Mercurial > hg > xemacs-beta
annotate src/regex.c @ 5534:68db75473fc6
Merge.
| author | Mats Lidell <mats.lidell@cag.se> |
|---|---|
| date | Sun, 07 Aug 2011 23:22:59 +0200 |
| parents | 308d34e9f07d |
| children | 3f4a234f4672 |
| rev | line source |
|---|---|
| 428 | 1 /* Extended regular expression matching and search library, |
| 2 version 0.12, extended for XEmacs. | |
| 3 (Implements POSIX draft P10003.2/D11.2, except for | |
| 4 internationalization features.) | |
| 5 | |
| 6 Copyright (C) 1993, 1994, 1995 Free Software Foundation, Inc. | |
| 7 Copyright (C) 1995 Sun Microsystems, Inc. | |
| 5041 | 8 Copyright (C) 1995, 2001, 2002, 2003, 2010 Ben Wing. |
| 428 | 9 |
|
5402
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
10 This file is part of XEmacs. |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
11 |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
12 XEmacs is free software: you can redistribute it and/or modify it |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
13 under the terms of the GNU General Public License as published by the |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
14 Free Software Foundation, either version 3 of the License, or (at your |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
15 option) any later version. |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
16 |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
17 XEmacs is distributed in the hope that it will be useful, but WITHOUT |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
18 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
19 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
20 for more details. |
|
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
21 |
| 428 | 22 You should have received a copy of the GNU General Public License |
|
5402
308d34e9f07d
Changed bulk of GPLv2 or later files identified by script
Mats Lidell <matsl@xemacs.org>
parents:
5041
diff
changeset
|
23 along with XEmacs. If not, see <http://www.gnu.org/licenses/>. */ |
| 428 | 24 /* Synched up with: FSF 19.29. */ |
| 25 | |
| 26 #ifdef HAVE_CONFIG_H | |
| 27 #include <config.h> | |
| 28 #endif | |
| 29 | |
| 30 #ifndef _GNU_SOURCE | |
| 31 #define _GNU_SOURCE 1 | |
| 32 #endif | |
| 33 | |
| 34 /* We assume non-Mule if emacs isn't defined. */ | |
| 35 #ifndef emacs | |
| 36 #undef MULE | |
| 37 #endif | |
| 38 | |
| 771 | 39 /* XEmacs addition */ |
| 40 #ifdef REL_ALLOC | |
| 41 #define REGEX_REL_ALLOC /* may be undefined below */ | |
| 42 #endif | |
| 43 | |
| 428 | 44 /* XEmacs: define this to add in a speedup for patterns anchored at |
| 45 the beginning of a line. Keep the ifdefs so that it's easier to | |
| 46 tell where/why this code has diverged from v19. */ | |
| 47 #define REGEX_BEGLINE_CHECK | |
| 48 | |
| 49 /* XEmacs: the current mmap-based ralloc handles small blocks very | |
| 50 poorly, so we disable it here. */ | |
| 51 | |
| 771 | 52 #if defined (HAVE_MMAP) || defined (DOUG_LEA_MALLOC) |
| 53 # undef REGEX_REL_ALLOC | |
| 428 | 54 #endif |
| 55 | |
| 56 /* The `emacs' switch turns on certain matching commands | |
| 57 that make sense only in Emacs. */ | |
| 58 #ifdef emacs | |
| 59 | |
| 60 #include "lisp.h" | |
| 61 #include "buffer.h" | |
| 62 #include "syntax.h" | |
| 63 | |
| 64 #if (defined (DEBUG_XEMACS) && !defined (DEBUG)) | |
| 65 #define DEBUG | |
| 66 #endif | |
| 67 | |
| 867 | 68 #define RE_TRANSLATE_1(ch) TRT_TABLE_OF (translate, (Ichar) ch) |
| 446 | 69 #define TRANSLATE_P(tr) (!NILP (tr)) |
| 428 | 70 |
| 826 | 71 /* Converts the pointer to the char to BEG-based offset from the start. */ |
| 72 #define PTR_TO_OFFSET(d) (MATCHING_IN_FIRST_STRING \ | |
| 73 ? (d) - string1 : (d) - (string2 - size1)) | |
| 74 | |
| 428 | 75 #else /* not emacs */ |
| 76 | |
| 2367 | 77 #include <stdlib.h> |
| 78 #include <sys/types.h> | |
| 79 #include <stddef.h> /* needed for ptrdiff_t under Solaris */ | |
| 80 #include <string.h> | |
| 81 | |
| 2286 | 82 #include "compiler.h" /* Get compiler-specific definitions like UNUSED */ |
| 83 | |
| 2500 | 84 #define ABORT abort |
| 85 | |
| 428 | 86 /* If we are not linking with Emacs proper, |
| 87 we can't use the relocating allocator | |
| 88 even if config.h says that we can. */ | |
| 771 | 89 #undef REGEX_REL_ALLOC |
| 428 | 90 |
| 544 | 91 /* defined in lisp.h */ |
| 92 #ifdef REGEX_MALLOC | |
| 93 #ifndef DECLARE_NOTHING | |
| 94 #define DECLARE_NOTHING struct nosuchstruct | |
| 95 #endif | |
| 96 #endif | |
| 97 | |
| 867 | 98 #define itext_ichar(str) ((Ichar) (str)[0]) |
| 99 #define itext_ichar_fmt(str, fmt, object) ((Ichar) (str)[0]) | |
| 100 #define itext_ichar_ascii_fmt(str, fmt, object) ((Ichar) (str)[0]) | |
| 428 | 101 |
| 102 #if (LONGBITS > INTBITS) | |
| 103 # define EMACS_INT long | |
| 104 #else | |
| 105 # define EMACS_INT int | |
| 106 #endif | |
| 107 | |
| 867 | 108 typedef int Ichar; |
| 109 | |
| 110 #define INC_IBYTEPTR(p) ((p)++) | |
| 111 #define INC_IBYTEPTR_FMT(p, fmt) ((p)++) | |
| 112 #define DEC_IBYTEPTR(p) ((p)--) | |
| 113 #define DEC_IBYTEPTR_FMT(p, fmt) ((p)--) | |
|
4750
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
114 #define MAX_ICHAR_LEN 1 |
| 867 | 115 #define itext_ichar_len(ptr) 1 |
| 116 #define itext_ichar_len_fmt(ptr, fmt) 1 | |
| 428 | 117 |
| 118 /* Define the syntax stuff for \<, \>, etc. */ | |
| 119 | |
| 120 /* This must be nonzero for the wordchar and notwordchar pattern | |
| 121 commands in re_match_2. */ | |
| 122 #ifndef Sword | |
| 123 #define Sword 1 | |
| 124 #endif | |
| 125 | |
| 126 #ifdef SYNTAX_TABLE | |
| 127 | |
| 128 extern char *re_syntax_table; | |
| 129 | |
| 130 #else /* not SYNTAX_TABLE */ | |
| 131 | |
| 132 /* How many characters in the character set. */ | |
| 133 #define CHAR_SET_SIZE 256 | |
| 134 | |
| 135 static char re_syntax_table[CHAR_SET_SIZE]; | |
| 136 | |
| 137 static void | |
| 138 init_syntax_once (void) | |
| 139 { | |
| 140 static int done = 0; | |
| 141 | |
| 142 if (!done) | |
| 143 { | |
| 442 | 144 const char *word_syntax_chars = |
| 428 | 145 "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789_"; |
| 146 | |
| 147 memset (re_syntax_table, 0, sizeof (re_syntax_table)); | |
| 148 | |
| 149 while (*word_syntax_chars) | |
| 647 | 150 re_syntax_table[(unsigned int) (*word_syntax_chars++)] = Sword; |
| 428 | 151 |
| 152 done = 1; | |
| 153 } | |
| 154 } | |
| 155 | |
| 446 | 156 #endif /* SYNTAX_TABLE */ |
| 428 | 157 |
| 826 | 158 #define SYNTAX(ignored, c) re_syntax_table[c] |
| 460 | 159 #undef SYNTAX_FROM_CACHE |
| 826 | 160 #define SYNTAX_FROM_CACHE SYNTAX |
| 161 | |
| 162 #define RE_TRANSLATE_1(c) translate[(unsigned char) (c)] | |
| 446 | 163 #define TRANSLATE_P(tr) tr |
| 164 | |
| 165 #endif /* emacs */ | |
| 428 | 166 |
| 2201 | 167 /* This is for other GNU distributions with internationalized messages. */ |
| 168 #if defined (I18N3) && (defined (HAVE_LIBINTL_H) || defined (_LIBC)) | |
| 169 # include <libintl.h> | |
| 170 #else | |
| 171 # define gettext(msgid) (msgid) | |
| 172 #endif | |
| 173 | |
| 428 | 174 |
| 175 /* Get the interface, including the syntax bits. */ | |
| 176 #include "regex.h" | |
| 177 | |
| 178 /* isalpha etc. are used for the character classes. */ | |
| 179 #include <ctype.h> | |
| 180 | |
| 181 /* Jim Meyering writes: | |
| 182 | |
| 183 "... Some ctype macros are valid only for character codes that | |
| 184 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when | |
| 185 using /bin/cc or gcc but without giving an ansi option). So, all | |
| 186 ctype uses should be through macros like ISPRINT... If | |
| 187 STDC_HEADERS is defined, then autoconf has verified that the ctype | |
| 188 macros don't need to be guarded with references to isascii. ... | |
| 189 Defining isascii to 1 should let any compiler worth its salt | |
| 190 eliminate the && through constant folding." */ | |
| 191 | |
| 192 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII)) | |
| 193 #define ISASCII_1(c) 1 | |
| 194 #else | |
| 195 #define ISASCII_1(c) isascii(c) | |
| 196 #endif | |
| 197 | |
| 198 #ifdef MULE | |
| 199 /* The IS*() macros can be passed any character, including an extended | |
| 200 one. We need to make sure there are no crashes, which would occur | |
| 201 otherwise due to out-of-bounds array references. */ | |
| 202 #define ISASCII(c) (((EMACS_UINT) (c)) < 0x100 && ISASCII_1 (c)) | |
| 203 #else | |
| 204 #define ISASCII(c) ISASCII_1 (c) | |
| 205 #endif /* MULE */ | |
| 206 | |
| 207 #ifdef isblank | |
| 208 #define ISBLANK(c) (ISASCII (c) && isblank (c)) | |
| 209 #else | |
| 210 #define ISBLANK(c) ((c) == ' ' || (c) == '\t') | |
| 211 #endif | |
| 212 #ifdef isgraph | |
| 213 #define ISGRAPH(c) (ISASCII (c) && isgraph (c)) | |
| 214 #else | |
| 215 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) | |
| 216 #endif | |
| 217 | |
| 218 #define ISPRINT(c) (ISASCII (c) && isprint (c)) | |
| 219 #define ISDIGIT(c) (ISASCII (c) && isdigit (c)) | |
| 220 #define ISALNUM(c) (ISASCII (c) && isalnum (c)) | |
| 221 #define ISALPHA(c) (ISASCII (c) && isalpha (c)) | |
| 222 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) | |
| 223 #define ISLOWER(c) (ISASCII (c) && islower (c)) | |
| 224 #define ISPUNCT(c) (ISASCII (c) && ispunct (c)) | |
| 225 #define ISSPACE(c) (ISASCII (c) && isspace (c)) | |
| 226 #define ISUPPER(c) (ISASCII (c) && isupper (c)) | |
| 227 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) | |
| 228 | |
| 229 #ifndef NULL | |
| 230 #define NULL (void *)0 | |
| 231 #endif | |
| 232 | |
| 233 /* We remove any previous definition of `SIGN_EXTEND_CHAR', | |
| 234 since ours (we hope) works properly with all combinations of | |
| 235 machines, compilers, `char' and `unsigned char' argument types. | |
| 236 (Per Bothner suggested the basic approach.) */ | |
| 237 #undef SIGN_EXTEND_CHAR | |
| 238 #if __STDC__ | |
| 239 #define SIGN_EXTEND_CHAR(c) ((signed char) (c)) | |
| 240 #else /* not __STDC__ */ | |
| 241 /* As in Harbison and Steele. */ | |
| 242 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) | |
| 243 #endif | |
| 244 | |
| 245 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we | |
| 246 use `alloca' instead of `malloc'. This is because using malloc in | |
| 247 re_search* or re_match* could cause memory leaks when C-g is used in | |
| 248 Emacs; also, malloc is slower and causes storage fragmentation. On | |
| 249 the other hand, malloc is more portable, and easier to debug. | |
| 250 | |
| 251 Because we sometimes use alloca, some routines have to be macros, | |
| 252 not functions -- `alloca'-allocated space disappears at the end of the | |
| 253 function it is called in. */ | |
| 254 | |
| 1333 | 255 #ifndef emacs |
| 256 #define ALLOCA alloca | |
| 257 #define xmalloc malloc | |
| 258 #define xrealloc realloc | |
|
4976
16112448d484
Rename xfree(FOO, TYPE) -> xfree(FOO)
Ben Wing <ben@xemacs.org>
parents:
4832
diff
changeset
|
259 #define xfree free |
| 1333 | 260 #endif |
| 261 | |
| 262 #ifdef emacs | |
| 263 #define ALLOCA_GARBAGE_COLLECT() \ | |
| 264 do \ | |
| 265 { \ | |
| 266 if (need_to_check_c_alloca) \ | |
| 267 xemacs_c_alloca (0); \ | |
| 268 } while (0) | |
| 269 #elif defined (C_ALLOCA) | |
| 270 #define ALLOCA_GARBAGE_COLLECT() alloca (0) | |
| 271 #else | |
| 272 #define ALLOCA_GARBAGE_COLLECT() | |
| 273 #endif | |
| 274 | |
| 275 #ifndef emacs | |
| 276 /* So we can use just it to conditionalize on */ | |
| 277 #undef ERROR_CHECK_MALLOC | |
| 278 #endif | |
| 279 | |
| 280 #ifdef ERROR_CHECK_MALLOC | |
| 281 /* When REL_ALLOC, malloc() is problematic because it could potentially | |
| 282 cause all rel-alloc()ed data -- including buffer text -- to be relocated. | |
| 283 We deal with this by checking for such relocation whenever we have | |
| 284 executed a statement that may call malloc() -- or alloca(), which may | |
| 285 end up calling malloc() in some circumstances -- and recomputing all | |
| 286 of our string pointers in re_match_2_internal() and re_search_2(). | |
| 287 However, if malloc() or alloca() happens and we don't know about it, | |
| 288 we could still be screwed. So we set up a system where we indicate all | |
| 289 places where we are prepared for malloc() or alloca(), and in any | |
| 290 other circumstances, calls to those functions (from anywhere inside of | |
| 2500 | 291 XEmacs!) will ABORT(). We do this even when REL_ALLOC is not defined |
| 1333 | 292 so that we catch these problems sooner, since many developers and beta |
| 293 testers will not be running with REL_ALLOC. */ | |
| 294 int regex_malloc_disallowed; | |
| 295 #define BEGIN_REGEX_MALLOC_OK() regex_malloc_disallowed = 0 | |
| 296 #define END_REGEX_MALLOC_OK() regex_malloc_disallowed = 1 | |
| 297 #define UNBIND_REGEX_MALLOC_CHECK() unbind_to (depth) | |
| 298 #else | |
| 299 #define BEGIN_REGEX_MALLOC_OK() | |
| 300 #define END_REGEX_MALLOC_OK() | |
| 301 #define UNBIND_REGEX_MALLOC_CHECK() | |
| 302 #endif | |
| 303 | |
| 304 | |
| 428 | 305 #ifdef REGEX_MALLOC |
| 306 | |
| 1333 | 307 #define REGEX_ALLOCATE xmalloc |
| 308 #define REGEX_REALLOCATE(source, osize, nsize) xrealloc (source, nsize) | |
| 309 #define REGEX_FREE xfree | |
| 428 | 310 |
| 311 #else /* not REGEX_MALLOC */ | |
| 312 | |
| 313 /* Emacs already defines alloca, sometimes. */ | |
| 314 #ifndef alloca | |
| 315 | |
| 316 /* Make alloca work the best possible way. */ | |
| 317 #ifdef __GNUC__ | |
| 318 #define alloca __builtin_alloca | |
| 771 | 319 #elif defined (__DECC) /* XEmacs: added next 3 lines, similar to config.h.in */ |
| 320 #include <alloca.h> | |
| 321 #pragma intrinsic(alloca) | |
| 428 | 322 #else /* not __GNUC__ */ |
| 323 #if HAVE_ALLOCA_H | |
| 324 #include <alloca.h> | |
| 325 #else /* not __GNUC__ or HAVE_ALLOCA_H */ | |
| 326 #ifndef _AIX /* Already did AIX, up at the top. */ | |
| 444 | 327 void *alloca (); |
| 428 | 328 #endif /* not _AIX */ |
| 446 | 329 #endif /* HAVE_ALLOCA_H */ |
| 330 #endif /* __GNUC__ */ | |
| 428 | 331 |
| 332 #endif /* not alloca */ | |
| 333 | |
| 1333 | 334 #define REGEX_ALLOCATE ALLOCA |
| 428 | 335 |
| 2367 | 336 /* !!#### Needs review */ |
| 428 | 337 /* Assumes a `char *destination' variable. */ |
| 338 #define REGEX_REALLOCATE(source, osize, nsize) \ | |
| 1333 | 339 (destination = (char *) ALLOCA (nsize), \ |
| 428 | 340 memmove (destination, source, osize), \ |
| 341 destination) | |
| 342 | |
| 1726 | 343 /* No need to do anything to free, after alloca. |
| 344 Do nothing! But inhibit gcc warning. */ | |
| 345 #define REGEX_FREE(arg,type) ((void)0) | |
| 428 | 346 |
| 446 | 347 #endif /* REGEX_MALLOC */ |
| 428 | 348 |
| 349 /* Define how to allocate the failure stack. */ | |
| 350 | |
| 771 | 351 #ifdef REGEX_REL_ALLOC |
| 428 | 352 #define REGEX_ALLOCATE_STACK(size) \ |
| 1346 | 353 r_alloc ((unsigned char **) &failure_stack_ptr, (size)) |
| 428 | 354 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
| 1346 | 355 r_re_alloc ((unsigned char **) &failure_stack_ptr, (nsize)) |
| 428 | 356 #define REGEX_FREE_STACK(ptr) \ |
| 1346 | 357 r_alloc_free ((unsigned char **) &failure_stack_ptr) |
| 428 | 358 |
| 771 | 359 #else /* not REGEX_REL_ALLOC */ |
| 428 | 360 |
| 361 #ifdef REGEX_MALLOC | |
| 362 | |
| 1333 | 363 #define REGEX_ALLOCATE_STACK xmalloc |
| 364 #define REGEX_REALLOCATE_STACK(source, osize, nsize) xrealloc (source, nsize) | |
|
4976
16112448d484
Rename xfree(FOO, TYPE) -> xfree(FOO)
Ben Wing <ben@xemacs.org>
parents:
4832
diff
changeset
|
365 #define REGEX_FREE_STACK(arg) xfree (arg) |
| 428 | 366 |
| 367 #else /* not REGEX_MALLOC */ | |
| 368 | |
| 1333 | 369 #define REGEX_ALLOCATE_STACK ALLOCA |
| 428 | 370 |
| 371 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \ | |
| 372 REGEX_REALLOCATE (source, osize, nsize) | |
| 373 /* No need to explicitly free anything. */ | |
| 374 #define REGEX_FREE_STACK(arg) | |
| 375 | |
| 446 | 376 #endif /* REGEX_MALLOC */ |
| 771 | 377 #endif /* REGEX_REL_ALLOC */ |
| 428 | 378 |
| 379 | |
| 380 /* True if `size1' is non-NULL and PTR is pointing anywhere inside | |
| 381 `string1' or just past its end. This works if PTR is NULL, which is | |
| 382 a good thing. */ | |
| 383 #define FIRST_STRING_P(ptr) \ | |
| 384 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) | |
| 385 | |
| 386 /* (Re)Allocate N items of type T using malloc, or fail. */ | |
| 1333 | 387 #define TALLOC(n, t) ((t *) xmalloc ((n) * sizeof (t))) |
| 388 #define RETALLOC(addr, n, t) ((addr) = (t *) xrealloc (addr, (n) * sizeof (t))) | |
| 428 | 389 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) |
| 390 | |
| 391 #define BYTEWIDTH 8 /* In bits. */ | |
| 392 | |
| 434 | 393 #define STREQ(s1, s2) (strcmp (s1, s2) == 0) |
| 428 | 394 |
| 395 #undef MAX | |
| 396 #undef MIN | |
| 397 #define MAX(a, b) ((a) > (b) ? (a) : (b)) | |
| 398 #define MIN(a, b) ((a) < (b) ? (a) : (b)) | |
| 399 | |
| 446 | 400 /* Type of source-pattern and string chars. */ |
| 401 typedef const unsigned char re_char; | |
| 402 | |
| 460 | 403 typedef char re_bool; |
| 428 | 404 #define false 0 |
| 405 #define true 1 | |
| 406 | |
| 407 | |
| 1346 | 408 #ifdef emacs |
| 409 | |
| 410 #ifdef MULE | |
| 411 | |
| 412 Lisp_Object Vthe_lisp_rangetab; | |
| 413 | |
| 414 void | |
| 415 vars_of_regex (void) | |
| 416 { | |
| 2421 | 417 Vthe_lisp_rangetab = Fmake_range_table (Qstart_closed_end_closed); |
| 1346 | 418 staticpro (&Vthe_lisp_rangetab); |
| 419 } | |
| 420 | |
| 421 #else /* not MULE */ | |
| 422 | |
| 423 void | |
| 424 vars_of_regex (void) | |
| 425 { | |
| 426 } | |
| 427 | |
| 428 #endif /* MULE */ | |
| 429 | |
| 430 /* Convert an offset from the start of the logical text string formed by | |
| 431 concatenating the two strings together into a character position in the | |
| 432 Lisp buffer or string that the text represents. Knows that | |
| 433 when handling buffer text, the "string" we're passed in is always | |
| 434 BEGV - ZV. */ | |
| 435 | |
| 436 static Charxpos | |
| 437 offset_to_charxpos (Lisp_Object lispobj, int off) | |
| 438 { | |
| 439 if (STRINGP (lispobj)) | |
| 440 return string_index_byte_to_char (lispobj, off); | |
| 441 else if (BUFFERP (lispobj)) | |
| 442 return bytebpos_to_charbpos (XBUFFER (lispobj), | |
| 443 off + BYTE_BUF_BEGV (XBUFFER (lispobj))); | |
| 444 else | |
| 445 return 0; | |
| 446 } | |
| 447 | |
| 448 #ifdef REL_ALLOC | |
| 449 | |
| 450 /* STRING1 is the value of STRING1 given to re_match_2(). LISPOBJ is | |
| 451 the Lisp object (if any) from which the string is taken. If LISPOBJ | |
| 452 is a buffer, return a relocation offset to be added to all pointers to | |
| 453 string data so that they will be accurate again, after an allocation or | |
| 454 reallocation that potentially relocated the buffer data. | |
| 455 */ | |
| 456 static Bytecount | |
| 457 offset_post_relocation (Lisp_Object lispobj, Ibyte *orig_buftext) | |
| 458 { | |
| 459 if (!BUFFERP (lispobj)) | |
| 460 return 0; | |
| 461 return (BYTE_BUF_BYTE_ADDRESS (XBUFFER (lispobj), | |
| 462 BYTE_BUF_BEGV (XBUFFER (lispobj))) - | |
| 463 orig_buftext); | |
| 464 } | |
| 465 | |
| 466 #endif /* REL_ALLOC */ | |
| 467 | |
| 468 #ifdef ERROR_CHECK_MALLOC | |
| 469 | |
| 470 /* NOTE that this can run malloc() so you need to adjust afterwards. */ | |
| 471 | |
| 472 static int | |
| 473 bind_regex_malloc_disallowed (int value) | |
| 474 { | |
| 475 /* Tricky, because the act of binding can run malloc(). */ | |
| 476 int old_regex_malloc_disallowed = regex_malloc_disallowed; | |
| 477 int depth; | |
| 478 regex_malloc_disallowed = 0; | |
| 479 depth = record_unwind_protect_restoring_int (®ex_malloc_disallowed, | |
| 480 old_regex_malloc_disallowed); | |
| 481 regex_malloc_disallowed = value; | |
| 482 return depth; | |
| 483 } | |
| 484 | |
| 485 #endif /* ERROR_CHECK_MALLOC */ | |
| 486 | |
| 487 #endif /* emacs */ | |
| 488 | |
| 489 | |
| 428 | 490 /* These are the command codes that appear in compiled regular |
| 491 expressions. Some opcodes are followed by argument bytes. A | |
| 492 command code can specify any interpretation whatsoever for its | |
| 493 arguments. Zero bytes may appear in the compiled regular expression. */ | |
| 494 | |
| 495 typedef enum | |
| 496 { | |
| 497 no_op = 0, | |
| 498 | |
| 499 /* Succeed right away--no more backtracking. */ | |
| 500 succeed, | |
| 501 | |
| 502 /* Followed by one byte giving n, then by n literal bytes. */ | |
| 503 exactn, | |
| 504 | |
| 505 /* Matches any (more or less) character. */ | |
| 506 anychar, | |
| 507 | |
| 508 /* Matches any one char belonging to specified set. First | |
| 509 following byte is number of bitmap bytes. Then come bytes | |
| 510 for a bitmap saying which chars are in. Bits in each byte | |
| 511 are ordered low-bit-first. A character is in the set if its | |
| 512 bit is 1. A character too large to have a bit in the map is | |
| 513 automatically not in the set. */ | |
| 514 charset, | |
| 515 | |
| 516 /* Same parameters as charset, but match any character that is | |
| 517 not one of those specified. */ | |
| 518 charset_not, | |
| 519 | |
| 520 /* Start remembering the text that is matched, for storing in a | |
| 521 register. Followed by one byte with the register number, in | |
| 502 | 522 the range 1 to the pattern buffer's re_ngroups |
| 428 | 523 field. Then followed by one byte with the number of groups |
| 524 inner to this one. (This last has to be part of the | |
| 525 start_memory only because we need it in the on_failure_jump | |
| 526 of re_match_2.) */ | |
| 527 start_memory, | |
| 528 | |
| 529 /* Stop remembering the text that is matched and store it in a | |
| 530 memory register. Followed by one byte with the register | |
| 502 | 531 number, in the range 1 to `re_ngroups' in the |
| 428 | 532 pattern buffer, and one byte with the number of inner groups, |
| 533 just like `start_memory'. (We need the number of inner | |
| 534 groups here because we don't have any easy way of finding the | |
| 535 corresponding start_memory when we're at a stop_memory.) */ | |
| 536 stop_memory, | |
| 537 | |
| 538 /* Match a duplicate of something remembered. Followed by one | |
| 539 byte containing the register number. */ | |
| 540 duplicate, | |
| 541 | |
| 542 /* Fail unless at beginning of line. */ | |
| 543 begline, | |
| 544 | |
| 545 /* Fail unless at end of line. */ | |
| 546 endline, | |
| 547 | |
| 548 /* Succeeds if at beginning of buffer (if emacs) or at beginning | |
| 549 of string to be matched (if not). */ | |
| 550 begbuf, | |
| 551 | |
| 552 /* Analogously, for end of buffer/string. */ | |
| 553 endbuf, | |
| 554 | |
| 555 /* Followed by two byte relative address to which to jump. */ | |
| 556 jump, | |
| 557 | |
| 558 /* Same as jump, but marks the end of an alternative. */ | |
| 559 jump_past_alt, | |
| 560 | |
| 561 /* Followed by two-byte relative address of place to resume at | |
| 562 in case of failure. */ | |
| 563 on_failure_jump, | |
| 564 | |
| 565 /* Like on_failure_jump, but pushes a placeholder instead of the | |
| 566 current string position when executed. */ | |
| 567 on_failure_keep_string_jump, | |
| 568 | |
| 569 /* Throw away latest failure point and then jump to following | |
| 570 two-byte relative address. */ | |
| 571 pop_failure_jump, | |
| 572 | |
| 573 /* Change to pop_failure_jump if know won't have to backtrack to | |
| 574 match; otherwise change to jump. This is used to jump | |
| 575 back to the beginning of a repeat. If what follows this jump | |
| 576 clearly won't match what the repeat does, such that we can be | |
| 577 sure that there is no use backtracking out of repetitions | |
| 578 already matched, then we change it to a pop_failure_jump. | |
| 579 Followed by two-byte address. */ | |
| 580 maybe_pop_jump, | |
| 581 | |
| 582 /* Jump to following two-byte address, and push a dummy failure | |
| 583 point. This failure point will be thrown away if an attempt | |
| 584 is made to use it for a failure. A `+' construct makes this | |
| 585 before the first repeat. Also used as an intermediary kind | |
| 586 of jump when compiling an alternative. */ | |
| 587 dummy_failure_jump, | |
| 588 | |
| 589 /* Push a dummy failure point and continue. Used at the end of | |
| 590 alternatives. */ | |
| 591 push_dummy_failure, | |
| 592 | |
| 593 /* Followed by two-byte relative address and two-byte number n. | |
| 594 After matching N times, jump to the address upon failure. */ | |
| 595 succeed_n, | |
| 596 | |
| 597 /* Followed by two-byte relative address, and two-byte number n. | |
| 598 Jump to the address N times, then fail. */ | |
| 599 jump_n, | |
| 600 | |
| 601 /* Set the following two-byte relative address to the | |
| 602 subsequent two-byte number. The address *includes* the two | |
| 603 bytes of number. */ | |
| 604 set_number_at, | |
| 605 | |
| 606 wordchar, /* Matches any word-constituent character. */ | |
| 607 notwordchar, /* Matches any char that is not a word-constituent. */ | |
| 608 | |
| 609 wordbeg, /* Succeeds if at word beginning. */ | |
| 610 wordend, /* Succeeds if at word end. */ | |
| 611 | |
| 612 wordbound, /* Succeeds if at a word boundary. */ | |
| 613 notwordbound /* Succeeds if not at a word boundary. */ | |
| 614 | |
| 615 #ifdef emacs | |
| 616 ,before_dot, /* Succeeds if before point. */ | |
| 617 at_dot, /* Succeeds if at point. */ | |
| 618 after_dot, /* Succeeds if after point. */ | |
| 619 | |
| 620 /* Matches any character whose syntax is specified. Followed by | |
| 621 a byte which contains a syntax code, e.g., Sword. */ | |
| 622 syntaxspec, | |
| 623 | |
| 624 /* Matches any character whose syntax is not that specified. */ | |
| 625 notsyntaxspec | |
| 626 | |
| 627 #endif /* emacs */ | |
| 628 | |
| 629 #ifdef MULE | |
| 630 /* need extra stuff to be able to properly work with XEmacs/Mule | |
| 631 characters (which may take up more than one byte) */ | |
| 632 | |
| 633 ,charset_mule, /* Matches any character belonging to specified set. | |
| 634 The set is stored in "unified range-table | |
| 635 format"; see rangetab.c. Unlike the `charset' | |
| 636 opcode, this can handle arbitrary characters. */ | |
| 637 | |
| 638 charset_mule_not /* Same parameters as charset_mule, but match any | |
| 639 character that is not one of those specified. */ | |
| 640 | |
| 641 /* 97/2/17 jhod: The following two were merged back in from the Mule | |
| 642 2.3 code to enable some language specific processing */ | |
| 643 ,categoryspec, /* Matches entries in the character category tables */ | |
| 644 notcategoryspec /* The opposite of the above */ | |
| 645 #endif /* MULE */ | |
| 646 | |
| 647 } re_opcode_t; | |
| 648 | |
| 649 /* Common operations on the compiled pattern. */ | |
| 650 | |
| 651 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ | |
| 652 | |
| 653 #define STORE_NUMBER(destination, number) \ | |
| 654 do { \ | |
| 655 (destination)[0] = (number) & 0377; \ | |
| 656 (destination)[1] = (number) >> 8; \ | |
| 657 } while (0) | |
| 658 | |
| 659 /* Same as STORE_NUMBER, except increment DESTINATION to | |
| 660 the byte after where the number is stored. Therefore, DESTINATION | |
| 661 must be an lvalue. */ | |
| 662 | |
| 663 #define STORE_NUMBER_AND_INCR(destination, number) \ | |
| 664 do { \ | |
| 665 STORE_NUMBER (destination, number); \ | |
| 666 (destination) += 2; \ | |
| 667 } while (0) | |
| 668 | |
| 669 /* Put into DESTINATION a number stored in two contiguous bytes starting | |
| 670 at SOURCE. */ | |
| 671 | |
| 672 #define EXTRACT_NUMBER(destination, source) \ | |
| 673 do { \ | |
| 674 (destination) = *(source) & 0377; \ | |
| 675 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ | |
| 676 } while (0) | |
| 677 | |
| 678 #ifdef DEBUG | |
| 679 static void | |
| 446 | 680 extract_number (int *dest, re_char *source) |
| 428 | 681 { |
| 682 int temp = SIGN_EXTEND_CHAR (*(source + 1)); | |
| 683 *dest = *source & 0377; | |
| 684 *dest += temp << 8; | |
| 685 } | |
| 686 | |
| 687 #ifndef EXTRACT_MACROS /* To debug the macros. */ | |
| 688 #undef EXTRACT_NUMBER | |
| 689 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) | |
| 690 #endif /* not EXTRACT_MACROS */ | |
| 691 | |
| 692 #endif /* DEBUG */ | |
| 693 | |
| 694 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. | |
| 695 SOURCE must be an lvalue. */ | |
| 696 | |
| 697 #define EXTRACT_NUMBER_AND_INCR(destination, source) \ | |
| 698 do { \ | |
| 699 EXTRACT_NUMBER (destination, source); \ | |
| 700 (source) += 2; \ | |
| 701 } while (0) | |
| 702 | |
| 703 #ifdef DEBUG | |
| 704 static void | |
| 705 extract_number_and_incr (int *destination, unsigned char **source) | |
| 706 { | |
| 707 extract_number (destination, *source); | |
| 708 *source += 2; | |
| 709 } | |
| 710 | |
| 711 #ifndef EXTRACT_MACROS | |
| 712 #undef EXTRACT_NUMBER_AND_INCR | |
| 713 #define EXTRACT_NUMBER_AND_INCR(dest, src) \ | |
| 714 extract_number_and_incr (&dest, &src) | |
| 715 #endif /* not EXTRACT_MACROS */ | |
| 716 | |
| 717 #endif /* DEBUG */ | |
| 718 | |
| 719 /* If DEBUG is defined, Regex prints many voluminous messages about what | |
| 720 it is doing (if the variable `debug' is nonzero). If linked with the | |
| 721 main program in `iregex.c', you can enter patterns and strings | |
| 722 interactively. And if linked with the main program in `main.c' and | |
| 723 the other test files, you can run the already-written tests. */ | |
| 724 | |
| 725 #if defined (DEBUG) | |
| 726 | |
| 727 /* We use standard I/O for debugging. */ | |
| 728 #include <stdio.h> | |
| 729 | |
| 730 #ifndef emacs | |
| 731 /* XEmacs provides its own version of assert() */ | |
| 732 /* It is useful to test things that ``must'' be true when debugging. */ | |
| 733 #include <assert.h> | |
| 734 #endif | |
| 735 | |
| 5041 | 736 extern int debug_regexps; |
| 428 | 737 |
| 738 #define DEBUG_STATEMENT(e) e | |
| 5041 | 739 |
| 740 #define DEBUG_PRINT1(x) if (debug_regexps) printf (x) | |
| 741 #define DEBUG_PRINT2(x1, x2) if (debug_regexps) printf (x1, x2) | |
| 742 #define DEBUG_PRINT3(x1, x2, x3) if (debug_regexps) printf (x1, x2, x3) | |
| 743 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug_regexps) printf (x1, x2, x3, x4) | |
| 428 | 744 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
| 5041 | 745 if (debug_regexps) print_partial_compiled_pattern (s, e) |
| 428 | 746 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
| 5041 | 747 if (debug_regexps) print_double_string (w, s1, sz1, s2, sz2) |
| 748 | |
| 749 #define DEBUG_FAIL_PRINT1(x) \ | |
| 750 if (debug_regexps & RE_DEBUG_FAILURE_POINT) printf (x) | |
| 751 #define DEBUG_FAIL_PRINT2(x1, x2) \ | |
| 752 if (debug_regexps & RE_DEBUG_FAILURE_POINT) printf (x1, x2) | |
| 753 #define DEBUG_FAIL_PRINT3(x1, x2, x3) \ | |
| 754 if (debug_regexps & RE_DEBUG_FAILURE_POINT) printf (x1, x2, x3) | |
| 755 #define DEBUG_FAIL_PRINT4(x1, x2, x3, x4) \ | |
| 756 if (debug_regexps & RE_DEBUG_FAILURE_POINT) printf (x1, x2, x3, x4) | |
| 757 #define DEBUG_FAIL_PRINT_COMPILED_PATTERN(p, s, e) \ | |
| 758 if (debug_regexps & RE_DEBUG_FAILURE_POINT) \ | |
| 759 print_partial_compiled_pattern (s, e) | |
| 760 #define DEBUG_FAIL_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
| 761 if (debug_regexps & RE_DEBUG_FAILURE_POINT) \ | |
| 762 print_double_string (w, s1, sz1, s2, sz2) | |
| 763 | |
| 764 #define DEBUG_MATCH_PRINT1(x) \ | |
| 765 if (debug_regexps & RE_DEBUG_MATCHING) printf (x) | |
| 766 #define DEBUG_MATCH_PRINT2(x1, x2) \ | |
| 767 if (debug_regexps & RE_DEBUG_MATCHING) printf (x1, x2) | |
| 768 #define DEBUG_MATCH_PRINT3(x1, x2, x3) \ | |
| 769 if (debug_regexps & RE_DEBUG_MATCHING) printf (x1, x2, x3) | |
| 770 #define DEBUG_MATCH_PRINT4(x1, x2, x3, x4) \ | |
| 771 if (debug_regexps & RE_DEBUG_MATCHING) printf (x1, x2, x3, x4) | |
| 772 #define DEBUG_MATCH_PRINT_COMPILED_PATTERN(p, s, e) \ | |
| 773 if (debug_regexps & RE_DEBUG_MATCHING) \ | |
| 774 print_partial_compiled_pattern (s, e) | |
| 775 #define DEBUG_MATCH_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ | |
| 776 if (debug_regexps & RE_DEBUG_MATCHING) \ | |
| 777 print_double_string (w, s1, sz1, s2, sz2) | |
| 428 | 778 |
| 779 | |
| 780 /* Print the fastmap in human-readable form. */ | |
| 781 | |
| 782 static void | |
| 783 print_fastmap (char *fastmap) | |
| 784 { | |
| 647 | 785 int was_a_range = 0; |
| 786 int i = 0; | |
| 428 | 787 |
| 788 while (i < (1 << BYTEWIDTH)) | |
| 789 { | |
| 790 if (fastmap[i++]) | |
| 791 { | |
| 792 was_a_range = 0; | |
| 793 putchar (i - 1); | |
| 794 while (i < (1 << BYTEWIDTH) && fastmap[i]) | |
| 795 { | |
| 796 was_a_range = 1; | |
| 797 i++; | |
| 798 } | |
| 799 if (was_a_range) | |
| 800 { | |
| 801 putchar ('-'); | |
| 802 putchar (i - 1); | |
| 803 } | |
| 804 } | |
| 805 } | |
| 806 putchar ('\n'); | |
| 807 } | |
| 808 | |
| 809 | |
| 810 /* Print a compiled pattern string in human-readable form, starting at | |
| 811 the START pointer into it and ending just before the pointer END. */ | |
| 812 | |
| 813 static void | |
| 446 | 814 print_partial_compiled_pattern (re_char *start, re_char *end) |
| 428 | 815 { |
| 816 int mcnt, mcnt2; | |
| 446 | 817 unsigned char *p = (unsigned char *) start; |
| 818 re_char *pend = end; | |
| 428 | 819 |
| 820 if (start == NULL) | |
| 821 { | |
| 822 puts ("(null)"); | |
| 823 return; | |
| 824 } | |
| 825 | |
| 826 /* Loop over pattern commands. */ | |
| 827 while (p < pend) | |
| 828 { | |
| 829 printf ("%ld:\t", (long)(p - start)); | |
| 830 | |
| 831 switch ((re_opcode_t) *p++) | |
| 832 { | |
| 833 case no_op: | |
| 834 printf ("/no_op"); | |
| 835 break; | |
| 836 | |
| 837 case exactn: | |
| 838 mcnt = *p++; | |
| 839 printf ("/exactn/%d", mcnt); | |
|
4750
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
840 while (mcnt--) |
| 428 | 841 { |
|
4750
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
842 putchar ('/'); |
| 428 | 843 putchar (*p++); |
| 844 } | |
| 845 break; | |
| 846 | |
| 847 case start_memory: | |
| 848 mcnt = *p++; | |
| 849 printf ("/start_memory/%d/%d", mcnt, *p++); | |
| 850 break; | |
| 851 | |
| 852 case stop_memory: | |
| 853 mcnt = *p++; | |
| 854 printf ("/stop_memory/%d/%d", mcnt, *p++); | |
| 855 break; | |
| 856 | |
| 857 case duplicate: | |
| 858 printf ("/duplicate/%d", *p++); | |
| 859 break; | |
| 860 | |
| 861 case anychar: | |
| 862 printf ("/anychar"); | |
| 863 break; | |
| 864 | |
| 865 case charset: | |
| 866 case charset_not: | |
| 867 { | |
| 868 REGISTER int c, last = -100; | |
| 869 REGISTER int in_range = 0; | |
| 870 | |
| 871 printf ("/charset [%s", | |
| 872 (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); | |
| 873 | |
| 874 assert (p + *p < pend); | |
| 875 | |
| 876 for (c = 0; c < 256; c++) | |
| 877 if (((unsigned char) (c / 8) < *p) | |
| 878 && (p[1 + (c/8)] & (1 << (c % 8)))) | |
| 879 { | |
| 880 /* Are we starting a range? */ | |
| 881 if (last + 1 == c && ! in_range) | |
| 882 { | |
| 883 putchar ('-'); | |
| 884 in_range = 1; | |
| 885 } | |
| 886 /* Have we broken a range? */ | |
| 887 else if (last + 1 != c && in_range) | |
| 888 { | |
| 889 putchar (last); | |
| 890 in_range = 0; | |
| 891 } | |
| 892 | |
| 893 if (! in_range) | |
| 894 putchar (c); | |
| 895 | |
| 896 last = c; | |
| 897 } | |
| 898 | |
| 899 if (in_range) | |
| 900 putchar (last); | |
| 901 | |
| 902 putchar (']'); | |
| 903 | |
| 904 p += 1 + *p; | |
| 905 } | |
| 906 break; | |
| 907 | |
| 908 #ifdef MULE | |
| 909 case charset_mule: | |
| 910 case charset_mule_not: | |
| 911 { | |
| 912 int nentries, i; | |
| 913 | |
| 914 printf ("/charset_mule [%s", | |
| 915 (re_opcode_t) *(p - 1) == charset_mule_not ? "^" : ""); | |
| 916 nentries = unified_range_table_nentries (p); | |
| 917 for (i = 0; i < nentries; i++) | |
| 918 { | |
| 919 EMACS_INT first, last; | |
| 920 Lisp_Object dummy_val; | |
| 921 | |
| 922 unified_range_table_get_range (p, i, &first, &last, | |
| 923 &dummy_val); | |
| 924 if (first < 0x100) | |
| 925 putchar (first); | |
| 926 else | |
| 927 printf ("(0x%lx)", (long)first); | |
| 928 if (first != last) | |
| 929 { | |
| 930 putchar ('-'); | |
| 931 if (last < 0x100) | |
| 932 putchar (last); | |
| 933 else | |
| 934 printf ("(0x%lx)", (long)last); | |
| 935 } | |
| 936 } | |
| 937 putchar (']'); | |
| 938 p += unified_range_table_bytes_used (p); | |
| 939 } | |
| 940 break; | |
| 941 #endif | |
| 942 | |
| 943 case begline: | |
| 944 printf ("/begline"); | |
| 945 break; | |
| 946 | |
| 947 case endline: | |
| 948 printf ("/endline"); | |
| 949 break; | |
| 950 | |
| 951 case on_failure_jump: | |
| 952 extract_number_and_incr (&mcnt, &p); | |
| 953 printf ("/on_failure_jump to %ld", (long)(p + mcnt - start)); | |
| 954 break; | |
| 955 | |
| 956 case on_failure_keep_string_jump: | |
| 957 extract_number_and_incr (&mcnt, &p); | |
| 958 printf ("/on_failure_keep_string_jump to %ld", (long)(p + mcnt - start)); | |
| 959 break; | |
| 960 | |
| 961 case dummy_failure_jump: | |
| 962 extract_number_and_incr (&mcnt, &p); | |
| 963 printf ("/dummy_failure_jump to %ld", (long)(p + mcnt - start)); | |
| 964 break; | |
| 965 | |
| 966 case push_dummy_failure: | |
| 967 printf ("/push_dummy_failure"); | |
| 968 break; | |
| 969 | |
| 970 case maybe_pop_jump: | |
| 971 extract_number_and_incr (&mcnt, &p); | |
| 972 printf ("/maybe_pop_jump to %ld", (long)(p + mcnt - start)); | |
| 973 break; | |
| 974 | |
| 975 case pop_failure_jump: | |
| 976 extract_number_and_incr (&mcnt, &p); | |
| 977 printf ("/pop_failure_jump to %ld", (long)(p + mcnt - start)); | |
| 978 break; | |
| 979 | |
| 980 case jump_past_alt: | |
| 981 extract_number_and_incr (&mcnt, &p); | |
| 982 printf ("/jump_past_alt to %ld", (long)(p + mcnt - start)); | |
| 983 break; | |
| 984 | |
| 985 case jump: | |
| 986 extract_number_and_incr (&mcnt, &p); | |
| 987 printf ("/jump to %ld", (long)(p + mcnt - start)); | |
| 988 break; | |
| 989 | |
| 990 case succeed_n: | |
| 991 extract_number_and_incr (&mcnt, &p); | |
| 992 extract_number_and_incr (&mcnt2, &p); | |
| 993 printf ("/succeed_n to %ld, %d times", (long)(p + mcnt - start), mcnt2); | |
| 994 break; | |
| 995 | |
| 996 case jump_n: | |
| 997 extract_number_and_incr (&mcnt, &p); | |
| 998 extract_number_and_incr (&mcnt2, &p); | |
| 999 printf ("/jump_n to %ld, %d times", (long)(p + mcnt - start), mcnt2); | |
| 1000 break; | |
| 1001 | |
| 1002 case set_number_at: | |
| 1003 extract_number_and_incr (&mcnt, &p); | |
| 1004 extract_number_and_incr (&mcnt2, &p); | |
| 1005 printf ("/set_number_at location %ld to %d", (long)(p + mcnt - start), mcnt2); | |
| 1006 break; | |
| 1007 | |
| 1008 case wordbound: | |
| 1009 printf ("/wordbound"); | |
| 1010 break; | |
| 1011 | |
| 1012 case notwordbound: | |
| 1013 printf ("/notwordbound"); | |
| 1014 break; | |
| 1015 | |
| 1016 case wordbeg: | |
| 1017 printf ("/wordbeg"); | |
| 1018 break; | |
| 1019 | |
| 1020 case wordend: | |
| 1021 printf ("/wordend"); | |
| 1022 | |
| 1023 #ifdef emacs | |
| 1024 case before_dot: | |
| 1025 printf ("/before_dot"); | |
| 1026 break; | |
| 1027 | |
| 1028 case at_dot: | |
| 1029 printf ("/at_dot"); | |
| 1030 break; | |
| 1031 | |
| 1032 case after_dot: | |
| 1033 printf ("/after_dot"); | |
| 1034 break; | |
| 1035 | |
| 1036 case syntaxspec: | |
| 1037 printf ("/syntaxspec"); | |
| 1038 mcnt = *p++; | |
| 1039 printf ("/%d", mcnt); | |
| 1040 break; | |
| 1041 | |
| 1042 case notsyntaxspec: | |
| 1043 printf ("/notsyntaxspec"); | |
| 1044 mcnt = *p++; | |
| 1045 printf ("/%d", mcnt); | |
| 1046 break; | |
| 1047 | |
| 1048 #ifdef MULE | |
| 1049 /* 97/2/17 jhod Mule category patch */ | |
| 1050 case categoryspec: | |
| 1051 printf ("/categoryspec"); | |
| 1052 mcnt = *p++; | |
| 1053 printf ("/%d", mcnt); | |
| 1054 break; | |
| 1055 | |
| 1056 case notcategoryspec: | |
| 1057 printf ("/notcategoryspec"); | |
| 1058 mcnt = *p++; | |
| 1059 printf ("/%d", mcnt); | |
| 1060 break; | |
| 1061 /* end of category patch */ | |
| 1062 #endif /* MULE */ | |
| 1063 #endif /* emacs */ | |
| 1064 | |
| 1065 case wordchar: | |
| 1066 printf ("/wordchar"); | |
| 1067 break; | |
| 1068 | |
| 1069 case notwordchar: | |
| 1070 printf ("/notwordchar"); | |
| 1071 break; | |
| 1072 | |
| 1073 case begbuf: | |
| 1074 printf ("/begbuf"); | |
| 1075 break; | |
| 1076 | |
| 1077 case endbuf: | |
| 1078 printf ("/endbuf"); | |
| 1079 break; | |
| 1080 | |
| 1081 default: | |
| 1082 printf ("?%d", *(p-1)); | |
| 1083 } | |
| 1084 | |
| 1085 putchar ('\n'); | |
| 1086 } | |
| 1087 | |
| 1088 printf ("%ld:\tend of pattern.\n", (long)(p - start)); | |
| 1089 } | |
| 1090 | |
| 1091 | |
| 1092 static void | |
| 1093 print_compiled_pattern (struct re_pattern_buffer *bufp) | |
| 1094 { | |
| 446 | 1095 re_char *buffer = bufp->buffer; |
| 428 | 1096 |
| 1097 print_partial_compiled_pattern (buffer, buffer + bufp->used); | |
| 1098 printf ("%ld bytes used/%ld bytes allocated.\n", bufp->used, | |
| 1099 bufp->allocated); | |
| 1100 | |
| 1101 if (bufp->fastmap_accurate && bufp->fastmap) | |
| 1102 { | |
| 1103 printf ("fastmap: "); | |
| 1104 print_fastmap (bufp->fastmap); | |
| 1105 } | |
| 1106 | |
| 1107 printf ("re_nsub: %ld\t", (long)bufp->re_nsub); | |
| 502 | 1108 printf ("re_ngroups: %ld\t", (long)bufp->re_ngroups); |
| 428 | 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); | |
| 1115 printf ("syntax: %d\n", bufp->syntax); | |
| 1116 /* Perhaps we should print the translate table? */ | |
| 1117 /* and maybe the category table? */ | |
| 502 | 1118 |
| 1119 if (bufp->external_to_internal_register) | |
| 1120 { | |
| 1121 int i; | |
| 1122 | |
| 1123 printf ("external_to_internal_register:\n"); | |
| 1124 for (i = 0; i <= bufp->re_nsub; i++) | |
| 1125 { | |
| 1126 if (i > 0) | |
| 1127 printf (", "); | |
| 1128 printf ("%d -> %d", i, bufp->external_to_internal_register[i]); | |
| 1129 } | |
| 1130 printf ("\n"); | |
| 1131 } | |
| 428 | 1132 } |
| 1133 | |
| 1134 | |
| 1135 static void | |
| 446 | 1136 print_double_string (re_char *where, re_char *string1, int size1, |
| 1137 re_char *string2, int size2) | |
| 428 | 1138 { |
| 1139 if (where == NULL) | |
| 1140 printf ("(null)"); | |
| 1141 else | |
| 1142 { | |
| 647 | 1143 int this_char; |
| 428 | 1144 |
| 1145 if (FIRST_STRING_P (where)) | |
| 1146 { | |
| 1147 for (this_char = where - string1; this_char < size1; this_char++) | |
| 1148 putchar (string1[this_char]); | |
| 1149 | |
| 1150 where = string2; | |
| 1151 } | |
| 1152 | |
| 1153 for (this_char = where - string2; this_char < size2; this_char++) | |
| 1154 putchar (string2[this_char]); | |
| 1155 } | |
| 1156 } | |
| 1157 | |
| 1158 #else /* not DEBUG */ | |
| 1159 | |
| 771 | 1160 #ifndef emacs |
| 428 | 1161 #undef assert |
| 771 | 1162 #define assert(e) ((void) (1)) |
| 1163 #endif | |
| 428 | 1164 |
| 1165 #define DEBUG_STATEMENT(e) | |
| 5041 | 1166 |
| 428 | 1167 #define DEBUG_PRINT1(x) |
| 1168 #define DEBUG_PRINT2(x1, x2) | |
| 1169 #define DEBUG_PRINT3(x1, x2, x3) | |
| 1170 #define DEBUG_PRINT4(x1, x2, x3, x4) | |
| 1171 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) | |
| 1172 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
| 1173 | |
| 5041 | 1174 #define DEBUG_FAIL_PRINT1(x) |
| 1175 #define DEBUG_FAIL_PRINT2(x1, x2) | |
| 1176 #define DEBUG_FAIL_PRINT3(x1, x2, x3) | |
| 1177 #define DEBUG_FAIL_PRINT4(x1, x2, x3, x4) | |
| 1178 #define DEBUG_FAIL_PRINT_COMPILED_PATTERN(p, s, e) | |
| 1179 #define DEBUG_FAIL_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
| 1180 | |
| 1181 #define DEBUG_MATCH_PRINT1(x) | |
| 1182 #define DEBUG_MATCH_PRINT2(x1, x2) | |
| 1183 #define DEBUG_MATCH_PRINT3(x1, x2, x3) | |
| 1184 #define DEBUG_MATCH_PRINT4(x1, x2, x3, x4) | |
| 1185 #define DEBUG_MATCH_PRINT_COMPILED_PATTERN(p, s, e) | |
| 1186 #define DEBUG_MATCH_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) | |
| 1187 | |
| 446 | 1188 #endif /* DEBUG */ |
| 428 | 1189 |
| 1190 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can | |
| 1191 also be assigned to arbitrarily: each pattern buffer stores its own | |
| 1192 syntax, so it can be changed between regex compilations. */ | |
| 1193 /* This has no initializer because initialized variables in Emacs | |
| 1194 become read-only after dumping. */ | |
| 1195 reg_syntax_t re_syntax_options; | |
| 1196 | |
| 1197 | |
| 1198 /* Specify the precise syntax of regexps for compilation. This provides | |
| 1199 for compatibility for various utilities which historically have | |
| 1200 different, incompatible syntaxes. | |
| 1201 | |
| 1202 The argument SYNTAX is a bit mask comprised of the various bits | |
| 1203 defined in regex.h. We return the old syntax. */ | |
| 1204 | |
| 1205 reg_syntax_t | |
| 1206 re_set_syntax (reg_syntax_t syntax) | |
| 1207 { | |
| 1208 reg_syntax_t ret = re_syntax_options; | |
| 1209 | |
| 1210 re_syntax_options = syntax; | |
| 1211 return ret; | |
| 1212 } | |
| 1213 | |
| 1214 /* This table gives an error message for each of the error codes listed | |
| 1215 in regex.h. Obviously the order here has to be same as there. | |
| 1216 POSIX doesn't require that we do anything for REG_NOERROR, | |
| 1217 but why not be nice? */ | |
| 1218 | |
| 442 | 1219 static const char *re_error_msgid[] = |
| 428 | 1220 { |
| 1221 "Success", /* REG_NOERROR */ | |
| 1222 "No match", /* REG_NOMATCH */ | |
| 1223 "Invalid regular expression", /* REG_BADPAT */ | |
| 1224 "Invalid collation character", /* REG_ECOLLATE */ | |
| 1225 "Invalid character class name", /* REG_ECTYPE */ | |
| 1226 "Trailing backslash", /* REG_EESCAPE */ | |
| 1227 "Invalid back reference", /* REG_ESUBREG */ | |
| 1228 "Unmatched [ or [^", /* REG_EBRACK */ | |
| 1229 "Unmatched ( or \\(", /* REG_EPAREN */ | |
| 1230 "Unmatched \\{", /* REG_EBRACE */ | |
| 1231 "Invalid content of \\{\\}", /* REG_BADBR */ | |
| 1232 "Invalid range end", /* REG_ERANGE */ | |
| 1233 "Memory exhausted", /* REG_ESPACE */ | |
| 1234 "Invalid preceding regular expression", /* REG_BADRPT */ | |
| 1235 "Premature end of regular expression", /* REG_EEND */ | |
| 1236 "Regular expression too big", /* REG_ESIZE */ | |
| 1237 "Unmatched ) or \\)", /* REG_ERPAREN */ | |
| 1238 #ifdef emacs | |
| 1239 "Invalid syntax designator", /* REG_ESYNTAX */ | |
| 1240 #endif | |
| 1241 #ifdef MULE | |
| 1242 "Ranges may not span charsets", /* REG_ERANGESPAN */ | |
| 1243 "Invalid category designator", /* REG_ECATEGORY */ | |
| 1244 #endif | |
| 1245 }; | |
| 1246 | |
| 1247 /* Avoiding alloca during matching, to placate r_alloc. */ | |
| 1248 | |
| 1333 | 1249 /* About these various flags: |
| 1250 | |
| 1251 MATCH_MAY_ALLOCATE indicates that it's OK to do allocation in the | |
| 1252 searching and matching functions. In this case, we use local variables | |
| 1253 to hold the values allocated. If not, we use *global* variables, which | |
| 1254 are pre-allocated. NOTE: XEmacs ***MUST*** run with MATCH_MAY_ALLOCATE, | |
| 1255 because the regexp routines may get called reentrantly as a result of | |
| 1256 QUIT processing (e.g. under Windows: re_match -> QUIT -> quit_p -> drain | |
| 1257 events -> process WM_INITMENU -> call filter -> re_match; see stack | |
| 1258 trace in signal.c), so we cannot have any global variables (unless we do | |
| 1259 lots of trickiness including some unwind-protects, which isn't worth it | |
| 1260 at this point). | |
| 1261 | |
| 1262 REL_ALLOC means that the relocating allocator is in use, for buffers | |
| 1263 and such. REGEX_REL_ALLOC means that we use rel-alloc to manage the | |
| 1264 fail stack, which may grow quite large. REGEX_MALLOC means we use | |
| 1265 malloc() in place of alloca() to allocate the fail stack -- only | |
| 1266 applicable if REGEX_REL_ALLOC is not defined. | |
| 1267 */ | |
| 1268 | |
| 428 | 1269 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the |
| 1270 searching and matching functions should not call alloca. On some | |
| 1271 systems, alloca is implemented in terms of malloc, and if we're | |
| 1272 using the relocating allocator routines, then malloc could cause a | |
| 1273 relocation, which might (if the strings being searched are in the | |
| 1274 ralloc heap) shift the data out from underneath the regexp | |
| 771 | 1275 routines. [To clarify: The purpose of rel-alloc is to allow data to |
| 1276 be moved in memory from one place to another so that all data | |
| 1277 blocks can be consolidated together and excess memory released back | |
| 1278 to the operating system. This requires that all the blocks that | |
| 1279 are managed by rel-alloc go at the very end of the program's heap, | |
| 1280 after all regularly malloc()ed data. malloc(), however, is used to | |
| 1281 owning the end of the heap, so that when more memory is needed, it | |
| 1282 just expands the heap using sbrk(). This is reconciled by using a | |
| 1283 malloc() (such as malloc.c, gmalloc.c, or recent versions of | |
| 1284 malloc() in libc) where the sbrk() call can be replaced with a | |
| 1285 user-specified call -- in this case, to rel-alloc's r_alloc_sbrk() | |
| 1286 routine. This routine calls the real sbrk(), but then shifts all | |
| 1287 the rel-alloc-managed blocks forward to the end of the heap again, | |
| 1288 so that malloc() gets the memory it needs in the location it needs | |
| 1289 it at. The regex routines may well have pointers to buffer data as | |
| 1290 their arguments, and buffers are managed by rel-alloc if rel-alloc | |
| 1291 has been enabled, so calling malloc() may potentially screw things | |
| 1292 up badly if it runs out of space and asks for more from the OS.] | |
| 1293 | |
| 1294 [[Here's another reason to avoid allocation: Emacs processes input | |
| 1295 from X in a signal handler; processing X input may call malloc; if | |
| 1296 input arrives while a matching routine is calling malloc, then | |
| 1297 we're scrod. But Emacs can't just block input while calling | |
| 1298 matching routines; then we don't notice interrupts when they come | |
| 1299 in. So, Emacs blocks input around all regexp calls except the | |
| 1300 matching calls, which it leaves unprotected, in the faith that they | |
| 1333 | 1301 will not malloc.]] This previous paragraph is irrelevant under XEmacs, |
| 1302 as we *do not* do anything so stupid as process input from within a | |
| 1303 signal handler. | |
| 1304 | |
| 1305 However, the regexp routines may get called reentrantly as a result of | |
| 1306 QUIT processing (e.g. under Windows: re_match -> QUIT -> quit_p -> drain | |
| 1307 events -> process WM_INITMENU -> call filter -> re_match; see stack | |
| 1308 trace in signal.c), so we cannot have any global variables (unless we do | |
| 1309 lots of trickiness including some unwind-protects, which isn't worth it | |
| 1310 at this point). Hence we MUST have MATCH_MAY_ALLOCATE defined. | |
| 1311 | |
| 1312 Also, the first paragraph does not make complete sense to me -- what | |
| 1313 about the use of rel-alloc to handle the fail stacks? Shouldn't these | |
| 1314 reallocations potentially cause buffer data to be relocated as well? I | |
| 826 | 1315 must be missing something, though -- perhaps the writer above is |
| 1316 assuming that the failure stack(s) will always be allocated after the | |
| 1317 buffer data, and thus reallocating them with rel-alloc won't move buffer | |
| 1333 | 1318 data. (In fact, a cursory glance at the code in ralloc.c seems to |
| 1319 confirm this.) --ben */ | |
| 428 | 1320 |
| 1321 /* Normally, this is fine. */ | |
| 1322 #define MATCH_MAY_ALLOCATE | |
| 1323 | |
| 1324 /* When using GNU C, we are not REALLY using the C alloca, no matter | |
| 1325 what config.h may say. So don't take precautions for it. */ | |
| 1326 #ifdef __GNUC__ | |
| 1327 #undef C_ALLOCA | |
| 1328 #endif | |
| 1329 | |
| 1330 /* The match routines may not allocate if (1) they would do it with malloc | |
| 1331 and (2) it's not safe for them to use malloc. | |
| 1332 Note that if REL_ALLOC is defined, matching would not use malloc for the | |
| 1333 failure stack, but we would still use it for the register vectors; | |
| 1334 so REL_ALLOC should not affect this. */ | |
| 771 | 1335 |
| 1333 | 1336 /* XEmacs can handle REL_ALLOC and malloc() OK */ |
| 1337 #if !defined (emacs) && (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (REL_ALLOC) | |
| 428 | 1338 #undef MATCH_MAY_ALLOCATE |
| 1339 #endif | |
| 1340 | |
| 1333 | 1341 #if !defined (MATCH_MAY_ALLOCATE) && defined (emacs) |
| 771 | 1342 #error regex must be handle reentrancy; MATCH_MAY_ALLOCATE must be defined |
| 1343 #endif | |
| 1344 | |
| 428 | 1345 |
| 1346 /* Failure stack declarations and macros; both re_compile_fastmap and | |
| 1347 re_match_2 use a failure stack. These have to be macros because of | |
| 1348 REGEX_ALLOCATE_STACK. */ | |
| 1349 | |
| 1350 | |
| 1351 /* Number of failure points for which to initially allocate space | |
| 1352 when matching. If this number is exceeded, we allocate more | |
| 1353 space, so it is not a hard limit. */ | |
| 1354 #ifndef INIT_FAILURE_ALLOC | |
| 3300 | 1355 #define INIT_FAILURE_ALLOC 20 |
| 428 | 1356 #endif |
| 1357 | |
| 1358 /* Roughly the maximum number of failure points on the stack. Would be | |
| 1359 exactly that if always used MAX_FAILURE_SPACE each time we failed. | |
| 1360 This is a variable only so users of regex can assign to it; we never | |
| 1361 change it ourselves. */ | |
| 1362 #if defined (MATCH_MAY_ALLOCATE) | |
| 1363 /* 4400 was enough to cause a crash on Alpha OSF/1, | |
| 1364 whose default stack limit is 2mb. */ | |
| 3300 | 1365 int re_max_failures = 40000; |
| 428 | 1366 #else |
| 3300 | 1367 int re_max_failures = 4000; |
| 428 | 1368 #endif |
| 1369 | |
| 1370 union fail_stack_elt | |
| 1371 { | |
| 446 | 1372 re_char *pointer; |
| 428 | 1373 int integer; |
| 1374 }; | |
| 1375 | |
| 1376 typedef union fail_stack_elt fail_stack_elt_t; | |
| 1377 | |
| 1378 typedef struct | |
| 1379 { | |
| 1380 fail_stack_elt_t *stack; | |
| 665 | 1381 Elemcount size; |
| 1382 Elemcount avail; /* Offset of next open position. */ | |
| 428 | 1383 } fail_stack_type; |
| 1384 | |
| 1385 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0) | |
| 1386 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) | |
| 1387 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) | |
| 1388 | |
| 1389 | |
| 1390 /* Define macros to initialize and free the failure stack. | |
| 1391 Do `return -2' if the alloc fails. */ | |
| 1392 | |
| 1393 #ifdef MATCH_MAY_ALLOCATE | |
| 1333 | 1394 #define INIT_FAIL_STACK() \ |
| 1395 do { \ | |
| 1396 fail_stack.stack = (fail_stack_elt_t *) \ | |
| 1397 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * \ | |
| 1398 sizeof (fail_stack_elt_t)); \ | |
| 1399 \ | |
| 1400 if (fail_stack.stack == NULL) \ | |
| 1401 { \ | |
| 1402 UNBIND_REGEX_MALLOC_CHECK (); \ | |
| 1403 return -2; \ | |
| 1404 } \ | |
| 1405 \ | |
| 1406 fail_stack.size = INIT_FAILURE_ALLOC; \ | |
| 1407 fail_stack.avail = 0; \ | |
| 428 | 1408 } while (0) |
| 1409 | |
| 1410 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) | |
| 1411 #else | |
| 1412 #define INIT_FAIL_STACK() \ | |
| 1413 do { \ | |
| 1414 fail_stack.avail = 0; \ | |
| 1415 } while (0) | |
| 1416 | |
| 1417 #define RESET_FAIL_STACK() | |
| 1418 #endif | |
| 1419 | |
| 1420 | |
| 1421 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. | |
| 1422 | |
| 1423 Return 1 if succeeds, and 0 if either ran out of memory | |
| 1424 allocating space for it or it was already too large. | |
| 1425 | |
| 1426 REGEX_REALLOCATE_STACK requires `destination' be declared. */ | |
| 1427 | |
| 1428 #define DOUBLE_FAIL_STACK(fail_stack) \ | |
| 1429 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \ | |
| 1430 ? 0 \ | |
| 1431 : ((fail_stack).stack = (fail_stack_elt_t *) \ | |
| 1432 REGEX_REALLOCATE_STACK ((fail_stack).stack, \ | |
| 1433 (fail_stack).size * sizeof (fail_stack_elt_t), \ | |
| 1434 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ | |
| 1435 \ | |
| 1436 (fail_stack).stack == NULL \ | |
| 1437 ? 0 \ | |
| 1438 : ((fail_stack).size <<= 1, \ | |
| 1439 1))) | |
| 1440 | |
| 1333 | 1441 #if !defined (emacs) || !defined (REL_ALLOC) |
| 1442 #define RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS() | |
| 1443 #else | |
| 1444 /* Don't change NULL pointers */ | |
| 1445 #define ADD_IF_NZ(val) if (val) val += rmdp_offset | |
| 1346 | 1446 #define RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS() \ |
| 1447 do \ | |
| 1448 { \ | |
| 1449 Bytecount rmdp_offset = offset_post_relocation (lispobj, orig_buftext); \ | |
| 1450 \ | |
| 1451 if (rmdp_offset) \ | |
| 1452 { \ | |
| 1453 int i; \ | |
| 1454 \ | |
| 1455 ADD_IF_NZ (string1); \ | |
| 1456 ADD_IF_NZ (string2); \ | |
| 1457 ADD_IF_NZ (d); \ | |
| 1458 ADD_IF_NZ (dend); \ | |
| 1459 ADD_IF_NZ (end1); \ | |
| 1460 ADD_IF_NZ (end2); \ | |
| 1461 ADD_IF_NZ (end_match_1); \ | |
| 1462 ADD_IF_NZ (end_match_2); \ | |
| 1463 \ | |
| 1464 if (bufp->re_ngroups) \ | |
| 1465 { \ | |
| 1466 for (i = 0; i < num_regs; i++) \ | |
| 1467 { \ | |
| 1468 ADD_IF_NZ (regstart[i]); \ | |
| 1469 ADD_IF_NZ (regend[i]); \ | |
| 1470 ADD_IF_NZ (old_regstart[i]); \ | |
| 1471 ADD_IF_NZ (old_regend[i]); \ | |
| 1472 ADD_IF_NZ (best_regstart[i]); \ | |
| 1473 ADD_IF_NZ (best_regend[i]); \ | |
| 1474 ADD_IF_NZ (reg_dummy[i]); \ | |
| 1475 } \ | |
| 1476 } \ | |
| 1477 \ | |
| 1478 ADD_IF_NZ (match_end); \ | |
| 1479 } \ | |
| 1333 | 1480 } while (0) |
| 1481 #endif /* !defined (emacs) || !defined (REL_ALLOC) */ | |
| 1482 | |
| 1483 #if !defined (emacs) || !defined (REL_ALLOC) | |
| 1484 #define RE_SEARCH_RELOCATE_MOVEABLE_DATA_POINTERS() | |
| 1485 #else | |
| 1346 | 1486 #define RE_SEARCH_RELOCATE_MOVEABLE_DATA_POINTERS() \ |
| 1487 do \ | |
| 1488 { \ | |
| 1489 Bytecount rmdp_offset = offset_post_relocation (lispobj, orig_buftext); \ | |
| 1490 \ | |
| 1491 if (rmdp_offset) \ | |
| 1492 { \ | |
| 1493 ADD_IF_NZ (str1); \ | |
| 1494 ADD_IF_NZ (str2); \ | |
| 1495 ADD_IF_NZ (string1); \ | |
| 1496 ADD_IF_NZ (string2); \ | |
| 1497 ADD_IF_NZ (d); \ | |
| 1498 } \ | |
| 1333 | 1499 } while (0) |
| 1500 | |
| 1501 #endif /* emacs */ | |
| 428 | 1502 |
| 1503 /* Push pointer POINTER on FAIL_STACK. | |
| 1504 Return 1 if was able to do so and 0 if ran out of memory allocating | |
| 1505 space to do so. */ | |
| 1506 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ | |
| 1507 ((FAIL_STACK_FULL () \ | |
| 1508 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \ | |
| 1509 ? 0 \ | |
| 1510 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ | |
| 1511 1)) | |
| 1512 | |
| 1513 /* Push a pointer value onto the failure stack. | |
| 1514 Assumes the variable `fail_stack'. Probably should only | |
| 1515 be called from within `PUSH_FAILURE_POINT'. */ | |
| 1516 #define PUSH_FAILURE_POINTER(item) \ | |
| 1517 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item) | |
| 1518 | |
| 1519 /* This pushes an integer-valued item onto the failure stack. | |
| 1520 Assumes the variable `fail_stack'. Probably should only | |
| 1521 be called from within `PUSH_FAILURE_POINT'. */ | |
| 1522 #define PUSH_FAILURE_INT(item) \ | |
| 1523 fail_stack.stack[fail_stack.avail++].integer = (item) | |
| 1524 | |
| 1525 /* Push a fail_stack_elt_t value onto the failure stack. | |
| 1526 Assumes the variable `fail_stack'. Probably should only | |
| 1527 be called from within `PUSH_FAILURE_POINT'. */ | |
| 1528 #define PUSH_FAILURE_ELT(item) \ | |
| 1529 fail_stack.stack[fail_stack.avail++] = (item) | |
| 1530 | |
| 1531 /* These three POP... operations complement the three PUSH... operations. | |
| 1532 All assume that `fail_stack' is nonempty. */ | |
| 1533 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer | |
| 1534 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer | |
| 1535 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] | |
| 1536 | |
| 1537 /* Used to omit pushing failure point id's when we're not debugging. */ | |
| 1538 #ifdef DEBUG | |
| 1539 #define DEBUG_PUSH PUSH_FAILURE_INT | |
| 1540 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () | |
| 1541 #else | |
| 1542 #define DEBUG_PUSH(item) | |
| 1543 #define DEBUG_POP(item_addr) | |
| 1544 #endif | |
| 1545 | |
| 1546 | |
| 1547 /* Push the information about the state we will need | |
| 1548 if we ever fail back to it. | |
| 1549 | |
| 1550 Requires variables fail_stack, regstart, regend, reg_info, and | |
| 1551 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be | |
| 1552 declared. | |
| 1553 | |
| 1554 Does `return FAILURE_CODE' if runs out of memory. */ | |
| 1555 | |
| 771 | 1556 #if !defined (REGEX_MALLOC) && !defined (REGEX_REL_ALLOC) |
| 456 | 1557 #define DECLARE_DESTINATION char *destination |
| 428 | 1558 #else |
| 456 | 1559 #define DECLARE_DESTINATION DECLARE_NOTHING |
| 428 | 1560 #endif |
| 1561 | |
| 1562 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ | |
| 456 | 1563 do { \ |
| 1564 DECLARE_DESTINATION; \ | |
| 1565 /* Must be int, so when we don't save any registers, the arithmetic \ | |
| 1566 of 0 + -1 isn't done as unsigned. */ \ | |
| 1567 int this_reg; \ | |
| 428 | 1568 \ |
| 456 | 1569 DEBUG_STATEMENT (failure_id++); \ |
| 1570 DEBUG_STATEMENT (nfailure_points_pushed++); \ | |
| 5041 | 1571 DEBUG_FAIL_PRINT2 ("\nPUSH_FAILURE_POINT #%d:\n", failure_id); \ |
| 1572 DEBUG_FAIL_PRINT2 (" Before push, next avail: %ld\n", \ | |
| 647 | 1573 (long) (fail_stack).avail); \ |
| 5041 | 1574 DEBUG_FAIL_PRINT2 (" size: %ld\n", \ |
| 647 | 1575 (long) (fail_stack).size); \ |
| 456 | 1576 \ |
| 5041 | 1577 DEBUG_FAIL_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \ |
| 1578 DEBUG_FAIL_PRINT2 (" available: %ld\n", \ | |
| 456 | 1579 (long) REMAINING_AVAIL_SLOTS); \ |
| 428 | 1580 \ |
| 456 | 1581 /* Ensure we have enough space allocated for what we will push. */ \ |
| 1582 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ | |
| 1583 { \ | |
| 1333 | 1584 BEGIN_REGEX_MALLOC_OK (); \ |
| 456 | 1585 if (!DOUBLE_FAIL_STACK (fail_stack)) \ |
| 1333 | 1586 { \ |
| 1587 END_REGEX_MALLOC_OK (); \ | |
| 1588 UNBIND_REGEX_MALLOC_CHECK (); \ | |
| 1589 return failure_code; \ | |
| 1590 } \ | |
| 1591 END_REGEX_MALLOC_OK (); \ | |
| 5041 | 1592 DEBUG_FAIL_PRINT2 ("\n Doubled stack; size now: %ld\n", \ |
| 647 | 1593 (long) (fail_stack).size); \ |
| 5041 | 1594 DEBUG_FAIL_PRINT2 (" slots available: %ld\n", \ |
| 456 | 1595 (long) REMAINING_AVAIL_SLOTS); \ |
| 1333 | 1596 \ |
| 1597 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); \ | |
| 456 | 1598 } \ |
| 428 | 1599 \ |
| 456 | 1600 /* Push the info, starting with the registers. */ \ |
| 5041 | 1601 DEBUG_FAIL_PRINT1 ("\n"); \ |
| 428 | 1602 \ |
| 456 | 1603 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ |
| 1604 this_reg++) \ | |
| 1605 { \ | |
| 5041 | 1606 DEBUG_FAIL_PRINT2 (" Pushing reg: %d\n", this_reg); \ |
| 456 | 1607 DEBUG_STATEMENT (num_regs_pushed++); \ |
| 428 | 1608 \ |
| 5041 | 1609 DEBUG_FAIL_PRINT2 (" start: 0x%lx\n", (long) regstart[this_reg]); \ |
| 456 | 1610 PUSH_FAILURE_POINTER (regstart[this_reg]); \ |
| 1611 \ | |
| 5041 | 1612 DEBUG_FAIL_PRINT2 (" end: 0x%lx\n", (long) regend[this_reg]); \ |
| 456 | 1613 PUSH_FAILURE_POINTER (regend[this_reg]); \ |
| 428 | 1614 \ |
| 5041 | 1615 DEBUG_FAIL_PRINT2 (" info: 0x%lx\n ", \ |
| 456 | 1616 * (long *) (®_info[this_reg])); \ |
| 5041 | 1617 DEBUG_FAIL_PRINT2 (" match_null=%d", \ |
| 456 | 1618 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ |
| 5041 | 1619 DEBUG_FAIL_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ |
| 1620 DEBUG_FAIL_PRINT2 (" matched_something=%d", \ | |
| 456 | 1621 MATCHED_SOMETHING (reg_info[this_reg])); \ |
| 5041 | 1622 DEBUG_FAIL_PRINT2 (" ever_matched_something=%d", \ |
| 456 | 1623 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ |
| 5041 | 1624 DEBUG_FAIL_PRINT1 ("\n"); \ |
| 456 | 1625 PUSH_FAILURE_ELT (reg_info[this_reg].word); \ |
| 1626 } \ | |
| 428 | 1627 \ |
| 5041 | 1628 DEBUG_FAIL_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg); \ |
| 456 | 1629 PUSH_FAILURE_INT (lowest_active_reg); \ |
| 428 | 1630 \ |
| 5041 | 1631 DEBUG_FAIL_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg); \ |
| 456 | 1632 PUSH_FAILURE_INT (highest_active_reg); \ |
| 428 | 1633 \ |
| 5041 | 1634 DEBUG_FAIL_PRINT2 (" Pushing pattern 0x%lx: \n", (long) pattern_place); \ |
| 1635 DEBUG_FAIL_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ | |
| 456 | 1636 PUSH_FAILURE_POINTER (pattern_place); \ |
| 428 | 1637 \ |
| 5041 | 1638 DEBUG_FAIL_PRINT2 (" Pushing string 0x%lx: `", (long) string_place); \ |
| 1639 DEBUG_FAIL_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ | |
| 456 | 1640 size2); \ |
| 5041 | 1641 DEBUG_FAIL_PRINT1 ("'\n"); \ |
| 456 | 1642 PUSH_FAILURE_POINTER (string_place); \ |
| 428 | 1643 \ |
| 5041 | 1644 DEBUG_FAIL_PRINT2 (" Pushing failure id: %u\n", failure_id); \ |
| 456 | 1645 DEBUG_PUSH (failure_id); \ |
| 1646 } while (0) | |
| 428 | 1647 |
| 1648 /* This is the number of items that are pushed and popped on the stack | |
| 1649 for each register. */ | |
| 1650 #define NUM_REG_ITEMS 3 | |
| 1651 | |
| 1652 /* Individual items aside from the registers. */ | |
| 1653 #ifdef DEBUG | |
| 1654 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ | |
| 1655 #else | |
| 1656 #define NUM_NONREG_ITEMS 4 | |
| 1657 #endif | |
| 1658 | |
| 1659 /* We push at most this many items on the stack. */ | |
| 1660 /* We used to use (num_regs - 1), which is the number of registers | |
| 1661 this regexp will save; but that was changed to 5 | |
| 1662 to avoid stack overflow for a regexp with lots of parens. */ | |
| 1663 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS) | |
| 1664 | |
| 1665 /* We actually push this many items. */ | |
| 1666 #define NUM_FAILURE_ITEMS \ | |
| 1667 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \ | |
| 1668 + NUM_NONREG_ITEMS) | |
| 1669 | |
| 1670 /* How many items can still be added to the stack without overflowing it. */ | |
| 1671 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) | |
| 1672 | |
| 1673 | |
| 1674 /* Pops what PUSH_FAIL_STACK pushes. | |
| 1675 | |
| 1676 We restore into the parameters, all of which should be lvalues: | |
| 1677 STR -- the saved data position. | |
| 1678 PAT -- the saved pattern position. | |
| 1679 LOW_REG, HIGH_REG -- the highest and lowest active registers. | |
| 1680 REGSTART, REGEND -- arrays of string positions. | |
| 1681 REG_INFO -- array of information about each subexpression. | |
| 1682 | |
| 1683 Also assumes the variables `fail_stack' and (if debugging), `bufp', | |
| 1684 `pend', `string1', `size1', `string2', and `size2'. */ | |
| 1685 | |
| 456 | 1686 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, \ |
| 1687 regstart, regend, reg_info) \ | |
| 1688 do { \ | |
| 428 | 1689 DEBUG_STATEMENT (fail_stack_elt_t ffailure_id;) \ |
| 1690 int this_reg; \ | |
| 442 | 1691 const unsigned char *string_temp; \ |
| 428 | 1692 \ |
| 1693 assert (!FAIL_STACK_EMPTY ()); \ | |
| 1694 \ | |
| 1695 /* Remove failure points and point to how many regs pushed. */ \ | |
| 5041 | 1696 DEBUG_FAIL_PRINT1 ("POP_FAILURE_POINT:\n"); \ |
| 1697 DEBUG_FAIL_PRINT2 (" Before pop, next avail: %ld\n", \ | |
| 647 | 1698 (long) fail_stack.avail); \ |
| 5041 | 1699 DEBUG_FAIL_PRINT2 (" size: %ld\n", \ |
| 647 | 1700 (long) fail_stack.size); \ |
| 428 | 1701 \ |
| 1702 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ | |
| 1703 \ | |
| 1704 DEBUG_POP (&ffailure_id.integer); \ | |
| 5041 | 1705 DEBUG_FAIL_PRINT2 (" Popping failure id: %d\n", \ |
| 647 | 1706 * (int *) &ffailure_id); \ |
| 428 | 1707 \ |
| 1708 /* If the saved string location is NULL, it came from an \ | |
| 1709 on_failure_keep_string_jump opcode, and we want to throw away the \ | |
| 1710 saved NULL, thus retaining our current position in the string. */ \ | |
| 1711 string_temp = POP_FAILURE_POINTER (); \ | |
| 1712 if (string_temp != NULL) \ | |
| 446 | 1713 str = string_temp; \ |
| 428 | 1714 \ |
| 5041 | 1715 DEBUG_FAIL_PRINT2 (" Popping string 0x%lx: `", (long) str); \ |
| 1716 DEBUG_FAIL_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ | |
| 1717 DEBUG_FAIL_PRINT1 ("'\n"); \ | |
| 428 | 1718 \ |
| 1719 pat = (unsigned char *) POP_FAILURE_POINTER (); \ | |
| 5041 | 1720 DEBUG_FAIL_PRINT2 (" Popping pattern 0x%lx: ", (long) pat); \ |
| 1721 DEBUG_FAIL_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ | |
| 428 | 1722 \ |
| 1723 /* Restore register info. */ \ | |
| 647 | 1724 high_reg = POP_FAILURE_INT (); \ |
| 5041 | 1725 DEBUG_FAIL_PRINT2 (" Popping high active reg: %d\n", high_reg); \ |
| 428 | 1726 \ |
| 647 | 1727 low_reg = POP_FAILURE_INT (); \ |
| 5041 | 1728 DEBUG_FAIL_PRINT2 (" Popping low active reg: %d\n", low_reg); \ |
| 428 | 1729 \ |
| 1730 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ | |
| 1731 { \ | |
| 5041 | 1732 DEBUG_FAIL_PRINT2 (" Popping reg: %d\n", this_reg); \ |
| 428 | 1733 \ |
| 1734 reg_info[this_reg].word = POP_FAILURE_ELT (); \ | |
| 5041 | 1735 DEBUG_FAIL_PRINT2 (" info: 0x%lx\n", \ |
| 428 | 1736 * (long *) ®_info[this_reg]); \ |
| 1737 \ | |
| 446 | 1738 regend[this_reg] = POP_FAILURE_POINTER (); \ |
| 5041 | 1739 DEBUG_FAIL_PRINT2 (" end: 0x%lx\n", (long) regend[this_reg]); \ |
| 428 | 1740 \ |
| 446 | 1741 regstart[this_reg] = POP_FAILURE_POINTER (); \ |
| 5041 | 1742 DEBUG_FAIL_PRINT2 (" start: 0x%lx\n", (long) regstart[this_reg]); \ |
| 428 | 1743 } \ |
| 1744 \ | |
| 1745 set_regs_matched_done = 0; \ | |
| 1746 DEBUG_STATEMENT (nfailure_points_popped++); \ | |
| 456 | 1747 } while (0) /* POP_FAILURE_POINT */ |
| 428 | 1748 |
| 1749 | |
| 1750 | |
| 1751 /* Structure for per-register (a.k.a. per-group) information. | |
| 1752 Other register information, such as the | |
| 1753 starting and ending positions (which are addresses), and the list of | |
| 1754 inner groups (which is a bits list) are maintained in separate | |
| 1755 variables. | |
| 1756 | |
| 1757 We are making a (strictly speaking) nonportable assumption here: that | |
| 1758 the compiler will pack our bit fields into something that fits into | |
| 1759 the type of `word', i.e., is something that fits into one item on the | |
| 1760 failure stack. */ | |
| 1761 | |
| 1762 typedef union | |
| 1763 { | |
| 1764 fail_stack_elt_t word; | |
| 1765 struct | |
| 1766 { | |
| 1767 /* This field is one if this group can match the empty string, | |
| 1768 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ | |
| 1769 #define MATCH_NULL_UNSET_VALUE 3 | |
| 647 | 1770 unsigned int match_null_string_p : 2; |
| 1771 unsigned int is_active : 1; | |
| 1772 unsigned int matched_something : 1; | |
| 1773 unsigned int ever_matched_something : 1; | |
| 428 | 1774 } bits; |
| 1775 } register_info_type; | |
| 1776 | |
| 1777 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) | |
| 1778 #define IS_ACTIVE(R) ((R).bits.is_active) | |
| 1779 #define MATCHED_SOMETHING(R) ((R).bits.matched_something) | |
| 1780 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) | |
| 1781 | |
| 1782 | |
| 1783 /* Call this when have matched a real character; it sets `matched' flags | |
| 1784 for the subexpressions which we are currently inside. Also records | |
| 1785 that those subexprs have matched. */ | |
| 1786 #define SET_REGS_MATCHED() \ | |
| 1787 do \ | |
| 1788 { \ | |
| 1789 if (!set_regs_matched_done) \ | |
| 1790 { \ | |
| 647 | 1791 int r; \ |
| 428 | 1792 set_regs_matched_done = 1; \ |
| 1793 for (r = lowest_active_reg; r <= highest_active_reg; r++) \ | |
| 1794 { \ | |
| 1795 MATCHED_SOMETHING (reg_info[r]) \ | |
| 1796 = EVER_MATCHED_SOMETHING (reg_info[r]) \ | |
| 1797 = 1; \ | |
| 1798 } \ | |
| 1799 } \ | |
| 1800 } \ | |
| 1801 while (0) | |
| 1802 | |
| 1803 /* Registers are set to a sentinel when they haven't yet matched. */ | |
| 446 | 1804 static unsigned char reg_unset_dummy; |
| 428 | 1805 #define REG_UNSET_VALUE (®_unset_dummy) |
| 1806 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE) | |
| 1807 | |
| 1808 /* Subroutine declarations and macros for regex_compile. */ | |
| 1809 | |
| 1810 /* Fetch the next character in the uncompiled pattern---translating it | |
| 826 | 1811 if necessary. */ |
| 428 | 1812 #define PATFETCH(c) \ |
| 446 | 1813 do { \ |
| 1814 PATFETCH_RAW (c); \ | |
| 826 | 1815 c = RE_TRANSLATE (c); \ |
| 428 | 1816 } while (0) |
| 1817 | |
| 1818 /* Fetch the next character in the uncompiled pattern, with no | |
| 1819 translation. */ | |
| 1820 #define PATFETCH_RAW(c) \ | |
| 1821 do {if (p == pend) return REG_EEND; \ | |
| 1822 assert (p < pend); \ | |
| 867 | 1823 c = itext_ichar (p); \ |
| 1824 INC_IBYTEPTR (p); \ | |
| 428 | 1825 } while (0) |
| 1826 | |
| 1827 /* Go backwards one character in the pattern. */ | |
| 867 | 1828 #define PATUNFETCH DEC_IBYTEPTR (p) |
| 428 | 1829 |
| 1830 /* If `translate' is non-null, return translate[D], else just D. We | |
| 1831 cast the subscript to translate because some data is declared as | |
| 1832 `char *', to avoid warnings when a string constant is passed. But | |
| 1833 when we use a character as a subscript we must make it unsigned. */ | |
| 826 | 1834 #define RE_TRANSLATE(d) \ |
| 1835 (TRANSLATE_P (translate) ? RE_TRANSLATE_1 (d) : (d)) | |
| 428 | 1836 |
| 1837 /* Macros for outputting the compiled pattern into `buffer'. */ | |
| 1838 | |
| 1839 /* If the buffer isn't allocated when it comes in, use this. */ | |
| 1840 #define INIT_BUF_SIZE 32 | |
| 1841 | |
| 1842 /* Make sure we have at least N more bytes of space in buffer. */ | |
| 1843 #define GET_BUFFER_SPACE(n) \ | |
| 647 | 1844 while (buf_end - bufp->buffer + (n) > (ptrdiff_t) bufp->allocated) \ |
| 428 | 1845 EXTEND_BUFFER () |
| 1846 | |
| 1847 /* Make sure we have one more byte of buffer space and then add C to it. */ | |
| 1848 #define BUF_PUSH(c) \ | |
| 1849 do { \ | |
| 1850 GET_BUFFER_SPACE (1); \ | |
| 446 | 1851 *buf_end++ = (unsigned char) (c); \ |
| 428 | 1852 } while (0) |
| 1853 | |
| 1854 | |
| 1855 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ | |
| 1856 #define BUF_PUSH_2(c1, c2) \ | |
| 1857 do { \ | |
| 1858 GET_BUFFER_SPACE (2); \ | |
| 446 | 1859 *buf_end++ = (unsigned char) (c1); \ |
| 1860 *buf_end++ = (unsigned char) (c2); \ | |
| 428 | 1861 } while (0) |
| 1862 | |
| 1863 | |
| 1864 /* As with BUF_PUSH_2, except for three bytes. */ | |
| 1865 #define BUF_PUSH_3(c1, c2, c3) \ | |
| 1866 do { \ | |
| 1867 GET_BUFFER_SPACE (3); \ | |
| 446 | 1868 *buf_end++ = (unsigned char) (c1); \ |
| 1869 *buf_end++ = (unsigned char) (c2); \ | |
| 1870 *buf_end++ = (unsigned char) (c3); \ | |
| 428 | 1871 } while (0) |
| 1872 | |
| 1873 | |
| 1874 /* Store a jump with opcode OP at LOC to location TO. We store a | |
| 1875 relative address offset by the three bytes the jump itself occupies. */ | |
| 1876 #define STORE_JUMP(op, loc, to) \ | |
| 1877 store_op1 (op, loc, (to) - (loc) - 3) | |
| 1878 | |
| 1879 /* Likewise, for a two-argument jump. */ | |
| 1880 #define STORE_JUMP2(op, loc, to, arg) \ | |
| 1881 store_op2 (op, loc, (to) - (loc) - 3, arg) | |
| 1882 | |
| 446 | 1883 /* Like `STORE_JUMP', but for inserting. Assume `buf_end' is the |
| 1884 buffer end. */ | |
| 428 | 1885 #define INSERT_JUMP(op, loc, to) \ |
| 446 | 1886 insert_op1 (op, loc, (to) - (loc) - 3, buf_end) |
| 1887 | |
| 1888 /* Like `STORE_JUMP2', but for inserting. Assume `buf_end' is the | |
| 1889 buffer end. */ | |
| 428 | 1890 #define INSERT_JUMP2(op, loc, to, arg) \ |
| 446 | 1891 insert_op2 (op, loc, (to) - (loc) - 3, arg, buf_end) |
| 428 | 1892 |
| 1893 | |
| 1894 /* This is not an arbitrary limit: the arguments which represent offsets | |
| 1895 into the pattern are two bytes long. So if 2^16 bytes turns out to | |
| 1896 be too small, many things would have to change. */ | |
| 1897 #define MAX_BUF_SIZE (1L << 16) | |
| 1898 | |
| 1899 | |
| 1900 /* Extend the buffer by twice its current size via realloc and | |
| 1901 reset the pointers that pointed into the old block to point to the | |
| 1902 correct places in the new one. If extending the buffer results in it | |
| 1903 being larger than MAX_BUF_SIZE, then flag memory exhausted. */ | |
| 1333 | 1904 #define EXTEND_BUFFER() \ |
| 1905 do { \ | |
| 1906 re_char *old_buffer = bufp->buffer; \ | |
| 1907 if (bufp->allocated == MAX_BUF_SIZE) \ | |
| 1908 return REG_ESIZE; \ | |
| 1909 bufp->allocated <<= 1; \ | |
| 1910 if (bufp->allocated > MAX_BUF_SIZE) \ | |
| 1911 bufp->allocated = MAX_BUF_SIZE; \ | |
| 1912 bufp->buffer = \ | |
| 1913 (unsigned char *) xrealloc (bufp->buffer, bufp->allocated); \ | |
| 1914 if (bufp->buffer == NULL) \ | |
| 1915 return REG_ESPACE; \ | |
| 1916 /* If the buffer moved, move all the pointers into it. */ \ | |
| 1917 if (old_buffer != bufp->buffer) \ | |
| 1918 { \ | |
| 1919 buf_end = (buf_end - old_buffer) + bufp->buffer; \ | |
| 1920 begalt = (begalt - old_buffer) + bufp->buffer; \ | |
| 1921 if (fixup_alt_jump) \ | |
| 1922 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer; \ | |
| 1923 if (laststart) \ | |
| 1924 laststart = (laststart - old_buffer) + bufp->buffer; \ | |
| 1925 if (pending_exact) \ | |
| 1926 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ | |
| 1927 } \ | |
| 428 | 1928 } while (0) |
| 1929 | |
| 1930 | |
| 1931 /* Since we have one byte reserved for the register number argument to | |
| 1932 {start,stop}_memory, the maximum number of groups we can report | |
| 1933 things about is what fits in that byte. */ | |
| 1934 #define MAX_REGNUM 255 | |
| 1935 | |
| 1936 /* But patterns can have more than `MAX_REGNUM' registers. We just | |
| 502 | 1937 ignore the excess. |
| 1938 #### not true! groups past this will fail in lots of ways, if we | |
| 1939 ever have to backtrack. | |
| 1940 */ | |
| 647 | 1941 typedef int regnum_t; |
| 428 | 1942 |
| 502 | 1943 #define INIT_REG_TRANSLATE_SIZE 5 |
| 428 | 1944 |
| 1945 /* Macros for the compile stack. */ | |
| 1946 | |
| 1947 /* Since offsets can go either forwards or backwards, this type needs to | |
| 1948 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ | |
| 1949 typedef int pattern_offset_t; | |
| 1950 | |
| 1951 typedef struct | |
| 1952 { | |
| 1953 pattern_offset_t begalt_offset; | |
| 1954 pattern_offset_t fixup_alt_jump; | |
| 1955 pattern_offset_t inner_group_offset; | |
| 1956 pattern_offset_t laststart_offset; | |
| 1957 regnum_t regnum; | |
| 1958 } compile_stack_elt_t; | |
| 1959 | |
| 1960 | |
| 1961 typedef struct | |
| 1962 { | |
| 1963 compile_stack_elt_t *stack; | |
| 647 | 1964 int size; |
| 1965 int avail; /* Offset of next open position. */ | |
| 428 | 1966 } compile_stack_type; |
| 1967 | |
| 1968 | |
| 1969 #define INIT_COMPILE_STACK_SIZE 32 | |
| 1970 | |
| 1971 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) | |
| 1972 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) | |
| 1973 | |
| 1974 /* The next available element. */ | |
| 1975 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) | |
| 1976 | |
| 1977 | |
| 1978 /* Set the bit for character C in a bit vector. */ | |
| 1979 #define SET_LIST_BIT(c) \ | |
| 446 | 1980 (buf_end[((unsigned char) (c)) / BYTEWIDTH] \ |
| 428 | 1981 |= 1 << (((unsigned char) c) % BYTEWIDTH)) |
| 1982 | |
| 1983 #ifdef MULE | |
| 1984 | |
| 1985 /* Set the "bit" for character C in a range table. */ | |
| 1986 #define SET_RANGETAB_BIT(c) put_range_table (rtab, c, c, Qt) | |
| 1987 | |
| 1988 /* Set the "bit" for character c in the appropriate table. */ | |
| 1989 #define SET_EITHER_BIT(c) \ | |
| 1990 do { \ | |
| 1991 if (has_extended_chars) \ | |
| 1992 SET_RANGETAB_BIT (c); \ | |
| 1993 else \ | |
| 1994 SET_LIST_BIT (c); \ | |
| 1995 } while (0) | |
| 1996 | |
| 1997 #else /* not MULE */ | |
| 1998 | |
| 1999 #define SET_EITHER_BIT(c) SET_LIST_BIT (c) | |
| 2000 | |
| 2001 #endif | |
| 2002 | |
| 2003 | |
| 2004 /* Get the next unsigned number in the uncompiled pattern. */ | |
| 2005 #define GET_UNSIGNED_NUMBER(num) \ | |
| 2006 { if (p != pend) \ | |
| 2007 { \ | |
| 2008 PATFETCH (c); \ | |
| 2009 while (ISDIGIT (c)) \ | |
| 2010 { \ | |
| 2011 if (num < 0) \ | |
| 2012 num = 0; \ | |
| 2013 num = num * 10 + c - '0'; \ | |
| 2014 if (p == pend) \ | |
| 2015 break; \ | |
| 2016 PATFETCH (c); \ | |
| 2017 } \ | |
| 2018 } \ | |
| 2019 } | |
| 2020 | |
| 2021 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ | |
| 2022 | |
| 2023 #define IS_CHAR_CLASS(string) \ | |
| 2024 (STREQ (string, "alpha") || STREQ (string, "upper") \ | |
| 2025 || STREQ (string, "lower") || STREQ (string, "digit") \ | |
| 2026 || STREQ (string, "alnum") || STREQ (string, "xdigit") \ | |
| 2027 || STREQ (string, "space") || STREQ (string, "print") \ | |
| 2028 || STREQ (string, "punct") || STREQ (string, "graph") \ | |
| 2029 || STREQ (string, "cntrl") || STREQ (string, "blank")) | |
| 2030 | |
| 2031 static void store_op1 (re_opcode_t op, unsigned char *loc, int arg); | |
| 2032 static void store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2); | |
| 2033 static void insert_op1 (re_opcode_t op, unsigned char *loc, int arg, | |
| 2034 unsigned char *end); | |
| 2035 static void insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, | |
| 2036 unsigned char *end); | |
| 460 | 2037 static re_bool at_begline_loc_p (re_char *pattern, re_char *p, |
| 428 | 2038 reg_syntax_t syntax); |
| 460 | 2039 static re_bool at_endline_loc_p (re_char *p, re_char *pend, int syntax); |
| 2040 static re_bool group_in_compile_stack (compile_stack_type compile_stack, | |
| 428 | 2041 regnum_t regnum); |
| 446 | 2042 static reg_errcode_t compile_range (re_char **p_ptr, re_char *pend, |
| 2043 RE_TRANSLATE_TYPE translate, | |
| 2044 reg_syntax_t syntax, | |
| 428 | 2045 unsigned char *b); |
| 2046 #ifdef MULE | |
| 446 | 2047 static reg_errcode_t compile_extended_range (re_char **p_ptr, |
| 2048 re_char *pend, | |
| 2049 RE_TRANSLATE_TYPE translate, | |
| 428 | 2050 reg_syntax_t syntax, |
| 2051 Lisp_Object rtab); | |
| 2052 #endif /* MULE */ | |
| 460 | 2053 static re_bool group_match_null_string_p (unsigned char **p, |
| 428 | 2054 unsigned char *end, |
| 2055 register_info_type *reg_info); | |
| 460 | 2056 static re_bool alt_match_null_string_p (unsigned char *p, unsigned char *end, |
| 428 | 2057 register_info_type *reg_info); |
| 460 | 2058 static re_bool common_op_match_null_string_p (unsigned char **p, |
| 428 | 2059 unsigned char *end, |
| 2060 register_info_type *reg_info); | |
| 826 | 2061 static int bcmp_translate (re_char *s1, re_char *s2, |
| 2062 REGISTER int len, RE_TRANSLATE_TYPE translate | |
| 2063 #ifdef emacs | |
| 2064 , Internal_Format fmt, Lisp_Object lispobj | |
| 2065 #endif | |
| 2066 ); | |
| 428 | 2067 static int re_match_2_internal (struct re_pattern_buffer *bufp, |
| 446 | 2068 re_char *string1, int size1, |
| 2069 re_char *string2, int size2, int pos, | |
| 826 | 2070 struct re_registers *regs, int stop |
| 2071 RE_LISP_CONTEXT_ARGS_DECL); | |
| 428 | 2072 |
| 2073 #ifndef MATCH_MAY_ALLOCATE | |
| 2074 | |
| 2075 /* If we cannot allocate large objects within re_match_2_internal, | |
| 2076 we make the fail stack and register vectors global. | |
| 2077 The fail stack, we grow to the maximum size when a regexp | |
| 2078 is compiled. | |
| 2079 The register vectors, we adjust in size each time we | |
| 2080 compile a regexp, according to the number of registers it needs. */ | |
| 2081 | |
| 2082 static fail_stack_type fail_stack; | |
| 2083 | |
| 2084 /* Size with which the following vectors are currently allocated. | |
| 2085 That is so we can make them bigger as needed, | |
| 2086 but never make them smaller. */ | |
| 2087 static int regs_allocated_size; | |
| 2088 | |
| 446 | 2089 static re_char ** regstart, ** regend; |
| 2090 static re_char ** old_regstart, ** old_regend; | |
| 2091 static re_char **best_regstart, **best_regend; | |
| 428 | 2092 static register_info_type *reg_info; |
| 446 | 2093 static re_char **reg_dummy; |
| 428 | 2094 static register_info_type *reg_info_dummy; |
| 2095 | |
| 2096 /* Make the register vectors big enough for NUM_REGS registers, | |
| 2097 but don't make them smaller. */ | |
| 2098 | |
| 2099 static | |
| 2100 regex_grow_registers (int num_regs) | |
| 2101 { | |
| 2102 if (num_regs > regs_allocated_size) | |
| 2103 { | |
| 551 | 2104 RETALLOC (regstart, num_regs, re_char *); |
| 2105 RETALLOC (regend, num_regs, re_char *); | |
| 2106 RETALLOC (old_regstart, num_regs, re_char *); | |
| 2107 RETALLOC (old_regend, num_regs, re_char *); | |
| 2108 RETALLOC (best_regstart, num_regs, re_char *); | |
| 2109 RETALLOC (best_regend, num_regs, re_char *); | |
| 2110 RETALLOC (reg_info, num_regs, register_info_type); | |
| 2111 RETALLOC (reg_dummy, num_regs, re_char *); | |
| 2112 RETALLOC (reg_info_dummy, num_regs, register_info_type); | |
| 428 | 2113 |
| 2114 regs_allocated_size = num_regs; | |
| 2115 } | |
| 2116 } | |
| 2117 | |
| 2118 #endif /* not MATCH_MAY_ALLOCATE */ | |
| 2119 | |
| 2120 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. | |
| 2121 Returns one of error codes defined in `regex.h', or zero for success. | |
| 2122 | |
| 2123 Assumes the `allocated' (and perhaps `buffer') and `translate' | |
| 2124 fields are set in BUFP on entry. | |
| 2125 | |
| 2126 If it succeeds, results are put in BUFP (if it returns an error, the | |
| 2127 contents of BUFP are undefined): | |
| 2128 `buffer' is the compiled pattern; | |
| 2129 `syntax' is set to SYNTAX; | |
| 2130 `used' is set to the length of the compiled pattern; | |
| 2131 `fastmap_accurate' is zero; | |
| 502 | 2132 `re_ngroups' is the number of groups/subexpressions (including shy |
| 2133 groups) in PATTERN; | |
| 2134 `re_nsub' is the number of non-shy groups in PATTERN; | |
| 428 | 2135 `not_bol' and `not_eol' are zero; |
| 2136 | |
| 2137 The `fastmap' and `newline_anchor' fields are neither | |
| 2138 examined nor set. */ | |
| 2139 | |
| 2140 /* Return, freeing storage we allocated. */ | |
| 1726 | 2141 #define FREE_STACK_RETURN(value) \ |
| 2142 do \ | |
| 2143 { \ | |
|
4976
16112448d484
Rename xfree(FOO, TYPE) -> xfree(FOO)
Ben Wing <ben@xemacs.org>
parents:
4832
diff
changeset
|
2144 xfree (compile_stack.stack); \ |
| 1726 | 2145 return value; \ |
| 1333 | 2146 } while (0) |
| 428 | 2147 |
| 2148 static reg_errcode_t | |
| 446 | 2149 regex_compile (re_char *pattern, int size, reg_syntax_t syntax, |
| 428 | 2150 struct re_pattern_buffer *bufp) |
| 2151 { | |
| 2152 /* We fetch characters from PATTERN here. We declare these as int | |
| 2153 (or possibly long) so that chars above 127 can be used as | |
| 2154 array indices. The macros that fetch a character from the pattern | |
| 2155 make sure to coerce to unsigned char before assigning, so we won't | |
| 2156 get bitten by negative numbers here. */ | |
| 2157 /* XEmacs change: used to be unsigned char. */ | |
| 2158 REGISTER EMACS_INT c, c1; | |
| 2159 | |
| 2160 /* A random temporary spot in PATTERN. */ | |
| 446 | 2161 re_char *p1; |
| 428 | 2162 |
| 2163 /* Points to the end of the buffer, where we should append. */ | |
| 446 | 2164 REGISTER unsigned char *buf_end; |
| 428 | 2165 |
| 2166 /* Keeps track of unclosed groups. */ | |
| 2167 compile_stack_type compile_stack; | |
| 2168 | |
| 2169 /* Points to the current (ending) position in the pattern. */ | |
| 446 | 2170 re_char *p = pattern; |
| 2171 re_char *pend = pattern + size; | |
| 428 | 2172 |
| 2173 /* How to translate the characters in the pattern. */ | |
| 446 | 2174 RE_TRANSLATE_TYPE translate = bufp->translate; |
| 428 | 2175 |
| 2176 /* Address of the count-byte of the most recently inserted `exactn' | |
| 2177 command. This makes it possible to tell if a new exact-match | |
| 2178 character can be added to that command or if the character requires | |
| 2179 a new `exactn' command. */ | |
| 2180 unsigned char *pending_exact = 0; | |
| 2181 | |
| 2182 /* Address of start of the most recently finished expression. | |
| 2183 This tells, e.g., postfix * where to find the start of its | |
| 2184 operand. Reset at the beginning of groups and alternatives. */ | |
| 2185 unsigned char *laststart = 0; | |
| 2186 | |
| 2187 /* Address of beginning of regexp, or inside of last group. */ | |
| 2188 unsigned char *begalt; | |
| 2189 | |
| 2190 /* Place in the uncompiled pattern (i.e., the {) to | |
| 2191 which to go back if the interval is invalid. */ | |
| 446 | 2192 re_char *beg_interval; |
| 428 | 2193 |
| 2194 /* Address of the place where a forward jump should go to the end of | |
| 2195 the containing expression. Each alternative of an `or' -- except the | |
| 2196 last -- ends with a forward jump of this sort. */ | |
| 2197 unsigned char *fixup_alt_jump = 0; | |
| 2198 | |
| 2199 /* Counts open-groups as they are encountered. Remembered for the | |
| 2200 matching close-group on the compile stack, so the same register | |
| 2201 number is put in the stop_memory as the start_memory. */ | |
| 2202 regnum_t regnum = 0; | |
| 2203 | |
| 2204 #ifdef DEBUG | |
| 5041 | 2205 if (debug_regexps & RE_DEBUG_COMPILATION) |
| 428 | 2206 { |
| 647 | 2207 int debug_count; |
| 428 | 2208 |
| 5041 | 2209 DEBUG_PRINT1 ("\nCompiling pattern: "); |
| 428 | 2210 for (debug_count = 0; debug_count < size; debug_count++) |
| 2211 putchar (pattern[debug_count]); | |
| 2212 putchar ('\n'); | |
| 2213 } | |
| 2214 #endif /* DEBUG */ | |
| 2215 | |
| 2216 /* Initialize the compile stack. */ | |
| 2217 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); | |
| 2218 if (compile_stack.stack == NULL) | |
| 2219 return REG_ESPACE; | |
| 2220 | |
| 2221 compile_stack.size = INIT_COMPILE_STACK_SIZE; | |
| 2222 compile_stack.avail = 0; | |
| 2223 | |
| 2224 /* Initialize the pattern buffer. */ | |
| 2225 bufp->syntax = syntax; | |
| 2226 bufp->fastmap_accurate = 0; | |
| 2227 bufp->not_bol = bufp->not_eol = 0; | |
| 2228 | |
| 2229 /* Set `used' to zero, so that if we return an error, the pattern | |
| 2230 printer (for debugging) will think there's no pattern. We reset it | |
| 2231 at the end. */ | |
| 2232 bufp->used = 0; | |
| 2233 | |
| 2234 /* Always count groups, whether or not bufp->no_sub is set. */ | |
| 2235 bufp->re_nsub = 0; | |
| 502 | 2236 bufp->re_ngroups = 0; |
| 2237 | |
| 2238 bufp->warned_about_incompatible_back_references = 0; | |
| 2239 | |
| 2240 if (bufp->external_to_internal_register == 0) | |
| 2241 { | |
| 2242 bufp->external_to_internal_register_size = INIT_REG_TRANSLATE_SIZE; | |
| 2243 RETALLOC (bufp->external_to_internal_register, | |
| 2244 bufp->external_to_internal_register_size, | |
| 2245 int); | |
| 2246 } | |
| 2247 | |
| 2248 { | |
| 2249 int i; | |
| 2250 | |
| 2251 bufp->external_to_internal_register[0] = 0; | |
| 2252 for (i = 1; i < bufp->external_to_internal_register_size; i++) | |
| 2253 bufp->external_to_internal_register[i] = (int) 0xDEADBEEF; | |
| 2254 } | |
| 428 | 2255 |
| 2256 #if !defined (emacs) && !defined (SYNTAX_TABLE) | |
| 2257 /* Initialize the syntax table. */ | |
| 2258 init_syntax_once (); | |
| 2259 #endif | |
| 2260 | |
| 2261 if (bufp->allocated == 0) | |
| 2262 { | |
| 2263 if (bufp->buffer) | |
| 2264 { /* If zero allocated, but buffer is non-null, try to realloc | |
| 2265 enough space. This loses if buffer's address is bogus, but | |
| 2266 that is the user's responsibility. */ | |
| 2267 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); | |
| 2268 } | |
| 2269 else | |
| 2270 { /* Caller did not allocate a buffer. Do it for them. */ | |
| 2271 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); | |
| 2272 } | |
| 2273 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); | |
| 2274 | |
| 2275 bufp->allocated = INIT_BUF_SIZE; | |
| 2276 } | |
| 2277 | |
| 446 | 2278 begalt = buf_end = bufp->buffer; |
| 428 | 2279 |
| 2280 /* Loop through the uncompiled pattern until we're at the end. */ | |
| 2281 while (p != pend) | |
| 2282 { | |
| 2283 PATFETCH (c); | |
| 2284 | |
| 2285 switch (c) | |
| 2286 { | |
| 2287 case '^': | |
| 2288 { | |
| 2289 if ( /* If at start of pattern, it's an operator. */ | |
| 2290 p == pattern + 1 | |
| 2291 /* If context independent, it's an operator. */ | |
| 2292 || syntax & RE_CONTEXT_INDEP_ANCHORS | |
| 2293 /* Otherwise, depends on what's come before. */ | |
| 2294 || at_begline_loc_p (pattern, p, syntax)) | |
| 2295 BUF_PUSH (begline); | |
| 2296 else | |
| 2297 goto normal_char; | |
| 2298 } | |
| 2299 break; | |
| 2300 | |
| 2301 | |
| 2302 case '$': | |
| 2303 { | |
| 2304 if ( /* If at end of pattern, it's an operator. */ | |
| 2305 p == pend | |
| 2306 /* If context independent, it's an operator. */ | |
| 2307 || syntax & RE_CONTEXT_INDEP_ANCHORS | |
| 2308 /* Otherwise, depends on what's next. */ | |
| 2309 || at_endline_loc_p (p, pend, syntax)) | |
| 2310 BUF_PUSH (endline); | |
| 2311 else | |
| 2312 goto normal_char; | |
| 2313 } | |
| 2314 break; | |
| 2315 | |
| 2316 | |
| 2317 case '+': | |
| 2318 case '?': | |
| 2319 if ((syntax & RE_BK_PLUS_QM) | |
| 2320 || (syntax & RE_LIMITED_OPS)) | |
| 2321 goto normal_char; | |
| 2322 handle_plus: | |
| 2323 case '*': | |
| 2324 /* If there is no previous pattern... */ | |
| 2325 if (!laststart) | |
| 2326 { | |
| 2327 if (syntax & RE_CONTEXT_INVALID_OPS) | |
| 2328 FREE_STACK_RETURN (REG_BADRPT); | |
| 2329 else if (!(syntax & RE_CONTEXT_INDEP_OPS)) | |
| 2330 goto normal_char; | |
| 2331 } | |
| 2332 | |
| 2333 { | |
| 2334 /* true means zero/many matches are allowed. */ | |
| 460 | 2335 re_bool zero_times_ok = c != '+'; |
| 2336 re_bool many_times_ok = c != '?'; | |
| 428 | 2337 |
| 2338 /* true means match shortest string possible. */ | |
| 460 | 2339 re_bool minimal = false; |
| 428 | 2340 |
| 2341 /* If there is a sequence of repetition chars, collapse it | |
| 2342 down to just one (the right one). We can't combine | |
| 2343 interval operators with these because of, e.g., `a{2}*', | |
| 2344 which should only match an even number of `a's. */ | |
| 2345 while (p != pend) | |
| 2346 { | |
| 2347 PATFETCH (c); | |
| 2348 | |
| 2349 if (c == '*' || (!(syntax & RE_BK_PLUS_QM) | |
| 2350 && (c == '+' || c == '?'))) | |
| 2351 ; | |
| 2352 | |
| 2353 else if (syntax & RE_BK_PLUS_QM && c == '\\') | |
| 2354 { | |
| 2355 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
| 2356 | |
| 2357 PATFETCH (c1); | |
| 2358 if (!(c1 == '+' || c1 == '?')) | |
| 2359 { | |
| 2360 PATUNFETCH; | |
| 2361 PATUNFETCH; | |
| 2362 break; | |
| 2363 } | |
| 2364 | |
| 2365 c = c1; | |
| 2366 } | |
| 2367 else | |
| 2368 { | |
| 2369 PATUNFETCH; | |
| 2370 break; | |
| 2371 } | |
| 2372 | |
| 2373 /* If we get here, we found another repeat character. */ | |
| 2374 if (!(syntax & RE_NO_MINIMAL_MATCHING)) | |
| 2375 { | |
| 440 | 2376 /* "*?" and "+?" and "??" are okay (and mean match |
| 2377 minimally), but other sequences (such as "*??" and | |
| 2378 "+++") are rejected (reserved for future use). */ | |
| 428 | 2379 if (minimal || c != '?') |
| 2380 FREE_STACK_RETURN (REG_BADRPT); | |
| 2381 minimal = true; | |
| 2382 } | |
| 2383 else | |
| 2384 { | |
| 2385 zero_times_ok |= c != '+'; | |
| 2386 many_times_ok |= c != '?'; | |
| 2387 } | |
| 2388 } | |
| 2389 | |
| 2390 /* Star, etc. applied to an empty pattern is equivalent | |
| 2391 to an empty pattern. */ | |
| 2392 if (!laststart) | |
| 2393 break; | |
| 2394 | |
| 2395 /* Now we know whether zero matches is allowed | |
| 2396 and whether two or more matches is allowed | |
| 2397 and whether we want minimal or maximal matching. */ | |
| 2398 if (minimal) | |
| 2399 { | |
| 2400 if (!many_times_ok) | |
| 2401 { | |
| 2402 /* "a??" becomes: | |
| 2403 0: /on_failure_jump to 6 | |
| 2404 3: /jump to 9 | |
| 2405 6: /exactn/1/A | |
| 2406 9: end of pattern. | |
| 2407 */ | |
| 2408 GET_BUFFER_SPACE (6); | |
| 446 | 2409 INSERT_JUMP (jump, laststart, buf_end + 3); |
| 2410 buf_end += 3; | |
| 428 | 2411 INSERT_JUMP (on_failure_jump, laststart, laststart + 6); |
| 446 | 2412 buf_end += 3; |
| 428 | 2413 } |
| 2414 else if (zero_times_ok) | |
| 2415 { | |
| 2416 /* "a*?" becomes: | |
| 2417 0: /jump to 6 | |
| 2418 3: /exactn/1/A | |
| 2419 6: /on_failure_jump to 3 | |
| 2420 9: end of pattern. | |
| 2421 */ | |
| 2422 GET_BUFFER_SPACE (6); | |
| 446 | 2423 INSERT_JUMP (jump, laststart, buf_end + 3); |
| 2424 buf_end += 3; | |
| 2425 STORE_JUMP (on_failure_jump, buf_end, laststart + 3); | |
| 2426 buf_end += 3; | |
| 428 | 2427 } |
| 2428 else | |
| 2429 { | |
| 2430 /* "a+?" becomes: | |
| 2431 0: /exactn/1/A | |
| 2432 3: /on_failure_jump to 0 | |
| 2433 6: end of pattern. | |
| 2434 */ | |
| 2435 GET_BUFFER_SPACE (3); | |
| 446 | 2436 STORE_JUMP (on_failure_jump, buf_end, laststart); |
| 2437 buf_end += 3; | |
| 428 | 2438 } |
| 2439 } | |
| 2440 else | |
| 2441 { | |
| 2442 /* Are we optimizing this jump? */ | |
| 460 | 2443 re_bool keep_string_p = false; |
| 428 | 2444 |
| 2445 if (many_times_ok) | |
| 446 | 2446 { /* More than one repetition is allowed, so put in |
| 2447 at the end a backward relative jump from | |
| 2448 `buf_end' to before the next jump we're going | |
| 2449 to put in below (which jumps from laststart to | |
| 2450 after this jump). | |
| 428 | 2451 |
| 2452 But if we are at the `*' in the exact sequence `.*\n', | |
| 2453 insert an unconditional jump backwards to the ., | |
| 2454 instead of the beginning of the loop. This way we only | |
| 2455 push a failure point once, instead of every time | |
| 2456 through the loop. */ | |
| 2457 assert (p - 1 > pattern); | |
| 2458 | |
| 2459 /* Allocate the space for the jump. */ | |
| 2460 GET_BUFFER_SPACE (3); | |
| 2461 | |
| 2462 /* We know we are not at the first character of the | |
| 2463 pattern, because laststart was nonzero. And we've | |
| 2464 already incremented `p', by the way, to be the | |
| 2465 character after the `*'. Do we have to do something | |
| 2466 analogous here for null bytes, because of | |
| 2467 RE_DOT_NOT_NULL? */ | |
| 446 | 2468 if (*(p - 2) == '.' |
| 428 | 2469 && zero_times_ok |
| 446 | 2470 && p < pend && *p == '\n' |
| 428 | 2471 && !(syntax & RE_DOT_NEWLINE)) |
| 2472 { /* We have .*\n. */ | |
| 446 | 2473 STORE_JUMP (jump, buf_end, laststart); |
| 428 | 2474 keep_string_p = true; |
| 2475 } | |
| 2476 else | |
| 2477 /* Anything else. */ | |
| 446 | 2478 STORE_JUMP (maybe_pop_jump, buf_end, laststart - 3); |
| 428 | 2479 |
| 2480 /* We've added more stuff to the buffer. */ | |
| 446 | 2481 buf_end += 3; |
| 428 | 2482 } |
| 2483 | |
| 446 | 2484 /* On failure, jump from laststart to buf_end + 3, |
| 2485 which will be the end of the buffer after this jump | |
| 2486 is inserted. */ | |
| 428 | 2487 GET_BUFFER_SPACE (3); |
| 2488 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump | |
| 2489 : on_failure_jump, | |
| 446 | 2490 laststart, buf_end + 3); |
| 2491 buf_end += 3; | |
| 428 | 2492 |
| 2493 if (!zero_times_ok) | |
| 2494 { | |
| 2495 /* At least one repetition is required, so insert a | |
| 2496 `dummy_failure_jump' before the initial | |
| 2497 `on_failure_jump' instruction of the loop. This | |
| 2498 effects a skip over that instruction the first time | |
| 2499 we hit that loop. */ | |
| 2500 GET_BUFFER_SPACE (3); | |
| 2501 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); | |
| 446 | 2502 buf_end += 3; |
| 428 | 2503 } |
| 2504 } | |
| 2505 pending_exact = 0; | |
| 2506 } | |
| 2507 break; | |
| 2508 | |
| 2509 | |
| 2510 case '.': | |
| 446 | 2511 laststart = buf_end; |
| 428 | 2512 BUF_PUSH (anychar); |
| 2513 break; | |
| 2514 | |
| 2515 | |
| 2516 case '[': | |
| 2517 { | |
| 2518 /* XEmacs change: this whole section */ | |
| 460 | 2519 re_bool had_char_class = false; |
| 428 | 2520 #ifdef MULE |
| 460 | 2521 re_bool has_extended_chars = false; |
| 428 | 2522 REGISTER Lisp_Object rtab = Qnil; |
| 2523 #endif | |
| 2524 | |
| 2525 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
| 2526 | |
| 2527 /* Ensure that we have enough space to push a charset: the | |
| 2528 opcode, the length count, and the bitset; 34 bytes in all. */ | |
| 2529 GET_BUFFER_SPACE (34); | |
| 2530 | |
| 446 | 2531 laststart = buf_end; |
| 428 | 2532 |
| 2533 /* We test `*p == '^' twice, instead of using an if | |
| 2534 statement, so we only need one BUF_PUSH. */ | |
| 2535 BUF_PUSH (*p == '^' ? charset_not : charset); | |
| 2536 if (*p == '^') | |
| 2537 p++; | |
| 2538 | |
| 2539 /* Remember the first position in the bracket expression. */ | |
| 2540 p1 = p; | |
| 2541 | |
| 2542 /* Push the number of bytes in the bitmap. */ | |
| 2543 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); | |
| 2544 | |
| 2545 /* Clear the whole map. */ | |
| 446 | 2546 memset (buf_end, 0, (1 << BYTEWIDTH) / BYTEWIDTH); |
| 428 | 2547 |
| 2548 /* charset_not matches newline according to a syntax bit. */ | |
| 446 | 2549 if ((re_opcode_t) buf_end[-2] == charset_not |
| 428 | 2550 && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) |
| 2551 SET_LIST_BIT ('\n'); | |
| 2552 | |
| 2553 #ifdef MULE | |
| 2554 start_over_with_extended: | |
| 2555 if (has_extended_chars) | |
| 2556 { | |
| 2557 /* There are extended chars here, which means we need to start | |
| 2558 over and shift to unified range-table format. */ | |
| 446 | 2559 if (buf_end[-2] == charset) |
| 2560 buf_end[-2] = charset_mule; | |
| 428 | 2561 else |
| 446 | 2562 buf_end[-2] = charset_mule_not; |
| 2563 buf_end--; | |
| 428 | 2564 p = p1; /* go back to the beginning of the charset, after |
| 2565 a possible ^. */ | |
| 2566 rtab = Vthe_lisp_rangetab; | |
| 2567 Fclear_range_table (rtab); | |
| 2568 | |
| 2569 /* charset_not matches newline according to a syntax bit. */ | |
| 446 | 2570 if ((re_opcode_t) buf_end[-1] == charset_mule_not |
| 428 | 2571 && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) |
| 2572 SET_EITHER_BIT ('\n'); | |
| 2573 } | |
| 2574 #endif /* MULE */ | |
| 2575 | |
| 2576 /* Read in characters and ranges, setting map bits. */ | |
| 2577 for (;;) | |
| 2578 { | |
| 2579 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
| 2580 | |
| 446 | 2581 PATFETCH (c); |
| 428 | 2582 |
| 2583 #ifdef MULE | |
| 2584 if (c >= 0x80 && !has_extended_chars) | |
| 2585 { | |
| 2586 has_extended_chars = 1; | |
| 2587 /* Frumble-bumble, we've found some extended chars. | |
| 2588 Need to start over, process everything using | |
| 2589 the general extended-char mechanism, and need | |
| 2590 to use charset_mule and charset_mule_not instead | |
| 2591 of charset and charset_not. */ | |
| 2592 goto start_over_with_extended; | |
| 2593 } | |
| 2594 #endif /* MULE */ | |
| 2595 /* \ might escape characters inside [...] and [^...]. */ | |
| 2596 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') | |
| 2597 { | |
| 2598 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
| 2599 | |
| 446 | 2600 PATFETCH (c1); |
| 428 | 2601 #ifdef MULE |
| 2602 if (c1 >= 0x80 && !has_extended_chars) | |
| 2603 { | |
| 2604 has_extended_chars = 1; | |
| 2605 goto start_over_with_extended; | |
| 2606 } | |
| 2607 #endif /* MULE */ | |
| 2608 SET_EITHER_BIT (c1); | |
| 2609 continue; | |
| 2610 } | |
| 2611 | |
| 2612 /* Could be the end of the bracket expression. If it's | |
| 2613 not (i.e., when the bracket expression is `[]' so | |
| 2614 far), the ']' character bit gets set way below. */ | |
| 2615 if (c == ']' && p != p1 + 1) | |
| 2616 break; | |
| 2617 | |
| 2618 /* Look ahead to see if it's a range when the last thing | |
| 2619 was a character class. */ | |
| 2620 if (had_char_class && c == '-' && *p != ']') | |
| 2621 FREE_STACK_RETURN (REG_ERANGE); | |
| 2622 | |
| 2623 /* Look ahead to see if it's a range when the last thing | |
| 2624 was a character: if this is a hyphen not at the | |
| 2625 beginning or the end of a list, then it's the range | |
| 2626 operator. */ | |
| 2627 if (c == '-' | |
| 2628 && !(p - 2 >= pattern && p[-2] == '[') | |
| 446 | 2629 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') |
| 428 | 2630 && *p != ']') |
| 2631 { | |
| 2632 reg_errcode_t ret; | |
| 2633 | |
| 2634 #ifdef MULE | |
| 2635 if (* (unsigned char *) p >= 0x80 && !has_extended_chars) | |
| 2636 { | |
| 2637 has_extended_chars = 1; | |
| 2638 goto start_over_with_extended; | |
| 2639 } | |
| 2640 if (has_extended_chars) | |
| 2641 ret = compile_extended_range (&p, pend, translate, | |
| 2642 syntax, rtab); | |
| 2643 else | |
| 2644 #endif /* MULE */ | |
| 446 | 2645 ret = compile_range (&p, pend, translate, syntax, buf_end); |
| 428 | 2646 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
| 2647 } | |
| 2648 | |
| 2649 else if (p[0] == '-' && p[1] != ']') | |
| 2650 { /* This handles ranges made up of characters only. */ | |
| 2651 reg_errcode_t ret; | |
| 2652 | |
| 2653 /* Move past the `-'. */ | |
| 2654 PATFETCH (c1); | |
| 2655 | |
| 2656 #ifdef MULE | |
| 2657 if (* (unsigned char *) p >= 0x80 && !has_extended_chars) | |
| 2658 { | |
| 2659 has_extended_chars = 1; | |
| 2660 goto start_over_with_extended; | |
| 2661 } | |
| 2662 if (has_extended_chars) | |
| 2663 ret = compile_extended_range (&p, pend, translate, | |
| 2664 syntax, rtab); | |
| 2665 else | |
| 2666 #endif /* MULE */ | |
| 446 | 2667 ret = compile_range (&p, pend, translate, syntax, buf_end); |
| 428 | 2668 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
| 2669 } | |
| 2670 | |
| 2671 /* See if we're at the beginning of a possible character | |
| 2672 class. */ | |
| 2673 | |
| 2674 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') | |
| 2675 { /* Leave room for the null. */ | |
| 2676 char str[CHAR_CLASS_MAX_LENGTH + 1]; | |
| 2677 | |
| 2678 PATFETCH (c); | |
| 2679 c1 = 0; | |
| 2680 | |
| 2681 /* If pattern is `[[:'. */ | |
| 2682 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
| 2683 | |
| 2684 for (;;) | |
| 2685 { | |
| 446 | 2686 /* #### This code is unused. |
| 2687 Correctness is not checked after TRT | |
| 2688 table change. */ | |
| 428 | 2689 PATFETCH (c); |
| 2690 if (c == ':' || c == ']' || p == pend | |
| 2691 || c1 == CHAR_CLASS_MAX_LENGTH) | |
| 2692 break; | |
| 442 | 2693 str[c1++] = (char) c; |
| 428 | 2694 } |
| 2695 str[c1] = '\0'; | |
| 2696 | |
| 446 | 2697 /* If isn't a word bracketed by `[:' and `:]': |
| 428 | 2698 undo the ending character, the letters, and leave |
| 2699 the leading `:' and `[' (but set bits for them). */ | |
| 2700 if (c == ':' && *p == ']') | |
| 2701 { | |
| 2702 int ch; | |
| 460 | 2703 re_bool is_alnum = STREQ (str, "alnum"); |
| 2704 re_bool is_alpha = STREQ (str, "alpha"); | |
| 2705 re_bool is_blank = STREQ (str, "blank"); | |
| 2706 re_bool is_cntrl = STREQ (str, "cntrl"); | |
| 2707 re_bool is_digit = STREQ (str, "digit"); | |
| 2708 re_bool is_graph = STREQ (str, "graph"); | |
| 2709 re_bool is_lower = STREQ (str, "lower"); | |
| 2710 re_bool is_print = STREQ (str, "print"); | |
| 2711 re_bool is_punct = STREQ (str, "punct"); | |
| 2712 re_bool is_space = STREQ (str, "space"); | |
| 2713 re_bool is_upper = STREQ (str, "upper"); | |
| 2714 re_bool is_xdigit = STREQ (str, "xdigit"); | |
| 428 | 2715 |
| 2716 if (!IS_CHAR_CLASS (str)) | |
| 2717 FREE_STACK_RETURN (REG_ECTYPE); | |
| 2718 | |
| 2719 /* Throw away the ] at the end of the character | |
| 2720 class. */ | |
| 2721 PATFETCH (c); | |
| 2722 | |
| 2723 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); | |
| 2724 | |
| 2725 for (ch = 0; ch < 1 << BYTEWIDTH; ch++) | |
| 2726 { | |
| 2727 /* This was split into 3 if's to | |
| 2728 avoid an arbitrary limit in some compiler. */ | |
| 2729 if ( (is_alnum && ISALNUM (ch)) | |
| 2730 || (is_alpha && ISALPHA (ch)) | |
| 2731 || (is_blank && ISBLANK (ch)) | |
| 2732 || (is_cntrl && ISCNTRL (ch))) | |
| 2733 SET_EITHER_BIT (ch); | |
| 2734 if ( (is_digit && ISDIGIT (ch)) | |
| 2735 || (is_graph && ISGRAPH (ch)) | |
| 2736 || (is_lower && ISLOWER (ch)) | |
| 2737 || (is_print && ISPRINT (ch))) | |
| 2738 SET_EITHER_BIT (ch); | |
| 2739 if ( (is_punct && ISPUNCT (ch)) | |
| 2740 || (is_space && ISSPACE (ch)) | |
| 2741 || (is_upper && ISUPPER (ch)) | |
| 2742 || (is_xdigit && ISXDIGIT (ch))) | |
| 2743 SET_EITHER_BIT (ch); | |
| 2744 } | |
| 2745 had_char_class = true; | |
| 2746 } | |
| 2747 else | |
| 2748 { | |
| 2749 c1++; | |
| 2750 while (c1--) | |
| 2751 PATUNFETCH; | |
| 2752 SET_EITHER_BIT ('['); | |
| 2753 SET_EITHER_BIT (':'); | |
| 2754 had_char_class = false; | |
| 2755 } | |
| 2756 } | |
| 2757 else | |
| 2758 { | |
| 2759 had_char_class = false; | |
| 2760 SET_EITHER_BIT (c); | |
| 2761 } | |
| 2762 } | |
| 2763 | |
| 2764 #ifdef MULE | |
| 2765 if (has_extended_chars) | |
| 2766 { | |
| 2767 /* We have a range table, not a bit vector. */ | |
| 2768 int bytes_needed = | |
| 2769 unified_range_table_bytes_needed (rtab); | |
| 2770 GET_BUFFER_SPACE (bytes_needed); | |
| 446 | 2771 unified_range_table_copy_data (rtab, buf_end); |
| 2772 buf_end += unified_range_table_bytes_used (buf_end); | |
| 428 | 2773 break; |
| 2774 } | |
| 2775 #endif /* MULE */ | |
| 2776 /* Discard any (non)matching list bytes that are all 0 at the | |
| 2777 end of the map. Decrease the map-length byte too. */ | |
| 446 | 2778 while ((int) buf_end[-1] > 0 && buf_end[buf_end[-1] - 1] == 0) |
| 2779 buf_end[-1]--; | |
| 2780 buf_end += buf_end[-1]; | |
| 428 | 2781 } |
| 2782 break; | |
| 2783 | |
| 2784 | |
| 2785 case '(': | |
| 2786 if (syntax & RE_NO_BK_PARENS) | |
| 2787 goto handle_open; | |
| 2788 else | |
| 2789 goto normal_char; | |
| 2790 | |
| 2791 | |
| 2792 case ')': | |
| 2793 if (syntax & RE_NO_BK_PARENS) | |
| 2794 goto handle_close; | |
| 2795 else | |
| 2796 goto normal_char; | |
| 2797 | |
| 2798 | |
| 2799 case '\n': | |
| 2800 if (syntax & RE_NEWLINE_ALT) | |
| 2801 goto handle_alt; | |
| 2802 else | |
| 2803 goto normal_char; | |
| 2804 | |
| 2805 | |
| 2806 case '|': | |
| 2807 if (syntax & RE_NO_BK_VBAR) | |
| 2808 goto handle_alt; | |
| 2809 else | |
| 2810 goto normal_char; | |
| 2811 | |
| 2812 | |
| 2813 case '{': | |
| 2814 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) | |
| 2815 goto handle_interval; | |
| 2816 else | |
| 2817 goto normal_char; | |
| 2818 | |
| 2819 | |
| 2820 case '\\': | |
| 2821 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); | |
| 2822 | |
| 2823 /* Do not translate the character after the \, so that we can | |
| 2824 distinguish, e.g., \B from \b, even if we normally would | |
| 2825 translate, e.g., B to b. */ | |
| 2826 PATFETCH_RAW (c); | |
| 2827 | |
| 2828 switch (c) | |
| 2829 { | |
| 2830 case '(': | |
| 2831 if (syntax & RE_NO_BK_PARENS) | |
| 2832 goto normal_backslash; | |
| 2833 | |
| 2834 handle_open: | |
| 2835 { | |
| 2836 regnum_t r; | |
| 502 | 2837 int shy = 0; |
| 428 | 2838 |
| 2839 if (!(syntax & RE_NO_SHY_GROUPS) | |
| 2840 && p != pend | |
| 446 | 2841 && *p == '?') |
| 428 | 2842 { |
| 2843 p++; | |
| 446 | 2844 PATFETCH (c); |
| 428 | 2845 switch (c) |
| 2846 { | |
| 2847 case ':': /* shy groups */ | |
| 502 | 2848 shy = 1; |
| 428 | 2849 break; |
| 2850 | |
| 2851 /* All others are reserved for future constructs. */ | |
| 2852 default: | |
| 2853 FREE_STACK_RETURN (REG_BADPAT); | |
| 2854 } | |
| 2855 } | |
| 502 | 2856 |
| 2857 r = ++regnum; | |
| 2858 bufp->re_ngroups++; | |
| 2859 if (!shy) | |
| 2860 { | |
| 2861 bufp->re_nsub++; | |
| 2862 while (bufp->external_to_internal_register_size <= | |
| 2863 bufp->re_nsub) | |
| 2864 { | |
| 2865 int i; | |
| 2866 int old_size = | |
| 2867 bufp->external_to_internal_register_size; | |
| 2868 bufp->external_to_internal_register_size += 5; | |
| 2869 RETALLOC (bufp->external_to_internal_register, | |
| 2870 bufp->external_to_internal_register_size, | |
| 2871 int); | |
| 2872 /* debugging */ | |
| 2873 for (i = old_size; | |
| 2874 i < bufp->external_to_internal_register_size; i++) | |
| 2875 bufp->external_to_internal_register[i] = | |
| 2876 (int) 0xDEADBEEF; | |
| 2877 } | |
| 2878 | |
| 2879 bufp->external_to_internal_register[bufp->re_nsub] = | |
| 2880 bufp->re_ngroups; | |
| 2881 } | |
| 428 | 2882 |
| 2883 if (COMPILE_STACK_FULL) | |
| 2884 { | |
| 2885 RETALLOC (compile_stack.stack, compile_stack.size << 1, | |
| 2886 compile_stack_elt_t); | |
| 2887 if (compile_stack.stack == NULL) return REG_ESPACE; | |
| 2888 | |
| 2889 compile_stack.size <<= 1; | |
| 2890 } | |
| 2891 | |
| 2892 /* These are the values to restore when we hit end of this | |
| 2893 group. They are all relative offsets, so that if the | |
| 2894 whole pattern moves because of realloc, they will still | |
| 2895 be valid. */ | |
| 2896 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; | |
| 2897 COMPILE_STACK_TOP.fixup_alt_jump | |
| 2898 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; | |
| 446 | 2899 COMPILE_STACK_TOP.laststart_offset = buf_end - bufp->buffer; |
| 428 | 2900 COMPILE_STACK_TOP.regnum = r; |
| 2901 | |
| 2902 /* We will eventually replace the 0 with the number of | |
| 2903 groups inner to this one. But do not push a | |
| 2904 start_memory for groups beyond the last one we can | |
| 502 | 2905 represent in the compiled pattern. |
| 2906 #### bad bad bad. this will fail in lots of ways, if we | |
| 2907 ever have to backtrack for these groups. | |
| 2908 */ | |
| 428 | 2909 if (r <= MAX_REGNUM) |
| 2910 { | |
| 2911 COMPILE_STACK_TOP.inner_group_offset | |
| 446 | 2912 = buf_end - bufp->buffer + 2; |
| 428 | 2913 BUF_PUSH_3 (start_memory, r, 0); |
| 2914 } | |
| 2915 | |
| 2916 compile_stack.avail++; | |
| 2917 | |
| 2918 fixup_alt_jump = 0; | |
| 2919 laststart = 0; | |
| 446 | 2920 begalt = buf_end; |
| 428 | 2921 /* If we've reached MAX_REGNUM groups, then this open |
| 2922 won't actually generate any code, so we'll have to | |
| 2923 clear pending_exact explicitly. */ | |
| 2924 pending_exact = 0; | |
| 2925 } | |
| 2926 break; | |
| 2927 | |
| 2928 | |
| 2929 case ')': | |
| 2930 if (syntax & RE_NO_BK_PARENS) goto normal_backslash; | |
| 2931 | |
| 2932 if (COMPILE_STACK_EMPTY) { | |
| 2933 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
| 2934 goto normal_backslash; | |
| 2935 else | |
| 2936 FREE_STACK_RETURN (REG_ERPAREN); | |
| 2937 } | |
| 2938 | |
| 2939 handle_close: | |
| 2940 if (fixup_alt_jump) | |
| 2941 { /* Push a dummy failure point at the end of the | |
| 2942 alternative for a possible future | |
| 2943 `pop_failure_jump' to pop. See comments at | |
| 2944 `push_dummy_failure' in `re_match_2'. */ | |
| 2945 BUF_PUSH (push_dummy_failure); | |
| 2946 | |
| 2947 /* We allocated space for this jump when we assigned | |
| 2948 to `fixup_alt_jump', in the `handle_alt' case below. */ | |
| 446 | 2949 STORE_JUMP (jump_past_alt, fixup_alt_jump, buf_end - 1); |
| 428 | 2950 } |
| 2951 | |
| 2952 /* See similar code for backslashed left paren above. */ | |
| 2953 if (COMPILE_STACK_EMPTY) { | |
| 2954 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) | |
| 2955 goto normal_char; | |
| 2956 else | |
| 2957 FREE_STACK_RETURN (REG_ERPAREN); | |
| 2958 } | |
| 2959 | |
| 2960 /* Since we just checked for an empty stack above, this | |
| 2961 ``can't happen''. */ | |
| 2962 assert (compile_stack.avail != 0); | |
| 2963 { | |
| 2964 /* We don't just want to restore into `regnum', because | |
| 2965 later groups should continue to be numbered higher, | |
| 2966 as in `(ab)c(de)' -- the second group is #2. */ | |
| 2967 regnum_t this_group_regnum; | |
| 2968 | |
| 2969 compile_stack.avail--; | |
| 2970 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; | |
| 2971 fixup_alt_jump | |
| 2972 = COMPILE_STACK_TOP.fixup_alt_jump | |
| 2973 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 | |
| 2974 : 0; | |
| 2975 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; | |
| 2976 this_group_regnum = COMPILE_STACK_TOP.regnum; | |
| 2977 /* If we've reached MAX_REGNUM groups, then this open | |
| 2978 won't actually generate any code, so we'll have to | |
| 2979 clear pending_exact explicitly. */ | |
| 2980 pending_exact = 0; | |
| 2981 | |
| 2982 /* We're at the end of the group, so now we know how many | |
| 2983 groups were inside this one. */ | |
| 2984 if (this_group_regnum <= MAX_REGNUM) | |
| 2985 { | |
| 2986 unsigned char *inner_group_loc | |
| 2987 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; | |
| 2988 | |
| 2989 *inner_group_loc = regnum - this_group_regnum; | |
| 2990 BUF_PUSH_3 (stop_memory, this_group_regnum, | |
| 2991 regnum - this_group_regnum); | |
| 2992 } | |
| 2993 } | |
| 2994 break; | |
| 2995 | |
| 2996 | |
| 2997 case '|': /* `\|'. */ | |
| 2998 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) | |
| 2999 goto normal_backslash; | |
| 3000 handle_alt: | |
| 3001 if (syntax & RE_LIMITED_OPS) | |
| 3002 goto normal_char; | |
| 3003 | |
| 3004 /* Insert before the previous alternative a jump which | |
| 3005 jumps to this alternative if the former fails. */ | |
| 3006 GET_BUFFER_SPACE (3); | |
| 446 | 3007 INSERT_JUMP (on_failure_jump, begalt, buf_end + 6); |
| 428 | 3008 pending_exact = 0; |
| 446 | 3009 buf_end += 3; |
| 428 | 3010 |
| 3011 /* The alternative before this one has a jump after it | |
| 3012 which gets executed if it gets matched. Adjust that | |
| 3013 jump so it will jump to this alternative's analogous | |
| 3014 jump (put in below, which in turn will jump to the next | |
| 3015 (if any) alternative's such jump, etc.). The last such | |
| 3016 jump jumps to the correct final destination. A picture: | |
| 3017 _____ _____ | |
| 3018 | | | | | |
| 3019 | v | v | |
| 3020 a | b | c | |
| 3021 | |
| 3022 If we are at `b', then fixup_alt_jump right now points to a | |
| 3023 three-byte space after `a'. We'll put in the jump, set | |
| 3024 fixup_alt_jump to right after `b', and leave behind three | |
| 3025 bytes which we'll fill in when we get to after `c'. */ | |
| 3026 | |
| 3027 if (fixup_alt_jump) | |
| 446 | 3028 STORE_JUMP (jump_past_alt, fixup_alt_jump, buf_end); |
| 428 | 3029 |
| 3030 /* Mark and leave space for a jump after this alternative, | |
| 3031 to be filled in later either by next alternative or | |
| 3032 when know we're at the end of a series of alternatives. */ | |
| 446 | 3033 fixup_alt_jump = buf_end; |
| 428 | 3034 GET_BUFFER_SPACE (3); |
| 446 | 3035 buf_end += 3; |
| 428 | 3036 |
| 3037 laststart = 0; | |
| 446 | 3038 begalt = buf_end; |
| 428 | 3039 break; |
| 3040 | |
| 3041 | |
| 3042 case '{': | |
| 3043 /* If \{ is a literal. */ | |
| 3044 if (!(syntax & RE_INTERVALS) | |
| 3045 /* If we're at `\{' and it's not the open-interval | |
| 3046 operator. */ | |
| 3047 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) | |
| 3048 || (p - 2 == pattern && p == pend)) | |
| 3049 goto normal_backslash; | |
| 3050 | |
| 3051 handle_interval: | |
| 3052 { | |
| 3053 /* If got here, then the syntax allows intervals. */ | |
| 3054 | |
| 3055 /* At least (most) this many matches must be made. */ | |
| 3056 int lower_bound = -1, upper_bound = -1; | |
| 3057 | |
| 3058 beg_interval = p - 1; | |
| 3059 | |
| 3060 if (p == pend) | |
| 3061 { | |
| 3062 if (syntax & RE_NO_BK_BRACES) | |
| 3063 goto unfetch_interval; | |
| 3064 else | |
| 3065 FREE_STACK_RETURN (REG_EBRACE); | |
| 3066 } | |
| 3067 | |
| 3068 GET_UNSIGNED_NUMBER (lower_bound); | |
| 3069 | |
| 3070 if (c == ',') | |
| 3071 { | |
| 3072 GET_UNSIGNED_NUMBER (upper_bound); | |
| 3073 if (upper_bound < 0) upper_bound = RE_DUP_MAX; | |
| 3074 } | |
| 3075 else | |
| 3076 /* Interval such as `{1}' => match exactly once. */ | |
| 3077 upper_bound = lower_bound; | |
| 3078 | |
| 3079 if (lower_bound < 0 || upper_bound > RE_DUP_MAX | |
| 3080 || lower_bound > upper_bound) | |
| 3081 { | |
| 3082 if (syntax & RE_NO_BK_BRACES) | |
| 3083 goto unfetch_interval; | |
| 3084 else | |
| 3085 FREE_STACK_RETURN (REG_BADBR); | |
| 3086 } | |
| 3087 | |
| 3088 if (!(syntax & RE_NO_BK_BRACES)) | |
| 3089 { | |
| 3090 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE); | |
| 3091 | |
| 3092 PATFETCH (c); | |
| 3093 } | |
| 3094 | |
| 3095 if (c != '}') | |
| 3096 { | |
| 3097 if (syntax & RE_NO_BK_BRACES) | |
| 3098 goto unfetch_interval; | |
| 3099 else | |
| 3100 FREE_STACK_RETURN (REG_BADBR); | |
| 3101 } | |
| 3102 | |
| 3103 /* We just parsed a valid interval. */ | |
| 3104 | |
| 3105 /* If it's invalid to have no preceding re. */ | |
| 3106 if (!laststart) | |
| 3107 { | |
| 3108 if (syntax & RE_CONTEXT_INVALID_OPS) | |
| 3109 FREE_STACK_RETURN (REG_BADRPT); | |
| 3110 else if (syntax & RE_CONTEXT_INDEP_OPS) | |
| 446 | 3111 laststart = buf_end; |
| 428 | 3112 else |
| 3113 goto unfetch_interval; | |
| 3114 } | |
| 3115 | |
| 3116 /* If the upper bound is zero, don't want to succeed at | |
| 3117 all; jump from `laststart' to `b + 3', which will be | |
| 3118 the end of the buffer after we insert the jump. */ | |
| 3119 if (upper_bound == 0) | |
| 3120 { | |
| 3121 GET_BUFFER_SPACE (3); | |
| 446 | 3122 INSERT_JUMP (jump, laststart, buf_end + 3); |
| 3123 buf_end += 3; | |
| 428 | 3124 } |
| 3125 | |
| 3126 /* Otherwise, we have a nontrivial interval. When | |
| 3127 we're all done, the pattern will look like: | |
| 3128 set_number_at <jump count> <upper bound> | |
| 3129 set_number_at <succeed_n count> <lower bound> | |
| 3130 succeed_n <after jump addr> <succeed_n count> | |
| 3131 <body of loop> | |
| 3132 jump_n <succeed_n addr> <jump count> | |
| 3133 (The upper bound and `jump_n' are omitted if | |
| 3134 `upper_bound' is 1, though.) */ | |
| 3135 else | |
| 3136 { /* If the upper bound is > 1, we need to insert | |
| 3137 more at the end of the loop. */ | |
| 647 | 3138 int nbytes = 10 + (upper_bound > 1) * 10; |
| 428 | 3139 |
| 3140 GET_BUFFER_SPACE (nbytes); | |
| 3141 | |
| 3142 /* Initialize lower bound of the `succeed_n', even | |
| 3143 though it will be set during matching by its | |
| 3144 attendant `set_number_at' (inserted next), | |
| 3145 because `re_compile_fastmap' needs to know. | |
| 3146 Jump to the `jump_n' we might insert below. */ | |
| 3147 INSERT_JUMP2 (succeed_n, laststart, | |
| 446 | 3148 buf_end + 5 + (upper_bound > 1) * 5, |
| 428 | 3149 lower_bound); |
| 446 | 3150 buf_end += 5; |
| 428 | 3151 |
| 3152 /* Code to initialize the lower bound. Insert | |
| 3153 before the `succeed_n'. The `5' is the last two | |
| 3154 bytes of this `set_number_at', plus 3 bytes of | |
| 3155 the following `succeed_n'. */ | |
| 446 | 3156 insert_op2 (set_number_at, laststart, 5, lower_bound, buf_end); |
| 3157 buf_end += 5; | |
| 428 | 3158 |
| 3159 if (upper_bound > 1) | |
| 3160 { /* More than one repetition is allowed, so | |
| 3161 append a backward jump to the `succeed_n' | |
| 3162 that starts this interval. | |
| 3163 | |
| 3164 When we've reached this during matching, | |
| 3165 we'll have matched the interval once, so | |
| 3166 jump back only `upper_bound - 1' times. */ | |
| 446 | 3167 STORE_JUMP2 (jump_n, buf_end, laststart + 5, |
| 428 | 3168 upper_bound - 1); |
| 446 | 3169 buf_end += 5; |
| 428 | 3170 |
| 3171 /* The location we want to set is the second | |
| 3172 parameter of the `jump_n'; that is `b-2' as | |
| 3173 an absolute address. `laststart' will be | |
| 3174 the `set_number_at' we're about to insert; | |
| 3175 `laststart+3' the number to set, the source | |
| 3176 for the relative address. But we are | |
| 3177 inserting into the middle of the pattern -- | |
| 3178 so everything is getting moved up by 5. | |
| 3179 Conclusion: (b - 2) - (laststart + 3) + 5, | |
| 3180 i.e., b - laststart. | |
| 3181 | |
| 3182 We insert this at the beginning of the loop | |
| 3183 so that if we fail during matching, we'll | |
| 3184 reinitialize the bounds. */ | |
| 446 | 3185 insert_op2 (set_number_at, laststart, |
| 3186 buf_end - laststart, | |
| 3187 upper_bound - 1, buf_end); | |
| 3188 buf_end += 5; | |
| 428 | 3189 } |
| 3190 } | |
| 3191 pending_exact = 0; | |
| 3192 beg_interval = NULL; | |
| 3193 } | |
| 3194 break; | |
| 3195 | |
| 3196 unfetch_interval: | |
| 3197 /* If an invalid interval, match the characters as literals. */ | |
| 3198 assert (beg_interval); | |
| 3199 p = beg_interval; | |
| 3200 beg_interval = NULL; | |
| 3201 | |
| 3202 /* normal_char and normal_backslash need `c'. */ | |
| 3203 PATFETCH (c); | |
| 3204 | |
| 3205 if (!(syntax & RE_NO_BK_BRACES)) | |
| 3206 { | |
| 3207 if (p > pattern && p[-1] == '\\') | |
| 3208 goto normal_backslash; | |
| 3209 } | |
| 3210 goto normal_char; | |
| 3211 | |
| 3212 #ifdef emacs | |
| 3213 /* There is no way to specify the before_dot and after_dot | |
| 3214 operators. rms says this is ok. --karl */ | |
| 3215 case '=': | |
| 3216 BUF_PUSH (at_dot); | |
| 3217 break; | |
| 3218 | |
| 3219 case 's': | |
| 446 | 3220 laststart = buf_end; |
| 428 | 3221 PATFETCH (c); |
| 3222 /* XEmacs addition */ | |
| 3223 if (c >= 0x80 || syntax_spec_code[c] == 0377) | |
| 3224 FREE_STACK_RETURN (REG_ESYNTAX); | |
| 3225 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); | |
| 3226 break; | |
| 3227 | |
| 3228 case 'S': | |
| 446 | 3229 laststart = buf_end; |
| 428 | 3230 PATFETCH (c); |
| 3231 /* XEmacs addition */ | |
| 3232 if (c >= 0x80 || syntax_spec_code[c] == 0377) | |
| 3233 FREE_STACK_RETURN (REG_ESYNTAX); | |
| 3234 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); | |
| 3235 break; | |
| 3236 | |
| 3237 #ifdef MULE | |
| 3238 /* 97.2.17 jhod merged in to XEmacs from mule-2.3 */ | |
| 3239 case 'c': | |
| 446 | 3240 laststart = buf_end; |
| 428 | 3241 PATFETCH_RAW (c); |
| 3242 if (c < 32 || c > 127) | |
| 3243 FREE_STACK_RETURN (REG_ECATEGORY); | |
| 3244 BUF_PUSH_2 (categoryspec, c); | |
| 3245 break; | |
| 3246 | |
| 3247 case 'C': | |
| 446 | 3248 laststart = buf_end; |
| 428 | 3249 PATFETCH_RAW (c); |
| 3250 if (c < 32 || c > 127) | |
| 3251 FREE_STACK_RETURN (REG_ECATEGORY); | |
| 3252 BUF_PUSH_2 (notcategoryspec, c); | |
| 3253 break; | |
| 3254 /* end of category patch */ | |
| 3255 #endif /* MULE */ | |
| 3256 #endif /* emacs */ | |
| 3257 | |
| 3258 | |
| 3259 case 'w': | |
| 446 | 3260 laststart = buf_end; |
| 428 | 3261 BUF_PUSH (wordchar); |
| 3262 break; | |
| 3263 | |
| 3264 | |
| 3265 case 'W': | |
| 446 | 3266 laststart = buf_end; |
| 428 | 3267 BUF_PUSH (notwordchar); |
| 3268 break; | |
| 3269 | |
| 3270 | |
| 3271 case '<': | |
| 3272 BUF_PUSH (wordbeg); | |
| 3273 break; | |
| 3274 | |
| 3275 case '>': | |
| 3276 BUF_PUSH (wordend); | |
| 3277 break; | |
| 3278 | |
| 3279 case 'b': | |
| 3280 BUF_PUSH (wordbound); | |
| 3281 break; | |
| 3282 | |
| 3283 case 'B': | |
| 3284 BUF_PUSH (notwordbound); | |
| 3285 break; | |
| 3286 | |
| 3287 case '`': | |
| 3288 BUF_PUSH (begbuf); | |
| 3289 break; | |
| 3290 | |
| 3291 case '\'': | |
| 3292 BUF_PUSH (endbuf); | |
| 3293 break; | |
| 3294 | |
| 3295 case '1': case '2': case '3': case '4': case '5': | |
| 3296 case '6': case '7': case '8': case '9': | |
| 446 | 3297 { |
| 502 | 3298 regnum_t reg, regint; |
| 3299 int may_need_to_unfetch = 0; | |
| 446 | 3300 if (syntax & RE_NO_BK_REFS) |
| 3301 goto normal_char; | |
| 3302 | |
| 502 | 3303 /* This only goes up to 99. It could be extended to work |
| 3304 up to 255 (the maximum number of registers that can be | |
| 3305 handled by the current regexp engine, because it stores | |
| 3306 its register numbers in the compiled pattern as one byte, | |
| 3307 ugh). Doing that's a bit trickier, because you might | |
| 3308 have the case where \25 a back-ref but \255 is not, ... */ | |
| 446 | 3309 reg = c - '0'; |
| 502 | 3310 if (p < pend) |
| 3311 { | |
| 3312 PATFETCH (c); | |
| 3313 if (c >= '0' && c <= '9') | |
| 3314 { | |
| 3315 regnum_t new_reg = reg * 10 + c - '0'; | |
| 3316 if (new_reg <= bufp->re_nsub) | |
| 3317 { | |
| 3318 reg = new_reg; | |
| 3319 may_need_to_unfetch = 1; | |
| 3320 } | |
| 3321 else | |
| 3322 PATUNFETCH; | |
| 3323 } | |
| 523 | 3324 else |
| 3325 PATUNFETCH; | |
| 502 | 3326 } |
| 3327 | |
| 3328 if (reg > bufp->re_nsub) | |
| 446 | 3329 FREE_STACK_RETURN (REG_ESUBREG); |
| 3330 | |
| 502 | 3331 regint = bufp->external_to_internal_register[reg]; |
| 446 | 3332 /* Can't back reference to a subexpression if inside of it. */ |
| 502 | 3333 if (group_in_compile_stack (compile_stack, regint)) |
| 3334 { | |
| 3335 if (may_need_to_unfetch) | |
| 3336 PATUNFETCH; | |
| 3337 goto normal_char; | |
| 3338 } | |
| 3339 | |
| 3340 #ifdef emacs | |
| 3341 if (reg > 9 && | |
| 3342 bufp->warned_about_incompatible_back_references == 0) | |
| 3343 { | |
| 3344 bufp->warned_about_incompatible_back_references = 1; | |
| 3345 warn_when_safe (intern ("regex"), Qinfo, | |
| 3346 "Back reference \\%d now has new " | |
| 3347 "semantics in %s", reg, pattern); | |
| 3348 } | |
| 3349 #endif | |
| 446 | 3350 |
| 3351 laststart = buf_end; | |
| 502 | 3352 BUF_PUSH_2 (duplicate, regint); |
| 446 | 3353 } |
| 428 | 3354 break; |
| 3355 | |
| 3356 | |
| 3357 case '+': | |
| 3358 case '?': | |
| 3359 if (syntax & RE_BK_PLUS_QM) | |
| 3360 goto handle_plus; | |
| 3361 else | |
| 3362 goto normal_backslash; | |
| 3363 | |
| 3364 default: | |
| 3365 normal_backslash: | |
| 3366 /* You might think it would be useful for \ to mean | |
| 3367 not to translate; but if we don't translate it, | |
| 3368 it will never match anything. */ | |
| 826 | 3369 c = RE_TRANSLATE (c); |
| 428 | 3370 goto normal_char; |
| 3371 } | |
| 3372 break; | |
| 3373 | |
| 3374 | |
| 3375 default: | |
| 3376 /* Expects the character in `c'. */ | |
| 3377 /* `p' points to the location after where `c' came from. */ | |
| 3378 normal_char: | |
| 3379 { | |
|
4750
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3380 /* The following conditional synced to GNU Emacs 22.1. */ |
| 428 | 3381 /* If no exactn currently being built. */ |
| 3382 if (!pending_exact | |
| 3383 | |
| 3384 /* If last exactn not at current position. */ | |
| 446 | 3385 || pending_exact + *pending_exact + 1 != buf_end |
| 428 | 3386 |
| 3387 /* We have only one byte following the exactn for the count. */ | |
|
4750
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3388 || *pending_exact >= (1 << BYTEWIDTH) - MAX_ICHAR_LEN |
|
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3389 |
|
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3390 /* If followed by a repetition operator. |
|
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3391 If the lookahead fails because of end of pattern, any |
|
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3392 trailing backslash will get caught later. */ |
|
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3393 || (p != pend && (*p == '*' || *p == '^')) |
| 428 | 3394 || ((syntax & RE_BK_PLUS_QM) |
|
4750
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3395 ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?') |
|
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3396 : p != pend && (*p == '+' || *p == '?')) |
| 428 | 3397 || ((syntax & RE_INTERVALS) |
| 3398 && ((syntax & RE_NO_BK_BRACES) | |
|
4750
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3399 ? p != pend && *p == '{' |
|
b5f21bb36684
Fix crash in regex.c (closes issue630).
Stephen J. Turnbull <stephen@xemacs.org>
parents:
4527
diff
changeset
|
3400 : p + 1 < pend && (p[0] == '\\' && p[1] == '{')))) |
| 428 | 3401 { |
| 3402 /* Start building a new exactn. */ | |
| 3403 | |
| 446 | 3404 laststart = buf_end; |
| 428 | 3405 |
| 3406 BUF_PUSH_2 (exactn, 0); | |
| 446 | 3407 pending_exact = buf_end - 1; |
| 428 | 3408 } |
| 3409 | |
| 446 | 3410 #ifndef MULE |
| 428 | 3411 BUF_PUSH (c); |
| 3412 (*pending_exact)++; | |
| 446 | 3413 #else |
| 3414 { | |
| 3415 Bytecount bt_count; | |
| 867 | 3416 Ibyte tmp_buf[MAX_ICHAR_LEN]; |
| 446 | 3417 int i; |
| 3418 | |
| 867 | 3419 bt_count = set_itext_ichar (tmp_buf, c); |
| 446 | 3420 |
| 3421 for (i = 0; i < bt_count; i++) | |
| 3422 { | |
| 3423 BUF_PUSH (tmp_buf[i]); | |
| 3424 (*pending_exact)++; | |
| 3425 } | |
| 3426 } | |
| 3427 #endif | |
| 428 | 3428 break; |
| 3429 } | |
| 3430 } /* switch (c) */ | |
| 3431 } /* while p != pend */ | |
| 3432 | |
| 3433 | |
| 3434 /* Through the pattern now. */ | |
| 3435 | |
| 3436 if (fixup_alt_jump) | |
| 446 | 3437 STORE_JUMP (jump_past_alt, fixup_alt_jump, buf_end); |
| 428 | 3438 |
| 3439 if (!COMPILE_STACK_EMPTY) | |
| 3440 FREE_STACK_RETURN (REG_EPAREN); | |
| 3441 | |
| 3442 /* If we don't want backtracking, force success | |
| 3443 the first time we reach the end of the compiled pattern. */ | |
| 3444 if (syntax & RE_NO_POSIX_BACKTRACKING) | |
| 3445 BUF_PUSH (succeed); | |
| 3446 | |
|
4976
16112448d484
Rename xfree(FOO, TYPE) -> xfree(FOO)
Ben Wing <ben@xemacs.org>
parents:
4832
diff
changeset
|
3447 xfree (compile_stack.stack); |
| 428 | 3448 |
| 3449 /* We have succeeded; set the length of the buffer. */ | |
| 446 | 3450 bufp->used = buf_end - bufp->buffer; |
| 428 | 3451 |
| 3452 #ifdef DEBUG | |
| 5041 | 3453 if (debug_regexps & RE_DEBUG_COMPILATION) |
| 428 | 3454 { |
| 3455 DEBUG_PRINT1 ("\nCompiled pattern: \n"); | |
| 3456 print_compiled_pattern (bufp); | |
| 3457 } | |
| 3458 #endif /* DEBUG */ | |
| 3459 | |
| 3460 #ifndef MATCH_MAY_ALLOCATE | |
| 3461 /* Initialize the failure stack to the largest possible stack. This | |
| 3462 isn't necessary unless we're trying to avoid calling alloca in | |
| 3463 the search and match routines. */ | |
| 3464 { | |
| 502 | 3465 int num_regs = bufp->re_ngroups + 1; |
| 428 | 3466 |
| 3467 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size | |
| 3468 is strictly greater than re_max_failures, the largest possible stack | |
| 3469 is 2 * re_max_failures failure points. */ | |
| 3470 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) | |
| 3471 { | |
| 3472 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); | |
| 3473 | |
| 3474 if (! fail_stack.stack) | |
| 3475 fail_stack.stack | |
| 3476 = (fail_stack_elt_t *) xmalloc (fail_stack.size | |
| 3477 * sizeof (fail_stack_elt_t)); | |
| 3478 else | |
| 3479 fail_stack.stack | |
| 3480 = (fail_stack_elt_t *) xrealloc (fail_stack.stack, | |
| 3481 (fail_stack.size | |
| 3482 * sizeof (fail_stack_elt_t))); | |
| 3483 } | |
| 3484 | |
| 3485 regex_grow_registers (num_regs); | |
| 3486 } | |
| 3487 #endif /* not MATCH_MAY_ALLOCATE */ | |
| 3488 | |
| 3489 return REG_NOERROR; | |
| 3490 } /* regex_compile */ | |
| 3491 | |
| 3492 /* Subroutines for `regex_compile'. */ | |
| 3493 | |
| 3494 /* Store OP at LOC followed by two-byte integer parameter ARG. */ | |
| 3495 | |
| 3496 static void | |
| 3497 store_op1 (re_opcode_t op, unsigned char *loc, int arg) | |
| 3498 { | |
| 3499 *loc = (unsigned char) op; | |
| 3500 STORE_NUMBER (loc + 1, arg); | |
| 3501 } | |
| 3502 | |
| 3503 | |
| 3504 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
| 3505 | |
| 3506 static void | |
| 3507 store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2) | |
| 3508 { | |
| 3509 *loc = (unsigned char) op; | |
| 3510 STORE_NUMBER (loc + 1, arg1); | |
| 3511 STORE_NUMBER (loc + 3, arg2); | |
| 3512 } | |
| 3513 | |
| 3514 | |
| 3515 /* Copy the bytes from LOC to END to open up three bytes of space at LOC | |
| 3516 for OP followed by two-byte integer parameter ARG. */ | |
| 3517 | |
| 3518 static void | |
| 3519 insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end) | |
| 3520 { | |
| 3521 REGISTER unsigned char *pfrom = end; | |
| 3522 REGISTER unsigned char *pto = end + 3; | |
| 3523 | |
| 3524 while (pfrom != loc) | |
| 3525 *--pto = *--pfrom; | |
| 3526 | |
| 3527 store_op1 (op, loc, arg); | |
| 3528 } | |
| 3529 | |
| 3530 | |
| 3531 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ | |
| 3532 | |
| 3533 static void | |
| 3534 insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, | |
| 3535 unsigned char *end) | |
| 3536 { | |
| 3537 REGISTER unsigned char *pfrom = end; | |
| 3538 REGISTER unsigned char *pto = end + 5; | |
| 3539 | |
| 3540 while (pfrom != loc) | |
| 3541 *--pto = *--pfrom; | |
| 3542 | |
| 3543 store_op2 (op, loc, arg1, arg2); | |
| 3544 } | |
| 3545 | |
| 3546 | |
| 3547 /* P points to just after a ^ in PATTERN. Return true if that ^ comes | |
| 3548 after an alternative or a begin-subexpression. We assume there is at | |
| 3549 least one character before the ^. */ | |
| 3550 | |
| 460 | 3551 static re_bool |
| 446 | 3552 at_begline_loc_p (re_char *pattern, re_char *p, reg_syntax_t syntax) |
| 428 | 3553 { |
| 446 | 3554 re_char *prev = p - 2; |
| 460 | 3555 re_bool prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
| 428 | 3556 |
| 3557 return | |
| 3558 /* After a subexpression? */ | |
| 3559 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) | |
| 3560 /* After an alternative? */ | |
| 3561 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); | |
| 3562 } | |
| 3563 | |
| 3564 | |
| 3565 /* The dual of at_begline_loc_p. This one is for $. We assume there is | |
| 3566 at least one character after the $, i.e., `P < PEND'. */ | |
| 3567 | |
| 460 | 3568 static re_bool |
| 446 | 3569 at_endline_loc_p (re_char *p, re_char *pend, int syntax) |
| 428 | 3570 { |
| 446 | 3571 re_char *next = p; |
| 460 | 3572 re_bool next_backslash = *next == '\\'; |
| 446 | 3573 re_char *next_next = p + 1 < pend ? p + 1 : 0; |
| 428 | 3574 |
| 3575 return | |
| 3576 /* Before a subexpression? */ | |
| 3577 (syntax & RE_NO_BK_PARENS ? *next == ')' | |
| 3578 : next_backslash && next_next && *next_next == ')') | |
| 3579 /* Before an alternative? */ | |
| 3580 || (syntax & RE_NO_BK_VBAR ? *next == '|' | |
| 3581 : next_backslash && next_next && *next_next == '|'); | |
| 3582 } | |
| 3583 | |
| 3584 | |
| 3585 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and | |
| 3586 false if it's not. */ | |
| 3587 | |
| 460 | 3588 static re_bool |
| 428 | 3589 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum) |
| 3590 { | |
| 3591 int this_element; | |
| 3592 | |
| 3593 for (this_element = compile_stack.avail - 1; | |
| 3594 this_element >= 0; | |
| 3595 this_element--) | |
| 3596 if (compile_stack.stack[this_element].regnum == regnum) | |
| 3597 return true; | |
| 3598 | |
| 3599 return false; | |
| 3600 } | |
| 3601 | |
| 3602 | |
| 3603 /* Read the ending character of a range (in a bracket expression) from the | |
| 3604 uncompiled pattern *P_PTR (which ends at PEND). We assume the | |
| 3605 starting character is in `P[-2]'. (`P[-1]' is the character `-'.) | |
| 3606 Then we set the translation of all bits between the starting and | |
| 3607 ending characters (inclusive) in the compiled pattern B. | |
| 3608 | |
| 3609 Return an error code. | |
| 3610 | |
| 3611 We use these short variable names so we can use the same macros as | |
| 826 | 3612 `regex_compile' itself. |
| 3613 | |
| 3614 Under Mule, this is only called when both chars of the range are | |
| 3615 ASCII. */ | |
| 428 | 3616 |
| 3617 static reg_errcode_t | |
| 446 | 3618 compile_range (re_char **p_ptr, re_char *pend, RE_TRANSLATE_TYPE translate, |
| 3619 reg_syntax_t syntax, unsigned char *buf_end) | |
| 428 | 3620 { |
| 867 | 3621 Ichar this_char; |
| 428 | 3622 |
| 446 | 3623 re_char *p = *p_ptr; |
| 428 | 3624 int range_start, range_end; |
| 3625 | |
| 3626 if (p == pend) | |
| 3627 return REG_ERANGE; | |
| 3628 | |
| 3629 /* Even though the pattern is a signed `char *', we need to fetch | |
| 3630 with unsigned char *'s; if the high bit of the pattern character | |
| 3631 is set, the range endpoints will be negative if we fetch using a | |
| 3632 signed char *. | |
| 3633 | |
| 3634 We also want to fetch the endpoints without translating them; the | |
| 3635 appropriate translation is done in the bit-setting loop below. */ | |
| 442 | 3636 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */ |
| 3637 range_start = ((const unsigned char *) p)[-2]; | |
| 3638 range_end = ((const unsigned char *) p)[0]; | |
| 428 | 3639 |
| 3640 /* Have to increment the pointer into the pattern string, so the | |
| 3641 caller isn't still at the ending character. */ | |
| 3642 (*p_ptr)++; | |
| 3643 | |
| 3644 /* If the start is after the end, the range is empty. */ | |
| 3645 if (range_start > range_end) | |
| 3646 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
| 3647 | |
| 3648 /* Here we see why `this_char' has to be larger than an `unsigned | |
| 3649 char' -- the range is inclusive, so if `range_end' == 0xff | |
| 3650 (assuming 8-bit characters), we would otherwise go into an infinite | |
| 3651 loop, since all characters <= 0xff. */ | |
| 3652 for (this_char = range_start; this_char <= range_end; this_char++) | |
| 3653 { | |
| 826 | 3654 SET_LIST_BIT (RE_TRANSLATE (this_char)); |
| 428 | 3655 } |
| 3656 | |
| 3657 return REG_NOERROR; | |
| 3658 } | |
| 3659 | |
| 3660 #ifdef MULE | |
| 3661 | |
| 3662 static reg_errcode_t | |
| 446 | 3663 compile_extended_range (re_char **p_ptr, re_char *pend, |
| 3664 RE_TRANSLATE_TYPE translate, | |
| 428 | 3665 reg_syntax_t syntax, Lisp_Object rtab) |
| 3666 { | |
| 867 | 3667 Ichar this_char, range_start, range_end; |
| 3668 const Ibyte *p; | |
| 428 | 3669 |
| 3670 if (*p_ptr == pend) | |
| 3671 return REG_ERANGE; | |
| 3672 | |
| 867 | 3673 p = (const Ibyte *) *p_ptr; |
| 3674 range_end = itext_ichar (p); | |
| 428 | 3675 p--; /* back to '-' */ |
| 867 | 3676 DEC_IBYTEPTR (p); /* back to start of range */ |
| 428 | 3677 /* We also want to fetch the endpoints without translating them; the |
| 3678 appropriate translation is done in the bit-setting loop below. */ | |
| 867 | 3679 range_start = itext_ichar (p); |
| 3680 INC_IBYTEPTR (*p_ptr); | |
| 428 | 3681 |
| 3682 /* If the start is after the end, the range is empty. */ | |
| 3683 if (range_start > range_end) | |
| 3684 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; | |
| 3685 | |
| 3686 /* Can't have ranges spanning different charsets, except maybe for | |
| 3687 ranges entirely within the first 256 chars. */ | |
| 3688 | |
| 3689 if ((range_start >= 0x100 || range_end >= 0x100) | |
| 867 | 3690 && ichar_leading_byte (range_start) != |
| 3691 ichar_leading_byte (range_end)) | |
| 428 | 3692 return REG_ERANGESPAN; |
| 3693 | |
| 826 | 3694 /* #### This might be way inefficient if the range encompasses 10,000 |
| 3695 chars or something. To be efficient, you'd have to do something like | |
| 3696 this: | |
| 428 | 3697 |
| 3698 range_table a; | |
| 3699 range_table b; | |
| 3700 map over translation table in [range_start, range_end] of | |
| 3701 (put the mapped range in a; | |
| 3702 put the translation in b) | |
| 3703 invert the range in a and truncate to [range_start, range_end] | |
| 3704 compute the union of a, b | |
| 3705 union the result into rtab | |
| 3706 */ | |
| 826 | 3707 for (this_char = range_start; this_char <= range_end; this_char++) |
| 428 | 3708 { |
| 826 | 3709 SET_RANGETAB_BIT (RE_TRANSLATE (this_char)); |
| 428 | 3710 } |
| 3711 | |
| 3712 if (this_char <= range_end) | |
| 3713 put_range_table (rtab, this_char, range_end, Qt); | |
| 3714 | |
| 3715 return REG_NOERROR; | |
| 3716 } | |
| 3717 | |
| 3718 #endif /* MULE */ | |
| 3719 | |
| 3720 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in | |
| 3721 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible | |
| 3722 characters can start a string that matches the pattern. This fastmap | |
| 3723 is used by re_search to skip quickly over impossible starting points. | |
| 3724 | |
| 3725 The caller must supply the address of a (1 << BYTEWIDTH)-byte data | |
| 3726 area as BUFP->fastmap. | |
| 3727 | |
| 3728 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in | |
| 3729 the pattern buffer. | |
| 3730 | |
| 3731 Returns 0 if we succeed, -2 if an internal error. */ | |
| 3732 | |
| 3733 int | |
| 826 | 3734 re_compile_fastmap (struct re_pattern_buffer *bufp |
| 3735 RE_LISP_SHORT_CONTEXT_ARGS_DECL) | |
| 428 | 3736 { |
| 3737 int j, k; | |
| 3738 #ifdef MATCH_MAY_ALLOCATE | |
| 3739 fail_stack_type fail_stack; | |
| 3740 #endif | |
| 456 | 3741 DECLARE_DESTINATION; |
| 428 | 3742 /* We don't push any register information onto the failure stack. */ |
| 3743 | |
| 826 | 3744 /* &&#### this should be changed for 8-bit-fixed, for efficiency. see |
| 3745 comment marked with &&#### in re_search_2. */ | |
| 3746 | |
| 428 | 3747 REGISTER char *fastmap = bufp->fastmap; |
| 3748 unsigned char *pattern = bufp->buffer; | |
| 647 | 3749 long size = bufp->used; |
| 428 | 3750 unsigned char *p = pattern; |
| 3751 REGISTER unsigned char *pend = pattern + size; | |
| 3752 | |
| 771 | 3753 #ifdef REGEX_REL_ALLOC |
| 428 | 3754 /* This holds the pointer to the failure stack, when |
| 3755 it is allocated relocatably. */ | |
| 3756 fail_stack_elt_t *failure_stack_ptr; | |
| 3757 #endif | |
| 3758 | |
| 3759 /* Assume that each path through the pattern can be null until | |
| 3760 proven otherwise. We set this false at the bottom of switch | |
| 3761 statement, to which we get only if a particular path doesn't | |
| 3762 match the empty string. */ | |
| 460 | 3763 re_bool path_can_be_null = true; |
| 428 | 3764 |
| 3765 /* We aren't doing a `succeed_n' to begin with. */ | |
| 460 | 3766 re_bool succeed_n_p = false; |
| 428 | 3767 |
| 1333 | 3768 #ifdef ERROR_CHECK_MALLOC |
| 3769 /* The pattern comes from string data, not buffer data. We don't access | |
| 3770 any buffer data, so we don't have to worry about malloc() (but the | |
| 3771 disallowed flag may have been set by a caller). */ | |
| 3772 int depth = bind_regex_malloc_disallowed (0); | |
| 3773 #endif | |
| 3774 | |
| 428 | 3775 assert (fastmap != NULL && p != NULL); |
| 3776 | |
| 3777 INIT_FAIL_STACK (); | |
| 3778 memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */ | |
| 3779 bufp->fastmap_accurate = 1; /* It will be when we're done. */ | |
| 3780 bufp->can_be_null = 0; | |
| 3781 | |
| 3782 while (1) | |
| 3783 { | |
| 3784 if (p == pend || *p == succeed) | |
| 3785 { | |
| 3786 /* We have reached the (effective) end of pattern. */ | |
| 3787 if (!FAIL_STACK_EMPTY ()) | |
| 3788 { | |
| 3789 bufp->can_be_null |= path_can_be_null; | |
| 3790 | |
| 3791 /* Reset for next path. */ | |
| 3792 path_can_be_null = true; | |
| 3793 | |
| 446 | 3794 p = (unsigned char *) fail_stack.stack[--fail_stack.avail].pointer; |
| 428 | 3795 |
| 3796 continue; | |
| 3797 } | |
| 3798 else | |
| 3799 break; | |
| 3800 } | |
| 3801 | |
| 3802 /* We should never be about to go beyond the end of the pattern. */ | |
| 3803 assert (p < pend); | |
| 3804 | |
|
4759
aa5ed11f473b
Remove support for obsolete systems. See xemacs-patches message with ID
Jerry James <james@xemacs.org>
parents:
4750
diff
changeset
|
3805 switch ((re_opcode_t) *p++) |
| 428 | 3806 { |
| 3807 | |
| 3808 /* I guess the idea here is to simply not bother with a fastmap | |
| 3809 if a backreference is used, since it's too hard to figure out | |
| 3810 the fastmap for the corresponding group. Setting | |
| 3811 `can_be_null' stops `re_search_2' from using the fastmap, so | |
| 3812 that is all we do. */ | |
| 3813 case duplicate: | |
| 3814 bufp->can_be_null = 1; | |
| 3815 goto done; | |
| 3816 | |
| 3817 | |
| 3818 /* Following are the cases which match a character. These end | |
| 3819 with `break'. */ | |
| 3820 | |
| 3821 case exactn: | |
| 3822 fastmap[p[1]] = 1; | |
| 3823 break; | |
| 3824 | |
| 3825 | |
| 3826 case charset: | |
| 3827 /* XEmacs: Under Mule, these bit vectors will | |
| 3828 only contain values for characters below 0x80. */ | |
| 3829 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
| 3830 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) | |
| 3831 fastmap[j] = 1; | |
| 3832 break; | |
| 3833 | |
| 3834 | |
| 3835 case charset_not: | |
| 3836 /* Chars beyond end of map must be allowed. */ | |
| 3837 #ifdef MULE | |
| 3838 for (j = *p * BYTEWIDTH; j < 0x80; j++) | |
| 3839 fastmap[j] = 1; | |
| 3840 /* And all extended characters must be allowed, too. */ | |
| 3841 for (j = 0x80; j < 0xA0; j++) | |
| 3842 fastmap[j] = 1; | |
| 446 | 3843 #else /* not MULE */ |
| 428 | 3844 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) |
| 3845 fastmap[j] = 1; | |
| 446 | 3846 #endif /* MULE */ |
| 428 | 3847 |
| 3848 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) | |
| 3849 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) | |
| 3850 fastmap[j] = 1; | |
| 3851 break; | |
| 3852 | |
| 3853 #ifdef MULE | |
| 3854 case charset_mule: | |
| 3855 { | |
| 3856 int nentries; | |
| 3857 int i; | |
| 3858 | |
| 3859 nentries = unified_range_table_nentries (p); | |
| 3860 for (i = 0; i < nentries; i++) | |
| 3861 { | |
| 3862 EMACS_INT first, last; | |
| 3863 Lisp_Object dummy_val; | |
| 3864 int jj; | |
| 867 | 3865 Ibyte strr[MAX_ICHAR_LEN]; |
| 428 | 3866 |
| 3867 unified_range_table_get_range (p, i, &first, &last, | |
| 3868 &dummy_val); | |
| 3869 for (jj = first; jj <= last && jj < 0x80; jj++) | |
| 3870 fastmap[jj] = 1; | |
| 3871 /* Ranges below 0x100 can span charsets, but there | |
| 3872 are only two (Control-1 and Latin-1), and | |
| 3873 either first or last has to be in them. */ | |
| 867 | 3874 set_itext_ichar (strr, first); |
| 428 | 3875 fastmap[*strr] = 1; |
| 3876 if (last < 0x100) | |
| 3877 { | |
| 867 | 3878 set_itext_ichar (strr, last); |
| 428 | 3879 fastmap[*strr] = 1; |
| 3880 } | |
| 3881 } | |
| 3882 } | |
| 3883 break; | |
| 3884 | |
| 3885 case charset_mule_not: | |
| 3886 { | |
| 3887 int nentries; | |
| 3888 int i; | |
|
4832
07fa38c30fdf
fix messed-up fastmap calculation in charset_mule_not
Ben Wing <ben@xemacs.org>
parents:
4759
diff
changeset
|
3889 int smallest_prev = 0; |
| 428 | 3890 |
| 3891 nentries = unified_range_table_nentries (p); | |
| 3892 for (i = 0; i < nentries; i++) | |
| 3893 { | |
| 3894 EMACS_INT first, last; | |
| 3895 Lisp_Object dummy_val; | |
| 3896 int jj; | |
| 3897 | |
| 3898 unified_range_table_get_range (p, i, &first, &last, | |
| 3899 &dummy_val); | |
| 3900 for (jj = smallest_prev; jj < first && jj < 0x80; jj++) | |
| 3901 fastmap[jj] = 1; | |
| 3902 smallest_prev = last + 1; | |
| 3903 if (smallest_prev >= 0x80) | |
| 3904 break; | |
| 3905 } | |
|
4832
07fa38c30fdf
fix messed-up fastmap calculation in charset_mule_not
Ben Wing <ben@xemacs.org>
parents:
4759
diff
changeset
|
3906 |
|
07fa38c30fdf
fix messed-up fastmap calculation in charset_mule_not
Ben Wing <ben@xemacs.org>
parents:
4759
diff
changeset
|
3907 /* Also set lead bytes after the end */ |
|
07fa38c30fdf
fix messed-up fastmap calculation in charset_mule_not
Ben Wing <ben@xemacs.org>
parents:
4759
diff
changeset
|
3908 for (i = smallest_prev; i < 0x80; i++) |
|
07fa38c30fdf
fix messed-up fastmap calculation in charset_mule_not
Ben Wing <ben@xemacs.org>
parents:
4759
diff
changeset
|
3909 fastmap[i] = 1; |
|
07fa38c30fdf
fix messed-up fastmap calculation in charset_mule_not
Ben Wing <ben@xemacs.org>
parents:
4759
diff
changeset
|
3910 |
| 428 | 3911 /* Calculating which leading bytes are actually allowed |
| 3912 here is rather difficult, so we just punt and allow | |
| 3913 all of them. */ | |
| 3914 for (i = 0x80; i < 0xA0; i++) | |
| 3915 fastmap[i] = 1; | |
| 3916 } | |
| 3917 break; | |
| 3918 #endif /* MULE */ | |
| 3919 | |
| 3920 | |
| 3921 case anychar: | |
| 3922 { | |
| 3923 int fastmap_newline = fastmap['\n']; | |
| 3924 | |
| 3925 /* `.' matches anything ... */ | |
| 3926 #ifdef MULE | |
| 3927 /* "anything" only includes bytes that can be the | |
| 3928 first byte of a character. */ | |
| 3929 for (j = 0; j < 0xA0; j++) | |
| 3930 fastmap[j] = 1; | |
| 3931 #else | |
| 3932 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
| 3933 fastmap[j] = 1; | |
| 3934 #endif | |
| 3935 | |
| 3936 /* ... except perhaps newline. */ | |
| 3937 if (!(bufp->syntax & RE_DOT_NEWLINE)) | |
| 3938 fastmap['\n'] = fastmap_newline; | |
| 3939 | |
| 3940 /* Return if we have already set `can_be_null'; if we have, | |
| 3941 then the fastmap is irrelevant. Something's wrong here. */ | |
| 3942 else if (bufp->can_be_null) | |
| 3943 goto done; | |
| 3944 | |
| 3945 /* Otherwise, have to check alternative paths. */ | |
| 3946 break; | |
| 3947 } | |
| 3948 | |
| 826 | 3949 #ifndef emacs |
| 3950 case wordchar: | |
| 3951 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
| 3952 if (SYNTAX (ignored, j) == Sword) | |
| 3953 fastmap[j] = 1; | |
| 3954 break; | |
| 3955 | |
| 3956 case notwordchar: | |
| 3957 for (j = 0; j < (1 << BYTEWIDTH); j++) | |
| 3958 if (SYNTAX (ignored, j) != Sword) | |
| 3959 fastmap[j] = 1; | |
| 3960 break; | |
| 3961 #else /* emacs */ | |
| 3962 case wordchar: | |
| 3963 case notwordchar: | |
| 460 | 3964 case wordbound: |
| 3965 case notwordbound: | |
| 3966 case wordbeg: | |
| 3967 case wordend: | |
| 3968 case notsyntaxspec: | |
| 3969 case syntaxspec: | |
| 3970 /* This match depends on text properties. These end with | |
| 3971 aborting optimizations. */ | |
| 3972 bufp->can_be_null = 1; | |
| 3973 goto done; | |
| 826 | 3974 #if 0 /* all of the following code is unused now that the `syntax-table' |
| 3975 property exists -- it's trickier to do this than just look in | |
| 3976 the buffer. &&#### but we could just use the syntax-cache stuff | |
| 3977 instead; why don't we? --ben */ | |
| 3978 case wordchar: | |
| 3979 k = (int) Sword; | |
| 3980 goto matchsyntax; | |
| 3981 | |
| 3982 case notwordchar: | |
| 3983 k = (int) Sword; | |
| 3984 goto matchnotsyntax; | |
| 3985 | |
| 428 | 3986 case syntaxspec: |
| 3987 k = *p++; | |
| 826 | 3988 matchsyntax: |
| 428 | 3989 #ifdef MULE |
| 3990 for (j = 0; j < 0x80; j++) | |
| 826 | 3991 if (SYNTAX |
| 3992 (XCHAR_TABLE (BUFFER_MIRROR_SYNTAX_TABLE (lispbuf)), j) == | |
| 428 | 3993 (enum syntaxcode) k) |
| 3994 fastmap[j] = 1; | |
| 3995 for (j = 0x80; j < 0xA0; j++) | |
| 3996 { | |
| 826 | 3997 if (leading_byte_prefix_p ((unsigned char) j)) |
| 428 | 3998 /* too complicated to calculate this right */ |
| 3999 fastmap[j] = 1; | |
| 4000 else | |
| 4001 { | |
| 4002 int multi_p; | |
| 4003 Lisp_Object cset; | |
| 4004 | |
| 826 | 4005 cset = charset_by_leading_byte (j); |
| 428 | 4006 if (CHARSETP (cset)) |
| 4007 { | |
| 826 | 4008 if (charset_syntax (lispbuf, cset, &multi_p) |
| 428 | 4009 == Sword || multi_p) |
| 4010 fastmap[j] = 1; | |
| 4011 } | |
| 4012 } | |
| 4013 } | |
| 446 | 4014 #else /* not MULE */ |
| 428 | 4015 for (j = 0; j < (1 << BYTEWIDTH); j++) |
| 826 | 4016 if (SYNTAX |
| 4017 (XCHAR_TABLE (BUFFER_MIRROR_SYNTAX_TABLE (lispbuf)), j) == | |
| 428 | 4018 (enum syntaxcode) k) |
| 4019 fastmap[j] = 1; | |
| 446 | 4020 #endif /* MULE */ |
| 428 | 4021 break; |
| 4022 | |
| 4023 | |
| 4024 case notsyntaxspec: | |
| 4025 k = *p++; | |
| 826 | 4026 matchnotsyntax: |
| 428 | 4027 #ifdef MULE |
| 4028 for (j = 0; j < 0x80; j++) | |
| 826 | 4029 if (SYNTAX |
| 428 | 4030 (XCHAR_TABLE |
| 826 | 4031 (BUFFER_MIRROR_SYNTAX_TABLE (lispbuf)), j) != |
| 428 | 4032 (enum syntaxcode) k) |
| 4033 fastmap[j] = 1; | |
| 4034 for (j = 0x80; j < 0xA0; j++) | |
| 4035 { | |
| 826 | 4036 if (leading_byte_prefix_p ((unsigned char) j)) |
| 428 | 4037 /* too complicated to calculate this right */ |
| 4038 fastmap[j] = 1; | |
| 4039 else | |
| 4040 { | |
| 4041 int multi_p; | |
| 4042 Lisp_Object cset; | |
| 4043 | |
| 826 | 4044 cset = charset_by_leading_byte (j); |
| 428 | 4045 if (CHARSETP (cset)) |
| 4046 { | |
| 826 | 4047 if (charset_syntax (lispbuf, cset, &multi_p) |
| 428 | 4048 != Sword || multi_p) |
| 4049 fastmap[j] = 1; | |
| 4050 } | |
| 4051 } | |
| 4052 } | |
| 446 | 4053 #else /* not MULE */ |
| 428 | 4054 for (j = 0; j < (1 << BYTEWIDTH); j++) |
| 826 | 4055 if (SYNTAX |
| 428 | 4056 (XCHAR_TABLE |
| 826 | 4057 (BUFFER_MIRROR_SYNTAX_TABLE (lispbuf)), j) != |
| 428 | 4058 (enum syntaxcode) k) |
| 4059 fastmap[j] = 1; | |
| 446 | 4060 #endif /* MULE */ |
| 428 | 4061 break; |
| 826 | 4062 #endif /* 0 */ |
| 428 | 4063 |
| 4064 #ifdef MULE | |
| 4065 /* 97/2/17 jhod category patch */ | |
| 4066 case categoryspec: | |
| 4067 case notcategoryspec: | |
| 4068 bufp->can_be_null = 1; | |
| 1333 | 4069 UNBIND_REGEX_MALLOC_CHECK (); |
| 428 | 4070 return 0; |
| 4071 /* end if category patch */ | |
| 4072 #endif /* MULE */ | |
| 4073 | |
| 4074 /* All cases after this match the empty string. These end with | |
| 4075 `continue'. */ | |
| 4076 case before_dot: | |
| 4077 case at_dot: | |
| 4078 case after_dot: | |
| 4079 continue; | |
| 826 | 4080 #endif /* emacs */ |
| 428 | 4081 |
| 4082 | |
| 4083 case no_op: | |
| 4084 case begline: | |
| 4085 case endline: | |
| 4086 case begbuf: | |
| 4087 case endbuf: | |
| 460 | 4088 #ifndef emacs |
| 428 | 4089 case wordbound: |
| 4090 case notwordbound: | |
| 4091 case wordbeg: | |
| 4092 case wordend: | |
| 460 | 4093 #endif |
| 428 | 4094 case push_dummy_failure: |
| 4095 continue; | |
| 4096 | |
| 4097 | |
| 4098 case jump_n: | |
| 4099 case pop_failure_jump: | |
| 4100 case maybe_pop_jump: | |
| 4101 case jump: | |
| 4102 case jump_past_alt: | |
| 4103 case dummy_failure_jump: | |
| 4104 EXTRACT_NUMBER_AND_INCR (j, p); | |
| 4105 p += j; | |
| 4106 if (j > 0) | |
| 4107 continue; | |
| 4108 | |
| 4109 /* Jump backward implies we just went through the body of a | |
| 4110 loop and matched nothing. Opcode jumped to should be | |
| 4111 `on_failure_jump' or `succeed_n'. Just treat it like an | |
| 4112 ordinary jump. For a * loop, it has pushed its failure | |
| 4113 point already; if so, discard that as redundant. */ | |
| 4114 if ((re_opcode_t) *p != on_failure_jump | |
| 4115 && (re_opcode_t) *p != succeed_n) | |
| 4116 continue; | |
| 4117 | |
| 4118 p++; | |
| 4119 EXTRACT_NUMBER_AND_INCR (j, p); | |
| 4120 p += j; | |
| 4121 | |
| 4122 /* If what's on the stack is where we are now, pop it. */ | |
| 4123 if (!FAIL_STACK_EMPTY () | |
| 4124 && fail_stack.stack[fail_stack.avail - 1].pointer == p) | |
| 4125 fail_stack.avail--; | |
| 4126 | |
| 4127 continue; | |
| 4128 | |
| 4129 | |
| 4130 case on_failure_jump: | |
| 4131 case on_failure_keep_string_jump: | |
| 4132 handle_on_failure_jump: | |
| 4133 EXTRACT_NUMBER_AND_INCR (j, p); | |
| 4134 | |
| 4135 /* For some patterns, e.g., `(a?)?', `p+j' here points to the | |
| 4136 end of the pattern. We don't want to push such a point, | |
| 4137 since when we restore it above, entering the switch will | |
| 4138 increment `p' past the end of the pattern. We don't need | |
| 4139 to push such a point since we obviously won't find any more | |
| 4140 fastmap entries beyond `pend'. Such a pattern can match | |
| 4141 the null string, though. */ | |
| 4142 if (p + j < pend) | |
| 4143 { | |
| 4144 if (!PUSH_PATTERN_OP (p + j, fail_stack)) | |
| 4145 { | |
| 4146 RESET_FAIL_STACK (); | |
| 1333 | 4147 UNBIND_REGEX_MALLOC_CHECK (); |
| 428 | 4148 return -2; |
| 4149 } | |
| 4150 } | |
| 4151 else | |
| 4152 bufp->can_be_null = 1; | |
| 4153 | |
| 4154 if (succeed_n_p) | |
| 4155 { | |
| 4156 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ | |
| 4157 succeed_n_p = false; | |
| 4158 } | |
| 4159 | |
| 4160 continue; | |
| 4161 | |
| 4162 | |
| 4163 case succeed_n: | |
| 4164 /* Get to the number of times to succeed. */ | |
| 4165 p += 2; | |
| 4166 | |
| 4167 /* Increment p past the n for when k != 0. */ | |
| 4168 EXTRACT_NUMBER_AND_INCR (k, p); | |
| 4169 if (k == 0) | |
| 4170 { | |
| 4171 p -= 4; | |
| 4172 succeed_n_p = true; /* Spaghetti code alert. */ | |
| 4173 goto handle_on_failure_jump; | |
| 4174 } | |
| 4175 continue; | |
| 4176 | |
| 4177 | |
| 4178 case set_number_at: | |
| 4179 p += 4; | |
| 4180 continue; | |
| 4181 | |
| 4182 | |
| 4183 case start_memory: | |
| 4184 case stop_memory: | |
| 4185 p += 2; | |
| 4186 continue; | |
| 4187 | |
| 4188 | |
| 4189 default: | |
| 2500 | 4190 ABORT (); /* We have listed all the cases. */ |
| 428 | 4191 } /* switch *p++ */ |
| 4192 | |
| 4193 /* Getting here means we have found the possible starting | |
| 4194 characters for one path of the pattern -- and that the empty | |
| 4195 string does not match. We need not follow this path further. | |
| 4196 Instead, look at the next alternative (remembered on the | |
| 4197 stack), or quit if no more. The test at the top of the loop | |
| 4198 does these things. */ | |
| 4199 path_can_be_null = false; | |
| 4200 p = pend; | |
| 4201 } /* while p */ | |
| 4202 | |
| 4203 /* Set `can_be_null' for the last path (also the first path, if the | |
| 4204 pattern is empty). */ | |
| 4205 bufp->can_be_null |= path_can_be_null; | |
| 4206 | |
| 4207 done: | |
| 4208 RESET_FAIL_STACK (); | |
| 1333 | 4209 UNBIND_REGEX_MALLOC_CHECK (); |
| 428 | 4210 return 0; |
| 4211 } /* re_compile_fastmap */ | |
| 4212 | |
| 4213 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and | |
| 4214 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use | |
| 4215 this memory for recording register information. STARTS and ENDS | |
| 4216 must be allocated using the malloc library routine, and must each | |
| 4217 be at least NUM_REGS * sizeof (regoff_t) bytes long. | |
| 4218 | |
| 4219 If NUM_REGS == 0, then subsequent matches should allocate their own | |
| 4220 register data. | |
| 4221 | |
| 4222 Unless this function is called, the first search or match using | |
| 4223 PATTERN_BUFFER will allocate its own register data, without | |
| 4224 freeing the old data. */ | |
| 4225 | |
| 4226 void | |
| 4227 re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, | |
| 647 | 4228 int num_regs, regoff_t *starts, regoff_t *ends) |
| 428 | 4229 { |
| 4230 if (num_regs) | |
| 4231 { | |
| 4232 bufp->regs_allocated = REGS_REALLOCATE; | |
| 4233 regs->num_regs = num_regs; | |
| 4234 regs->start = starts; | |
| 4235 regs->end = ends; | |
| 4236 } | |
| 4237 else | |
| 4238 { | |
| 4239 bufp->regs_allocated = REGS_UNALLOCATED; | |
| 4240 regs->num_regs = 0; | |
| 4241 regs->start = regs->end = (regoff_t *) 0; | |
| 4242 } | |
| 4243 } | |
| 4244 | |
| 4245 /* Searching routines. */ | |
| 4246 | |
| 4247 /* Like re_search_2, below, but only one string is specified, and | |
| 4248 doesn't let you say where to stop matching. */ | |
| 4249 | |
| 4250 int | |
| 442 | 4251 re_search (struct re_pattern_buffer *bufp, const char *string, int size, |
| 826 | 4252 int startpos, int range, struct re_registers *regs |
| 4253 RE_LISP_CONTEXT_ARGS_DECL) | |
| 428 | 4254 { |
| 4255 return re_search_2 (bufp, NULL, 0, string, size, startpos, range, | |
| 826 | 4256 regs, size RE_LISP_CONTEXT_ARGS); |
| 428 | 4257 } |
| 4258 | |
| 4259 /* Using the compiled pattern in BUFP->buffer, first tries to match the | |
| 4260 virtual concatenation of STRING1 and STRING2, starting first at index | |
| 4261 STARTPOS, then at STARTPOS + 1, and so on. | |
| 4262 | |
| 4263 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. | |
| 4264 | |
| 4265 RANGE is how far to scan while trying to match. RANGE = 0 means try | |
| 4266 only at STARTPOS; in general, the last start tried is STARTPOS + | |
| 4267 RANGE. | |
| 4268 | |
| 826 | 4269 All sizes and positions refer to bytes (not chars); under Mule, the code |
| 4270 knows about the format of the text and will only check at positions | |
| 4271 where a character starts. | |
| 4272 | |
| 428 | 4273 With MULE, RANGE is a byte position, not a char position. The last |
| 4274 start tried is the character starting <= STARTPOS + RANGE. | |
| 4275 | |
| 4276 In REGS, return the indices of the virtual concatenation of STRING1 | |
| 4277 and STRING2 that matched the entire BUFP->buffer and its contained | |
| 4278 subexpressions. | |
| 4279 | |
| 4280 Do not consider matching one past the index STOP in the virtual | |
| 4281 concatenation of STRING1 and STRING2. | |
| 4282 | |
| 4283 We return either the position in the strings at which the match was | |
| 4284 found, -1 if no match, or -2 if error (such as failure | |
| 4285 stack overflow). */ | |
| 4286 | |
| 4287 int | |
| 446 | 4288 re_search_2 (struct re_pattern_buffer *bufp, const char *str1, |
| 4289 int size1, const char *str2, int size2, int startpos, | |
| 826 | 4290 int range, struct re_registers *regs, int stop |
| 4291 RE_LISP_CONTEXT_ARGS_DECL) | |
| 428 | 4292 { |
| 4293 int val; | |
| 446 | 4294 re_char *string1 = (re_char *) str1; |
| 4295 re_char *string2 = (re_char *) str2; | |
| 428 | 4296 REGISTER char *fastmap = bufp->fastmap; |
| 446 | 4297 REGISTER RE_TRANSLATE_TYPE translate = bufp->translate; |
| 428 | 4298 int total_size = size1 + size2; |
| 4299 int endpos = startpos + range; | |
| 4300 #ifdef REGEX_BEGLINE_CHECK | |
| 4301 int anchored_at_begline = 0; | |
| 4302 #endif | |
| 446 | 4303 re_char *d; |
| 826 | 4304 #ifdef emacs |
| 4305 Internal_Format fmt = buffer_or_other_internal_format (lispobj); | |
| 1346 | 4306 #ifdef REL_ALLOC |
| 4307 Ibyte *orig_buftext = | |
| 4308 BUFFERP (lispobj) ? | |
| 4309 BYTE_BUF_BYTE_ADDRESS (XBUFFER (lispobj), | |
| 4310 BYTE_BUF_BEGV (XBUFFER (lispobj))) : | |
| 4311 0; | |
| 4312 #endif | |
| 1333 | 4313 #ifdef ERROR_CHECK_MALLOC |
| 4314 int depth; | |
| 4315 #endif | |
| 826 | 4316 #endif /* emacs */ |
| 4317 #if 1 | |
| 4318 int forward_search_p; | |
| 4319 #endif | |
| 428 | 4320 |
| 4321 /* Check for out-of-range STARTPOS. */ | |
| 4322 if (startpos < 0 || startpos > total_size) | |
| 4323 return -1; | |
| 4324 | |
| 4325 /* Fix up RANGE if it might eventually take us outside | |
| 4326 the virtual concatenation of STRING1 and STRING2. */ | |
| 4327 if (endpos < 0) | |
| 4328 range = 0 - startpos; | |
| 4329 else if (endpos > total_size) | |
| 4330 range = total_size - startpos; | |
| 4331 | |
| 826 | 4332 #if 1 |
| 4333 forward_search_p = range > 0; | |
| 4334 #endif | |
| 4335 | |
| 428 | 4336 /* If the search isn't to be a backwards one, don't waste time in a |
| 4337 search for a pattern that must be anchored. */ | |
| 4338 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) | |
| 4339 { | |
| 4340 if (startpos > 0) | |
| 4341 return -1; | |
| 4342 else | |
| 4343 { | |
| 442 | 4344 d = ((const unsigned char *) |
| 428 | 4345 (startpos >= size1 ? string2 - size1 : string1) + startpos); |
| 867 | 4346 range = itext_ichar_len_fmt (d, fmt); |
| 428 | 4347 } |
| 4348 } | |
| 4349 | |
| 460 | 4350 #ifdef emacs |
| 4351 /* In a forward search for something that starts with \=. | |
| 4352 don't keep searching past point. */ | |
| 4353 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) | |
| 4354 { | |
| 826 | 4355 if (!BUFFERP (lispobj)) |
| 4356 return -1; | |
|
4527
8418d1ad4944
Fix at_dot regex under Mule. <87hc6rv53v.fsf@uwakimon.sk.tsukuba.ac.jp>
Stephen J. Turnbull <stephen@xemacs.org>
parents:
3300
diff
changeset
|
4357 range = (BYTE_BUF_PT (XBUFFER (lispobj)) |
|
8418d1ad4944
Fix at_dot regex under Mule. <87hc6rv53v.fsf@uwakimon.sk.tsukuba.ac.jp>
Stephen J. Turnbull <stephen@xemacs.org>
parents:
3300
diff
changeset
|
4358 - BYTE_BUF_BEGV (XBUFFER (lispobj)) - startpos); |
| 460 | 4359 if (range < 0) |
| 4360 return -1; | |
| 4361 } | |
| 4362 #endif /* emacs */ | |
| 4363 | |
| 1333 | 4364 #ifdef ERROR_CHECK_MALLOC |
| 4365 /* Do this after the above return()s. */ | |
| 4366 depth = bind_regex_malloc_disallowed (1); | |
| 4367 #endif | |
| 4368 | |
| 428 | 4369 /* Update the fastmap now if not correct already. */ |
| 1333 | 4370 BEGIN_REGEX_MALLOC_OK (); |
| 428 | 4371 if (fastmap && !bufp->fastmap_accurate) |
| 826 | 4372 if (re_compile_fastmap (bufp RE_LISP_SHORT_CONTEXT_ARGS) == -2) |
| 1333 | 4373 { |
| 4374 END_REGEX_MALLOC_OK (); | |
| 4375 UNBIND_REGEX_MALLOC_CHECK (); | |
| 4376 return -2; | |
| 4377 } | |
| 4378 | |
| 4379 END_REGEX_MALLOC_OK (); | |
| 4380 RE_SEARCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 428 | 4381 |
| 4382 #ifdef REGEX_BEGLINE_CHECK | |
| 4383 { | |
| 647 | 4384 long i = 0; |
| 428 | 4385 |
| 4386 while (i < bufp->used) | |
| 4387 { | |
| 4388 if (bufp->buffer[i] == start_memory || | |
| 4389 bufp->buffer[i] == stop_memory) | |
| 4390 i += 2; | |
| 4391 else | |
| 4392 break; | |
| 4393 } | |
| 4394 anchored_at_begline = i < bufp->used && bufp->buffer[i] == begline; | |
| 4395 } | |
| 4396 #endif | |
| 4397 | |
| 460 | 4398 #ifdef emacs |
| 1333 | 4399 BEGIN_REGEX_MALLOC_OK (); |
| 826 | 4400 scache = setup_syntax_cache (scache, lispobj, lispbuf, |
| 4401 offset_to_charxpos (lispobj, startpos), | |
| 4402 1); | |
| 1333 | 4403 END_REGEX_MALLOC_OK (); |
| 4404 RE_SEARCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 460 | 4405 #endif |
| 4406 | |
| 428 | 4407 /* Loop through the string, looking for a place to start matching. */ |
| 4408 for (;;) | |
| 4409 { | |
| 4410 #ifdef REGEX_BEGLINE_CHECK | |
| 826 | 4411 /* If the regex is anchored at the beginning of a line (i.e. with a |
| 4412 ^), then we can speed things up by skipping to the next | |
| 4413 beginning-of-line. However, to determine "beginning of line" we | |
| 4414 need to look at the previous char, so can't do this check if at | |
| 4415 beginning of either string. (Well, we could if at the beginning of | |
| 4416 the second string, but it would require additional code, and this | |
| 4417 is just an optimization.) */ | |
| 4418 if (anchored_at_begline && startpos > 0 && startpos != size1) | |
| 428 | 4419 { |
| 826 | 4420 if (range > 0) |
| 4421 { | |
| 4422 /* whose stupid idea was it anyway to make this | |
| 4423 function take two strings to match?? */ | |
| 4424 int lim = 0; | |
| 4425 re_char *orig_d; | |
| 4426 re_char *stop_d; | |
| 4427 | |
| 4428 /* Compute limit as below in fastmap code, so we are guaranteed | |
| 4429 to remain within a single string. */ | |
| 4430 if (startpos < size1 && startpos + range >= size1) | |
| 4431 lim = range - (size1 - startpos); | |
| 4432 | |
| 4433 d = ((const unsigned char *) | |
| 4434 (startpos >= size1 ? string2 - size1 : string1) + startpos); | |
| 4435 orig_d = d; | |
| 4436 stop_d = d + range - lim; | |
| 4437 | |
| 4438 /* We want to find the next location (including the current | |
| 4439 one) where the previous char is a newline, so back up one | |
| 4440 and search forward for a newline. */ | |
| 867 | 4441 DEC_IBYTEPTR_FMT (d, fmt); /* Ok, since startpos != size1. */ |
| 826 | 4442 |
| 4443 /* Written out as an if-else to avoid testing `translate' | |
| 4444 inside the loop. */ | |
| 4445 if (TRANSLATE_P (translate)) | |
| 4446 while (d < stop_d && | |
| 867 | 4447 RE_TRANSLATE_1 (itext_ichar_fmt (d, fmt, lispobj)) |
| 826 | 4448 != '\n') |
| 867 | 4449 INC_IBYTEPTR_FMT (d, fmt); |
| 826 | 4450 else |
| 4451 while (d < stop_d && | |
| 867 | 4452 itext_ichar_ascii_fmt (d, fmt, lispobj) != '\n') |
| 4453 INC_IBYTEPTR_FMT (d, fmt); | |
| 826 | 4454 |
| 4455 /* If we were stopped by a newline, skip forward over it. | |
| 4456 Otherwise we will get in an infloop when our start position | |
| 4457 was at begline. */ | |
| 4458 if (d < stop_d) | |
| 867 | 4459 INC_IBYTEPTR_FMT (d, fmt); |
| 826 | 4460 range -= d - orig_d; |
| 4461 startpos += d - orig_d; | |
| 4462 #if 1 | |
| 4463 assert (!forward_search_p || range >= 0); | |
| 4464 #endif | |
| 4465 } | |
| 4466 else if (range < 0) | |
| 4467 { | |
| 4468 /* We're lazy, like in the fastmap code below */ | |
| 867 | 4469 Ichar c; |
| 826 | 4470 |
| 4471 d = ((const unsigned char *) | |
| 4472 (startpos >= size1 ? string2 - size1 : string1) + startpos); | |
| 867 | 4473 DEC_IBYTEPTR_FMT (d, fmt); |
| 4474 c = itext_ichar_fmt (d, fmt, lispobj); | |
| 826 | 4475 c = RE_TRANSLATE (c); |
| 4476 if (c != '\n') | |
| 4477 goto advance; | |
| 4478 } | |
| 428 | 4479 } |
| 4480 #endif /* REGEX_BEGLINE_CHECK */ | |
| 4481 | |
| 4482 /* If a fastmap is supplied, skip quickly over characters that | |
| 4483 cannot be the start of a match. If the pattern can match the | |
| 4484 null string, however, we don't need to skip characters; we want | |
| 4485 the first null string. */ | |
| 4486 if (fastmap && startpos < total_size && !bufp->can_be_null) | |
| 4487 { | |
| 826 | 4488 /* For the moment, fastmap always works as if buffer |
| 4489 is in default format, so convert chars in the search strings | |
| 4490 into default format as we go along, if necessary. | |
| 4491 | |
| 4492 &&#### fastmap needs rethinking for 8-bit-fixed so | |
| 4493 it's faster. We need it to reflect the raw | |
| 4494 8-bit-fixed values. That isn't so hard if we assume | |
| 4495 that the top 96 bytes represent a single 1-byte | |
| 4496 charset. For 16-bit/32-bit stuff it's probably not | |
| 4497 worth it to make the fastmap represent the raw, due to | |
| 4498 its nature -- we'd have to use the LSB for the | |
| 4499 fastmap, and that causes lots of problems with Mule | |
| 4500 chars, where it essentially wipes out the usefulness | |
| 4501 of the fastmap entirely. */ | |
| 428 | 4502 if (range > 0) /* Searching forwards. */ |
| 4503 { | |
| 4504 int lim = 0; | |
| 4505 int irange = range; | |
| 4506 | |
| 4507 if (startpos < size1 && startpos + range >= size1) | |
| 4508 lim = range - (size1 - startpos); | |
| 4509 | |
| 442 | 4510 d = ((const unsigned char *) |
| 428 | 4511 (startpos >= size1 ? string2 - size1 : string1) + startpos); |
| 4512 | |
| 4513 /* Written out as an if-else to avoid testing `translate' | |
| 4514 inside the loop. */ | |
| 446 | 4515 if (TRANSLATE_P (translate)) |
| 826 | 4516 { |
| 4517 while (range > lim) | |
| 4518 { | |
| 4519 re_char *old_d = d; | |
| 428 | 4520 #ifdef MULE |
| 867 | 4521 Ibyte tempch[MAX_ICHAR_LEN]; |
| 4522 Ichar buf_ch = | |
| 4523 RE_TRANSLATE_1 (itext_ichar_fmt (d, fmt, lispobj)); | |
| 4524 set_itext_ichar (tempch, buf_ch); | |
| 826 | 4525 if (fastmap[*tempch]) |
| 4526 break; | |
| 446 | 4527 #else |
| 826 | 4528 if (fastmap[(unsigned char) RE_TRANSLATE_1 (*d)]) |
| 4529 break; | |
| 446 | 4530 #endif /* MULE */ |
| 867 | 4531 INC_IBYTEPTR_FMT (d, fmt); |
| 826 | 4532 range -= (d - old_d); |
| 4533 #if 1 | |
| 1333 | 4534 assert (!forward_search_p || range >= 0); |
| 826 | 4535 #endif |
| 4536 } | |
| 4537 } | |
| 4538 #ifdef MULE | |
| 4539 else if (fmt != FORMAT_DEFAULT) | |
| 4540 { | |
| 4541 while (range > lim) | |
| 4542 { | |
| 4543 re_char *old_d = d; | |
| 867 | 4544 Ibyte tempch[MAX_ICHAR_LEN]; |
| 4545 Ichar buf_ch = itext_ichar_fmt (d, fmt, lispobj); | |
| 4546 set_itext_ichar (tempch, buf_ch); | |
| 826 | 4547 if (fastmap[*tempch]) |
| 4548 break; | |
| 867 | 4549 INC_IBYTEPTR_FMT (d, fmt); |
| 826 | 4550 range -= (d - old_d); |
| 4551 #if 1 | |
| 1333 | 4552 assert (!forward_search_p || range >= 0); |
| 826 | 4553 #endif |
| 4554 } | |
| 4555 } | |
| 4556 #endif /* MULE */ | |
| 428 | 4557 else |
| 826 | 4558 { |
| 4559 while (range > lim && !fastmap[*d]) | |
| 4560 { | |
| 4561 re_char *old_d = d; | |
| 867 | 4562 INC_IBYTEPTR (d); |
| 826 | 4563 range -= (d - old_d); |
| 4564 #if 1 | |
| 4565 assert (!forward_search_p || range >= 0); | |
| 4566 #endif | |
| 4567 } | |
| 4568 } | |
| 428 | 4569 |
| 4570 startpos += irange - range; | |
| 4571 } | |
| 4572 else /* Searching backwards. */ | |
| 4573 { | |
| 826 | 4574 /* #### It's not clear why we don't just write a loop, like |
| 4575 for the moving-forward case. Perhaps the writer got lazy, | |
| 4576 since backward searches aren't so common. */ | |
| 4577 d = ((const unsigned char *) | |
| 4578 (startpos >= size1 ? string2 - size1 : string1) + startpos); | |
| 428 | 4579 #ifdef MULE |
| 826 | 4580 { |
| 867 | 4581 Ibyte tempch[MAX_ICHAR_LEN]; |
| 4582 Ichar buf_ch = | |
| 4583 RE_TRANSLATE (itext_ichar_fmt (d, fmt, lispobj)); | |
| 4584 set_itext_ichar (tempch, buf_ch); | |
| 826 | 4585 if (!fastmap[*tempch]) |
| 4586 goto advance; | |
| 4587 } | |
| 428 | 4588 #else |
| 826 | 4589 if (!fastmap[(unsigned char) RE_TRANSLATE (*d)]) |
| 446 | 4590 goto advance; |
| 826 | 4591 #endif /* MULE */ |
| 428 | 4592 } |
| 4593 } | |
| 4594 | |
| 4595 /* If can't match the null string, and that's all we have left, fail. */ | |
| 4596 if (range >= 0 && startpos == total_size && fastmap | |
| 4597 && !bufp->can_be_null) | |
| 1333 | 4598 { |
| 4599 UNBIND_REGEX_MALLOC_CHECK (); | |
| 4600 return -1; | |
| 4601 } | |
| 428 | 4602 |
| 4603 #ifdef emacs /* XEmacs added, w/removal of immediate_quit */ | |
| 4604 if (!no_quit_in_re_search) | |
| 1333 | 4605 { |
| 4606 BEGIN_REGEX_MALLOC_OK (); | |
| 4607 QUIT; | |
| 4608 END_REGEX_MALLOC_OK (); | |
| 4609 RE_SEARCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 4610 } | |
| 4611 | |
| 428 | 4612 #endif |
| 1333 | 4613 BEGIN_REGEX_MALLOC_OK (); |
| 428 | 4614 val = re_match_2_internal (bufp, string1, size1, string2, size2, |
| 826 | 4615 startpos, regs, stop |
| 4616 RE_LISP_CONTEXT_ARGS); | |
| 428 | 4617 #ifndef REGEX_MALLOC |
| 1333 | 4618 ALLOCA_GARBAGE_COLLECT (); |
| 428 | 4619 #endif |
| 1333 | 4620 END_REGEX_MALLOC_OK (); |
| 4621 RE_SEARCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 428 | 4622 |
| 4623 if (val >= 0) | |
| 1333 | 4624 { |
| 4625 UNBIND_REGEX_MALLOC_CHECK (); | |
| 4626 return startpos; | |
| 4627 } | |
| 428 | 4628 |
| 4629 if (val == -2) | |
| 1333 | 4630 { |
| 4631 UNBIND_REGEX_MALLOC_CHECK (); | |
| 4632 return -2; | |
| 4633 } | |
| 4634 | |
| 4635 RE_SEARCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 428 | 4636 advance: |
| 4637 if (!range) | |
| 4638 break; | |
| 4639 else if (range > 0) | |
| 4640 { | |
| 826 | 4641 Bytecount d_size; |
| 442 | 4642 d = ((const unsigned char *) |
| 428 | 4643 (startpos >= size1 ? string2 - size1 : string1) + startpos); |
| 867 | 4644 d_size = itext_ichar_len_fmt (d, fmt); |
| 428 | 4645 range -= d_size; |
| 826 | 4646 #if 1 |
| 4647 assert (!forward_search_p || range >= 0); | |
| 4648 #endif | |
| 428 | 4649 startpos += d_size; |
| 4650 } | |
| 4651 else | |
| 4652 { | |
| 826 | 4653 Bytecount d_size; |
| 428 | 4654 /* Note startpos > size1 not >=. If we are on the |
| 4655 string1/string2 boundary, we want to backup into string1. */ | |
| 442 | 4656 d = ((const unsigned char *) |
| 428 | 4657 (startpos > size1 ? string2 - size1 : string1) + startpos); |
| 867 | 4658 DEC_IBYTEPTR_FMT (d, fmt); |
| 4659 d_size = itext_ichar_len_fmt (d, fmt); | |
| 428 | 4660 range += d_size; |
| 826 | 4661 #if 1 |
| 4662 assert (!forward_search_p || range >= 0); | |
| 4663 #endif | |
| 428 | 4664 startpos -= d_size; |
| 4665 } | |
| 4666 } | |
| 1333 | 4667 UNBIND_REGEX_MALLOC_CHECK (); |
| 428 | 4668 return -1; |
| 4669 } /* re_search_2 */ | |
| 826 | 4670 |
| 428 | 4671 |
| 4672 /* Declarations and macros for re_match_2. */ | |
| 4673 | |
| 4674 /* This converts PTR, a pointer into one of the search strings `string1' | |
| 4675 and `string2' into an offset from the beginning of that string. */ | |
| 4676 #define POINTER_TO_OFFSET(ptr) \ | |
| 4677 (FIRST_STRING_P (ptr) \ | |
| 4678 ? ((regoff_t) ((ptr) - string1)) \ | |
| 4679 : ((regoff_t) ((ptr) - string2 + size1))) | |
| 4680 | |
| 4681 /* Macros for dealing with the split strings in re_match_2. */ | |
| 4682 | |
| 4683 #define MATCHING_IN_FIRST_STRING (dend == end_match_1) | |
| 4684 | |
| 4685 /* Call before fetching a character with *d. This switches over to | |
| 4686 string2 if necessary. */ | |
| 826 | 4687 #define REGEX_PREFETCH() \ |
| 428 | 4688 while (d == dend) \ |
| 4689 { \ | |
| 4690 /* End of string2 => fail. */ \ | |
| 4691 if (dend == end_match_2) \ | |
| 4692 goto fail; \ | |
| 4693 /* End of string1 => advance to string2. */ \ | |
| 4694 d = string2; \ | |
| 4695 dend = end_match_2; \ | |
| 4696 } | |
| 4697 | |
| 4698 | |
| 4699 /* Test if at very beginning or at very end of the virtual concatenation | |
| 4700 of `string1' and `string2'. If only one string, it's `string2'. */ | |
| 4701 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) | |
| 4702 #define AT_STRINGS_END(d) ((d) == end2) | |
| 4703 | |
| 4704 /* XEmacs change: | |
| 4705 If the given position straddles the string gap, return the equivalent | |
| 4706 position that is before or after the gap, respectively; otherwise, | |
| 4707 return the same position. */ | |
| 4708 #define POS_BEFORE_GAP_UNSAFE(d) ((d) == string2 ? end1 : (d)) | |
| 4709 #define POS_AFTER_GAP_UNSAFE(d) ((d) == end1 ? string2 : (d)) | |
| 4710 | |
| 4711 /* Test if CH is a word-constituent character. (XEmacs change) */ | |
| 826 | 4712 #define WORDCHAR_P(ch) \ |
| 4713 (SYNTAX (BUFFER_MIRROR_SYNTAX_TABLE (lispbuf), ch) == Sword) | |
| 428 | 4714 |
| 4715 /* Free everything we malloc. */ | |
| 4716 #ifdef MATCH_MAY_ALLOCATE | |
| 1726 | 4717 #define FREE_VAR(var,type) if (var) REGEX_FREE (var, type); var = NULL |
| 428 | 4718 #define FREE_VARIABLES() \ |
| 4719 do { \ | |
| 1333 | 4720 UNBIND_REGEX_MALLOC_CHECK (); \ |
| 428 | 4721 REGEX_FREE_STACK (fail_stack.stack); \ |
| 1726 | 4722 FREE_VAR (regstart, re_char **); \ |
| 4723 FREE_VAR (regend, re_char **); \ | |
| 4724 FREE_VAR (old_regstart, re_char **); \ | |
| 4725 FREE_VAR (old_regend, re_char **); \ | |
| 4726 FREE_VAR (best_regstart, re_char **); \ | |
| 4727 FREE_VAR (best_regend, re_char **); \ | |
| 4728 FREE_VAR (reg_info, register_info_type *); \ | |
| 4729 FREE_VAR (reg_dummy, re_char **); \ | |
| 4730 FREE_VAR (reg_info_dummy, register_info_type *); \ | |
| 428 | 4731 } while (0) |
| 446 | 4732 #else /* not MATCH_MAY_ALLOCATE */ |
| 1333 | 4733 #define FREE_VARIABLES() \ |
| 4734 do { \ | |
| 4735 UNBIND_REGEX_MALLOC_CHECK (); \ | |
| 4736 } while (0) | |
| 446 | 4737 #endif /* MATCH_MAY_ALLOCATE */ |
| 428 | 4738 |
| 4739 /* These values must meet several constraints. They must not be valid | |
| 4740 register values; since we have a limit of 255 registers (because | |
| 4741 we use only one byte in the pattern for the register number), we can | |
| 4742 use numbers larger than 255. They must differ by 1, because of | |
| 4743 NUM_FAILURE_ITEMS above. And the value for the lowest register must | |
| 4744 be larger than the value for the highest register, so we do not try | |
| 4745 to actually save any registers when none are active. */ | |
| 4746 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) | |
| 4747 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) | |
| 4748 | |
| 4749 /* Matching routines. */ | |
| 4750 | |
| 826 | 4751 #ifndef emacs /* XEmacs never uses this. */ |
| 428 | 4752 /* re_match is like re_match_2 except it takes only a single string. */ |
| 4753 | |
| 4754 int | |
| 442 | 4755 re_match (struct re_pattern_buffer *bufp, const char *string, int size, |
| 826 | 4756 int pos, struct re_registers *regs |
| 4757 RE_LISP_CONTEXT_ARGS_DECL) | |
| 428 | 4758 { |
| 446 | 4759 int result = re_match_2_internal (bufp, NULL, 0, (re_char *) string, size, |
| 826 | 4760 pos, regs, size |
| 4761 RE_LISP_CONTEXT_ARGS); | |
| 1333 | 4762 ALLOCA_GARBAGE_COLLECT (); |
| 428 | 4763 return result; |
| 4764 } | |
| 4765 #endif /* not emacs */ | |
| 4766 | |
| 4767 /* re_match_2 matches the compiled pattern in BUFP against the | |
| 4768 (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 and | |
| 4769 SIZE2, respectively). We start matching at POS, and stop matching | |
| 4770 at STOP. | |
| 4771 | |
| 4772 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we | |
| 4773 store offsets for the substring each group matched in REGS. See the | |
| 4774 documentation for exactly how many groups we fill. | |
| 4775 | |
| 4776 We return -1 if no match, -2 if an internal error (such as the | |
| 4777 failure stack overflowing). Otherwise, we return the length of the | |
| 4778 matched substring. */ | |
| 4779 | |
| 4780 int | |
| 442 | 4781 re_match_2 (struct re_pattern_buffer *bufp, const char *string1, |
| 4782 int size1, const char *string2, int size2, int pos, | |
| 826 | 4783 struct re_registers *regs, int stop |
| 4784 RE_LISP_CONTEXT_ARGS_DECL) | |
| 428 | 4785 { |
| 460 | 4786 int result; |
| 4787 | |
| 4788 #ifdef emacs | |
| 826 | 4789 scache = setup_syntax_cache (scache, lispobj, lispbuf, |
| 4790 offset_to_charxpos (lispobj, pos), | |
| 4791 1); | |
| 460 | 4792 #endif |
| 4793 | |
| 4794 result = re_match_2_internal (bufp, (re_char *) string1, size1, | |
| 4795 (re_char *) string2, size2, | |
| 826 | 4796 pos, regs, stop |
| 4797 RE_LISP_CONTEXT_ARGS); | |
| 460 | 4798 |
| 1333 | 4799 ALLOCA_GARBAGE_COLLECT (); |
| 428 | 4800 return result; |
| 4801 } | |
| 4802 | |
| 4803 /* This is a separate function so that we can force an alloca cleanup | |
| 4804 afterwards. */ | |
| 4805 static int | |
| 446 | 4806 re_match_2_internal (struct re_pattern_buffer *bufp, re_char *string1, |
| 4807 int size1, re_char *string2, int size2, int pos, | |
| 826 | 4808 struct re_registers *regs, int stop |
| 2333 | 4809 RE_LISP_CONTEXT_ARGS_MULE_DECL) |
| 428 | 4810 { |
| 4811 /* General temporaries. */ | |
| 4812 int mcnt; | |
| 4813 unsigned char *p1; | |
| 4814 int should_succeed; /* XEmacs change */ | |
| 4815 | |
| 4816 /* Just past the end of the corresponding string. */ | |
| 446 | 4817 re_char *end1, *end2; |
| 428 | 4818 |
| 4819 /* Pointers into string1 and string2, just past the last characters in | |
| 4820 each to consider matching. */ | |
| 446 | 4821 re_char *end_match_1, *end_match_2; |
| 428 | 4822 |
| 4823 /* Where we are in the data, and the end of the current string. */ | |
| 446 | 4824 re_char *d, *dend; |
| 428 | 4825 |
| 4826 /* Where we are in the pattern, and the end of the pattern. */ | |
| 4827 unsigned char *p = bufp->buffer; | |
| 4828 REGISTER unsigned char *pend = p + bufp->used; | |
| 4829 | |
| 4830 /* Mark the opcode just after a start_memory, so we can test for an | |
| 4831 empty subpattern when we get to the stop_memory. */ | |
| 446 | 4832 re_char *just_past_start_mem = 0; |
| 428 | 4833 |
| 4834 /* We use this to map every character in the string. */ | |
| 446 | 4835 RE_TRANSLATE_TYPE translate = bufp->translate; |
| 428 | 4836 |
| 4837 /* Failure point stack. Each place that can handle a failure further | |
| 4838 down the line pushes a failure point on this stack. It consists of | |
| 4839 restart, regend, and reg_info for all registers corresponding to | |
| 4840 the subexpressions we're currently inside, plus the number of such | |
| 4841 registers, and, finally, two char *'s. The first char * is where | |
| 4842 to resume scanning the pattern; the second one is where to resume | |
| 4843 scanning the strings. If the latter is zero, the failure point is | |
| 4844 a ``dummy''; if a failure happens and the failure point is a dummy, | |
| 4845 it gets discarded and the next one is tried. */ | |
| 4846 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
| 4847 fail_stack_type fail_stack; | |
| 4848 #endif | |
| 4849 #ifdef DEBUG | |
| 647 | 4850 static int failure_id; |
| 4851 int nfailure_points_pushed = 0, nfailure_points_popped = 0; | |
| 428 | 4852 #endif |
| 4853 | |
| 771 | 4854 #ifdef REGEX_REL_ALLOC |
| 428 | 4855 /* This holds the pointer to the failure stack, when |
| 4856 it is allocated relocatably. */ | |
| 4857 fail_stack_elt_t *failure_stack_ptr; | |
| 4858 #endif | |
| 4859 | |
| 4860 /* We fill all the registers internally, independent of what we | |
| 4861 return, for use in backreferences. The number here includes | |
| 4862 an element for register zero. */ | |
| 647 | 4863 int num_regs = bufp->re_ngroups + 1; |
| 428 | 4864 |
| 4865 /* The currently active registers. */ | |
| 647 | 4866 int lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
| 4867 int highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
| 428 | 4868 |
| 4869 /* Information on the contents of registers. These are pointers into | |
| 4870 the input strings; they record just what was matched (on this | |
| 4871 attempt) by a subexpression part of the pattern, that is, the | |
| 4872 regnum-th regstart pointer points to where in the pattern we began | |
| 4873 matching and the regnum-th regend points to right after where we | |
| 4874 stopped matching the regnum-th subexpression. (The zeroth register | |
| 4875 keeps track of what the whole pattern matches.) */ | |
| 4876 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
| 446 | 4877 re_char **regstart, **regend; |
| 428 | 4878 #endif |
| 4879 | |
| 4880 /* If a group that's operated upon by a repetition operator fails to | |
| 4881 match anything, then the register for its start will need to be | |
| 4882 restored because it will have been set to wherever in the string we | |
| 4883 are when we last see its open-group operator. Similarly for a | |
| 4884 register's end. */ | |
| 4885 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
| 446 | 4886 re_char **old_regstart, **old_regend; |
| 428 | 4887 #endif |
| 4888 | |
| 4889 /* The is_active field of reg_info helps us keep track of which (possibly | |
| 4890 nested) subexpressions we are currently in. The matched_something | |
| 4891 field of reg_info[reg_num] helps us tell whether or not we have | |
| 4892 matched any of the pattern so far this time through the reg_num-th | |
| 4893 subexpression. These two fields get reset each time through any | |
| 4894 loop their register is in. */ | |
| 4895 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ | |
| 4896 register_info_type *reg_info; | |
| 4897 #endif | |
| 4898 | |
| 4899 /* The following record the register info as found in the above | |
| 4900 variables when we find a match better than any we've seen before. | |
| 4901 This happens as we backtrack through the failure points, which in | |
| 4902 turn happens only if we have not yet matched the entire string. */ | |
| 647 | 4903 int best_regs_set = false; |
| 428 | 4904 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
| 446 | 4905 re_char **best_regstart, **best_regend; |
| 428 | 4906 #endif |
| 4907 | |
| 4908 /* Logically, this is `best_regend[0]'. But we don't want to have to | |
| 4909 allocate space for that if we're not allocating space for anything | |
| 4910 else (see below). Also, we never need info about register 0 for | |
| 4911 any of the other register vectors, and it seems rather a kludge to | |
| 4912 treat `best_regend' differently than the rest. So we keep track of | |
| 4913 the end of the best match so far in a separate variable. We | |
| 4914 initialize this to NULL so that when we backtrack the first time | |
| 4915 and need to test it, it's not garbage. */ | |
| 446 | 4916 re_char *match_end = NULL; |
| 428 | 4917 |
| 4918 /* This helps SET_REGS_MATCHED avoid doing redundant work. */ | |
| 4919 int set_regs_matched_done = 0; | |
| 4920 | |
| 4921 /* Used when we pop values we don't care about. */ | |
| 4922 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ | |
| 446 | 4923 re_char **reg_dummy; |
| 428 | 4924 register_info_type *reg_info_dummy; |
| 4925 #endif | |
| 4926 | |
| 4927 #ifdef DEBUG | |
| 4928 /* Counts the total number of registers pushed. */ | |
| 647 | 4929 int num_regs_pushed = 0; |
| 428 | 4930 #endif |
| 4931 | |
| 4932 /* 1 if this match ends in the same string (string1 or string2) | |
| 4933 as the best previous match. */ | |
| 460 | 4934 re_bool same_str_p; |
| 428 | 4935 |
| 4936 /* 1 if this match is the best seen so far. */ | |
| 460 | 4937 re_bool best_match_p; |
| 428 | 4938 |
| 826 | 4939 #ifdef emacs |
| 4940 Internal_Format fmt = buffer_or_other_internal_format (lispobj); | |
| 1346 | 4941 #ifdef REL_ALLOC |
| 4942 Ibyte *orig_buftext = | |
| 4943 BUFFERP (lispobj) ? | |
| 4944 BYTE_BUF_BYTE_ADDRESS (XBUFFER (lispobj), | |
| 4945 BYTE_BUF_BEGV (XBUFFER (lispobj))) : | |
| 4946 0; | |
| 4947 #endif | |
| 4948 | |
| 1333 | 4949 #ifdef ERROR_CHECK_MALLOC |
| 4950 int depth = bind_regex_malloc_disallowed (1); | |
| 4951 #endif | |
| 826 | 4952 #endif /* emacs */ |
| 771 | 4953 |
| 5041 | 4954 DEBUG_MATCH_PRINT1 ("\n\nEntering re_match_2.\n"); |
| 428 | 4955 |
| 1333 | 4956 BEGIN_REGEX_MALLOC_OK (); |
| 428 | 4957 INIT_FAIL_STACK (); |
| 1333 | 4958 END_REGEX_MALLOC_OK (); |
| 428 | 4959 |
| 4960 #ifdef MATCH_MAY_ALLOCATE | |
| 4961 /* Do not bother to initialize all the register variables if there are | |
| 4962 no groups in the pattern, as it takes a fair amount of time. If | |
| 4963 there are groups, we include space for register 0 (the whole | |
| 4964 pattern), even though we never use it, since it simplifies the | |
| 4965 array indexing. We should fix this. */ | |
| 502 | 4966 if (bufp->re_ngroups) |
| 428 | 4967 { |
| 1333 | 4968 BEGIN_REGEX_MALLOC_OK (); |
| 446 | 4969 regstart = REGEX_TALLOC (num_regs, re_char *); |
| 4970 regend = REGEX_TALLOC (num_regs, re_char *); | |
| 4971 old_regstart = REGEX_TALLOC (num_regs, re_char *); | |
| 4972 old_regend = REGEX_TALLOC (num_regs, re_char *); | |
| 4973 best_regstart = REGEX_TALLOC (num_regs, re_char *); | |
| 4974 best_regend = REGEX_TALLOC (num_regs, re_char *); | |
| 428 | 4975 reg_info = REGEX_TALLOC (num_regs, register_info_type); |
| 446 | 4976 reg_dummy = REGEX_TALLOC (num_regs, re_char *); |
| 428 | 4977 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); |
| 1333 | 4978 END_REGEX_MALLOC_OK (); |
| 428 | 4979 |
| 4980 if (!(regstart && regend && old_regstart && old_regend && reg_info | |
| 4981 && best_regstart && best_regend && reg_dummy && reg_info_dummy)) | |
| 4982 { | |
| 4983 FREE_VARIABLES (); | |
| 4984 return -2; | |
| 4985 } | |
| 4986 } | |
| 4987 else | |
| 4988 { | |
| 4989 /* We must initialize all our variables to NULL, so that | |
| 4990 `FREE_VARIABLES' doesn't try to free them. */ | |
| 4991 regstart = regend = old_regstart = old_regend = best_regstart | |
| 4992 = best_regend = reg_dummy = NULL; | |
| 4993 reg_info = reg_info_dummy = (register_info_type *) NULL; | |
| 4994 } | |
| 4995 #endif /* MATCH_MAY_ALLOCATE */ | |
| 4996 | |
| 1333 | 4997 #if defined (emacs) && defined (REL_ALLOC) |
| 4998 { | |
| 4999 /* If the allocations above (or the call to setup_syntax_cache() in | |
| 5000 re_match_2) caused a rel-alloc relocation, then fix up the data | |
| 5001 pointers */ | |
| 1346 | 5002 Bytecount offset = offset_post_relocation (lispobj, orig_buftext); |
| 1333 | 5003 if (offset) |
| 5004 { | |
| 5005 string1 += offset; | |
| 5006 string2 += offset; | |
| 5007 } | |
| 5008 } | |
| 5009 #endif /* defined (emacs) && defined (REL_ALLOC) */ | |
| 5010 | |
| 428 | 5011 /* The starting position is bogus. */ |
| 5012 if (pos < 0 || pos > size1 + size2) | |
| 5013 { | |
| 5014 FREE_VARIABLES (); | |
| 5015 return -1; | |
| 5016 } | |
| 5017 | |
| 5018 /* Initialize subexpression text positions to -1 to mark ones that no | |
| 5019 start_memory/stop_memory has been seen for. Also initialize the | |
| 5020 register information struct. */ | |
| 5021 for (mcnt = 1; mcnt < num_regs; mcnt++) | |
| 5022 { | |
| 5023 regstart[mcnt] = regend[mcnt] | |
| 5024 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; | |
| 5025 | |
| 5026 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; | |
| 5027 IS_ACTIVE (reg_info[mcnt]) = 0; | |
| 5028 MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
| 5029 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; | |
| 5030 } | |
| 5031 /* We move `string1' into `string2' if the latter's empty -- but not if | |
| 5032 `string1' is null. */ | |
| 5033 if (size2 == 0 && string1 != NULL) | |
| 5034 { | |
| 5035 string2 = string1; | |
| 5036 size2 = size1; | |
| 5037 string1 = 0; | |
| 5038 size1 = 0; | |
| 5039 } | |
| 5040 end1 = string1 + size1; | |
| 5041 end2 = string2 + size2; | |
| 5042 | |
| 5043 /* Compute where to stop matching, within the two strings. */ | |
| 5044 if (stop <= size1) | |
| 5045 { | |
| 5046 end_match_1 = string1 + stop; | |
| 5047 end_match_2 = string2; | |
| 5048 } | |
| 5049 else | |
| 5050 { | |
| 5051 end_match_1 = end1; | |
| 5052 end_match_2 = string2 + stop - size1; | |
| 5053 } | |
| 5054 | |
| 5055 /* `p' scans through the pattern as `d' scans through the data. | |
| 5056 `dend' is the end of the input string that `d' points within. `d' | |
| 5057 is advanced into the following input string whenever necessary, but | |
| 5058 this happens before fetching; therefore, at the beginning of the | |
| 5059 loop, `d' can be pointing at the end of a string, but it cannot | |
| 5060 equal `string2'. */ | |
| 5061 if (size1 > 0 && pos <= size1) | |
| 5062 { | |
| 5063 d = string1 + pos; | |
| 5064 dend = end_match_1; | |
| 5065 } | |
| 5066 else | |
| 5067 { | |
| 5068 d = string2 + pos - size1; | |
| 5069 dend = end_match_2; | |
| 5070 } | |
| 5071 | |
| 5041 | 5072 DEBUG_MATCH_PRINT1 ("The compiled pattern is: \n"); |
| 5073 DEBUG_MATCH_PRINT_COMPILED_PATTERN (bufp, p, pend); | |
| 5074 DEBUG_MATCH_PRINT1 ("The string to match is: `"); | |
| 5075 DEBUG_MATCH_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); | |
| 5076 DEBUG_MATCH_PRINT1 ("'\n"); | |
| 428 | 5077 |
| 5078 /* This loops over pattern commands. It exits by returning from the | |
| 5079 function if the match is complete, or it drops through if the match | |
| 5080 fails at this starting point in the input data. */ | |
| 5081 for (;;) | |
| 5082 { | |
| 5041 | 5083 DEBUG_MATCH_PRINT2 ("\n0x%lx: ", (long) p); |
| 428 | 5084 #ifdef emacs /* XEmacs added, w/removal of immediate_quit */ |
| 5085 if (!no_quit_in_re_search) | |
| 1333 | 5086 { |
| 5087 BEGIN_REGEX_MALLOC_OK (); | |
| 5088 QUIT; | |
| 5089 END_REGEX_MALLOC_OK (); | |
| 1346 | 5090 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); |
| 1333 | 5091 } |
| 428 | 5092 #endif |
| 5093 | |
| 5094 if (p == pend) | |
| 5095 { /* End of pattern means we might have succeeded. */ | |
| 5041 | 5096 DEBUG_MATCH_PRINT1 ("end of pattern ... "); |
| 428 | 5097 |
| 5098 /* If we haven't matched the entire string, and we want the | |
| 5099 longest match, try backtracking. */ | |
| 5100 if (d != end_match_2) | |
| 5101 { | |
| 5102 same_str_p = (FIRST_STRING_P (match_end) | |
| 5103 == MATCHING_IN_FIRST_STRING); | |
| 5104 | |
| 5105 /* AIX compiler got confused when this was combined | |
| 5106 with the previous declaration. */ | |
| 5107 if (same_str_p) | |
| 5108 best_match_p = d > match_end; | |
| 5109 else | |
| 5110 best_match_p = !MATCHING_IN_FIRST_STRING; | |
| 5111 | |
| 5041 | 5112 DEBUG_MATCH_PRINT1 ("backtracking.\n"); |
| 428 | 5113 |
| 5114 if (!FAIL_STACK_EMPTY ()) | |
| 5115 { /* More failure points to try. */ | |
| 5116 | |
| 5117 /* If exceeds best match so far, save it. */ | |
| 5118 if (!best_regs_set || best_match_p) | |
| 5119 { | |
| 5120 best_regs_set = true; | |
| 5121 match_end = d; | |
| 5122 | |
| 5041 | 5123 DEBUG_MATCH_PRINT1 ("\nSAVING match as best so far.\n"); |
| 428 | 5124 |
| 5125 for (mcnt = 1; mcnt < num_regs; mcnt++) | |
| 5126 { | |
| 5127 best_regstart[mcnt] = regstart[mcnt]; | |
| 5128 best_regend[mcnt] = regend[mcnt]; | |
| 5129 } | |
| 5130 } | |
| 5131 goto fail; | |
| 5132 } | |
| 5133 | |
| 5134 /* If no failure points, don't restore garbage. And if | |
| 5135 last match is real best match, don't restore second | |
| 5136 best one. */ | |
| 5137 else if (best_regs_set && !best_match_p) | |
| 5138 { | |
| 5139 restore_best_regs: | |
| 5140 /* Restore best match. It may happen that `dend == | |
| 5141 end_match_1' while the restored d is in string2. | |
| 5142 For example, the pattern `x.*y.*z' against the | |
| 5143 strings `x-' and `y-z-', if the two strings are | |
| 5144 not consecutive in memory. */ | |
| 5041 | 5145 DEBUG_MATCH_PRINT1 ("Restoring best registers.\n"); |
| 428 | 5146 |
| 5147 d = match_end; | |
| 5148 dend = ((d >= string1 && d <= end1) | |
| 5149 ? end_match_1 : end_match_2); | |
| 5150 | |
| 5151 for (mcnt = 1; mcnt < num_regs; mcnt++) | |
| 5152 { | |
| 5153 regstart[mcnt] = best_regstart[mcnt]; | |
| 5154 regend[mcnt] = best_regend[mcnt]; | |
| 5155 } | |
| 5156 } | |
| 5157 } /* d != end_match_2 */ | |
| 5158 | |
| 5159 succeed_label: | |
| 5041 | 5160 DEBUG_MATCH_PRINT1 ("Accepting match.\n"); |
| 428 | 5161 |
| 5162 /* If caller wants register contents data back, do it. */ | |
| 1028 | 5163 { |
| 5164 int num_nonshy_regs = bufp->re_nsub + 1; | |
| 5165 if (regs && !bufp->no_sub) | |
| 5166 { | |
| 5167 /* Have the register data arrays been allocated? */ | |
| 5168 if (bufp->regs_allocated == REGS_UNALLOCATED) | |
| 5169 { /* No. So allocate them with malloc. We need one | |
| 5170 extra element beyond `num_regs' for the `-1' marker | |
| 5171 GNU code uses. */ | |
| 5172 regs->num_regs = MAX (RE_NREGS, num_nonshy_regs + 1); | |
| 1333 | 5173 BEGIN_REGEX_MALLOC_OK (); |
| 1028 | 5174 regs->start = TALLOC (regs->num_regs, regoff_t); |
| 5175 regs->end = TALLOC (regs->num_regs, regoff_t); | |
| 1333 | 5176 END_REGEX_MALLOC_OK (); |
| 5177 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 1028 | 5178 if (regs->start == NULL || regs->end == NULL) |
| 5179 { | |
| 5180 FREE_VARIABLES (); | |
| 5181 return -2; | |
| 5182 } | |
| 5183 bufp->regs_allocated = REGS_REALLOCATE; | |
| 5184 } | |
| 5185 else if (bufp->regs_allocated == REGS_REALLOCATE) | |
| 5186 { /* Yes. If we need more elements than were already | |
| 5187 allocated, reallocate them. If we need fewer, just | |
| 5188 leave it alone. */ | |
| 5189 if (regs->num_regs < num_nonshy_regs + 1) | |
| 5190 { | |
| 5191 regs->num_regs = num_nonshy_regs + 1; | |
| 1333 | 5192 BEGIN_REGEX_MALLOC_OK (); |
| 1028 | 5193 RETALLOC (regs->start, regs->num_regs, regoff_t); |
| 5194 RETALLOC (regs->end, regs->num_regs, regoff_t); | |
| 1333 | 5195 END_REGEX_MALLOC_OK (); |
| 5196 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 1028 | 5197 if (regs->start == NULL || regs->end == NULL) |
| 5198 { | |
| 5199 FREE_VARIABLES (); | |
| 5200 return -2; | |
| 5201 } | |
| 5202 } | |
| 5203 } | |
| 5204 else | |
| 5205 { | |
| 5206 /* The braces fend off a "empty body in an else-statement" | |
| 5207 warning under GCC when assert expands to nothing. */ | |
| 5208 assert (bufp->regs_allocated == REGS_FIXED); | |
| 5209 } | |
| 5210 | |
| 5211 /* Convert the pointer data in `regstart' and `regend' to | |
| 5212 indices. Register zero has to be set differently, | |
| 5213 since we haven't kept track of any info for it. */ | |
| 5214 if (regs->num_regs > 0) | |
| 5215 { | |
| 5216 regs->start[0] = pos; | |
| 5217 regs->end[0] = (MATCHING_IN_FIRST_STRING | |
| 5218 ? ((regoff_t) (d - string1)) | |
| 5219 : ((regoff_t) (d - string2 + size1))); | |
| 5220 } | |
| 5221 | |
| 2639 | 5222 /* Map over the NUM_NONSHY_REGS non-shy internal registers. |
| 5223 Copy each into the corresponding external register. | |
| 5224 MCNT indexes external registers. */ | |
| 1028 | 5225 for (mcnt = 1; mcnt < MIN (num_nonshy_regs, regs->num_regs); |
| 5226 mcnt++) | |
| 5227 { | |
| 5228 int internal_reg = bufp->external_to_internal_register[mcnt]; | |
| 5229 if (REG_UNSET (regstart[internal_reg]) || | |
| 5230 REG_UNSET (regend[internal_reg])) | |
| 5231 regs->start[mcnt] = regs->end[mcnt] = -1; | |
| 5232 else | |
| 5233 { | |
| 5234 regs->start[mcnt] = | |
| 5235 (regoff_t) POINTER_TO_OFFSET (regstart[internal_reg]); | |
| 5236 regs->end[mcnt] = | |
| 5237 (regoff_t) POINTER_TO_OFFSET (regend[internal_reg]); | |
| 5238 } | |
| 5239 } | |
| 5240 } /* regs && !bufp->no_sub */ | |
| 5241 | |
| 5242 /* If we have regs and the regs structure has more elements than | |
| 2639 | 5243 were in the pattern, set the extra elements starting with |
| 5244 NUM_NONSHY_REGS to -1. If we (re)allocated the registers, | |
| 5245 this is the case, because we always allocate enough to have | |
| 5246 at least one -1 at the end. | |
| 1028 | 5247 |
| 5248 We do this even when no_sub is set because some applications | |
| 5249 (XEmacs) reuse register structures which may contain stale | |
| 5250 information, and permit attempts to access those registers. | |
| 5251 | |
| 5252 It would be possible to require the caller to do this, but we'd | |
| 5253 have to change the API for this function to reflect that, and | |
| 1425 | 5254 audit all callers. Note: as of 2003-04-17 callers in XEmacs |
| 5255 do clear the registers, but it's safer to leave this code in | |
| 5256 because of reallocation. | |
| 5257 */ | |
| 1028 | 5258 if (regs && regs->num_regs > 0) |
| 5259 for (mcnt = num_nonshy_regs; mcnt < regs->num_regs; mcnt++) | |
| 5260 regs->start[mcnt] = regs->end[mcnt] = -1; | |
| 5261 } | |
| 5041 | 5262 DEBUG_MATCH_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
| 428 | 5263 nfailure_points_pushed, nfailure_points_popped, |
| 5264 nfailure_points_pushed - nfailure_points_popped); | |
| 5041 | 5265 DEBUG_MATCH_PRINT2 ("%u registers pushed.\n", num_regs_pushed); |
| 428 | 5266 |
| 5267 mcnt = d - pos - (MATCHING_IN_FIRST_STRING | |
| 5268 ? string1 | |
| 5269 : string2 - size1); | |
| 5270 | |
| 5041 | 5271 DEBUG_MATCH_PRINT2 ("Returning %d from re_match_2.\n", mcnt); |
| 428 | 5272 |
| 5273 FREE_VARIABLES (); | |
| 5274 return mcnt; | |
| 5275 } | |
| 5276 | |
| 5277 /* Otherwise match next pattern command. */ | |
|
4759
aa5ed11f473b
Remove support for obsolete systems. See xemacs-patches message with ID
Jerry James <james@xemacs.org>
parents:
4750
diff
changeset
|
5278 switch ((re_opcode_t) *p++) |
| 428 | 5279 { |
| 5280 /* Ignore these. Used to ignore the n of succeed_n's which | |
| 5281 currently have n == 0. */ | |
| 5282 case no_op: | |
| 5041 | 5283 DEBUG_MATCH_PRINT1 ("EXECUTING no_op.\n"); |
| 428 | 5284 break; |
| 5285 | |
| 5286 case succeed: | |
| 5041 | 5287 DEBUG_MATCH_PRINT1 ("EXECUTING succeed.\n"); |
| 428 | 5288 goto succeed_label; |
| 5289 | |
| 826 | 5290 /* Match exactly a string of length n in the pattern. The |
| 5291 following byte in the pattern defines n, and the n bytes after | |
| 5292 that make up the string to match. (Under Mule, this will be in | |
| 5293 the default internal format.) */ | |
| 428 | 5294 case exactn: |
| 5295 mcnt = *p++; | |
| 5041 | 5296 DEBUG_MATCH_PRINT2 ("EXECUTING exactn %d.\n", mcnt); |
| 428 | 5297 |
| 5298 /* This is written out as an if-else so we don't waste time | |
| 5299 testing `translate' inside the loop. */ | |
| 446 | 5300 if (TRANSLATE_P (translate)) |
| 428 | 5301 { |
| 5302 do | |
| 5303 { | |
| 446 | 5304 #ifdef MULE |
| 5305 Bytecount pat_len; | |
| 5306 | |
| 450 | 5307 REGEX_PREFETCH (); |
| 867 | 5308 if (RE_TRANSLATE_1 (itext_ichar_fmt (d, fmt, lispobj)) |
| 5309 != itext_ichar (p)) | |
| 428 | 5310 goto fail; |
| 446 | 5311 |
| 867 | 5312 pat_len = itext_ichar_len (p); |
| 446 | 5313 p += pat_len; |
| 867 | 5314 INC_IBYTEPTR_FMT (d, fmt); |
| 446 | 5315 |
| 5316 mcnt -= pat_len; | |
| 5317 #else /* not MULE */ | |
| 450 | 5318 REGEX_PREFETCH (); |
| 826 | 5319 if ((unsigned char) RE_TRANSLATE_1 (*d++) != *p++) |
| 446 | 5320 goto fail; |
| 5321 mcnt--; | |
| 5322 #endif | |
| 428 | 5323 } |
| 446 | 5324 while (mcnt > 0); |
| 428 | 5325 } |
| 5326 else | |
| 5327 { | |
| 826 | 5328 #ifdef MULE |
| 5329 /* If buffer format is default, then we can shortcut and just | |
| 5330 compare the text directly, byte by byte. Otherwise, we | |
| 5331 need to go character by character. */ | |
| 5332 if (fmt != FORMAT_DEFAULT) | |
| 428 | 5333 { |
| 826 | 5334 do |
| 5335 { | |
| 5336 Bytecount pat_len; | |
| 5337 | |
| 5338 REGEX_PREFETCH (); | |
| 867 | 5339 if (itext_ichar_fmt (d, fmt, lispobj) != |
| 5340 itext_ichar (p)) | |
| 826 | 5341 goto fail; |
| 5342 | |
| 867 | 5343 pat_len = itext_ichar_len (p); |
| 826 | 5344 p += pat_len; |
| 867 | 5345 INC_IBYTEPTR_FMT (d, fmt); |
| 826 | 5346 |
| 5347 mcnt -= pat_len; | |
| 5348 } | |
| 5349 while (mcnt > 0); | |
| 428 | 5350 } |
| 826 | 5351 else |
| 5352 #endif | |
| 5353 { | |
| 5354 do | |
| 5355 { | |
| 5356 REGEX_PREFETCH (); | |
| 5357 if (*d++ != *p++) goto fail; | |
| 5358 mcnt--; | |
| 5359 } | |
| 5360 while (mcnt > 0); | |
| 5361 } | |
| 428 | 5362 } |
| 5363 SET_REGS_MATCHED (); | |
| 5364 break; | |
| 5365 | |
| 5366 | |
| 5367 /* Match any character except possibly a newline or a null. */ | |
| 5368 case anychar: | |
| 5041 | 5369 DEBUG_MATCH_PRINT1 ("EXECUTING anychar.\n"); |
| 428 | 5370 |
| 450 | 5371 REGEX_PREFETCH (); |
| 428 | 5372 |
| 826 | 5373 if ((!(bufp->syntax & RE_DOT_NEWLINE) && |
| 867 | 5374 RE_TRANSLATE (itext_ichar_fmt (d, fmt, lispobj)) == '\n') |
| 826 | 5375 || (bufp->syntax & RE_DOT_NOT_NULL && |
| 867 | 5376 RE_TRANSLATE (itext_ichar_fmt (d, fmt, lispobj)) == |
| 826 | 5377 '\000')) |
| 428 | 5378 goto fail; |
| 5379 | |
| 5380 SET_REGS_MATCHED (); | |
| 5041 | 5381 DEBUG_MATCH_PRINT2 (" Matched `%d'.\n", *d); |
| 867 | 5382 INC_IBYTEPTR_FMT (d, fmt); /* XEmacs change */ |
| 428 | 5383 break; |
| 5384 | |
| 5385 | |
| 5386 case charset: | |
| 5387 case charset_not: | |
| 5388 { | |
| 1414 | 5389 REGISTER Ichar c; |
| 460 | 5390 re_bool not_p = (re_opcode_t) *(p - 1) == charset_not; |
| 458 | 5391 |
| 5041 | 5392 DEBUG_MATCH_PRINT2 ("EXECUTING charset%s.\n", not_p ? "_not" : ""); |
| 428 | 5393 |
| 450 | 5394 REGEX_PREFETCH (); |
| 867 | 5395 c = itext_ichar_fmt (d, fmt, lispobj); |
| 826 | 5396 c = RE_TRANSLATE (c); /* The character to match. */ |
| 428 | 5397 |
| 647 | 5398 /* Cast to `unsigned int' instead of `unsigned char' in case the |
| 428 | 5399 bit list is a full 32 bytes long. */ |
| 1414 | 5400 if ((unsigned int)c < (unsigned int) (*p * BYTEWIDTH) |
| 428 | 5401 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
| 458 | 5402 not_p = !not_p; |
| 428 | 5403 |
| 5404 p += 1 + *p; | |
| 5405 | |
| 458 | 5406 if (!not_p) goto fail; |
| 428 | 5407 |
| 5408 SET_REGS_MATCHED (); | |
| 867 | 5409 INC_IBYTEPTR_FMT (d, fmt); /* XEmacs change */ |
| 428 | 5410 break; |
| 5411 } | |
| 5412 | |
| 5413 #ifdef MULE | |
| 5414 case charset_mule: | |
| 5415 case charset_mule_not: | |
| 5416 { | |
| 867 | 5417 REGISTER Ichar c; |
| 460 | 5418 re_bool not_p = (re_opcode_t) *(p - 1) == charset_mule_not; |
| 458 | 5419 |
| 5041 | 5420 DEBUG_MATCH_PRINT2 ("EXECUTING charset_mule%s.\n", not_p ? "_not" : ""); |
| 428 | 5421 |
| 450 | 5422 REGEX_PREFETCH (); |
| 867 | 5423 c = itext_ichar_fmt (d, fmt, lispobj); |
| 826 | 5424 c = RE_TRANSLATE (c); /* The character to match. */ |
| 428 | 5425 |
| 5426 if (EQ (Qt, unified_range_table_lookup (p, c, Qnil))) | |
| 458 | 5427 not_p = !not_p; |
| 428 | 5428 |
| 5429 p += unified_range_table_bytes_used (p); | |
| 5430 | |
| 458 | 5431 if (!not_p) goto fail; |
| 428 | 5432 |
| 5433 SET_REGS_MATCHED (); | |
| 867 | 5434 INC_IBYTEPTR_FMT (d, fmt); |
| 428 | 5435 break; |
| 5436 } | |
| 5437 #endif /* MULE */ | |
| 5438 | |
| 5439 | |
| 5440 /* The beginning of a group is represented by start_memory. | |
| 5441 The arguments are the register number in the next byte, and the | |
| 5442 number of groups inner to this one in the next. The text | |
| 5443 matched within the group is recorded (in the internal | |
| 5444 registers data structure) under the register number. */ | |
| 5445 case start_memory: | |
| 5041 | 5446 DEBUG_MATCH_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); |
| 428 | 5447 |
| 5448 /* Find out if this group can match the empty string. */ | |
| 5449 p1 = p; /* To send to group_match_null_string_p. */ | |
| 5450 | |
| 5451 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) | |
| 2639 | 5452 REG_MATCH_NULL_STRING_P (reg_info[*p]) |
| 5453 = group_match_null_string_p (&p1, pend, reg_info); | |
| 5454 | |
| 5041 | 5455 DEBUG_MATCH_PRINT2 (" group CAN%s match null string\n", |
| 2639 | 5456 REG_MATCH_NULL_STRING_P (reg_info[*p]) ? "NOT" : ""); |
| 428 | 5457 |
| 5458 /* Save the position in the string where we were the last time | |
| 5459 we were at this open-group operator in case the group is | |
| 5460 operated upon by a repetition operator, e.g., with `(a*)*b' | |
| 5461 against `ab'; then we want to ignore where we are now in | |
| 5462 the string in case this attempt to match fails. */ | |
| 5463 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
| 5464 ? REG_UNSET (regstart[*p]) ? d : regstart[*p] | |
| 5465 : regstart[*p]; | |
| 5041 | 5466 DEBUG_MATCH_PRINT2 (" old_regstart: %d\n", |
| 428 | 5467 POINTER_TO_OFFSET (old_regstart[*p])); |
| 5468 | |
| 5469 regstart[*p] = d; | |
| 5041 | 5470 DEBUG_MATCH_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); |
| 428 | 5471 |
| 5472 IS_ACTIVE (reg_info[*p]) = 1; | |
| 5473 MATCHED_SOMETHING (reg_info[*p]) = 0; | |
| 5474 | |
| 5475 /* Clear this whenever we change the register activity status. */ | |
| 5476 set_regs_matched_done = 0; | |
| 5477 | |
| 5478 /* This is the new highest active register. */ | |
| 5479 highest_active_reg = *p; | |
| 5480 | |
| 5481 /* If nothing was active before, this is the new lowest active | |
| 5482 register. */ | |
| 5483 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
| 5484 lowest_active_reg = *p; | |
| 5485 | |
| 5486 /* Move past the register number and inner group count. */ | |
| 5487 p += 2; | |
| 5488 just_past_start_mem = p; | |
| 5489 | |
| 5490 break; | |
| 5491 | |
| 5492 | |
| 5493 /* The stop_memory opcode represents the end of a group. Its | |
| 5494 arguments are the same as start_memory's: the register | |
| 5495 number, and the number of inner groups. */ | |
| 5496 case stop_memory: | |
| 5041 | 5497 DEBUG_MATCH_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); |
| 428 | 5498 |
| 5499 /* We need to save the string position the last time we were at | |
| 5500 this close-group operator in case the group is operated | |
| 5501 upon by a repetition operator, e.g., with `((a*)*(b*)*)*' | |
| 5502 against `aba'; then we want to ignore where we are now in | |
| 5503 the string in case this attempt to match fails. */ | |
| 5504 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) | |
| 5505 ? REG_UNSET (regend[*p]) ? d : regend[*p] | |
| 5506 : regend[*p]; | |
| 5041 | 5507 DEBUG_MATCH_PRINT2 (" old_regend: %d\n", |
| 428 | 5508 POINTER_TO_OFFSET (old_regend[*p])); |
| 5509 | |
| 5510 regend[*p] = d; | |
| 5041 | 5511 DEBUG_MATCH_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); |
| 428 | 5512 |
| 5513 /* This register isn't active anymore. */ | |
| 5514 IS_ACTIVE (reg_info[*p]) = 0; | |
| 5515 | |
| 5516 /* Clear this whenever we change the register activity status. */ | |
| 5517 set_regs_matched_done = 0; | |
| 5518 | |
| 5519 /* If this was the only register active, nothing is active | |
| 5520 anymore. */ | |
| 5521 if (lowest_active_reg == highest_active_reg) | |
| 5522 { | |
| 5523 lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
| 5524 highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
| 5525 } | |
| 5526 else | |
| 5527 { /* We must scan for the new highest active register, since | |
| 5528 it isn't necessarily one less than now: consider | |
| 5529 (a(b)c(d(e)f)g). When group 3 ends, after the f), the | |
| 5530 new highest active register is 1. */ | |
| 5531 unsigned char r = *p - 1; | |
| 5532 while (r > 0 && !IS_ACTIVE (reg_info[r])) | |
| 5533 r--; | |
| 5534 | |
| 5535 /* If we end up at register zero, that means that we saved | |
| 5536 the registers as the result of an `on_failure_jump', not | |
| 5537 a `start_memory', and we jumped to past the innermost | |
| 5538 `stop_memory'. For example, in ((.)*) we save | |
| 5539 registers 1 and 2 as a result of the *, but when we pop | |
| 5540 back to the second ), we are at the stop_memory 1. | |
| 5541 Thus, nothing is active. */ | |
| 5542 if (r == 0) | |
| 5543 { | |
| 5544 lowest_active_reg = NO_LOWEST_ACTIVE_REG; | |
| 5545 highest_active_reg = NO_HIGHEST_ACTIVE_REG; | |
| 5546 } | |
| 5547 else | |
| 5548 { | |
| 5549 highest_active_reg = r; | |
| 5550 | |
| 5551 /* 98/9/21 jhod: We've also gotta set lowest_active_reg, don't we? */ | |
| 5552 r = 1; | |
| 5553 while (r < highest_active_reg && !IS_ACTIVE(reg_info[r])) | |
| 5554 r++; | |
| 5555 lowest_active_reg = r; | |
| 5556 } | |
| 5557 } | |
| 5558 | |
| 5559 /* If just failed to match something this time around with a | |
| 5560 group that's operated on by a repetition operator, try to | |
| 5561 force exit from the ``loop'', and restore the register | |
| 5562 information for this group that we had before trying this | |
| 5563 last match. */ | |
| 5564 if ((!MATCHED_SOMETHING (reg_info[*p]) | |
| 5565 || just_past_start_mem == p - 1) | |
| 5566 && (p + 2) < pend) | |
| 5567 { | |
| 460 | 5568 re_bool is_a_jump_n = false; |
| 428 | 5569 |
| 5570 p1 = p + 2; | |
| 5571 mcnt = 0; | |
| 5572 switch ((re_opcode_t) *p1++) | |
| 5573 { | |
| 5574 case jump_n: | |
| 5575 is_a_jump_n = true; | |
| 5576 case pop_failure_jump: | |
| 5577 case maybe_pop_jump: | |
| 5578 case jump: | |
| 5579 case dummy_failure_jump: | |
| 5580 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
| 5581 if (is_a_jump_n) | |
| 5582 p1 += 2; | |
| 5583 break; | |
| 5584 | |
| 5585 default: | |
| 5586 /* do nothing */ ; | |
| 5587 } | |
| 5588 p1 += mcnt; | |
| 5589 | |
| 5590 /* If the next operation is a jump backwards in the pattern | |
| 5591 to an on_failure_jump right before the start_memory | |
| 5592 corresponding to this stop_memory, exit from the loop | |
| 5593 by forcing a failure after pushing on the stack the | |
| 5594 on_failure_jump's jump in the pattern, and d. */ | |
| 5595 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump | |
| 5596 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) | |
| 5597 { | |
| 5598 /* If this group ever matched anything, then restore | |
| 5599 what its registers were before trying this last | |
| 5600 failed match, e.g., with `(a*)*b' against `ab' for | |
| 5601 regstart[1], and, e.g., with `((a*)*(b*)*)*' | |
| 5602 against `aba' for regend[3]. | |
| 5603 | |
| 5604 Also restore the registers for inner groups for, | |
| 5605 e.g., `((a*)(b*))*' against `aba' (register 3 would | |
| 5606 otherwise get trashed). */ | |
| 5607 | |
| 5608 if (EVER_MATCHED_SOMETHING (reg_info[*p])) | |
| 5609 { | |
| 647 | 5610 int r; |
| 428 | 5611 |
| 5612 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; | |
| 5613 | |
| 5614 /* Restore this and inner groups' (if any) registers. */ | |
| 5615 for (r = *p; r < *p + *(p + 1); r++) | |
| 5616 { | |
| 5617 regstart[r] = old_regstart[r]; | |
| 5618 | |
| 5619 /* xx why this test? */ | |
| 5620 if (old_regend[r] >= regstart[r]) | |
| 5621 regend[r] = old_regend[r]; | |
| 5622 } | |
| 5623 } | |
| 5624 p1++; | |
| 5625 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
| 5626 PUSH_FAILURE_POINT (p1 + mcnt, d, -2); | |
| 5627 | |
| 5628 goto fail; | |
| 5629 } | |
| 5630 } | |
| 5631 | |
| 5632 /* Move past the register number and the inner group count. */ | |
| 5633 p += 2; | |
| 5634 break; | |
| 5635 | |
| 5636 | |
| 5637 /* \<digit> has been turned into a `duplicate' command which is | |
| 502 | 5638 followed by the numeric value of <digit> as the register number. |
| 5639 (Already passed through external-to-internal-register mapping, | |
| 5640 so it refers to the actual group number, not the non-shy-only | |
| 5641 numbering used in the external world.) */ | |
| 428 | 5642 case duplicate: |
| 5643 { | |
| 446 | 5644 REGISTER re_char *d2, *dend2; |
| 502 | 5645 /* Get which register to match against. */ |
| 5646 int regno = *p++; | |
| 5041 | 5647 DEBUG_MATCH_PRINT2 ("EXECUTING duplicate %d.\n", regno); |
| 428 | 5648 |
| 5649 /* Can't back reference a group which we've never matched. */ | |
| 5650 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) | |
| 5651 goto fail; | |
| 5652 | |
| 5653 /* Where in input to try to start matching. */ | |
| 5654 d2 = regstart[regno]; | |
| 5655 | |
| 5656 /* Where to stop matching; if both the place to start and | |
| 5657 the place to stop matching are in the same string, then | |
| 5658 set to the place to stop, otherwise, for now have to use | |
| 5659 the end of the first string. */ | |
| 5660 | |
| 5661 dend2 = ((FIRST_STRING_P (regstart[regno]) | |
| 5662 == FIRST_STRING_P (regend[regno])) | |
| 5663 ? regend[regno] : end_match_1); | |
| 5664 for (;;) | |
| 5665 { | |
| 5666 /* If necessary, advance to next segment in register | |
| 5667 contents. */ | |
| 5668 while (d2 == dend2) | |
| 5669 { | |
| 5670 if (dend2 == end_match_2) break; | |
| 5671 if (dend2 == regend[regno]) break; | |
| 5672 | |
| 5673 /* End of string1 => advance to string2. */ | |
| 5674 d2 = string2; | |
| 5675 dend2 = regend[regno]; | |
| 5676 } | |
| 5677 /* At end of register contents => success */ | |
| 5678 if (d2 == dend2) break; | |
| 5679 | |
| 5680 /* If necessary, advance to next segment in data. */ | |
| 450 | 5681 REGEX_PREFETCH (); |
| 428 | 5682 |
| 5683 /* How many characters left in this segment to match. */ | |
| 5684 mcnt = dend - d; | |
| 5685 | |
| 5686 /* Want how many consecutive characters we can match in | |
| 5687 one shot, so, if necessary, adjust the count. */ | |
| 5688 if (mcnt > dend2 - d2) | |
| 5689 mcnt = dend2 - d2; | |
| 5690 | |
| 5691 /* Compare that many; failure if mismatch, else move | |
| 5692 past them. */ | |
| 446 | 5693 if (TRANSLATE_P (translate) |
| 826 | 5694 ? bcmp_translate (d, d2, mcnt, translate |
| 5695 #ifdef emacs | |
| 5696 , fmt, lispobj | |
| 5697 #endif | |
| 5698 ) | |
| 428 | 5699 : memcmp (d, d2, mcnt)) |
| 5700 goto fail; | |
| 5701 d += mcnt, d2 += mcnt; | |
| 5702 | |
| 5703 /* Do this because we've match some characters. */ | |
| 5704 SET_REGS_MATCHED (); | |
| 5705 } | |
| 5706 } | |
| 5707 break; | |
| 5708 | |
| 5709 | |
| 5710 /* begline matches the empty string at the beginning of the string | |
| 5711 (unless `not_bol' is set in `bufp'), and, if | |
| 5712 `newline_anchor' is set, after newlines. */ | |
| 5713 case begline: | |
| 5041 | 5714 DEBUG_MATCH_PRINT1 ("EXECUTING begline.\n"); |
| 428 | 5715 |
| 5716 if (AT_STRINGS_BEG (d)) | |
| 5717 { | |
| 5718 if (!bufp->not_bol) break; | |
| 5719 } | |
| 826 | 5720 else |
| 5721 { | |
| 5722 re_char *d2 = d; | |
| 867 | 5723 DEC_IBYTEPTR (d2); |
| 5724 if (itext_ichar_ascii_fmt (d2, fmt, lispobj) == '\n' && | |
| 826 | 5725 bufp->newline_anchor) |
| 5726 break; | |
| 5727 } | |
| 428 | 5728 /* In all other cases, we fail. */ |
| 5729 goto fail; | |
| 5730 | |
| 5731 | |
| 5732 /* endline is the dual of begline. */ | |
| 5733 case endline: | |
| 5041 | 5734 DEBUG_MATCH_PRINT1 ("EXECUTING endline.\n"); |
| 428 | 5735 |
| 5736 if (AT_STRINGS_END (d)) | |
| 5737 { | |
| 5738 if (!bufp->not_eol) break; | |
| 5739 } | |
| 5740 | |
| 5741 /* We have to ``prefetch'' the next character. */ | |
| 826 | 5742 else if ((d == end1 ? |
| 867 | 5743 itext_ichar_ascii_fmt (string2, fmt, lispobj) : |
| 5744 itext_ichar_ascii_fmt (d, fmt, lispobj)) == '\n' | |
| 428 | 5745 && bufp->newline_anchor) |
| 5746 { | |
| 5747 break; | |
| 5748 } | |
| 5749 goto fail; | |
| 5750 | |
| 5751 | |
| 5752 /* Match at the very beginning of the data. */ | |
| 5753 case begbuf: | |
| 5041 | 5754 DEBUG_MATCH_PRINT1 ("EXECUTING begbuf.\n"); |
| 428 | 5755 if (AT_STRINGS_BEG (d)) |
| 5756 break; | |
| 5757 goto fail; | |
| 5758 | |
| 5759 | |
| 5760 /* Match at the very end of the data. */ | |
| 5761 case endbuf: | |
| 5041 | 5762 DEBUG_MATCH_PRINT1 ("EXECUTING endbuf.\n"); |
| 428 | 5763 if (AT_STRINGS_END (d)) |
| 5764 break; | |
| 5765 goto fail; | |
| 5766 | |
| 5767 | |
| 5768 /* on_failure_keep_string_jump is used to optimize `.*\n'. It | |
| 5769 pushes NULL as the value for the string on the stack. Then | |
| 5770 `pop_failure_point' will keep the current value for the | |
| 5771 string, instead of restoring it. To see why, consider | |
| 5772 matching `foo\nbar' against `.*\n'. The .* matches the foo; | |
| 5773 then the . fails against the \n. But the next thing we want | |
| 5774 to do is match the \n against the \n; if we restored the | |
| 5775 string value, we would be back at the foo. | |
| 5776 | |
| 5777 Because this is used only in specific cases, we don't need to | |
| 5778 check all the things that `on_failure_jump' does, to make | |
| 5779 sure the right things get saved on the stack. Hence we don't | |
| 5780 share its code. The only reason to push anything on the | |
| 5781 stack at all is that otherwise we would have to change | |
| 5782 `anychar's code to do something besides goto fail in this | |
| 5783 case; that seems worse than this. */ | |
| 5784 case on_failure_keep_string_jump: | |
| 5041 | 5785 DEBUG_MATCH_PRINT1 ("EXECUTING on_failure_keep_string_jump"); |
| 428 | 5786 |
| 5787 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
| 5041 | 5788 DEBUG_MATCH_PRINT3 (" %d (to 0x%lx):\n", mcnt, (long) (p + mcnt)); |
| 428 | 5789 |
| 446 | 5790 PUSH_FAILURE_POINT (p + mcnt, (unsigned char *) 0, -2); |
| 428 | 5791 break; |
| 5792 | |
| 5793 | |
| 5794 /* Uses of on_failure_jump: | |
| 5795 | |
| 5796 Each alternative starts with an on_failure_jump that points | |
| 5797 to the beginning of the next alternative. Each alternative | |
| 5798 except the last ends with a jump that in effect jumps past | |
| 5799 the rest of the alternatives. (They really jump to the | |
| 5800 ending jump of the following alternative, because tensioning | |
| 5801 these jumps is a hassle.) | |
| 5802 | |
| 5803 Repeats start with an on_failure_jump that points past both | |
| 5804 the repetition text and either the following jump or | |
| 5805 pop_failure_jump back to this on_failure_jump. */ | |
| 5806 case on_failure_jump: | |
| 5807 on_failure: | |
| 5041 | 5808 DEBUG_MATCH_PRINT1 ("EXECUTING on_failure_jump"); |
| 428 | 5809 |
| 5810 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
| 5041 | 5811 DEBUG_MATCH_PRINT3 (" %d (to 0x%lx)", mcnt, (long) (p + mcnt)); |
| 428 | 5812 |
| 5813 /* If this on_failure_jump comes right before a group (i.e., | |
| 5814 the original * applied to a group), save the information | |
| 5815 for that group and all inner ones, so that if we fail back | |
| 5816 to this point, the group's information will be correct. | |
| 5817 For example, in \(a*\)*\1, we need the preceding group, | |
| 5818 and in \(\(a*\)b*\)\2, we need the inner group. */ | |
| 5819 | |
| 5820 /* We can't use `p' to check ahead because we push | |
| 5821 a failure point to `p + mcnt' after we do this. */ | |
| 5822 p1 = p; | |
| 5823 | |
| 5824 /* We need to skip no_op's before we look for the | |
| 5825 start_memory in case this on_failure_jump is happening as | |
| 5826 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 | |
| 5827 against aba. */ | |
| 5828 while (p1 < pend && (re_opcode_t) *p1 == no_op) | |
| 5829 p1++; | |
| 5830 | |
| 5831 if (p1 < pend && (re_opcode_t) *p1 == start_memory) | |
| 5832 { | |
| 5833 /* We have a new highest active register now. This will | |
| 5834 get reset at the start_memory we are about to get to, | |
| 5835 but we will have saved all the registers relevant to | |
| 5836 this repetition op, as described above. */ | |
| 5837 highest_active_reg = *(p1 + 1) + *(p1 + 2); | |
| 5838 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) | |
| 5839 lowest_active_reg = *(p1 + 1); | |
| 5840 } | |
| 5841 | |
| 5041 | 5842 DEBUG_MATCH_PRINT1 (":\n"); |
| 428 | 5843 PUSH_FAILURE_POINT (p + mcnt, d, -2); |
| 5844 break; | |
| 5845 | |
| 5846 | |
| 5847 /* A smart repeat ends with `maybe_pop_jump'. | |
| 5848 We change it to either `pop_failure_jump' or `jump'. */ | |
| 5849 case maybe_pop_jump: | |
| 5850 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
| 5041 | 5851 DEBUG_MATCH_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); |
| 428 | 5852 { |
| 5853 REGISTER unsigned char *p2 = p; | |
| 5854 | |
| 5855 /* Compare the beginning of the repeat with what in the | |
| 5856 pattern follows its end. If we can establish that there | |
| 5857 is nothing that they would both match, i.e., that we | |
| 5858 would have to backtrack because of (as in, e.g., `a*a') | |
| 5859 then we can change to pop_failure_jump, because we'll | |
| 5860 never have to backtrack. | |
| 5861 | |
| 5862 This is not true in the case of alternatives: in | |
| 5863 `(a|ab)*' we do need to backtrack to the `ab' alternative | |
| 5864 (e.g., if the string was `ab'). But instead of trying to | |
| 5865 detect that here, the alternative has put on a dummy | |
| 5866 failure point which is what we will end up popping. */ | |
| 5867 | |
| 5868 /* Skip over open/close-group commands. | |
| 5869 If what follows this loop is a ...+ construct, | |
| 5870 look at what begins its body, since we will have to | |
| 5871 match at least one of that. */ | |
| 5872 while (1) | |
| 5873 { | |
| 5874 if (p2 + 2 < pend | |
| 5875 && ((re_opcode_t) *p2 == stop_memory | |
| 5876 || (re_opcode_t) *p2 == start_memory)) | |
| 5877 p2 += 3; | |
| 5878 else if (p2 + 6 < pend | |
| 5879 && (re_opcode_t) *p2 == dummy_failure_jump) | |
| 5880 p2 += 6; | |
| 5881 else | |
| 5882 break; | |
| 5883 } | |
| 5884 | |
| 5885 p1 = p + mcnt; | |
| 5886 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding | |
| 5887 to the `maybe_finalize_jump' of this case. Examine what | |
| 5888 follows. */ | |
| 5889 | |
| 5890 /* If we're at the end of the pattern, we can change. */ | |
| 5891 if (p2 == pend) | |
| 5892 { | |
| 5893 /* Consider what happens when matching ":\(.*\)" | |
| 5894 against ":/". I don't really understand this code | |
| 5895 yet. */ | |
| 5896 p[-3] = (unsigned char) pop_failure_jump; | |
| 5041 | 5897 DEBUG_MATCH_PRINT1 |
| 428 | 5898 (" End of pattern: change to `pop_failure_jump'.\n"); |
| 5899 } | |
| 5900 | |
| 5901 else if ((re_opcode_t) *p2 == exactn | |
| 5902 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) | |
| 5903 { | |
| 5904 REGISTER unsigned char c | |
| 5905 = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
| 5906 | |
| 5907 if ((re_opcode_t) p1[3] == exactn && p1[5] != c) | |
| 5908 { | |
| 5909 p[-3] = (unsigned char) pop_failure_jump; | |
| 5041 | 5910 DEBUG_MATCH_PRINT3 (" %c != %c => pop_failure_jump.\n", |
| 428 | 5911 c, p1[5]); |
| 5912 } | |
| 5913 | |
| 5914 else if ((re_opcode_t) p1[3] == charset | |
| 5915 || (re_opcode_t) p1[3] == charset_not) | |
| 5916 { | |
| 458 | 5917 int not_p = (re_opcode_t) p1[3] == charset_not; |
| 428 | 5918 |
| 5919 if (c < (unsigned char) (p1[4] * BYTEWIDTH) | |
| 5920 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) | |
| 458 | 5921 not_p = !not_p; |
| 5922 | |
| 5923 /* `not_p' is equal to 1 if c would match, which means | |
| 428 | 5924 that we can't change to pop_failure_jump. */ |
| 458 | 5925 if (!not_p) |
| 428 | 5926 { |
| 5927 p[-3] = (unsigned char) pop_failure_jump; | |
| 5041 | 5928 DEBUG_MATCH_PRINT1 (" No match => pop_failure_jump.\n"); |
| 428 | 5929 } |
| 5930 } | |
| 5931 } | |
| 5932 else if ((re_opcode_t) *p2 == charset) | |
| 5933 { | |
| 5934 #ifdef DEBUG | |
| 5935 REGISTER unsigned char c | |
| 5936 = *p2 == (unsigned char) endline ? '\n' : p2[2]; | |
| 5937 #endif | |
| 5938 | |
| 5939 if ((re_opcode_t) p1[3] == exactn | |
| 5940 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5] | |
| 5941 && (p2[2 + p1[5] / BYTEWIDTH] | |
| 5942 & (1 << (p1[5] % BYTEWIDTH))))) | |
| 5943 { | |
| 5944 p[-3] = (unsigned char) pop_failure_jump; | |
| 5041 | 5945 DEBUG_MATCH_PRINT3 (" %c != %c => pop_failure_jump.\n", |
| 428 | 5946 c, p1[5]); |
| 5947 } | |
| 5948 | |
| 5949 else if ((re_opcode_t) p1[3] == charset_not) | |
| 5950 { | |
| 5951 int idx; | |
| 5952 /* We win if the charset_not inside the loop | |
| 5953 lists every character listed in the charset after. */ | |
| 5954 for (idx = 0; idx < (int) p2[1]; idx++) | |
| 5955 if (! (p2[2 + idx] == 0 | |
| 5956 || (idx < (int) p1[4] | |
| 5957 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) | |
| 5958 break; | |
| 5959 | |
| 5960 if (idx == p2[1]) | |
| 5961 { | |
| 5962 p[-3] = (unsigned char) pop_failure_jump; | |
| 5041 | 5963 DEBUG_MATCH_PRINT1 (" No match => pop_failure_jump.\n"); |
| 428 | 5964 } |
| 5965 } | |
| 5966 else if ((re_opcode_t) p1[3] == charset) | |
| 5967 { | |
| 5968 int idx; | |
| 5969 /* We win if the charset inside the loop | |
| 5970 has no overlap with the one after the loop. */ | |
| 5971 for (idx = 0; | |
| 5972 idx < (int) p2[1] && idx < (int) p1[4]; | |
| 5973 idx++) | |
| 5974 if ((p2[2 + idx] & p1[5 + idx]) != 0) | |
| 5975 break; | |
| 5976 | |
| 5977 if (idx == p2[1] || idx == p1[4]) | |
| 5978 { | |
| 5979 p[-3] = (unsigned char) pop_failure_jump; | |
| 5041 | 5980 DEBUG_MATCH_PRINT1 (" No match => pop_failure_jump.\n"); |
| 428 | 5981 } |
| 5982 } | |
| 5983 } | |
| 5984 } | |
| 5985 p -= 2; /* Point at relative address again. */ | |
| 5986 if ((re_opcode_t) p[-1] != pop_failure_jump) | |
| 5987 { | |
| 5988 p[-1] = (unsigned char) jump; | |
| 5041 | 5989 DEBUG_MATCH_PRINT1 (" Match => jump.\n"); |
| 428 | 5990 goto unconditional_jump; |
| 5991 } | |
| 5992 /* Note fall through. */ | |
| 5993 | |
| 5994 | |
| 5995 /* The end of a simple repeat has a pop_failure_jump back to | |
| 5996 its matching on_failure_jump, where the latter will push a | |
| 5997 failure point. The pop_failure_jump takes off failure | |
| 5998 points put on by this pop_failure_jump's matching | |
| 5999 on_failure_jump; we got through the pattern to here from the | |
| 6000 matching on_failure_jump, so didn't fail. */ | |
| 6001 case pop_failure_jump: | |
| 6002 { | |
| 6003 /* We need to pass separate storage for the lowest and | |
| 6004 highest registers, even though we don't care about the | |
| 6005 actual values. Otherwise, we will restore only one | |
| 6006 register from the stack, since lowest will == highest in | |
| 6007 `pop_failure_point'. */ | |
| 647 | 6008 int dummy_low_reg, dummy_high_reg; |
| 428 | 6009 unsigned char *pdummy; |
| 446 | 6010 re_char *sdummy = NULL; |
| 428 | 6011 |
| 5041 | 6012 DEBUG_MATCH_PRINT1 ("EXECUTING pop_failure_jump.\n"); |
| 428 | 6013 POP_FAILURE_POINT (sdummy, pdummy, |
| 6014 dummy_low_reg, dummy_high_reg, | |
| 6015 reg_dummy, reg_dummy, reg_info_dummy); | |
| 6016 } | |
| 6017 /* Note fall through. */ | |
| 6018 | |
| 6019 | |
| 6020 /* Unconditionally jump (without popping any failure points). */ | |
| 6021 case jump: | |
| 6022 unconditional_jump: | |
| 6023 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ | |
| 5041 | 6024 DEBUG_MATCH_PRINT2 ("EXECUTING jump %d ", mcnt); |
| 428 | 6025 p += mcnt; /* Do the jump. */ |
| 5041 | 6026 DEBUG_MATCH_PRINT2 ("(to 0x%lx).\n", (long) p); |
| 428 | 6027 break; |
| 6028 | |
| 6029 | |
| 6030 /* We need this opcode so we can detect where alternatives end | |
| 6031 in `group_match_null_string_p' et al. */ | |
| 6032 case jump_past_alt: | |
| 5041 | 6033 DEBUG_MATCH_PRINT1 ("EXECUTING jump_past_alt.\n"); |
| 428 | 6034 goto unconditional_jump; |
| 6035 | |
| 6036 | |
| 6037 /* Normally, the on_failure_jump pushes a failure point, which | |
| 6038 then gets popped at pop_failure_jump. We will end up at | |
| 6039 pop_failure_jump, also, and with a pattern of, say, `a+', we | |
| 6040 are skipping over the on_failure_jump, so we have to push | |
| 6041 something meaningless for pop_failure_jump to pop. */ | |
| 6042 case dummy_failure_jump: | |
| 5041 | 6043 DEBUG_MATCH_PRINT1 ("EXECUTING dummy_failure_jump.\n"); |
| 428 | 6044 /* It doesn't matter what we push for the string here. What |
| 6045 the code at `fail' tests is the value for the pattern. */ | |
| 446 | 6046 PUSH_FAILURE_POINT ((unsigned char *) 0, (unsigned char *) 0, -2); |
| 428 | 6047 goto unconditional_jump; |
| 6048 | |
| 6049 | |
| 6050 /* At the end of an alternative, we need to push a dummy failure | |
| 6051 point in case we are followed by a `pop_failure_jump', because | |
| 6052 we don't want the failure point for the alternative to be | |
| 6053 popped. For example, matching `(a|ab)*' against `aab' | |
| 6054 requires that we match the `ab' alternative. */ | |
| 6055 case push_dummy_failure: | |
| 5041 | 6056 DEBUG_MATCH_PRINT1 ("EXECUTING push_dummy_failure.\n"); |
| 428 | 6057 /* See comments just above at `dummy_failure_jump' about the |
| 6058 two zeroes. */ | |
| 446 | 6059 PUSH_FAILURE_POINT ((unsigned char *) 0, (unsigned char *) 0, -2); |
| 428 | 6060 break; |
| 6061 | |
| 6062 /* Have to succeed matching what follows at least n times. | |
| 6063 After that, handle like `on_failure_jump'. */ | |
| 6064 case succeed_n: | |
| 6065 EXTRACT_NUMBER (mcnt, p + 2); | |
| 5041 | 6066 DEBUG_MATCH_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); |
| 428 | 6067 |
| 6068 assert (mcnt >= 0); | |
| 6069 /* Originally, this is how many times we HAVE to succeed. */ | |
| 6070 if (mcnt > 0) | |
| 6071 { | |
| 6072 mcnt--; | |
| 6073 p += 2; | |
| 6074 STORE_NUMBER_AND_INCR (p, mcnt); | |
| 5041 | 6075 DEBUG_MATCH_PRINT3 (" Setting 0x%lx to %d.\n", (long) p, mcnt); |
| 428 | 6076 } |
| 6077 else if (mcnt == 0) | |
| 6078 { | |
| 5041 | 6079 DEBUG_MATCH_PRINT2 (" Setting two bytes from 0x%lx to no_op.\n", |
| 428 | 6080 (long) (p+2)); |
| 6081 p[2] = (unsigned char) no_op; | |
| 6082 p[3] = (unsigned char) no_op; | |
| 6083 goto on_failure; | |
| 6084 } | |
| 6085 break; | |
| 6086 | |
| 6087 case jump_n: | |
| 6088 EXTRACT_NUMBER (mcnt, p + 2); | |
| 5041 | 6089 DEBUG_MATCH_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); |
| 428 | 6090 |
| 6091 /* Originally, this is how many times we CAN jump. */ | |
| 6092 if (mcnt) | |
| 6093 { | |
| 6094 mcnt--; | |
| 6095 STORE_NUMBER (p + 2, mcnt); | |
| 6096 goto unconditional_jump; | |
| 6097 } | |
| 6098 /* If don't have to jump any more, skip over the rest of command. */ | |
| 6099 else | |
| 6100 p += 4; | |
| 6101 break; | |
| 6102 | |
| 6103 case set_number_at: | |
| 6104 { | |
| 5041 | 6105 DEBUG_MATCH_PRINT1 ("EXECUTING set_number_at.\n"); |
| 428 | 6106 |
| 6107 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
| 6108 p1 = p + mcnt; | |
| 6109 EXTRACT_NUMBER_AND_INCR (mcnt, p); | |
| 5041 | 6110 DEBUG_MATCH_PRINT3 (" Setting 0x%lx to %d.\n", (long) p1, mcnt); |
| 428 | 6111 STORE_NUMBER (p1, mcnt); |
| 6112 break; | |
| 6113 } | |
| 6114 | |
| 6115 case wordbound: | |
| 5041 | 6116 DEBUG_MATCH_PRINT1 ("EXECUTING wordbound.\n"); |
| 428 | 6117 should_succeed = 1; |
| 6118 matchwordbound: | |
| 6119 { | |
| 6120 /* XEmacs change */ | |
| 1377 | 6121 /* Straightforward and (I hope) correct implementation. |
| 6122 Probably should be optimized by arranging to compute | |
| 1497 | 6123 charpos only once. */ |
| 1377 | 6124 /* emch1 is the character before d, syn1 is the syntax of |
| 6125 emch1, emch2 is the character at d, and syn2 is the | |
| 6126 syntax of emch2. */ | |
| 6127 Ichar emch1, emch2; | |
| 1468 | 6128 int syn1 = 0, |
| 6129 syn2 = 0; | |
| 1377 | 6130 re_char *d_before, *d_after; |
| 6131 int result, | |
| 6132 at_beg = AT_STRINGS_BEG (d), | |
| 6133 at_end = AT_STRINGS_END (d); | |
| 6134 #ifdef emacs | |
| 1497 | 6135 Charxpos charpos; |
| 1377 | 6136 #endif |
| 6137 | |
| 6138 if (at_beg && at_end) | |
| 6139 { | |
| 6140 result = 0; | |
| 6141 } | |
| 428 | 6142 else |
| 6143 { | |
| 1377 | 6144 if (!at_beg) |
| 6145 { | |
| 6146 d_before = POS_BEFORE_GAP_UNSAFE (d); | |
| 6147 DEC_IBYTEPTR_FMT (d_before, fmt); | |
| 6148 emch1 = itext_ichar_fmt (d_before, fmt, lispobj); | |
| 460 | 6149 #ifdef emacs |
| 1497 | 6150 charpos = offset_to_charxpos (lispobj, |
| 6151 PTR_TO_OFFSET (d)) - 1; | |
| 1377 | 6152 BEGIN_REGEX_MALLOC_OK (); |
| 1497 | 6153 UPDATE_SYNTAX_CACHE (scache, charpos); |
| 460 | 6154 #endif |
| 1377 | 6155 syn1 = SYNTAX_FROM_CACHE (scache, emch1); |
| 6156 END_REGEX_MALLOC_OK (); | |
| 6157 } | |
| 6158 if (!at_end) | |
| 6159 { | |
| 6160 d_after = POS_AFTER_GAP_UNSAFE (d); | |
| 6161 emch2 = itext_ichar_fmt (d_after, fmt, lispobj); | |
| 460 | 6162 #ifdef emacs |
| 1497 | 6163 charpos = offset_to_charxpos (lispobj, PTR_TO_OFFSET (d)); |
| 1377 | 6164 BEGIN_REGEX_MALLOC_OK (); |
| 1497 | 6165 UPDATE_SYNTAX_CACHE_FORWARD (scache, charpos); |
| 460 | 6166 #endif |
| 1377 | 6167 syn2 = SYNTAX_FROM_CACHE (scache, emch2); |
| 6168 END_REGEX_MALLOC_OK (); | |
| 6169 } | |
| 1333 | 6170 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); |
| 1377 | 6171 |
| 6172 if (at_beg) | |
| 6173 result = (syn2 == Sword); | |
| 6174 else if (at_end) | |
| 6175 result = (syn1 == Sword); | |
| 6176 else | |
| 6177 result = ((syn1 == Sword) != (syn2 == Sword)); | |
| 428 | 6178 } |
| 1377 | 6179 |
| 428 | 6180 if (result == should_succeed) |
| 6181 break; | |
| 6182 goto fail; | |
| 6183 } | |
| 6184 | |
| 6185 case notwordbound: | |
| 5041 | 6186 DEBUG_MATCH_PRINT1 ("EXECUTING notwordbound.\n"); |
| 428 | 6187 should_succeed = 0; |
| 6188 goto matchwordbound; | |
| 6189 | |
| 6190 case wordbeg: | |
| 5041 | 6191 DEBUG_MATCH_PRINT1 ("EXECUTING wordbeg.\n"); |
| 460 | 6192 if (AT_STRINGS_END (d)) |
| 6193 goto fail; | |
| 428 | 6194 { |
| 6195 /* XEmacs: this originally read: | |
| 6196 | |
| 6197 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) | |
| 6198 break; | |
| 6199 | |
| 6200 */ | |
| 460 | 6201 re_char *dtmp = POS_AFTER_GAP_UNSAFE (d); |
| 867 | 6202 Ichar emch = itext_ichar_fmt (dtmp, fmt, lispobj); |
| 1333 | 6203 int tempres; |
| 1347 | 6204 #ifdef emacs |
| 6205 Charxpos charpos = offset_to_charxpos (lispobj, PTR_TO_OFFSET (d)); | |
| 6206 #endif | |
| 1333 | 6207 BEGIN_REGEX_MALLOC_OK (); |
| 460 | 6208 #ifdef emacs |
| 826 | 6209 UPDATE_SYNTAX_CACHE (scache, charpos); |
| 460 | 6210 #endif |
| 1333 | 6211 tempres = (SYNTAX_FROM_CACHE (scache, emch) != Sword); |
| 6212 END_REGEX_MALLOC_OK (); | |
| 6213 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 6214 if (tempres) | |
| 428 | 6215 goto fail; |
| 6216 if (AT_STRINGS_BEG (d)) | |
| 6217 break; | |
| 460 | 6218 dtmp = POS_BEFORE_GAP_UNSAFE (d); |
| 867 | 6219 DEC_IBYTEPTR_FMT (dtmp, fmt); |
| 6220 emch = itext_ichar_fmt (dtmp, fmt, lispobj); | |
| 1333 | 6221 BEGIN_REGEX_MALLOC_OK (); |
| 460 | 6222 #ifdef emacs |
| 826 | 6223 UPDATE_SYNTAX_CACHE_BACKWARD (scache, charpos - 1); |
| 460 | 6224 #endif |
| 1333 | 6225 tempres = (SYNTAX_FROM_CACHE (scache, emch) != Sword); |
| 6226 END_REGEX_MALLOC_OK (); | |
| 6227 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 6228 if (tempres) | |
| 428 | 6229 break; |
| 6230 goto fail; | |
| 6231 } | |
| 6232 | |
| 6233 case wordend: | |
| 5041 | 6234 DEBUG_MATCH_PRINT1 ("EXECUTING wordend.\n"); |
| 460 | 6235 if (AT_STRINGS_BEG (d)) |
| 6236 goto fail; | |
| 428 | 6237 { |
| 6238 /* XEmacs: this originally read: | |
| 6239 | |
| 6240 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) | |
| 6241 && (!WORDCHAR_P (d) || AT_STRINGS_END (d))) | |
| 6242 break; | |
| 6243 | |
| 6244 The or condition is incorrect (reversed). | |
| 6245 */ | |
| 460 | 6246 re_char *dtmp; |
| 867 | 6247 Ichar emch; |
| 1333 | 6248 int tempres; |
| 460 | 6249 #ifdef emacs |
| 826 | 6250 Charxpos charpos = offset_to_charxpos (lispobj, PTR_TO_OFFSET (d)); |
| 1347 | 6251 BEGIN_REGEX_MALLOC_OK (); |
| 826 | 6252 UPDATE_SYNTAX_CACHE (scache, charpos); |
| 1333 | 6253 END_REGEX_MALLOC_OK (); |
| 6254 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 1347 | 6255 #endif |
| 460 | 6256 dtmp = POS_BEFORE_GAP_UNSAFE (d); |
| 867 | 6257 DEC_IBYTEPTR_FMT (dtmp, fmt); |
| 6258 emch = itext_ichar_fmt (dtmp, fmt, lispobj); | |
| 1333 | 6259 BEGIN_REGEX_MALLOC_OK (); |
| 6260 tempres = (SYNTAX_FROM_CACHE (scache, emch) != Sword); | |
| 6261 END_REGEX_MALLOC_OK (); | |
| 6262 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 6263 if (tempres) | |
| 428 | 6264 goto fail; |
| 6265 if (AT_STRINGS_END (d)) | |
| 6266 break; | |
| 460 | 6267 dtmp = POS_AFTER_GAP_UNSAFE (d); |
| 867 | 6268 emch = itext_ichar_fmt (dtmp, fmt, lispobj); |
| 1333 | 6269 BEGIN_REGEX_MALLOC_OK (); |
| 460 | 6270 #ifdef emacs |
| 826 | 6271 UPDATE_SYNTAX_CACHE_FORWARD (scache, charpos + 1); |
| 460 | 6272 #endif |
| 1333 | 6273 tempres = (SYNTAX_FROM_CACHE (scache, emch) != Sword); |
| 6274 END_REGEX_MALLOC_OK (); | |
| 6275 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 6276 if (tempres) | |
| 428 | 6277 break; |
| 6278 goto fail; | |
| 6279 } | |
| 6280 | |
| 6281 #ifdef emacs | |
| 6282 case before_dot: | |
| 5041 | 6283 DEBUG_MATCH_PRINT1 ("EXECUTING before_dot.\n"); |
| 826 | 6284 if (!BUFFERP (lispobj) |
| 6285 || (BUF_PTR_BYTE_POS (XBUFFER (lispobj), (unsigned char *) d) | |
| 6286 >= BUF_PT (XBUFFER (lispobj)))) | |
| 428 | 6287 goto fail; |
| 6288 break; | |
| 6289 | |
| 6290 case at_dot: | |
| 5041 | 6291 DEBUG_MATCH_PRINT1 ("EXECUTING at_dot.\n"); |
| 826 | 6292 if (!BUFFERP (lispobj) |
| 6293 || (BUF_PTR_BYTE_POS (XBUFFER (lispobj), (unsigned char *) d) | |
| 6294 != BUF_PT (XBUFFER (lispobj)))) | |
| 428 | 6295 goto fail; |
| 6296 break; | |
| 6297 | |
| 6298 case after_dot: | |
| 5041 | 6299 DEBUG_MATCH_PRINT1 ("EXECUTING after_dot.\n"); |
| 826 | 6300 if (!BUFFERP (lispobj) |
| 6301 || (BUF_PTR_BYTE_POS (XBUFFER (lispobj), (unsigned char *) d) | |
| 6302 <= BUF_PT (XBUFFER (lispobj)))) | |
| 428 | 6303 goto fail; |
| 6304 break; | |
| 6305 | |
| 6306 case syntaxspec: | |
| 5041 | 6307 DEBUG_MATCH_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); |
| 428 | 6308 mcnt = *p++; |
| 6309 goto matchsyntax; | |
| 6310 | |
| 6311 case wordchar: | |
| 5041 | 6312 DEBUG_MATCH_PRINT1 ("EXECUTING Emacs wordchar.\n"); |
| 428 | 6313 mcnt = (int) Sword; |
| 6314 matchsyntax: | |
| 6315 should_succeed = 1; | |
| 6316 matchornotsyntax: | |
| 6317 { | |
| 6318 int matches; | |
| 867 | 6319 Ichar emch; |
| 428 | 6320 |
| 450 | 6321 REGEX_PREFETCH (); |
| 1333 | 6322 BEGIN_REGEX_MALLOC_OK (); |
| 826 | 6323 UPDATE_SYNTAX_CACHE |
| 6324 (scache, offset_to_charxpos (lispobj, PTR_TO_OFFSET (d))); | |
| 1333 | 6325 END_REGEX_MALLOC_OK (); |
| 6326 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 826 | 6327 |
| 867 | 6328 emch = itext_ichar_fmt (d, fmt, lispobj); |
| 1333 | 6329 BEGIN_REGEX_MALLOC_OK (); |
| 826 | 6330 matches = (SYNTAX_FROM_CACHE (scache, emch) == |
| 6331 (enum syntaxcode) mcnt); | |
| 1333 | 6332 END_REGEX_MALLOC_OK (); |
| 6333 RE_MATCH_RELOCATE_MOVEABLE_DATA_POINTERS (); | |
| 867 | 6334 INC_IBYTEPTR_FMT (d, fmt); |
| 428 | 6335 if (matches != should_succeed) |
| 6336 goto fail; | |
| 6337 SET_REGS_MATCHED (); | |
| 6338 } | |
| 6339 break; | |
| 6340 | |
| 6341 case notsyntaxspec: | |
| 5041 | 6342 DEBUG_MATCH_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); |
| 428 | 6343 mcnt = *p++; |
| 6344 goto matchnotsyntax; | |
| 6345 | |
| 6346 case notwordchar: | |
| 5041 | 6347 DEBUG_MATCH_PRINT1 ("EXECUTING Emacs notwordchar.\n"); |
| 428 | 6348 mcnt = (int) Sword; |
| 6349 matchnotsyntax: | |
| 6350 should_succeed = 0; | |
| 6351 goto matchornotsyntax; | |
| 6352 | |
| 6353 #ifdef MULE | |
| 6354 /* 97/2/17 jhod Mule category code patch */ | |
| 6355 case categoryspec: | |
| 6356 should_succeed = 1; | |
| 6357 matchornotcategory: | |
| 6358 { | |
| 867 | 6359 Ichar emch; |
| 428 | 6360 |
| 6361 mcnt = *p++; | |
| 450 | 6362 REGEX_PREFETCH (); |
| 867 | 6363 emch = itext_ichar_fmt (d, fmt, lispobj); |
| 6364 INC_IBYTEPTR_FMT (d, fmt); | |
| 826 | 6365 if (check_category_char (emch, BUFFER_CATEGORY_TABLE (lispbuf), |
| 6366 mcnt, should_succeed)) | |
| 428 | 6367 goto fail; |
| 6368 SET_REGS_MATCHED (); | |
| 6369 } | |
| 6370 break; | |
| 6371 | |
| 6372 case notcategoryspec: | |
| 6373 should_succeed = 0; | |
| 6374 goto matchornotcategory; | |
| 6375 /* end of category patch */ | |
| 6376 #endif /* MULE */ | |
| 6377 #else /* not emacs */ | |
| 6378 case wordchar: | |
| 5041 | 6379 DEBUG_MATCH_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); |
| 450 | 6380 REGEX_PREFETCH (); |
| 826 | 6381 if (!WORDCHAR_P ((int) (*d))) |
| 428 | 6382 goto fail; |
| 6383 SET_REGS_MATCHED (); | |
| 6384 d++; | |
| 6385 break; | |
| 6386 | |
| 6387 case notwordchar: | |
| 5041 | 6388 DEBUG_MATCH_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); |
| 450 | 6389 REGEX_PREFETCH (); |
| 826 | 6390 if (!WORDCHAR_P ((int) (*d))) |
| 428 | 6391 goto fail; |
| 6392 SET_REGS_MATCHED (); | |
| 6393 d++; | |
| 6394 break; | |
| 446 | 6395 #endif /* emacs */ |
| 428 | 6396 |
| 6397 default: | |
| 2500 | 6398 ABORT (); |
| 428 | 6399 } |
| 6400 continue; /* Successfully executed one pattern command; keep going. */ | |
| 6401 | |
| 6402 | |
| 6403 /* We goto here if a matching operation fails. */ | |
| 6404 fail: | |
| 6405 if (!FAIL_STACK_EMPTY ()) | |
| 6406 { /* A restart point is known. Restore to that state. */ | |
| 5041 | 6407 DEBUG_MATCH_PRINT1 ("\nFAIL:\n"); |
| 428 | 6408 POP_FAILURE_POINT (d, p, |
| 6409 lowest_active_reg, highest_active_reg, | |
| 6410 regstart, regend, reg_info); | |
| 6411 | |
| 6412 /* If this failure point is a dummy, try the next one. */ | |
| 6413 if (!p) | |
| 6414 goto fail; | |
| 6415 | |
| 6416 /* If we failed to the end of the pattern, don't examine *p. */ | |
| 6417 assert (p <= pend); | |
| 6418 if (p < pend) | |
| 6419 { | |
| 460 | 6420 re_bool is_a_jump_n = false; |
| 428 | 6421 |
| 6422 /* If failed to a backwards jump that's part of a repetition | |
| 6423 loop, need to pop this failure point and use the next one. */ | |
| 6424 switch ((re_opcode_t) *p) | |
| 6425 { | |
| 6426 case jump_n: | |
| 6427 is_a_jump_n = true; | |
| 6428 case maybe_pop_jump: | |
| 6429 case pop_failure_jump: | |
| 6430 case jump: | |
| 6431 p1 = p + 1; | |
| 6432 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
| 6433 p1 += mcnt; | |
| 6434 | |
| 6435 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) | |
| 6436 || (!is_a_jump_n | |
| 6437 && (re_opcode_t) *p1 == on_failure_jump)) | |
| 6438 goto fail; | |
| 6439 break; | |
| 6440 default: | |
| 6441 /* do nothing */ ; | |
| 6442 } | |
| 6443 } | |
| 6444 | |
| 6445 if (d >= string1 && d <= end1) | |
| 6446 dend = end_match_1; | |
| 6447 } | |
| 6448 else | |
| 6449 break; /* Matching at this starting point really fails. */ | |
| 6450 } /* for (;;) */ | |
| 6451 | |
| 6452 if (best_regs_set) | |
| 6453 goto restore_best_regs; | |
| 6454 | |
| 6455 FREE_VARIABLES (); | |
| 6456 | |
| 6457 return -1; /* Failure to match. */ | |
| 1333 | 6458 } /* re_match_2_internal */ |
| 428 | 6459 |
| 6460 /* Subroutine definitions for re_match_2. */ | |
| 6461 | |
| 6462 | |
| 6463 /* We are passed P pointing to a register number after a start_memory. | |
| 6464 | |
| 6465 Return true if the pattern up to the corresponding stop_memory can | |
| 6466 match the empty string, and false otherwise. | |
| 6467 | |
| 6468 If we find the matching stop_memory, sets P to point to one past its number. | |
| 6469 Otherwise, sets P to an undefined byte less than or equal to END. | |
| 6470 | |
| 6471 We don't handle duplicates properly (yet). */ | |
| 6472 | |
| 460 | 6473 static re_bool |
| 428 | 6474 group_match_null_string_p (unsigned char **p, unsigned char *end, |
| 6475 register_info_type *reg_info) | |
| 6476 { | |
| 6477 int mcnt; | |
| 6478 /* Point to after the args to the start_memory. */ | |
| 6479 unsigned char *p1 = *p + 2; | |
| 6480 | |
| 6481 while (p1 < end) | |
| 6482 { | |
| 6483 /* Skip over opcodes that can match nothing, and return true or | |
| 6484 false, as appropriate, when we get to one that can't, or to the | |
| 6485 matching stop_memory. */ | |
| 6486 | |
| 6487 switch ((re_opcode_t) *p1) | |
| 6488 { | |
| 6489 /* Could be either a loop or a series of alternatives. */ | |
| 6490 case on_failure_jump: | |
| 6491 p1++; | |
| 6492 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
| 6493 | |
| 6494 /* If the next operation is not a jump backwards in the | |
| 6495 pattern. */ | |
| 6496 | |
| 6497 if (mcnt >= 0) | |
| 6498 { | |
| 6499 /* Go through the on_failure_jumps of the alternatives, | |
| 6500 seeing if any of the alternatives cannot match nothing. | |
| 6501 The last alternative starts with only a jump, | |
| 6502 whereas the rest start with on_failure_jump and end | |
| 6503 with a jump, e.g., here is the pattern for `a|b|c': | |
| 6504 | |
| 6505 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 | |
| 6506 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 | |
| 6507 /exactn/1/c | |
| 6508 | |
| 6509 So, we have to first go through the first (n-1) | |
| 6510 alternatives and then deal with the last one separately. */ | |
| 6511 | |
| 6512 | |
| 6513 /* Deal with the first (n-1) alternatives, which start | |
| 6514 with an on_failure_jump (see above) that jumps to right | |
| 6515 past a jump_past_alt. */ | |
| 6516 | |
| 6517 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) | |
| 6518 { | |
| 6519 /* `mcnt' holds how many bytes long the alternative | |
| 6520 is, including the ending `jump_past_alt' and | |
| 6521 its number. */ | |
| 6522 | |
| 6523 if (!alt_match_null_string_p (p1, p1 + mcnt - 3, | |
| 6524 reg_info)) | |
| 6525 return false; | |
| 6526 | |
| 6527 /* Move to right after this alternative, including the | |
| 6528 jump_past_alt. */ | |
| 6529 p1 += mcnt; | |
| 6530 | |
| 6531 /* Break if it's the beginning of an n-th alternative | |
| 6532 that doesn't begin with an on_failure_jump. */ | |
| 6533 if ((re_opcode_t) *p1 != on_failure_jump) | |
| 6534 break; | |
| 6535 | |
| 6536 /* Still have to check that it's not an n-th | |
| 6537 alternative that starts with an on_failure_jump. */ | |
| 6538 p1++; | |
| 6539 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
| 6540 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) | |
| 6541 { | |
| 6542 /* Get to the beginning of the n-th alternative. */ | |
| 6543 p1 -= 3; | |
| 6544 break; | |
| 6545 } | |
| 6546 } | |
| 6547 | |
| 6548 /* Deal with the last alternative: go back and get number | |
| 6549 of the `jump_past_alt' just before it. `mcnt' contains | |
| 6550 the length of the alternative. */ | |
| 6551 EXTRACT_NUMBER (mcnt, p1 - 2); | |
| 6552 | |
| 6553 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) | |
| 6554 return false; | |
| 6555 | |
| 6556 p1 += mcnt; /* Get past the n-th alternative. */ | |
| 6557 } /* if mcnt > 0 */ | |
| 6558 break; | |
| 6559 | |
| 6560 | |
| 6561 case stop_memory: | |
| 6562 assert (p1[1] == **p); | |
| 6563 *p = p1 + 2; | |
| 6564 return true; | |
| 6565 | |
| 6566 | |
| 6567 default: | |
| 6568 if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
| 6569 return false; | |
| 6570 } | |
| 6571 } /* while p1 < end */ | |
| 6572 | |
| 6573 return false; | |
| 6574 } /* group_match_null_string_p */ | |
| 6575 | |
| 6576 | |
| 6577 /* Similar to group_match_null_string_p, but doesn't deal with alternatives: | |
| 6578 It expects P to be the first byte of a single alternative and END one | |
| 6579 byte past the last. The alternative can contain groups. */ | |
| 6580 | |
| 460 | 6581 static re_bool |
| 428 | 6582 alt_match_null_string_p (unsigned char *p, unsigned char *end, |
| 6583 register_info_type *reg_info) | |
| 6584 { | |
| 6585 int mcnt; | |
| 6586 unsigned char *p1 = p; | |
| 6587 | |
| 6588 while (p1 < end) | |
| 6589 { | |
| 6590 /* Skip over opcodes that can match nothing, and break when we get | |
| 6591 to one that can't. */ | |
| 6592 | |
| 6593 switch ((re_opcode_t) *p1) | |
| 6594 { | |
| 6595 /* It's a loop. */ | |
| 6596 case on_failure_jump: | |
| 6597 p1++; | |
| 6598 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
| 6599 p1 += mcnt; | |
| 6600 break; | |
| 6601 | |
| 6602 default: | |
| 6603 if (!common_op_match_null_string_p (&p1, end, reg_info)) | |
| 6604 return false; | |
| 6605 } | |
| 6606 } /* while p1 < end */ | |
| 6607 | |
| 6608 return true; | |
| 6609 } /* alt_match_null_string_p */ | |
| 6610 | |
| 6611 | |
| 6612 /* Deals with the ops common to group_match_null_string_p and | |
| 6613 alt_match_null_string_p. | |
| 6614 | |
| 6615 Sets P to one after the op and its arguments, if any. */ | |
| 6616 | |
| 460 | 6617 static re_bool |
| 428 | 6618 common_op_match_null_string_p (unsigned char **p, unsigned char *end, |
| 6619 register_info_type *reg_info) | |
| 6620 { | |
| 6621 int mcnt; | |
| 460 | 6622 re_bool ret; |
| 428 | 6623 int reg_no; |
| 6624 unsigned char *p1 = *p; | |
| 6625 | |
| 6626 switch ((re_opcode_t) *p1++) | |
| 6627 { | |
| 6628 case no_op: | |
| 6629 case begline: | |
| 6630 case endline: | |
| 6631 case begbuf: | |
| 6632 case endbuf: | |
| 6633 case wordbeg: | |
| 6634 case wordend: | |
| 6635 case wordbound: | |
| 6636 case notwordbound: | |
| 6637 #ifdef emacs | |
| 6638 case before_dot: | |
| 6639 case at_dot: | |
| 6640 case after_dot: | |
| 6641 #endif | |
| 6642 break; | |
| 6643 | |
| 6644 case start_memory: | |
| 6645 reg_no = *p1; | |
| 6646 assert (reg_no > 0 && reg_no <= MAX_REGNUM); | |
| 6647 ret = group_match_null_string_p (&p1, end, reg_info); | |
| 6648 | |
| 6649 /* Have to set this here in case we're checking a group which | |
| 6650 contains a group and a back reference to it. */ | |
| 6651 | |
| 6652 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) | |
| 6653 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; | |
| 6654 | |
| 6655 if (!ret) | |
| 6656 return false; | |
| 6657 break; | |
| 6658 | |
| 6659 /* If this is an optimized succeed_n for zero times, make the jump. */ | |
| 6660 case jump: | |
| 6661 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
| 6662 if (mcnt >= 0) | |
| 6663 p1 += mcnt; | |
| 6664 else | |
| 6665 return false; | |
| 6666 break; | |
| 6667 | |
| 6668 case succeed_n: | |
| 6669 /* Get to the number of times to succeed. */ | |
| 6670 p1 += 2; | |
| 6671 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
| 6672 | |
| 6673 if (mcnt == 0) | |
| 6674 { | |
| 6675 p1 -= 4; | |
| 6676 EXTRACT_NUMBER_AND_INCR (mcnt, p1); | |
| 6677 p1 += mcnt; | |
| 6678 } | |
| 6679 else | |
| 6680 return false; | |
| 6681 break; | |
| 6682 | |
| 6683 case duplicate: | |
| 6684 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) | |
| 6685 return false; | |
| 6686 break; | |
| 6687 | |
| 6688 case set_number_at: | |
| 6689 p1 += 4; | |
| 6690 | |
| 6691 default: | |
| 6692 /* All other opcodes mean we cannot match the empty string. */ | |
| 6693 return false; | |
| 6694 } | |
| 6695 | |
| 6696 *p = p1; | |
| 6697 return true; | |
| 6698 } /* common_op_match_null_string_p */ | |
| 6699 | |
| 6700 | |
| 6701 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN | |
| 6702 bytes; nonzero otherwise. */ | |
| 6703 | |
| 6704 static int | |
| 446 | 6705 bcmp_translate (re_char *s1, re_char *s2, |
| 826 | 6706 REGISTER int len, RE_TRANSLATE_TYPE translate |
| 6707 #ifdef emacs | |
| 2333 | 6708 , Internal_Format USED_IF_MULE (fmt), |
| 6709 Lisp_Object USED_IF_MULE (lispobj) | |
| 826 | 6710 #endif |
| 6711 ) | |
| 428 | 6712 { |
| 826 | 6713 REGISTER re_char *p1 = s1, *p2 = s2; |
| 446 | 6714 #ifdef MULE |
| 826 | 6715 re_char *p1_end = s1 + len; |
| 6716 re_char *p2_end = s2 + len; | |
| 446 | 6717 |
| 6718 while (p1 != p1_end && p2 != p2_end) | |
| 6719 { | |
| 867 | 6720 Ichar p1_ch, p2_ch; |
| 6721 | |
| 6722 p1_ch = itext_ichar_fmt (p1, fmt, lispobj); | |
| 6723 p2_ch = itext_ichar_fmt (p2, fmt, lispobj); | |
| 826 | 6724 |
| 6725 if (RE_TRANSLATE_1 (p1_ch) | |
| 6726 != RE_TRANSLATE_1 (p2_ch)) | |
| 446 | 6727 return 1; |
| 867 | 6728 INC_IBYTEPTR_FMT (p1, fmt); |
| 6729 INC_IBYTEPTR_FMT (p2, fmt); | |
| 446 | 6730 } |
| 6731 #else /* not MULE */ | |
| 428 | 6732 while (len) |
| 6733 { | |
| 826 | 6734 if (RE_TRANSLATE_1 (*p1++) != RE_TRANSLATE_1 (*p2++)) return 1; |
| 428 | 6735 len--; |
| 6736 } | |
| 446 | 6737 #endif /* MULE */ |
| 428 | 6738 return 0; |
| 6739 } | |
| 6740 | |
| 6741 /* Entry points for GNU code. */ | |
| 6742 | |
| 6743 /* re_compile_pattern is the GNU regular expression compiler: it | |
| 6744 compiles PATTERN (of length SIZE) and puts the result in BUFP. | |
| 6745 Returns 0 if the pattern was valid, otherwise an error string. | |
| 6746 | |
| 6747 Assumes the `allocated' (and perhaps `buffer') and `translate' fields | |
| 6748 are set in BUFP on entry. | |
| 6749 | |
| 6750 We call regex_compile to do the actual compilation. */ | |
| 6751 | |
| 442 | 6752 const char * |
| 6753 re_compile_pattern (const char *pattern, int length, | |
| 428 | 6754 struct re_pattern_buffer *bufp) |
| 6755 { | |
| 6756 reg_errcode_t ret; | |
| 6757 | |
| 6758 /* GNU code is written to assume at least RE_NREGS registers will be set | |
| 6759 (and at least one extra will be -1). */ | |
| 6760 bufp->regs_allocated = REGS_UNALLOCATED; | |
| 6761 | |
| 6762 /* And GNU code determines whether or not to get register information | |
| 6763 by passing null for the REGS argument to re_match, etc., not by | |
| 6764 setting no_sub. */ | |
| 6765 bufp->no_sub = 0; | |
| 6766 | |
| 6767 /* Match anchors at newline. */ | |
| 6768 bufp->newline_anchor = 1; | |
| 6769 | |
| 826 | 6770 ret = regex_compile ((unsigned char *) pattern, length, re_syntax_options, |
| 6771 bufp); | |
| 428 | 6772 |
| 6773 if (!ret) | |
| 6774 return NULL; | |
| 6775 return gettext (re_error_msgid[(int) ret]); | |
| 6776 } | |
| 6777 | |
| 6778 /* Entry points compatible with 4.2 BSD regex library. We don't define | |
| 6779 them unless specifically requested. */ | |
| 6780 | |
| 6781 #ifdef _REGEX_RE_COMP | |
| 6782 | |
| 6783 /* BSD has one and only one pattern buffer. */ | |
| 6784 static struct re_pattern_buffer re_comp_buf; | |
| 6785 | |
| 6786 char * | |
| 442 | 6787 re_comp (const char *s) |
| 428 | 6788 { |
| 6789 reg_errcode_t ret; | |
| 6790 | |
| 6791 if (!s) | |
| 6792 { | |
| 6793 if (!re_comp_buf.buffer) | |
| 6794 return gettext ("No previous regular expression"); | |
| 6795 return 0; | |
| 6796 } | |
| 6797 | |
| 6798 if (!re_comp_buf.buffer) | |
| 6799 { | |
| 1333 | 6800 re_comp_buf.buffer = (unsigned char *) xmalloc (200); |
| 428 | 6801 if (re_comp_buf.buffer == NULL) |
| 6802 return gettext (re_error_msgid[(int) REG_ESPACE]); | |
| 6803 re_comp_buf.allocated = 200; | |
| 6804 | |
| 1333 | 6805 re_comp_buf.fastmap = (char *) xmalloc (1 << BYTEWIDTH); |
| 428 | 6806 if (re_comp_buf.fastmap == NULL) |
| 6807 return gettext (re_error_msgid[(int) REG_ESPACE]); | |
| 6808 } | |
| 6809 | |
| 6810 /* Since `re_exec' always passes NULL for the `regs' argument, we | |
| 6811 don't need to initialize the pattern buffer fields which affect it. */ | |
| 6812 | |
| 6813 /* Match anchors at newlines. */ | |
| 6814 re_comp_buf.newline_anchor = 1; | |
| 6815 | |
| 826 | 6816 ret = regex_compile ((unsigned char *)s, strlen (s), re_syntax_options, |
| 6817 &re_comp_buf); | |
| 428 | 6818 |
| 6819 if (!ret) | |
| 6820 return NULL; | |
| 6821 | |
| 442 | 6822 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
| 428 | 6823 return (char *) gettext (re_error_msgid[(int) ret]); |
| 6824 } | |
| 6825 | |
| 6826 | |
| 6827 int | |
| 442 | 6828 re_exec (const char *s) |
| 428 | 6829 { |
| 442 | 6830 const int len = strlen (s); |
| 428 | 6831 return |
| 6832 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); | |
| 6833 } | |
| 6834 #endif /* _REGEX_RE_COMP */ | |
| 6835 | |
| 6836 /* POSIX.2 functions. Don't define these for Emacs. */ | |
| 6837 | |
| 6838 #ifndef emacs | |
| 6839 | |
| 6840 /* regcomp takes a regular expression as a string and compiles it. | |
| 6841 | |
| 6842 PREG is a regex_t *. We do not expect any fields to be initialized, | |
| 6843 since POSIX says we shouldn't. Thus, we set | |
| 6844 | |
| 6845 `buffer' to the compiled pattern; | |
| 6846 `used' to the length of the compiled pattern; | |
| 6847 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the | |
| 6848 REG_EXTENDED bit in CFLAGS is set; otherwise, to | |
| 6849 RE_SYNTAX_POSIX_BASIC; | |
| 6850 `newline_anchor' to REG_NEWLINE being set in CFLAGS; | |
| 6851 `fastmap' and `fastmap_accurate' to zero; | |
| 6852 `re_nsub' to the number of subexpressions in PATTERN. | |
| 502 | 6853 (non-shy of course. POSIX probably doesn't know about |
| 6854 shy ones, and in any case they should be invisible.) | |
| 428 | 6855 |
| 6856 PATTERN is the address of the pattern string. | |
| 6857 | |
| 6858 CFLAGS is a series of bits which affect compilation. | |
| 6859 | |
| 6860 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we | |
| 6861 use POSIX basic syntax. | |
| 6862 | |
| 6863 If REG_NEWLINE is set, then . and [^...] don't match newline. | |
| 6864 Also, regexec will try a match beginning after every newline. | |
| 6865 | |
| 6866 If REG_ICASE is set, then we considers upper- and lowercase | |
| 6867 versions of letters to be equivalent when matching. | |
| 6868 | |
| 6869 If REG_NOSUB is set, then when PREG is passed to regexec, that | |
| 6870 routine will report only success or failure, and nothing about the | |
| 6871 registers. | |
| 6872 | |
| 6873 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for | |
| 6874 the return codes and their meanings.) */ | |
| 6875 | |
| 6876 int | |
| 442 | 6877 regcomp (regex_t *preg, const char *pattern, int cflags) |
| 428 | 6878 { |
| 6879 reg_errcode_t ret; | |
| 647 | 6880 unsigned int syntax |
| 428 | 6881 = (cflags & REG_EXTENDED) ? |
| 6882 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; | |
| 6883 | |
| 6884 /* regex_compile will allocate the space for the compiled pattern. */ | |
| 6885 preg->buffer = 0; | |
| 6886 preg->allocated = 0; | |
| 6887 preg->used = 0; | |
| 6888 | |
| 6889 /* Don't bother to use a fastmap when searching. This simplifies the | |
| 6890 REG_NEWLINE case: if we used a fastmap, we'd have to put all the | |
| 6891 characters after newlines into the fastmap. This way, we just try | |
| 6892 every character. */ | |
| 6893 preg->fastmap = 0; | |
| 6894 | |
| 6895 if (cflags & REG_ICASE) | |
| 6896 { | |
| 647 | 6897 int i; |
| 428 | 6898 |
| 1333 | 6899 preg->translate = (char *) xmalloc (CHAR_SET_SIZE); |
| 428 | 6900 if (preg->translate == NULL) |
| 6901 return (int) REG_ESPACE; | |
| 6902 | |
| 6903 /* Map uppercase characters to corresponding lowercase ones. */ | |
| 6904 for (i = 0; i < CHAR_SET_SIZE; i++) | |
| 6905 preg->translate[i] = ISUPPER (i) ? tolower (i) : i; | |
| 6906 } | |
| 6907 else | |
| 6908 preg->translate = NULL; | |
| 6909 | |
| 6910 /* If REG_NEWLINE is set, newlines are treated differently. */ | |
| 6911 if (cflags & REG_NEWLINE) | |
| 6912 { /* REG_NEWLINE implies neither . nor [^...] match newline. */ | |
| 6913 syntax &= ~RE_DOT_NEWLINE; | |
| 6914 syntax |= RE_HAT_LISTS_NOT_NEWLINE; | |
| 6915 /* It also changes the matching behavior. */ | |
| 6916 preg->newline_anchor = 1; | |
| 6917 } | |
| 6918 else | |
| 6919 preg->newline_anchor = 0; | |
| 6920 | |
| 6921 preg->no_sub = !!(cflags & REG_NOSUB); | |
| 6922 | |
| 6923 /* POSIX says a null character in the pattern terminates it, so we | |
| 6924 can use strlen here in compiling the pattern. */ | |
| 446 | 6925 ret = regex_compile ((unsigned char *) pattern, strlen (pattern), syntax, preg); |
| 428 | 6926 |
| 6927 /* POSIX doesn't distinguish between an unmatched open-group and an | |
| 6928 unmatched close-group: both are REG_EPAREN. */ | |
| 6929 if (ret == REG_ERPAREN) ret = REG_EPAREN; | |
| 6930 | |
| 6931 return (int) ret; | |
| 6932 } | |
| 6933 | |
| 6934 | |
| 6935 /* regexec searches for a given pattern, specified by PREG, in the | |
| 6936 string STRING. | |
| 6937 | |
| 6938 If NMATCH is zero or REG_NOSUB was set in the cflags argument to | |
| 6939 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at | |
| 6940 least NMATCH elements, and we set them to the offsets of the | |
| 6941 corresponding matched substrings. | |
| 6942 | |
| 6943 EFLAGS specifies `execution flags' which affect matching: if | |
| 6944 REG_NOTBOL is set, then ^ does not match at the beginning of the | |
| 6945 string; if REG_NOTEOL is set, then $ does not match at the end. | |
| 6946 | |
| 6947 We return 0 if we find a match and REG_NOMATCH if not. */ | |
| 6948 | |
| 6949 int | |
| 442 | 6950 regexec (const regex_t *preg, const char *string, size_t nmatch, |
| 428 | 6951 regmatch_t pmatch[], int eflags) |
| 6952 { | |
| 6953 int ret; | |
| 6954 struct re_registers regs; | |
| 6955 regex_t private_preg; | |
| 6956 int len = strlen (string); | |
| 460 | 6957 re_bool want_reg_info = !preg->no_sub && nmatch > 0; |
| 428 | 6958 |
| 6959 private_preg = *preg; | |
| 6960 | |
| 6961 private_preg.not_bol = !!(eflags & REG_NOTBOL); | |
| 6962 private_preg.not_eol = !!(eflags & REG_NOTEOL); | |
| 6963 | |
| 6964 /* The user has told us exactly how many registers to return | |
| 6965 information about, via `nmatch'. We have to pass that on to the | |
| 6966 matching routines. */ | |
| 6967 private_preg.regs_allocated = REGS_FIXED; | |
| 6968 | |
| 6969 if (want_reg_info) | |
| 6970 { | |
| 647 | 6971 regs.num_regs = (int) nmatch; |
| 6972 regs.start = TALLOC ((int) nmatch, regoff_t); | |
| 6973 regs.end = TALLOC ((int) nmatch, regoff_t); | |
| 428 | 6974 if (regs.start == NULL || regs.end == NULL) |
| 6975 return (int) REG_NOMATCH; | |
| 6976 } | |
| 6977 | |
| 6978 /* Perform the searching operation. */ | |
| 6979 ret = re_search (&private_preg, string, len, | |
| 6980 /* start: */ 0, /* range: */ len, | |
| 6981 want_reg_info ? ®s : (struct re_registers *) 0); | |
| 6982 | |
| 6983 /* Copy the register information to the POSIX structure. */ | |
| 6984 if (want_reg_info) | |
| 6985 { | |
| 6986 if (ret >= 0) | |
| 6987 { | |
| 647 | 6988 int r; |
| 6989 | |
| 6990 for (r = 0; r < (int) nmatch; r++) | |
| 428 | 6991 { |
| 6992 pmatch[r].rm_so = regs.start[r]; | |
| 6993 pmatch[r].rm_eo = regs.end[r]; | |
| 6994 } | |
| 6995 } | |
| 6996 | |
| 6997 /* If we needed the temporary register info, free the space now. */ | |
|
4976
16112448d484
Rename xfree(FOO, TYPE) -> xfree(FOO)
Ben Wing <ben@xemacs.org>
parents:
4832
diff
changeset
|
6998 xfree (regs.start); |
|
16112448d484
Rename xfree(FOO, TYPE) -> xfree(FOO)
Ben Wing <ben@xemacs.org>
parents:
4832
diff
changeset
|
6999 xfree (regs.end); |
| 428 | 7000 } |
| 7001 | |
| 7002 /* We want zero return to mean success, unlike `re_search'. */ | |
| 7003 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; | |
| 7004 } | |
| 7005 | |
| 7006 | |
| 7007 /* Returns a message corresponding to an error code, ERRCODE, returned | |
| 7008 from either regcomp or regexec. We don't use PREG here. */ | |
| 7009 | |
| 7010 size_t | |
| 2286 | 7011 regerror (int errcode, const regex_t *UNUSED (preg), char *errbuf, |
| 647 | 7012 size_t errbuf_size) |
| 428 | 7013 { |
| 442 | 7014 const char *msg; |
| 665 | 7015 Bytecount msg_size; |
| 428 | 7016 |
| 7017 if (errcode < 0 | |
| 647 | 7018 || errcode >= (int) (sizeof (re_error_msgid) / |
| 7019 sizeof (re_error_msgid[0]))) | |
| 428 | 7020 /* Only error codes returned by the rest of the code should be passed |
| 7021 to this routine. If we are given anything else, or if other regex | |
| 7022 code generates an invalid error code, then the program has a bug. | |
| 7023 Dump core so we can fix it. */ | |
| 2500 | 7024 ABORT (); |
| 428 | 7025 |
| 7026 msg = gettext (re_error_msgid[errcode]); | |
| 7027 | |
| 7028 msg_size = strlen (msg) + 1; /* Includes the null. */ | |
| 7029 | |
| 7030 if (errbuf_size != 0) | |
| 7031 { | |
| 665 | 7032 if (msg_size > (Bytecount) errbuf_size) |
| 428 | 7033 { |
| 7034 strncpy (errbuf, msg, errbuf_size - 1); | |
| 7035 errbuf[errbuf_size - 1] = 0; | |
| 7036 } | |
| 7037 else | |
| 7038 strcpy (errbuf, msg); | |
| 7039 } | |
| 7040 | |
| 647 | 7041 return (size_t) msg_size; |
| 428 | 7042 } |
| 7043 | |
| 7044 | |
| 7045 /* Free dynamically allocated space used by PREG. */ | |
| 7046 | |
| 7047 void | |
| 7048 regfree (regex_t *preg) | |
| 7049 { | |
| 7050 if (preg->buffer != NULL) | |
|
4976
16112448d484
Rename xfree(FOO, TYPE) -> xfree(FOO)
Ben Wing <ben@xemacs.org>
parents:
4832
diff
changeset
|
7051 xfree (preg->buffer); |
| 428 | 7052 preg->buffer = NULL; |
| 7053 | |
| 7054 preg->allocated = 0; | |
| 7055 preg->used = 0; | |
| 7056 | |
| 7057 if (preg->fastmap != NULL) | |
|
4976
16112448d484
Rename xfree(FOO, TYPE) -> xfree(FOO)
Ben Wing <ben@xemacs.org>
parents:
4832
diff
changeset
|
7058 xfree (preg->fastmap); |
| 428 | 7059 preg->fastmap = NULL; |
| 7060 preg->fastmap_accurate = 0; | |
| 7061 | |
| 7062 if (preg->translate != NULL) | |
|
4976
16112448d484
Rename xfree(FOO, TYPE) -> xfree(FOO)
Ben Wing <ben@xemacs.org>
parents:
4832
diff
changeset
|
7063 xfree (preg->translate); |
| 428 | 7064 preg->translate = NULL; |
| 7065 } | |
| 7066 | |
| 7067 #endif /* not emacs */ | |
| 7068 |