | ||
2211 | static | |
2212 | regex_grow_registers (num_regs) | |
2213 | int num_regs; | |
2214 | { | |
2215 | if (num_regs > regs_allocated_size) | |
2216 | { | |
2217 | RETALLOC_IF (regstart, num_regs, const char *); | |
2218 | RETALLOC_IF (regend, num_regs, const char *); | |
2219 | RETALLOC_IF (old_regstart, num_regs, const char *); | |
2220 | RETALLOC_IF (old_regend, num_regs, const char *); | |
2221 | RETALLOC_IF (best_regstart, num_regs, const char *); | |
2222 | RETALLOC_IF (best_regend, num_regs, const char *); | |
2223 | RETALLOC_IF (reg_info, num_regs, register_info_type); | |
2224 | RETALLOC_IF (reg_dummy, num_regs, const char *); | |
2225 | RETALLOC_IF (reg_info_dummy, num_regs, register_info_type); | |
2226 | ||
2227 | regs_allocated_size = num_regs; | |
2228 | } | |
2229 | } | |
2230 | ||
2231 | #endif /* not MATCH_MAY_ALLOCATE */ | |
2232 | \f | |
4cca6b86 UD |
2233 | static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type |
2234 | compile_stack, | |
2235 | regnum_t regnum)); | |
2236 | ||
2b83a2a4 RM |
2237 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
2238 | Returns one of error codes defined in `regex.h', or zero for success. | |
2239 | ||
2240 | Assumes the `allocated' (and perhaps `buffer') and `translate' | |
2241 | fields are set in BUFP on entry. | |
2242 | ||
2243 | If it succeeds, results are put in BUFP (if it returns an error, the | |
2244 | contents of BUFP are undefined): | |
2245 | `buffer' is the compiled pattern; | |
2246 | `syntax' is set to SYNTAX; | |
2247 | `used' is set to the length of the compiled pattern; | |
2248 | `fastmap_accurate' is zero; | |
2249 | `re_nsub' is the number of subexpressions in PATTERN; | |
2250 | `not_bol' and `not_eol' are zero; | |
91c7b85d | 2251 | |
2b83a2a4 RM |
2252 | The `fastmap' and `newline_anchor' fields are neither |
2253 | examined nor set. */ | |
2254 | ||
2255 | /* Return, freeing storage we allocated. */ | |
e4c785c8 UD |
2256 | #ifdef MBS_SUPPORT |
2257 | # define FREE_STACK_RETURN(value) \ | |
2258 | return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value) | |
2259 | #else | |
2260 | # define FREE_STACK_RETURN(value) \ | |
2b83a2a4 | 2261 | return (free (compile_stack.stack), value) |
e4c785c8 | 2262 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
2263 | |
2264 | static reg_errcode_t | |
e4c785c8 UD |
2265 | #ifdef MBS_SUPPORT |
2266 | regex_compile (cpattern, csize, syntax, bufp) | |
2267 | const char *cpattern; | |
2268 | size_t csize; | |
2269 | #else | |
2b83a2a4 RM |
2270 | regex_compile (pattern, size, syntax, bufp) |
2271 | const char *pattern; | |
4cca6b86 | 2272 | size_t size; |
e4c785c8 | 2273 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
2274 | reg_syntax_t syntax; |
2275 | struct re_pattern_buffer *bufp; | |
2276 | { | |
2277 | /* We fetch characters from PATTERN here. Even though PATTERN is | |
2278 | `char *' (i.e., signed), we declare these variables as unsigned, so | |
2279 | they can be reliably used as array indices. */ | |
e4c785c8 UD |
2280 | register US_CHAR_TYPE c, c1; |
2281 | ||
2282 | #ifdef MBS_SUPPORT | |
2283 | /* A temporary space to keep wchar_t pattern and compiled pattern. */ | |
2284 | CHAR_TYPE *pattern, *COMPILED_BUFFER_VAR; | |
2285 | size_t size; | |
2286 | /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ | |
2287 | int *mbs_offset = NULL; | |
2288 | /* It hold whether each wchar_t is binary data or not. */ | |
770d454d | 2289 | char *is_binary = NULL; |
e4c785c8 | 2290 | /* A flag whether exactn is handling binary data or not. */ |
770d454d | 2291 | char is_exactn_bin = FALSE; |
e4c785c8 | 2292 | #endif /* MBS_SUPPORT */ |
91c7b85d | 2293 | |
2b83a2a4 | 2294 | /* A random temporary spot in PATTERN. */ |
e4c785c8 | 2295 | const CHAR_TYPE *p1; |
2b83a2a4 RM |
2296 | |
2297 | /* Points to the end of the buffer, where we should append. */ | |
e4c785c8 | 2298 | register US_CHAR_TYPE *b; |
91c7b85d | 2299 | |
2b83a2a4 RM |
2300 | /* Keeps track of unclosed groups. */ |
2301 | compile_stack_type compile_stack; | |
2302 | ||
2303 | /* Points to the current (ending) position in the pattern. */ | |
e4c785c8 UD |
2304 | #ifdef MBS_SUPPORT |
2305 | const CHAR_TYPE *p; | |
2306 | const CHAR_TYPE *pend; | |
2307 | #else | |
2308 | const CHAR_TYPE *p = pattern; | |
2309 | const CHAR_TYPE *pend = pattern + size; | |
2310 | #endif /* MBS_SUPPORT */ | |
91c7b85d | 2311 | |
2b83a2a4 | 2312 | /* How to translate the characters in the pattern. */ |
03a75825 | 2313 | RE_TRANSLATE_TYPE translate = bufp->translate; |
2b83a2a4 RM |
2314 | |
2315 | /* Address of the count-byte of the most recently inserted `exactn' | |
2316 | command. This makes it possible to tell if a new exact-match | |
2317 | character can be added to that command or if the character requires | |
2318 | a new `exactn' command. */ | |
e4c785c8 | 2319 | US_CHAR_TYPE *pending_exact = 0; |
2b83a2a4 RM |
2320 | |
2321 | /* Address of start of the most recently finished expression. | |
2322 | This tells, e.g., postfix * where to find the start of its | |
2323 | operand. Reset at the beginning of groups and alternatives. */ | |
e4c785c8 | 2324 | US_CHAR_TYPE *laststart = 0; |
2b83a2a4 RM |
2325 | |
2326 | /* Address of beginning of regexp, or inside of last group. */ | |
e4c785c8 | 2327 | US_CHAR_TYPE *begalt; |
2b83a2a4 | 2328 | |
2b83a2a4 RM |
2329 | /* Address of the place where a forward jump should go to the end of |
2330 | the containing expression. Each alternative of an `or' -- except the | |
2331 | last -- ends with a forward jump of this sort. */ | |
e4c785c8 | 2332 | US_CHAR_TYPE *fixup_alt_jump = 0; |
2b83a2a4 RM |
2333 | |
2334 | /* Counts open-groups as they are encountered. Remembered for the | |
2335 | matching close-group on the compile stack, so the same register | |
2336 | number is put in the stop_memory as the start_memory. */ | |
2337 | regnum_t regnum = 0; | |
2338 | ||
e4c785c8 UD |
2339 | #ifdef MBS_SUPPORT |
2340 | /* Initialize the wchar_t PATTERN and offset_buffer. */ | |
28d2fb9a | 2341 | p = pend = pattern = TALLOC(csize + 1, CHAR_TYPE); |
e4c785c8 | 2342 | mbs_offset = TALLOC(csize + 1, int); |
770d454d | 2343 | is_binary = TALLOC(csize + 1, char); |
e4c785c8 UD |
2344 | if (pattern == NULL || mbs_offset == NULL || is_binary == NULL) |
2345 | { | |
8bca0bd4 UD |
2346 | free(pattern); |
2347 | free(mbs_offset); | |
2348 | free(is_binary); | |
e4c785c8 UD |
2349 | return REG_ESPACE; |
2350 | } | |
220e7575 | 2351 | pattern[csize] = L'\0'; /* sentinel */ |
e4c785c8 UD |
2352 | size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary); |
2353 | pend = p + size; | |
2354 | if (size < 0) | |
2355 | { | |
8bca0bd4 UD |
2356 | free(pattern); |
2357 | free(mbs_offset); | |
2358 | free(is_binary); | |
e4c785c8 UD |
2359 | return REG_BADPAT; |
2360 | } | |
2361 | #endif | |
2362 | ||
2b83a2a4 RM |
2363 | #ifdef DEBUG |
2364 | DEBUG_PRINT1 ("\nCompiling pattern: "); | |
2365 | if (debug) | |
2366 | { | |
2367 | unsigned debug_count; | |
91c7b85d | 2368 | |
2b83a2a4 | 2369 | for (debug_count = 0; debug_count < size; debug_count++) |
e4c785c8 | 2370 | PUT_CHAR (pattern[debug_count]); |
2b83a2a4 RM |
2371 | putchar ('\n'); |
2372 | } | |
2373 | #endif /* DEBUG */ | |
2374 | ||
2375 | /* Initialize the compile stack. */ | |
2376 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
2377 | if (compile_stack.stack == NULL) | |
e4c785c8 UD |
2378 | { |
2379 | #ifdef MBS_SUPPORT | |
8bca0bd4 UD |
2380 | free(pattern); |
2381 | free(mbs_offset); | |
2382 | free(is_binary); | |
e4c785c8 UD |
2383 | #endif |
2384 | return REG_ESPACE; | |
2385 | } | |
2b83a2a4 RM |
2386 | |
2387 | compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
2388 | compile_stack.avail = 0; | |
2389 | ||
2390 | /* Initialize the pattern buffer. */ | |
2391 | bufp->syntax = syntax; | |
2392 | bufp->fastmap_accurate = 0; | |
2393 | bufp->not_bol = bufp->not_eol = 0; | |
2394 | ||
2395 | /* Set `used' to zero, so that if we return an error, the pattern | |
2396 | printer (for debugging) will think there's no pattern. We reset it | |
2397 | at the end. */ | |
2398 | bufp->used = 0; | |
91c7b85d | 2399 | |
2b83a2a4 | 2400 | /* Always count groups, whether or not bufp->no_sub is set. */ |
91c7b85d | 2401 | bufp->re_nsub = 0; |
2b83a2a4 | 2402 | |
86187531 | 2403 | #if !defined emacs && !defined SYNTAX_TABLE |
2b83a2a4 RM |
2404 | /* Initialize the syntax table. */ |
2405 | init_syntax_once (); | |
2406 | #endif | |
2407 | ||
2408 | if (bufp->allocated == 0) | |
2409 | { | |
2410 | if (bufp->buffer) | |
2411 | { /* If zero allocated, but buffer is non-null, try to realloc | |
2412 | enough space. This loses if buffer's address is bogus, but | |
2413 | that is the user's responsibility. */ | |
e4c785c8 UD |
2414 | #ifdef MBS_SUPPORT |
2415 | /* Free bufp->buffer and allocate an array for wchar_t pattern | |
2416 | buffer. */ | |
2417 | free(bufp->buffer); | |
2418 | COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(US_CHAR_TYPE), | |
2419 | US_CHAR_TYPE); | |
2420 | #else | |
2421 | RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, US_CHAR_TYPE); | |
2422 | #endif /* MBS_SUPPORT */ | |
2b83a2a4 RM |
2423 | } |
2424 | else | |
2425 | { /* Caller did not allocate a buffer. Do it for them. */ | |
e4c785c8 UD |
2426 | COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(US_CHAR_TYPE), |
2427 | US_CHAR_TYPE); | |
2b83a2a4 | 2428 | } |
2b83a2a4 | 2429 | |
e4c785c8 UD |
2430 | if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE); |
2431 | #ifdef MBS_SUPPORT | |
2432 | bufp->buffer = (char*)COMPILED_BUFFER_VAR; | |
2433 | #endif /* MBS_SUPPORT */ | |
2b83a2a4 RM |
2434 | bufp->allocated = INIT_BUF_SIZE; |
2435 | } | |
e4c785c8 UD |
2436 | #ifdef MBS_SUPPORT |
2437 | else | |
2438 | COMPILED_BUFFER_VAR = (US_CHAR_TYPE*) bufp->buffer; | |
2439 | #endif | |
2b83a2a4 | 2440 | |
e4c785c8 | 2441 | begalt = b = COMPILED_BUFFER_VAR; |
2b83a2a4 RM |
2442 | |
2443 | /* Loop through the uncompiled pattern until we're at the end. */ | |
2444 | while (p != pend) | |
2445 | { | |
2446 | PATFETCH (c); | |
2447 | ||
2448 | switch (c) | |
2449 | { | |
2450 | case '^': | |
2451 | { | |
2452 | if ( /* If at start of pattern, it's an operator. */ | |
2453 | p == pattern + 1 | |
2454 | /* If context independent, it's an operator. */ | |
2455 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2456 | /* Otherwise, depends on what's come before. */ | |
2457 | || at_begline_loc_p (pattern, p, syntax)) | |
2458 | BUF_PUSH (begline); | |
2459 | else | |
2460 | goto normal_char; | |
2461 | } | |
2462 | break; | |
2463 | ||
2464 | ||
2465 | case '$': | |
2466 | { | |
2467 | if ( /* If at end of pattern, it's an operator. */ | |
91c7b85d | 2468 | p == pend |
2b83a2a4 RM |
2469 | /* If context independent, it's an operator. */ |
2470 | || syntax & RE_CONTEXT_INDEP_ANCHORS | |
2471 | /* Otherwise, depends on what's next. */ | |
2472 | || at_endline_loc_p (p, pend, syntax)) | |
2473 | BUF_PUSH (endline); | |
2474 | else | |
2475 | goto normal_char; | |
2476 | } | |
2477 | break; | |
2478 | ||
2479 | ||
2480 | case '+': | |
2481 | case '?': | |
2482 | if ((syntax & RE_BK_PLUS_QM) | |
2483 | || (syntax & RE_LIMITED_OPS)) | |
2484 | goto normal_char; | |
2485 | handle_plus: | |
2486 | case '*': | |
2487 | /* If there is no previous pattern... */ | |
2488 | if (!laststart) | |
2489 | { | |
2490 | if (syntax & RE_CONTEXT_INVALID_OPS) | |
2491 | FREE_STACK_RETURN (REG_BADRPT); | |
2492 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
2493 | goto normal_char; | |
2494 | } | |
2495 | ||
2496 | { | |
2497 | /* Are we optimizing this jump? */ | |
2498 | boolean keep_string_p = false; | |
91c7b85d | 2499 | |
2b83a2a4 RM |
2500 | /* 1 means zero (many) matches is allowed. */ |
2501 | char zero_times_ok = 0, many_times_ok = 0; | |
2502 | ||
2503 | /* If there is a sequence of repetition chars, collapse it | |
2504 | down to just one (the right one). We can't combine | |
2505 | interval operators with these because of, e.g., `a{2}*', | |
2506 | which should only match an even number of `a's. */ | |
2507 | ||
2508 | for (;;) | |
2509 | { | |
2510 | zero_times_ok |= c != '+'; | |
2511 | many_times_ok |= c != '?'; | |
2512 | ||
2513 | if (p == pend) | |
2514 | break; | |
2515 | ||
2516 | PATFETCH (c); | |
2517 | ||
2518 | if (c == '*' | |
2519 | || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) | |
2520 | ; | |
2521 | ||
2522 | else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
2523 | { | |
2524 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
2525 | ||
2526 | PATFETCH (c1); | |
2527 | if (!(c1 == '+' || c1 == '?')) | |
2528 | { | |
2529 | PATUNFETCH; | |
2530 | PATUNFETCH; | |
2531 | break; | |
2532 | } | |
2533 | ||
2534 | c = c1; | |
2535 | } | |
2536 | else | |
2537 | { | |
2538 | PATUNFETCH; | |
2539 | break; | |
2540 | } | |
2541 | ||
2542 | /* If we get here, we found another repeat character. */ | |
2543 | } | |
2544 | ||
2545 | /* Star, etc. applied to an empty pattern is equivalent | |
2546 | to an empty pattern. */ | |
91c7b85d | 2547 | if (!laststart) |
2b83a2a4 RM |
2548 | break; |
2549 | ||
2550 | /* Now we know whether or not zero matches is allowed | |
2551 | and also whether or not two or more matches is allowed. */ | |
2552 | if (many_times_ok) | |
2553 | { /* More than one repetition is allowed, so put in at the | |
2554 | end a backward relative jump from `b' to before the next | |
2555 | jump we're going to put in below (which jumps from | |
91c7b85d | 2556 | laststart to after this jump). |
2b83a2a4 RM |
2557 | |
2558 | But if we are at the `*' in the exact sequence `.*\n', | |
2559 | insert an unconditional jump backwards to the ., | |
2560 | instead of the beginning of the loop. This way we only | |
2561 | push a failure point once, instead of every time | |
2562 | through the loop. */ | |
2563 | assert (p - 1 > pattern); | |
2564 | ||
2565 | /* Allocate the space for the jump. */ | |
e4c785c8 | 2566 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
2b83a2a4 RM |
2567 | |
2568 | /* We know we are not at the first character of the pattern, | |
2569 | because laststart was nonzero. And we've already | |
2570 | incremented `p', by the way, to be the character after | |
2571 | the `*'. Do we have to do something analogous here | |
2572 | for null bytes, because of RE_DOT_NOT_NULL? */ | |
2573 | if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') | |
2574 | && zero_times_ok | |
2575 | && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') | |
2576 | && !(syntax & RE_DOT_NEWLINE)) | |
2577 | { /* We have .*\n. */ | |
2578 | STORE_JUMP (jump, b, laststart); | |
2579 | keep_string_p = true; | |
2580 | } | |
2581 | else | |
2582 | /* Anything else. */ | |
e4c785c8 UD |
2583 | STORE_JUMP (maybe_pop_jump, b, laststart - |
2584 | (1 + OFFSET_ADDRESS_SIZE)); | |
2b83a2a4 RM |
2585 | |
2586 | /* We've added more stuff to the buffer. */ | |
e4c785c8 | 2587 | b += 1 + OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
2588 | } |
2589 | ||
2590 | /* On failure, jump from laststart to b + 3, which will be the | |
2591 | end of the buffer after this jump is inserted. */ | |
e4c785c8 UD |
2592 | /* ifdef MBS_SUPPORT, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of |
2593 | 'b + 3'. */ | |
2594 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); | |
2b83a2a4 RM |
2595 | INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump |
2596 | : on_failure_jump, | |
e4c785c8 | 2597 | laststart, b + 1 + OFFSET_ADDRESS_SIZE); |
2b83a2a4 | 2598 | pending_exact = 0; |
e4c785c8 | 2599 | b += 1 + OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
2600 | |
2601 | if (!zero_times_ok) | |
2602 | { | |
2603 | /* At least one repetition is required, so insert a | |
2604 | `dummy_failure_jump' before the initial | |
2605 | `on_failure_jump' instruction of the loop. This | |
2606 | effects a skip over that instruction the first time | |
2607 | we hit that loop. */ | |
e4c785c8 UD |
2608 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
2609 | INSERT_JUMP (dummy_failure_jump, laststart, laststart + | |
2610 | 2 + 2 * OFFSET_ADDRESS_SIZE); | |
2611 | b += 1 + OFFSET_ADDRESS_SIZE; | |
2b83a2a4 RM |
2612 | } |
2613 | } | |
2614 | break; | |
2615 | ||
2616 | ||
2617 | case '.': | |
2618 | laststart = b; | |
2619 | BUF_PUSH (anychar); | |
2620 | break; | |
2621 | ||
2622 | ||
2623 | case '[': | |
2624 | { | |
2625 | boolean had_char_class = false; | |
e4c785c8 UD |
2626 | #ifdef MBS_SUPPORT |
2627 | CHAR_TYPE range_start = 0xffffffff; | |
2628 | #else | |
ac8295d2 | 2629 | unsigned int range_start = 0xffffffff; |
e4c785c8 | 2630 | #endif |
2b83a2a4 RM |
2631 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
2632 | ||
e4c785c8 UD |
2633 | #ifdef MBS_SUPPORT |
2634 | /* We assume a charset(_not) structure as a wchar_t array. | |
2635 | charset[0] = (re_opcode_t) charset(_not) | |
2636 | charset[1] = l (= length of char_classes) | |
2637 | charset[2] = m (= length of collating_symbols) | |
2638 | charset[3] = n (= length of equivalence_classes) | |
2639 | charset[4] = o (= length of char_ranges) | |
2640 | charset[5] = p (= length of chars) | |
2641 | ||
2642 | charset[6] = char_class (wctype_t) | |
054d2bf7 | 2643 | charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t) |
e4c785c8 UD |
2644 | ... |
2645 | charset[l+5] = char_class (wctype_t) | |
2646 | ||
2647 | charset[l+6] = collating_symbol (wchar_t) | |
2648 | ... | |
2649 | charset[l+m+5] = collating_symbol (wchar_t) | |
2650 | ifdef _LIBC we use the index if | |
2651 | _NL_COLLATE_SYMB_EXTRAMB instead of | |
2652 | wchar_t string. | |
2653 | ||
2654 | charset[l+m+6] = equivalence_classes (wchar_t) | |
2655 | ... | |
2656 | charset[l+m+n+5] = equivalence_classes (wchar_t) | |
2657 | ifdef _LIBC we use the index in | |
2658 | _NL_COLLATE_WEIGHT instead of | |
2659 | wchar_t string. | |
2660 | ||
2661 | charset[l+m+n+6] = range_start | |
2662 | charset[l+m+n+7] = range_end | |
2663 | ... | |
2664 | charset[l+m+n+2o+4] = range_start | |
2665 | charset[l+m+n+2o+5] = range_end | |
2666 | ifdef _LIBC we use the value looked up | |
2667 | in _NL_COLLATE_COLLSEQ instead of | |
2668 | wchar_t character. | |
2669 | ||
2670 | charset[l+m+n+2o+6] = char | |
2671 | ... | |
2672 | charset[l+m+n+2o+p+5] = char | |
2673 | ||
2674 | */ | |
2675 | ||
2676 | /* We need at least 6 spaces: the opcode, the length of | |
2677 | char_classes, the length of collating_symbols, the length of | |
2678 | equivalence_classes, the length of char_ranges, the length of | |
2679 | chars. */ | |
2680 | GET_BUFFER_SPACE (6); | |
2681 | ||
2682 | /* Save b as laststart. And We use laststart as the pointer | |
2683 | to the first element of the charset here. | |
2684 | In other words, laststart[i] indicates charset[i]. */ | |
2685 | laststart = b; | |
2686 | ||
2687 | /* We test `*p == '^' twice, instead of using an if | |
2688 | statement, so we only need one BUF_PUSH. */ | |
2689 | BUF_PUSH (*p == '^' ? charset_not : charset); | |
2690 | if (*p == '^') | |
2691 | p++; | |
2692 | ||
2693 | /* Push the length of char_classes, the length of | |
2694 | collating_symbols, the length of equivalence_classes, the | |
2695 | length of char_ranges and the length of chars. */ | |
2696 | BUF_PUSH_3 (0, 0, 0); | |
2697 | BUF_PUSH_2 (0, 0); | |
2698 | ||
2699 | /* Remember the first position in the bracket expression. */ | |
2700 | p1 = p; | |
2701 | ||
2702 | /* charset_not matches newline according to a syntax bit. */ | |
2703 | if ((re_opcode_t) b[-6] == charset_not | |
2704 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
2705 | { | |
2706 | BUF_PUSH('\n'); | |
2707 | laststart[5]++; /* Update the length of characters */ | |
2708 | } | |
2709 | ||
2710 | /* Read in characters and ranges, setting map bits. */ | |
2711 | for (;;) | |
2712 | { | |
2713 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2714 | ||
2715 | PATFETCH (c); | |
2716 | ||
2717 | /* \ might escape characters inside [...] and [^...]. */ | |
2718 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
2719 | { | |
2720 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
2721 | ||
2722 | PATFETCH (c1); | |
2723 | BUF_PUSH(c1); | |
2724 | laststart[5]++; /* Update the length of chars */ | |
2725 | range_start = c1; | |
2726 | continue; | |
2727 | } | |
2728 | ||
2729 | /* Could be the end of the bracket expression. If it's | |
2730 | not (i.e., when the bracket expression is `[]' so | |
2731 | far), the ']' character bit gets set way below. */ | |
2732 | if (c == ']' && p != p1 + 1) | |
2733 | break; | |
2734 | ||
2735 | /* Look ahead to see if it's a range when the last thing | |
2736 | was a character class. */ | |
2737 | if (had_char_class && c == '-' && *p != ']') | |
2738 | FREE_STACK_RETURN (REG_ERANGE); | |
2739 | ||
2740 | /* Look ahead to see if it's a range when the last thing | |
2741 | was a character: if this is a hyphen not at the | |
2742 | beginning or the end of a list, then it's the range | |
2743 | operator. */ | |
2744 | if (c == '-' | |
2745 | && !(p - 2 >= pattern && p[-2] == '[') | |
2746 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') | |
2747 | && *p != ']') | |
2748 | { | |
2749 | reg_errcode_t ret; | |
2750 | /* Allocate the space for range_start and range_end. */ | |
2751 | GET_BUFFER_SPACE (2); | |
2752 | /* Update the pointer to indicate end of buffer. */ | |
2753 | b += 2; | |
2754 | ret = compile_range (range_start, &p, pend, translate, | |
2755 | syntax, b, laststart); | |
2756 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); | |
2757 | range_start = 0xffffffff; | |
2758 | } | |
2759 | else if (p[0] == '-' && p[1] != ']') | |
2760 | { /* This handles ranges made up of characters only. */ | |
2761 | reg_errcode_t ret; | |
2762 | ||
2763 | /* Move past the `-'. */ | |
2764 | PATFETCH (c1); | |
2765 | /* Allocate the space for range_start and range_end. */ | |
2766 | GET_BUFFER_SPACE (2); | |
2767 | /* Update the pointer to indicate end of buffer. */ | |
2768 | b += 2; | |
2769 | ret = compile_range (c, &p, pend, translate, syntax, b, | |
2770 | laststart); | |
2771 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); | |
2772 | range_start = 0xffffffff; | |
2773 | } | |
2774 | ||
2775 | /* See if we're at the beginning of a possible character | |
2776 | class. */ | |
2777 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
2778 | { /* Leave room for the null. */ | |
2779 | char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
2780 | ||
2781 | PATFETCH (c); | |
2782 | c1 = 0; | |
2783 | ||
2784 | /* If pattern is `[[:'. */ | |
2785 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2786 | ||
2787 | for (;;) | |
2788 | { | |
2789 | PATFETCH (c); | |
2790 | if ((c == ':' && *p == ']') || p == pend) | |
2791 | break; | |
2792 | if (c1 < CHAR_CLASS_MAX_LENGTH) | |
2793 | str[c1++] = c; | |
2794 | else | |
2795 | /* This is in any case an invalid class name. */ | |
2796 | str[0] = '\0'; | |
2797 | } | |
2798 | str[c1] = '\0'; | |
2799 | ||
2800 | /* If isn't a word bracketed by `[:' and `:]': | |
2801 | undo the ending character, the letters, and leave | |
2802 | the leading `:' and `[' (but store them as character). */ | |
2803 | if (c == ':' && *p == ']') | |
2804 | { | |
2805 | wctype_t wt; | |
441f7d1e UD |
2806 | uintptr_t alignedp; |
2807 | ||
e4c785c8 UD |
2808 | /* Query the character class as wctype_t. */ |
2809 | wt = IS_CHAR_CLASS (str); | |
2810 | if (wt == 0) | |
2811 | FREE_STACK_RETURN (REG_ECTYPE); | |
2812 | ||
2813 | /* Throw away the ] at the end of the character | |
2814 | class. */ | |
2815 | PATFETCH (c); | |
2816 | ||
2817 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2818 | ||
2819 | /* Allocate the space for character class. */ | |
054d2bf7 | 2820 | GET_BUFFER_SPACE(CHAR_CLASS_SIZE); |
e4c785c8 | 2821 | /* Update the pointer to indicate end of buffer. */ |
054d2bf7 | 2822 | b += CHAR_CLASS_SIZE; |
e4c785c8 UD |
2823 | /* Move data which follow character classes |
2824 | not to violate the data. */ | |
441f7d1e UD |
2825 | insert_space(CHAR_CLASS_SIZE, |
2826 | laststart + 6 + laststart[1], | |
2827 | b - 1); | |
2828 | alignedp = ((uintptr_t)(laststart + 6 + laststart[1]) | |
2829 | + __alignof__(wctype_t) - 1) | |
2830 | & ~(uintptr_t)(__alignof__(wctype_t) - 1); | |
e4c785c8 | 2831 | /* Store the character class. */ |
441f7d1e | 2832 | *((wctype_t*)alignedp) = wt; |
054d2bf7 UD |
2833 | /* Update length of char_classes */ |
2834 | laststart[1] += CHAR_CLASS_SIZE; | |
e4c785c8 UD |
2835 | |
2836 | had_char_class = true; | |
2837 | } | |
2838 | else | |
2839 | { | |
2840 | c1++; | |
2841 | while (c1--) | |
2842 | PATUNFETCH; | |
2843 | BUF_PUSH ('['); | |
2844 | BUF_PUSH (':'); | |
2845 | laststart[5] += 2; /* Update the length of characters */ | |
2846 | range_start = ':'; | |
2847 | had_char_class = false; | |
2848 | } | |
2849 | } | |
2850 | else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '=' | |
2851 | || *p == '.')) | |
2852 | { | |
2853 | CHAR_TYPE str[128]; /* Should be large enough. */ | |
2854 | CHAR_TYPE delim = *p; /* '=' or '.' */ | |
2855 | # ifdef _LIBC | |
2856 | uint32_t nrules = | |
2857 | _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
2858 | # endif | |
2859 | PATFETCH (c); | |
2860 | c1 = 0; | |
2861 | ||
2862 | /* If pattern is `[[=' or '[[.'. */ | |
2863 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
2864 | ||
2865 | for (;;) | |
2866 | { | |
2867 | PATFETCH (c); | |
2868 | if ((c == delim && *p == ']') || p == pend) | |
2869 | break; | |
2870 | if (c1 < sizeof (str) - 1) | |
2871 | str[c1++] = c; | |
2872 | else | |
2873 | /* This is in any case an invalid class name. */ | |
2874 | str[0] = '\0'; | |
2875 | } | |
2876 | str[c1] = '\0'; | |
2877 | ||
2878 | if (c == delim && *p == ']' && str[0] != '\0') | |
2879 | { | |
2880 | unsigned int i, offset; | |
2881 | /* If we have no collation data we use the default | |
2882 | collation in which each character is in a class | |
2883 | by itself. It also means that ASCII is the | |
2884 | character set and therefore we cannot have character | |
2885 | with more than one byte in the multibyte | |
2886 | representation. */ | |
2887 | ||
2888 | /* If not defined _LIBC, we push the name and | |
2889 | `\0' for the sake of matching performance. */ | |
2890 | int datasize = c1 + 1; | |
2891 | ||
2892 | # ifdef _LIBC | |
2893 | int32_t idx = 0; | |
2894 | if (nrules == 0) | |
2895 | # endif | |
2896 | { | |
2897 | if (c1 != 1) | |
2898 | FREE_STACK_RETURN (REG_ECOLLATE); | |
2899 | } | |
2900 | # ifdef _LIBC | |
2901 | else | |
2902 | { | |
2903 | const int32_t *table; | |
2904 | const int32_t *weights; | |
2905 | const int32_t *extra; | |
2906 | const int32_t *indirect; | |
2907 | wint_t *cp; | |
2908 | ||
2909 | /* This #include defines a local function! */ | |
2910 | # include <locale/weightwc.h> | |
2911 | ||
2912 | if(delim == '=') | |
2913 | { | |
2914 | /* We push the index for equivalence class. */ | |
2915 | cp = (wint_t*)str; | |
2916 | ||
2917 | table = (const int32_t *) | |
2918 | _NL_CURRENT (LC_COLLATE, | |
2919 | _NL_COLLATE_TABLEWC); | |
2920 | weights = (const int32_t *) | |
2921 | _NL_CURRENT (LC_COLLATE, | |
2922 | _NL_COLLATE_WEIGHTWC); | |
2923 | extra = (const int32_t *) | |
2924 | _NL_CURRENT (LC_COLLATE, | |
2925 | _NL_COLLATE_EXTRAWC); | |
2926 | indirect = (const int32_t *) | |
2927 | _NL_CURRENT (LC_COLLATE, | |
2928 | _NL_COLLATE_INDIRECTWC); | |
2929 | ||
2930 | idx = findidx ((const wint_t**)&cp); | |
2931 | if (idx == 0 || cp < (wint_t*) str + c1) | |
2932 | /* This is no valid character. */ | |
2933 | FREE_STACK_RETURN (REG_ECOLLATE); | |
2934 | ||
2935 | str[0] = (wchar_t)idx; | |
2936 | } | |
2937 | else /* delim == '.' */ | |
2938 | { | |
2939 | /* We push collation sequence value | |
2940 | for collating symbol. */ | |
2941 | int32_t table_size; | |
2942 | const int32_t *symb_table; | |
2943 | const unsigned char *extra; | |
2944 | int32_t idx; | |
2945 | int32_t elem; | |
2946 | int32_t second; | |
2947 | int32_t hash; | |
2948 | char char_str[c1]; | |
2949 | ||
2950 | /* We have to convert the name to a single-byte | |
2951 | string. This is possible since the names | |
2952 | consist of ASCII characters and the internal | |
2953 | representation is UCS4. */ | |
2954 | for (i = 0; i < c1; ++i) | |
2955 | char_str[i] = str[i]; | |
2956 | ||
2957 | table_size = | |
2958 | _NL_CURRENT_WORD (LC_COLLATE, | |
2959 | _NL_COLLATE_SYMB_HASH_SIZEMB); | |
2960 | symb_table = (const int32_t *) | |
2961 | _NL_CURRENT (LC_COLLATE, | |
2962 | _NL_COLLATE_SYMB_TABLEMB); | |
2963 | extra = (const unsigned char *) | |
2964 | _NL_CURRENT (LC_COLLATE, | |
2965 | _NL_COLLATE_SYMB_EXTRAMB); | |
2966 | ||
2967 | /* Locate the character in the hashing table. */ | |
2968 | hash = elem_hash (char_str, c1); | |
2969 | ||
2970 | idx = 0; | |
2971 | elem = hash % table_size; | |
2972 | second = hash % (table_size - 2); | |
2973 | while (symb_table[2 * elem] != 0) | |
2974 | { | |
2975 | /* First compare the hashing value. */ | |
2976 | if (symb_table[2 * elem] == hash | |
2977 | && c1 == extra[symb_table[2 * elem + 1]] | |
2978 | && memcmp (str, | |
2979 | &extra[symb_table[2 * elem + 1] | |
2980 | + 1], c1) == 0) | |
2981 | { | |
2982 | /* Yep, this is the entry. */ | |
2983 | idx = symb_table[2 * elem + 1]; | |
2984 | idx += 1 + extra[idx]; | |
2985 | break; | |
2986 | } | |
2987 | ||
2988 | /* Next entry. */ | |
2989 | elem += second; | |
2990 | } | |
2991 | ||
2992 | if (symb_table[2 * elem] != 0) | |
2993 | { | |
2994 | /* Compute the index of the byte sequence | |
2995 | in the table. */ | |
2996 | idx += 1 + extra[idx]; | |
2997 | /* Adjust for the alignment. */ | |
2998 | idx = (idx + 3) & ~4; | |
2999 | ||
054d2bf7 | 3000 | str[0] = (wchar_t) idx + 4; |
e4c785c8 UD |
3001 | } |
3002 | else if (symb_table[2 * elem] == 0 && c1 == 1) | |
3003 | { | |
3004 | /* No valid character. Match it as a | |
3005 | single byte character. */ | |
3006 | had_char_class = false; | |
3007 | BUF_PUSH(str[0]); | |
3008 | /* Update the length of characters */ | |
3009 | laststart[5]++; | |
3010 | range_start = str[0]; | |
3011 | ||
3012 | /* Throw away the ] at the end of the | |
3013 | collating symbol. */ | |
3014 | PATFETCH (c); | |
3015 | /* exit from the switch block. */ | |
3016 | continue; | |
3017 | } | |
3018 | else | |
3019 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3020 | } | |
3021 | datasize = 1; | |
3022 | } | |
3023 | # endif | |
3024 | /* Throw away the ] at the end of the equivalence | |
3025 | class (or collating symbol). */ | |
3026 | PATFETCH (c); | |
3027 | ||
3028 | /* Allocate the space for the equivalence class | |
3029 | (or collating symbol) (and '\0' if needed). */ | |
3030 | GET_BUFFER_SPACE(datasize); | |
3031 | /* Update the pointer to indicate end of buffer. */ | |
3032 | b += datasize; | |
3033 | ||
3034 | if (delim == '=') | |
3035 | { /* equivalence class */ | |
3036 | /* Calculate the offset of char_ranges, | |
3037 | which is next to equivalence_classes. */ | |
3038 | offset = laststart[1] + laststart[2] | |
3039 | + laststart[3] +6; | |
3040 | /* Insert space. */ | |
3041 | insert_space(datasize, laststart + offset, b - 1); | |
3042 | ||
3043 | /* Write the equivalence_class and \0. */ | |
3044 | for (i = 0 ; i < datasize ; i++) | |
3045 | laststart[offset + i] = str[i]; | |
3046 | ||
3047 | /* Update the length of equivalence_classes. */ | |
3048 | laststart[3] += datasize; | |
3049 | had_char_class = true; | |
3050 | } | |
3051 | else /* delim == '.' */ | |
3052 | { /* collating symbol */ | |
3053 | /* Calculate the offset of the equivalence_classes, | |
3054 | which is next to collating_symbols. */ | |
3055 | offset = laststart[1] + laststart[2] + 6; | |
3056 | /* Insert space and write the collationg_symbol | |
3057 | and \0. */ | |
3058 | insert_space(datasize, laststart + offset, b-1); | |
3059 | for (i = 0 ; i < datasize ; i++) | |
3060 | laststart[offset + i] = str[i]; | |
3061 | ||
3062 | /* In re_match_2_internal if range_start < -1, we | |
3063 | assume -range_start is the offset of the | |
3064 | collating symbol which is specified as | |
3065 | the character of the range start. So we assign | |
3066 | -(laststart[1] + laststart[2] + 6) to | |
3067 | range_start. */ | |
3068 | range_start = -(laststart[1] + laststart[2] + 6); | |
3069 | /* Update the length of collating_symbol. */ | |
3070 | laststart[2] += datasize; | |
3071 | had_char_class = false; | |
3072 | } | |
3073 | } | |
3074 | else | |
3075 | { | |
3076 | c1++; | |
3077 | while (c1--) | |
3078 | PATUNFETCH; | |
3079 | BUF_PUSH ('['); | |
3080 | BUF_PUSH (delim); | |
3081 | laststart[5] += 2; /* Update the length of characters */ | |
3082 | range_start = delim; | |
3083 | had_char_class = false; | |
3084 | } | |
3085 | } | |
3086 | else | |
3087 | { | |
3088 | had_char_class = false; | |
3089 | BUF_PUSH(c); | |
3090 | laststart[5]++; /* Update the length of characters */ | |
3091 | range_start = c; | |
3092 | } | |
3093 | } | |
3094 | ||
3095 | #else /* not MBS_SUPPORT */ | |
2b83a2a4 RM |
3096 | /* Ensure that we have enough space to push a charset: the |
3097 | opcode, the length count, and the bitset; 34 bytes in all. */ | |
3098 | GET_BUFFER_SPACE (34); | |
3099 | ||
3100 | laststart = b; | |
3101 | ||
3102 | /* We test `*p == '^' twice, instead of using an if | |
3103 | statement, so we only need one BUF_PUSH. */ | |
91c7b85d | 3104 | BUF_PUSH (*p == '^' ? charset_not : charset); |
2b83a2a4 RM |
3105 | if (*p == '^') |
3106 | p++; | |
3107 | ||
3108 | /* Remember the first position in the bracket expression. */ | |
3109 | p1 = p; | |
3110 | ||
3111 | /* Push the number of bytes in the bitmap. */ | |
3112 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
3113 | ||
3114 | /* Clear the whole map. */ | |
3115 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); | |
3116 | ||
3117 | /* charset_not matches newline according to a syntax bit. */ | |
3118 | if ((re_opcode_t) b[-2] == charset_not | |
3119 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) | |
3120 | SET_LIST_BIT ('\n'); | |
3121 | ||
3122 | /* Read in characters and ranges, setting map bits. */ | |
3123 | for (;;) | |
3124 | { | |
3125 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3126 | ||
3127 | PATFETCH (c); | |
3128 | ||
3129 | /* \ might escape characters inside [...] and [^...]. */ | |
3130 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
3131 | { | |
3132 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
3133 | ||
3134 | PATFETCH (c1); | |
3135 | SET_LIST_BIT (c1); | |
ac8295d2 | 3136 | range_start = c1; |
2b83a2a4 RM |
3137 | continue; |
3138 | } | |
3139 | ||
3140 | /* Could be the end of the bracket expression. If it's | |
3141 | not (i.e., when the bracket expression is `[]' so | |
3142 | far), the ']' character bit gets set way below. */ | |
3143 | if (c == ']' && p != p1 + 1) | |
3144 | break; | |
3145 | ||
3146 | /* Look ahead to see if it's a range when the last thing | |
3147 | was a character class. */ | |
3148 | if (had_char_class && c == '-' && *p != ']') | |
3149 | FREE_STACK_RETURN (REG_ERANGE); | |
3150 | ||
3151 | /* Look ahead to see if it's a range when the last thing | |
3152 | was a character: if this is a hyphen not at the | |
3153 | beginning or the end of a list, then it's the range | |
3154 | operator. */ | |
91c7b85d RM |
3155 | if (c == '-' |
3156 | && !(p - 2 >= pattern && p[-2] == '[') | |
2b83a2a4 RM |
3157 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') |
3158 | && *p != ']') | |
3159 | { | |
3160 | reg_errcode_t ret | |
ac8295d2 UD |
3161 | = compile_range (range_start, &p, pend, translate, |
3162 | syntax, b); | |
2b83a2a4 | 3163 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
ac8295d2 | 3164 | range_start = 0xffffffff; |
2b83a2a4 RM |
3165 | } |
3166 | ||
3167 | else if (p[0] == '-' && p[1] != ']') | |
3168 | { /* This handles ranges made up of characters only. */ | |
3169 | reg_errcode_t ret; | |
3170 | ||
3171 | /* Move past the `-'. */ | |
3172 | PATFETCH (c1); | |
91c7b85d | 3173 | |
ac8295d2 | 3174 | ret = compile_range (c, &p, pend, translate, syntax, b); |
2b83a2a4 | 3175 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
ac8295d2 | 3176 | range_start = 0xffffffff; |
2b83a2a4 RM |
3177 | } |
3178 | ||
3179 | /* See if we're at the beginning of a possible character | |
3180 | class. */ | |
3181 | ||
3182 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
3183 | { /* Leave room for the null. */ | |
3184 | char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
3185 | ||
3186 | PATFETCH (c); | |
3187 | c1 = 0; | |
3188 | ||
3189 | /* If pattern is `[[:'. */ | |
3190 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3191 | ||
3192 | for (;;) | |
3193 | { | |
3194 | PATFETCH (c); | |
bece5ca7 | 3195 | if ((c == ':' && *p == ']') || p == pend) |
2b83a2a4 | 3196 | break; |
bece5ca7 UD |
3197 | if (c1 < CHAR_CLASS_MAX_LENGTH) |
3198 | str[c1++] = c; | |
3199 | else | |
3200 | /* This is in any case an invalid class name. */ | |
3201 | str[0] = '\0'; | |
2b83a2a4 RM |
3202 | } |
3203 | str[c1] = '\0'; | |
3204 | ||
68b50604 | 3205 | /* If isn't a word bracketed by `[:' and `:]': |
91c7b85d | 3206 | undo the ending character, the letters, and leave |
2b83a2a4 RM |
3207 | the leading `:' and `[' (but set bits for them). */ |
3208 | if (c == ':' && *p == ']') | |
3209 | { | |
e4c785c8 | 3210 | # if defined _LIBC || WIDE_CHAR_SUPPORT |
51702635 UD |
3211 | boolean is_lower = STREQ (str, "lower"); |
3212 | boolean is_upper = STREQ (str, "upper"); | |
3213 | wctype_t wt; | |
3214 | int ch; | |
3215 | ||
2ad4fab2 | 3216 | wt = IS_CHAR_CLASS (str); |
51702635 UD |
3217 | if (wt == 0) |
3218 | FREE_STACK_RETURN (REG_ECTYPE); | |
3219 | ||
3220 | /* Throw away the ] at the end of the character | |
3221 | class. */ | |
3222 | PATFETCH (c); | |
3223 | ||
3224 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3225 | ||
3226 | for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) | |
3227 | { | |
e4c785c8 | 3228 | # ifdef _LIBC |
2ad4fab2 UD |
3229 | if (__iswctype (__btowc (ch), wt)) |
3230 | SET_LIST_BIT (ch); | |
e4c785c8 | 3231 | # else |
51702635 UD |
3232 | if (iswctype (btowc (ch), wt)) |
3233 | SET_LIST_BIT (ch); | |
e4c785c8 | 3234 | # endif |
51702635 UD |
3235 | |
3236 | if (translate && (is_upper || is_lower) | |
3237 | && (ISUPPER (ch) || ISLOWER (ch))) | |
3238 | SET_LIST_BIT (ch); | |
3239 | } | |
3240 | ||
3241 | had_char_class = true; | |
e4c785c8 | 3242 | # else |
2b83a2a4 RM |
3243 | int ch; |
3244 | boolean is_alnum = STREQ (str, "alnum"); | |
3245 | boolean is_alpha = STREQ (str, "alpha"); | |
3246 | boolean is_blank = STREQ (str, "blank"); | |
3247 | boolean is_cntrl = STREQ (str, "cntrl"); | |
3248 | boolean is_digit = STREQ (str, "digit"); | |
3249 | boolean is_graph = STREQ (str, "graph"); | |
3250 | boolean is_lower = STREQ (str, "lower"); | |
3251 | boolean is_print = STREQ (str, "print"); | |
3252 | boolean is_punct = STREQ (str, "punct"); | |
3253 | boolean is_space = STREQ (str, "space"); | |
3254 | boolean is_upper = STREQ (str, "upper"); | |
3255 | boolean is_xdigit = STREQ (str, "xdigit"); | |
91c7b85d | 3256 | |
2b83a2a4 RM |
3257 | if (!IS_CHAR_CLASS (str)) |
3258 | FREE_STACK_RETURN (REG_ECTYPE); | |
3259 | ||
3260 | /* Throw away the ] at the end of the character | |
3261 | class. */ | |
91c7b85d | 3262 | PATFETCH (c); |
2b83a2a4 RM |
3263 | |
3264 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3265 | ||
3266 | for (ch = 0; ch < 1 << BYTEWIDTH; ch++) | |
3267 | { | |
3268 | /* This was split into 3 if's to | |
3269 | avoid an arbitrary limit in some compiler. */ | |
3270 | if ( (is_alnum && ISALNUM (ch)) | |
3271 | || (is_alpha && ISALPHA (ch)) | |
3272 | || (is_blank && ISBLANK (ch)) | |
3273 | || (is_cntrl && ISCNTRL (ch))) | |
3274 | SET_LIST_BIT (ch); | |
3275 | if ( (is_digit && ISDIGIT (ch)) | |
3276 | || (is_graph && ISGRAPH (ch)) | |
3277 | || (is_lower && ISLOWER (ch)) | |
3278 | || (is_print && ISPRINT (ch))) | |
3279 | SET_LIST_BIT (ch); | |
3280 | if ( (is_punct && ISPUNCT (ch)) | |
3281 | || (is_space && ISSPACE (ch)) | |
3282 | || (is_upper && ISUPPER (ch)) | |
3283 | || (is_xdigit && ISXDIGIT (ch))) | |
3284 | SET_LIST_BIT (ch); | |
4cca6b86 UD |
3285 | if ( translate && (is_upper || is_lower) |
3286 | && (ISUPPER (ch) || ISLOWER (ch))) | |
3287 | SET_LIST_BIT (ch); | |
2b83a2a4 RM |
3288 | } |
3289 | had_char_class = true; | |
e4c785c8 | 3290 | # endif /* libc || wctype.h */ |
2b83a2a4 RM |
3291 | } |
3292 | else | |
3293 | { | |
3294 | c1++; | |
91c7b85d | 3295 | while (c1--) |
2b83a2a4 RM |
3296 | PATUNFETCH; |
3297 | SET_LIST_BIT ('['); | |
3298 | SET_LIST_BIT (':'); | |
ac8295d2 | 3299 | range_start = ':'; |
2b83a2a4 RM |
3300 | had_char_class = false; |
3301 | } | |
3302 | } | |
a63a3c2c UD |
3303 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=') |
3304 | { | |
3305 | unsigned char str[MB_LEN_MAX + 1]; | |
e4c785c8 | 3306 | # ifdef _LIBC |
a63a3c2c UD |
3307 | uint32_t nrules = |
3308 | _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
e4c785c8 | 3309 | # endif |
a63a3c2c UD |
3310 | |
3311 | PATFETCH (c); | |
3312 | c1 = 0; | |
3313 | ||
3314 | /* If pattern is `[[='. */ | |
3315 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
3316 | ||
3317 | for (;;) | |
3318 | { | |
3319 | PATFETCH (c); | |
3320 | if ((c == '=' && *p == ']') || p == pend) | |
3321 | break; | |
3322 | if (c1 < MB_LEN_MAX) | |
3323 | str[c1++] = c; | |
3324 | else | |
3325 | /* This is in any case an invalid class name. */ | |
3326 | str[0] = '\0'; | |
3327 | } | |
3328 | str[c1] = '\0'; | |
3329 | ||
3330 | if (c == '=' && *p == ']' && str[0] != '\0') | |
3331 | { | |
3332 | /* If we have no collation data we use the default | |
3333 | collation in which each character is in a class | |
3334 | by itself. It also means that ASCII is the | |
3335 | character set and therefore we cannot have character | |
3336 | with more than one byte in the multibyte | |
3337 | representation. */ | |
e4c785c8 | 3338 | # ifdef _LIBC |
a63a3c2c | 3339 | if (nrules == 0) |
e4c785c8 | 3340 | # endif |
a63a3c2c UD |
3341 | { |
3342 | if (c1 != 1) | |
3343 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3344 | ||
3345 | /* Throw away the ] at the end of the equivalence | |
3346 | class. */ | |
3347 | PATFETCH (c); | |
3348 | ||
3349 | /* Set the bit for the character. */ | |
3350 | SET_LIST_BIT (str[0]); | |
3351 | } | |
e4c785c8 | 3352 | # ifdef _LIBC |
a63a3c2c UD |
3353 | else |
3354 | { | |
3355 | /* Try to match the byte sequence in `str' against | |
3356 | those known to the collate implementation. | |
3357 | First find out whether the bytes in `str' are | |
3358 | actually from exactly one character. */ | |
3359 | const int32_t *table; | |
3360 | const unsigned char *weights; | |
3361 | const unsigned char *extra; | |
3362 | const int32_t *indirect; | |
3363 | int32_t idx; | |
3364 | const unsigned char *cp = str; | |
a63a3c2c UD |
3365 | int ch; |
3366 | ||
3367 | /* This #include defines a local function! */ | |
e4c785c8 | 3368 | # include <locale/weight.h> |
a63a3c2c UD |
3369 | |
3370 | table = (const int32_t *) | |
3371 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB); | |
3372 | weights = (const unsigned char *) | |
3373 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB); | |
3374 | extra = (const unsigned char *) | |
3375 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB); | |
3376 | indirect = (const int32_t *) | |
3377 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB); | |
3378 | ||
3379 | idx = findidx (&cp); | |
3380 | if (idx == 0 || cp < str + c1) | |
3381 | /* This is no valid character. */ | |
3382 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3383 | ||
3384 | /* Throw away the ] at the end of the equivalence | |
3385 | class. */ | |
3386 | PATFETCH (c); | |
3387 | ||
3388 | /* Now we have to go throught the whole table | |
3389 | and find all characters which have the same | |
3390 | first level weight. | |
3391 | ||
3392 | XXX Note that this is not entirely correct. | |
3393 | we would have to match multibyte sequences | |
3394 | but this is not possible with the current | |
3395 | implementation. */ | |
3396 | for (ch = 1; ch < 256; ++ch) | |
3397 | /* XXX This test would have to be changed if we | |
3398 | would allow matching multibyte sequences. */ | |
3399 | if (table[ch] > 0) | |
3400 | { | |
3401 | int32_t idx2 = table[ch]; | |
3402 | size_t len = weights[idx2]; | |
3403 | ||
3404 | /* Test whether the lenghts match. */ | |
3405 | if (weights[idx] == len) | |
3406 | { | |
3407 | /* They do. New compare the bytes of | |
3408 | the weight. */ | |
3409 | size_t cnt = 0; | |
3410 | ||
3411 | while (cnt < len | |
3412 | && (weights[idx + 1 + cnt] | |
3413 | == weights[idx2 + 1 + cnt])) | |
e4c785c8 | 3414 | ++cnt; |
a63a3c2c UD |
3415 | |
3416 | if (cnt == len) | |
3417 | /* They match. Mark the character as | |
3418 | acceptable. */ | |
3419 | SET_LIST_BIT (ch); | |
3420 | } | |
3421 | } | |
3422 | } | |
e4c785c8 | 3423 | # endif |
a63a3c2c UD |
3424 | had_char_class = true; |
3425 | } | |
ac8295d2 UD |
3426 | else |
3427 | { | |
3428 | c1++; | |
3429 | while (c1--) | |
3430 | PATUNFETCH; | |
3431 | SET_LIST_BIT ('['); | |
3432 | SET_LIST_BIT ('='); | |
3433 | range_start = '='; | |
3434 | had_char_class = false; | |
3435 | } | |
3216711f UD |
3436 | } |
3437 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.') | |
3438 | { | |
3439 | unsigned char str[128]; /* Should be large enough. */ | |
e4c785c8 | 3440 | # ifdef _LIBC |
3216711f UD |
3441 | uint32_t nrules = |
3442 | _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
e4c785c8 | 3443 | # endif |
3216711f UD |
3444 | |
3445 | PATFETCH (c); | |
3446 | c1 = 0; | |
3447 | ||
e4c785c8 | 3448 | /* If pattern is `[[.'. */ |
3216711f UD |
3449 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
3450 | ||
3451 | for (;;) | |
3452 | { | |
3453 | PATFETCH (c); | |
3454 | if ((c == '.' && *p == ']') || p == pend) | |
3455 | break; | |
3456 | if (c1 < sizeof (str)) | |
3457 | str[c1++] = c; | |
3458 | else | |
3459 | /* This is in any case an invalid class name. */ | |
3460 | str[0] = '\0'; | |
3461 | } | |
3462 | str[c1] = '\0'; | |
3463 | ||
3464 | if (c == '.' && *p == ']' && str[0] != '\0') | |
3465 | { | |
3466 | /* If we have no collation data we use the default | |
3467 | collation in which each character is the name | |
3468 | for its own class which contains only the one | |
3469 | character. It also means that ASCII is the | |
3470 | character set and therefore we cannot have character | |
3471 | with more than one byte in the multibyte | |
3472 | representation. */ | |
e4c785c8 | 3473 | # ifdef _LIBC |
3216711f | 3474 | if (nrules == 0) |
e4c785c8 | 3475 | # endif |
3216711f UD |
3476 | { |
3477 | if (c1 != 1) | |
3478 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3479 | ||
3480 | /* Throw away the ] at the end of the equivalence | |
3481 | class. */ | |
3482 | PATFETCH (c); | |
3483 | ||
3484 | /* Set the bit for the character. */ | |
3485 | SET_LIST_BIT (str[0]); | |
ac8295d2 | 3486 | range_start = ((const unsigned char *) str)[0]; |
3216711f | 3487 | } |
e4c785c8 | 3488 | # ifdef _LIBC |
3216711f UD |
3489 | else |
3490 | { | |
3491 | /* Try to match the byte sequence in `str' against | |
3492 | those known to the collate implementation. | |
3493 | First find out whether the bytes in `str' are | |
3494 | actually from exactly one character. */ | |
3216711f | 3495 | int32_t table_size; |
3216711f UD |
3496 | const int32_t *symb_table; |
3497 | const unsigned char *extra; | |
3498 | int32_t idx; | |
3499 | int32_t elem; | |
3216711f UD |
3500 | int32_t second; |
3501 | int32_t hash; | |
3216711f | 3502 | |
3216711f UD |
3503 | table_size = |
3504 | _NL_CURRENT_WORD (LC_COLLATE, | |
3505 | _NL_COLLATE_SYMB_HASH_SIZEMB); | |
3506 | symb_table = (const int32_t *) | |
3507 | _NL_CURRENT (LC_COLLATE, | |
3508 | _NL_COLLATE_SYMB_TABLEMB); | |
3509 | extra = (const unsigned char *) | |
3510 | _NL_CURRENT (LC_COLLATE, | |
3511 | _NL_COLLATE_SYMB_EXTRAMB); | |
3512 | ||
3513 | /* Locate the character in the hashing table. */ | |
3514 | hash = elem_hash (str, c1); | |
3515 | ||
3516 | idx = 0; | |
3517 | elem = hash % table_size; | |
3518 | second = hash % (table_size - 2); | |
3519 | while (symb_table[2 * elem] != 0) | |
3520 | { | |
3521 | /* First compare the hashing value. */ | |
3522 | if (symb_table[2 * elem] == hash | |
ac8295d2 | 3523 | && c1 == extra[symb_table[2 * elem + 1]] |
3216711f UD |
3524 | && memcmp (str, |
3525 | &extra[symb_table[2 * elem + 1] | |
ac8295d2 | 3526 | + 1], |
3216711f UD |
3527 | c1) == 0) |
3528 | { | |
3529 | /* Yep, this is the entry. */ | |
ac8295d2 UD |
3530 | idx = symb_table[2 * elem + 1]; |
3531 | idx += 1 + extra[idx]; | |
3216711f UD |
3532 | break; |
3533 | } | |
3534 | ||
3535 | /* Next entry. */ | |
3536 | elem += second; | |
3537 | } | |
3538 | ||
3539 | if (symb_table[2 * elem] == 0) | |
3540 | /* This is no valid character. */ | |
3541 | FREE_STACK_RETURN (REG_ECOLLATE); | |
3542 | ||
3543 | /* Throw away the ] at the end of the equivalence | |
3544 | class. */ | |
3545 | PATFETCH (c); | |
3546 | ||
ac8295d2 | 3547 | /* Now add the multibyte character(s) we found |
f3e29a1a | 3548 | to the accept list. |
3216711f UD |
3549 | |
3550 | XXX Note that this is not entirely correct. | |
3551 | we would have to match multibyte sequences | |
3552 | but this is not possible with the current | |
ac8295d2 UD |
3553 | implementation. Also, we have to match |
3554 | collating symbols, which expand to more than | |
3555 | one file, as a whole and not allow the | |
3556 | individual bytes. */ | |
3557 | c1 = extra[idx++]; | |
3558 | if (c1 == 1) | |
3559 | range_start = extra[idx]; | |
3560 | while (c1-- > 0) | |
cd3cd00c AJ |
3561 | { |
3562 | SET_LIST_BIT (extra[idx]); | |
3563 | ++idx; | |
3564 | } | |
3216711f | 3565 | } |
e4c785c8 | 3566 | # endif |
3216711f UD |
3567 | had_char_class = false; |
3568 | } | |
a63a3c2c UD |
3569 | else |
3570 | { | |
3571 | c1++; | |
3572 | while (c1--) | |
3573 | PATUNFETCH; | |
3574 | SET_LIST_BIT ('['); | |
ac8295d2 UD |
3575 | SET_LIST_BIT ('.'); |
3576 | range_start = '.'; | |
a63a3c2c UD |
3577 | had_char_class = false; |
3578 | } | |
3579 | } | |
2b83a2a4 RM |
3580 | else |
3581 | { | |
3582 | had_char_class = false; | |
3583 | SET_LIST_BIT (c); | |
ac8295d2 | 3584 | range_start = c; |
2b83a2a4 RM |
3585 | } |
3586 | } | |
3587 | ||
3588 | /* Discard any (non)matching list bytes that are all 0 at the | |
3589 | end of the map. Decrease the map-length byte too. */ | |
91c7b85d RM |
3590 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) |
3591 | b[-1]--; | |
2b83a2a4 | 3592 | b += b[-1]; |
e4c785c8 | 3593 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
3594 | } |
3595 | break; | |
3596 | ||
3597 | ||
3598 | case '(': | |
3599 | if (syntax & RE_NO_BK_PARENS) | |
3600 | goto handle_open; | |
3601 | else | |
3602 | goto normal_char; | |
3603 | ||
3604 | ||
3605 | case ')': | |
3606 | if (syntax & RE_NO_BK_PARENS) | |
3607 | goto handle_close; | |
3608 | else | |
3609 | goto normal_char; | |
3610 | ||
3611 | ||
3612 | case '\n': | |
3613 | if (syntax & RE_NEWLINE_ALT) | |
3614 | goto handle_alt; | |
3615 | else | |
3616 | goto normal_char; | |
3617 | ||
3618 | ||
3619 | case '|': | |
3620 | if (syntax & RE_NO_BK_VBAR) | |
3621 | goto handle_alt; | |
3622 | else | |
3623 | goto normal_char; | |
3624 | ||
3625 | ||
3626 | case '{': | |
3627 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
3628 | goto handle_interval; | |
3629 | else | |
3630 | goto normal_char; | |
3631 | ||
3632 | ||
3633 | case '\\': | |
3634 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
3635 | ||
3636 | /* Do not translate the character after the \, so that we can | |
3637 | distinguish, e.g., \B from \b, even if we normally would | |
3638 | translate, e.g., B to b. */ | |
3639 | PATFETCH_RAW (c); | |
3640 | ||
3641 | switch (c) | |
3642 | { | |
3643 | case '(': | |
3644 | if (syntax & RE_NO_BK_PARENS) | |
3645 | goto normal_backslash; | |
3646 | ||
3647 | handle_open: | |
3648 | bufp->re_nsub++; | |
3649 | regnum++; | |
3650 | ||
3651 | if (COMPILE_STACK_FULL) | |
91c7b85d | 3652 | { |
2b83a2a4 RM |
3653 | RETALLOC (compile_stack.stack, compile_stack.size << 1, |
3654 | compile_stack_elt_t); | |
3655 | if (compile_stack.stack == NULL) return REG_ESPACE; | |
3656 | ||
3657 | compile_stack.size <<= 1; | |
3658 | } | |
3659 | ||
3660 | /* These are the values to restore when we hit end of this | |
3661 | group. They are all relative offsets, so that if the | |
3662 | whole pattern moves because of realloc, they will still | |
3663 | be valid. */ | |
e4c785c8 | 3664 | COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR; |
91c7b85d | 3665 | COMPILE_STACK_TOP.fixup_alt_jump |
e4c785c8 UD |
3666 | = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0; |
3667 | COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR; | |
2b83a2a4 RM |
3668 | COMPILE_STACK_TOP.regnum = regnum; |
3669 | ||
3670 | /* We will eventually replace the 0 with the number of | |
3671 | groups inner to this one. But do not push a | |
3672 | start_memory for groups beyond the last one we can | |
3673 | represent in the compiled pattern. */ | |
3674 | if (regnum <= MAX_REGNUM) | |
3675 | { | |
e4c785c8 UD |
3676 | COMPILE_STACK_TOP.inner_group_offset = b |
3677 | - COMPILED_BUFFER_VAR + 2; | |
2b83a2a4 RM |
3678 | BUF_PUSH_3 (start_memory, regnum, 0); |
3679 | } | |
91c7b85d | 3680 | |
2b83a2a4 RM |
3681 | compile_stack.avail++; |
3682 | ||
3683 | fixup_alt_jump = 0; | |
3684 | laststart = 0; | |
3685 | begalt = b; | |
3686 | /* If we've reached MAX_REGNUM groups, then this open | |
3687 | won't actually generate any code, so we'll have to | |
3688 | clear pending_exact explicitly. */ | |
3689 | pending_exact = 0; | |
3690 | break; | |
3691 | ||
3692 | ||
3693 | case ')': | |
3694 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
3695 | ||
3696 | if (COMPILE_STACK_EMPTY) | |
07b51ba5 UD |
3697 | { |
3698 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3699 | goto normal_backslash; | |
3700 | else | |
3701 | FREE_STACK_RETURN (REG_ERPAREN); | |
3702 | } | |
2b83a2a4 RM |
3703 | |
3704 | handle_close: | |
3705 | if (fixup_alt_jump) | |
3706 | { /* Push a dummy failure point at the end of the | |
3707 | alternative for a possible future | |
3708 | `pop_failure_jump' to pop. See comments at | |
3709 | `push_dummy_failure' in `re_match_2'. */ | |
3710 | BUF_PUSH (push_dummy_failure); | |
91c7b85d | 3711 | |
2b83a2a4 RM |
3712 | /* We allocated space for this jump when we assigned |
3713 | to `fixup_alt_jump', in the `handle_alt' case below. */ | |
3714 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); | |
3715 | } | |
3716 | ||
3717 | /* See similar code for backslashed left paren above. */ | |
3718 | if (COMPILE_STACK_EMPTY) | |
07b51ba5 UD |
3719 | { |
3720 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
3721 | goto normal_char; | |
3722 | else | |
3723 | FREE_STACK_RETURN (REG_ERPAREN); | |
3724 | } | |
2b83a2a4 RM |
3725 | |
3726 | /* Since we just checked for an empty stack above, this | |
3727 | ``can't happen''. */ | |
3728 | assert (compile_stack.avail != 0); | |
3729 | { | |
3730 | /* We don't just want to restore into `regnum', because | |
3731 | later groups should continue to be numbered higher, | |
3732 | as in `(ab)c(de)' -- the second group is #2. */ | |
3733 | regnum_t this_group_regnum; | |
3734 | ||
91c7b85d | 3735 | compile_stack.avail--; |
e4c785c8 | 3736 | begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset; |
2b83a2a4 RM |
3737 | fixup_alt_jump |
3738 | = COMPILE_STACK_TOP.fixup_alt_jump | |
e4c785c8 | 3739 | ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1 |
2b83a2a4 | 3740 | : 0; |
e4c785c8 | 3741 | laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset; |
2b83a2a4 RM |
3742 | this_group_regnum = COMPILE_STACK_TOP.regnum; |
3743 | /* If we've reached MAX_REGNUM groups, then this open | |
3744 | won't actually generate any code, so we'll have to | |
3745 | clear pending_exact explicitly. */ | |
3746 | pending_exact = 0; | |
3747 | ||
3748 | /* We're at the end of the group, so now we know how many | |
3749 | groups were inside this one. */ | |
3750 | if (this_group_regnum <= MAX_REGNUM) | |
3751 | { | |
e4c785c8 UD |
3752 | US_CHAR_TYPE *inner_group_loc |
3753 | = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset; | |
91c7b85d | 3754 | |
2b83a2a4 RM |
3755 | *inner_group_loc = regnum - this_group_regnum; |
3756 | BUF_PUSH_3 (stop_memory, this_group_regnum, | |
3757 | regnum - this_group_regnum); | |
3758 | } | |
3759 | } | |
3760 | break; | |
3761 | ||
3762 | ||
3763 | case '|': /* `\|'. */ | |
3764 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
3765 | goto normal_backslash; | |
3766 | handle_alt: | |
3767 | if (syntax & RE_LIMITED_OPS) | |
3768 | goto normal_char; | |
3769 | ||
3770 | /* Insert before the previous alternative a jump which | |
3771 | jumps to this alternative if the former fails. */ | |
e4c785c8 UD |
3772 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
3773 | INSERT_JUMP (on_failure_jump, begalt, | |
3774 | b + 2 + 2 * OFFSET_ADDRESS_SIZE); | |
2b83a2a4 | 3775 | pending_exact = 0; |
e4c785c8 | 3776 | b += 1 + OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
3777 | |
3778 | /* The alternative before this one has a jump after it | |
3779 | which gets executed if it gets matched. Adjust that | |
3780 | jump so it will jump to this alternative's analogous | |
3781 | jump (put in below, which in turn will jump to the next | |
3782 | (if any) alternative's such jump, etc.). The last such | |
3783 | jump jumps to the correct final destination. A picture: | |
91c7b85d RM |
3784 | _____ _____ |
3785 | | | | | | |
3786 | | v | v | |
3787 | a | b | c | |
2b83a2a4 RM |
3788 | |
3789 | If we are at `b', then fixup_alt_jump right now points to a | |
3790 | three-byte space after `a'. We'll put in the jump, set | |
3791 | fixup_alt_jump to right after `b', and leave behind three | |
3792 | bytes which we'll fill in when we get to after `c'. */ | |
3793 | ||
3794 | if (fixup_alt_jump) | |
3795 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
3796 | ||
3797 | /* Mark and leave space for a jump after this alternative, | |
3798 | to be filled in later either by next alternative or | |
3799 | when know we're at the end of a series of alternatives. */ | |
3800 | fixup_alt_jump = b; | |
e4c785c8 UD |
3801 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
3802 | b += 1 + OFFSET_ADDRESS_SIZE; | |
2b83a2a4 RM |
3803 | |
3804 | laststart = 0; | |
3805 | begalt = b; | |
3806 | break; | |
3807 | ||
3808 | ||
91c7b85d | 3809 | case '{': |
2b83a2a4 RM |
3810 | /* If \{ is a literal. */ |
3811 | if (!(syntax & RE_INTERVALS) | |
91c7b85d | 3812 | /* If we're at `\{' and it's not the open-interval |
2b83a2a4 | 3813 | operator. */ |
460e040a | 3814 | || (syntax & RE_NO_BK_BRACES)) |
2b83a2a4 RM |
3815 | goto normal_backslash; |
3816 | ||
3817 | handle_interval: | |
3818 | { | |
3819 | /* If got here, then the syntax allows intervals. */ | |
3820 | ||
3821 | /* At least (most) this many matches must be made. */ | |
3822 | int lower_bound = -1, upper_bound = -1; | |
0a45b76c UD |
3823 | |
3824 | /* Place in the uncompiled pattern (i.e., just after | |
3825 | the '{') to go back to if the interval is invalid. */ | |
3826 | const CHAR_TYPE *beg_interval = p; | |
2b83a2a4 RM |
3827 | |
3828 | if (p == pend) | |
0a45b76c | 3829 | goto invalid_interval; |
2b83a2a4 RM |
3830 | |
3831 | GET_UNSIGNED_NUMBER (lower_bound); | |
3832 | ||
3833 | if (c == ',') | |
3834 | { | |
3835 | GET_UNSIGNED_NUMBER (upper_bound); | |
9281f45d UD |
3836 | if (upper_bound < 0) |
3837 | upper_bound = RE_DUP_MAX; | |
2b83a2a4 RM |
3838 | } |
3839 | else | |
3840 | /* Interval such as `{1}' => match exactly once. */ | |
3841 | upper_bound = lower_bound; | |
3842 | ||
0a45b76c UD |
3843 | if (! (0 <= lower_bound && lower_bound <= upper_bound)) |
3844 | goto invalid_interval; | |
2b83a2a4 | 3845 | |
91c7b85d | 3846 | if (!(syntax & RE_NO_BK_BRACES)) |
2b83a2a4 | 3847 | { |
0a45b76c UD |
3848 | if (c != '\\' || p == pend) |
3849 | goto invalid_interval; | |
2b83a2a4 RM |
3850 | PATFETCH (c); |
3851 | } | |
3852 | ||
3853 | if (c != '}') | |
0a45b76c | 3854 | goto invalid_interval; |
2b83a2a4 RM |
3855 | |
3856 | /* If it's invalid to have no preceding re. */ | |
3857 | if (!laststart) | |
3858 | { | |
0a45b76c UD |
3859 | if (syntax & RE_CONTEXT_INVALID_OPS |
3860 | && !(syntax & RE_INVALID_INTERVAL_ORD)) | |
2b83a2a4 RM |
3861 | FREE_STACK_RETURN (REG_BADRPT); |
3862 | else if (syntax & RE_CONTEXT_INDEP_OPS) | |
3863 | laststart = b; | |
3864 | else | |
3865 | goto unfetch_interval; | |
3866 | } | |
3867 | ||
0a45b76c UD |
3868 | /* We just parsed a valid interval. */ |
3869 | ||
3870 | if (RE_DUP_MAX < upper_bound) | |
3871 | FREE_STACK_RETURN (REG_BADBR); | |
3872 | ||
2b83a2a4 RM |
3873 | /* If the upper bound is zero, don't want to succeed at |
3874 | all; jump from `laststart' to `b + 3', which will be | |
e4c785c8 UD |
3875 | the end of the buffer after we insert the jump. */ |
3876 | /* ifdef MBS_SUPPORT, 'b + 1 + OFFSET_ADDRESS_SIZE' | |
3877 | instead of 'b + 3'. */ | |
2b83a2a4 RM |
3878 | if (upper_bound == 0) |
3879 | { | |
e4c785c8 UD |
3880 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
3881 | INSERT_JUMP (jump, laststart, b + 1 | |
3882 | + OFFSET_ADDRESS_SIZE); | |
3883 | b += 1 + OFFSET_ADDRESS_SIZE; | |
2b83a2a4 RM |
3884 | } |
3885 | ||
3886 | /* Otherwise, we have a nontrivial interval. When | |
3887 | we're all done, the pattern will look like: | |
3888 | set_number_at <jump count> <upper bound> | |
3889 | set_number_at <succeed_n count> <lower bound> | |
3890 | succeed_n <after jump addr> <succeed_n count> | |
3891 | <body of loop> | |
3892 | jump_n <succeed_n addr> <jump count> | |
3893 | (The upper bound and `jump_n' are omitted if | |
3894 | `upper_bound' is 1, though.) */ | |
91c7b85d | 3895 | else |
2b83a2a4 RM |
3896 | { /* If the upper bound is > 1, we need to insert |
3897 | more at the end of the loop. */ | |
e4c785c8 UD |
3898 | unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE + |
3899 | (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE); | |
2b83a2a4 RM |
3900 | |
3901 | GET_BUFFER_SPACE (nbytes); | |
3902 | ||
3903 | /* Initialize lower bound of the `succeed_n', even | |
3904 | though it will be set during matching by its | |
3905 | attendant `set_number_at' (inserted next), | |
3906 | because `re_compile_fastmap' needs to know. | |
3907 | Jump to the `jump_n' we might insert below. */ | |
3908 | INSERT_JUMP2 (succeed_n, laststart, | |
e4c785c8 UD |
3909 | b + 1 + 2 * OFFSET_ADDRESS_SIZE |
3910 | + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE) | |
3911 | , lower_bound); | |
3912 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; | |
2b83a2a4 | 3913 | |
91c7b85d | 3914 | /* Code to initialize the lower bound. Insert |
2b83a2a4 RM |
3915 | before the `succeed_n'. The `5' is the last two |
3916 | bytes of this `set_number_at', plus 3 bytes of | |
3917 | the following `succeed_n'. */ | |
e4c785c8 UD |
3918 | /* ifdef MBS_SUPPORT, The '1+2*OFFSET_ADDRESS_SIZE' |
3919 | is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE' | |
3920 | of the following `succeed_n'. */ | |
3921 | insert_op2 (set_number_at, laststart, 1 | |
3922 | + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b); | |
3923 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; | |
2b83a2a4 RM |
3924 | |
3925 | if (upper_bound > 1) | |
3926 | { /* More than one repetition is allowed, so | |
3927 | append a backward jump to the `succeed_n' | |
3928 | that starts this interval. | |
91c7b85d | 3929 | |
2b83a2a4 RM |
3930 | When we've reached this during matching, |
3931 | we'll have matched the interval once, so | |
3932 | jump back only `upper_bound - 1' times. */ | |
e4c785c8 UD |
3933 | STORE_JUMP2 (jump_n, b, laststart |
3934 | + 2 * OFFSET_ADDRESS_SIZE + 1, | |
2b83a2a4 | 3935 | upper_bound - 1); |
e4c785c8 | 3936 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
3937 | |
3938 | /* The location we want to set is the second | |
3939 | parameter of the `jump_n'; that is `b-2' as | |
3940 | an absolute address. `laststart' will be | |
3941 | the `set_number_at' we're about to insert; | |
3942 | `laststart+3' the number to set, the source | |
3943 | for the relative address. But we are | |
3944 | inserting into the middle of the pattern -- | |
3945 | so everything is getting moved up by 5. | |
3946 | Conclusion: (b - 2) - (laststart + 3) + 5, | |
3947 | i.e., b - laststart. | |
91c7b85d | 3948 | |
2b83a2a4 RM |
3949 | We insert this at the beginning of the loop |
3950 | so that if we fail during matching, we'll | |
3951 | reinitialize the bounds. */ | |
3952 | insert_op2 (set_number_at, laststart, b - laststart, | |
3953 | upper_bound - 1, b); | |
e4c785c8 | 3954 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
3955 | } |
3956 | } | |
3957 | pending_exact = 0; | |
0a45b76c UD |
3958 | break; |
3959 | ||
3960 | invalid_interval: | |
3961 | if (!(syntax & RE_INVALID_INTERVAL_ORD)) | |
3962 | FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR); | |
3963 | unfetch_interval: | |
3964 | /* Match the characters as literals. */ | |
3965 | p = beg_interval; | |
3966 | c = '{'; | |
3967 | if (syntax & RE_NO_BK_BRACES) | |
3968 | goto normal_char; | |
3969 | else | |
3970 | goto normal_backslash; | |
3971 | } | |
2b83a2a4 RM |
3972 | |
3973 | #ifdef emacs | |
3974 | /* There is no way to specify the before_dot and after_dot | |
3975 | operators. rms says this is ok. --karl */ | |
3976 | case '=': | |
3977 | BUF_PUSH (at_dot); | |
3978 | break; | |
3979 | ||
91c7b85d | 3980 | case 's': |
2b83a2a4 RM |
3981 | laststart = b; |
3982 | PATFETCH (c); | |
3983 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
3984 | break; | |
3985 | ||
3986 | case 'S': | |
3987 | laststart = b; | |
3988 | PATFETCH (c); | |
3989 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
3990 | break; | |
3991 | #endif /* emacs */ | |
3992 | ||
3993 | ||
3994 | case 'w': | |
310b3460 | 3995 | if (syntax & RE_NO_GNU_OPS) |
4cca6b86 | 3996 | goto normal_char; |
2b83a2a4 RM |
3997 | laststart = b; |
3998 | BUF_PUSH (wordchar); | |
3999 | break; | |
4000 | ||
4001 | ||
4002 | case 'W': | |
310b3460 | 4003 | if (syntax & RE_NO_GNU_OPS) |
4cca6b86 | 4004 | goto normal_char; |
2b83a2a4 RM |
4005 | laststart = b; |
4006 | BUF_PUSH (notwordchar); | |
4007 | break; | |
4008 | ||
4009 | ||
4010 | case '<': | |
310b3460 | 4011 | if (syntax & RE_NO_GNU_OPS) |
4cca6b86 | 4012 | goto normal_char; |
2b83a2a4 RM |
4013 | BUF_PUSH (wordbeg); |
4014 | break; | |
4015 | ||
4016 | case '>': | |
310b3460 | 4017 | if (syntax & RE_NO_GNU_OPS) |
4cca6b86 | 4018 | goto normal_char; |
2b83a2a4 RM |
4019 | BUF_PUSH (wordend); |
4020 | break; | |
4021 | ||
4022 | case 'b': | |
310b3460 | 4023 | if (syntax & RE_NO_GNU_OPS) |
4cca6b86 | 4024 | goto normal_char; |
2b83a2a4 RM |
4025 | BUF_PUSH (wordbound); |
4026 | break; | |
4027 | ||
4028 | case 'B': | |
310b3460 | 4029 | if (syntax & RE_NO_GNU_OPS) |
4cca6b86 | 4030 | goto normal_char; |
2b83a2a4 RM |
4031 | BUF_PUSH (notwordbound); |
4032 | break; | |
4033 | ||
4034 | case '`': | |
310b3460 | 4035 | if (syntax & RE_NO_GNU_OPS) |
4cca6b86 | 4036 | goto normal_char; |
2b83a2a4 RM |
4037 | BUF_PUSH (begbuf); |
4038 | break; | |
4039 | ||
4040 | case '\'': | |
310b3460 | 4041 | if (syntax & RE_NO_GNU_OPS) |
4cca6b86 | 4042 | goto normal_char; |
2b83a2a4 RM |
4043 | BUF_PUSH (endbuf); |
4044 | break; | |
4045 | ||
4046 | case '1': case '2': case '3': case '4': case '5': | |
4047 | case '6': case '7': case '8': case '9': | |
4048 | if (syntax & RE_NO_BK_REFS) | |
4049 | goto normal_char; | |
4050 | ||
4051 | c1 = c - '0'; | |
4052 | ||
4053 | if (c1 > regnum) | |
4054 | FREE_STACK_RETURN (REG_ESUBREG); | |
4055 | ||
4056 | /* Can't back reference to a subexpression if inside of it. */ | |
4cca6b86 | 4057 | if (group_in_compile_stack (compile_stack, (regnum_t) c1)) |
2b83a2a4 RM |
4058 | goto normal_char; |
4059 | ||
4060 | laststart = b; | |
4061 | BUF_PUSH_2 (duplicate, c1); | |
4062 | break; | |
4063 | ||
4064 | ||
4065 | case '+': | |
4066 | case '?': | |
4067 | if (syntax & RE_BK_PLUS_QM) | |
4068 | goto handle_plus; | |
4069 | else | |
4070 | goto normal_backslash; | |
4071 | ||
4072 | default: | |
4073 | normal_backslash: | |
4074 | /* You might think it would be useful for \ to mean | |
4075 | not to translate; but if we don't translate it | |
4076 | it will never match anything. */ | |
4077 | c = TRANSLATE (c); | |
4078 | goto normal_char; | |
4079 | } | |
4080 | break; | |
4081 | ||
4082 | ||
4083 | default: | |
4084 | /* Expects the character in `c'. */ | |
4085 | normal_char: | |
4086 | /* If no exactn currently being built. */ | |
91c7b85d | 4087 | if (!pending_exact |
e4c785c8 UD |
4088 | #ifdef MBS_SUPPORT |
4089 | /* If last exactn handle binary(or character) and | |
4090 | new exactn handle character(or binary). */ | |
4091 | || is_exactn_bin != is_binary[p - 1 - pattern] | |
4092 | #endif /* MBS_SUPPORT */ | |
2b83a2a4 RM |
4093 | |
4094 | /* If last exactn not at current position. */ | |
4095 | || pending_exact + *pending_exact + 1 != b | |
91c7b85d | 4096 | |
2b83a2a4 RM |
4097 | /* We have only one byte following the exactn for the count. */ |
4098 | || *pending_exact == (1 << BYTEWIDTH) - 1 | |
4099 | ||
4100 | /* If followed by a repetition operator. */ | |
4101 | || *p == '*' || *p == '^' | |
4102 | || ((syntax & RE_BK_PLUS_QM) | |
4103 | ? *p == '\\' && (p[1] == '+' || p[1] == '?') | |
4104 | : (*p == '+' || *p == '?')) | |
4105 | || ((syntax & RE_INTERVALS) | |
4106 | && ((syntax & RE_NO_BK_BRACES) | |
4107 | ? *p == '{' | |
4108 | : (p[0] == '\\' && p[1] == '{')))) | |
4109 | { | |
4110 | /* Start building a new exactn. */ | |
91c7b85d | 4111 | |
2b83a2a4 RM |
4112 | laststart = b; |
4113 | ||
e4c785c8 UD |
4114 | #ifdef MBS_SUPPORT |
4115 | /* Is this exactn binary data or character? */ | |
4116 | is_exactn_bin = is_binary[p - 1 - pattern]; | |
4117 | if (is_exactn_bin) | |
4118 | BUF_PUSH_2 (exactn_bin, 0); | |
4119 | else | |
4120 | BUF_PUSH_2 (exactn, 0); | |
4121 | #else | |
2b83a2a4 | 4122 | BUF_PUSH_2 (exactn, 0); |
e4c785c8 | 4123 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
4124 | pending_exact = b - 1; |
4125 | } | |
91c7b85d | 4126 | |
2b83a2a4 RM |
4127 | BUF_PUSH (c); |
4128 | (*pending_exact)++; | |
4129 | break; | |
4130 | } /* switch (c) */ | |
4131 | } /* while p != pend */ | |
4132 | ||
91c7b85d | 4133 | |
2b83a2a4 | 4134 | /* Through the pattern now. */ |
91c7b85d | 4135 | |
2b83a2a4 RM |
4136 | if (fixup_alt_jump) |
4137 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); | |
4138 | ||
91c7b85d | 4139 | if (!COMPILE_STACK_EMPTY) |
2b83a2a4 RM |
4140 | FREE_STACK_RETURN (REG_EPAREN); |
4141 | ||
4142 | /* If we don't want backtracking, force success | |
4143 | the first time we reach the end of the compiled pattern. */ | |
4144 | if (syntax & RE_NO_POSIX_BACKTRACKING) | |
4145 | BUF_PUSH (succeed); | |
4146 | ||
e4c785c8 UD |
4147 | #ifdef MBS_SUPPORT |
4148 | free (pattern); | |
4149 | free (mbs_offset); | |
4150 | free (is_binary); | |
4151 | #endif | |
2b83a2a4 RM |
4152 | free (compile_stack.stack); |
4153 | ||
4154 | /* We have succeeded; set the length of the buffer. */ | |
e4c785c8 | 4155 | #ifdef MBS_SUPPORT |
672fd41b | 4156 | bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR; |
e4c785c8 | 4157 | #else |
2b83a2a4 | 4158 | bufp->used = b - bufp->buffer; |
e4c785c8 | 4159 | #endif |
2b83a2a4 RM |
4160 | |
4161 | #ifdef DEBUG | |
4162 | if (debug) | |
4163 | { | |
4164 | DEBUG_PRINT1 ("\nCompiled pattern: \n"); | |
4165 | print_compiled_pattern (bufp); | |
4166 | } | |
4167 | #endif /* DEBUG */ | |
4168 | ||
4169 | #ifndef MATCH_MAY_ALLOCATE | |
4170 | /* Initialize the failure stack to the largest possible stack. This | |
4171 | isn't necessary unless we're trying to avoid calling alloca in | |
4172 | the search and match routines. */ | |
4173 | { | |
4174 | int num_regs = bufp->re_nsub + 1; | |
4175 | ||
4176 | /* Since DOUBLE_FAIL_STACK refuses to double only if the current size | |
4177 | is strictly greater than re_max_failures, the largest possible stack | |
4178 | is 2 * re_max_failures failure points. */ | |
4179 | if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) | |
4180 | { | |
4181 | fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); | |
4182 | ||
86187531 | 4183 | # ifdef emacs |
2b83a2a4 RM |
4184 | if (! fail_stack.stack) |
4185 | fail_stack.stack | |
91c7b85d | 4186 | = (fail_stack_elt_t *) xmalloc (fail_stack.size |
2b83a2a4 RM |
4187 | * sizeof (fail_stack_elt_t)); |
4188 | else | |
4189 | fail_stack.stack | |
4190 | = (fail_stack_elt_t *) xrealloc (fail_stack.stack, | |
4191 | (fail_stack.size | |
4192 | * sizeof (fail_stack_elt_t))); | |
86187531 | 4193 | # else /* not emacs */ |
2b83a2a4 RM |
4194 | if (! fail_stack.stack) |
4195 | fail_stack.stack | |
91c7b85d | 4196 | = (fail_stack_elt_t *) malloc (fail_stack.size |
2b83a2a4 RM |
4197 | * sizeof (fail_stack_elt_t)); |
4198 | else | |
4199 | fail_stack.stack | |
4200 | = (fail_stack_elt_t *) realloc (fail_stack.stack, | |
4201 | (fail_stack.size | |
4202 | * sizeof (fail_stack_elt_t))); | |
86187531 | 4203 | # endif /* not emacs */ |
2b83a2a4 RM |
4204 | } |
4205 | ||
4206 | regex_grow_registers (num_regs); | |
4207 | } | |
4208 | #endif /* not MATCH_MAY_ALLOCATE */ | |
4209 | ||
4210 | return REG_NOERROR; | |
4211 | } /* regex_compile */ | |
4212 | \f | |
4213 | /* Subroutines for `regex_compile'. */ | |
4214 | ||
4215 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ | |
e4c785c8 | 4216 | /* ifdef MBS_SUPPORT, integer parameter is 1 wchar_t. */ |
2b83a2a4 RM |
4217 | |
4218 | static void | |
4219 | store_op1 (op, loc, arg) | |
4220 | re_opcode_t op; | |
e4c785c8 | 4221 | US_CHAR_TYPE *loc; |
2b83a2a4 RM |
4222 | int arg; |
4223 | { | |
e4c785c8 | 4224 | *loc = (US_CHAR_TYPE) op; |
2b83a2a4 RM |
4225 | STORE_NUMBER (loc + 1, arg); |
4226 | } | |
4227 | ||
4228 | ||
4229 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
e4c785c8 | 4230 | /* ifdef MBS_SUPPORT, integer parameter is 1 wchar_t. */ |
2b83a2a4 RM |
4231 | |
4232 | static void | |
4233 | store_op2 (op, loc, arg1, arg2) | |
4234 | re_opcode_t op; | |
e4c785c8 | 4235 | US_CHAR_TYPE *loc; |
2b83a2a4 RM |
4236 | int arg1, arg2; |
4237 | { | |
e4c785c8 | 4238 | *loc = (US_CHAR_TYPE) op; |
2b83a2a4 | 4239 | STORE_NUMBER (loc + 1, arg1); |
e4c785c8 | 4240 | STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2); |
2b83a2a4 RM |
4241 | } |
4242 | ||
4243 | ||
4244 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
4245 | for OP followed by two-byte integer parameter ARG. */ | |
e4c785c8 | 4246 | /* ifdef MBS_SUPPORT, integer parameter is 1 wchar_t. */ |
2b83a2a4 RM |
4247 | |
4248 | static void | |
4249 | insert_op1 (op, loc, arg, end) | |
4250 | re_opcode_t op; | |
e4c785c8 | 4251 | US_CHAR_TYPE *loc; |
2b83a2a4 | 4252 | int arg; |
e4c785c8 | 4253 | US_CHAR_TYPE *end; |
2b83a2a4 | 4254 | { |
e4c785c8 UD |
4255 | register US_CHAR_TYPE *pfrom = end; |
4256 | register US_CHAR_TYPE *pto = end + 1 + OFFSET_ADDRESS_SIZE; | |
2b83a2a4 RM |
4257 | |
4258 | while (pfrom != loc) | |
4259 | *--pto = *--pfrom; | |
91c7b85d | 4260 | |
2b83a2a4 RM |
4261 | store_op1 (op, loc, arg); |
4262 | } | |
4263 | ||
4264 | ||
4265 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
e4c785c8 | 4266 | /* ifdef MBS_SUPPORT, integer parameter is 1 wchar_t. */ |
2b83a2a4 RM |
4267 | |
4268 | static void | |
4269 | insert_op2 (op, loc, arg1, arg2, end) | |
4270 | re_opcode_t op; | |
e4c785c8 | 4271 | US_CHAR_TYPE *loc; |
2b83a2a4 | 4272 | int arg1, arg2; |
e4c785c8 | 4273 | US_CHAR_TYPE *end; |
2b83a2a4 | 4274 | { |
e4c785c8 UD |
4275 | register US_CHAR_TYPE *pfrom = end; |
4276 | register US_CHAR_TYPE *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE; | |
2b83a2a4 RM |
4277 | |
4278 | while (pfrom != loc) | |
4279 | *--pto = *--pfrom; | |
91c7b85d | 4280 | |
2b83a2a4 RM |
4281 | store_op2 (op, loc, arg1, arg2); |
4282 | } | |
4283 | ||
4284 | ||
4285 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
4286 | after an alternative or a begin-subexpression. We assume there is at | |
4287 | least one character before the ^. */ | |
4288 | ||
4289 | static boolean | |
4290 | at_begline_loc_p (pattern, p, syntax) | |
e4c785c8 | 4291 | const CHAR_TYPE *pattern, *p; |
2b83a2a4 RM |
4292 | reg_syntax_t syntax; |
4293 | { | |
e4c785c8 | 4294 | const CHAR_TYPE *prev = p - 2; |
2b83a2a4 | 4295 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
91c7b85d | 4296 | |
2b83a2a4 RM |
4297 | return |
4298 | /* After a subexpression? */ | |
4299 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
4300 | /* After an alternative? */ | |
4301 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); | |
4302 | } | |
4303 | ||
4304 | ||
4305 | /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
4306 | at least one character after the $, i.e., `P < PEND'. */ | |
4307 | ||
4308 | static boolean | |
4309 | at_endline_loc_p (p, pend, syntax) | |
e4c785c8 | 4310 | const CHAR_TYPE *p, *pend; |
4cca6b86 | 4311 | reg_syntax_t syntax; |
2b83a2a4 | 4312 | { |
e4c785c8 | 4313 | const CHAR_TYPE *next = p; |
2b83a2a4 | 4314 | boolean next_backslash = *next == '\\'; |
e4c785c8 | 4315 | const CHAR_TYPE *next_next = p + 1 < pend ? p + 1 : 0; |
91c7b85d | 4316 | |
2b83a2a4 RM |
4317 | return |
4318 | /* Before a subexpression? */ | |
4319 | (syntax & RE_NO_BK_PARENS ? *next == ')' | |
4320 | : next_backslash && next_next && *next_next == ')') | |
4321 | /* Before an alternative? */ | |
4322 | || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
4323 | : next_backslash && next_next && *next_next == '|'); | |
4324 | } | |
4325 | ||
4326 | ||
91c7b85d | 4327 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
2b83a2a4 RM |
4328 | false if it's not. */ |
4329 | ||
4330 | static boolean | |
4331 | group_in_compile_stack (compile_stack, regnum) | |
4332 | compile_stack_type compile_stack; | |
4333 | regnum_t regnum; | |
4334 | { | |
4335 | int this_element; | |
4336 | ||
91c7b85d RM |
4337 | for (this_element = compile_stack.avail - 1; |
4338 | this_element >= 0; | |
2b83a2a4 RM |
4339 | this_element--) |
4340 | if (compile_stack.stack[this_element].regnum == regnum) | |
4341 | return true; | |
4342 | ||
4343 | return false; | |
4344 | } | |
4345 | ||
e4c785c8 | 4346 | #ifdef MBS_SUPPORT |
054d2bf7 UD |
4347 | /* This insert space, which size is "num", into the pattern at "loc". |
4348 | "end" must point the end of the allocated buffer. */ | |
e4c785c8 UD |
4349 | static void |
4350 | insert_space (num, loc, end) | |
4351 | int num; | |
4352 | CHAR_TYPE *loc; | |
4353 | CHAR_TYPE *end; | |
4354 | { | |
4355 | register CHAR_TYPE *pto = end; | |
4356 | register CHAR_TYPE *pfrom = end - num; | |
4357 | ||
4358 | while (pfrom >= loc) | |
4359 | *pto-- = *pfrom--; | |
4360 | } | |
4361 | #endif /* MBS_SUPPORT */ | |
4362 | ||
4363 | #ifdef MBS_SUPPORT | |
4364 | static reg_errcode_t | |
4365 | compile_range (range_start_char, p_ptr, pend, translate, syntax, b, | |
4366 | char_set) | |
4367 | CHAR_TYPE range_start_char; | |
4368 | const CHAR_TYPE **p_ptr, *pend; | |
4369 | CHAR_TYPE *char_set, *b; | |
4370 | RE_TRANSLATE_TYPE translate; | |
4371 | reg_syntax_t syntax; | |
4372 | { | |
4373 | const CHAR_TYPE *p = *p_ptr; | |
4374 | CHAR_TYPE range_start, range_end; | |
4375 | reg_errcode_t ret; | |
4376 | # ifdef _LIBC | |
4377 | uint32_t nrules; | |
4378 | uint32_t start_val, end_val; | |
4379 | # endif | |
4380 | if (p == pend) | |
4381 | return REG_ERANGE; | |
4382 | ||
4383 | # ifdef _LIBC | |
4384 | nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
4385 | if (nrules != 0) | |
4386 | { | |
4387 | const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE, | |
4388 | _NL_COLLATE_COLLSEQWC); | |
054d2bf7 UD |
4389 | const unsigned char *extra = (const unsigned char *) |
4390 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); | |
e4c785c8 UD |
4391 | |
4392 | if (range_start_char < -1) | |
4393 | { | |
4394 | /* range_start is a collating symbol. */ | |
4395 | int32_t *wextra; | |
4396 | /* Retreive the index and get collation sequence value. */ | |
054d2bf7 | 4397 | wextra = (int32_t*)(extra + char_set[-range_start_char]); |
e4c785c8 UD |
4398 | start_val = wextra[1 + *wextra]; |
4399 | } | |
4400 | else | |
4401 | start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char)); | |
4402 | ||
4403 | end_val = collseq_table_lookup (collseq, TRANSLATE (p[0])); | |
4404 | ||
4405 | /* Report an error if the range is empty and the syntax prohibits | |
4406 | this. */ | |
4407 | ret = ((syntax & RE_NO_EMPTY_RANGES) | |
4408 | && (start_val > end_val))? REG_ERANGE : REG_NOERROR; | |
4409 | ||
4410 | /* Insert space to the end of the char_ranges. */ | |
4411 | insert_space(2, b - char_set[5] - 2, b - 1); | |
4412 | *(b - char_set[5] - 2) = (wchar_t)start_val; | |
4413 | *(b - char_set[5] - 1) = (wchar_t)end_val; | |
4414 | char_set[4]++; /* ranges_index */ | |
4415 | } | |
4416 | else | |
4417 | # endif | |
4418 | { | |
4419 | range_start = (range_start_char >= 0)? TRANSLATE (range_start_char): | |
4420 | range_start_char; | |
4421 | range_end = TRANSLATE (p[0]); | |
4422 | /* Report an error if the range is empty and the syntax prohibits | |
4423 | this. */ | |
4424 | ret = ((syntax & RE_NO_EMPTY_RANGES) | |
4425 | && (range_start > range_end))? REG_ERANGE : REG_NOERROR; | |
4426 | ||
4427 | /* Insert space to the end of the char_ranges. */ | |
4428 | insert_space(2, b - char_set[5] - 2, b - 1); | |
4429 | *(b - char_set[5] - 2) = range_start; | |
4430 | *(b - char_set[5] - 1) = range_end; | |
4431 | char_set[4]++; /* ranges_index */ | |
4432 | } | |
4433 | /* Have to increment the pointer into the pattern string, so the | |
4434 | caller isn't still at the ending character. */ | |
4435 | (*p_ptr)++; | |
2b83a2a4 | 4436 | |
e4c785c8 UD |
4437 | return ret; |
4438 | } | |
4439 | #else | |
2b83a2a4 RM |
4440 | /* Read the ending character of a range (in a bracket expression) from the |
4441 | uncompiled pattern *P_PTR (which ends at PEND). We assume the | |
4442 | starting character is in `P[-2]'. (`P[-1]' is the character `-'.) | |
4443 | Then we set the translation of all bits between the starting and | |
4444 | ending characters (inclusive) in the compiled pattern B. | |
91c7b85d | 4445 | |
2b83a2a4 | 4446 | Return an error code. |
91c7b85d | 4447 | |
2b83a2a4 RM |
4448 | We use these short variable names so we can use the same macros as |
4449 | `regex_compile' itself. */ | |
4450 | ||
4451 | static reg_errcode_t | |
14a6b4e4 UD |
4452 | compile_range (range_start_char, p_ptr, pend, translate, syntax, b) |
4453 | unsigned int range_start_char; | |
4454 | const char **p_ptr, *pend; | |
4455 | RE_TRANSLATE_TYPE translate; | |
4456 | reg_syntax_t syntax; | |
4457 | unsigned char *b; | |
2b83a2a4 RM |
4458 | { |
4459 | unsigned this_char; | |
2b83a2a4 | 4460 | const char *p = *p_ptr; |
14a6b4e4 | 4461 | reg_errcode_t ret; |
e4c785c8 | 4462 | # if _LIBC |
8868f97b UD |
4463 | const unsigned char *collseq; |
4464 | unsigned int start_colseq; | |
4465 | unsigned int end_colseq; | |
e4c785c8 | 4466 | # else |
8868f97b | 4467 | unsigned end_char; |
e4c785c8 | 4468 | # endif |
91c7b85d | 4469 | |
2b83a2a4 RM |
4470 | if (p == pend) |
4471 | return REG_ERANGE; | |
4472 | ||
2b83a2a4 RM |
4473 | /* Have to increment the pointer into the pattern string, so the |
4474 | caller isn't still at the ending character. */ | |
4475 | (*p_ptr)++; | |
4476 | ||
14a6b4e4 UD |
4477 | /* Report an error if the range is empty and the syntax prohibits this. */ |
4478 | ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
2b83a2a4 | 4479 | |
e4c785c8 | 4480 | # if _LIBC |
8868f97b UD |
4481 | collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE, |
4482 | _NL_COLLATE_COLLSEQMB); | |
4483 | ||
b5e62988 UD |
4484 | start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)]; |
4485 | end_colseq = collseq[(unsigned char) TRANSLATE (p[0])]; | |
14a6b4e4 | 4486 | for (this_char = 0; this_char <= (unsigned char) -1; ++this_char) |
2b83a2a4 | 4487 | { |
b5e62988 | 4488 | unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)]; |
8868f97b UD |
4489 | |
4490 | if (start_colseq <= this_colseq && this_colseq <= end_colseq) | |
14a6b4e4 UD |
4491 | { |
4492 | SET_LIST_BIT (TRANSLATE (this_char)); | |
4493 | ret = REG_NOERROR; | |
4494 | } | |
2b83a2a4 | 4495 | } |
e4c785c8 | 4496 | # else |
8868f97b UD |
4497 | /* Here we see why `this_char' has to be larger than an `unsigned |
4498 | char' -- we would otherwise go into an infinite loop, since all | |
4499 | characters <= 0xff. */ | |
4500 | range_start_char = TRANSLATE (range_start_char); | |
e4c785c8 UD |
4501 | /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE, |
4502 | and some compilers cast it to int implicitly, so following for_loop | |
4503 | may fall to (almost) infinite loop. | |
4504 | e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff. | |
4505 | To avoid this, we cast p[0] to unsigned int and truncate it. */ | |
4506 | end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1)); | |
4507 | ||
8868f97b UD |
4508 | for (this_char = range_start_char; this_char <= end_char; ++this_char) |
4509 | { | |
4510 | SET_LIST_BIT (TRANSLATE (this_char)); | |
4511 | ret = REG_NOERROR; | |
4512 | } | |
e4c785c8 | 4513 | # endif |
91c7b85d | 4514 | |
14a6b4e4 | 4515 | return ret; |
2b83a2a4 | 4516 | } |
e4c785c8 | 4517 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
4518 | \f |
4519 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
4520 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
4521 | characters can start a string that matches the pattern. This fastmap | |
4522 | is used by re_search to skip quickly over impossible starting points. | |
4523 | ||
4524 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
4525 | area as BUFP->fastmap. | |
91c7b85d | 4526 | |
2b83a2a4 RM |
4527 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in |
4528 | the pattern buffer. | |
4529 | ||
4530 | Returns 0 if we succeed, -2 if an internal error. */ | |
4531 | ||
e4c785c8 UD |
4532 | #ifdef MBS_SUPPORT |
4533 | /* local function for re_compile_fastmap. | |
4534 | truncate wchar_t character to char. */ | |
d4620e04 AJ |
4535 | static unsigned char truncate_wchar (CHAR_TYPE c); |
4536 | ||
4537 | static unsigned char | |
4538 | truncate_wchar (c) | |
e4c785c8 UD |
4539 | CHAR_TYPE c; |
4540 | { | |
4541 | unsigned char buf[MB_LEN_MAX]; | |
4542 | int retval = wctomb(buf, c); | |
4543 | return retval > 0 ? buf[0] : (unsigned char)c; | |
4544 | } | |
4545 | #endif /* MBS_SUPPORT */ | |
4546 | ||
2b83a2a4 RM |
4547 | int |
4548 | re_compile_fastmap (bufp) | |
4549 | struct re_pattern_buffer *bufp; | |
4550 | { | |
4551 | int j, k; | |
4552 | #ifdef MATCH_MAY_ALLOCATE | |
4553 | fail_stack_type fail_stack; | |
4554 | #endif | |
4555 | #ifndef REGEX_MALLOC | |
4556 | char *destination; | |
4557 | #endif | |
91c7b85d | 4558 | |
2b83a2a4 | 4559 | register char *fastmap = bufp->fastmap; |
e4c785c8 UD |
4560 | |
4561 | #ifdef MBS_SUPPORT | |
4562 | /* We need to cast pattern to (wchar_t*), because we casted this compiled | |
4563 | pattern to (char*) in regex_compile. */ | |
4564 | US_CHAR_TYPE *pattern = (US_CHAR_TYPE*)bufp->buffer; | |
4565 | register US_CHAR_TYPE *pend = (US_CHAR_TYPE*) (bufp->buffer + bufp->used); | |
4566 | #else | |
4567 | US_CHAR_TYPE *pattern = bufp->buffer; | |
4568 | register US_CHAR_TYPE *pend = pattern + bufp->used; | |
4569 | #endif /* MBS_SUPPORT */ | |
4570 | US_CHAR_TYPE *p = pattern; | |
2b83a2a4 | 4571 | |
4cca6b86 | 4572 | #ifdef REL_ALLOC |
2b83a2a4 RM |
4573 | /* This holds the pointer to the failure stack, when |
4574 | it is allocated relocatably. */ | |
4575 | fail_stack_elt_t *failure_stack_ptr; | |
4cca6b86 | 4576 | #endif |
2b83a2a4 RM |
4577 | |
4578 | /* Assume that each path through the pattern can be null until | |
4579 | proven otherwise. We set this false at the bottom of switch | |
4580 | statement, to which we get only if a particular path doesn't | |
4581 | match the empty string. */ | |
4582 | boolean path_can_be_null = true; | |
4583 | ||
4584 | /* We aren't doing a `succeed_n' to begin with. */ | |
4585 | boolean succeed_n_p = false; | |
4586 | ||
4587 | assert (fastmap != NULL && p != NULL); | |
91c7b85d | 4588 | |
2b83a2a4 RM |
4589 | INIT_FAIL_STACK (); |
4590 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
4591 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
4592 | bufp->can_be_null = 0; | |
91c7b85d | 4593 | |
2b83a2a4 RM |
4594 | while (1) |
4595 | { | |
4596 | if (p == pend || *p == succeed) | |
4597 | { | |
4598 | /* We have reached the (effective) end of pattern. */ | |
4599 | if (!FAIL_STACK_EMPTY ()) | |
4600 | { | |
4601 | bufp->can_be_null |= path_can_be_null; | |
4602 | ||
4603 | /* Reset for next path. */ | |
4604 | path_can_be_null = true; | |
4605 | ||
4606 | p = fail_stack.stack[--fail_stack.avail].pointer; | |
4607 | ||
4608 | continue; | |
4609 | } | |
4610 | else | |
4611 | break; | |
4612 | } | |
4613 | ||
4614 | /* We should never be about to go beyond the end of the pattern. */ | |
4615 | assert (p < pend); | |
91c7b85d | 4616 | |
2b83a2a4 RM |
4617 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
4618 | { | |
4619 | ||
4620 | /* I guess the idea here is to simply not bother with a fastmap | |
4621 | if a backreference is used, since it's too hard to figure out | |
4622 | the fastmap for the corresponding group. Setting | |
4623 | `can_be_null' stops `re_search_2' from using the fastmap, so | |
4624 | that is all we do. */ | |
4625 | case duplicate: | |
4626 | bufp->can_be_null = 1; | |
4627 | goto done; | |
4628 | ||
4629 | ||
4630 | /* Following are the cases which match a character. These end | |
4631 | with `break'. */ | |
4632 | ||
e4c785c8 UD |
4633 | #ifdef MBS_SUPPORT |
4634 | case exactn: | |
4635 | fastmap[truncate_wchar(p[1])] = 1; | |
4636 | break; | |
4637 | case exactn_bin: | |
4638 | fastmap[p[1]] = 1; | |
4639 | break; | |
4640 | #else | |
2b83a2a4 RM |
4641 | case exactn: |
4642 | fastmap[p[1]] = 1; | |
4643 | break; | |
e4c785c8 | 4644 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
4645 | |
4646 | ||
e4c785c8 UD |
4647 | #ifdef MBS_SUPPORT |
4648 | /* It is hard to distinguish fastmap from (multi byte) characters | |
4649 | which depends on current locale. */ | |
4650 | case charset: | |
4651 | case charset_not: | |
4652 | case wordchar: | |
4653 | case notwordchar: | |
4654 | bufp->can_be_null = 1; | |
4655 | goto done; | |
4656 | #else | |
2b83a2a4 RM |
4657 | case charset: |
4658 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
4659 | if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
4660 | fastmap[j] = 1; | |
4661 | break; | |
4662 | ||
4663 | ||
4664 | case charset_not: | |
4665 | /* Chars beyond end of map must be allowed. */ | |
4666 | for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) | |
4667 | fastmap[j] = 1; | |
4668 | ||
4669 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
4670 | if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
4671 | fastmap[j] = 1; | |
4672 | break; | |
4673 | ||
4674 | ||
4675 | case wordchar: | |
4676 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4677 | if (SYNTAX (j) == Sword) | |
4678 | fastmap[j] = 1; | |
4679 | break; | |
4680 | ||
4681 | ||
4682 | case notwordchar: | |
4683 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4684 | if (SYNTAX (j) != Sword) | |
4685 | fastmap[j] = 1; | |
4686 | break; | |
e4c785c8 | 4687 | #endif |
2b83a2a4 RM |
4688 | |
4689 | case anychar: | |
4690 | { | |
4691 | int fastmap_newline = fastmap['\n']; | |
4692 | ||
4693 | /* `.' matches anything ... */ | |
4694 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4695 | fastmap[j] = 1; | |
4696 | ||
4697 | /* ... except perhaps newline. */ | |
4698 | if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
4699 | fastmap['\n'] = fastmap_newline; | |
4700 | ||
4701 | /* Return if we have already set `can_be_null'; if we have, | |
4702 | then the fastmap is irrelevant. Something's wrong here. */ | |
4703 | else if (bufp->can_be_null) | |
4704 | goto done; | |
4705 | ||
4706 | /* Otherwise, have to check alternative paths. */ | |
4707 | break; | |
4708 | } | |
4709 | ||
4710 | #ifdef emacs | |
4711 | case syntaxspec: | |
4712 | k = *p++; | |
4713 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4714 | if (SYNTAX (j) == (enum syntaxcode) k) | |
4715 | fastmap[j] = 1; | |
4716 | break; | |
4717 | ||
4718 | ||
4719 | case notsyntaxspec: | |
4720 | k = *p++; | |
4721 | for (j = 0; j < (1 << BYTEWIDTH); j++) | |
4722 | if (SYNTAX (j) != (enum syntaxcode) k) | |
4723 | fastmap[j] = 1; | |
4724 | break; | |
4725 | ||
4726 | ||
4727 | /* All cases after this match the empty string. These end with | |
4728 | `continue'. */ | |
4729 | ||
4730 | ||
4731 | case before_dot: | |
4732 | case at_dot: | |
4733 | case after_dot: | |
4734 | continue; | |
44c8d1a2 | 4735 | #endif /* emacs */ |
2b83a2a4 RM |
4736 | |
4737 | ||
4738 | case no_op: | |
4739 | case begline: | |
4740 | case endline: | |
4741 | case begbuf: | |
4742 | case endbuf: | |
4743 | case wordbound: | |
4744 | case notwordbound: | |
4745 | case wordbeg: | |
4746 | case wordend: | |
4747 | case push_dummy_failure: | |
4748 | continue; | |
4749 | ||
4750 | ||
4751 | case jump_n: | |
4752 | case pop_failure_jump: | |
4753 | case maybe_pop_jump: | |
4754 | case jump: | |
4755 | case jump_past_alt: | |
4756 | case dummy_failure_jump: | |
4757 | EXTRACT_NUMBER_AND_INCR (j, p); | |
91c7b85d | 4758 | p += j; |
2b83a2a4 RM |
4759 | if (j > 0) |
4760 | continue; | |
91c7b85d | 4761 | |
2b83a2a4 RM |
4762 | /* Jump backward implies we just went through the body of a |
4763 | loop and matched nothing. Opcode jumped to should be | |
4764 | `on_failure_jump' or `succeed_n'. Just treat it like an | |
4765 | ordinary jump. For a * loop, it has pushed its failure | |
4766 | point already; if so, discard that as redundant. */ | |
4767 | if ((re_opcode_t) *p != on_failure_jump | |
4768 | && (re_opcode_t) *p != succeed_n) | |
4769 | continue; | |
4770 | ||
4771 | p++; | |
4772 | EXTRACT_NUMBER_AND_INCR (j, p); | |
91c7b85d RM |
4773 | p += j; |
4774 | ||
2b83a2a4 | 4775 | /* If what's on the stack is where we are now, pop it. */ |
91c7b85d | 4776 | if (!FAIL_STACK_EMPTY () |
2b83a2a4 RM |
4777 | && fail_stack.stack[fail_stack.avail - 1].pointer == p) |
4778 | fail_stack.avail--; | |
4779 | ||
4780 | continue; | |
4781 | ||
4782 | ||
4783 | case on_failure_jump: | |
4784 | case on_failure_keep_string_jump: | |
4785 | handle_on_failure_jump: | |
4786 | EXTRACT_NUMBER_AND_INCR (j, p); | |
4787 | ||
4788 | /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
4789 | end of the pattern. We don't want to push such a point, | |
4790 | since when we restore it above, entering the switch will | |
4791 | increment `p' past the end of the pattern. We don't need | |
4792 | to push such a point since we obviously won't find any more | |
4793 | fastmap entries beyond `pend'. Such a pattern can match | |
4794 | the null string, though. */ | |
4795 | if (p + j < pend) | |
4796 | { | |
4797 | if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
4798 | { | |
4799 | RESET_FAIL_STACK (); | |
4800 | return -2; | |
4801 | } | |
4802 | } | |
4803 | else | |
4804 | bufp->can_be_null = 1; | |
4805 | ||
4806 | if (succeed_n_p) | |
4807 | { | |
4808 | EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
4809 | succeed_n_p = false; | |
4810 | } | |
4811 | ||
4812 | continue; | |
4813 | ||
4814 | ||
4815 | case succeed_n: | |
4816 | /* Get to the number of times to succeed. */ | |
e4c785c8 | 4817 | p += OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
4818 | |
4819 | /* Increment p past the n for when k != 0. */ | |
4820 | EXTRACT_NUMBER_AND_INCR (k, p); | |
4821 | if (k == 0) | |
4822 | { | |
e4c785c8 | 4823 | p -= 2 * OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
4824 | succeed_n_p = true; /* Spaghetti code alert. */ |
4825 | goto handle_on_failure_jump; | |
4826 | } | |
4827 | continue; | |
4828 | ||
4829 | ||
4830 | case set_number_at: | |
e4c785c8 | 4831 | p += 2 * OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
4832 | continue; |
4833 | ||
4834 | ||
4835 | case start_memory: | |
4836 | case stop_memory: | |
4837 | p += 2; | |
4838 | continue; | |
4839 | ||
4840 | ||
4841 | default: | |
4842 | abort (); /* We have listed all the cases. */ | |
4843 | } /* switch *p++ */ | |
4844 | ||
4845 | /* Getting here means we have found the possible starting | |
4846 | characters for one path of the pattern -- and that the empty | |
4847 | string does not match. We need not follow this path further. | |
4848 | Instead, look at the next alternative (remembered on the | |
4849 | stack), or quit if no more. The test at the top of the loop | |
4850 | does these things. */ | |
4851 | path_can_be_null = false; | |
4852 | p = pend; | |
4853 | } /* while p */ | |
4854 | ||
4855 | /* Set `can_be_null' for the last path (also the first path, if the | |
4856 | pattern is empty). */ | |
4857 | bufp->can_be_null |= path_can_be_null; | |
4858 | ||
4859 | done: | |
4860 | RESET_FAIL_STACK (); | |
4861 | return 0; | |
4862 | } /* re_compile_fastmap */ | |
2ad4fab2 UD |
4863 | #ifdef _LIBC |
4864 | weak_alias (__re_compile_fastmap, re_compile_fastmap) | |
4865 | #endif | |
2b83a2a4 RM |
4866 | \f |
4867 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
4868 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
4869 | this memory for recording register information. STARTS and ENDS | |
4870 | must be allocated using the malloc library routine, and must each | |
4871 | be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
4872 | ||
4873 | If NUM_REGS == 0, then subsequent matches should allocate their own | |
4874 | register data. | |
4875 | ||
4876 | Unless this function is called, the first search or match using | |
4877 | PATTERN_BUFFER will allocate its own register data, without | |
4878 | freeing the old data. */ | |
4879 | ||
4880 | void | |
4881 | re_set_registers (bufp, regs, num_regs, starts, ends) | |
4882 | struct re_pattern_buffer *bufp; | |
4883 | struct re_registers *regs; | |
4884 | unsigned num_regs; | |
4885 | regoff_t *starts, *ends; | |
4886 | { | |
4887 | if (num_regs) | |
4888 | { | |
4889 | bufp->regs_allocated = REGS_REALLOCATE; | |
4890 | regs->num_regs = num_regs; | |
4891 | regs->start = starts; | |
4892 | regs->end = ends; | |
4893 | } | |
4894 | else | |
4895 | { | |
4896 | bufp->regs_allocated = REGS_UNALLOCATED; | |
4897 | regs->num_regs = 0; | |
4898 | regs->start = regs->end = (regoff_t *) 0; | |
4899 | } | |
4900 | } | |
2ad4fab2 UD |
4901 | #ifdef _LIBC |
4902 | weak_alias (__re_set_registers, re_set_registers) | |
4903 | #endif | |
2b83a2a4 RM |
4904 | \f |
4905 | /* Searching routines. */ | |
4906 | ||
4907 | /* Like re_search_2, below, but only one string is specified, and | |
e4c785c8 | 4908 | doesn't let you say where to stop matching. */ |
2b83a2a4 RM |
4909 | |
4910 | int | |
4911 | re_search (bufp, string, size, startpos, range, regs) | |
4912 | struct re_pattern_buffer *bufp; | |
4913 | const char *string; | |
4914 | int size, startpos, range; | |
4915 | struct re_registers *regs; | |
4916 | { | |
91c7b85d | 4917 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
2b83a2a4 RM |
4918 | regs, size); |
4919 | } | |
2ad4fab2 UD |
4920 | #ifdef _LIBC |
4921 | weak_alias (__re_search, re_search) | |
4922 | #endif | |
2b83a2a4 RM |
4923 | |
4924 | ||
4925 | /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
4926 | virtual concatenation of STRING1 and STRING2, starting first at index | |
4927 | STARTPOS, then at STARTPOS + 1, and so on. | |
91c7b85d | 4928 | |
2b83a2a4 | 4929 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
91c7b85d | 4930 | |
2b83a2a4 RM |
4931 | RANGE is how far to scan while trying to match. RANGE = 0 means try |
4932 | only at STARTPOS; in general, the last start tried is STARTPOS + | |
4933 | RANGE. | |
91c7b85d | 4934 | |
2b83a2a4 RM |
4935 | In REGS, return the indices of the virtual concatenation of STRING1 |
4936 | and STRING2 that matched the entire BUFP->buffer and its contained | |
4937 | subexpressions. | |
91c7b85d | 4938 | |
2b83a2a4 RM |
4939 | Do not consider matching one past the index STOP in the virtual |
4940 | concatenation of STRING1 and STRING2. | |
4941 | ||
4942 | We return either the position in the strings at which the match was | |
4943 | found, -1 if no match, or -2 if error (such as failure | |
4944 | stack overflow). */ | |
4945 | ||
4946 | int | |
4947 | re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) | |
4948 | struct re_pattern_buffer *bufp; | |
4949 | const char *string1, *string2; | |
4950 | int size1, size2; | |
4951 | int startpos; | |
4952 | int range; | |
4953 | struct re_registers *regs; | |
4954 | int stop; | |
4955 | { | |
4956 | int val; | |
4957 | register char *fastmap = bufp->fastmap; | |
03a75825 | 4958 | register RE_TRANSLATE_TYPE translate = bufp->translate; |
2b83a2a4 RM |
4959 | int total_size = size1 + size2; |
4960 | int endpos = startpos + range; | |
4961 | ||
4962 | /* Check for out-of-range STARTPOS. */ | |
4963 | if (startpos < 0 || startpos > total_size) | |
4964 | return -1; | |
91c7b85d | 4965 | |
2b83a2a4 | 4966 | /* Fix up RANGE if it might eventually take us outside |
57aefafe | 4967 | the virtual concatenation of STRING1 and STRING2. |
91c7b85d | 4968 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
57aefafe RM |
4969 | if (endpos < 0) |
4970 | range = 0 - startpos; | |
2b83a2a4 RM |
4971 | else if (endpos > total_size) |
4972 | range = total_size - startpos; | |
4973 | ||
4974 | /* If the search isn't to be a backwards one, don't waste time in a | |
4975 | search for a pattern that must be anchored. */ | |
7cabd57c UD |
4976 | if (bufp->used > 0 && range > 0 |
4977 | && ((re_opcode_t) bufp->buffer[0] == begbuf | |
4978 | /* `begline' is like `begbuf' if it cannot match at newlines. */ | |
4979 | || ((re_opcode_t) bufp->buffer[0] == begline | |
4980 | && !bufp->newline_anchor))) | |
2b83a2a4 RM |
4981 | { |
4982 | if (startpos > 0) | |
4983 | return -1; | |
4984 | else | |
4985 | range = 1; | |
4986 | } | |
4987 | ||
44c8d1a2 RM |
4988 | #ifdef emacs |
4989 | /* In a forward search for something that starts with \=. | |
4990 | don't keep searching past point. */ | |
4991 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
4992 | { | |
4993 | range = PT - startpos; | |
4994 | if (range <= 0) | |
4995 | return -1; | |
4996 | } | |
4997 | #endif /* emacs */ | |
4998 | ||
2b83a2a4 RM |
4999 | /* Update the fastmap now if not correct already. */ |
5000 | if (fastmap && !bufp->fastmap_accurate) | |
5001 | if (re_compile_fastmap (bufp) == -2) | |
5002 | return -2; | |
91c7b85d | 5003 | |
2b83a2a4 RM |
5004 | /* Loop through the string, looking for a place to start matching. */ |
5005 | for (;;) | |
91c7b85d | 5006 | { |
2b83a2a4 RM |
5007 | /* If a fastmap is supplied, skip quickly over characters that |
5008 | cannot be the start of a match. If the pattern can match the | |
5009 | null string, however, we don't need to skip characters; we want | |
5010 | the first null string. */ | |
5011 | if (fastmap && startpos < total_size && !bufp->can_be_null) | |
5012 | { | |
5013 | if (range > 0) /* Searching forwards. */ | |
5014 | { | |
5015 | register const char *d; | |
5016 | register int lim = 0; | |
5017 | int irange = range; | |
5018 | ||
5019 | if (startpos < size1 && startpos + range >= size1) | |
5020 | lim = range - (size1 - startpos); | |
5021 | ||
5022 | d = (startpos >= size1 ? string2 - size1 : string1) + startpos; | |
91c7b85d | 5023 | |
2b83a2a4 RM |
5024 | /* Written out as an if-else to avoid testing `translate' |
5025 | inside the loop. */ | |
5026 | if (translate) | |
5027 | while (range > lim | |
5028 | && !fastmap[(unsigned char) | |
5029 | translate[(unsigned char) *d++]]) | |
5030 | range--; | |
5031 | else | |
5032 | while (range > lim && !fastmap[(unsigned char) *d++]) | |
5033 | range--; | |
5034 | ||
5035 | startpos += irange - range; | |
5036 | } | |
5037 | else /* Searching backwards. */ | |
5038 | { | |
2d0aea11 UD |
5039 | register CHAR_TYPE c = (size1 == 0 || startpos >= size1 |
5040 | ? string2[startpos - size1] | |
5041 | : string1[startpos]); | |
2b83a2a4 RM |
5042 | |
5043 | if (!fastmap[(unsigned char) TRANSLATE (c)]) | |
5044 | goto advance; | |
5045 | } | |
5046 | } | |
5047 | ||
5048 | /* If can't match the null string, and that's all we have left, fail. */ | |
5049 | if (range >= 0 && startpos == total_size && fastmap | |
5050 | && !bufp->can_be_null) | |
5051 | return -1; | |
5052 | ||
5053 | val = re_match_2_internal (bufp, string1, size1, string2, size2, | |
5054 | startpos, regs, stop); | |
5055 | #ifndef REGEX_MALLOC | |
86187531 | 5056 | # ifdef C_ALLOCA |
2b83a2a4 | 5057 | alloca (0); |
86187531 | 5058 | # endif |
2b83a2a4 RM |
5059 | #endif |
5060 | ||
5061 | if (val >= 0) | |
5062 | return startpos; | |
91c7b85d | 5063 | |
2b83a2a4 RM |
5064 | if (val == -2) |
5065 | return -2; | |
5066 | ||
5067 | advance: | |
91c7b85d | 5068 | if (!range) |
2b83a2a4 | 5069 | break; |
91c7b85d | 5070 | else if (range > 0) |
2b83a2a4 | 5071 | { |
91c7b85d | 5072 | range--; |
2b83a2a4 RM |
5073 | startpos++; |
5074 | } | |
5075 | else | |
5076 | { | |
91c7b85d | 5077 | range++; |
2b83a2a4 RM |
5078 | startpos--; |
5079 | } | |
5080 | } | |
5081 | return -1; | |
5082 | } /* re_search_2 */ | |
2ad4fab2 UD |
5083 | #ifdef _LIBC |
5084 | weak_alias (__re_search_2, re_search_2) | |
5085 | #endif | |
2b83a2a4 | 5086 | \f |
e4c785c8 UD |
5087 | #ifdef MBS_SUPPORT |
5088 | /* This converts PTR, a pointer into one of the search wchar_t strings | |
5089 | `string1' and `string2' into an multibyte string offset from the | |
5090 | beginning of that string. We use mbs_offset to optimize. | |
5091 | See convert_mbs_to_wcs. */ | |
5092 | # define POINTER_TO_OFFSET(ptr) \ | |
5093 | (FIRST_STRING_P (ptr) \ | |
5094 | ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \ | |
5095 | : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \ | |
5096 | + csize1))) | |
5097 | #else | |
2b83a2a4 RM |
5098 | /* This converts PTR, a pointer into one of the search strings `string1' |
5099 | and `string2' into an offset from the beginning of that string. */ | |
e4c785c8 | 5100 | # define POINTER_TO_OFFSET(ptr) \ |
2b83a2a4 RM |
5101 | (FIRST_STRING_P (ptr) \ |
5102 | ? ((regoff_t) ((ptr) - string1)) \ | |
5103 | : ((regoff_t) ((ptr) - string2 + size1))) | |
e4c785c8 | 5104 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
5105 | |
5106 | /* Macros for dealing with the split strings in re_match_2. */ | |
5107 | ||
5108 | #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
5109 | ||
5110 | /* Call before fetching a character with *d. This switches over to | |
5111 | string2 if necessary. */ | |
5112 | #define PREFETCH() \ | |
5113 | while (d == dend) \ | |
5114 | { \ | |
5115 | /* End of string2 => fail. */ \ | |
5116 | if (dend == end_match_2) \ | |
5117 | goto fail; \ | |
5118 | /* End of string1 => advance to string2. */ \ | |
5119 | d = string2; \ | |
5120 | dend = end_match_2; \ | |
5121 | } | |
5122 | ||
5123 | ||
5124 | /* Test if at very beginning or at very end of the virtual concatenation | |
5125 | of `string1' and `string2'. If only one string, it's `string2'. */ | |
5126 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) | |
91c7b85d | 5127 | #define AT_STRINGS_END(d) ((d) == end2) |
2b83a2a4 RM |
5128 | |
5129 | ||
5130 | /* Test if D points to a character which is word-constituent. We have | |
5131 | two special cases to check for: if past the end of string1, look at | |
5132 | the first character in string2; and if before the beginning of | |
5133 | string2, look at the last character in string1. */ | |
e4c785c8 UD |
5134 | #ifdef MBS_SUPPORT |
5135 | /* Use internationalized API instead of SYNTAX. */ | |
5136 | # define WORDCHAR_P(d) \ | |
5137 | (iswalnum ((wint_t)((d) == end1 ? *string2 \ | |
5138 | : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0) | |
5139 | #else | |
5140 | # define WORDCHAR_P(d) \ | |
2b83a2a4 RM |
5141 | (SYNTAX ((d) == end1 ? *string2 \ |
5142 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ | |
5143 | == Sword) | |
e4c785c8 | 5144 | #endif /* MBS_SUPPORT */ |
2b83a2a4 | 5145 | |
51702635 UD |
5146 | /* Disabled due to a compiler bug -- see comment at case wordbound */ |
5147 | #if 0 | |
2b83a2a4 RM |
5148 | /* Test if the character before D and the one at D differ with respect |
5149 | to being word-constituent. */ | |
5150 | #define AT_WORD_BOUNDARY(d) \ | |
5151 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ | |
5152 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) | |
51702635 | 5153 | #endif |
2b83a2a4 RM |
5154 | |
5155 | /* Free everything we malloc. */ | |
5156 | #ifdef MATCH_MAY_ALLOCATE | |
86187531 | 5157 | # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL |
e4c785c8 UD |
5158 | # ifdef MBS_SUPPORT |
5159 | # define FREE_VARIABLES() \ | |
5160 | do { \ | |
5161 | REGEX_FREE_STACK (fail_stack.stack); \ | |
5162 | FREE_VAR (regstart); \ | |
5163 | FREE_VAR (regend); \ | |
5164 | FREE_VAR (old_regstart); \ | |
5165 | FREE_VAR (old_regend); \ | |
5166 | FREE_VAR (best_regstart); \ | |
5167 | FREE_VAR (best_regend); \ | |
5168 | FREE_VAR (reg_info); \ | |
5169 | FREE_VAR (reg_dummy); \ | |
5170 | FREE_VAR (reg_info_dummy); \ | |
5171 | FREE_VAR (string1); \ | |
5172 | FREE_VAR (string2); \ | |
5173 | FREE_VAR (mbs_offset1); \ | |
5174 | FREE_VAR (mbs_offset2); \ | |
e4c785c8 UD |
5175 | } while (0) |
5176 | # else /* not MBS_SUPPORT */ | |
5177 | # define FREE_VARIABLES() \ | |
2b83a2a4 RM |
5178 | do { \ |
5179 | REGEX_FREE_STACK (fail_stack.stack); \ | |
5180 | FREE_VAR (regstart); \ | |
5181 | FREE_VAR (regend); \ | |
5182 | FREE_VAR (old_regstart); \ | |
5183 | FREE_VAR (old_regend); \ | |
5184 | FREE_VAR (best_regstart); \ | |
5185 | FREE_VAR (best_regend); \ | |
5186 | FREE_VAR (reg_info); \ | |
5187 | FREE_VAR (reg_dummy); \ | |
5188 | FREE_VAR (reg_info_dummy); \ | |
5189 | } while (0) | |
e4c785c8 | 5190 | # endif /* MBS_SUPPORT */ |
2b83a2a4 | 5191 | #else |
770d454d | 5192 | # define FREE_VAR(var) if (var) free (var); var = NULL |
e4c785c8 UD |
5193 | # ifdef MBS_SUPPORT |
5194 | # define FREE_VARIABLES() \ | |
5195 | do { \ | |
770d454d UD |
5196 | FREE_VAR (string1); \ |
5197 | FREE_VAR (string2); \ | |
5198 | FREE_VAR (mbs_offset1); \ | |
5199 | FREE_VAR (mbs_offset2); \ | |
e4c785c8 | 5200 | } while (0) |
770d454d | 5201 | # else |
e4c785c8 UD |
5202 | # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ |
5203 | # endif /* MBS_SUPPORT */ | |
2b83a2a4 RM |
5204 | #endif /* not MATCH_MAY_ALLOCATE */ |
5205 | ||
5206 | /* These values must meet several constraints. They must not be valid | |
5207 | register values; since we have a limit of 255 registers (because | |
5208 | we use only one byte in the pattern for the register number), we can | |
5209 | use numbers larger than 255. They must differ by 1, because of | |
5210 | NUM_FAILURE_ITEMS above. And the value for the lowest register must | |
5211 | be larger than the value for the highest register, so we do not try | |
5212 | to actually save any registers when none are active. */ | |
5213 | #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) | |
5214 | #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
5215 | \f | |
5216 | /* Matching routines. */ | |
5217 | ||
5218 | #ifndef emacs /* Emacs never uses this. */ | |
5219 | /* re_match is like re_match_2 except it takes only a single string. */ | |
5220 | ||
5221 | int | |
5222 | re_match (bufp, string, size, pos, regs) | |
5223 | struct re_pattern_buffer *bufp; | |
5224 | const char *string; | |
5225 | int size, pos; | |
5226 | struct re_registers *regs; | |
5227 | { | |
5228 | int result = re_match_2_internal (bufp, NULL, 0, string, size, | |
5229 | pos, regs, size); | |
86187531 UD |
5230 | # ifndef REGEX_MALLOC |
5231 | # ifdef C_ALLOCA | |
2b83a2a4 | 5232 | alloca (0); |
86187531 UD |
5233 | # endif |
5234 | # endif | |
2b83a2a4 RM |
5235 | return result; |
5236 | } | |
2ad4fab2 UD |
5237 | # ifdef _LIBC |
5238 | weak_alias (__re_match, re_match) | |
5239 | # endif | |
2b83a2a4 RM |
5240 | #endif /* not emacs */ |
5241 | ||
e4c785c8 UD |
5242 | static boolean group_match_null_string_p _RE_ARGS ((US_CHAR_TYPE **p, |
5243 | US_CHAR_TYPE *end, | |
4cca6b86 | 5244 | register_info_type *reg_info)); |
e4c785c8 UD |
5245 | static boolean alt_match_null_string_p _RE_ARGS ((US_CHAR_TYPE *p, |
5246 | US_CHAR_TYPE *end, | |
4cca6b86 | 5247 | register_info_type *reg_info)); |
e4c785c8 UD |
5248 | static boolean common_op_match_null_string_p _RE_ARGS ((US_CHAR_TYPE **p, |
5249 | US_CHAR_TYPE *end, | |
4cca6b86 | 5250 | register_info_type *reg_info)); |
e4c785c8 | 5251 | static int bcmp_translate _RE_ARGS ((const CHAR_TYPE *s1, const CHAR_TYPE *s2, |
4cca6b86 | 5252 | int len, char *translate)); |
2b83a2a4 RM |
5253 | |
5254 | /* re_match_2 matches the compiled pattern in BUFP against the | |
5255 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 | |
5256 | and SIZE2, respectively). We start matching at POS, and stop | |
5257 | matching at STOP. | |
91c7b85d | 5258 | |
2b83a2a4 RM |
5259 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
5260 | store offsets for the substring each group matched in REGS. See the | |
5261 | documentation for exactly how many groups we fill. | |
5262 | ||
5263 | We return -1 if no match, -2 if an internal error (such as the | |
5264 | failure stack overflowing). Otherwise, we return the length of the | |
5265 | matched substring. */ | |
5266 | ||
5267 | int | |
5268 | re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) | |
5269 | struct re_pattern_buffer *bufp; | |
5270 | const char *string1, *string2; | |
5271 | int size1, size2; | |
5272 | int pos; | |
5273 | struct re_registers *regs; | |
5274 | int stop; | |
5275 | { | |
5276 | int result = re_match_2_internal (bufp, string1, size1, string2, size2, | |
5277 | pos, regs, stop); | |
4cca6b86 | 5278 | #ifndef REGEX_MALLOC |
86187531 | 5279 | # ifdef C_ALLOCA |
2b83a2a4 | 5280 | alloca (0); |
86187531 | 5281 | # endif |
4cca6b86 | 5282 | #endif |
2b83a2a4 RM |
5283 | return result; |
5284 | } | |
2ad4fab2 UD |
5285 | #ifdef _LIBC |
5286 | weak_alias (__re_match_2, re_match_2) | |
5287 | #endif | |
2b83a2a4 | 5288 | |
e4c785c8 | 5289 | #ifdef MBS_SUPPORT |
2d0aea11 UD |
5290 | |
5291 | static int count_mbs_length PARAMS ((int *, int)); | |
5292 | ||
e4c785c8 UD |
5293 | /* This check the substring (from 0, to length) of the multibyte string, |
5294 | to which offset_buffer correspond. And count how many wchar_t_characters | |
5295 | the substring occupy. We use offset_buffer to optimization. | |
5296 | See convert_mbs_to_wcs. */ | |
2d0aea11 | 5297 | |
e4c785c8 UD |
5298 | static int |
5299 | count_mbs_length(offset_buffer, length) | |
5300 | int *offset_buffer; | |
5301 | int length; | |
5302 | { | |
5303 | int wcs_size; | |
5304 | ||
5305 | /* Check whether the size is valid. */ | |
5306 | if (length < 0) | |
5307 | return -1; | |
5308 | ||
5309 | if (offset_buffer == NULL) | |
5310 | return 0; | |
5311 | ||
5312 | for (wcs_size = 0 ; offset_buffer[wcs_size] != -1 ; wcs_size++) | |
5313 | { | |
5314 | if (offset_buffer[wcs_size] == length) | |
5315 | return wcs_size; | |
5316 | if (offset_buffer[wcs_size] > length) | |
5317 | /* It is a fragment of a wide character. */ | |
5318 | return -1; | |
5319 | } | |
5320 | ||
5321 | /* We reached at the sentinel. */ | |
5322 | return -1; | |
5323 | } | |
5324 | #endif /* MBS_SUPPORT */ | |
5325 | ||
2b83a2a4 RM |
5326 | /* This is a separate function so that we can force an alloca cleanup |
5327 | afterwards. */ | |
5328 | static int | |
e4c785c8 UD |
5329 | #ifdef MBS_SUPPORT |
5330 | re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos, regs, stop) | |
5331 | struct re_pattern_buffer *bufp; | |
5332 | const char *cstring1, *cstring2; | |
5333 | int csize1, csize2; | |
5334 | #else | |
2b83a2a4 RM |
5335 | re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop) |
5336 | struct re_pattern_buffer *bufp; | |
5337 | const char *string1, *string2; | |
5338 | int size1, size2; | |
e4c785c8 | 5339 | #endif |
2b83a2a4 RM |
5340 | int pos; |
5341 | struct re_registers *regs; | |
5342 | int stop; | |
5343 | { | |
5344 | /* General temporaries. */ | |
5345 | int mcnt; | |
e4c785c8 UD |
5346 | US_CHAR_TYPE *p1; |
5347 | #ifdef MBS_SUPPORT | |
5348 | /* We need wchar_t* buffers correspond to string1, string2. */ | |
5349 | CHAR_TYPE *string1 = NULL, *string2 = NULL; | |
5350 | /* We need the size of wchar_t buffers correspond to csize1, csize2. */ | |
5351 | int size1 = 0, size2 = 0; | |
5352 | /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ | |
5353 | int *mbs_offset1 = NULL, *mbs_offset2 = NULL; | |
5354 | /* They hold whether each wchar_t is binary data or not. */ | |
770d454d | 5355 | char *is_binary = NULL; |
e4c785c8 | 5356 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
5357 | |
5358 | /* Just past the end of the corresponding string. */ | |
e4c785c8 | 5359 | const CHAR_TYPE *end1, *end2; |
2b83a2a4 RM |
5360 | |
5361 | /* Pointers into string1 and string2, just past the last characters in | |
5362 | each to consider matching. */ | |
e4c785c8 | 5363 | const CHAR_TYPE *end_match_1, *end_match_2; |
2b83a2a4 RM |
5364 | |
5365 | /* Where we are in the data, and the end of the current string. */ | |
e4c785c8 | 5366 | const CHAR_TYPE *d, *dend; |
91c7b85d | 5367 | |
2b83a2a4 | 5368 | /* Where we are in the pattern, and the end of the pattern. */ |
e4c785c8 UD |
5369 | #ifdef MBS_SUPPORT |
5370 | US_CHAR_TYPE *pattern, *p; | |
5371 | register US_CHAR_TYPE *pend; | |
5372 | #else | |
5373 | US_CHAR_TYPE *p = bufp->buffer; | |
5374 | register US_CHAR_TYPE *pend = p + bufp->used; | |
5375 | #endif /* MBS_SUPPORT */ | |
2b83a2a4 RM |
5376 | |
5377 | /* Mark the opcode just after a start_memory, so we can test for an | |
5378 | empty subpattern when we get to the stop_memory. */ | |
e4c785c8 | 5379 | US_CHAR_TYPE *just_past_start_mem = 0; |
2b83a2a4 RM |
5380 | |
5381 | /* We use this to map every character in the string. */ | |
03a75825 | 5382 | RE_TRANSLATE_TYPE translate = bufp->translate; |
2b83a2a4 RM |
5383 | |
5384 | /* Failure point stack. Each place that can handle a failure further | |
5385 | down the line pushes a failure point on this stack. It consists of | |
5386 | restart, regend, and reg_info for all registers corresponding to | |
5387 | the subexpressions we're currently inside, plus the number of such | |
5388 | registers, and, finally, two char *'s. The first char * is where | |
5389 | to resume scanning the pattern; the second one is where to resume | |
5390 | scanning the strings. If the latter is zero, the failure point is | |
5391 | a ``dummy''; if a failure happens and the failure point is a dummy, | |
5392 | it gets discarded and the next next one is tried. */ | |
5393 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
5394 | fail_stack_type fail_stack; | |
5395 | #endif | |
5396 | #ifdef DEBUG | |
c4563d2d | 5397 | static unsigned failure_id; |
2b83a2a4 RM |
5398 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
5399 | #endif | |
5400 | ||
4cca6b86 | 5401 | #ifdef REL_ALLOC |
2b83a2a4 RM |
5402 | /* This holds the pointer to the failure stack, when |
5403 | it is allocated relocatably. */ | |
5404 | fail_stack_elt_t *failure_stack_ptr; | |
4cca6b86 | 5405 | #endif |
2b83a2a4 RM |
5406 | |
5407 | /* We fill all the registers internally, independent of what we | |
5408 | return, for use in backreferences. The number here includes | |
5409 | an element for register zero. */ | |
4cca6b86 | 5410 | size_t num_regs = bufp->re_nsub + 1; |
91c7b85d | 5411 | |
2b83a2a4 | 5412 | /* The currently active registers. */ |
4cca6b86 UD |
5413 | active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
5414 | active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
2b83a2a4 RM |
5415 | |
5416 | /* Information on the contents of registers. These are pointers into | |
5417 | the input strings; they record just what was matched (on this | |
5418 | attempt) by a subexpression part of the pattern, that is, the | |
5419 | regnum-th regstart pointer points to where in the pattern we began | |
5420 | matching and the regnum-th regend points to right after where we | |
5421 | stopped matching the regnum-th subexpression. (The zeroth register | |
5422 | keeps track of what the whole pattern matches.) */ | |
5423 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
e4c785c8 | 5424 | const CHAR_TYPE **regstart, **regend; |
2b83a2a4 RM |
5425 | #endif |
5426 | ||
5427 | /* If a group that's operated upon by a repetition operator fails to | |
5428 | match anything, then the register for its start will need to be | |
5429 | restored because it will have been set to wherever in the string we | |
5430 | are when we last see its open-group operator. Similarly for a | |
5431 | register's end. */ | |
5432 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
e4c785c8 | 5433 | const CHAR_TYPE **old_regstart, **old_regend; |
2b83a2a4 RM |
5434 | #endif |
5435 | ||
5436 | /* The is_active field of reg_info helps us keep track of which (possibly | |
5437 | nested) subexpressions we are currently in. The matched_something | |
5438 | field of reg_info[reg_num] helps us tell whether or not we have | |
5439 | matched any of the pattern so far this time through the reg_num-th | |
5440 | subexpression. These two fields get reset each time through any | |
5441 | loop their register is in. */ | |
5442 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
91c7b85d | 5443 | register_info_type *reg_info; |
2b83a2a4 RM |
5444 | #endif |
5445 | ||
5446 | /* The following record the register info as found in the above | |
91c7b85d | 5447 | variables when we find a match better than any we've seen before. |
2b83a2a4 RM |
5448 | This happens as we backtrack through the failure points, which in |
5449 | turn happens only if we have not yet matched the entire string. */ | |
5450 | unsigned best_regs_set = false; | |
5451 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
e4c785c8 | 5452 | const CHAR_TYPE **best_regstart, **best_regend; |
2b83a2a4 | 5453 | #endif |
91c7b85d | 5454 | |
2b83a2a4 RM |
5455 | /* Logically, this is `best_regend[0]'. But we don't want to have to |
5456 | allocate space for that if we're not allocating space for anything | |
5457 | else (see below). Also, we never need info about register 0 for | |
5458 | any of the other register vectors, and it seems rather a kludge to | |
5459 | treat `best_regend' differently than the rest. So we keep track of | |
5460 | the end of the best match so far in a separate variable. We | |
5461 | initialize this to NULL so that when we backtrack the first time | |
5462 | and need to test it, it's not garbage. */ | |
e4c785c8 | 5463 | const CHAR_TYPE *match_end = NULL; |
2b83a2a4 RM |
5464 | |
5465 | /* This helps SET_REGS_MATCHED avoid doing redundant work. */ | |
5466 | int set_regs_matched_done = 0; | |
5467 | ||
5468 | /* Used when we pop values we don't care about. */ | |
5469 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
e4c785c8 | 5470 | const CHAR_TYPE **reg_dummy; |
2b83a2a4 RM |
5471 | register_info_type *reg_info_dummy; |
5472 | #endif | |
5473 | ||
5474 | #ifdef DEBUG | |
5475 | /* Counts the total number of registers pushed. */ | |
91c7b85d | 5476 | unsigned num_regs_pushed = 0; |
2b83a2a4 RM |
5477 | #endif |
5478 | ||
5479 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); | |
91c7b85d | 5480 | |
2b83a2a4 | 5481 | INIT_FAIL_STACK (); |
91c7b85d | 5482 | |
2b83a2a4 RM |
5483 | #ifdef MATCH_MAY_ALLOCATE |
5484 | /* Do not bother to initialize all the register variables if there are | |
5485 | no groups in the pattern, as it takes a fair amount of time. If | |
5486 | there are groups, we include space for register 0 (the whole | |
5487 | pattern), even though we never use it, since it simplifies the | |
5488 | array indexing. We should fix this. */ | |
5489 | if (bufp->re_nsub) | |
5490 | { | |
e4c785c8 UD |
5491 | regstart = REGEX_TALLOC (num_regs, const CHAR_TYPE *); |
5492 | regend = REGEX_TALLOC (num_regs, const CHAR_TYPE *); | |
5493 | old_regstart = REGEX_TALLOC (num_regs, const CHAR_TYPE *); | |
5494 | old_regend = REGEX_TALLOC (num_regs, const CHAR_TYPE *); | |
5495 | best_regstart = REGEX_TALLOC (num_regs, const CHAR_TYPE *); | |
5496 | best_regend = REGEX_TALLOC (num_regs, const CHAR_TYPE *); | |
2b83a2a4 | 5497 | reg_info = REGEX_TALLOC (num_regs, register_info_type); |
e4c785c8 | 5498 | reg_dummy = REGEX_TALLOC (num_regs, const CHAR_TYPE *); |
2b83a2a4 RM |
5499 | reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); |
5500 | ||
91c7b85d RM |
5501 | if (!(regstart && regend && old_regstart && old_regend && reg_info |
5502 | && best_regstart && best_regend && reg_dummy && reg_info_dummy)) | |
2b83a2a4 RM |
5503 | { |
5504 | FREE_VARIABLES (); | |
5505 | return -2; | |
5506 | } | |
5507 | } | |
5508 | else | |
5509 | { | |
5510 | /* We must initialize all our variables to NULL, so that | |
5511 | `FREE_VARIABLES' doesn't try to free them. */ | |
5512 | regstart = regend = old_regstart = old_regend = best_regstart | |
5513 | = best_regend = reg_dummy = NULL; | |
5514 | reg_info = reg_info_dummy = (register_info_type *) NULL; | |
5515 | } | |
5516 | #endif /* MATCH_MAY_ALLOCATE */ | |
5517 | ||
5518 | /* The starting position is bogus. */ | |
e4c785c8 UD |
5519 | #ifdef MBS_SUPPORT |
5520 | if (pos < 0 || pos > csize1 + csize2) | |
5521 | #else | |
2b83a2a4 | 5522 | if (pos < 0 || pos > size1 + size2) |
e4c785c8 | 5523 | #endif |
2b83a2a4 RM |
5524 | { |
5525 | FREE_VARIABLES (); | |
5526 | return -1; | |
5527 | } | |
91c7b85d | 5528 | |
e4c785c8 UD |
5529 | #ifdef MBS_SUPPORT |
5530 | /* Allocate wchar_t array for string1 and string2 and | |
5531 | fill them with converted string. */ | |
5532 | if (csize1 != 0) | |
5533 | { | |
770d454d UD |
5534 | string1 = REGEX_TALLOC (csize1 + 1, CHAR_TYPE); |
5535 | mbs_offset1 = REGEX_TALLOC (csize1 + 1, int); | |
5536 | is_binary = REGEX_TALLOC (csize1 + 1, char); | |
5537 | if (!string1 || !mbs_offset1 || !is_binary) | |
e4c785c8 | 5538 | { |
770d454d UD |
5539 | FREE_VAR (string1); |
5540 | FREE_VAR (mbs_offset1); | |
5541 | FREE_VAR (is_binary); | |
e4c785c8 UD |
5542 | return -2; |
5543 | } | |
5544 | size1 = convert_mbs_to_wcs(string1, cstring1, csize1, | |
770d454d | 5545 | mbs_offset1, is_binary); |
e4c785c8 | 5546 | string1[size1] = L'\0'; /* for a sentinel */ |
770d454d | 5547 | FREE_VAR (is_binary); |
e4c785c8 UD |
5548 | } |
5549 | if (csize2 != 0) | |
5550 | { | |
5551 | string2 = REGEX_TALLOC (csize2 + 1, CHAR_TYPE); | |
5552 | mbs_offset2 = REGEX_TALLOC (csize2 + 1, int); | |
770d454d UD |
5553 | is_binary = REGEX_TALLOC (csize2 + 1, char); |
5554 | if (!string2 || !mbs_offset2 || !is_binary) | |
e4c785c8 | 5555 | { |
770d454d UD |
5556 | FREE_VAR (string1); |
5557 | FREE_VAR (mbs_offset1); | |
5558 | FREE_VAR (string2); | |
5559 | FREE_VAR (mbs_offset2); | |
5560 | FREE_VAR (is_binary); | |
e4c785c8 UD |
5561 | return -2; |
5562 | } | |
5563 | size2 = convert_mbs_to_wcs(string2, cstring2, csize2, | |
770d454d | 5564 | mbs_offset2, is_binary); |
e4c785c8 | 5565 | string2[size2] = L'\0'; /* for a sentinel */ |
770d454d | 5566 | FREE_VAR (is_binary); |
e4c785c8 UD |
5567 | } |
5568 | ||
5569 | /* We need to cast pattern to (wchar_t*), because we casted this compiled | |
5570 | pattern to (char*) in regex_compile. */ | |
5571 | p = pattern = (CHAR_TYPE*)bufp->buffer; | |
5572 | pend = (CHAR_TYPE*)(bufp->buffer + bufp->used); | |
5573 | ||
5574 | #endif /* MBS_SUPPORT */ | |
5575 | ||
2b83a2a4 RM |
5576 | /* Initialize subexpression text positions to -1 to mark ones that no |
5577 | start_memory/stop_memory has been seen for. Also initialize the | |
5578 | register information struct. */ | |
cccda09f | 5579 | for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) |
2b83a2a4 | 5580 | { |
91c7b85d | 5581 | regstart[mcnt] = regend[mcnt] |
2b83a2a4 | 5582 | = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; |
91c7b85d | 5583 | |
2b83a2a4 RM |
5584 | REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; |
5585 | IS_ACTIVE (reg_info[mcnt]) = 0; | |
5586 | MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
5587 | EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
5588 | } | |
91c7b85d | 5589 | |
2b83a2a4 RM |
5590 | /* We move `string1' into `string2' if the latter's empty -- but not if |
5591 | `string1' is null. */ | |
5592 | if (size2 == 0 && string1 != NULL) | |
5593 | { | |
5594 | string2 = string1; | |
5595 | size2 = size1; | |
5596 | string1 = 0; | |
5597 | size1 = 0; | |
3eab00bd AJ |
5598 | #ifdef MBS_SUPPORT |
5599 | mbs_offset2 = mbs_offset1; | |
5600 | csize2 = csize1; | |
5601 | mbs_offset1 = NULL; | |
5602 | csize1 = 0; | |
5603 | #endif | |
2b83a2a4 RM |
5604 | } |
5605 | end1 = string1 + size1; | |
5606 | end2 = string2 + size2; | |
5607 | ||
5608 | /* Compute where to stop matching, within the two strings. */ | |
e4c785c8 UD |
5609 | #ifdef MBS_SUPPORT |
5610 | if (stop <= csize1) | |
5611 | { | |
5612 | mcnt = count_mbs_length(mbs_offset1, stop); | |
5613 | end_match_1 = string1 + mcnt; | |
5614 | end_match_2 = string2; | |
5615 | } | |
5616 | else | |
5617 | { | |
3eab00bd AJ |
5618 | if (stop > csize1 + csize2) |
5619 | stop = csize1 + csize2; | |
e4c785c8 UD |
5620 | end_match_1 = end1; |
5621 | mcnt = count_mbs_length(mbs_offset2, stop-csize1); | |
5622 | end_match_2 = string2 + mcnt; | |
5623 | } | |
5624 | if (mcnt < 0) | |
5625 | { /* count_mbs_length return error. */ | |
5626 | FREE_VARIABLES (); | |
5627 | return -1; | |
5628 | } | |
5629 | #else | |
2b83a2a4 RM |
5630 | if (stop <= size1) |
5631 | { | |
5632 | end_match_1 = string1 + stop; | |
5633 | end_match_2 = string2; | |
5634 | } | |
5635 | else | |
5636 | { | |
5637 | end_match_1 = end1; | |
5638 | end_match_2 = string2 + stop - size1; | |
5639 | } | |
e4c785c8 | 5640 | #endif /* MBS_SUPPORT */ |
2b83a2a4 | 5641 | |
91c7b85d | 5642 | /* `p' scans through the pattern as `d' scans through the data. |
2b83a2a4 RM |
5643 | `dend' is the end of the input string that `d' points within. `d' |
5644 | is advanced into the following input string whenever necessary, but | |
5645 | this happens before fetching; therefore, at the beginning of the | |
5646 | loop, `d' can be pointing at the end of a string, but it cannot | |
5647 | equal `string2'. */ | |
e4c785c8 UD |
5648 | #ifdef MBS_SUPPORT |
5649 | if (size1 > 0 && pos <= csize1) | |
5650 | { | |
5651 | mcnt = count_mbs_length(mbs_offset1, pos); | |
5652 | d = string1 + mcnt; | |
5653 | dend = end_match_1; | |
5654 | } | |
5655 | else | |
5656 | { | |
5657 | mcnt = count_mbs_length(mbs_offset2, pos-csize1); | |
5658 | d = string2 + mcnt; | |
5659 | dend = end_match_2; | |
5660 | } | |
5661 | ||
5662 | if (mcnt < 0) | |
5663 | { /* count_mbs_length return error. */ | |
5664 | FREE_VARIABLES (); | |
5665 | return -1; | |
5666 | } | |
5667 | #else | |
2b83a2a4 RM |
5668 | if (size1 > 0 && pos <= size1) |
5669 | { | |
5670 | d = string1 + pos; | |
5671 | dend = end_match_1; | |
5672 | } | |
5673 | else | |
5674 | { | |
5675 | d = string2 + pos - size1; | |
5676 | dend = end_match_2; | |
5677 | } | |
e4c785c8 | 5678 | #endif /* MBS_SUPPORT */ |
2b83a2a4 | 5679 | |
5929563f | 5680 | DEBUG_PRINT1 ("The compiled pattern is:\n"); |
2b83a2a4 RM |
5681 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); |
5682 | DEBUG_PRINT1 ("The string to match is: `"); | |
5683 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
5684 | DEBUG_PRINT1 ("'\n"); | |
91c7b85d | 5685 | |
2b83a2a4 RM |
5686 | /* This loops over pattern commands. It exits by returning from the |
5687 | function if the match is complete, or it drops through if the match | |
5688 | fails at this starting point in the input data. */ | |
5689 | for (;;) | |
5690 | { | |
5929563f UD |
5691 | #ifdef _LIBC |
5692 | DEBUG_PRINT2 ("\n%p: ", p); | |
5693 | #else | |
2b83a2a4 | 5694 | DEBUG_PRINT2 ("\n0x%x: ", p); |
5929563f | 5695 | #endif |
2b83a2a4 RM |
5696 | |
5697 | if (p == pend) | |
5698 | { /* End of pattern means we might have succeeded. */ | |
5699 | DEBUG_PRINT1 ("end of pattern ... "); | |
91c7b85d | 5700 | |
2b83a2a4 RM |
5701 | /* If we haven't matched the entire string, and we want the |
5702 | longest match, try backtracking. */ | |
5703 | if (d != end_match_2) | |
5704 | { | |
5705 | /* 1 if this match ends in the same string (string1 or string2) | |
5706 | as the best previous match. */ | |
91c7b85d | 5707 | boolean same_str_p = (FIRST_STRING_P (match_end) |
2b83a2a4 RM |
5708 | == MATCHING_IN_FIRST_STRING); |
5709 | /* 1 if this match is the best seen so far. */ | |
5710 | boolean best_match_p; | |
5711 | ||
5712 | /* AIX compiler got confused when this was combined | |
5713 | with the previous declaration. */ | |
5714 | if (same_str_p) | |
5715 | best_match_p = d > match_end; | |
5716 | else | |
5717 | best_match_p = !MATCHING_IN_FIRST_STRING; | |
5718 | ||
5719 | DEBUG_PRINT1 ("backtracking.\n"); | |
91c7b85d | 5720 | |
2b83a2a4 RM |
5721 | if (!FAIL_STACK_EMPTY ()) |
5722 | { /* More failure points to try. */ | |
5723 | ||
5724 | /* If exceeds best match so far, save it. */ | |
5725 | if (!best_regs_set || best_match_p) | |
5726 | { | |
5727 | best_regs_set = true; | |
5728 | match_end = d; | |
91c7b85d | 5729 | |
2b83a2a4 | 5730 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); |
91c7b85d | 5731 | |
cccda09f | 5732 | for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) |
2b83a2a4 RM |
5733 | { |
5734 | best_regstart[mcnt] = regstart[mcnt]; | |
5735 | best_regend[mcnt] = regend[mcnt]; | |
5736 | } | |
5737 | } | |
91c7b85d | 5738 | goto fail; |
2b83a2a4 RM |
5739 | } |
5740 | ||
5741 | /* If no failure points, don't restore garbage. And if | |
5742 | last match is real best match, don't restore second | |
5743 | best one. */ | |
5744 | else if (best_regs_set && !best_match_p) | |
5745 | { | |
5746 | restore_best_regs: | |
5747 | /* Restore best match. It may happen that `dend == | |
5748 | end_match_1' while the restored d is in string2. | |
5749 | For example, the pattern `x.*y.*z' against the | |
5750 | strings `x-' and `y-z-', if the two strings are | |
5751 | not consecutive in memory. */ | |
5752 | DEBUG_PRINT1 ("Restoring best registers.\n"); | |
91c7b85d | 5753 | |
2b83a2a4 RM |
5754 | d = match_end; |
5755 | dend = ((d >= string1 && d <= end1) | |
5756 | ? end_match_1 : end_match_2); | |
5757 | ||
cccda09f | 5758 | for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) |
2b83a2a4 RM |
5759 | { |
5760 | regstart[mcnt] = best_regstart[mcnt]; | |
5761 | regend[mcnt] = best_regend[mcnt]; | |
5762 | } | |
5763 | } | |
5764 | } /* d != end_match_2 */ | |
5765 | ||
5766 | succeed_label: | |
5767 | DEBUG_PRINT1 ("Accepting match.\n"); | |
2b83a2a4 RM |
5768 | /* If caller wants register contents data back, do it. */ |
5769 | if (regs && !bufp->no_sub) | |
5770 | { | |
e4c785c8 | 5771 | /* Have the register data arrays been allocated? */ |
2b83a2a4 RM |
5772 | if (bufp->regs_allocated == REGS_UNALLOCATED) |
5773 | { /* No. So allocate them with malloc. We need one | |
5774 | extra element beyond `num_regs' for the `-1' marker | |
5775 | GNU code uses. */ | |
5776 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); | |
5777 | regs->start = TALLOC (regs->num_regs, regoff_t); | |
5778 | regs->end = TALLOC (regs->num_regs, regoff_t); | |
5779 | if (regs->start == NULL || regs->end == NULL) | |
5780 | { | |
5781 | FREE_VARIABLES (); | |
5782 | return -2; | |
5783 | } | |
5784 | bufp->regs_allocated = REGS_REALLOCATE; | |
5785 | } | |
5786 | else if (bufp->regs_allocated == REGS_REALLOCATE) | |
5787 | { /* Yes. If we need more elements than were already | |
5788 | allocated, reallocate them. If we need fewer, just | |
5789 | leave it alone. */ | |
5790 | if (regs->num_regs < num_regs + 1) | |
5791 | { | |
5792 | regs->num_regs = num_regs + 1; | |
5793 | RETALLOC (regs->start, regs->num_regs, regoff_t); | |
5794 | RETALLOC (regs->end, regs->num_regs, regoff_t); | |
5795 | if (regs->start == NULL || regs->end == NULL) | |
5796 | { | |
5797 | FREE_VARIABLES (); | |
5798 | return -2; | |
5799 | } | |
5800 | } | |
5801 | } | |
5802 | else | |
5803 | { | |
5804 | /* These braces fend off a "empty body in an else-statement" | |
5805 | warning under GCC when assert expands to nothing. */ | |
5806 | assert (bufp->regs_allocated == REGS_FIXED); | |
5807 | } | |
5808 | ||
5809 | /* Convert the pointer data in `regstart' and `regend' to | |
5810 | indices. Register zero has to be set differently, | |
5811 | since we haven't kept track of any info for it. */ | |
5812 | if (regs->num_regs > 0) | |
5813 | { | |
5814 | regs->start[0] = pos; | |
e4c785c8 UD |
5815 | #ifdef MBS_SUPPORT |
5816 | if (MATCHING_IN_FIRST_STRING) | |
5817 | regs->end[0] = mbs_offset1 != NULL ? | |
5818 | mbs_offset1[d-string1] : 0; | |
5819 | else | |
5820 | regs->end[0] = csize1 + (mbs_offset2 != NULL ? | |
5821 | mbs_offset2[d-string2] : 0); | |
5822 | #else | |
2b83a2a4 RM |
5823 | regs->end[0] = (MATCHING_IN_FIRST_STRING |
5824 | ? ((regoff_t) (d - string1)) | |
5825 | : ((regoff_t) (d - string2 + size1))); | |
e4c785c8 | 5826 | #endif /* MBS_SUPPORT */ |
2b83a2a4 | 5827 | } |
91c7b85d | 5828 | |
2b83a2a4 RM |
5829 | /* Go through the first `min (num_regs, regs->num_regs)' |
5830 | registers, since that is all we initialized. */ | |
cccda09f UD |
5831 | for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs); |
5832 | mcnt++) | |
2b83a2a4 RM |
5833 | { |
5834 | if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) | |
5835 | regs->start[mcnt] = regs->end[mcnt] = -1; | |
5836 | else | |
5837 | { | |
5838 | regs->start[mcnt] | |
5839 | = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); | |
5840 | regs->end[mcnt] | |
5841 | = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); | |
5842 | } | |
5843 | } | |
91c7b85d | 5844 | |
2b83a2a4 RM |
5845 | /* If the regs structure we return has more elements than |
5846 | were in the pattern, set the extra elements to -1. If | |
5847 | we (re)allocated the registers, this is the case, | |
5848 | because we always allocate enough to have at least one | |
5849 | -1 at the end. */ | |
cccda09f | 5850 | for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++) |
2b83a2a4 RM |
5851 | regs->start[mcnt] = regs->end[mcnt] = -1; |
5852 | } /* regs && !bufp->no_sub */ | |
5853 | ||
5854 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", | |
5855 | nfailure_points_pushed, nfailure_points_popped, | |
5856 | nfailure_points_pushed - nfailure_points_popped); | |
5857 | DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); | |
5858 | ||
e4c785c8 UD |
5859 | #ifdef MBS_SUPPORT |
5860 | if (MATCHING_IN_FIRST_STRING) | |
5861 | mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0; | |
5862 | else | |
5863 | mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) + | |
5864 | csize1; | |
5865 | mcnt -= pos; | |
5866 | #else | |
91c7b85d RM |
5867 | mcnt = d - pos - (MATCHING_IN_FIRST_STRING |
5868 | ? string1 | |
2b83a2a4 | 5869 | : string2 - size1); |
e4c785c8 | 5870 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
5871 | |
5872 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); | |
5873 | ||
5874 | FREE_VARIABLES (); | |
5875 | return mcnt; | |
5876 | } | |
5877 | ||
5878 | /* Otherwise match next pattern command. */ | |
5879 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) | |
5880 | { | |
5881 | /* Ignore these. Used to ignore the n of succeed_n's which | |
5882 | currently have n == 0. */ | |
5883 | case no_op: | |
5884 | DEBUG_PRINT1 ("EXECUTING no_op.\n"); | |
5885 | break; | |
5886 | ||
5887 | case succeed: | |
5888 | DEBUG_PRINT1 ("EXECUTING succeed.\n"); | |
5889 | goto succeed_label; | |
5890 | ||
5891 | /* Match the next n pattern characters exactly. The following | |
5892 | byte in the pattern defines n, and the n bytes after that | |
5893 | are the characters to match. */ | |
5894 | case exactn: | |
e4c785c8 UD |
5895 | #ifdef MBS_SUPPORT |
5896 | case exactn_bin: | |
5897 | #endif | |
2b83a2a4 RM |
5898 | mcnt = *p++; |
5899 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); | |
5900 | ||
5901 | /* This is written out as an if-else so we don't waste time | |
5902 | testing `translate' inside the loop. */ | |
5903 | if (translate) | |
5904 | { | |
5905 | do | |
5906 | { | |
5907 | PREFETCH (); | |
e4c785c8 UD |
5908 | #ifdef MBS_SUPPORT |
5909 | if (*d <= 0xff) | |
5910 | { | |
5911 | if ((US_CHAR_TYPE) translate[(unsigned char) *d++] | |
5912 | != (US_CHAR_TYPE) *p++) | |
5913 | goto fail; | |
5914 | } | |
5915 | else | |
5916 | { | |
5917 | if (*d++ != (CHAR_TYPE) *p++) | |
5918 | goto fail; | |
5919 | } | |
5920 | #else | |
5921 | if ((US_CHAR_TYPE) translate[(unsigned char) *d++] | |
5922 | != (US_CHAR_TYPE) *p++) | |
2b83a2a4 | 5923 | goto fail; |
e4c785c8 | 5924 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
5925 | } |
5926 | while (--mcnt); | |
5927 | } | |
5928 | else | |
5929 | { | |
5930 | do | |
5931 | { | |
5932 | PREFETCH (); | |
e4c785c8 | 5933 | if (*d++ != (CHAR_TYPE) *p++) goto fail; |
2b83a2a4 RM |
5934 | } |
5935 | while (--mcnt); | |
5936 | } | |
5937 | SET_REGS_MATCHED (); | |
5938 | break; | |
5939 | ||
5940 | ||
5941 | /* Match any character except possibly a newline or a null. */ | |
5942 | case anychar: | |
5943 | DEBUG_PRINT1 ("EXECUTING anychar.\n"); | |
5944 | ||
5945 | PREFETCH (); | |
5946 | ||
5947 | if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') | |
5948 | || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) | |
5949 | goto fail; | |
5950 | ||
5951 | SET_REGS_MATCHED (); | |
672fd41b | 5952 | DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d); |
2b83a2a4 RM |
5953 | d++; |
5954 | break; | |
5955 | ||
5956 | ||
5957 | case charset: | |
5958 | case charset_not: | |
5959 | { | |
e4c785c8 UD |
5960 | register US_CHAR_TYPE c; |
5961 | #ifdef MBS_SUPPORT | |
5962 | unsigned int i, char_class_length, coll_symbol_length, | |
5963 | equiv_class_length, ranges_length, chars_length, length; | |
5964 | CHAR_TYPE *workp, *workp2, *charset_top; | |
5965 | #define WORK_BUFFER_SIZE 128 | |
5966 | CHAR_TYPE str_buf[WORK_BUFFER_SIZE]; | |
5967 | # ifdef _LIBC | |
5968 | uint32_t nrules; | |
5969 | # endif /* _LIBC */ | |
5970 | #endif /* MBS_SUPPORT */ | |
2b83a2a4 RM |
5971 | boolean not = (re_opcode_t) *(p - 1) == charset_not; |
5972 | ||
5973 | DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); | |
2b83a2a4 RM |
5974 | PREFETCH (); |
5975 | c = TRANSLATE (*d); /* The character to match. */ | |
e4c785c8 UD |
5976 | #ifdef MBS_SUPPORT |
5977 | # ifdef _LIBC | |
5978 | nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); | |
5979 | # endif /* _LIBC */ | |
5980 | charset_top = p - 1; | |
5981 | char_class_length = *p++; | |
5982 | coll_symbol_length = *p++; | |
5983 | equiv_class_length = *p++; | |
5984 | ranges_length = *p++; | |
5985 | chars_length = *p++; | |
5986 | /* p points charset[6], so the address of the next instruction | |
5987 | (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'], | |
5988 | where l=length of char_classes, m=length of collating_symbol, | |
5989 | n=equivalence_class, o=length of char_range, | |
5990 | p'=length of character. */ | |
5991 | workp = p; | |
5992 | /* Update p to indicate the next instruction. */ | |
5993 | p += char_class_length + coll_symbol_length+ equiv_class_length + | |
5994 | 2*ranges_length + chars_length; | |
5995 | ||
5996 | /* match with char_class? */ | |
054d2bf7 UD |
5997 | for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE) |
5998 | { | |
441f7d1e UD |
5999 | wctype_t wctype; |
6000 | uintptr_t alignedp = ((uintptr_t)workp | |
6001 | + __alignof__(wctype_t) - 1) | |
6002 | & ~(uintptr_t)(__alignof__(wctype_t) - 1); | |
6003 | wctype = *((wctype_t*)alignedp); | |
054d2bf7 UD |
6004 | workp += CHAR_CLASS_SIZE; |
6005 | if (iswctype((wint_t)c, wctype)) | |
6006 | goto char_set_matched; | |
6007 | } | |
e4c785c8 UD |
6008 | |
6009 | /* match with collating_symbol? */ | |
6010 | # ifdef _LIBC | |
6011 | if (nrules != 0) | |
6012 | { | |
054d2bf7 UD |
6013 | const unsigned char *extra = (const unsigned char *) |
6014 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); | |
6015 | ||
e4c785c8 UD |
6016 | for (workp2 = workp + coll_symbol_length ; workp < workp2 ; |
6017 | workp++) | |
6018 | { | |
6019 | int32_t *wextra; | |
054d2bf7 | 6020 | wextra = (int32_t*)(extra + *workp++); |
e4c785c8 UD |
6021 | for (i = 0; i < *wextra; ++i) |
6022 | if (TRANSLATE(d[i]) != wextra[1 + i]) | |
6023 | break; | |
6024 | ||
6025 | if (i == *wextra) | |
6026 | { | |
6027 | /* Update d, however d will be incremented at | |
6028 | char_set_matched:, we decrement d here. */ | |
6029 | d += i - 1; | |
6030 | goto char_set_matched; | |
6031 | } | |
6032 | } | |
6033 | } | |
6034 | else /* (nrules == 0) */ | |
6035 | # endif | |
6036 | /* If we can't look up collation data, we use wcscoll | |
6037 | instead. */ | |
6038 | { | |
6039 | for (workp2 = workp + coll_symbol_length ; workp < workp2 ;) | |
6040 | { | |
6041 | const CHAR_TYPE *backup_d = d, *backup_dend = dend; | |
6042 | length = wcslen(workp); | |
6043 | ||
6044 | /* If wcscoll(the collating symbol, whole string) > 0, | |
6045 | any substring of the string never match with the | |
6046 | collating symbol. */ | |
6047 | if (wcscoll(workp, d) > 0) | |
6048 | { | |
6049 | workp += length + 1; | |
6050 | continue; | |
6051 | } | |
6052 | ||
6053 | /* First, we compare the collating symbol with | |
6054 | the first character of the string. | |
6055 | If it don't match, we add the next character to | |
6056 | the compare buffer in turn. */ | |
6057 | for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++) | |
6058 | { | |
6059 | int match; | |
6060 | if (d == dend) | |
6061 | { | |
6062 | if (dend == end_match_2) | |
6063 | break; | |
6064 | d = string2; | |
6065 | dend = end_match_2; | |
6066 | } | |
6067 | ||
6068 | /* add next character to the compare buffer. */ | |
6069 | str_buf[i] = TRANSLATE(*d); | |
6070 | str_buf[i+1] = '\0'; | |
6071 | ||
6072 | match = wcscoll(workp, str_buf); | |
6073 | if (match == 0) | |
6074 | goto char_set_matched; | |
6075 | ||
6076 | if (match < 0) | |
6077 | /* (str_buf > workp) indicate (str_buf + X > workp), | |
6078 | because for all X (str_buf + X > str_buf). | |
6079 | So we don't need continue this loop. */ | |
6080 | break; | |
6081 | ||
6082 | /* Otherwise(str_buf < workp), | |
6083 | (str_buf+next_character) may equals (workp). | |
6084 | So we continue this loop. */ | |
6085 | } | |
6086 | /* not matched */ | |
6087 | d = backup_d; | |
6088 | dend = backup_dend; | |
6089 | workp += length + 1; | |
6090 | } | |
6091 | } | |
6092 | /* match with equivalence_class? */ | |
6093 | # ifdef _LIBC | |
6094 | if (nrules != 0) | |
6095 | { | |
6096 | const CHAR_TYPE *backup_d = d, *backup_dend = dend; | |
6097 | /* Try to match the equivalence class against | |
6098 | those known to the collate implementation. */ | |
6099 | const int32_t *table; | |
6100 | const int32_t *weights; | |
6101 | const int32_t *extra; | |
6102 | const int32_t *indirect; | |
6103 | int32_t idx, idx2; | |
6104 | wint_t *cp; | |
6105 | size_t len; | |
6106 | ||
6107 | /* This #include defines a local function! */ | |
6108 | # include <locale/weightwc.h> | |
6109 | ||
6110 | table = (const int32_t *) | |
6111 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC); | |
6112 | weights = (const wint_t *) | |
6113 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC); | |
6114 | extra = (const wint_t *) | |
6115 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC); | |
6116 | indirect = (const int32_t *) | |
6117 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC); | |
6118 | ||
6119 | /* Write 1 collating element to str_buf, and | |
6120 | get its index. */ | |
6121 | idx2 = 0; | |
6122 | ||
6123 | for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++) | |
6124 | { | |
6125 | cp = (wint_t*)str_buf; | |
6126 | if (d == dend) | |
6127 | { | |
6128 | if (dend == end_match_2) | |
6129 | break; | |
6130 | d = string2; | |
6131 | dend = end_match_2; | |
6132 | } | |
6133 | str_buf[i] = TRANSLATE(*(d+i)); | |
6134 | str_buf[i+1] = '\0'; /* sentinel */ | |
6135 | idx2 = findidx ((const wint_t**)&cp); | |
6136 | } | |
6137 | ||
6138 | /* Update d, however d will be incremented at | |
6139 | char_set_matched:, we decrement d here. */ | |
054d2bf7 | 6140 | d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1); |
e4c785c8 UD |
6141 | if (d >= dend) |
6142 | { | |
6143 | if (dend == end_match_2) | |
6144 | d = dend; | |
6145 | else | |
6146 | { | |
6147 | d = string2; | |
6148 | dend = end_match_2; | |
6149 | } | |
6150 | } | |
6151 | ||
6152 | len = weights[idx2]; | |
6153 | ||
6154 | for (workp2 = workp + equiv_class_length ; workp < workp2 ; | |
6155 | workp++) | |
6156 | { | |
6157 | idx = (int32_t)*workp; | |
6158 | /* We already checked idx != 0 in regex_compile. */ | |
6159 | ||
6160 | if (idx2 != 0 && len == weights[idx]) | |
6161 | { | |
6162 | int cnt = 0; | |
6163 | while (cnt < len && (weights[idx + 1 + cnt] | |
6164 | == weights[idx2 + 1 + cnt])) | |
6165 | ++cnt; | |
6166 | ||
6167 | if (cnt == len) | |
6168 | goto char_set_matched; | |
6169 | } | |
6170 | } | |
6171 | /* not matched */ | |
6172 | d = backup_d; | |
6173 | dend = backup_dend; | |
6174 | } | |
6175 | else /* (nrules == 0) */ | |
6176 | # endif | |
6177 | /* If we can't look up collation data, we use wcscoll | |
6178 | instead. */ | |
6179 | { | |
6180 | for (workp2 = workp + equiv_class_length ; workp < workp2 ;) | |
6181 | { | |
6182 | const CHAR_TYPE *backup_d = d, *backup_dend = dend; | |
6183 | length = wcslen(workp); | |
6184 | ||
6185 | /* If wcscoll(the collating symbol, whole string) > 0, | |
6186 | any substring of the string never match with the | |
6187 | collating symbol. */ | |
6188 | if (wcscoll(workp, d) > 0) | |
6189 | { | |
6190 | workp += length + 1; | |
6191 | break; | |
6192 | } | |
6193 | ||
6194 | /* First, we compare the equivalence class with | |
6195 | the first character of the string. | |
6196 | If it don't match, we add the next character to | |
6197 | the compare buffer in turn. */ | |
6198 | for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++) | |
6199 | { | |
6200 | int match; | |
6201 | if (d == dend) | |
6202 | { | |
6203 | if (dend == end_match_2) | |
6204 | break; | |
6205 | d = string2; | |
6206 | dend = end_match_2; | |
6207 | } | |
6208 | ||
6209 | /* add next character to the compare buffer. */ | |
6210 | str_buf[i] = TRANSLATE(*d); | |
6211 | str_buf[i+1] = '\0'; | |
6212 | ||
6213 | match = wcscoll(workp, str_buf); | |
6214 | ||
6215 | if (match == 0) | |
6216 | goto char_set_matched; | |
6217 | ||
6218 | if (match < 0) | |
6219 | /* (str_buf > workp) indicate (str_buf + X > workp), | |
6220 | because for all X (str_buf + X > str_buf). | |
6221 | So we don't need continue this loop. */ | |
6222 | break; | |
6223 | ||
6224 | /* Otherwise(str_buf < workp), | |
6225 | (str_buf+next_character) may equals (workp). | |
6226 | So we continue this loop. */ | |
6227 | } | |
6228 | /* not matched */ | |
6229 | d = backup_d; | |
6230 | dend = backup_dend; | |
6231 | workp += length + 1; | |
6232 | } | |
6233 | } | |
6234 | ||
6235 | /* match with char_range? */ | |
6236 | #ifdef _LIBC | |
6237 | if (nrules != 0) | |
6238 | { | |
6239 | uint32_t collseqval; | |
6240 | const char *collseq = (const char *) | |
6241 | _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC); | |
6242 | ||
6243 | collseqval = collseq_table_lookup (collseq, c); | |
6244 | ||
6245 | for (; workp < p - chars_length ;) | |
6246 | { | |
6247 | uint32_t start_val, end_val; | |
6248 | ||
6249 | /* We already compute the collation sequence value | |
6250 | of the characters (or collating symbols). */ | |
6251 | start_val = (uint32_t) *workp++; /* range_start */ | |
6252 | end_val = (uint32_t) *workp++; /* range_end */ | |
6253 | ||
6254 | if (start_val <= collseqval && collseqval <= end_val) | |
6255 | goto char_set_matched; | |
6256 | } | |
6257 | } | |
6258 | else | |
6259 | #endif | |
6260 | { | |
6261 | /* We set range_start_char at str_buf[0], range_end_char | |
6262 | at str_buf[4], and compared char at str_buf[2]. */ | |
6263 | str_buf[1] = 0; | |
6264 | str_buf[2] = c; | |
6265 | str_buf[3] = 0; | |
6266 | str_buf[5] = 0; | |
6267 | for (; workp < p - chars_length ;) | |
6268 | { | |
6269 | wchar_t *range_start_char, *range_end_char; | |
6270 | ||
6271 | /* match if (range_start_char <= c <= range_end_char). */ | |
6272 | ||
6273 | /* If range_start(or end) < 0, we assume -range_start(end) | |
6274 | is the offset of the collating symbol which is specified | |
6275 | as the character of the range start(end). */ | |
6276 | ||
6277 | /* range_start */ | |
6278 | if (*workp < 0) | |
6279 | range_start_char = charset_top - (*workp++); | |
6280 | else | |
6281 | { | |
6282 | str_buf[0] = *workp++; | |
6283 | range_start_char = str_buf; | |
6284 | } | |
6285 | ||
6286 | /* range_end */ | |
6287 | if (*workp < 0) | |
6288 | range_end_char = charset_top - (*workp++); | |
6289 | else | |
6290 | { | |
6291 | str_buf[4] = *workp++; | |
6292 | range_end_char = str_buf + 4; | |
6293 | } | |
2b83a2a4 | 6294 | |
e4c785c8 UD |
6295 | if (wcscoll(range_start_char, str_buf+2) <= 0 && |
6296 | wcscoll(str_buf+2, range_end_char) <= 0) | |
6297 | ||
6298 | goto char_set_matched; | |
6299 | } | |
6300 | } | |
6301 | ||
6302 | /* match with char? */ | |
6303 | for (; workp < p ; workp++) | |
6304 | if (c == *workp) | |
6305 | goto char_set_matched; | |
6306 | ||
6307 | not = !not; | |
6308 | ||
6309 | char_set_matched: | |
6310 | if (not) goto fail; | |
6311 | #else | |
2b83a2a4 RM |
6312 | /* Cast to `unsigned' instead of `unsigned char' in case the |
6313 | bit list is a full 32 bytes long. */ | |
6314 | if (c < (unsigned) (*p * BYTEWIDTH) | |
6315 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
6316 | not = !not; | |
6317 | ||
6318 | p += 1 + *p; | |
6319 | ||
6320 | if (!not) goto fail; | |
e4c785c8 UD |
6321 | #undef WORK_BUFFER_SIZE |
6322 | #endif /* MBS_SUPPORT */ | |
2b83a2a4 RM |
6323 | SET_REGS_MATCHED (); |
6324 | d++; | |
6325 | break; | |
6326 | } | |
6327 | ||
6328 | ||
6329 | /* The beginning of a group is represented by start_memory. | |
6330 | The arguments are the register number in the next byte, and the | |
6331 | number of groups inner to this one in the next. The text | |
6332 | matched within the group is recorded (in the internal | |
6333 | registers data structure) under the register number. */ | |
6334 | case start_memory: | |
672fd41b UD |
6335 | DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n", |
6336 | (long int) *p, (long int) p[1]); | |
2b83a2a4 RM |
6337 | |
6338 | /* Find out if this group can match the empty string. */ | |
6339 | p1 = p; /* To send to group_match_null_string_p. */ | |
91c7b85d | 6340 | |
2b83a2a4 | 6341 | if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) |
91c7b85d | 6342 | REG_MATCH_NULL_STRING_P (reg_info[*p]) |
2b83a2a4 RM |
6343 | = group_match_null_string_p (&p1, pend, reg_info); |
6344 | ||
6345 | /* Save the position in the string where we were the last time | |
6346 | we were at this open-group operator in case the group is | |
6347 | operated upon by a repetition operator, e.g., with `(a*)*b' | |
6348 | against `ab'; then we want to ignore where we are now in | |
6349 | the string in case this attempt to match fails. */ | |
6350 | old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
6351 | ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
6352 | : regstart[*p]; | |
91c7b85d | 6353 | DEBUG_PRINT2 (" old_regstart: %d\n", |
2b83a2a4 RM |
6354 | POINTER_TO_OFFSET (old_regstart[*p])); |
6355 | ||
6356 | regstart[*p] = d; | |
6357 | DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); | |
6358 | ||
6359 | IS_ACTIVE (reg_info[*p]) = 1; | |
6360 | MATCHED_SOMETHING (reg_info[*p]) = 0; | |
6361 | ||
6362 | /* Clear this whenever we change the register activity status. */ | |
6363 | set_regs_matched_done = 0; | |
91c7b85d | 6364 | |
2b83a2a4 RM |
6365 | /* This is the new highest active register. */ |
6366 | highest_active_reg = *p; | |
91c7b85d | 6367 | |
2b83a2a4 RM |
6368 | /* If nothing was active before, this is the new lowest active |
6369 | register. */ | |
6370 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
6371 | lowest_active_reg = *p; | |
6372 | ||
6373 | /* Move past the register number and inner group count. */ | |
6374 | p += 2; | |
6375 | just_past_start_mem = p; | |
6376 | ||
6377 | break; | |
6378 | ||
6379 | ||
6380 | /* The stop_memory opcode represents the end of a group. Its | |
6381 | arguments are the same as start_memory's: the register | |
6382 | number, and the number of inner groups. */ | |
6383 | case stop_memory: | |
672fd41b UD |
6384 | DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n", |
6385 | (long int) *p, (long int) p[1]); | |
91c7b85d | 6386 | |
2b83a2a4 RM |
6387 | /* We need to save the string position the last time we were at |
6388 | this close-group operator in case the group is operated | |
6389 | upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
6390 | against `aba'; then we want to ignore where we are now in | |
6391 | the string in case this attempt to match fails. */ | |
6392 | old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
6393 | ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
6394 | : regend[*p]; | |
91c7b85d | 6395 | DEBUG_PRINT2 (" old_regend: %d\n", |
2b83a2a4 RM |
6396 | POINTER_TO_OFFSET (old_regend[*p])); |
6397 | ||
6398 | regend[*p] = d; | |
6399 | DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); | |
6400 | ||
6401 | /* This register isn't active anymore. */ | |
6402 | IS_ACTIVE (reg_info[*p]) = 0; | |
6403 | ||
6404 | /* Clear this whenever we change the register activity status. */ | |
6405 | set_regs_matched_done = 0; | |
6406 | ||
6407 | /* If this was the only register active, nothing is active | |
6408 | anymore. */ | |
6409 | if (lowest_active_reg == highest_active_reg) | |
6410 | { | |
6411 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
6412 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
6413 | } | |
6414 | else | |
6415 | { /* We must scan for the new highest active register, since | |
6416 | it isn't necessarily one less than now: consider | |
6417 | (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
6418 | new highest active register is 1. */ | |
e4c785c8 | 6419 | US_CHAR_TYPE r = *p - 1; |
2b83a2a4 RM |
6420 | while (r > 0 && !IS_ACTIVE (reg_info[r])) |
6421 | r--; | |
91c7b85d | 6422 | |
2b83a2a4 RM |
6423 | /* If we end up at register zero, that means that we saved |
6424 | the registers as the result of an `on_failure_jump', not | |
6425 | a `start_memory', and we jumped to past the innermost | |
6426 | `stop_memory'. For example, in ((.)*) we save | |
6427 | registers 1 and 2 as a result of the *, but when we pop | |
6428 | back to the second ), we are at the stop_memory 1. | |
6429 | Thus, nothing is active. */ | |
6430 | if (r == 0) | |
6431 | { | |
6432 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
6433 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
6434 | } | |
6435 | else | |
6436 | highest_active_reg = r; | |
6437 | } | |
91c7b85d | 6438 | |
2b83a2a4 RM |
6439 | /* If just failed to match something this time around with a |
6440 | group that's operated on by a repetition operator, try to | |
6441 | force exit from the ``loop'', and restore the register | |
6442 | information for this group that we had before trying this | |
6443 | last match. */ | |
6444 | if ((!MATCHED_SOMETHING (reg_info[*p]) | |
6445 | || just_past_start_mem == p - 1) | |
91c7b85d | 6446 | && (p + 2) < pend) |
2b83a2a4 RM |
6447 | { |
6448 | boolean is_a_jump_n = false; | |
91c7b85d | 6449 | |
2b83a2a4 RM |
6450 | p1 = p + 2; |
6451 | mcnt = 0; | |
6452 | switch ((re_opcode_t) *p1++) | |
6453 | { | |
6454 | case jump_n: | |
6455 | is_a_jump_n = true; | |
6456 | case pop_failure_jump: | |
6457 | case maybe_pop_jump: | |
6458 | case jump: | |
6459 | case dummy_failure_jump: | |
6460 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
6461 | if (is_a_jump_n) | |
e4c785c8 | 6462 | p1 += OFFSET_ADDRESS_SIZE; |
2b83a2a4 | 6463 | break; |
91c7b85d | 6464 | |
2b83a2a4 RM |
6465 | default: |
6466 | /* do nothing */ ; | |
6467 | } | |
6468 | p1 += mcnt; | |
91c7b85d | 6469 | |
2b83a2a4 RM |
6470 | /* If the next operation is a jump backwards in the pattern |
6471 | to an on_failure_jump right before the start_memory | |
6472 | corresponding to this stop_memory, exit from the loop | |
6473 | by forcing a failure after pushing on the stack the | |
6474 | on_failure_jump's jump in the pattern, and d. */ | |
6475 | if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
e4c785c8 UD |
6476 | && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory |
6477 | && p1[2+OFFSET_ADDRESS_SIZE] == *p) | |
2b83a2a4 RM |
6478 | { |
6479 | /* If this group ever matched anything, then restore | |
6480 | what its registers were before trying this last | |
6481 | failed match, e.g., with `(a*)*b' against `ab' for | |
6482 | regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
6483 | against `aba' for regend[3]. | |
91c7b85d | 6484 | |
2b83a2a4 RM |
6485 | Also restore the registers for inner groups for, |
6486 | e.g., `((a*)(b*))*' against `aba' (register 3 would | |
6487 | otherwise get trashed). */ | |
91c7b85d | 6488 | |
2b83a2a4 RM |
6489 | if (EVER_MATCHED_SOMETHING (reg_info[*p])) |
6490 | { | |
91c7b85d RM |
6491 | unsigned r; |
6492 | ||
2b83a2a4 | 6493 | EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; |
91c7b85d | 6494 | |
2b83a2a4 | 6495 | /* Restore this and inner groups' (if any) registers. */ |
cccda09f UD |
6496 | for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1); |
6497 | r++) | |
2b83a2a4 RM |
6498 | { |
6499 | regstart[r] = old_regstart[r]; | |
6500 | ||
6501 | /* xx why this test? */ | |
6502 | if (old_regend[r] >= regstart[r]) | |
6503 | regend[r] = old_regend[r]; | |
91c7b85d | 6504 | } |
2b83a2a4 RM |
6505 | } |
6506 | p1++; | |
6507 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
6508 | PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
6509 | ||
6510 | goto fail; | |
6511 | } | |
6512 | } | |
91c7b85d | 6513 | |
2b83a2a4 RM |
6514 | /* Move past the register number and the inner group count. */ |
6515 | p += 2; | |
6516 | break; | |
6517 | ||
6518 | ||
6519 | /* \<digit> has been turned into a `duplicate' command which is | |
6520 | followed by the numeric value of <digit> as the register number. */ | |
6521 | case duplicate: | |
6522 | { | |
e4c785c8 | 6523 | register const CHAR_TYPE *d2, *dend2; |
2b83a2a4 RM |
6524 | int regno = *p++; /* Get which register to match against. */ |
6525 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); | |
6526 | ||
6527 | /* Can't back reference a group which we've never matched. */ | |
6528 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
6529 | goto fail; | |
91c7b85d | 6530 | |
2b83a2a4 RM |
6531 | /* Where in input to try to start matching. */ |
6532 | d2 = regstart[regno]; | |
91c7b85d | 6533 | |
2b83a2a4 RM |
6534 | /* Where to stop matching; if both the place to start and |
6535 | the place to stop matching are in the same string, then | |
6536 | set to the place to stop, otherwise, for now have to use | |
6537 | the end of the first string. */ | |
6538 | ||
91c7b85d | 6539 | dend2 = ((FIRST_STRING_P (regstart[regno]) |
2b83a2a4 RM |
6540 | == FIRST_STRING_P (regend[regno])) |
6541 | ? regend[regno] : end_match_1); | |
6542 | for (;;) | |
6543 | { | |
6544 | /* If necessary, advance to next segment in register | |
6545 | contents. */ | |
6546 | while (d2 == dend2) | |
6547 | { | |
6548 | if (dend2 == end_match_2) break; | |
6549 | if (dend2 == regend[regno]) break; | |
6550 | ||
6551 | /* End of string1 => advance to string2. */ | |
6552 | d2 = string2; | |
6553 | dend2 = regend[regno]; | |
6554 | } | |
6555 | /* At end of register contents => success */ | |
6556 | if (d2 == dend2) break; | |
6557 | ||
6558 | /* If necessary, advance to next segment in data. */ | |
6559 | PREFETCH (); | |
6560 | ||
6561 | /* How many characters left in this segment to match. */ | |
6562 | mcnt = dend - d; | |
91c7b85d | 6563 | |
2b83a2a4 RM |
6564 | /* Want how many consecutive characters we can match in |
6565 | one shot, so, if necessary, adjust the count. */ | |
6566 | if (mcnt > dend2 - d2) | |
6567 | mcnt = dend2 - d2; | |
91c7b85d | 6568 | |
2b83a2a4 RM |
6569 | /* Compare that many; failure if mismatch, else move |
6570 | past them. */ | |
91c7b85d RM |
6571 | if (translate |
6572 | ? bcmp_translate (d, d2, mcnt, translate) | |
e4c785c8 | 6573 | : memcmp (d, d2, mcnt*sizeof(US_CHAR_TYPE))) |
2b83a2a4 RM |
6574 | goto fail; |
6575 | d += mcnt, d2 += mcnt; | |
6576 | ||
6577 | /* Do this because we've match some characters. */ | |
6578 | SET_REGS_MATCHED (); | |
6579 | } | |
6580 | } | |
6581 | break; | |
6582 | ||
6583 | ||
6584 | /* begline matches the empty string at the beginning of the string | |
6585 | (unless `not_bol' is set in `bufp'), and, if | |
6586 | `newline_anchor' is set, after newlines. */ | |
6587 | case begline: | |
6588 | DEBUG_PRINT1 ("EXECUTING begline.\n"); | |
91c7b85d | 6589 | |
2b83a2a4 RM |
6590 | if (AT_STRINGS_BEG (d)) |
6591 | { | |
6592 | if (!bufp->not_bol) break; | |
6593 | } | |
6594 | else if (d[-1] == '\n' && bufp->newline_anchor) | |
6595 | { | |
6596 | break; | |
6597 | } | |
6598 | /* In all other cases, we fail. */ | |
6599 | goto fail; | |
6600 | ||
6601 | ||
6602 | /* endline is the dual of begline. */ | |
6603 | case endline: | |
6604 | DEBUG_PRINT1 ("EXECUTING endline.\n"); | |
6605 | ||
6606 | if (AT_STRINGS_END (d)) | |
6607 | { | |
6608 | if (!bufp->not_eol) break; | |
6609 | } | |
91c7b85d | 6610 | |
2b83a2a4 RM |
6611 | /* We have to ``prefetch'' the next character. */ |
6612 | else if ((d == end1 ? *string2 : *d) == '\n' | |
6613 | && bufp->newline_anchor) | |
6614 | { | |
6615 | break; | |
6616 | } | |
6617 | goto fail; | |
6618 | ||
6619 | ||
6620 | /* Match at the very beginning of the data. */ | |
6621 | case begbuf: | |
6622 | DEBUG_PRINT1 ("EXECUTING begbuf.\n"); | |
6623 | if (AT_STRINGS_BEG (d)) | |
6624 | break; | |
6625 | goto fail; | |
6626 | ||
6627 | ||
6628 | /* Match at the very end of the data. */ | |
6629 | case endbuf: | |
6630 | DEBUG_PRINT1 ("EXECUTING endbuf.\n"); | |
6631 | if (AT_STRINGS_END (d)) | |
6632 | break; | |
6633 | goto fail; | |
6634 | ||
6635 | ||
6636 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It | |
6637 | pushes NULL as the value for the string on the stack. Then | |
6638 | `pop_failure_point' will keep the current value for the | |
6639 | string, instead of restoring it. To see why, consider | |
6640 | matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
6641 | then the . fails against the \n. But the next thing we want | |
6642 | to do is match the \n against the \n; if we restored the | |
6643 | string value, we would be back at the foo. | |
91c7b85d | 6644 | |
2b83a2a4 RM |
6645 | Because this is used only in specific cases, we don't need to |
6646 | check all the things that `on_failure_jump' does, to make | |
6647 | sure the right things get saved on the stack. Hence we don't | |
6648 | share its code. The only reason to push anything on the | |
6649 | stack at all is that otherwise we would have to change | |
6650 | `anychar's code to do something besides goto fail in this | |
6651 | case; that seems worse than this. */ | |
6652 | case on_failure_keep_string_jump: | |
6653 | DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); | |
91c7b85d | 6654 | |
2b83a2a4 | 6655 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
5929563f UD |
6656 | #ifdef _LIBC |
6657 | DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt); | |
6658 | #else | |
2b83a2a4 | 6659 | DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); |
5929563f | 6660 | #endif |
2b83a2a4 RM |
6661 | |
6662 | PUSH_FAILURE_POINT (p + mcnt, NULL, -2); | |
6663 | break; | |
6664 | ||
6665 | ||
6666 | /* Uses of on_failure_jump: | |
91c7b85d | 6667 | |
2b83a2a4 RM |
6668 | Each alternative starts with an on_failure_jump that points |
6669 | to the beginning of the next alternative. Each alternative | |
6670 | except the last ends with a jump that in effect jumps past | |
6671 | the rest of the alternatives. (They really jump to the | |
6672 | ending jump of the following alternative, because tensioning | |
6673 | these jumps is a hassle.) | |
6674 | ||
6675 | Repeats start with an on_failure_jump that points past both | |
6676 | the repetition text and either the following jump or | |
6677 | pop_failure_jump back to this on_failure_jump. */ | |
6678 | case on_failure_jump: | |
6679 | on_failure: | |
6680 | DEBUG_PRINT1 ("EXECUTING on_failure_jump"); | |
6681 | ||
6682 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5929563f UD |
6683 | #ifdef _LIBC |
6684 | DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt); | |
6685 | #else | |
2b83a2a4 | 6686 | DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); |
5929563f | 6687 | #endif |
2b83a2a4 RM |
6688 | |
6689 | /* If this on_failure_jump comes right before a group (i.e., | |
6690 | the original * applied to a group), save the information | |
6691 | for that group and all inner ones, so that if we fail back | |
6692 | to this point, the group's information will be correct. | |
6693 | For example, in \(a*\)*\1, we need the preceding group, | |
8e3cc80f | 6694 | and in \(zz\(a*\)b*\)\2, we need the inner group. */ |
2b83a2a4 RM |
6695 | |
6696 | /* We can't use `p' to check ahead because we push | |
6697 | a failure point to `p + mcnt' after we do this. */ | |
6698 | p1 = p; | |
6699 | ||
6700 | /* We need to skip no_op's before we look for the | |
6701 | start_memory in case this on_failure_jump is happening as | |
6702 | the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
6703 | against aba. */ | |
6704 | while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
6705 | p1++; | |
6706 | ||
6707 | if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
6708 | { | |
6709 | /* We have a new highest active register now. This will | |
6710 | get reset at the start_memory we are about to get to, | |
6711 | but we will have saved all the registers relevant to | |
6712 | this repetition op, as described above. */ | |
6713 | highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
6714 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
6715 | lowest_active_reg = *(p1 + 1); | |
6716 | } | |
6717 | ||
6718 | DEBUG_PRINT1 (":\n"); | |
6719 | PUSH_FAILURE_POINT (p + mcnt, d, -2); | |
6720 | break; | |
6721 | ||
6722 | ||
6723 | /* A smart repeat ends with `maybe_pop_jump'. | |
6724 | We change it to either `pop_failure_jump' or `jump'. */ | |
6725 | case maybe_pop_jump: | |
6726 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
6727 | DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); | |
6728 | { | |
e4c785c8 | 6729 | register US_CHAR_TYPE *p2 = p; |
2b83a2a4 RM |
6730 | |
6731 | /* Compare the beginning of the repeat with what in the | |
6732 | pattern follows its end. If we can establish that there | |
6733 | is nothing that they would both match, i.e., that we | |
6734 | would have to backtrack because of (as in, e.g., `a*a') | |
6735 | then we can change to pop_failure_jump, because we'll | |
6736 | never have to backtrack. | |
91c7b85d | 6737 | |
2b83a2a4 RM |
6738 | This is not true in the case of alternatives: in |
6739 | `(a|ab)*' we do need to backtrack to the `ab' alternative | |
6740 | (e.g., if the string was `ab'). But instead of trying to | |
6741 | detect that here, the alternative has put on a dummy | |
6742 | failure point which is what we will end up popping. */ | |
6743 | ||
6744 | /* Skip over open/close-group commands. | |
6745 | If what follows this loop is a ...+ construct, | |
6746 | look at what begins its body, since we will have to | |
6747 | match at least one of that. */ | |
6748 | while (1) | |
6749 | { | |
6750 | if (p2 + 2 < pend | |
6751 | && ((re_opcode_t) *p2 == stop_memory | |
6752 | || (re_opcode_t) *p2 == start_memory)) | |
6753 | p2 += 3; | |
e4c785c8 | 6754 | else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend |
2b83a2a4 | 6755 | && (re_opcode_t) *p2 == dummy_failure_jump) |
e4c785c8 | 6756 | p2 += 2 + 2 * OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
6757 | else |
6758 | break; | |
6759 | } | |
6760 | ||
6761 | p1 = p + mcnt; | |
6762 | /* p1[0] ... p1[2] are the `on_failure_jump' corresponding | |
91c7b85d | 6763 | to the `maybe_finalize_jump' of this case. Examine what |
2b83a2a4 RM |
6764 | follows. */ |
6765 | ||
6766 | /* If we're at the end of the pattern, we can change. */ | |
6767 | if (p2 == pend) | |
6768 | { | |
6769 | /* Consider what happens when matching ":\(.*\)" | |
6770 | against ":/". I don't really understand this code | |
6771 | yet. */ | |
e4c785c8 UD |
6772 | p[-(1+OFFSET_ADDRESS_SIZE)] = (US_CHAR_TYPE) |
6773 | pop_failure_jump; | |
2b83a2a4 RM |
6774 | DEBUG_PRINT1 |
6775 | (" End of pattern: change to `pop_failure_jump'.\n"); | |
6776 | } | |
6777 | ||
6778 | else if ((re_opcode_t) *p2 == exactn | |
e4c785c8 UD |
6779 | #ifdef MBS_SUPPORT |
6780 | || (re_opcode_t) *p2 == exactn_bin | |
6781 | #endif | |
2b83a2a4 RM |
6782 | || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) |
6783 | { | |
e4c785c8 UD |
6784 | register US_CHAR_TYPE c |
6785 | = *p2 == (US_CHAR_TYPE) endline ? '\n' : p2[2]; | |
2b83a2a4 | 6786 | |
e4c785c8 UD |
6787 | if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn |
6788 | #ifdef MBS_SUPPORT | |
6789 | || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin | |
6790 | #endif | |
6791 | ) && p1[3+OFFSET_ADDRESS_SIZE] != c) | |
2b83a2a4 | 6792 | { |
e4c785c8 UD |
6793 | p[-(1+OFFSET_ADDRESS_SIZE)] = (US_CHAR_TYPE) |
6794 | pop_failure_jump; | |
672fd41b UD |
6795 | #ifdef MBS_SUPPORT |
6796 | if (MB_CUR_MAX != 1) | |
6797 | DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n", | |
6798 | (wint_t) c, | |
6799 | (wint_t) p1[3+OFFSET_ADDRESS_SIZE]); | |
6800 | else | |
6801 | #endif | |
6802 | DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", | |
6803 | (char) c, | |
6804 | (char) p1[3+OFFSET_ADDRESS_SIZE]); | |
2b83a2a4 | 6805 | } |
91c7b85d | 6806 | |
e4c785c8 | 6807 | #ifndef MBS_SUPPORT |
2b83a2a4 RM |
6808 | else if ((re_opcode_t) p1[3] == charset |
6809 | || (re_opcode_t) p1[3] == charset_not) | |
6810 | { | |
6811 | int not = (re_opcode_t) p1[3] == charset_not; | |
91c7b85d | 6812 | |
672fd41b | 6813 | if (c < (unsigned) (p1[4] * BYTEWIDTH) |
2b83a2a4 RM |
6814 | && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
6815 | not = !not; | |
6816 | ||
6817 | /* `not' is equal to 1 if c would match, which means | |
6818 | that we can't change to pop_failure_jump. */ | |
6819 | if (!not) | |
6820 | { | |
6821 | p[-3] = (unsigned char) pop_failure_jump; | |
6822 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
6823 | } | |
6824 | } | |
e4c785c8 | 6825 | #endif /* not MBS_SUPPORT */ |
2b83a2a4 | 6826 | } |
e4c785c8 | 6827 | #ifndef MBS_SUPPORT |
2b83a2a4 RM |
6828 | else if ((re_opcode_t) *p2 == charset) |
6829 | { | |
539491ac UD |
6830 | /* We win if the first character of the loop is not part |
6831 | of the charset. */ | |
6832 | if ((re_opcode_t) p1[3] == exactn | |
6833 | && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5] | |
6834 | && (p2[2 + p1[5] / BYTEWIDTH] | |
6835 | & (1 << (p1[5] % BYTEWIDTH))))) | |
a2860282 | 6836 | { |
539491ac UD |
6837 | p[-3] = (unsigned char) pop_failure_jump; |
6838 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
2b83a2a4 | 6839 | } |
91c7b85d | 6840 | |
2b83a2a4 RM |
6841 | else if ((re_opcode_t) p1[3] == charset_not) |
6842 | { | |
6843 | int idx; | |
6844 | /* We win if the charset_not inside the loop | |
6845 | lists every character listed in the charset after. */ | |
6846 | for (idx = 0; idx < (int) p2[1]; idx++) | |
6847 | if (! (p2[2 + idx] == 0 | |
6848 | || (idx < (int) p1[4] | |
6849 | && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) | |
6850 | break; | |
6851 | ||
6852 | if (idx == p2[1]) | |
6853 | { | |
6854 | p[-3] = (unsigned char) pop_failure_jump; | |
6855 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
6856 | } | |
6857 | } | |
6858 | else if ((re_opcode_t) p1[3] == charset) | |
6859 | { | |
6860 | int idx; | |
6861 | /* We win if the charset inside the loop | |
6862 | has no overlap with the one after the loop. */ | |
6863 | for (idx = 0; | |
6864 | idx < (int) p2[1] && idx < (int) p1[4]; | |
6865 | idx++) | |
6866 | if ((p2[2 + idx] & p1[5 + idx]) != 0) | |
6867 | break; | |
6868 | ||
6869 | if (idx == p2[1] || idx == p1[4]) | |
6870 | { | |
6871 | p[-3] = (unsigned char) pop_failure_jump; | |
6872 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); | |
6873 | } | |
6874 | } | |
6875 | } | |
e4c785c8 | 6876 | #endif /* not MBS_SUPPORT */ |
2b83a2a4 | 6877 | } |
e4c785c8 | 6878 | p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */ |
2b83a2a4 RM |
6879 | if ((re_opcode_t) p[-1] != pop_failure_jump) |
6880 | { | |
e4c785c8 | 6881 | p[-1] = (US_CHAR_TYPE) jump; |
2b83a2a4 RM |
6882 | DEBUG_PRINT1 (" Match => jump.\n"); |
6883 | goto unconditional_jump; | |
6884 | } | |
6885 | /* Note fall through. */ | |
6886 | ||
6887 | ||
6888 | /* The end of a simple repeat has a pop_failure_jump back to | |
6889 | its matching on_failure_jump, where the latter will push a | |
6890 | failure point. The pop_failure_jump takes off failure | |
6891 | points put on by this pop_failure_jump's matching | |
6892 | on_failure_jump; we got through the pattern to here from the | |
6893 | matching on_failure_jump, so didn't fail. */ | |
6894 | case pop_failure_jump: | |
6895 | { | |
6896 | /* We need to pass separate storage for the lowest and | |
6897 | highest registers, even though we don't care about the | |
6898 | actual values. Otherwise, we will restore only one | |
6899 | register from the stack, since lowest will == highest in | |
6900 | `pop_failure_point'. */ | |
4cca6b86 | 6901 | active_reg_t dummy_low_reg, dummy_high_reg; |
e4c785c8 UD |
6902 | US_CHAR_TYPE *pdummy = NULL; |
6903 | const CHAR_TYPE *sdummy = NULL; | |
2b83a2a4 RM |
6904 | |
6905 | DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); | |
6906 | POP_FAILURE_POINT (sdummy, pdummy, | |
6907 | dummy_low_reg, dummy_high_reg, | |
6908 | reg_dummy, reg_dummy, reg_info_dummy); | |
6909 | } | |
51702635 | 6910 | /* Note fall through. */ |
2b83a2a4 | 6911 | |
5929563f UD |
6912 | unconditional_jump: |
6913 | #ifdef _LIBC | |
6914 | DEBUG_PRINT2 ("\n%p: ", p); | |
6915 | #else | |
6916 | DEBUG_PRINT2 ("\n0x%x: ", p); | |
6917 | #endif | |
6918 | /* Note fall through. */ | |
91c7b85d | 6919 | |
2b83a2a4 RM |
6920 | /* Unconditionally jump (without popping any failure points). */ |
6921 | case jump: | |
2b83a2a4 RM |
6922 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
6923 | DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); | |
6924 | p += mcnt; /* Do the jump. */ | |
5929563f UD |
6925 | #ifdef _LIBC |
6926 | DEBUG_PRINT2 ("(to %p).\n", p); | |
6927 | #else | |
2b83a2a4 | 6928 | DEBUG_PRINT2 ("(to 0x%x).\n", p); |
5929563f | 6929 | #endif |
2b83a2a4 RM |
6930 | break; |
6931 | ||
91c7b85d | 6932 | |
2b83a2a4 RM |
6933 | /* We need this opcode so we can detect where alternatives end |
6934 | in `group_match_null_string_p' et al. */ | |
6935 | case jump_past_alt: | |
6936 | DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); | |
6937 | goto unconditional_jump; | |
6938 | ||
6939 | ||
6940 | /* Normally, the on_failure_jump pushes a failure point, which | |
6941 | then gets popped at pop_failure_jump. We will end up at | |
6942 | pop_failure_jump, also, and with a pattern of, say, `a+', we | |
6943 | are skipping over the on_failure_jump, so we have to push | |
6944 | something meaningless for pop_failure_jump to pop. */ | |
6945 | case dummy_failure_jump: | |
6946 | DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); | |
6947 | /* It doesn't matter what we push for the string here. What | |
6948 | the code at `fail' tests is the value for the pattern. */ | |
bca973bc | 6949 | PUSH_FAILURE_POINT (NULL, NULL, -2); |
2b83a2a4 RM |
6950 | goto unconditional_jump; |
6951 | ||
6952 | ||
6953 | /* At the end of an alternative, we need to push a dummy failure | |
6954 | point in case we are followed by a `pop_failure_jump', because | |
6955 | we don't want the failure point for the alternative to be | |
6956 | popped. For example, matching `(a|ab)*' against `aab' | |
6957 | requires that we match the `ab' alternative. */ | |
6958 | case push_dummy_failure: | |
6959 | DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); | |
6960 | /* See comments just above at `dummy_failure_jump' about the | |
6961 | two zeroes. */ | |
bca973bc | 6962 | PUSH_FAILURE_POINT (NULL, NULL, -2); |
2b83a2a4 RM |
6963 | break; |
6964 | ||
6965 | /* Have to succeed matching what follows at least n times. | |
6966 | After that, handle like `on_failure_jump'. */ | |
91c7b85d | 6967 | case succeed_n: |
e4c785c8 | 6968 | EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); |
2b83a2a4 RM |
6969 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); |
6970 | ||
6971 | assert (mcnt >= 0); | |
6972 | /* Originally, this is how many times we HAVE to succeed. */ | |
6973 | if (mcnt > 0) | |
6974 | { | |
6975 | mcnt--; | |
e4c785c8 | 6976 | p += OFFSET_ADDRESS_SIZE; |
2b83a2a4 | 6977 | STORE_NUMBER_AND_INCR (p, mcnt); |
5929563f | 6978 | #ifdef _LIBC |
e4c785c8 UD |
6979 | DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE |
6980 | , mcnt); | |
5929563f | 6981 | #else |
e4c785c8 UD |
6982 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE |
6983 | , mcnt); | |
5929563f | 6984 | #endif |
2b83a2a4 RM |
6985 | } |
6986 | else if (mcnt == 0) | |
6987 | { | |
5929563f | 6988 | #ifdef _LIBC |
e4c785c8 UD |
6989 | DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", |
6990 | p + OFFSET_ADDRESS_SIZE); | |
5929563f | 6991 | #else |
e4c785c8 UD |
6992 | DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", |
6993 | p + OFFSET_ADDRESS_SIZE); | |
6994 | #endif /* _LIBC */ | |
6995 | ||
6996 | #ifdef MBS_SUPPORT | |
6997 | p[1] = (US_CHAR_TYPE) no_op; | |
6998 | #else | |
6999 | p[2] = (US_CHAR_TYPE) no_op; | |
7000 | p[3] = (US_CHAR_TYPE) no_op; | |
7001 | #endif /* MBS_SUPPORT */ | |
2b83a2a4 RM |
7002 | goto on_failure; |
7003 | } | |
7004 | break; | |
91c7b85d RM |
7005 | |
7006 | case jump_n: | |
e4c785c8 | 7007 | EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); |
2b83a2a4 RM |
7008 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); |
7009 | ||
7010 | /* Originally, this is how many times we CAN jump. */ | |
7011 | if (mcnt) | |
7012 | { | |
7013 | mcnt--; | |
e4c785c8 UD |
7014 | STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt); |
7015 | ||
5929563f | 7016 | #ifdef _LIBC |
e4c785c8 UD |
7017 | DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE, |
7018 | mcnt); | |
5929563f | 7019 | #else |
e4c785c8 UD |
7020 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE, |
7021 | mcnt); | |
7022 | #endif /* _LIBC */ | |
91c7b85d | 7023 | goto unconditional_jump; |
2b83a2a4 RM |
7024 | } |
7025 | /* If don't have to jump any more, skip over the rest of command. */ | |
91c7b85d | 7026 | else |
e4c785c8 | 7027 | p += 2 * OFFSET_ADDRESS_SIZE; |
2b83a2a4 | 7028 | break; |
91c7b85d | 7029 | |
2b83a2a4 RM |
7030 | case set_number_at: |
7031 | { | |
7032 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); | |
7033 | ||
7034 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
7035 | p1 = p + mcnt; | |
7036 | EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
5929563f UD |
7037 | #ifdef _LIBC |
7038 | DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt); | |
7039 | #else | |
2b83a2a4 | 7040 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); |
5929563f | 7041 | #endif |
2b83a2a4 RM |
7042 | STORE_NUMBER (p1, mcnt); |
7043 | break; | |
7044 | } | |
7045 | ||
102800e0 RM |
7046 | #if 0 |
7047 | /* The DEC Alpha C compiler 3.x generates incorrect code for the | |
7048 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of | |
7049 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the | |
7050 | macro and introducing temporary variables works around the bug. */ | |
7051 | ||
7052 | case wordbound: | |
7053 | DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
7054 | if (AT_WORD_BOUNDARY (d)) | |
2b83a2a4 | 7055 | break; |
102800e0 | 7056 | goto fail; |
2b83a2a4 RM |
7057 | |
7058 | case notwordbound: | |
102800e0 | 7059 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); |
2b83a2a4 RM |
7060 | if (AT_WORD_BOUNDARY (d)) |
7061 | goto fail; | |
102800e0 RM |
7062 | break; |
7063 | #else | |
7064 | case wordbound: | |
7065 | { | |
7066 | boolean prevchar, thischar; | |
7067 | ||
7068 | DEBUG_PRINT1 ("EXECUTING wordbound.\n"); | |
7069 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) | |
7070 | break; | |
7071 | ||
7072 | prevchar = WORDCHAR_P (d - 1); | |
7073 | thischar = WORDCHAR_P (d); | |
7074 | if (prevchar != thischar) | |
7075 | break; | |
7076 | goto fail; | |
7077 | } | |
7078 | ||
7079 | case notwordbound: | |
7080 | { | |
7081 | boolean prevchar, thischar; | |
7082 | ||
7083 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); | |
7084 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) | |
7085 | goto fail; | |
7086 | ||
7087 | prevchar = WORDCHAR_P (d - 1); | |
7088 | thischar = WORDCHAR_P (d); | |
7089 | if (prevchar != thischar) | |
7090 | goto fail; | |
7091 | break; | |
7092 | } | |
7093 | #endif | |
2b83a2a4 RM |
7094 | |
7095 | case wordbeg: | |
7096 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); | |
e10a9eba UD |
7097 | if (!AT_STRINGS_END (d) && WORDCHAR_P (d) |
7098 | && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) | |
2b83a2a4 RM |
7099 | break; |
7100 | goto fail; | |
7101 | ||
7102 | case wordend: | |
7103 | DEBUG_PRINT1 ("EXECUTING wordend.\n"); | |
7104 | if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) | |
e10a9eba | 7105 | && (AT_STRINGS_END (d) || !WORDCHAR_P (d))) |
2b83a2a4 RM |
7106 | break; |
7107 | goto fail; | |
7108 | ||
7109 | #ifdef emacs | |
7110 | case before_dot: | |
7111 | DEBUG_PRINT1 ("EXECUTING before_dot.\n"); | |
7112 | if (PTR_CHAR_POS ((unsigned char *) d) >= point) | |
7113 | goto fail; | |
7114 | break; | |
91c7b85d | 7115 | |
2b83a2a4 RM |
7116 | case at_dot: |
7117 | DEBUG_PRINT1 ("EXECUTING at_dot.\n"); | |
7118 | if (PTR_CHAR_POS ((unsigned char *) d) != point) | |
7119 | goto fail; | |
7120 | break; | |
91c7b85d | 7121 | |
2b83a2a4 RM |
7122 | case after_dot: |
7123 | DEBUG_PRINT1 ("EXECUTING after_dot.\n"); | |
7124 | if (PTR_CHAR_POS ((unsigned char *) d) <= point) | |
7125 | goto fail; | |
7126 | break; | |
2b83a2a4 RM |
7127 | |
7128 | case syntaxspec: | |
7129 | DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); | |
7130 | mcnt = *p++; | |
7131 | goto matchsyntax; | |
7132 | ||
7133 | case wordchar: | |
7134 | DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); | |
7135 | mcnt = (int) Sword; | |
7136 | matchsyntax: | |
7137 | PREFETCH (); | |
7138 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ | |
7139 | d++; | |
7140 | if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt) | |
7141 | goto fail; | |
7142 | SET_REGS_MATCHED (); | |
7143 | break; | |
7144 | ||
7145 | case notsyntaxspec: | |
7146 | DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); | |
7147 | mcnt = *p++; | |
7148 | goto matchnotsyntax; | |
7149 | ||
7150 | case notwordchar: | |
7151 | DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); | |
7152 | mcnt = (int) Sword; | |
7153 | matchnotsyntax: | |
7154 | PREFETCH (); | |
7155 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ | |
7156 | d++; | |
7157 | if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt) | |
7158 | goto fail; | |
7159 | SET_REGS_MATCHED (); | |
7160 | break; | |
7161 | ||
7162 | #else /* not emacs */ | |
7163 | case wordchar: | |
7164 | DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); | |
7165 | PREFETCH (); | |
7166 | if (!WORDCHAR_P (d)) | |
7167 | goto fail; | |
7168 | SET_REGS_MATCHED (); | |
7169 | d++; | |
7170 | break; | |
91c7b85d | 7171 | |
2b83a2a4 RM |
7172 | case notwordchar: |
7173 | DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); | |
7174 | PREFETCH (); | |
7175 | if (WORDCHAR_P (d)) | |
7176 | goto fail; | |
7177 | SET_REGS_MATCHED (); | |
7178 | d++; | |
7179 | break; | |
7180 | #endif /* not emacs */ | |
91c7b85d | 7181 | |
2b83a2a4 RM |
7182 | default: |
7183 | abort (); | |
7184 | } | |
7185 | continue; /* Successfully executed one pattern command; keep going. */ | |
7186 | ||
7187 | ||
7188 | /* We goto here if a matching operation fails. */ | |
7189 | fail: | |
7190 | if (!FAIL_STACK_EMPTY ()) | |
7191 | { /* A restart point is known. Restore to that state. */ | |
7192 | DEBUG_PRINT1 ("\nFAIL:\n"); | |
7193 | POP_FAILURE_POINT (d, p, | |
7194 | lowest_active_reg, highest_active_reg, | |
7195 | regstart, regend, reg_info); | |
7196 | ||
7197 | /* If this failure point is a dummy, try the next one. */ | |
7198 | if (!p) | |
7199 | goto fail; | |
7200 | ||
7201 | /* If we failed to the end of the pattern, don't examine *p. */ | |
7202 | assert (p <= pend); | |
7203 | if (p < pend) | |
7204 | { | |
7205 | boolean is_a_jump_n = false; | |
91c7b85d | 7206 | |
2b83a2a4 RM |
7207 | /* If failed to a backwards jump that's part of a repetition |
7208 | loop, need to pop this failure point and use the next one. */ | |
7209 | switch ((re_opcode_t) *p) | |
7210 | { | |
7211 | case jump_n: | |
7212 | is_a_jump_n = true; | |
7213 | case maybe_pop_jump: | |
7214 | case pop_failure_jump: | |
7215 | case jump: | |
7216 | p1 = p + 1; | |
7217 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
91c7b85d | 7218 | p1 += mcnt; |
2b83a2a4 RM |
7219 | |
7220 | if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
7221 | || (!is_a_jump_n | |
7222 | && (re_opcode_t) *p1 == on_failure_jump)) | |
7223 | goto fail; | |
7224 | break; | |
7225 | default: | |
7226 | /* do nothing */ ; | |
7227 | } | |
7228 | } | |
7229 | ||
7230 | if (d >= string1 && d <= end1) | |
7231 | dend = end_match_1; | |
7232 | } | |
7233 | else | |
7234 | break; /* Matching at this starting point really fails. */ | |
7235 | } /* for (;;) */ | |
7236 | ||
7237 | if (best_regs_set) | |
7238 | goto restore_best_regs; | |
7239 | ||
7240 | FREE_VARIABLES (); | |
7241 | ||
7242 | return -1; /* Failure to match. */ | |
7243 | } /* re_match_2 */ | |
7244 | \f | |
7245 | /* Subroutine definitions for re_match_2. */ | |
7246 | ||
7247 | ||
7248 | /* We are passed P pointing to a register number after a start_memory. | |
91c7b85d | 7249 | |
2b83a2a4 RM |
7250 | Return true if the pattern up to the corresponding stop_memory can |
7251 | match the empty string, and false otherwise. | |
91c7b85d | 7252 | |
2b83a2a4 RM |
7253 | If we find the matching stop_memory, sets P to point to one past its number. |
7254 | Otherwise, sets P to an undefined byte less than or equal to END. | |
7255 | ||
7256 | We don't handle duplicates properly (yet). */ | |
7257 | ||
7258 | static boolean | |
7259 | group_match_null_string_p (p, end, reg_info) | |
e4c785c8 | 7260 | US_CHAR_TYPE **p, *end; |
2b83a2a4 RM |
7261 | register_info_type *reg_info; |
7262 | { | |
7263 | int mcnt; | |
7264 | /* Point to after the args to the start_memory. */ | |
e4c785c8 | 7265 | US_CHAR_TYPE *p1 = *p + 2; |
91c7b85d | 7266 | |
2b83a2a4 RM |
7267 | while (p1 < end) |
7268 | { | |
7269 | /* Skip over opcodes that can match nothing, and return true or | |
7270 | false, as appropriate, when we get to one that can't, or to the | |
7271 | matching stop_memory. */ | |
91c7b85d | 7272 | |
2b83a2a4 RM |
7273 | switch ((re_opcode_t) *p1) |
7274 | { | |
7275 | /* Could be either a loop or a series of alternatives. */ | |
7276 | case on_failure_jump: | |
7277 | p1++; | |
7278 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
91c7b85d | 7279 | |
2b83a2a4 RM |
7280 | /* If the next operation is not a jump backwards in the |
7281 | pattern. */ | |
7282 | ||
7283 | if (mcnt >= 0) | |
7284 | { | |
7285 | /* Go through the on_failure_jumps of the alternatives, | |
7286 | seeing if any of the alternatives cannot match nothing. | |
7287 | The last alternative starts with only a jump, | |
7288 | whereas the rest start with on_failure_jump and end | |
7289 | with a jump, e.g., here is the pattern for `a|b|c': | |
7290 | ||
7291 | /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 | |
7292 | /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
91c7b85d | 7293 | /exactn/1/c |
2b83a2a4 RM |
7294 | |
7295 | So, we have to first go through the first (n-1) | |
7296 | alternatives and then deal with the last one separately. */ | |
7297 | ||
7298 | ||
7299 | /* Deal with the first (n-1) alternatives, which start | |
7300 | with an on_failure_jump (see above) that jumps to right | |
7301 | past a jump_past_alt. */ | |
7302 | ||
e4c785c8 UD |
7303 | while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] == |
7304 | jump_past_alt) | |
2b83a2a4 RM |
7305 | { |
7306 | /* `mcnt' holds how many bytes long the alternative | |
7307 | is, including the ending `jump_past_alt' and | |
7308 | its number. */ | |
7309 | ||
e4c785c8 UD |
7310 | if (!alt_match_null_string_p (p1, p1 + mcnt - |
7311 | (1 + OFFSET_ADDRESS_SIZE), | |
7312 | reg_info)) | |
2b83a2a4 RM |
7313 | return false; |
7314 | ||
7315 | /* Move to right after this alternative, including the | |
7316 | jump_past_alt. */ | |
91c7b85d | 7317 | p1 += mcnt; |
2b83a2a4 RM |
7318 | |
7319 | /* Break if it's the beginning of an n-th alternative | |
7320 | that doesn't begin with an on_failure_jump. */ | |
7321 | if ((re_opcode_t) *p1 != on_failure_jump) | |
7322 | break; | |
91c7b85d | 7323 | |
2b83a2a4 RM |
7324 | /* Still have to check that it's not an n-th |
7325 | alternative that starts with an on_failure_jump. */ | |
7326 | p1++; | |
7327 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
e4c785c8 UD |
7328 | if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] != |
7329 | jump_past_alt) | |
2b83a2a4 RM |
7330 | { |
7331 | /* Get to the beginning of the n-th alternative. */ | |
e4c785c8 | 7332 | p1 -= 1 + OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
7333 | break; |
7334 | } | |
7335 | } | |
7336 | ||
7337 | /* Deal with the last alternative: go back and get number | |
7338 | of the `jump_past_alt' just before it. `mcnt' contains | |
7339 | the length of the alternative. */ | |
e4c785c8 | 7340 | EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE); |
2b83a2a4 RM |
7341 | |
7342 | if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) | |
7343 | return false; | |
7344 | ||
7345 | p1 += mcnt; /* Get past the n-th alternative. */ | |
7346 | } /* if mcnt > 0 */ | |
7347 | break; | |
7348 | ||
91c7b85d | 7349 | |
2b83a2a4 RM |
7350 | case stop_memory: |
7351 | assert (p1[1] == **p); | |
7352 | *p = p1 + 2; | |
7353 | return true; | |
7354 | ||
91c7b85d RM |
7355 | |
7356 | default: | |
2b83a2a4 RM |
7357 | if (!common_op_match_null_string_p (&p1, end, reg_info)) |
7358 | return false; | |
7359 | } | |
7360 | } /* while p1 < end */ | |
7361 | ||
7362 | return false; | |
7363 | } /* group_match_null_string_p */ | |
7364 | ||
7365 | ||
7366 | /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
7367 | It expects P to be the first byte of a single alternative and END one | |
7368 | byte past the last. The alternative can contain groups. */ | |
91c7b85d | 7369 | |
2b83a2a4 RM |
7370 | static boolean |
7371 | alt_match_null_string_p (p, end, reg_info) | |
e4c785c8 | 7372 | US_CHAR_TYPE *p, *end; |
2b83a2a4 RM |
7373 | register_info_type *reg_info; |
7374 | { | |
7375 | int mcnt; | |
e4c785c8 | 7376 | US_CHAR_TYPE *p1 = p; |
91c7b85d | 7377 | |
2b83a2a4 RM |
7378 | while (p1 < end) |
7379 | { | |
91c7b85d | 7380 | /* Skip over opcodes that can match nothing, and break when we get |
2b83a2a4 | 7381 | to one that can't. */ |
91c7b85d | 7382 | |
2b83a2a4 RM |
7383 | switch ((re_opcode_t) *p1) |
7384 | { | |
7385 | /* It's a loop. */ | |
7386 | case on_failure_jump: | |
7387 | p1++; | |
7388 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
7389 | p1 += mcnt; | |
7390 | break; | |
91c7b85d RM |
7391 | |
7392 | default: | |
2b83a2a4 RM |
7393 | if (!common_op_match_null_string_p (&p1, end, reg_info)) |
7394 | return false; | |
7395 | } | |
7396 | } /* while p1 < end */ | |
7397 | ||
7398 | return true; | |
7399 | } /* alt_match_null_string_p */ | |
7400 | ||
7401 | ||
7402 | /* Deals with the ops common to group_match_null_string_p and | |
91c7b85d RM |
7403 | alt_match_null_string_p. |
7404 | ||
2b83a2a4 RM |
7405 | Sets P to one after the op and its arguments, if any. */ |
7406 | ||
7407 | static boolean | |
7408 | common_op_match_null_string_p (p, end, reg_info) | |
e4c785c8 | 7409 | US_CHAR_TYPE **p, *end; |
2b83a2a4 RM |
7410 | register_info_type *reg_info; |
7411 | { | |
7412 | int mcnt; | |
7413 | boolean ret; | |
7414 | int reg_no; | |
e4c785c8 | 7415 | US_CHAR_TYPE *p1 = *p; |
2b83a2a4 RM |
7416 | |
7417 | switch ((re_opcode_t) *p1++) | |
7418 | { | |
7419 | case no_op: | |
7420 | case begline: | |
7421 | case endline: | |
7422 | case begbuf: | |
7423 | case endbuf: | |
7424 | case wordbeg: | |
7425 | case wordend: | |
7426 | case wordbound: | |
7427 | case notwordbound: | |
7428 | #ifdef emacs | |
7429 | case before_dot: | |
7430 | case at_dot: | |
7431 | case after_dot: | |
7432 | #endif | |
7433 | break; | |
7434 | ||
7435 | case start_memory: | |
7436 | reg_no = *p1; | |
7437 | assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
7438 | ret = group_match_null_string_p (&p1, end, reg_info); | |
91c7b85d | 7439 | |
2b83a2a4 RM |
7440 | /* Have to set this here in case we're checking a group which |
7441 | contains a group and a back reference to it. */ | |
7442 | ||
7443 | if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
7444 | REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; | |
7445 | ||
7446 | if (!ret) | |
7447 | return false; | |
7448 | break; | |
91c7b85d | 7449 | |
2b83a2a4 RM |
7450 | /* If this is an optimized succeed_n for zero times, make the jump. */ |
7451 | case jump: | |
7452 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
7453 | if (mcnt >= 0) | |
7454 | p1 += mcnt; | |
7455 | else | |
7456 | return false; | |
7457 | break; | |
7458 | ||
7459 | case succeed_n: | |
7460 | /* Get to the number of times to succeed. */ | |
e4c785c8 | 7461 | p1 += OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
7462 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
7463 | ||
7464 | if (mcnt == 0) | |
7465 | { | |
e4c785c8 | 7466 | p1 -= 2 * OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
7467 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
7468 | p1 += mcnt; | |
7469 | } | |
7470 | else | |
7471 | return false; | |
7472 | break; | |
7473 | ||
91c7b85d | 7474 | case duplicate: |
2b83a2a4 RM |
7475 | if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) |
7476 | return false; | |
7477 | break; | |
7478 | ||
7479 | case set_number_at: | |
e4c785c8 | 7480 | p1 += 2 * OFFSET_ADDRESS_SIZE; |
2b83a2a4 RM |
7481 | |
7482 | default: | |
7483 | /* All other opcodes mean we cannot match the empty string. */ | |
7484 | return false; | |
7485 | } | |
7486 | ||
7487 | *p = p1; | |
7488 | return true; | |
7489 | } /* common_op_match_null_string_p */ | |
7490 | ||
7491 | ||
7492 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
7493 | bytes; nonzero otherwise. */ | |
91c7b85d | 7494 | |
2b83a2a4 RM |
7495 | static int |
7496 | bcmp_translate (s1, s2, len, translate) | |
e4c785c8 | 7497 | const CHAR_TYPE *s1, *s2; |
2b83a2a4 | 7498 | register int len; |
03a75825 | 7499 | RE_TRANSLATE_TYPE translate; |
2b83a2a4 | 7500 | { |
e4c785c8 UD |
7501 | register const US_CHAR_TYPE *p1 = (const US_CHAR_TYPE *) s1; |
7502 | register const US_CHAR_TYPE *p2 = (const US_CHAR_TYPE *) s2; | |
2b83a2a4 RM |
7503 | while (len) |
7504 | { | |
e4c785c8 UD |
7505 | #ifdef MBS_SUPPORT |
7506 | if (((*p1<=0xff)?translate[*p1++]:*p1++) | |
7507 | != ((*p2<=0xff)?translate[*p2++]:*p2++)) | |
7508 | return 1; | |
7509 | #else | |
2b83a2a4 | 7510 | if (translate[*p1++] != translate[*p2++]) return 1; |
e4c785c8 | 7511 | #endif /* MBS_SUPPORT */ |
2b83a2a4 RM |
7512 | len--; |
7513 | } | |
7514 | return 0; | |
7515 | } | |
7516 | \f | |
7517 | /* Entry points for GNU code. */ | |
7518 | ||
7519 | /* re_compile_pattern is the GNU regular expression compiler: it | |
7520 | compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
7521 | Returns 0 if the pattern was valid, otherwise an error string. | |
91c7b85d | 7522 | |
2b83a2a4 RM |
7523 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
7524 | are set in BUFP on entry. | |
91c7b85d | 7525 | |
2b83a2a4 RM |
7526 | We call regex_compile to do the actual compilation. */ |
7527 | ||
7528 | const char * | |
7529 | re_compile_pattern (pattern, length, bufp) | |
7530 | const char *pattern; | |
4cca6b86 | 7531 | size_t length; |
2b83a2a4 RM |
7532 | struct re_pattern_buffer *bufp; |
7533 | { | |
7534 | reg_errcode_t ret; | |
91c7b85d | 7535 | |
2b83a2a4 RM |
7536 | /* GNU code is written to assume at least RE_NREGS registers will be set |
7537 | (and at least one extra will be -1). */ | |
7538 | bufp->regs_allocated = REGS_UNALLOCATED; | |
91c7b85d | 7539 | |
2b83a2a4 RM |
7540 | /* And GNU code determines whether or not to get register information |
7541 | by passing null for the REGS argument to re_match, etc., not by | |
7542 | setting no_sub. */ | |
7543 | bufp->no_sub = 0; | |
91c7b85d | 7544 | |
2b83a2a4 RM |
7545 | /* Match anchors at newline. */ |
7546 | bufp->newline_anchor = 1; | |
91c7b85d | 7547 | |
2b83a2a4 RM |
7548 | ret = regex_compile (pattern, length, re_syntax_options, bufp); |
7549 | ||
7550 | if (!ret) | |
7551 | return NULL; | |
c4563d2d | 7552 | return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]); |
91c7b85d | 7553 | } |
2ad4fab2 UD |
7554 | #ifdef _LIBC |
7555 | weak_alias (__re_compile_pattern, re_compile_pattern) | |
7556 | #endif | |
2b83a2a4 RM |
7557 | \f |
7558 | /* Entry points compatible with 4.2 BSD regex library. We don't define | |
7559 | them unless specifically requested. */ | |
7560 | ||
86187531 | 7561 | #if defined _REGEX_RE_COMP || defined _LIBC |
2b83a2a4 RM |
7562 | |
7563 | /* BSD has one and only one pattern buffer. */ | |
7564 | static struct re_pattern_buffer re_comp_buf; | |
7565 | ||
51702635 UD |
7566 | char * |
7567 | #ifdef _LIBC | |
7568 | /* Make these definitions weak in libc, so POSIX programs can redefine | |
7569 | these names if they don't use our functions, and still use | |
7570 | regcomp/regexec below without link errors. */ | |
7571 | weak_function | |
7572 | #endif | |
2b83a2a4 RM |
7573 | re_comp (s) |
7574 | const char *s; | |
7575 | { | |
7576 | reg_errcode_t ret; | |
91c7b85d | 7577 | |
2b83a2a4 RM |
7578 | if (!s) |
7579 | { | |
7580 | if (!re_comp_buf.buffer) | |
7581 | return gettext ("No previous regular expression"); | |
7582 | return 0; | |
7583 | } | |
7584 | ||
7585 | if (!re_comp_buf.buffer) | |
7586 | { | |
7587 | re_comp_buf.buffer = (unsigned char *) malloc (200); | |
7588 | if (re_comp_buf.buffer == NULL) | |
c4563d2d UD |
7589 | return (char *) gettext (re_error_msgid |
7590 | + re_error_msgid_idx[(int) REG_ESPACE]); | |
2b83a2a4 RM |
7591 | re_comp_buf.allocated = 200; |
7592 | ||
7593 | re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); | |
7594 | if (re_comp_buf.fastmap == NULL) | |
c4563d2d UD |
7595 | return (char *) gettext (re_error_msgid |
7596 | + re_error_msgid_idx[(int) REG_ESPACE]); | |
2b83a2a4 RM |
7597 | } |
7598 | ||
7599 | /* Since `re_exec' always passes NULL for the `regs' argument, we | |
7600 | don't need to initialize the pattern buffer fields which affect it. */ | |
7601 | ||
7602 | /* Match anchors at newlines. */ | |
7603 | re_comp_buf.newline_anchor = 1; | |
7604 | ||
7605 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); | |
91c7b85d | 7606 | |
2b83a2a4 RM |
7607 | if (!ret) |
7608 | return NULL; | |
7609 | ||
7610 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ | |
c4563d2d | 7611 | return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]); |
2b83a2a4 RM |
7612 | } |
7613 | ||
7614 | ||
51702635 UD |
7615 | int |
7616 | #ifdef _LIBC | |
7617 | weak_function | |
7618 | #endif | |
2b83a2a4 RM |
7619 | re_exec (s) |
7620 | const char *s; | |
7621 | { | |
7622 | const int len = strlen (s); | |
7623 | return | |
7624 | 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
7625 | } | |
dc997231 | 7626 | |
2b83a2a4 RM |
7627 | #endif /* _REGEX_RE_COMP */ |
7628 | \f | |
7629 | /* POSIX.2 functions. Don't define these for Emacs. */ | |
7630 | ||
7631 | #ifndef emacs | |
7632 | ||
7633 | /* regcomp takes a regular expression as a string and compiles it. | |
7634 | ||
7635 | PREG is a regex_t *. We do not expect any fields to be initialized, | |
7636 | since POSIX says we shouldn't. Thus, we set | |
7637 | ||
7638 | `buffer' to the compiled pattern; | |
7639 | `used' to the length of the compiled pattern; | |
7640 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
7641 | REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
7642 | RE_SYNTAX_POSIX_BASIC; | |
7643 | `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
a5d1d726 UD |
7644 | `fastmap' to an allocated space for the fastmap; |
7645 | `fastmap_accurate' to zero; | |
2b83a2a4 RM |
7646 | `re_nsub' to the number of subexpressions in PATTERN. |
7647 | ||
7648 | PATTERN is the address of the pattern string. | |
7649 | ||
7650 | CFLAGS is a series of bits which affect compilation. | |
7651 | ||
7652 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
7653 | use POSIX basic syntax. | |
7654 | ||
7655 | If REG_NEWLINE is set, then . and [^...] don't match newline. | |
7656 | Also, regexec will try a match beginning after every newline. | |
7657 | ||
7658 | If REG_ICASE is set, then we considers upper- and lowercase | |
7659 | versions of letters to be equivalent when matching. | |
7660 | ||
7661 | If REG_NOSUB is set, then when PREG is passed to regexec, that | |
7662 | routine will report only success or failure, and nothing about the | |
7663 | registers. | |
7664 | ||
7665 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for | |
7666 | the return codes and their meanings.) */ | |
7667 | ||
7668 | int | |
7669 | regcomp (preg, pattern, cflags) | |
7670 | regex_t *preg; | |
91c7b85d | 7671 | const char *pattern; |
2b83a2a4 RM |
7672 | int cflags; |
7673 | { | |
7674 | reg_errcode_t ret; | |
4cca6b86 | 7675 | reg_syntax_t syntax |
2b83a2a4 RM |
7676 | = (cflags & REG_EXTENDED) ? |
7677 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
7678 | ||
7679 | /* regex_compile will allocate the space for the compiled pattern. */ | |
7680 | preg->buffer = 0; | |
7681 | preg->allocated = 0; | |
7682 | preg->used = 0; | |
91c7b85d | 7683 | |
a5d1d726 UD |
7684 | /* Try to allocate space for the fastmap. */ |
7685 | preg->fastmap = (char *) malloc (1 << BYTEWIDTH); | |
91c7b85d | 7686 | |
2b83a2a4 RM |
7687 | if (cflags & REG_ICASE) |
7688 | { | |
7689 | unsigned i; | |
91c7b85d | 7690 | |
03a75825 RM |
7691 | preg->translate |
7692 | = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE | |
7693 | * sizeof (*(RE_TRANSLATE_TYPE)0)); | |
2b83a2a4 RM |
7694 | if (preg->translate == NULL) |
7695 | return (int) REG_ESPACE; | |
7696 | ||
7697 | /* Map uppercase characters to corresponding lowercase ones. */ | |
7698 | for (i = 0; i < CHAR_SET_SIZE; i++) | |
4caef86c | 7699 | preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; |
2b83a2a4 RM |
7700 | } |
7701 | else | |
7702 | preg->translate = NULL; | |
7703 | ||
7704 | /* If REG_NEWLINE is set, newlines are treated differently. */ | |
7705 | if (cflags & REG_NEWLINE) | |
7706 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
7707 | syntax &= ~RE_DOT_NEWLINE; | |
7708 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
7709 | /* It also changes the matching behavior. */ | |
7710 | preg->newline_anchor = 1; | |
7711 | } | |
7712 | else | |
7713 | preg->newline_anchor = 0; | |
7714 | ||
7715 | preg->no_sub = !!(cflags & REG_NOSUB); | |
7716 | ||
91c7b85d | 7717 | /* POSIX says a null character in the pattern terminates it, so we |
2b83a2a4 RM |
7718 | can use strlen here in compiling the pattern. */ |
7719 | ret = regex_compile (pattern, strlen (pattern), syntax, preg); | |
91c7b85d | 7720 | |
2b83a2a4 RM |
7721 | /* POSIX doesn't distinguish between an unmatched open-group and an |
7722 | unmatched close-group: both are REG_EPAREN. */ | |
7723 | if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
91c7b85d | 7724 | |
a5d1d726 UD |
7725 | if (ret == REG_NOERROR && preg->fastmap) |
7726 | { | |
7727 | /* Compute the fastmap now, since regexec cannot modify the pattern | |
7728 | buffer. */ | |
7729 | if (re_compile_fastmap (preg) == -2) | |
7730 | { | |
49c091e5 | 7731 | /* Some error occurred while computing the fastmap, just forget |
a5d1d726 UD |
7732 | about it. */ |
7733 | free (preg->fastmap); | |
7734 | preg->fastmap = NULL; | |
7735 | } | |
7736 | } | |
7737 | ||
2b83a2a4 RM |
7738 | return (int) ret; |
7739 | } | |
2ad4fab2 UD |
7740 | #ifdef _LIBC |
7741 | weak_alias (__regcomp, regcomp) | |
7742 | #endif | |
2b83a2a4 RM |
7743 | |
7744 | ||
7745 | /* regexec searches for a given pattern, specified by PREG, in the | |
7746 | string STRING. | |
91c7b85d | 7747 | |
2b83a2a4 RM |
7748 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
7749 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | |
7750 | least NMATCH elements, and we set them to the offsets of the | |
7751 | corresponding matched substrings. | |
91c7b85d | 7752 | |
2b83a2a4 RM |
7753 | EFLAGS specifies `execution flags' which affect matching: if |
7754 | REG_NOTBOL is set, then ^ does not match at the beginning of the | |
7755 | string; if REG_NOTEOL is set, then $ does not match at the end. | |
91c7b85d | 7756 | |
2b83a2a4 RM |
7757 | We return 0 if we find a match and REG_NOMATCH if not. */ |
7758 | ||
7759 | int | |
7760 | regexec (preg, string, nmatch, pmatch, eflags) | |
7761 | const regex_t *preg; | |
91c7b85d RM |
7762 | const char *string; |
7763 | size_t nmatch; | |
7764 | regmatch_t pmatch[]; | |
2b83a2a4 RM |
7765 | int eflags; |
7766 | { | |
7767 | int ret; | |
7768 | struct re_registers regs; | |
7769 | regex_t private_preg; | |
7770 | int len = strlen (string); | |
7771 | boolean want_reg_info = !preg->no_sub && nmatch > 0; | |
7772 | ||
7773 | private_preg = *preg; | |
91c7b85d | 7774 | |
2b83a2a4 RM |
7775 | private_preg.not_bol = !!(eflags & REG_NOTBOL); |
7776 | private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
91c7b85d | 7777 | |
2b83a2a4 RM |
7778 | /* The user has told us exactly how many registers to return |
7779 | information about, via `nmatch'. We have to pass that on to the | |
7780 | matching routines. */ | |
7781 | private_preg.regs_allocated = REGS_FIXED; | |
91c7b85d | 7782 | |
2b83a2a4 RM |
7783 | if (want_reg_info) |
7784 | { | |
7785 | regs.num_regs = nmatch; | |
a5d1d726 UD |
7786 | regs.start = TALLOC (nmatch * 2, regoff_t); |
7787 | if (regs.start == NULL) | |
2b83a2a4 | 7788 | return (int) REG_NOMATCH; |
a5d1d726 | 7789 | regs.end = regs.start + nmatch; |
2b83a2a4 RM |
7790 | } |
7791 | ||
7792 | /* Perform the searching operation. */ | |
7793 | ret = re_search (&private_preg, string, len, | |
7794 | /* start: */ 0, /* range: */ len, | |
7795 | want_reg_info ? ®s : (struct re_registers *) 0); | |
91c7b85d | 7796 | |
2b83a2a4 RM |
7797 | /* Copy the register information to the POSIX structure. */ |
7798 | if (want_reg_info) | |
7799 | { | |
7800 | if (ret >= 0) | |
7801 | { | |
7802 | unsigned r; | |
7803 | ||
7804 | for (r = 0; r < nmatch; r++) | |
7805 | { | |
7806 | pmatch[r].rm_so = regs.start[r]; | |
7807 | pmatch[r].rm_eo = regs.end[r]; | |
7808 | } | |
7809 | } | |
7810 | ||
7811 | /* If we needed the temporary register info, free the space now. */ | |
7812 | free (regs.start); | |
2b83a2a4 RM |
7813 | } |
7814 | ||
7815 | /* We want zero return to mean success, unlike `re_search'. */ | |
7816 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
7817 | } | |
2ad4fab2 UD |
7818 | #ifdef _LIBC |
7819 | weak_alias (__regexec, regexec) | |
7820 | #endif | |
2b83a2a4 RM |
7821 | |
7822 | ||
7823 | /* Returns a message corresponding to an error code, ERRCODE, returned | |
7824 | from either regcomp or regexec. We don't use PREG here. */ | |
7825 | ||
7826 | size_t | |
7cabd57c | 7827 | regerror (errcode, preg, errbuf, errbuf_size) |
2b83a2a4 RM |
7828 | int errcode; |
7829 | const regex_t *preg; | |
7830 | char *errbuf; | |
7831 | size_t errbuf_size; | |
7832 | { | |
7833 | const char *msg; | |
7834 | size_t msg_size; | |
7835 | ||
7836 | if (errcode < 0 | |
c4563d2d UD |
7837 | || errcode >= (int) (sizeof (re_error_msgid_idx) |
7838 | / sizeof (re_error_msgid_idx[0]))) | |
91c7b85d | 7839 | /* Only error codes returned by the rest of the code should be passed |
2b83a2a4 RM |
7840 | to this routine. If we are given anything else, or if other regex |
7841 | code generates an invalid error code, then the program has a bug. | |
7842 | Dump core so we can fix it. */ | |
7843 | abort (); | |
7844 | ||
c4563d2d | 7845 | msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]); |
2b83a2a4 RM |
7846 | |
7847 | msg_size = strlen (msg) + 1; /* Includes the null. */ | |
91c7b85d | 7848 | |
2b83a2a4 RM |
7849 | if (errbuf_size != 0) |
7850 | { | |
7851 | if (msg_size > errbuf_size) | |
7852 | { | |
86187531 UD |
7853 | #if defined HAVE_MEMPCPY || defined _LIBC |
7854 | *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0'; | |
7855 | #else | |
7856 | memcpy (errbuf, msg, errbuf_size - 1); | |
2b83a2a4 | 7857 | errbuf[errbuf_size - 1] = 0; |
86187531 | 7858 | #endif |
2b83a2a4 RM |
7859 | } |
7860 | else | |
86187531 | 7861 | memcpy (errbuf, msg, msg_size); |
2b83a2a4 RM |
7862 | } |
7863 | ||
7864 | return msg_size; | |
7865 | } | |
2ad4fab2 UD |
7866 | #ifdef _LIBC |
7867 | weak_alias (__regerror, regerror) | |
7868 | #endif | |
2b83a2a4 RM |
7869 | |
7870 | ||
7871 | /* Free dynamically allocated space used by PREG. */ | |
7872 | ||
7873 | void | |
7874 | regfree (preg) | |
7875 | regex_t *preg; | |
7876 | { | |
7877 | if (preg->buffer != NULL) | |
7878 | free (preg->buffer); | |
7879 | preg->buffer = NULL; | |
91c7b85d | 7880 | |
2b83a2a4 RM |
7881 | preg->allocated = 0; |
7882 | preg->used = 0; | |
7883 | ||
7884 | if (preg->fastmap != NULL) | |
7885 | free (preg->fastmap); | |
7886 | preg->fastmap = NULL; | |
7887 | preg->fastmap_accurate = 0; | |
7888 | ||
7889 | if (preg->translate != NULL) | |
7890 | free (preg->translate); | |
7891 | preg->translate = NULL; | |
7892 | } | |
2ad4fab2 UD |
7893 | #ifdef _LIBC |
7894 | weak_alias (__regfree, regfree) | |
7895 | #endif | |
2b83a2a4 RM |
7896 | |
7897 | #endif /* not emacs */ |