|
Line 0
Link Here
|
|
|
1 |
/* Extended regular expression matching and search library, |
| 2 |
version 0.12. |
| 3 |
(Implements POSIX draft P1003.2/D11.2, except for some of the |
| 4 |
internationalization features.) |
| 5 |
Copyright (C) 1993, 94, 95, 96, 97, 98 Free Software Foundation, Inc. |
| 6 |
|
| 7 |
The GNU C Library is free software; you can redistribute it and/or |
| 8 |
modify it under the terms of the GNU Library General Public License as |
| 9 |
published by the Free Software Foundation; either version 2 of the |
| 10 |
License, or (at your option) any later version. |
| 11 |
|
| 12 |
The GNU C Library is distributed in the hope that it will be useful, |
| 13 |
but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 |
Library General Public License for more details. |
| 16 |
|
| 17 |
You should have received a copy of the GNU Library General Public |
| 18 |
License along with the GNU C Library; see the file COPYING.LIB. If not, |
| 19 |
write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 20 |
Boston, MA 02111-1307, USA. */ |
| 21 |
|
| 22 |
/* AIX requires this to be the first thing in the file. */ |
| 23 |
#if defined _AIX && !defined REGEX_MALLOC |
| 24 |
#pragma alloca |
| 25 |
#endif |
| 26 |
|
| 27 |
#undef _GNU_SOURCE |
| 28 |
#define _GNU_SOURCE |
| 29 |
|
| 30 |
#ifdef HAVE_CONFIG_H |
| 31 |
# include <config.h> |
| 32 |
#endif |
| 33 |
|
| 34 |
#ifndef PARAMS |
| 35 |
# if defined __GNUC__ || (defined __STDC__ && __STDC__) |
| 36 |
# define PARAMS(args) args |
| 37 |
# else |
| 38 |
# define PARAMS(args) () |
| 39 |
# endif /* GCC. */ |
| 40 |
#endif /* Not PARAMS. */ |
| 41 |
|
| 42 |
#if defined STDC_HEADERS && !defined emacs |
| 43 |
# include <stddef.h> |
| 44 |
#else |
| 45 |
/* We need this for `regex.h', and perhaps for the Emacs include files. */ |
| 46 |
# include <sys/types.h> |
| 47 |
#endif |
| 48 |
|
| 49 |
#define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC) |
| 50 |
|
| 51 |
/* For platform which support the ISO C amendement 1 functionality we |
| 52 |
support user defined character classes. */ |
| 53 |
#if defined _LIBC || WIDE_CHAR_SUPPORT |
| 54 |
/* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ |
| 55 |
# include <wchar.h> |
| 56 |
# include <wctype.h> |
| 57 |
#endif |
| 58 |
|
| 59 |
#ifdef _LIBC |
| 60 |
/* We have to keep the namespace clean. */ |
| 61 |
# define regfree(preg) __regfree (preg) |
| 62 |
# define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) |
| 63 |
# define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) |
| 64 |
# define regerror(errcode, preg, errbuf, errbuf_size) \ |
| 65 |
__regerror(errcode, preg, errbuf, errbuf_size) |
| 66 |
# define re_set_registers(bu, re, nu, st, en) \ |
| 67 |
__re_set_registers (bu, re, nu, st, en) |
| 68 |
# define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ |
| 69 |
__re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) |
| 70 |
# define re_match(bufp, string, size, pos, regs) \ |
| 71 |
__re_match (bufp, string, size, pos, regs) |
| 72 |
# define re_search(bufp, string, size, startpos, range, regs) \ |
| 73 |
__re_search (bufp, string, size, startpos, range, regs) |
| 74 |
# define re_compile_pattern(pattern, length, bufp) \ |
| 75 |
__re_compile_pattern (pattern, length, bufp) |
| 76 |
# define re_set_syntax(syntax) __re_set_syntax (syntax) |
| 77 |
# define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ |
| 78 |
__re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) |
| 79 |
# define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) |
| 80 |
|
| 81 |
#define btowc __btowc |
| 82 |
#endif |
| 83 |
|
| 84 |
/* This is for other GNU distributions with internationalized messages. */ |
| 85 |
#if HAVE_LIBINTL_H || defined _LIBC |
| 86 |
# include <libintl.h> |
| 87 |
#else |
| 88 |
# define gettext(msgid) (msgid) |
| 89 |
#endif |
| 90 |
|
| 91 |
#ifndef gettext_noop |
| 92 |
/* This define is so xgettext can find the internationalizable |
| 93 |
strings. */ |
| 94 |
# define gettext_noop(String) String |
| 95 |
#endif |
| 96 |
|
| 97 |
/* The `emacs' switch turns on certain matching commands |
| 98 |
that make sense only in Emacs. */ |
| 99 |
#ifdef emacs |
| 100 |
|
| 101 |
# include "lisp.h" |
| 102 |
# include "buffer.h" |
| 103 |
# include "syntax.h" |
| 104 |
|
| 105 |
#else /* not emacs */ |
| 106 |
|
| 107 |
/* If we are not linking with Emacs proper, |
| 108 |
we can't use the relocating allocator |
| 109 |
even if config.h says that we can. */ |
| 110 |
# undef REL_ALLOC |
| 111 |
|
| 112 |
# if defined STDC_HEADERS || defined _LIBC |
| 113 |
# include <stdlib.h> |
| 114 |
# else |
| 115 |
char *malloc (); |
| 116 |
char *realloc (); |
| 117 |
# endif |
| 118 |
|
| 119 |
/* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. |
| 120 |
If nothing else has been done, use the method below. */ |
| 121 |
# ifdef INHIBIT_STRING_HEADER |
| 122 |
# if !(defined HAVE_BZERO && defined HAVE_BCOPY) |
| 123 |
# if !defined bzero && !defined bcopy |
| 124 |
# undef INHIBIT_STRING_HEADER |
| 125 |
# endif |
| 126 |
# endif |
| 127 |
# endif |
| 128 |
|
| 129 |
/* This is the normal way of making sure we have a bcopy and a bzero. |
| 130 |
This is used in most programs--a few other programs avoid this |
| 131 |
by defining INHIBIT_STRING_HEADER. */ |
| 132 |
# ifndef INHIBIT_STRING_HEADER |
| 133 |
# if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC |
| 134 |
# include <string.h> |
| 135 |
# ifndef bzero |
| 136 |
# ifndef _LIBC |
| 137 |
# define bzero(s, n) (memset (s, '\0', n), (s)) |
| 138 |
# else |
| 139 |
# define bzero(s, n) __bzero (s, n) |
| 140 |
# endif |
| 141 |
# endif |
| 142 |
# else |
| 143 |
# include <strings.h> |
| 144 |
# ifndef memcmp |
| 145 |
# define memcmp(s1, s2, n) bcmp (s1, s2, n) |
| 146 |
# endif |
| 147 |
# ifndef memcpy |
| 148 |
# define memcpy(d, s, n) (bcopy (s, d, n), (d)) |
| 149 |
# endif |
| 150 |
# endif |
| 151 |
# endif |
| 152 |
|
| 153 |
/* Define the syntax stuff for \<, \>, etc. */ |
| 154 |
|
| 155 |
/* This must be nonzero for the wordchar and notwordchar pattern |
| 156 |
commands in re_match_2. */ |
| 157 |
# ifndef Sword |
| 158 |
# define Sword 1 |
| 159 |
# endif |
| 160 |
|
| 161 |
# ifdef SWITCH_ENUM_BUG |
| 162 |
# define SWITCH_ENUM_CAST(x) ((int)(x)) |
| 163 |
# else |
| 164 |
# define SWITCH_ENUM_CAST(x) (x) |
| 165 |
# endif |
| 166 |
|
| 167 |
/* How many characters in the character set. */ |
| 168 |
# define CHAR_SET_SIZE 256 |
| 169 |
|
| 170 |
# ifdef SYNTAX_TABLE |
| 171 |
|
| 172 |
extern char *re_syntax_table; |
| 173 |
|
| 174 |
# else /* not SYNTAX_TABLE */ |
| 175 |
|
| 176 |
static char re_syntax_table[CHAR_SET_SIZE]; |
| 177 |
|
| 178 |
static void |
| 179 |
init_syntax_once () |
| 180 |
{ |
| 181 |
register int c; |
| 182 |
static int done = 0; |
| 183 |
|
| 184 |
if (done) |
| 185 |
return; |
| 186 |
|
| 187 |
bzero (re_syntax_table, sizeof re_syntax_table); |
| 188 |
|
| 189 |
for (c = 'a'; c <= 'z'; c++) |
| 190 |
re_syntax_table[c] = Sword; |
| 191 |
|
| 192 |
for (c = 'A'; c <= 'Z'; c++) |
| 193 |
re_syntax_table[c] = Sword; |
| 194 |
|
| 195 |
for (c = '0'; c <= '9'; c++) |
| 196 |
re_syntax_table[c] = Sword; |
| 197 |
|
| 198 |
re_syntax_table['_'] = Sword; |
| 199 |
|
| 200 |
done = 1; |
| 201 |
} |
| 202 |
|
| 203 |
# endif /* not SYNTAX_TABLE */ |
| 204 |
|
| 205 |
# define SYNTAX(c) re_syntax_table[c] |
| 206 |
|
| 207 |
#endif /* not emacs */ |
| 208 |
|
| 209 |
/* Get the interface, including the syntax bits. */ |
| 210 |
#include "regex.h" |
| 211 |
|
| 212 |
/* isalpha etc. are used for the character classes. */ |
| 213 |
#include <ctype.h> |
| 214 |
|
| 215 |
/* Jim Meyering writes: |
| 216 |
|
| 217 |
"... Some ctype macros are valid only for character codes that |
| 218 |
isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when |
| 219 |
using /bin/cc or gcc but without giving an ansi option). So, all |
| 220 |
ctype uses should be through macros like ISPRINT... If |
| 221 |
STDC_HEADERS is defined, then autoconf has verified that the ctype |
| 222 |
macros don't need to be guarded with references to isascii. ... |
| 223 |
Defining isascii to 1 should let any compiler worth its salt |
| 224 |
eliminate the && through constant folding." |
| 225 |
Solaris defines some of these symbols so we must undefine them first. */ |
| 226 |
|
| 227 |
#undef ISASCII |
| 228 |
#if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) |
| 229 |
# define ISASCII(c) 1 |
| 230 |
#else |
| 231 |
# define ISASCII(c) isascii(c) |
| 232 |
#endif |
| 233 |
|
| 234 |
#ifdef isblank |
| 235 |
# define ISBLANK(c) (ISASCII (c) && isblank (c)) |
| 236 |
#else |
| 237 |
# define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
| 238 |
#endif |
| 239 |
#ifdef isgraph |
| 240 |
# define ISGRAPH(c) (ISASCII (c) && isgraph (c)) |
| 241 |
#else |
| 242 |
# define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) |
| 243 |
#endif |
| 244 |
|
| 245 |
#undef ISPRINT |
| 246 |
#define ISPRINT(c) (ISASCII (c) && isprint (c)) |
| 247 |
#define ISDIGIT(c) (ISASCII (c) && isdigit (c)) |
| 248 |
#define ISALNUM(c) (ISASCII (c) && isalnum (c)) |
| 249 |
#define ISALPHA(c) (ISASCII (c) && isalpha (c)) |
| 250 |
#define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) |
| 251 |
#define ISLOWER(c) (ISASCII (c) && islower (c)) |
| 252 |
#define ISPUNCT(c) (ISASCII (c) && ispunct (c)) |
| 253 |
#define ISSPACE(c) (ISASCII (c) && isspace (c)) |
| 254 |
#define ISUPPER(c) (ISASCII (c) && isupper (c)) |
| 255 |
#define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) |
| 256 |
|
| 257 |
#ifndef NULL |
| 258 |
# define NULL (void *)0 |
| 259 |
#endif |
| 260 |
|
| 261 |
/* We remove any previous definition of `SIGN_EXTEND_CHAR', |
| 262 |
since ours (we hope) works properly with all combinations of |
| 263 |
machines, compilers, `char' and `unsigned char' argument types. |
| 264 |
(Per Bothner suggested the basic approach.) */ |
| 265 |
#undef SIGN_EXTEND_CHAR |
| 266 |
#if __STDC__ |
| 267 |
# define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
| 268 |
#else /* not __STDC__ */ |
| 269 |
/* As in Harbison and Steele. */ |
| 270 |
# define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) |
| 271 |
#endif |
| 272 |
|
| 273 |
/* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we |
| 274 |
use `alloca' instead of `malloc'. This is because using malloc in |
| 275 |
re_search* or re_match* could cause memory leaks when C-g is used in |
| 276 |
Emacs; also, malloc is slower and causes storage fragmentation. On |
| 277 |
the other hand, malloc is more portable, and easier to debug. |
| 278 |
|
| 279 |
Because we sometimes use alloca, some routines have to be macros, |
| 280 |
not functions -- `alloca'-allocated space disappears at the end of the |
| 281 |
function it is called in. */ |
| 282 |
|
| 283 |
#ifdef REGEX_MALLOC |
| 284 |
|
| 285 |
# define REGEX_ALLOCATE malloc |
| 286 |
# define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) |
| 287 |
# define REGEX_FREE free |
| 288 |
|
| 289 |
#else /* not REGEX_MALLOC */ |
| 290 |
|
| 291 |
/* Emacs already defines alloca, sometimes. */ |
| 292 |
# ifndef alloca |
| 293 |
|
| 294 |
/* Make alloca work the best possible way. */ |
| 295 |
# ifdef __GNUC__ |
| 296 |
# define alloca __builtin_alloca |
| 297 |
# else /* not __GNUC__ */ |
| 298 |
# if HAVE_ALLOCA_H |
| 299 |
# include <alloca.h> |
| 300 |
# endif /* HAVE_ALLOCA_H */ |
| 301 |
# endif /* not __GNUC__ */ |
| 302 |
|
| 303 |
# endif /* not alloca */ |
| 304 |
|
| 305 |
# define REGEX_ALLOCATE alloca |
| 306 |
|
| 307 |
/* Assumes a `char *destination' variable. */ |
| 308 |
# define REGEX_REALLOCATE(source, osize, nsize) \ |
| 309 |
(destination = (char *) alloca (nsize), \ |
| 310 |
memcpy (destination, source, osize)) |
| 311 |
|
| 312 |
/* No need to do anything to free, after alloca. */ |
| 313 |
# define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
| 314 |
|
| 315 |
#endif /* not REGEX_MALLOC */ |
| 316 |
|
| 317 |
/* Define how to allocate the failure stack. */ |
| 318 |
|
| 319 |
#if defined REL_ALLOC && defined REGEX_MALLOC |
| 320 |
|
| 321 |
# define REGEX_ALLOCATE_STACK(size) \ |
| 322 |
r_alloc (&failure_stack_ptr, (size)) |
| 323 |
# define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
| 324 |
r_re_alloc (&failure_stack_ptr, (nsize)) |
| 325 |
# define REGEX_FREE_STACK(ptr) \ |
| 326 |
r_alloc_free (&failure_stack_ptr) |
| 327 |
|
| 328 |
#else /* not using relocating allocator */ |
| 329 |
|
| 330 |
# ifdef REGEX_MALLOC |
| 331 |
|
| 332 |
# define REGEX_ALLOCATE_STACK malloc |
| 333 |
# define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) |
| 334 |
# define REGEX_FREE_STACK free |
| 335 |
|
| 336 |
# else /* not REGEX_MALLOC */ |
| 337 |
|
| 338 |
# define REGEX_ALLOCATE_STACK alloca |
| 339 |
|
| 340 |
# define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
| 341 |
REGEX_REALLOCATE (source, osize, nsize) |
| 342 |
/* No need to explicitly free anything. */ |
| 343 |
# define REGEX_FREE_STACK(arg) |
| 344 |
|
| 345 |
# endif /* not REGEX_MALLOC */ |
| 346 |
#endif /* not using relocating allocator */ |
| 347 |
|
| 348 |
|
| 349 |
/* True if `size1' is non-NULL and PTR is pointing anywhere inside |
| 350 |
`string1' or just past its end. This works if PTR is NULL, which is |
| 351 |
a good thing. */ |
| 352 |
#define FIRST_STRING_P(ptr) \ |
| 353 |
(size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
| 354 |
|
| 355 |
/* (Re)Allocate N items of type T using malloc, or fail. */ |
| 356 |
#define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) |
| 357 |
#define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) |
| 358 |
#define RETALLOC_IF(addr, n, t) \ |
| 359 |
if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) |
| 360 |
#define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) |
| 361 |
|
| 362 |
#define BYTEWIDTH 8 /* In bits. */ |
| 363 |
|
| 364 |
#define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) |
| 365 |
|
| 366 |
#undef MAX |
| 367 |
#undef MIN |
| 368 |
#define MAX(a, b) ((a) > (b) ? (a) : (b)) |
| 369 |
#define MIN(a, b) ((a) < (b) ? (a) : (b)) |
| 370 |
|
| 371 |
typedef char boolean; |
| 372 |
#define false 0 |
| 373 |
#define true 1 |
| 374 |
|
| 375 |
static int re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp, |
| 376 |
const char *string1, int size1, |
| 377 |
const char *string2, int size2, |
| 378 |
int pos, |
| 379 |
struct re_registers *regs, |
| 380 |
int stop)); |
| 381 |
|
| 382 |
/* These are the command codes that appear in compiled regular |
| 383 |
expressions. Some opcodes are followed by argument bytes. A |
| 384 |
command code can specify any interpretation whatsoever for its |
| 385 |
arguments. Zero bytes may appear in the compiled regular expression. */ |
| 386 |
|
| 387 |
typedef enum |
| 388 |
{ |
| 389 |
no_op = 0, |
| 390 |
|
| 391 |
/* Succeed right away--no more backtracking. */ |
| 392 |
succeed, |
| 393 |
|
| 394 |
/* Followed by one byte giving n, then by n literal bytes. */ |
| 395 |
exactn, |
| 396 |
|
| 397 |
/* Matches any (more or less) character. */ |
| 398 |
anychar, |
| 399 |
|
| 400 |
/* Matches any one char belonging to specified set. First |
| 401 |
following byte is number of bitmap bytes. Then come bytes |
| 402 |
for a bitmap saying which chars are in. Bits in each byte |
| 403 |
are ordered low-bit-first. A character is in the set if its |
| 404 |
bit is 1. A character too large to have a bit in the map is |
| 405 |
automatically not in the set. */ |
| 406 |
charset, |
| 407 |
|
| 408 |
/* Same parameters as charset, but match any character that is |
| 409 |
not one of those specified. */ |
| 410 |
charset_not, |
| 411 |
|
| 412 |
/* Start remembering the text that is matched, for storing in a |
| 413 |
register. Followed by one byte with the register number, in |
| 414 |
the range 0 to one less than the pattern buffer's re_nsub |
| 415 |
field. Then followed by one byte with the number of groups |
| 416 |
inner to this one. (This last has to be part of the |
| 417 |
start_memory only because we need it in the on_failure_jump |
| 418 |
of re_match_2.) */ |
| 419 |
start_memory, |
| 420 |
|
| 421 |
/* Stop remembering the text that is matched and store it in a |
| 422 |
memory register. Followed by one byte with the register |
| 423 |
number, in the range 0 to one less than `re_nsub' in the |
| 424 |
pattern buffer, and one byte with the number of inner groups, |
| 425 |
just like `start_memory'. (We need the number of inner |
| 426 |
groups here because we don't have any easy way of finding the |
| 427 |
corresponding start_memory when we're at a stop_memory.) */ |
| 428 |
stop_memory, |
| 429 |
|
| 430 |
/* Match a duplicate of something remembered. Followed by one |
| 431 |
byte containing the register number. */ |
| 432 |
duplicate, |
| 433 |
|
| 434 |
/* Fail unless at beginning of line. */ |
| 435 |
begline, |
| 436 |
|
| 437 |
/* Fail unless at end of line. */ |
| 438 |
endline, |
| 439 |
|
| 440 |
/* Succeeds if at beginning of buffer (if emacs) or at beginning |
| 441 |
of string to be matched (if not). */ |
| 442 |
begbuf, |
| 443 |
|
| 444 |
/* Analogously, for end of buffer/string. */ |
| 445 |
endbuf, |
| 446 |
|
| 447 |
/* Followed by two byte relative address to which to jump. */ |
| 448 |
jump, |
| 449 |
|
| 450 |
/* Same as jump, but marks the end of an alternative. */ |
| 451 |
jump_past_alt, |
| 452 |
|
| 453 |
/* Followed by two-byte relative address of place to resume at |
| 454 |
in case of failure. */ |
| 455 |
on_failure_jump, |
| 456 |
|
| 457 |
/* Like on_failure_jump, but pushes a placeholder instead of the |
| 458 |
current string position when executed. */ |
| 459 |
on_failure_keep_string_jump, |
| 460 |
|
| 461 |
/* Throw away latest failure point and then jump to following |
| 462 |
two-byte relative address. */ |
| 463 |
pop_failure_jump, |
| 464 |
|
| 465 |
/* Change to pop_failure_jump if know won't have to backtrack to |
| 466 |
match; otherwise change to jump. This is used to jump |
| 467 |
back to the beginning of a repeat. If what follows this jump |
| 468 |
clearly won't match what the repeat does, such that we can be |
| 469 |
sure that there is no use backtracking out of repetitions |
| 470 |
already matched, then we change it to a pop_failure_jump. |
| 471 |
Followed by two-byte address. */ |
| 472 |
maybe_pop_jump, |
| 473 |
|
| 474 |
/* Jump to following two-byte address, and push a dummy failure |
| 475 |
point. This failure point will be thrown away if an attempt |
| 476 |
is made to use it for a failure. A `+' construct makes this |
| 477 |
before the first repeat. Also used as an intermediary kind |
| 478 |
of jump when compiling an alternative. */ |
| 479 |
dummy_failure_jump, |
| 480 |
|
| 481 |
/* Push a dummy failure point and continue. Used at the end of |
| 482 |
alternatives. */ |
| 483 |
push_dummy_failure, |
| 484 |
|
| 485 |
/* Followed by two-byte relative address and two-byte number n. |
| 486 |
After matching N times, jump to the address upon failure. */ |
| 487 |
succeed_n, |
| 488 |
|
| 489 |
/* Followed by two-byte relative address, and two-byte number n. |
| 490 |
Jump to the address N times, then fail. */ |
| 491 |
jump_n, |
| 492 |
|
| 493 |
/* Set the following two-byte relative address to the |
| 494 |
subsequent two-byte number. The address *includes* the two |
| 495 |
bytes of number. */ |
| 496 |
set_number_at, |
| 497 |
|
| 498 |
wordchar, /* Matches any word-constituent character. */ |
| 499 |
notwordchar, /* Matches any char that is not a word-constituent. */ |
| 500 |
|
| 501 |
wordbeg, /* Succeeds if at word beginning. */ |
| 502 |
wordend, /* Succeeds if at word end. */ |
| 503 |
|
| 504 |
wordbound, /* Succeeds if at a word boundary. */ |
| 505 |
notwordbound /* Succeeds if not at a word boundary. */ |
| 506 |
|
| 507 |
#ifdef emacs |
| 508 |
,before_dot, /* Succeeds if before point. */ |
| 509 |
at_dot, /* Succeeds if at point. */ |
| 510 |
after_dot, /* Succeeds if after point. */ |
| 511 |
|
| 512 |
/* Matches any character whose syntax is specified. Followed by |
| 513 |
a byte which contains a syntax code, e.g., Sword. */ |
| 514 |
syntaxspec, |
| 515 |
|
| 516 |
/* Matches any character whose syntax is not that specified. */ |
| 517 |
notsyntaxspec |
| 518 |
#endif /* emacs */ |
| 519 |
} re_opcode_t; |
| 520 |
|
| 521 |
/* Common operations on the compiled pattern. */ |
| 522 |
|
| 523 |
/* Store NUMBER in two contiguous bytes starting at DESTINATION. */ |
| 524 |
|
| 525 |
#define STORE_NUMBER(destination, number) \ |
| 526 |
do { \ |
| 527 |
(destination)[0] = (number) & 0377; \ |
| 528 |
(destination)[1] = (number) >> 8; \ |
| 529 |
} while (0) |
| 530 |
|
| 531 |
/* Same as STORE_NUMBER, except increment DESTINATION to |
| 532 |
the byte after where the number is stored. Therefore, DESTINATION |
| 533 |
must be an lvalue. */ |
| 534 |
|
| 535 |
#define STORE_NUMBER_AND_INCR(destination, number) \ |
| 536 |
do { \ |
| 537 |
STORE_NUMBER (destination, number); \ |
| 538 |
(destination) += 2; \ |
| 539 |
} while (0) |
| 540 |
|
| 541 |
/* Put into DESTINATION a number stored in two contiguous bytes starting |
| 542 |
at SOURCE. */ |
| 543 |
|
| 544 |
#define EXTRACT_NUMBER(destination, source) \ |
| 545 |
do { \ |
| 546 |
(destination) = *(source) & 0377; \ |
| 547 |
(destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ |
| 548 |
} while (0) |
| 549 |
|
| 550 |
#ifdef DEBUG |
| 551 |
static void extract_number _RE_ARGS ((int *dest, unsigned char *source)); |
| 552 |
static void |
| 553 |
extract_number (dest, source) |
| 554 |
int *dest; |
| 555 |
unsigned char *source; |
| 556 |
{ |
| 557 |
int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
| 558 |
*dest = *source & 0377; |
| 559 |
*dest += temp << 8; |
| 560 |
} |
| 561 |
|
| 562 |
# ifndef EXTRACT_MACROS /* To debug the macros. */ |
| 563 |
# undef EXTRACT_NUMBER |
| 564 |
# define EXTRACT_NUMBER(dest, src) extract_number (&dest, src) |
| 565 |
# endif /* not EXTRACT_MACROS */ |
| 566 |
|
| 567 |
#endif /* DEBUG */ |
| 568 |
|
| 569 |
/* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. |
| 570 |
SOURCE must be an lvalue. */ |
| 571 |
|
| 572 |
#define EXTRACT_NUMBER_AND_INCR(destination, source) \ |
| 573 |
do { \ |
| 574 |
EXTRACT_NUMBER (destination, source); \ |
| 575 |
(source) += 2; \ |
| 576 |
} while (0) |
| 577 |
|
| 578 |
#ifdef DEBUG |
| 579 |
static void extract_number_and_incr _RE_ARGS ((int *destination, |
| 580 |
unsigned char **source)); |
| 581 |
static void |
| 582 |
extract_number_and_incr (destination, source) |
| 583 |
int *destination; |
| 584 |
unsigned char **source; |
| 585 |
{ |
| 586 |
extract_number (destination, *source); |
| 587 |
*source += 2; |
| 588 |
} |
| 589 |
|
| 590 |
# ifndef EXTRACT_MACROS |
| 591 |
# undef EXTRACT_NUMBER_AND_INCR |
| 592 |
# define EXTRACT_NUMBER_AND_INCR(dest, src) \ |
| 593 |
extract_number_and_incr (&dest, &src) |
| 594 |
# endif /* not EXTRACT_MACROS */ |
| 595 |
|
| 596 |
#endif /* DEBUG */ |
| 597 |
|
| 598 |
/* If DEBUG is defined, Regex prints many voluminous messages about what |
| 599 |
it is doing (if the variable `debug' is nonzero). If linked with the |
| 600 |
main program in `iregex.c', you can enter patterns and strings |
| 601 |
interactively. And if linked with the main program in `main.c' and |
| 602 |
the other test files, you can run the already-written tests. */ |
| 603 |
|
| 604 |
#ifdef DEBUG |
| 605 |
|
| 606 |
/* We use standard I/O for debugging. */ |
| 607 |
# include <stdio.h> |
| 608 |
|
| 609 |
/* It is useful to test things that ``must'' be true when debugging. */ |
| 610 |
# include <assert.h> |
| 611 |
|
| 612 |
static int debug = 0; |
| 613 |
|
| 614 |
# define DEBUG_STATEMENT(e) e |
| 615 |
# define DEBUG_PRINT1(x) if (debug) printf (x) |
| 616 |
# define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) |
| 617 |
# define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) |
| 618 |
# define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) |
| 619 |
# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
| 620 |
if (debug) print_partial_compiled_pattern (s, e) |
| 621 |
# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
| 622 |
if (debug) print_double_string (w, s1, sz1, s2, sz2) |
| 623 |
|
| 624 |
|
| 625 |
/* Print the fastmap in human-readable form. */ |
| 626 |
|
| 627 |
void |
| 628 |
print_fastmap (fastmap) |
| 629 |
char *fastmap; |
| 630 |
{ |
| 631 |
unsigned was_a_range = 0; |
| 632 |
unsigned i = 0; |
| 633 |
|
| 634 |
while (i < (1 << BYTEWIDTH)) |
| 635 |
{ |
| 636 |
if (fastmap[i++]) |
| 637 |
{ |
| 638 |
was_a_range = 0; |
| 639 |
putchar (i - 1); |
| 640 |
while (i < (1 << BYTEWIDTH) && fastmap[i]) |
| 641 |
{ |
| 642 |
was_a_range = 1; |
| 643 |
i++; |
| 644 |
} |
| 645 |
if (was_a_range) |
| 646 |
{ |
| 647 |
printf ("-"); |
| 648 |
putchar (i - 1); |
| 649 |
} |
| 650 |
} |
| 651 |
} |
| 652 |
putchar ('\n'); |
| 653 |
} |
| 654 |
|
| 655 |
|
| 656 |
/* Print a compiled pattern string in human-readable form, starting at |
| 657 |
the START pointer into it and ending just before the pointer END. */ |
| 658 |
|
| 659 |
void |
| 660 |
print_partial_compiled_pattern (start, end) |
| 661 |
unsigned char *start; |
| 662 |
unsigned char *end; |
| 663 |
{ |
| 664 |
int mcnt, mcnt2; |
| 665 |
unsigned char *p1; |
| 666 |
unsigned char *p = start; |
| 667 |
unsigned char *pend = end; |
| 668 |
|
| 669 |
if (start == NULL) |
| 670 |
{ |
| 671 |
printf ("(null)\n"); |
| 672 |
return; |
| 673 |
} |
| 674 |
|
| 675 |
/* Loop over pattern commands. */ |
| 676 |
while (p < pend) |
| 677 |
{ |
| 678 |
printf ("%d:\t", p - start); |
| 679 |
|
| 680 |
switch ((re_opcode_t) *p++) |
| 681 |
{ |
| 682 |
case no_op: |
| 683 |
printf ("/no_op"); |
| 684 |
break; |
| 685 |
|
| 686 |
case exactn: |
| 687 |
mcnt = *p++; |
| 688 |
printf ("/exactn/%d", mcnt); |
| 689 |
do |
| 690 |
{ |
| 691 |
putchar ('/'); |
| 692 |
putchar (*p++); |
| 693 |
} |
| 694 |
while (--mcnt); |
| 695 |
break; |
| 696 |
|
| 697 |
case start_memory: |
| 698 |
mcnt = *p++; |
| 699 |
printf ("/start_memory/%d/%d", mcnt, *p++); |
| 700 |
break; |
| 701 |
|
| 702 |
case stop_memory: |
| 703 |
mcnt = *p++; |
| 704 |
printf ("/stop_memory/%d/%d", mcnt, *p++); |
| 705 |
break; |
| 706 |
|
| 707 |
case duplicate: |
| 708 |
printf ("/duplicate/%d", *p++); |
| 709 |
break; |
| 710 |
|
| 711 |
case anychar: |
| 712 |
printf ("/anychar"); |
| 713 |
break; |
| 714 |
|
| 715 |
case charset: |
| 716 |
case charset_not: |
| 717 |
{ |
| 718 |
register int c, last = -100; |
| 719 |
register int in_range = 0; |
| 720 |
|
| 721 |
printf ("/charset [%s", |
| 722 |
(re_opcode_t) *(p - 1) == charset_not ? "^" : ""); |
| 723 |
|
| 724 |
assert (p + *p < pend); |
| 725 |
|
| 726 |
for (c = 0; c < 256; c++) |
| 727 |
if (c / 8 < *p |
| 728 |
&& (p[1 + (c/8)] & (1 << (c % 8)))) |
| 729 |
{ |
| 730 |
/* Are we starting a range? */ |
| 731 |
if (last + 1 == c && ! in_range) |
| 732 |
{ |
| 733 |
putchar ('-'); |
| 734 |
in_range = 1; |
| 735 |
} |
| 736 |
/* Have we broken a range? */ |
| 737 |
else if (last + 1 != c && in_range) |
| 738 |
{ |
| 739 |
putchar (last); |
| 740 |
in_range = 0; |
| 741 |
} |
| 742 |
|
| 743 |
if (! in_range) |
| 744 |
putchar (c); |
| 745 |
|
| 746 |
last = c; |
| 747 |
} |
| 748 |
|
| 749 |
if (in_range) |
| 750 |
putchar (last); |
| 751 |
|
| 752 |
putchar (']'); |
| 753 |
|
| 754 |
p += 1 + *p; |
| 755 |
} |
| 756 |
break; |
| 757 |
|
| 758 |
case begline: |
| 759 |
printf ("/begline"); |
| 760 |
break; |
| 761 |
|
| 762 |
case endline: |
| 763 |
printf ("/endline"); |
| 764 |
break; |
| 765 |
|
| 766 |
case on_failure_jump: |
| 767 |
extract_number_and_incr (&mcnt, &p); |
| 768 |
printf ("/on_failure_jump to %d", p + mcnt - start); |
| 769 |
break; |
| 770 |
|
| 771 |
case on_failure_keep_string_jump: |
| 772 |
extract_number_and_incr (&mcnt, &p); |
| 773 |
printf ("/on_failure_keep_string_jump to %d", p + mcnt - start); |
| 774 |
break; |
| 775 |
|
| 776 |
case dummy_failure_jump: |
| 777 |
extract_number_and_incr (&mcnt, &p); |
| 778 |
printf ("/dummy_failure_jump to %d", p + mcnt - start); |
| 779 |
break; |
| 780 |
|
| 781 |
case push_dummy_failure: |
| 782 |
printf ("/push_dummy_failure"); |
| 783 |
break; |
| 784 |
|
| 785 |
case maybe_pop_jump: |
| 786 |
extract_number_and_incr (&mcnt, &p); |
| 787 |
printf ("/maybe_pop_jump to %d", p + mcnt - start); |
| 788 |
break; |
| 789 |
|
| 790 |
case pop_failure_jump: |
| 791 |
extract_number_and_incr (&mcnt, &p); |
| 792 |
printf ("/pop_failure_jump to %d", p + mcnt - start); |
| 793 |
break; |
| 794 |
|
| 795 |
case jump_past_alt: |
| 796 |
extract_number_and_incr (&mcnt, &p); |
| 797 |
printf ("/jump_past_alt to %d", p + mcnt - start); |
| 798 |
break; |
| 799 |
|
| 800 |
case jump: |
| 801 |
extract_number_and_incr (&mcnt, &p); |
| 802 |
printf ("/jump to %d", p + mcnt - start); |
| 803 |
break; |
| 804 |
|
| 805 |
case succeed_n: |
| 806 |
extract_number_and_incr (&mcnt, &p); |
| 807 |
p1 = p + mcnt; |
| 808 |
extract_number_and_incr (&mcnt2, &p); |
| 809 |
printf ("/succeed_n to %d, %d times", p1 - start, mcnt2); |
| 810 |
break; |
| 811 |
|
| 812 |
case jump_n: |
| 813 |
extract_number_and_incr (&mcnt, &p); |
| 814 |
p1 = p + mcnt; |
| 815 |
extract_number_and_incr (&mcnt2, &p); |
| 816 |
printf ("/jump_n to %d, %d times", p1 - start, mcnt2); |
| 817 |
break; |
| 818 |
|
| 819 |
case set_number_at: |
| 820 |
extract_number_and_incr (&mcnt, &p); |
| 821 |
p1 = p + mcnt; |
| 822 |
extract_number_and_incr (&mcnt2, &p); |
| 823 |
printf ("/set_number_at location %d to %d", p1 - start, mcnt2); |
| 824 |
break; |
| 825 |
|
| 826 |
case wordbound: |
| 827 |
printf ("/wordbound"); |
| 828 |
break; |
| 829 |
|
| 830 |
case notwordbound: |
| 831 |
printf ("/notwordbound"); |
| 832 |
break; |
| 833 |
|
| 834 |
case wordbeg: |
| 835 |
printf ("/wordbeg"); |
| 836 |
break; |
| 837 |
|
| 838 |
case wordend: |
| 839 |
printf ("/wordend"); |
| 840 |
|
| 841 |
# ifdef emacs |
| 842 |
case before_dot: |
| 843 |
printf ("/before_dot"); |
| 844 |
break; |
| 845 |
|
| 846 |
case at_dot: |
| 847 |
printf ("/at_dot"); |
| 848 |
break; |
| 849 |
|
| 850 |
case after_dot: |
| 851 |
printf ("/after_dot"); |
| 852 |
break; |
| 853 |
|
| 854 |
case syntaxspec: |
| 855 |
printf ("/syntaxspec"); |
| 856 |
mcnt = *p++; |
| 857 |
printf ("/%d", mcnt); |
| 858 |
break; |
| 859 |
|
| 860 |
case notsyntaxspec: |
| 861 |
printf ("/notsyntaxspec"); |
| 862 |
mcnt = *p++; |
| 863 |
printf ("/%d", mcnt); |
| 864 |
break; |
| 865 |
# endif /* emacs */ |
| 866 |
|
| 867 |
case wordchar: |
| 868 |
printf ("/wordchar"); |
| 869 |
break; |
| 870 |
|
| 871 |
case notwordchar: |
| 872 |
printf ("/notwordchar"); |
| 873 |
break; |
| 874 |
|
| 875 |
case begbuf: |
| 876 |
printf ("/begbuf"); |
| 877 |
break; |
| 878 |
|
| 879 |
case endbuf: |
| 880 |
printf ("/endbuf"); |
| 881 |
break; |
| 882 |
|
| 883 |
default: |
| 884 |
printf ("?%d", *(p-1)); |
| 885 |
} |
| 886 |
|
| 887 |
putchar ('\n'); |
| 888 |
} |
| 889 |
|
| 890 |
printf ("%d:\tend of pattern.\n", p - start); |
| 891 |
} |
| 892 |
|
| 893 |
|
| 894 |
void |
| 895 |
print_compiled_pattern (bufp) |
| 896 |
struct re_pattern_buffer *bufp; |
| 897 |
{ |
| 898 |
unsigned char *buffer = bufp->buffer; |
| 899 |
|
| 900 |
print_partial_compiled_pattern (buffer, buffer + bufp->used); |
| 901 |
printf ("%ld bytes used/%ld bytes allocated.\n", |
| 902 |
bufp->used, bufp->allocated); |
| 903 |
|
| 904 |
if (bufp->fastmap_accurate && bufp->fastmap) |
| 905 |
{ |
| 906 |
printf ("fastmap: "); |
| 907 |
print_fastmap (bufp->fastmap); |
| 908 |
} |
| 909 |
|
| 910 |
printf ("re_nsub: %d\t", bufp->re_nsub); |
| 911 |
printf ("regs_alloc: %d\t", bufp->regs_allocated); |
| 912 |
printf ("can_be_null: %d\t", bufp->can_be_null); |
| 913 |
printf ("newline_anchor: %d\n", bufp->newline_anchor); |
| 914 |
printf ("no_sub: %d\t", bufp->no_sub); |
| 915 |
printf ("not_bol: %d\t", bufp->not_bol); |
| 916 |
printf ("not_eol: %d\t", bufp->not_eol); |
| 917 |
printf ("syntax: %lx\n", bufp->syntax); |
| 918 |
/* Perhaps we should print the translate table? */ |
| 919 |
} |
| 920 |
|
| 921 |
|
| 922 |
void |
| 923 |
print_double_string (where, string1, size1, string2, size2) |
| 924 |
const char *where; |
| 925 |
const char *string1; |
| 926 |
const char *string2; |
| 927 |
int size1; |
| 928 |
int size2; |
| 929 |
{ |
| 930 |
int this_char; |
| 931 |
|
| 932 |
if (where == NULL) |
| 933 |
printf ("(null)"); |
| 934 |
else |
| 935 |
{ |
| 936 |
if (FIRST_STRING_P (where)) |
| 937 |
{ |
| 938 |
for (this_char = where - string1; this_char < size1; this_char++) |
| 939 |
putchar (string1[this_char]); |
| 940 |
|
| 941 |
where = string2; |
| 942 |
} |
| 943 |
|
| 944 |
for (this_char = where - string2; this_char < size2; this_char++) |
| 945 |
putchar (string2[this_char]); |
| 946 |
} |
| 947 |
} |
| 948 |
|
| 949 |
void |
| 950 |
printchar (c) |
| 951 |
int c; |
| 952 |
{ |
| 953 |
putc (c, stderr); |
| 954 |
} |
| 955 |
|
| 956 |
#else /* not DEBUG */ |
| 957 |
|
| 958 |
# undef assert |
| 959 |
# define assert(e) |
| 960 |
|
| 961 |
# define DEBUG_STATEMENT(e) |
| 962 |
# define DEBUG_PRINT1(x) |
| 963 |
# define DEBUG_PRINT2(x1, x2) |
| 964 |
# define DEBUG_PRINT3(x1, x2, x3) |
| 965 |
# define DEBUG_PRINT4(x1, x2, x3, x4) |
| 966 |
# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) |
| 967 |
# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) |
| 968 |
|
| 969 |
#endif /* not DEBUG */ |
| 970 |
|
| 971 |
/* Set by `re_set_syntax' to the current regexp syntax to recognize. Can |
| 972 |
also be assigned to arbitrarily: each pattern buffer stores its own |
| 973 |
syntax, so it can be changed between regex compilations. */ |
| 974 |
/* This has no initializer because initialized variables in Emacs |
| 975 |
become read-only after dumping. */ |
| 976 |
reg_syntax_t re_syntax_options; |
| 977 |
|
| 978 |
|
| 979 |
/* Specify the precise syntax of regexps for compilation. This provides |
| 980 |
for compatibility for various utilities which historically have |
| 981 |
different, incompatible syntaxes. |
| 982 |
|
| 983 |
The argument SYNTAX is a bit mask comprised of the various bits |
| 984 |
defined in regex.h. We return the old syntax. */ |
| 985 |
|
| 986 |
reg_syntax_t |
| 987 |
re_set_syntax (syntax) |
| 988 |
reg_syntax_t syntax; |
| 989 |
{ |
| 990 |
reg_syntax_t ret = re_syntax_options; |
| 991 |
|
| 992 |
re_syntax_options = syntax; |
| 993 |
#ifdef DEBUG |
| 994 |
if (syntax & RE_DEBUG) |
| 995 |
debug = 1; |
| 996 |
else if (debug) /* was on but now is not */ |
| 997 |
debug = 0; |
| 998 |
#endif /* DEBUG */ |
| 999 |
return ret; |
| 1000 |
} |
| 1001 |
#ifdef _LIBC |
| 1002 |
weak_alias (__re_set_syntax, re_set_syntax) |
| 1003 |
#endif |
| 1004 |
|
| 1005 |
/* This table gives an error message for each of the error codes listed |
| 1006 |
in regex.h. Obviously the order here has to be same as there. |
| 1007 |
POSIX doesn't require that we do anything for REG_NOERROR, |
| 1008 |
but why not be nice? */ |
| 1009 |
|
| 1010 |
static const char *re_error_msgid[] = |
| 1011 |
{ |
| 1012 |
gettext_noop ("Success"), /* REG_NOERROR */ |
| 1013 |
gettext_noop ("No match"), /* REG_NOMATCH */ |
| 1014 |
gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ |
| 1015 |
gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ |
| 1016 |
gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ |
| 1017 |
gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ |
| 1018 |
gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ |
| 1019 |
gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ |
| 1020 |
gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ |
| 1021 |
gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ |
| 1022 |
gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ |
| 1023 |
gettext_noop ("Invalid range end"), /* REG_ERANGE */ |
| 1024 |
gettext_noop ("Memory exhausted"), /* REG_ESPACE */ |
| 1025 |
gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ |
| 1026 |
gettext_noop ("Premature end of regular expression"), /* REG_EEND */ |
| 1027 |
gettext_noop ("Regular expression too big"), /* REG_ESIZE */ |
| 1028 |
gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */ |
| 1029 |
}; |
| 1030 |
|
| 1031 |
/* Avoiding alloca during matching, to placate r_alloc. */ |
| 1032 |
|
| 1033 |
/* Define MATCH_MAY_ALLOCATE unless we need to make sure that the |
| 1034 |
searching and matching functions should not call alloca. On some |
| 1035 |
systems, alloca is implemented in terms of malloc, and if we're |
| 1036 |
using the relocating allocator routines, then malloc could cause a |
| 1037 |
relocation, which might (if the strings being searched are in the |
| 1038 |
ralloc heap) shift the data out from underneath the regexp |
| 1039 |
routines. |
| 1040 |
|
| 1041 |
Here's another reason to avoid allocation: Emacs |
| 1042 |
processes input from X in a signal handler; processing X input may |
| 1043 |
call malloc; if input arrives while a matching routine is calling |
| 1044 |
malloc, then we're scrod. But Emacs can't just block input while |
| 1045 |
calling matching routines; then we don't notice interrupts when |
| 1046 |
they come in. So, Emacs blocks input around all regexp calls |
| 1047 |
except the matching calls, which it leaves unprotected, in the |
| 1048 |
faith that they will not malloc. */ |
| 1049 |
|
| 1050 |
/* Normally, this is fine. */ |
| 1051 |
#define MATCH_MAY_ALLOCATE |
| 1052 |
|
| 1053 |
/* When using GNU C, we are not REALLY using the C alloca, no matter |
| 1054 |
what config.h may say. So don't take precautions for it. */ |
| 1055 |
#ifdef __GNUC__ |
| 1056 |
# undef C_ALLOCA |
| 1057 |
#endif |
| 1058 |
|
| 1059 |
/* The match routines may not allocate if (1) they would do it with malloc |
| 1060 |
and (2) it's not safe for them to use malloc. |
| 1061 |
Note that if REL_ALLOC is defined, matching would not use malloc for the |
| 1062 |
failure stack, but we would still use it for the register vectors; |
| 1063 |
so REL_ALLOC should not affect this. */ |
| 1064 |
#if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs |
| 1065 |
# undef MATCH_MAY_ALLOCATE |
| 1066 |
#endif |
| 1067 |
|
| 1068 |
|
| 1069 |
/* Failure stack declarations and macros; both re_compile_fastmap and |
| 1070 |
re_match_2 use a failure stack. These have to be macros because of |
| 1071 |
REGEX_ALLOCATE_STACK. */ |
| 1072 |
|
| 1073 |
|
| 1074 |
/* Number of failure points for which to initially allocate space |
| 1075 |
when matching. If this number is exceeded, we allocate more |
| 1076 |
space, so it is not a hard limit. */ |
| 1077 |
#ifndef INIT_FAILURE_ALLOC |
| 1078 |
# define INIT_FAILURE_ALLOC 5 |
| 1079 |
#endif |
| 1080 |
|
| 1081 |
/* Roughly the maximum number of failure points on the stack. Would be |
| 1082 |
exactly that if always used MAX_FAILURE_ITEMS items each time we failed. |
| 1083 |
This is a variable only so users of regex can assign to it; we never |
| 1084 |
change it ourselves. */ |
| 1085 |
|
| 1086 |
#ifdef INT_IS_16BIT |
| 1087 |
|
| 1088 |
# if defined MATCH_MAY_ALLOCATE |
| 1089 |
/* 4400 was enough to cause a crash on Alpha OSF/1, |
| 1090 |
whose default stack limit is 2mb. */ |
| 1091 |
long int re_max_failures = 4000; |
| 1092 |
# else |
| 1093 |
long int re_max_failures = 2000; |
| 1094 |
# endif |
| 1095 |
|
| 1096 |
union fail_stack_elt |
| 1097 |
{ |
| 1098 |
unsigned char *pointer; |
| 1099 |
long int integer; |
| 1100 |
}; |
| 1101 |
|
| 1102 |
typedef union fail_stack_elt fail_stack_elt_t; |
| 1103 |
|
| 1104 |
typedef struct |
| 1105 |
{ |
| 1106 |
fail_stack_elt_t *stack; |
| 1107 |
unsigned long int size; |
| 1108 |
unsigned long int avail; /* Offset of next open position. */ |
| 1109 |
} fail_stack_type; |
| 1110 |
|
| 1111 |
#else /* not INT_IS_16BIT */ |
| 1112 |
|
| 1113 |
# if defined MATCH_MAY_ALLOCATE |
| 1114 |
/* 4400 was enough to cause a crash on Alpha OSF/1, |
| 1115 |
whose default stack limit is 2mb. */ |
| 1116 |
int re_max_failures = 20000; |
| 1117 |
# else |
| 1118 |
int re_max_failures = 2000; |
| 1119 |
# endif |
| 1120 |
|
| 1121 |
union fail_stack_elt |
| 1122 |
{ |
| 1123 |
unsigned char *pointer; |
| 1124 |
int integer; |
| 1125 |
}; |
| 1126 |
|
| 1127 |
typedef union fail_stack_elt fail_stack_elt_t; |
| 1128 |
|
| 1129 |
typedef struct |
| 1130 |
{ |
| 1131 |
fail_stack_elt_t *stack; |
| 1132 |
unsigned size; |
| 1133 |
unsigned avail; /* Offset of next open position. */ |
| 1134 |
} fail_stack_type; |
| 1135 |
|
| 1136 |
#endif /* INT_IS_16BIT */ |
| 1137 |
|
| 1138 |
#define FAIL_STACK_EMPTY() (fail_stack.avail == 0) |
| 1139 |
#define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) |
| 1140 |
#define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) |
| 1141 |
|
| 1142 |
|
| 1143 |
/* Define macros to initialize and free the failure stack. |
| 1144 |
Do `return -2' if the alloc fails. */ |
| 1145 |
|
| 1146 |
#ifdef MATCH_MAY_ALLOCATE |
| 1147 |
# define INIT_FAIL_STACK() \ |
| 1148 |
do { \ |
| 1149 |
fail_stack.stack = (fail_stack_elt_t *) \ |
| 1150 |
REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \ |
| 1151 |
\ |
| 1152 |
if (fail_stack.stack == NULL) \ |
| 1153 |
return -2; \ |
| 1154 |
\ |
| 1155 |
fail_stack.size = INIT_FAILURE_ALLOC; \ |
| 1156 |
fail_stack.avail = 0; \ |
| 1157 |
} while (0) |
| 1158 |
|
| 1159 |
# define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) |
| 1160 |
#else |
| 1161 |
# define INIT_FAIL_STACK() \ |
| 1162 |
do { \ |
| 1163 |
fail_stack.avail = 0; \ |
| 1164 |
} while (0) |
| 1165 |
|
| 1166 |
# define RESET_FAIL_STACK() |
| 1167 |
#endif |
| 1168 |
|
| 1169 |
|
| 1170 |
/* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. |
| 1171 |
|
| 1172 |
Return 1 if succeeds, and 0 if either ran out of memory |
| 1173 |
allocating space for it or it was already too large. |
| 1174 |
|
| 1175 |
REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
| 1176 |
|
| 1177 |
#define DOUBLE_FAIL_STACK(fail_stack) \ |
| 1178 |
((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \ |
| 1179 |
? 0 \ |
| 1180 |
: ((fail_stack).stack = (fail_stack_elt_t *) \ |
| 1181 |
REGEX_REALLOCATE_STACK ((fail_stack).stack, \ |
| 1182 |
(fail_stack).size * sizeof (fail_stack_elt_t), \ |
| 1183 |
((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \ |
| 1184 |
\ |
| 1185 |
(fail_stack).stack == NULL \ |
| 1186 |
? 0 \ |
| 1187 |
: ((fail_stack).size <<= 1, \ |
| 1188 |
1))) |
| 1189 |
|
| 1190 |
|
| 1191 |
/* Push pointer POINTER on FAIL_STACK. |
| 1192 |
Return 1 if was able to do so and 0 if ran out of memory allocating |
| 1193 |
space to do so. */ |
| 1194 |
#define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ |
| 1195 |
((FAIL_STACK_FULL () \ |
| 1196 |
&& !DOUBLE_FAIL_STACK (FAIL_STACK)) \ |
| 1197 |
? 0 \ |
| 1198 |
: ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ |
| 1199 |
1)) |
| 1200 |
|
| 1201 |
/* Push a pointer value onto the failure stack. |
| 1202 |
Assumes the variable `fail_stack'. Probably should only |
| 1203 |
be called from within `PUSH_FAILURE_POINT'. */ |
| 1204 |
#define PUSH_FAILURE_POINTER(item) \ |
| 1205 |
fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item) |
| 1206 |
|
| 1207 |
/* This pushes an integer-valued item onto the failure stack. |
| 1208 |
Assumes the variable `fail_stack'. Probably should only |
| 1209 |
be called from within `PUSH_FAILURE_POINT'. */ |
| 1210 |
#define PUSH_FAILURE_INT(item) \ |
| 1211 |
fail_stack.stack[fail_stack.avail++].integer = (item) |
| 1212 |
|
| 1213 |
/* Push a fail_stack_elt_t value onto the failure stack. |
| 1214 |
Assumes the variable `fail_stack'. Probably should only |
| 1215 |
be called from within `PUSH_FAILURE_POINT'. */ |
| 1216 |
#define PUSH_FAILURE_ELT(item) \ |
| 1217 |
fail_stack.stack[fail_stack.avail++] = (item) |
| 1218 |
|
| 1219 |
/* These three POP... operations complement the three PUSH... operations. |
| 1220 |
All assume that `fail_stack' is nonempty. */ |
| 1221 |
#define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer |
| 1222 |
#define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer |
| 1223 |
#define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] |
| 1224 |
|
| 1225 |
/* Used to omit pushing failure point id's when we're not debugging. */ |
| 1226 |
#ifdef DEBUG |
| 1227 |
# define DEBUG_PUSH PUSH_FAILURE_INT |
| 1228 |
# define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () |
| 1229 |
#else |
| 1230 |
# define DEBUG_PUSH(item) |
| 1231 |
# define DEBUG_POP(item_addr) |
| 1232 |
#endif |
| 1233 |
|
| 1234 |
|
| 1235 |
/* Push the information about the state we will need |
| 1236 |
if we ever fail back to it. |
| 1237 |
|
| 1238 |
Requires variables fail_stack, regstart, regend, reg_info, and |
| 1239 |
num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination' |
| 1240 |
be declared. |
| 1241 |
|
| 1242 |
Does `return FAILURE_CODE' if runs out of memory. */ |
| 1243 |
|
| 1244 |
#define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ |
| 1245 |
do { \ |
| 1246 |
char *destination; \ |
| 1247 |
/* Must be int, so when we don't save any registers, the arithmetic \ |
| 1248 |
of 0 + -1 isn't done as unsigned. */ \ |
| 1249 |
/* Can't be int, since there is not a shred of a guarantee that int \ |
| 1250 |
is wide enough to hold a value of something to which pointer can \ |
| 1251 |
be assigned */ \ |
| 1252 |
active_reg_t this_reg; \ |
| 1253 |
\ |
| 1254 |
DEBUG_STATEMENT (failure_id++); \ |
| 1255 |
DEBUG_STATEMENT (nfailure_points_pushed++); \ |
| 1256 |
DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ |
| 1257 |
DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ |
| 1258 |
DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ |
| 1259 |
\ |
| 1260 |
DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \ |
| 1261 |
DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ |
| 1262 |
\ |
| 1263 |
/* Ensure we have enough space allocated for what we will push. */ \ |
| 1264 |
while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ |
| 1265 |
{ \ |
| 1266 |
if (!DOUBLE_FAIL_STACK (fail_stack)) \ |
| 1267 |
return failure_code; \ |
| 1268 |
\ |
| 1269 |
DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ |
| 1270 |
(fail_stack).size); \ |
| 1271 |
DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ |
| 1272 |
} \ |
| 1273 |
\ |
| 1274 |
/* Push the info, starting with the registers. */ \ |
| 1275 |
DEBUG_PRINT1 ("\n"); \ |
| 1276 |
\ |
| 1277 |
if (1) \ |
| 1278 |
for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ |
| 1279 |
this_reg++) \ |
| 1280 |
{ \ |
| 1281 |
DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \ |
| 1282 |
DEBUG_STATEMENT (num_regs_pushed++); \ |
| 1283 |
\ |
| 1284 |
DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ |
| 1285 |
PUSH_FAILURE_POINTER (regstart[this_reg]); \ |
| 1286 |
\ |
| 1287 |
DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ |
| 1288 |
PUSH_FAILURE_POINTER (regend[this_reg]); \ |
| 1289 |
\ |
| 1290 |
DEBUG_PRINT2 (" info: %p\n ", \ |
| 1291 |
reg_info[this_reg].word.pointer); \ |
| 1292 |
DEBUG_PRINT2 (" match_null=%d", \ |
| 1293 |
REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ |
| 1294 |
DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ |
| 1295 |
DEBUG_PRINT2 (" matched_something=%d", \ |
| 1296 |
MATCHED_SOMETHING (reg_info[this_reg])); \ |
| 1297 |
DEBUG_PRINT2 (" ever_matched=%d", \ |
| 1298 |
EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ |
| 1299 |
DEBUG_PRINT1 ("\n"); \ |
| 1300 |
PUSH_FAILURE_ELT (reg_info[this_reg].word); \ |
| 1301 |
} \ |
| 1302 |
\ |
| 1303 |
DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\ |
| 1304 |
PUSH_FAILURE_INT (lowest_active_reg); \ |
| 1305 |
\ |
| 1306 |
DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\ |
| 1307 |
PUSH_FAILURE_INT (highest_active_reg); \ |
| 1308 |
\ |
| 1309 |
DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \ |
| 1310 |
DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ |
| 1311 |
PUSH_FAILURE_POINTER (pattern_place); \ |
| 1312 |
\ |
| 1313 |
DEBUG_PRINT2 (" Pushing string %p: `", string_place); \ |
| 1314 |
DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ |
| 1315 |
size2); \ |
| 1316 |
DEBUG_PRINT1 ("'\n"); \ |
| 1317 |
PUSH_FAILURE_POINTER (string_place); \ |
| 1318 |
\ |
| 1319 |
DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ |
| 1320 |
DEBUG_PUSH (failure_id); \ |
| 1321 |
} while (0) |
| 1322 |
|
| 1323 |
/* This is the number of items that are pushed and popped on the stack |
| 1324 |
for each register. */ |
| 1325 |
#define NUM_REG_ITEMS 3 |
| 1326 |
|
| 1327 |
/* Individual items aside from the registers. */ |
| 1328 |
#ifdef DEBUG |
| 1329 |
# define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ |
| 1330 |
#else |
| 1331 |
# define NUM_NONREG_ITEMS 4 |
| 1332 |
#endif |
| 1333 |
|
| 1334 |
/* We push at most this many items on the stack. */ |
| 1335 |
/* We used to use (num_regs - 1), which is the number of registers |
| 1336 |
this regexp will save; but that was changed to 5 |
| 1337 |
to avoid stack overflow for a regexp with lots of parens. */ |
| 1338 |
#define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS) |
| 1339 |
|
| 1340 |
/* We actually push this many items. */ |
| 1341 |
#define NUM_FAILURE_ITEMS \ |
| 1342 |
(((0 \ |
| 1343 |
? 0 : highest_active_reg - lowest_active_reg + 1) \ |
| 1344 |
* NUM_REG_ITEMS) \ |
| 1345 |
+ NUM_NONREG_ITEMS) |
| 1346 |
|
| 1347 |
/* How many items can still be added to the stack without overflowing it. */ |
| 1348 |
#define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) |
| 1349 |
|
| 1350 |
|
| 1351 |
/* Pops what PUSH_FAIL_STACK pushes. |
| 1352 |
|
| 1353 |
We restore into the parameters, all of which should be lvalues: |
| 1354 |
STR -- the saved data position. |
| 1355 |
PAT -- the saved pattern position. |
| 1356 |
LOW_REG, HIGH_REG -- the highest and lowest active registers. |
| 1357 |
REGSTART, REGEND -- arrays of string positions. |
| 1358 |
REG_INFO -- array of information about each subexpression. |
| 1359 |
|
| 1360 |
Also assumes the variables `fail_stack' and (if debugging), `bufp', |
| 1361 |
`pend', `string1', `size1', `string2', and `size2'. */ |
| 1362 |
|
| 1363 |
#define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ |
| 1364 |
{ \ |
| 1365 |
DEBUG_STATEMENT (unsigned failure_id;) \ |
| 1366 |
active_reg_t this_reg; \ |
| 1367 |
const unsigned char *string_temp; \ |
| 1368 |
\ |
| 1369 |
assert (!FAIL_STACK_EMPTY ()); \ |
| 1370 |
\ |
| 1371 |
/* Remove failure points and point to how many regs pushed. */ \ |
| 1372 |
DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ |
| 1373 |
DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ |
| 1374 |
DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ |
| 1375 |
\ |
| 1376 |
assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ |
| 1377 |
\ |
| 1378 |
DEBUG_POP (&failure_id); \ |
| 1379 |
DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ |
| 1380 |
\ |
| 1381 |
/* If the saved string location is NULL, it came from an \ |
| 1382 |
on_failure_keep_string_jump opcode, and we want to throw away the \ |
| 1383 |
saved NULL, thus retaining our current position in the string. */ \ |
| 1384 |
string_temp = POP_FAILURE_POINTER (); \ |
| 1385 |
if (string_temp != NULL) \ |
| 1386 |
str = (const char *) string_temp; \ |
| 1387 |
\ |
| 1388 |
DEBUG_PRINT2 (" Popping string %p: `", str); \ |
| 1389 |
DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
| 1390 |
DEBUG_PRINT1 ("'\n"); \ |
| 1391 |
\ |
| 1392 |
pat = (unsigned char *) POP_FAILURE_POINTER (); \ |
| 1393 |
DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \ |
| 1394 |
DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ |
| 1395 |
\ |
| 1396 |
/* Restore register info. */ \ |
| 1397 |
high_reg = (active_reg_t) POP_FAILURE_INT (); \ |
| 1398 |
DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \ |
| 1399 |
\ |
| 1400 |
low_reg = (active_reg_t) POP_FAILURE_INT (); \ |
| 1401 |
DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \ |
| 1402 |
\ |
| 1403 |
if (1) \ |
| 1404 |
for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ |
| 1405 |
{ \ |
| 1406 |
DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \ |
| 1407 |
\ |
| 1408 |
reg_info[this_reg].word = POP_FAILURE_ELT (); \ |
| 1409 |
DEBUG_PRINT2 (" info: %p\n", \ |
| 1410 |
reg_info[this_reg].word.pointer); \ |
| 1411 |
\ |
| 1412 |
regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \ |
| 1413 |
DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ |
| 1414 |
\ |
| 1415 |
regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \ |
| 1416 |
DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ |
| 1417 |
} \ |
| 1418 |
else \ |
| 1419 |
{ \ |
| 1420 |
for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \ |
| 1421 |
{ \ |
| 1422 |
reg_info[this_reg].word.integer = 0; \ |
| 1423 |
regend[this_reg] = 0; \ |
| 1424 |
regstart[this_reg] = 0; \ |
| 1425 |
} \ |
| 1426 |
highest_active_reg = high_reg; \ |
| 1427 |
} \ |
| 1428 |
\ |
| 1429 |
set_regs_matched_done = 0; \ |
| 1430 |
DEBUG_STATEMENT (nfailure_points_popped++); \ |
| 1431 |
} /* POP_FAILURE_POINT */ |
| 1432 |
|
| 1433 |
|
| 1434 |
|
| 1435 |
/* Structure for per-register (a.k.a. per-group) information. |
| 1436 |
Other register information, such as the |
| 1437 |
starting and ending positions (which are addresses), and the list of |
| 1438 |
inner groups (which is a bits list) are maintained in separate |
| 1439 |
variables. |
| 1440 |
|
| 1441 |
We are making a (strictly speaking) nonportable assumption here: that |
| 1442 |
the compiler will pack our bit fields into something that fits into |
| 1443 |
the type of `word', i.e., is something that fits into one item on the |
| 1444 |
failure stack. */ |
| 1445 |
|
| 1446 |
|
| 1447 |
/* Declarations and macros for re_match_2. */ |
| 1448 |
|
| 1449 |
typedef union |
| 1450 |
{ |
| 1451 |
fail_stack_elt_t word; |
| 1452 |
struct |
| 1453 |
{ |
| 1454 |
/* This field is one if this group can match the empty string, |
| 1455 |
zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ |
| 1456 |
#define MATCH_NULL_UNSET_VALUE 3 |
| 1457 |
unsigned match_null_string_p : 2; |
| 1458 |
unsigned is_active : 1; |
| 1459 |
unsigned matched_something : 1; |
| 1460 |
unsigned ever_matched_something : 1; |
| 1461 |
} bits; |
| 1462 |
} register_info_type; |
| 1463 |
|
| 1464 |
#define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) |
| 1465 |
#define IS_ACTIVE(R) ((R).bits.is_active) |
| 1466 |
#define MATCHED_SOMETHING(R) ((R).bits.matched_something) |
| 1467 |
#define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) |
| 1468 |
|
| 1469 |
|
| 1470 |
/* Call this when have matched a real character; it sets `matched' flags |
| 1471 |
for the subexpressions which we are currently inside. Also records |
| 1472 |
that those subexprs have matched. */ |
| 1473 |
#define SET_REGS_MATCHED() \ |
| 1474 |
do \ |
| 1475 |
{ \ |
| 1476 |
if (!set_regs_matched_done) \ |
| 1477 |
{ \ |
| 1478 |
active_reg_t r; \ |
| 1479 |
set_regs_matched_done = 1; \ |
| 1480 |
for (r = lowest_active_reg; r <= highest_active_reg; r++) \ |
| 1481 |
{ \ |
| 1482 |
MATCHED_SOMETHING (reg_info[r]) \ |
| 1483 |
= EVER_MATCHED_SOMETHING (reg_info[r]) \ |
| 1484 |
= 1; \ |
| 1485 |
} \ |
| 1486 |
} \ |
| 1487 |
} \ |
| 1488 |
while (0) |
| 1489 |
|
| 1490 |
/* Registers are set to a sentinel when they haven't yet matched. */ |
| 1491 |
static char reg_unset_dummy; |
| 1492 |
#define REG_UNSET_VALUE (®_unset_dummy) |
| 1493 |
#define REG_UNSET(e) ((e) == REG_UNSET_VALUE) |
| 1494 |
|
| 1495 |
/* Subroutine declarations and macros for regex_compile. */ |
| 1496 |
|
| 1497 |
static reg_errcode_t regex_compile _RE_ARGS ((const char *pattern, size_t size, |
| 1498 |
reg_syntax_t syntax, |
| 1499 |
struct re_pattern_buffer *bufp)); |
| 1500 |
static void store_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, int arg)); |
| 1501 |
static void store_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, |
| 1502 |
int arg1, int arg2)); |
| 1503 |
static void insert_op1 _RE_ARGS ((re_opcode_t op, unsigned char *loc, |
| 1504 |
int arg, unsigned char *end)); |
| 1505 |
static void insert_op2 _RE_ARGS ((re_opcode_t op, unsigned char *loc, |
| 1506 |
int arg1, int arg2, unsigned char *end)); |
| 1507 |
static boolean at_begline_loc_p _RE_ARGS ((const char *pattern, const char *p, |
| 1508 |
reg_syntax_t syntax)); |
| 1509 |
static boolean at_endline_loc_p _RE_ARGS ((const char *p, const char *pend, |
| 1510 |
reg_syntax_t syntax)); |
| 1511 |
static reg_errcode_t compile_range _RE_ARGS ((const char **p_ptr, |
| 1512 |
const char *pend, |
| 1513 |
char *translate, |
| 1514 |
reg_syntax_t syntax, |
| 1515 |
unsigned char *b)); |
| 1516 |
|
| 1517 |
/* Fetch the next character in the uncompiled pattern---translating it |
| 1518 |
if necessary. Also cast from a signed character in the constant |
| 1519 |
string passed to us by the user to an unsigned char that we can use |
| 1520 |
as an array index (in, e.g., `translate'). */ |
| 1521 |
#ifndef PATFETCH |
| 1522 |
# define PATFETCH(c) \ |
| 1523 |
do {if (p == pend) return REG_EEND; \ |
| 1524 |
c = (unsigned char) *p++; \ |
| 1525 |
if (translate) c = (unsigned char) translate[c]; \ |
| 1526 |
} while (0) |
| 1527 |
#endif |
| 1528 |
|
| 1529 |
/* Fetch the next character in the uncompiled pattern, with no |
| 1530 |
translation. */ |
| 1531 |
#define PATFETCH_RAW(c) \ |
| 1532 |
do {if (p == pend) return REG_EEND; \ |
| 1533 |
c = (unsigned char) *p++; \ |
| 1534 |
} while (0) |
| 1535 |
|
| 1536 |
/* Go backwards one character in the pattern. */ |
| 1537 |
#define PATUNFETCH p-- |
| 1538 |
|
| 1539 |
|
| 1540 |
/* If `translate' is non-null, return translate[D], else just D. We |
| 1541 |
cast the subscript to translate because some data is declared as |
| 1542 |
`char *', to avoid warnings when a string constant is passed. But |
| 1543 |
when we use a character as a subscript we must make it unsigned. */ |
| 1544 |
#ifndef TRANSLATE |
| 1545 |
# define TRANSLATE(d) \ |
| 1546 |
(translate ? (char) translate[(unsigned char) (d)] : (d)) |
| 1547 |
#endif |
| 1548 |
|
| 1549 |
|
| 1550 |
/* Macros for outputting the compiled pattern into `buffer'. */ |
| 1551 |
|
| 1552 |
/* If the buffer isn't allocated when it comes in, use this. */ |
| 1553 |
#define INIT_BUF_SIZE 32 |
| 1554 |
|
| 1555 |
/* Make sure we have at least N more bytes of space in buffer. */ |
| 1556 |
#define GET_BUFFER_SPACE(n) \ |
| 1557 |
while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \ |
| 1558 |
EXTEND_BUFFER () |
| 1559 |
|
| 1560 |
/* Make sure we have one more byte of buffer space and then add C to it. */ |
| 1561 |
#define BUF_PUSH(c) \ |
| 1562 |
do { \ |
| 1563 |
GET_BUFFER_SPACE (1); \ |
| 1564 |
*b++ = (unsigned char) (c); \ |
| 1565 |
} while (0) |
| 1566 |
|
| 1567 |
|
| 1568 |
/* Ensure we have two more bytes of buffer space and then append C1 and C2. */ |
| 1569 |
#define BUF_PUSH_2(c1, c2) \ |
| 1570 |
do { \ |
| 1571 |
GET_BUFFER_SPACE (2); \ |
| 1572 |
*b++ = (unsigned char) (c1); \ |
| 1573 |
*b++ = (unsigned char) (c2); \ |
| 1574 |
} while (0) |
| 1575 |
|
| 1576 |
|
| 1577 |
/* As with BUF_PUSH_2, except for three bytes. */ |
| 1578 |
#define BUF_PUSH_3(c1, c2, c3) \ |
| 1579 |
do { \ |
| 1580 |
GET_BUFFER_SPACE (3); \ |
| 1581 |
*b++ = (unsigned char) (c1); \ |
| 1582 |
*b++ = (unsigned char) (c2); \ |
| 1583 |
*b++ = (unsigned char) (c3); \ |
| 1584 |
} while (0) |
| 1585 |
|
| 1586 |
|
| 1587 |
/* Store a jump with opcode OP at LOC to location TO. We store a |
| 1588 |
relative address offset by the three bytes the jump itself occupies. */ |
| 1589 |
#define STORE_JUMP(op, loc, to) \ |
| 1590 |
store_op1 (op, loc, (int) ((to) - (loc) - 3)) |
| 1591 |
|
| 1592 |
/* Likewise, for a two-argument jump. */ |
| 1593 |
#define STORE_JUMP2(op, loc, to, arg) \ |
| 1594 |
store_op2 (op, loc, (int) ((to) - (loc) - 3), arg) |
| 1595 |
|
| 1596 |
/* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
| 1597 |
#define INSERT_JUMP(op, loc, to) \ |
| 1598 |
insert_op1 (op, loc, (int) ((to) - (loc) - 3), b) |
| 1599 |
|
| 1600 |
/* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ |
| 1601 |
#define INSERT_JUMP2(op, loc, to, arg) \ |
| 1602 |
insert_op2 (op, loc, (int) ((to) - (loc) - 3), arg, b) |
| 1603 |
|
| 1604 |
|
| 1605 |
/* This is not an arbitrary limit: the arguments which represent offsets |
| 1606 |
into the pattern are two bytes long. So if 2^16 bytes turns out to |
| 1607 |
be too small, many things would have to change. */ |
| 1608 |
/* Any other compiler which, like MSC, has allocation limit below 2^16 |
| 1609 |
bytes will have to use approach similar to what was done below for |
| 1610 |
MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up |
| 1611 |
reallocating to 0 bytes. Such thing is not going to work too well. |
| 1612 |
You have been warned!! */ |
| 1613 |
#if defined _MSC_VER && !defined WIN32 |
| 1614 |
/* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. |
| 1615 |
The REALLOC define eliminates a flurry of conversion warnings, |
| 1616 |
but is not required. */ |
| 1617 |
# define MAX_BUF_SIZE 65500L |
| 1618 |
# define REALLOC(p,s) realloc ((p), (size_t) (s)) |
| 1619 |
#else |
| 1620 |
# define MAX_BUF_SIZE (1L << 16) |
| 1621 |
# define REALLOC(p,s) realloc ((p), (s)) |
| 1622 |
#endif |
| 1623 |
|
| 1624 |
/* Extend the buffer by twice its current size via realloc and |
| 1625 |
reset the pointers that pointed into the old block to point to the |
| 1626 |
correct places in the new one. If extending the buffer results in it |
| 1627 |
being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
| 1628 |
#define EXTEND_BUFFER() \ |
| 1629 |
do { \ |
| 1630 |
unsigned char *old_buffer = bufp->buffer; \ |
| 1631 |
if (bufp->allocated == MAX_BUF_SIZE) \ |
| 1632 |
return REG_ESIZE; \ |
| 1633 |
bufp->allocated <<= 1; \ |
| 1634 |
if (bufp->allocated > MAX_BUF_SIZE) \ |
| 1635 |
bufp->allocated = MAX_BUF_SIZE; \ |
| 1636 |
bufp->buffer = (unsigned char *) REALLOC (bufp->buffer, bufp->allocated);\ |
| 1637 |
if (bufp->buffer == NULL) \ |
| 1638 |
return REG_ESPACE; \ |
| 1639 |
/* If the buffer moved, move all the pointers into it. */ \ |
| 1640 |
if (old_buffer != bufp->buffer) \ |
| 1641 |
{ \ |
| 1642 |
b = (b - old_buffer) + bufp->buffer; \ |
| 1643 |
begalt = (begalt - old_buffer) + bufp->buffer; \ |
| 1644 |
if (fixup_alt_jump) \ |
| 1645 |
fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\ |
| 1646 |
if (laststart) \ |
| 1647 |
laststart = (laststart - old_buffer) + bufp->buffer; \ |
| 1648 |
if (pending_exact) \ |
| 1649 |
pending_exact = (pending_exact - old_buffer) + bufp->buffer; \ |
| 1650 |
} \ |
| 1651 |
} while (0) |
| 1652 |
|
| 1653 |
|
| 1654 |
/* Since we have one byte reserved for the register number argument to |
| 1655 |
{start,stop}_memory, the maximum number of groups we can report |
| 1656 |
things about is what fits in that byte. */ |
| 1657 |
#define MAX_REGNUM 255 |
| 1658 |
|
| 1659 |
/* But patterns can have more than `MAX_REGNUM' registers. We just |
| 1660 |
ignore the excess. */ |
| 1661 |
typedef unsigned regnum_t; |
| 1662 |
|
| 1663 |
|
| 1664 |
/* Macros for the compile stack. */ |
| 1665 |
|
| 1666 |
/* Since offsets can go either forwards or backwards, this type needs to |
| 1667 |
be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
| 1668 |
/* int may be not enough when sizeof(int) == 2. */ |
| 1669 |
typedef long pattern_offset_t; |
| 1670 |
|
| 1671 |
typedef struct |
| 1672 |
{ |
| 1673 |
pattern_offset_t begalt_offset; |
| 1674 |
pattern_offset_t fixup_alt_jump; |
| 1675 |
pattern_offset_t inner_group_offset; |
| 1676 |
pattern_offset_t laststart_offset; |
| 1677 |
regnum_t regnum; |
| 1678 |
} compile_stack_elt_t; |
| 1679 |
|
| 1680 |
|
| 1681 |
typedef struct |
| 1682 |
{ |
| 1683 |
compile_stack_elt_t *stack; |
| 1684 |
unsigned size; |
| 1685 |
unsigned avail; /* Offset of next open position. */ |
| 1686 |
} compile_stack_type; |
| 1687 |
|
| 1688 |
|
| 1689 |
#define INIT_COMPILE_STACK_SIZE 32 |
| 1690 |
|
| 1691 |
#define COMPILE_STACK_EMPTY (compile_stack.avail == 0) |
| 1692 |
#define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) |
| 1693 |
|
| 1694 |
/* The next available element. */ |
| 1695 |
#define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
| 1696 |
|
| 1697 |
|
| 1698 |
/* Set the bit for character C in a list. */ |
| 1699 |
#define SET_LIST_BIT(c) \ |
| 1700 |
(b[((unsigned char) (c)) / BYTEWIDTH] \ |
| 1701 |
|= 1 << (((unsigned char) c) % BYTEWIDTH)) |
| 1702 |
|
| 1703 |
|
| 1704 |
/* Get the next unsigned number in the uncompiled pattern. */ |
| 1705 |
#define GET_UNSIGNED_NUMBER(num) \ |
| 1706 |
{ if (p != pend) \ |
| 1707 |
{ \ |
| 1708 |
PATFETCH (c); \ |
| 1709 |
while (ISDIGIT (c)) \ |
| 1710 |
{ \ |
| 1711 |
if (num < 0) \ |
| 1712 |
num = 0; \ |
| 1713 |
num = num * 10 + c - '0'; \ |
| 1714 |
if (p == pend) \ |
| 1715 |
break; \ |
| 1716 |
PATFETCH (c); \ |
| 1717 |
} \ |
| 1718 |
} \ |
| 1719 |
} |
| 1720 |
|
| 1721 |
#if defined _LIBC || WIDE_CHAR_SUPPORT |
| 1722 |
/* The GNU C library provides support for user-defined character classes |
| 1723 |
and the functions from ISO C amendement 1. */ |
| 1724 |
# ifdef CHARCLASS_NAME_MAX |
| 1725 |
# define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX |
| 1726 |
# else |
| 1727 |
/* This shouldn't happen but some implementation might still have this |
| 1728 |
problem. Use a reasonable default value. */ |
| 1729 |
# define CHAR_CLASS_MAX_LENGTH 256 |
| 1730 |
# endif |
| 1731 |
|
| 1732 |
# ifdef _LIBC |
| 1733 |
# define IS_CHAR_CLASS(string) __wctype (string) |
| 1734 |
# else |
| 1735 |
# define IS_CHAR_CLASS(string) wctype (string) |
| 1736 |
# endif |
| 1737 |
#else |
| 1738 |
# define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ |
| 1739 |
|
| 1740 |
# define IS_CHAR_CLASS(string) \ |
| 1741 |
(STREQ (string, "alpha") || STREQ (string, "upper") \ |
| 1742 |
|| STREQ (string, "lower") || STREQ (string, "digit") \ |
| 1743 |
|| STREQ (string, "alnum") || STREQ (string, "xdigit") \ |
| 1744 |
|| STREQ (string, "space") || STREQ (string, "print") \ |
| 1745 |
|| STREQ (string, "punct") || STREQ (string, "graph") \ |
| 1746 |
|| STREQ (string, "cntrl") || STREQ (string, "blank")) |
| 1747 |
#endif |
| 1748 |
|
| 1749 |
#ifndef MATCH_MAY_ALLOCATE |
| 1750 |
|
| 1751 |
/* If we cannot allocate large objects within re_match_2_internal, |
| 1752 |
we make the fail stack and register vectors global. |
| 1753 |
The fail stack, we grow to the maximum size when a regexp |
| 1754 |
is compiled. |
| 1755 |
The register vectors, we adjust in size each time we |
| 1756 |
compile a regexp, according to the number of registers it needs. */ |
| 1757 |
|
| 1758 |
static fail_stack_type fail_stack; |
| 1759 |
|
| 1760 |
/* Size with which the following vectors are currently allocated. |
| 1761 |
That is so we can make them bigger as needed, |
| 1762 |
but never make them smaller. */ |
| 1763 |
static int regs_allocated_size; |
| 1764 |
|
| 1765 |
static const char ** regstart, ** regend; |
| 1766 |
static const char ** old_regstart, ** old_regend; |
| 1767 |
static const char **best_regstart, **best_regend; |
| 1768 |
static register_info_type *reg_info; |
| 1769 |
static const char **reg_dummy; |
| 1770 |
static register_info_type *reg_info_dummy; |
| 1771 |
|
| 1772 |
/* Make the register vectors big enough for NUM_REGS registers, |
| 1773 |
but don't make them smaller. */ |
| 1774 |
|
| 1775 |
static |
| 1776 |
regex_grow_registers (num_regs) |
| 1777 |
int num_regs; |
| 1778 |
{ |
| 1779 |
if (num_regs > regs_allocated_size) |
| 1780 |
{ |
| 1781 |
RETALLOC_IF (regstart, num_regs, const char *); |
| 1782 |
RETALLOC_IF (regend, num_regs, const char *); |
| 1783 |
RETALLOC_IF (old_regstart, num_regs, const char *); |
| 1784 |
RETALLOC_IF (old_regend, num_regs, const char *); |
| 1785 |
RETALLOC_IF (best_regstart, num_regs, const char *); |
| 1786 |
RETALLOC_IF (best_regend, num_regs, const char *); |
| 1787 |
RETALLOC_IF (reg_info, num_regs, register_info_type); |
| 1788 |
RETALLOC_IF (reg_dummy, num_regs, const char *); |
| 1789 |
RETALLOC_IF (reg_info_dummy, num_regs, register_info_type); |
| 1790 |
|
| 1791 |
regs_allocated_size = num_regs; |
| 1792 |
} |
| 1793 |
} |
| 1794 |
|
| 1795 |
#endif /* not MATCH_MAY_ALLOCATE */ |
| 1796 |
|
| 1797 |
static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type |
| 1798 |
compile_stack, |
| 1799 |
regnum_t regnum)); |
| 1800 |
|
| 1801 |
/* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
| 1802 |
Returns one of error codes defined in `regex.h', or zero for success. |
| 1803 |
|
| 1804 |
Assumes the `allocated' (and perhaps `buffer') and `translate' |
| 1805 |
fields are set in BUFP on entry. |
| 1806 |
|
| 1807 |
If it succeeds, results are put in BUFP (if it returns an error, the |
| 1808 |
contents of BUFP are undefined): |
| 1809 |
`buffer' is the compiled pattern; |
| 1810 |
`syntax' is set to SYNTAX; |
| 1811 |
`used' is set to the length of the compiled pattern; |
| 1812 |
`fastmap_accurate' is zero; |
| 1813 |
`re_nsub' is the number of subexpressions in PATTERN; |
| 1814 |
`not_bol' and `not_eol' are zero; |
| 1815 |
|
| 1816 |
The `fastmap' and `newline_anchor' fields are neither |
| 1817 |
examined nor set. */ |
| 1818 |
|
| 1819 |
/* Return, freeing storage we allocated. */ |
| 1820 |
#define FREE_STACK_RETURN(value) \ |
| 1821 |
return (free (compile_stack.stack), value) |
| 1822 |
|
| 1823 |
static reg_errcode_t |
| 1824 |
regex_compile (pattern, size, syntax, bufp) |
| 1825 |
const char *pattern; |
| 1826 |
size_t size; |
| 1827 |
reg_syntax_t syntax; |
| 1828 |
struct re_pattern_buffer *bufp; |
| 1829 |
{ |
| 1830 |
/* We fetch characters from PATTERN here. Even though PATTERN is |
| 1831 |
`char *' (i.e., signed), we declare these variables as unsigned, so |
| 1832 |
they can be reliably used as array indices. */ |
| 1833 |
register unsigned char c, c1; |
| 1834 |
|
| 1835 |
/* A random temporary spot in PATTERN. */ |
| 1836 |
const char *p1; |
| 1837 |
|
| 1838 |
/* Points to the end of the buffer, where we should append. */ |
| 1839 |
register unsigned char *b; |
| 1840 |
|
| 1841 |
/* Keeps track of unclosed groups. */ |
| 1842 |
compile_stack_type compile_stack; |
| 1843 |
|
| 1844 |
/* Points to the current (ending) position in the pattern. */ |
| 1845 |
const char *p = pattern; |
| 1846 |
const char *pend = pattern + size; |
| 1847 |
|
| 1848 |
/* How to translate the characters in the pattern. */ |
| 1849 |
RE_TRANSLATE_TYPE translate = bufp->translate; |
| 1850 |
|
| 1851 |
/* Address of the count-byte of the most recently inserted `exactn' |
| 1852 |
command. This makes it possible to tell if a new exact-match |
| 1853 |
character can be added to that command or if the character requires |
| 1854 |
a new `exactn' command. */ |
| 1855 |
unsigned char *pending_exact = 0; |
| 1856 |
|
| 1857 |
/* Address of start of the most recently finished expression. |
| 1858 |
This tells, e.g., postfix * where to find the start of its |
| 1859 |
operand. Reset at the beginning of groups and alternatives. */ |
| 1860 |
unsigned char *laststart = 0; |
| 1861 |
|
| 1862 |
/* Address of beginning of regexp, or inside of last group. */ |
| 1863 |
unsigned char *begalt; |
| 1864 |
|
| 1865 |
/* Place in the uncompiled pattern (i.e., the {) to |
| 1866 |
which to go back if the interval is invalid. */ |
| 1867 |
const char *beg_interval; |
| 1868 |
|
| 1869 |
/* Address of the place where a forward jump should go to the end of |
| 1870 |
the containing expression. Each alternative of an `or' -- except the |
| 1871 |
last -- ends with a forward jump of this sort. */ |
| 1872 |
unsigned char *fixup_alt_jump = 0; |
| 1873 |
|
| 1874 |
/* Counts open-groups as they are encountered. Remembered for the |
| 1875 |
matching close-group on the compile stack, so the same register |
| 1876 |
number is put in the stop_memory as the start_memory. */ |
| 1877 |
regnum_t regnum = 0; |
| 1878 |
|
| 1879 |
#ifdef DEBUG |
| 1880 |
DEBUG_PRINT1 ("\nCompiling pattern: "); |
| 1881 |
if (debug) |
| 1882 |
{ |
| 1883 |
unsigned debug_count; |
| 1884 |
|
| 1885 |
for (debug_count = 0; debug_count < size; debug_count++) |
| 1886 |
putchar (pattern[debug_count]); |
| 1887 |
putchar ('\n'); |
| 1888 |
} |
| 1889 |
#endif /* DEBUG */ |
| 1890 |
|
| 1891 |
/* Initialize the compile stack. */ |
| 1892 |
compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); |
| 1893 |
if (compile_stack.stack == NULL) |
| 1894 |
return REG_ESPACE; |
| 1895 |
|
| 1896 |
compile_stack.size = INIT_COMPILE_STACK_SIZE; |
| 1897 |
compile_stack.avail = 0; |
| 1898 |
|
| 1899 |
/* Initialize the pattern buffer. */ |
| 1900 |
bufp->syntax = syntax; |
| 1901 |
bufp->fastmap_accurate = 0; |
| 1902 |
bufp->not_bol = bufp->not_eol = 0; |
| 1903 |
|
| 1904 |
/* Set `used' to zero, so that if we return an error, the pattern |
| 1905 |
printer (for debugging) will think there's no pattern. We reset it |
| 1906 |
at the end. */ |
| 1907 |
bufp->used = 0; |
| 1908 |
|
| 1909 |
/* Always count groups, whether or not bufp->no_sub is set. */ |
| 1910 |
bufp->re_nsub = 0; |
| 1911 |
|
| 1912 |
#if !defined emacs && !defined SYNTAX_TABLE |
| 1913 |
/* Initialize the syntax table. */ |
| 1914 |
init_syntax_once (); |
| 1915 |
#endif |
| 1916 |
|
| 1917 |
if (bufp->allocated == 0) |
| 1918 |
{ |
| 1919 |
if (bufp->buffer) |
| 1920 |
{ /* If zero allocated, but buffer is non-null, try to realloc |
| 1921 |
enough space. This loses if buffer's address is bogus, but |
| 1922 |
that is the user's responsibility. */ |
| 1923 |
RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); |
| 1924 |
} |
| 1925 |
else |
| 1926 |
{ /* Caller did not allocate a buffer. Do it for them. */ |
| 1927 |
bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); |
| 1928 |
} |
| 1929 |
if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); |
| 1930 |
|
| 1931 |
bufp->allocated = INIT_BUF_SIZE; |
| 1932 |
} |
| 1933 |
|
| 1934 |
begalt = b = bufp->buffer; |
| 1935 |
|
| 1936 |
/* Loop through the uncompiled pattern until we're at the end. */ |
| 1937 |
while (p != pend) |
| 1938 |
{ |
| 1939 |
PATFETCH (c); |
| 1940 |
|
| 1941 |
switch (c) |
| 1942 |
{ |
| 1943 |
case '^': |
| 1944 |
{ |
| 1945 |
if ( /* If at start of pattern, it's an operator. */ |
| 1946 |
p == pattern + 1 |
| 1947 |
/* If context independent, it's an operator. */ |
| 1948 |
|| syntax & RE_CONTEXT_INDEP_ANCHORS |
| 1949 |
/* Otherwise, depends on what's come before. */ |
| 1950 |
|| at_begline_loc_p (pattern, p, syntax)) |
| 1951 |
BUF_PUSH (begline); |
| 1952 |
else |
| 1953 |
goto normal_char; |
| 1954 |
} |
| 1955 |
break; |
| 1956 |
|
| 1957 |
|
| 1958 |
case '$': |
| 1959 |
{ |
| 1960 |
if ( /* If at end of pattern, it's an operator. */ |
| 1961 |
p == pend |
| 1962 |
/* If context independent, it's an operator. */ |
| 1963 |
|| syntax & RE_CONTEXT_INDEP_ANCHORS |
| 1964 |
/* Otherwise, depends on what's next. */ |
| 1965 |
|| at_endline_loc_p (p, pend, syntax)) |
| 1966 |
BUF_PUSH (endline); |
| 1967 |
else |
| 1968 |
goto normal_char; |
| 1969 |
} |
| 1970 |
break; |
| 1971 |
|
| 1972 |
|
| 1973 |
case '+': |
| 1974 |
case '?': |
| 1975 |
if ((syntax & RE_BK_PLUS_QM) |
| 1976 |
|| (syntax & RE_LIMITED_OPS)) |
| 1977 |
goto normal_char; |
| 1978 |
handle_plus: |
| 1979 |
case '*': |
| 1980 |
/* If there is no previous pattern... */ |
| 1981 |
if (!laststart) |
| 1982 |
{ |
| 1983 |
if (syntax & RE_CONTEXT_INVALID_OPS) |
| 1984 |
FREE_STACK_RETURN (REG_BADRPT); |
| 1985 |
else if (!(syntax & RE_CONTEXT_INDEP_OPS)) |
| 1986 |
goto normal_char; |
| 1987 |
} |
| 1988 |
|
| 1989 |
{ |
| 1990 |
/* Are we optimizing this jump? */ |
| 1991 |
boolean keep_string_p = false; |
| 1992 |
|
| 1993 |
/* 1 means zero (many) matches is allowed. */ |
| 1994 |
char zero_times_ok = 0, many_times_ok = 0; |
| 1995 |
|
| 1996 |
/* If there is a sequence of repetition chars, collapse it |
| 1997 |
down to just one (the right one). We can't combine |
| 1998 |
interval operators with these because of, e.g., `a{2}*', |
| 1999 |
which should only match an even number of `a's. */ |
| 2000 |
|
| 2001 |
for (;;) |
| 2002 |
{ |
| 2003 |
zero_times_ok |= c != '+'; |
| 2004 |
many_times_ok |= c != '?'; |
| 2005 |
|
| 2006 |
if (p == pend) |
| 2007 |
break; |
| 2008 |
|
| 2009 |
PATFETCH (c); |
| 2010 |
|
| 2011 |
if (c == '*' |
| 2012 |
|| (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) |
| 2013 |
; |
| 2014 |
|
| 2015 |
else if (syntax & RE_BK_PLUS_QM && c == '\\') |
| 2016 |
{ |
| 2017 |
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
| 2018 |
|
| 2019 |
PATFETCH (c1); |
| 2020 |
if (!(c1 == '+' || c1 == '?')) |
| 2021 |
{ |
| 2022 |
PATUNFETCH; |
| 2023 |
PATUNFETCH; |
| 2024 |
break; |
| 2025 |
} |
| 2026 |
|
| 2027 |
c = c1; |
| 2028 |
} |
| 2029 |
else |
| 2030 |
{ |
| 2031 |
PATUNFETCH; |
| 2032 |
break; |
| 2033 |
} |
| 2034 |
|
| 2035 |
/* If we get here, we found another repeat character. */ |
| 2036 |
} |
| 2037 |
|
| 2038 |
/* Star, etc. applied to an empty pattern is equivalent |
| 2039 |
to an empty pattern. */ |
| 2040 |
if (!laststart) |
| 2041 |
break; |
| 2042 |
|
| 2043 |
/* Now we know whether or not zero matches is allowed |
| 2044 |
and also whether or not two or more matches is allowed. */ |
| 2045 |
if (many_times_ok) |
| 2046 |
{ /* More than one repetition is allowed, so put in at the |
| 2047 |
end a backward relative jump from `b' to before the next |
| 2048 |
jump we're going to put in below (which jumps from |
| 2049 |
laststart to after this jump). |
| 2050 |
|
| 2051 |
But if we are at the `*' in the exact sequence `.*\n', |
| 2052 |
insert an unconditional jump backwards to the ., |
| 2053 |
instead of the beginning of the loop. This way we only |
| 2054 |
push a failure point once, instead of every time |
| 2055 |
through the loop. */ |
| 2056 |
assert (p - 1 > pattern); |
| 2057 |
|
| 2058 |
/* Allocate the space for the jump. */ |
| 2059 |
GET_BUFFER_SPACE (3); |
| 2060 |
|
| 2061 |
/* We know we are not at the first character of the pattern, |
| 2062 |
because laststart was nonzero. And we've already |
| 2063 |
incremented `p', by the way, to be the character after |
| 2064 |
the `*'. Do we have to do something analogous here |
| 2065 |
for null bytes, because of RE_DOT_NOT_NULL? */ |
| 2066 |
if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') |
| 2067 |
&& zero_times_ok |
| 2068 |
&& p < pend && TRANSLATE (*p) == TRANSLATE ('\n') |
| 2069 |
&& !(syntax & RE_DOT_NEWLINE)) |
| 2070 |
{ /* We have .*\n. */ |
| 2071 |
STORE_JUMP (jump, b, laststart); |
| 2072 |
keep_string_p = true; |
| 2073 |
} |
| 2074 |
else |
| 2075 |
/* Anything else. */ |
| 2076 |
STORE_JUMP (maybe_pop_jump, b, laststart - 3); |
| 2077 |
|
| 2078 |
/* We've added more stuff to the buffer. */ |
| 2079 |
b += 3; |
| 2080 |
} |
| 2081 |
|
| 2082 |
/* On failure, jump from laststart to b + 3, which will be the |
| 2083 |
end of the buffer after this jump is inserted. */ |
| 2084 |
GET_BUFFER_SPACE (3); |
| 2085 |
INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump |
| 2086 |
: on_failure_jump, |
| 2087 |
laststart, b + 3); |
| 2088 |
pending_exact = 0; |
| 2089 |
b += 3; |
| 2090 |
|
| 2091 |
if (!zero_times_ok) |
| 2092 |
{ |
| 2093 |
/* At least one repetition is required, so insert a |
| 2094 |
`dummy_failure_jump' before the initial |
| 2095 |
`on_failure_jump' instruction of the loop. This |
| 2096 |
effects a skip over that instruction the first time |
| 2097 |
we hit that loop. */ |
| 2098 |
GET_BUFFER_SPACE (3); |
| 2099 |
INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6); |
| 2100 |
b += 3; |
| 2101 |
} |
| 2102 |
} |
| 2103 |
break; |
| 2104 |
|
| 2105 |
|
| 2106 |
case '.': |
| 2107 |
laststart = b; |
| 2108 |
BUF_PUSH (anychar); |
| 2109 |
break; |
| 2110 |
|
| 2111 |
|
| 2112 |
case '[': |
| 2113 |
{ |
| 2114 |
boolean had_char_class = false; |
| 2115 |
|
| 2116 |
if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
| 2117 |
|
| 2118 |
/* Ensure that we have enough space to push a charset: the |
| 2119 |
opcode, the length count, and the bitset; 34 bytes in all. */ |
| 2120 |
GET_BUFFER_SPACE (34); |
| 2121 |
|
| 2122 |
laststart = b; |
| 2123 |
|
| 2124 |
/* We test `*p == '^' twice, instead of using an if |
| 2125 |
statement, so we only need one BUF_PUSH. */ |
| 2126 |
BUF_PUSH (*p == '^' ? charset_not : charset); |
| 2127 |
if (*p == '^') |
| 2128 |
p++; |
| 2129 |
|
| 2130 |
/* Remember the first position in the bracket expression. */ |
| 2131 |
p1 = p; |
| 2132 |
|
| 2133 |
/* Push the number of bytes in the bitmap. */ |
| 2134 |
BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); |
| 2135 |
|
| 2136 |
/* Clear the whole map. */ |
| 2137 |
bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); |
| 2138 |
|
| 2139 |
/* charset_not matches newline according to a syntax bit. */ |
| 2140 |
if ((re_opcode_t) b[-2] == charset_not |
| 2141 |
&& (syntax & RE_HAT_LISTS_NOT_NEWLINE)) |
| 2142 |
SET_LIST_BIT ('\n'); |
| 2143 |
|
| 2144 |
/* Read in characters and ranges, setting map bits. */ |
| 2145 |
for (;;) |
| 2146 |
{ |
| 2147 |
if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
| 2148 |
|
| 2149 |
PATFETCH (c); |
| 2150 |
|
| 2151 |
/* \ might escape characters inside [...] and [^...]. */ |
| 2152 |
if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') |
| 2153 |
{ |
| 2154 |
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
| 2155 |
|
| 2156 |
PATFETCH (c1); |
| 2157 |
SET_LIST_BIT (c1); |
| 2158 |
continue; |
| 2159 |
} |
| 2160 |
|
| 2161 |
/* Could be the end of the bracket expression. If it's |
| 2162 |
not (i.e., when the bracket expression is `[]' so |
| 2163 |
far), the ']' character bit gets set way below. */ |
| 2164 |
if (c == ']' && p != p1 + 1) |
| 2165 |
break; |
| 2166 |
|
| 2167 |
/* Look ahead to see if it's a range when the last thing |
| 2168 |
was a character class. */ |
| 2169 |
if (had_char_class && c == '-' && *p != ']') |
| 2170 |
FREE_STACK_RETURN (REG_ERANGE); |
| 2171 |
|
| 2172 |
/* Look ahead to see if it's a range when the last thing |
| 2173 |
was a character: if this is a hyphen not at the |
| 2174 |
beginning or the end of a list, then it's the range |
| 2175 |
operator. */ |
| 2176 |
if (c == '-' |
| 2177 |
&& !(p - 2 >= pattern && p[-2] == '[') |
| 2178 |
&& !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') |
| 2179 |
&& *p != ']') |
| 2180 |
{ |
| 2181 |
reg_errcode_t ret |
| 2182 |
= compile_range (&p, pend, translate, syntax, b); |
| 2183 |
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
| 2184 |
} |
| 2185 |
|
| 2186 |
else if (p[0] == '-' && p[1] != ']') |
| 2187 |
{ /* This handles ranges made up of characters only. */ |
| 2188 |
reg_errcode_t ret; |
| 2189 |
|
| 2190 |
/* Move past the `-'. */ |
| 2191 |
PATFETCH (c1); |
| 2192 |
|
| 2193 |
ret = compile_range (&p, pend, translate, syntax, b); |
| 2194 |
if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
| 2195 |
} |
| 2196 |
|
| 2197 |
/* See if we're at the beginning of a possible character |
| 2198 |
class. */ |
| 2199 |
|
| 2200 |
else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') |
| 2201 |
{ /* Leave room for the null. */ |
| 2202 |
char str[CHAR_CLASS_MAX_LENGTH + 1]; |
| 2203 |
|
| 2204 |
PATFETCH (c); |
| 2205 |
c1 = 0; |
| 2206 |
|
| 2207 |
/* If pattern is `[[:'. */ |
| 2208 |
if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
| 2209 |
|
| 2210 |
for (;;) |
| 2211 |
{ |
| 2212 |
PATFETCH (c); |
| 2213 |
if ((c == ':' && *p == ']') || p == pend) |
| 2214 |
break; |
| 2215 |
if (c1 < CHAR_CLASS_MAX_LENGTH) |
| 2216 |
str[c1++] = c; |
| 2217 |
else |
| 2218 |
/* This is in any case an invalid class name. */ |
| 2219 |
str[0] = '\0'; |
| 2220 |
} |
| 2221 |
str[c1] = '\0'; |
| 2222 |
|
| 2223 |
/* If isn't a word bracketed by `[:' and `:]': |
| 2224 |
undo the ending character, the letters, and leave |
| 2225 |
the leading `:' and `[' (but set bits for them). */ |
| 2226 |
if (c == ':' && *p == ']') |
| 2227 |
{ |
| 2228 |
#if defined _LIBC || WIDE_CHAR_SUPPORT |
| 2229 |
boolean is_lower = STREQ (str, "lower"); |
| 2230 |
boolean is_upper = STREQ (str, "upper"); |
| 2231 |
wctype_t wt; |
| 2232 |
int ch; |
| 2233 |
|
| 2234 |
wt = IS_CHAR_CLASS (str); |
| 2235 |
if (wt == 0) |
| 2236 |
FREE_STACK_RETURN (REG_ECTYPE); |
| 2237 |
|
| 2238 |
/* Throw away the ] at the end of the character |
| 2239 |
class. */ |
| 2240 |
PATFETCH (c); |
| 2241 |
|
| 2242 |
if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
| 2243 |
|
| 2244 |
for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) |
| 2245 |
{ |
| 2246 |
# ifdef _LIBC |
| 2247 |
if (__iswctype (__btowc (ch), wt)) |
| 2248 |
SET_LIST_BIT (ch); |
| 2249 |
# else |
| 2250 |
if (iswctype (btowc (ch), wt)) |
| 2251 |
SET_LIST_BIT (ch); |
| 2252 |
# endif |
| 2253 |
|
| 2254 |
if (translate && (is_upper || is_lower) |
| 2255 |
&& (ISUPPER (ch) || ISLOWER (ch))) |
| 2256 |
SET_LIST_BIT (ch); |
| 2257 |
} |
| 2258 |
|
| 2259 |
had_char_class = true; |
| 2260 |
#else |
| 2261 |
int ch; |
| 2262 |
boolean is_alnum = STREQ (str, "alnum"); |
| 2263 |
boolean is_alpha = STREQ (str, "alpha"); |
| 2264 |
boolean is_blank = STREQ (str, "blank"); |
| 2265 |
boolean is_cntrl = STREQ (str, "cntrl"); |
| 2266 |
boolean is_digit = STREQ (str, "digit"); |
| 2267 |
boolean is_graph = STREQ (str, "graph"); |
| 2268 |
boolean is_lower = STREQ (str, "lower"); |
| 2269 |
boolean is_print = STREQ (str, "print"); |
| 2270 |
boolean is_punct = STREQ (str, "punct"); |
| 2271 |
boolean is_space = STREQ (str, "space"); |
| 2272 |
boolean is_upper = STREQ (str, "upper"); |
| 2273 |
boolean is_xdigit = STREQ (str, "xdigit"); |
| 2274 |
|
| 2275 |
if (!IS_CHAR_CLASS (str)) |
| 2276 |
FREE_STACK_RETURN (REG_ECTYPE); |
| 2277 |
|
| 2278 |
/* Throw away the ] at the end of the character |
| 2279 |
class. */ |
| 2280 |
PATFETCH (c); |
| 2281 |
|
| 2282 |
if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
| 2283 |
|
| 2284 |
for (ch = 0; ch < 1 << BYTEWIDTH; ch++) |
| 2285 |
{ |
| 2286 |
/* This was split into 3 if's to |
| 2287 |
avoid an arbitrary limit in some compiler. */ |
| 2288 |
if ( (is_alnum && ISALNUM (ch)) |
| 2289 |
|| (is_alpha && ISALPHA (ch)) |
| 2290 |
|| (is_blank && ISBLANK (ch)) |
| 2291 |
|| (is_cntrl && ISCNTRL (ch))) |
| 2292 |
SET_LIST_BIT (ch); |
| 2293 |
if ( (is_digit && ISDIGIT (ch)) |
| 2294 |
|| (is_graph && ISGRAPH (ch)) |
| 2295 |
|| (is_lower && ISLOWER (ch)) |
| 2296 |
|| (is_print && ISPRINT (ch))) |
| 2297 |
SET_LIST_BIT (ch); |
| 2298 |
if ( (is_punct && ISPUNCT (ch)) |
| 2299 |
|| (is_space && ISSPACE (ch)) |
| 2300 |
|| (is_upper && ISUPPER (ch)) |
| 2301 |
|| (is_xdigit && ISXDIGIT (ch))) |
| 2302 |
SET_LIST_BIT (ch); |
| 2303 |
if ( translate && (is_upper || is_lower) |
| 2304 |
&& (ISUPPER (ch) || ISLOWER (ch))) |
| 2305 |
SET_LIST_BIT (ch); |
| 2306 |
} |
| 2307 |
had_char_class = true; |
| 2308 |
#endif /* libc || wctype.h */ |
| 2309 |
} |
| 2310 |
else |
| 2311 |
{ |
| 2312 |
c1++; |
| 2313 |
while (c1--) |
| 2314 |
PATUNFETCH; |
| 2315 |
SET_LIST_BIT ('['); |
| 2316 |
SET_LIST_BIT (':'); |
| 2317 |
had_char_class = false; |
| 2318 |
} |
| 2319 |
} |
| 2320 |
else |
| 2321 |
{ |
| 2322 |
had_char_class = false; |
| 2323 |
SET_LIST_BIT (c); |
| 2324 |
} |
| 2325 |
} |
| 2326 |
|
| 2327 |
/* Discard any (non)matching list bytes that are all 0 at the |
| 2328 |
end of the map. Decrease the map-length byte too. */ |
| 2329 |
while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) |
| 2330 |
b[-1]--; |
| 2331 |
b += b[-1]; |
| 2332 |
} |
| 2333 |
break; |
| 2334 |
|
| 2335 |
|
| 2336 |
case '(': |
| 2337 |
if (syntax & RE_NO_BK_PARENS) |
| 2338 |
goto handle_open; |
| 2339 |
else |
| 2340 |
goto normal_char; |
| 2341 |
|
| 2342 |
|
| 2343 |
case ')': |
| 2344 |
if (syntax & RE_NO_BK_PARENS) |
| 2345 |
goto handle_close; |
| 2346 |
else |
| 2347 |
goto normal_char; |
| 2348 |
|
| 2349 |
|
| 2350 |
case '\n': |
| 2351 |
if (syntax & RE_NEWLINE_ALT) |
| 2352 |
goto handle_alt; |
| 2353 |
else |
| 2354 |
goto normal_char; |
| 2355 |
|
| 2356 |
|
| 2357 |
case '|': |
| 2358 |
if (syntax & RE_NO_BK_VBAR) |
| 2359 |
goto handle_alt; |
| 2360 |
else |
| 2361 |
goto normal_char; |
| 2362 |
|
| 2363 |
|
| 2364 |
case '{': |
| 2365 |
if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) |
| 2366 |
goto handle_interval; |
| 2367 |
else |
| 2368 |
goto normal_char; |
| 2369 |
|
| 2370 |
|
| 2371 |
case '\\': |
| 2372 |
if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
| 2373 |
|
| 2374 |
/* Do not translate the character after the \, so that we can |
| 2375 |
distinguish, e.g., \B from \b, even if we normally would |
| 2376 |
translate, e.g., B to b. */ |
| 2377 |
PATFETCH_RAW (c); |
| 2378 |
|
| 2379 |
switch (c) |
| 2380 |
{ |
| 2381 |
case '(': |
| 2382 |
if (syntax & RE_NO_BK_PARENS) |
| 2383 |
goto normal_backslash; |
| 2384 |
|
| 2385 |
handle_open: |
| 2386 |
bufp->re_nsub++; |
| 2387 |
regnum++; |
| 2388 |
|
| 2389 |
if (COMPILE_STACK_FULL) |
| 2390 |
{ |
| 2391 |
RETALLOC (compile_stack.stack, compile_stack.size << 1, |
| 2392 |
compile_stack_elt_t); |
| 2393 |
if (compile_stack.stack == NULL) return REG_ESPACE; |
| 2394 |
|
| 2395 |
compile_stack.size <<= 1; |
| 2396 |
} |
| 2397 |
|
| 2398 |
/* These are the values to restore when we hit end of this |
| 2399 |
group. They are all relative offsets, so that if the |
| 2400 |
whole pattern moves because of realloc, they will still |
| 2401 |
be valid. */ |
| 2402 |
COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; |
| 2403 |
COMPILE_STACK_TOP.fixup_alt_jump |
| 2404 |
= fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; |
| 2405 |
COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; |
| 2406 |
COMPILE_STACK_TOP.regnum = regnum; |
| 2407 |
|
| 2408 |
/* We will eventually replace the 0 with the number of |
| 2409 |
groups inner to this one. But do not push a |
| 2410 |
start_memory for groups beyond the last one we can |
| 2411 |
represent in the compiled pattern. */ |
| 2412 |
if (regnum <= MAX_REGNUM) |
| 2413 |
{ |
| 2414 |
COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2; |
| 2415 |
BUF_PUSH_3 (start_memory, regnum, 0); |
| 2416 |
} |
| 2417 |
|
| 2418 |
compile_stack.avail++; |
| 2419 |
|
| 2420 |
fixup_alt_jump = 0; |
| 2421 |
laststart = 0; |
| 2422 |
begalt = b; |
| 2423 |
/* If we've reached MAX_REGNUM groups, then this open |
| 2424 |
won't actually generate any code, so we'll have to |
| 2425 |
clear pending_exact explicitly. */ |
| 2426 |
pending_exact = 0; |
| 2427 |
break; |
| 2428 |
|
| 2429 |
|
| 2430 |
case ')': |
| 2431 |
if (syntax & RE_NO_BK_PARENS) goto normal_backslash; |
| 2432 |
|
| 2433 |
if (COMPILE_STACK_EMPTY) |
| 2434 |
{ |
| 2435 |
if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) |
| 2436 |
goto normal_backslash; |
| 2437 |
else |
| 2438 |
FREE_STACK_RETURN (REG_ERPAREN); |
| 2439 |
} |
| 2440 |
|
| 2441 |
handle_close: |
| 2442 |
if (fixup_alt_jump) |
| 2443 |
{ /* Push a dummy failure point at the end of the |
| 2444 |
alternative for a possible future |
| 2445 |
`pop_failure_jump' to pop. See comments at |
| 2446 |
`push_dummy_failure' in `re_match_2'. */ |
| 2447 |
BUF_PUSH (push_dummy_failure); |
| 2448 |
|
| 2449 |
/* We allocated space for this jump when we assigned |
| 2450 |
to `fixup_alt_jump', in the `handle_alt' case below. */ |
| 2451 |
STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); |
| 2452 |
} |
| 2453 |
|
| 2454 |
/* See similar code for backslashed left paren above. */ |
| 2455 |
if (COMPILE_STACK_EMPTY) |
| 2456 |
{ |
| 2457 |
if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) |
| 2458 |
goto normal_char; |
| 2459 |
else |
| 2460 |
FREE_STACK_RETURN (REG_ERPAREN); |
| 2461 |
} |
| 2462 |
|
| 2463 |
/* Since we just checked for an empty stack above, this |
| 2464 |
``can't happen''. */ |
| 2465 |
assert (compile_stack.avail != 0); |
| 2466 |
{ |
| 2467 |
/* We don't just want to restore into `regnum', because |
| 2468 |
later groups should continue to be numbered higher, |
| 2469 |
as in `(ab)c(de)' -- the second group is #2. */ |
| 2470 |
regnum_t this_group_regnum; |
| 2471 |
|
| 2472 |
compile_stack.avail--; |
| 2473 |
begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; |
| 2474 |
fixup_alt_jump |
| 2475 |
= COMPILE_STACK_TOP.fixup_alt_jump |
| 2476 |
? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 |
| 2477 |
: 0; |
| 2478 |
laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; |
| 2479 |
this_group_regnum = COMPILE_STACK_TOP.regnum; |
| 2480 |
/* If we've reached MAX_REGNUM groups, then this open |
| 2481 |
won't actually generate any code, so we'll have to |
| 2482 |
clear pending_exact explicitly. */ |
| 2483 |
pending_exact = 0; |
| 2484 |
|
| 2485 |
/* We're at the end of the group, so now we know how many |
| 2486 |
groups were inside this one. */ |
| 2487 |
if (this_group_regnum <= MAX_REGNUM) |
| 2488 |
{ |
| 2489 |
unsigned char *inner_group_loc |
| 2490 |
= bufp->buffer + COMPILE_STACK_TOP.inner_group_offset; |
| 2491 |
|
| 2492 |
*inner_group_loc = regnum - this_group_regnum; |
| 2493 |
BUF_PUSH_3 (stop_memory, this_group_regnum, |
| 2494 |
regnum - this_group_regnum); |
| 2495 |
} |
| 2496 |
} |
| 2497 |
break; |
| 2498 |
|
| 2499 |
|
| 2500 |
case '|': /* `\|'. */ |
| 2501 |
if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) |
| 2502 |
goto normal_backslash; |
| 2503 |
handle_alt: |
| 2504 |
if (syntax & RE_LIMITED_OPS) |
| 2505 |
goto normal_char; |
| 2506 |
|
| 2507 |
/* Insert before the previous alternative a jump which |
| 2508 |
jumps to this alternative if the former fails. */ |
| 2509 |
GET_BUFFER_SPACE (3); |
| 2510 |
INSERT_JUMP (on_failure_jump, begalt, b + 6); |
| 2511 |
pending_exact = 0; |
| 2512 |
b += 3; |
| 2513 |
|
| 2514 |
/* The alternative before this one has a jump after it |
| 2515 |
which gets executed if it gets matched. Adjust that |
| 2516 |
jump so it will jump to this alternative's analogous |
| 2517 |
jump (put in below, which in turn will jump to the next |
| 2518 |
(if any) alternative's such jump, etc.). The last such |
| 2519 |
jump jumps to the correct final destination. A picture: |
| 2520 |
_____ _____ |
| 2521 |
| | | | |
| 2522 |
| v | v |
| 2523 |
a | b | c |
| 2524 |
|
| 2525 |
If we are at `b', then fixup_alt_jump right now points to a |
| 2526 |
three-byte space after `a'. We'll put in the jump, set |
| 2527 |
fixup_alt_jump to right after `b', and leave behind three |
| 2528 |
bytes which we'll fill in when we get to after `c'. */ |
| 2529 |
|
| 2530 |
if (fixup_alt_jump) |
| 2531 |
STORE_JUMP (jump_past_alt, fixup_alt_jump, b); |
| 2532 |
|
| 2533 |
/* Mark and leave space for a jump after this alternative, |
| 2534 |
to be filled in later either by next alternative or |
| 2535 |
when know we're at the end of a series of alternatives. */ |
| 2536 |
fixup_alt_jump = b; |
| 2537 |
GET_BUFFER_SPACE (3); |
| 2538 |
b += 3; |
| 2539 |
|
| 2540 |
laststart = 0; |
| 2541 |
begalt = b; |
| 2542 |
break; |
| 2543 |
|
| 2544 |
|
| 2545 |
case '{': |
| 2546 |
/* If \{ is a literal. */ |
| 2547 |
if (!(syntax & RE_INTERVALS) |
| 2548 |
/* If we're at `\{' and it's not the open-interval |
| 2549 |
operator. */ |
| 2550 |
|| ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES)) |
| 2551 |
|| (p - 2 == pattern && p == pend)) |
| 2552 |
goto normal_backslash; |
| 2553 |
|
| 2554 |
handle_interval: |
| 2555 |
{ |
| 2556 |
/* If got here, then the syntax allows intervals. */ |
| 2557 |
|
| 2558 |
/* At least (most) this many matches must be made. */ |
| 2559 |
int lower_bound = -1, upper_bound = -1; |
| 2560 |
|
| 2561 |
beg_interval = p - 1; |
| 2562 |
|
| 2563 |
if (p == pend) |
| 2564 |
{ |
| 2565 |
if (syntax & RE_NO_BK_BRACES) |
| 2566 |
goto unfetch_interval; |
| 2567 |
else |
| 2568 |
FREE_STACK_RETURN (REG_EBRACE); |
| 2569 |
} |
| 2570 |
|
| 2571 |
GET_UNSIGNED_NUMBER (lower_bound); |
| 2572 |
|
| 2573 |
if (c == ',') |
| 2574 |
{ |
| 2575 |
GET_UNSIGNED_NUMBER (upper_bound); |
| 2576 |
if (upper_bound < 0) upper_bound = RE_DUP_MAX; |
| 2577 |
} |
| 2578 |
else |
| 2579 |
/* Interval such as `{1}' => match exactly once. */ |
| 2580 |
upper_bound = lower_bound; |
| 2581 |
|
| 2582 |
if (lower_bound < 0 || upper_bound > RE_DUP_MAX |
| 2583 |
|| lower_bound > upper_bound) |
| 2584 |
{ |
| 2585 |
if (syntax & RE_NO_BK_BRACES) |
| 2586 |
goto unfetch_interval; |
| 2587 |
else |
| 2588 |
FREE_STACK_RETURN (REG_BADBR); |
| 2589 |
} |
| 2590 |
|
| 2591 |
if (!(syntax & RE_NO_BK_BRACES)) |
| 2592 |
{ |
| 2593 |
if (c != '\\') FREE_STACK_RETURN (REG_EBRACE); |
| 2594 |
|
| 2595 |
PATFETCH (c); |
| 2596 |
} |
| 2597 |
|
| 2598 |
if (c != '}') |
| 2599 |
{ |
| 2600 |
if (syntax & RE_NO_BK_BRACES) |
| 2601 |
goto unfetch_interval; |
| 2602 |
else |
| 2603 |
FREE_STACK_RETURN (REG_BADBR); |
| 2604 |
} |
| 2605 |
|
| 2606 |
/* We just parsed a valid interval. */ |
| 2607 |
|
| 2608 |
/* If it's invalid to have no preceding re. */ |
| 2609 |
if (!laststart) |
| 2610 |
{ |
| 2611 |
if (syntax & RE_CONTEXT_INVALID_OPS) |
| 2612 |
FREE_STACK_RETURN (REG_BADRPT); |
| 2613 |
else if (syntax & RE_CONTEXT_INDEP_OPS) |
| 2614 |
laststart = b; |
| 2615 |
else |
| 2616 |
goto unfetch_interval; |
| 2617 |
} |
| 2618 |
|
| 2619 |
/* If the upper bound is zero, don't want to succeed at |
| 2620 |
all; jump from `laststart' to `b + 3', which will be |
| 2621 |
the end of the buffer after we insert the jump. */ |
| 2622 |
if (upper_bound == 0) |
| 2623 |
{ |
| 2624 |
GET_BUFFER_SPACE (3); |
| 2625 |
INSERT_JUMP (jump, laststart, b + 3); |
| 2626 |
b += 3; |
| 2627 |
} |
| 2628 |
|
| 2629 |
/* Otherwise, we have a nontrivial interval. When |
| 2630 |
we're all done, the pattern will look like: |
| 2631 |
set_number_at <jump count> <upper bound> |
| 2632 |
set_number_at <succeed_n count> <lower bound> |
| 2633 |
succeed_n <after jump addr> <succeed_n count> |
| 2634 |
<body of loop> |
| 2635 |
jump_n <succeed_n addr> <jump count> |
| 2636 |
(The upper bound and `jump_n' are omitted if |
| 2637 |
`upper_bound' is 1, though.) */ |
| 2638 |
else |
| 2639 |
{ /* If the upper bound is > 1, we need to insert |
| 2640 |
more at the end of the loop. */ |
| 2641 |
unsigned nbytes = 10 + (upper_bound > 1) * 10; |
| 2642 |
|
| 2643 |
GET_BUFFER_SPACE (nbytes); |
| 2644 |
|
| 2645 |
/* Initialize lower bound of the `succeed_n', even |
| 2646 |
though it will be set during matching by its |
| 2647 |
attendant `set_number_at' (inserted next), |
| 2648 |
because `re_compile_fastmap' needs to know. |
| 2649 |
Jump to the `jump_n' we might insert below. */ |
| 2650 |
INSERT_JUMP2 (succeed_n, laststart, |
| 2651 |
b + 5 + (upper_bound > 1) * 5, |
| 2652 |
lower_bound); |
| 2653 |
b += 5; |
| 2654 |
|
| 2655 |
/* Code to initialize the lower bound. Insert |
| 2656 |
before the `succeed_n'. The `5' is the last two |
| 2657 |
bytes of this `set_number_at', plus 3 bytes of |
| 2658 |
the following `succeed_n'. */ |
| 2659 |
insert_op2 (set_number_at, laststart, 5, lower_bound, b); |
| 2660 |
b += 5; |
| 2661 |
|
| 2662 |
if (upper_bound > 1) |
| 2663 |
{ /* More than one repetition is allowed, so |
| 2664 |
append a backward jump to the `succeed_n' |
| 2665 |
that starts this interval. |
| 2666 |
|
| 2667 |
When we've reached this during matching, |
| 2668 |
we'll have matched the interval once, so |
| 2669 |
jump back only `upper_bound - 1' times. */ |
| 2670 |
STORE_JUMP2 (jump_n, b, laststart + 5, |
| 2671 |
upper_bound - 1); |
| 2672 |
b += 5; |
| 2673 |
|
| 2674 |
/* The location we want to set is the second |
| 2675 |
parameter of the `jump_n'; that is `b-2' as |
| 2676 |
an absolute address. `laststart' will be |
| 2677 |
the `set_number_at' we're about to insert; |
| 2678 |
`laststart+3' the number to set, the source |
| 2679 |
for the relative address. But we are |
| 2680 |
inserting into the middle of the pattern -- |
| 2681 |
so everything is getting moved up by 5. |
| 2682 |
Conclusion: (b - 2) - (laststart + 3) + 5, |
| 2683 |
i.e., b - laststart. |
| 2684 |
|
| 2685 |
We insert this at the beginning of the loop |
| 2686 |
so that if we fail during matching, we'll |
| 2687 |
reinitialize the bounds. */ |
| 2688 |
insert_op2 (set_number_at, laststart, b - laststart, |
| 2689 |
upper_bound - 1, b); |
| 2690 |
b += 5; |
| 2691 |
} |
| 2692 |
} |
| 2693 |
pending_exact = 0; |
| 2694 |
beg_interval = NULL; |
| 2695 |
} |
| 2696 |
break; |
| 2697 |
|
| 2698 |
unfetch_interval: |
| 2699 |
/* If an invalid interval, match the characters as literals. */ |
| 2700 |
assert (beg_interval); |
| 2701 |
p = beg_interval; |
| 2702 |
beg_interval = NULL; |
| 2703 |
|
| 2704 |
/* normal_char and normal_backslash need `c'. */ |
| 2705 |
PATFETCH (c); |
| 2706 |
|
| 2707 |
if (!(syntax & RE_NO_BK_BRACES)) |
| 2708 |
{ |
| 2709 |
if (p > pattern && p[-1] == '\\') |
| 2710 |
goto normal_backslash; |
| 2711 |
} |
| 2712 |
goto normal_char; |
| 2713 |
|
| 2714 |
#ifdef emacs |
| 2715 |
/* There is no way to specify the before_dot and after_dot |
| 2716 |
operators. rms says this is ok. --karl */ |
| 2717 |
case '=': |
| 2718 |
BUF_PUSH (at_dot); |
| 2719 |
break; |
| 2720 |
|
| 2721 |
case 's': |
| 2722 |
laststart = b; |
| 2723 |
PATFETCH (c); |
| 2724 |
BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); |
| 2725 |
break; |
| 2726 |
|
| 2727 |
case 'S': |
| 2728 |
laststart = b; |
| 2729 |
PATFETCH (c); |
| 2730 |
BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); |
| 2731 |
break; |
| 2732 |
#endif /* emacs */ |
| 2733 |
|
| 2734 |
|
| 2735 |
case 'w': |
| 2736 |
if (syntax & RE_NO_GNU_OPS) |
| 2737 |
goto normal_char; |
| 2738 |
laststart = b; |
| 2739 |
BUF_PUSH (wordchar); |
| 2740 |
break; |
| 2741 |
|
| 2742 |
|
| 2743 |
case 'W': |
| 2744 |
if (syntax & RE_NO_GNU_OPS) |
| 2745 |
goto normal_char; |
| 2746 |
laststart = b; |
| 2747 |
BUF_PUSH (notwordchar); |
| 2748 |
break; |
| 2749 |
|
| 2750 |
|
| 2751 |
case '<': |
| 2752 |
if (syntax & RE_NO_GNU_OPS) |
| 2753 |
goto normal_char; |
| 2754 |
BUF_PUSH (wordbeg); |
| 2755 |
break; |
| 2756 |
|
| 2757 |
case '>': |
| 2758 |
if (syntax & RE_NO_GNU_OPS) |
| 2759 |
goto normal_char; |
| 2760 |
BUF_PUSH (wordend); |
| 2761 |
break; |
| 2762 |
|
| 2763 |
case 'b': |
| 2764 |
if (syntax & RE_NO_GNU_OPS) |
| 2765 |
goto normal_char; |
| 2766 |
BUF_PUSH (wordbound); |
| 2767 |
break; |
| 2768 |
|
| 2769 |
case 'B': |
| 2770 |
if (syntax & RE_NO_GNU_OPS) |
| 2771 |
goto normal_char; |
| 2772 |
BUF_PUSH (notwordbound); |
| 2773 |
break; |
| 2774 |
|
| 2775 |
case '`': |
| 2776 |
if (syntax & RE_NO_GNU_OPS) |
| 2777 |
goto normal_char; |
| 2778 |
BUF_PUSH (begbuf); |
| 2779 |
break; |
| 2780 |
|
| 2781 |
case '\'': |
| 2782 |
if (syntax & RE_NO_GNU_OPS) |
| 2783 |
goto normal_char; |
| 2784 |
BUF_PUSH (endbuf); |
| 2785 |
break; |
| 2786 |
|
| 2787 |
case '1': case '2': case '3': case '4': case '5': |
| 2788 |
case '6': case '7': case '8': case '9': |
| 2789 |
if (syntax & RE_NO_BK_REFS) |
| 2790 |
goto normal_char; |
| 2791 |
|
| 2792 |
c1 = c - '0'; |
| 2793 |
|
| 2794 |
if (c1 > regnum) |
| 2795 |
FREE_STACK_RETURN (REG_ESUBREG); |
| 2796 |
|
| 2797 |
/* Can't back reference to a subexpression if inside of it. */ |
| 2798 |
if (group_in_compile_stack (compile_stack, (regnum_t) c1)) |
| 2799 |
goto normal_char; |
| 2800 |
|
| 2801 |
laststart = b; |
| 2802 |
BUF_PUSH_2 (duplicate, c1); |
| 2803 |
break; |
| 2804 |
|
| 2805 |
|
| 2806 |
case '+': |
| 2807 |
case '?': |
| 2808 |
if (syntax & RE_BK_PLUS_QM) |
| 2809 |
goto handle_plus; |
| 2810 |
else |
| 2811 |
goto normal_backslash; |
| 2812 |
|
| 2813 |
default: |
| 2814 |
normal_backslash: |
| 2815 |
/* You might think it would be useful for \ to mean |
| 2816 |
not to translate; but if we don't translate it |
| 2817 |
it will never match anything. */ |
| 2818 |
c = TRANSLATE (c); |
| 2819 |
goto normal_char; |
| 2820 |
} |
| 2821 |
break; |
| 2822 |
|
| 2823 |
|
| 2824 |
default: |
| 2825 |
/* Expects the character in `c'. */ |
| 2826 |
normal_char: |
| 2827 |
/* If no exactn currently being built. */ |
| 2828 |
if (!pending_exact |
| 2829 |
|
| 2830 |
/* If last exactn not at current position. */ |
| 2831 |
|| pending_exact + *pending_exact + 1 != b |
| 2832 |
|
| 2833 |
/* We have only one byte following the exactn for the count. */ |
| 2834 |
|| *pending_exact == (1 << BYTEWIDTH) - 1 |
| 2835 |
|
| 2836 |
/* If followed by a repetition operator. */ |
| 2837 |
|| *p == '*' || *p == '^' |
| 2838 |
|| ((syntax & RE_BK_PLUS_QM) |
| 2839 |
? *p == '\\' && (p[1] == '+' || p[1] == '?') |
| 2840 |
: (*p == '+' || *p == '?')) |
| 2841 |
|| ((syntax & RE_INTERVALS) |
| 2842 |
&& ((syntax & RE_NO_BK_BRACES) |
| 2843 |
? *p == '{' |
| 2844 |
: (p[0] == '\\' && p[1] == '{')))) |
| 2845 |
{ |
| 2846 |
/* Start building a new exactn. */ |
| 2847 |
|
| 2848 |
laststart = b; |
| 2849 |
|
| 2850 |
BUF_PUSH_2 (exactn, 0); |
| 2851 |
pending_exact = b - 1; |
| 2852 |
} |
| 2853 |
|
| 2854 |
BUF_PUSH (c); |
| 2855 |
(*pending_exact)++; |
| 2856 |
break; |
| 2857 |
} /* switch (c) */ |
| 2858 |
} /* while p != pend */ |
| 2859 |
|
| 2860 |
|
| 2861 |
/* Through the pattern now. */ |
| 2862 |
|
| 2863 |
if (fixup_alt_jump) |
| 2864 |
STORE_JUMP (jump_past_alt, fixup_alt_jump, b); |
| 2865 |
|
| 2866 |
if (!COMPILE_STACK_EMPTY) |
| 2867 |
FREE_STACK_RETURN (REG_EPAREN); |
| 2868 |
|
| 2869 |
/* If we don't want backtracking, force success |
| 2870 |
the first time we reach the end of the compiled pattern. */ |
| 2871 |
if (syntax & RE_NO_POSIX_BACKTRACKING) |
| 2872 |
BUF_PUSH (succeed); |
| 2873 |
|
| 2874 |
free (compile_stack.stack); |
| 2875 |
|
| 2876 |
/* We have succeeded; set the length of the buffer. */ |
| 2877 |
bufp->used = b - bufp->buffer; |
| 2878 |
|
| 2879 |
#ifdef DEBUG |
| 2880 |
if (debug) |
| 2881 |
{ |
| 2882 |
DEBUG_PRINT1 ("\nCompiled pattern: \n"); |
| 2883 |
print_compiled_pattern (bufp); |
| 2884 |
} |
| 2885 |
#endif /* DEBUG */ |
| 2886 |
|
| 2887 |
#ifndef MATCH_MAY_ALLOCATE |
| 2888 |
/* Initialize the failure stack to the largest possible stack. This |
| 2889 |
isn't necessary unless we're trying to avoid calling alloca in |
| 2890 |
the search and match routines. */ |
| 2891 |
{ |
| 2892 |
int num_regs = bufp->re_nsub + 1; |
| 2893 |
|
| 2894 |
/* Since DOUBLE_FAIL_STACK refuses to double only if the current size |
| 2895 |
is strictly greater than re_max_failures, the largest possible stack |
| 2896 |
is 2 * re_max_failures failure points. */ |
| 2897 |
if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) |
| 2898 |
{ |
| 2899 |
fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); |
| 2900 |
|
| 2901 |
# ifdef emacs |
| 2902 |
if (! fail_stack.stack) |
| 2903 |
fail_stack.stack |
| 2904 |
= (fail_stack_elt_t *) xmalloc (fail_stack.size |
| 2905 |
* sizeof (fail_stack_elt_t)); |
| 2906 |
else |
| 2907 |
fail_stack.stack |
| 2908 |
= (fail_stack_elt_t *) xrealloc (fail_stack.stack, |
| 2909 |
(fail_stack.size |
| 2910 |
* sizeof (fail_stack_elt_t))); |
| 2911 |
# else /* not emacs */ |
| 2912 |
if (! fail_stack.stack) |
| 2913 |
fail_stack.stack |
| 2914 |
= (fail_stack_elt_t *) malloc (fail_stack.size |
| 2915 |
* sizeof (fail_stack_elt_t)); |
| 2916 |
else |
| 2917 |
fail_stack.stack |
| 2918 |
= (fail_stack_elt_t *) realloc (fail_stack.stack, |
| 2919 |
(fail_stack.size |
| 2920 |
* sizeof (fail_stack_elt_t))); |
| 2921 |
# endif /* not emacs */ |
| 2922 |
} |
| 2923 |
|
| 2924 |
regex_grow_registers (num_regs); |
| 2925 |
} |
| 2926 |
#endif /* not MATCH_MAY_ALLOCATE */ |
| 2927 |
|
| 2928 |
return REG_NOERROR; |
| 2929 |
} /* regex_compile */ |
| 2930 |
|
| 2931 |
/* Subroutines for `regex_compile'. */ |
| 2932 |
|
| 2933 |
/* Store OP at LOC followed by two-byte integer parameter ARG. */ |
| 2934 |
|
| 2935 |
static void |
| 2936 |
store_op1 (op, loc, arg) |
| 2937 |
re_opcode_t op; |
| 2938 |
unsigned char *loc; |
| 2939 |
int arg; |
| 2940 |
{ |
| 2941 |
*loc = (unsigned char) op; |
| 2942 |
STORE_NUMBER (loc + 1, arg); |
| 2943 |
} |
| 2944 |
|
| 2945 |
|
| 2946 |
/* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ |
| 2947 |
|
| 2948 |
static void |
| 2949 |
store_op2 (op, loc, arg1, arg2) |
| 2950 |
re_opcode_t op; |
| 2951 |
unsigned char *loc; |
| 2952 |
int arg1, arg2; |
| 2953 |
{ |
| 2954 |
*loc = (unsigned char) op; |
| 2955 |
STORE_NUMBER (loc + 1, arg1); |
| 2956 |
STORE_NUMBER (loc + 3, arg2); |
| 2957 |
} |
| 2958 |
|
| 2959 |
|
| 2960 |
/* Copy the bytes from LOC to END to open up three bytes of space at LOC |
| 2961 |
for OP followed by two-byte integer parameter ARG. */ |
| 2962 |
|
| 2963 |
static void |
| 2964 |
insert_op1 (op, loc, arg, end) |
| 2965 |
re_opcode_t op; |
| 2966 |
unsigned char *loc; |
| 2967 |
int arg; |
| 2968 |
unsigned char *end; |
| 2969 |
{ |
| 2970 |
register unsigned char *pfrom = end; |
| 2971 |
register unsigned char *pto = end + 3; |
| 2972 |
|
| 2973 |
while (pfrom != loc) |
| 2974 |
*--pto = *--pfrom; |
| 2975 |
|
| 2976 |
store_op1 (op, loc, arg); |
| 2977 |
} |
| 2978 |
|
| 2979 |
|
| 2980 |
/* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ |
| 2981 |
|
| 2982 |
static void |
| 2983 |
insert_op2 (op, loc, arg1, arg2, end) |
| 2984 |
re_opcode_t op; |
| 2985 |
unsigned char *loc; |
| 2986 |
int arg1, arg2; |
| 2987 |
unsigned char *end; |
| 2988 |
{ |
| 2989 |
register unsigned char *pfrom = end; |
| 2990 |
register unsigned char *pto = end + 5; |
| 2991 |
|
| 2992 |
while (pfrom != loc) |
| 2993 |
*--pto = *--pfrom; |
| 2994 |
|
| 2995 |
store_op2 (op, loc, arg1, arg2); |
| 2996 |
} |
| 2997 |
|
| 2998 |
|
| 2999 |
/* P points to just after a ^ in PATTERN. Return true if that ^ comes |
| 3000 |
after an alternative or a begin-subexpression. We assume there is at |
| 3001 |
least one character before the ^. */ |
| 3002 |
|
| 3003 |
static boolean |
| 3004 |
at_begline_loc_p (pattern, p, syntax) |
| 3005 |
const char *pattern, *p; |
| 3006 |
reg_syntax_t syntax; |
| 3007 |
{ |
| 3008 |
const char *prev = p - 2; |
| 3009 |
boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
| 3010 |
|
| 3011 |
return |
| 3012 |
/* After a subexpression? */ |
| 3013 |
(*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) |
| 3014 |
/* After an alternative? */ |
| 3015 |
|| (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); |
| 3016 |
} |
| 3017 |
|
| 3018 |
|
| 3019 |
/* The dual of at_begline_loc_p. This one is for $. We assume there is |
| 3020 |
at least one character after the $, i.e., `P < PEND'. */ |
| 3021 |
|
| 3022 |
static boolean |
| 3023 |
at_endline_loc_p (p, pend, syntax) |
| 3024 |
const char *p, *pend; |
| 3025 |
reg_syntax_t syntax; |
| 3026 |
{ |
| 3027 |
const char *next = p; |
| 3028 |
boolean next_backslash = *next == '\\'; |
| 3029 |
const char *next_next = p + 1 < pend ? p + 1 : 0; |
| 3030 |
|
| 3031 |
return |
| 3032 |
/* Before a subexpression? */ |
| 3033 |
(syntax & RE_NO_BK_PARENS ? *next == ')' |
| 3034 |
: next_backslash && next_next && *next_next == ')') |
| 3035 |
/* Before an alternative? */ |
| 3036 |
|| (syntax & RE_NO_BK_VBAR ? *next == '|' |
| 3037 |
: next_backslash && next_next && *next_next == '|'); |
| 3038 |
} |
| 3039 |
|
| 3040 |
|
| 3041 |
/* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
| 3042 |
false if it's not. */ |
| 3043 |
|
| 3044 |
static boolean |
| 3045 |
group_in_compile_stack (compile_stack, regnum) |
| 3046 |
compile_stack_type compile_stack; |
| 3047 |
regnum_t regnum; |
| 3048 |
{ |
| 3049 |
int this_element; |
| 3050 |
|
| 3051 |
for (this_element = compile_stack.avail - 1; |
| 3052 |
this_element >= 0; |
| 3053 |
this_element--) |
| 3054 |
if (compile_stack.stack[this_element].regnum == regnum) |
| 3055 |
return true; |
| 3056 |
|
| 3057 |
return false; |
| 3058 |
} |
| 3059 |
|
| 3060 |
|
| 3061 |
/* Read the ending character of a range (in a bracket expression) from the |
| 3062 |
uncompiled pattern *P_PTR (which ends at PEND). We assume the |
| 3063 |
starting character is in `P[-2]'. (`P[-1]' is the character `-'.) |
| 3064 |
Then we set the translation of all bits between the starting and |
| 3065 |
ending characters (inclusive) in the compiled pattern B. |
| 3066 |
|
| 3067 |
Return an error code. |
| 3068 |
|
| 3069 |
We use these short variable names so we can use the same macros as |
| 3070 |
`regex_compile' itself. */ |
| 3071 |
|
| 3072 |
static reg_errcode_t |
| 3073 |
compile_range (p_ptr, pend, translate, syntax, b) |
| 3074 |
const char **p_ptr, *pend; |
| 3075 |
RE_TRANSLATE_TYPE translate; |
| 3076 |
reg_syntax_t syntax; |
| 3077 |
unsigned char *b; |
| 3078 |
{ |
| 3079 |
unsigned this_char; |
| 3080 |
|
| 3081 |
const char *p = *p_ptr; |
| 3082 |
unsigned int range_start, range_end; |
| 3083 |
|
| 3084 |
if (p == pend) |
| 3085 |
return REG_ERANGE; |
| 3086 |
|
| 3087 |
/* Even though the pattern is a signed `char *', we need to fetch |
| 3088 |
with unsigned char *'s; if the high bit of the pattern character |
| 3089 |
is set, the range endpoints will be negative if we fetch using a |
| 3090 |
signed char *. |
| 3091 |
|
| 3092 |
We also want to fetch the endpoints without translating them; the |
| 3093 |
appropriate translation is done in the bit-setting loop below. */ |
| 3094 |
/* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */ |
| 3095 |
range_start = ((const unsigned char *) p)[-2]; |
| 3096 |
range_end = ((const unsigned char *) p)[0]; |
| 3097 |
|
| 3098 |
/* Have to increment the pointer into the pattern string, so the |
| 3099 |
caller isn't still at the ending character. */ |
| 3100 |
(*p_ptr)++; |
| 3101 |
|
| 3102 |
/* If the start is after the end, the range is empty. */ |
| 3103 |
if (range_start > range_end) |
| 3104 |
return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; |
| 3105 |
|
| 3106 |
/* Here we see why `this_char' has to be larger than an `unsigned |
| 3107 |
char' -- the range is inclusive, so if `range_end' == 0xff |
| 3108 |
(assuming 8-bit characters), we would otherwise go into an infinite |
| 3109 |
loop, since all characters <= 0xff. */ |
| 3110 |
for (this_char = range_start; this_char <= range_end; this_char++) |
| 3111 |
{ |
| 3112 |
SET_LIST_BIT (TRANSLATE (this_char)); |
| 3113 |
} |
| 3114 |
|
| 3115 |
return REG_NOERROR; |
| 3116 |
} |
| 3117 |
|
| 3118 |
/* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in |
| 3119 |
BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible |
| 3120 |
characters can start a string that matches the pattern. This fastmap |
| 3121 |
is used by re_search to skip quickly over impossible starting points. |
| 3122 |
|
| 3123 |
The caller must supply the address of a (1 << BYTEWIDTH)-byte data |
| 3124 |
area as BUFP->fastmap. |
| 3125 |
|
| 3126 |
We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in |
| 3127 |
the pattern buffer. |
| 3128 |
|
| 3129 |
Returns 0 if we succeed, -2 if an internal error. */ |
| 3130 |
|
| 3131 |
int |
| 3132 |
re_compile_fastmap (bufp) |
| 3133 |
struct re_pattern_buffer *bufp; |
| 3134 |
{ |
| 3135 |
int j, k; |
| 3136 |
#ifdef MATCH_MAY_ALLOCATE |
| 3137 |
fail_stack_type fail_stack; |
| 3138 |
#endif |
| 3139 |
#ifndef REGEX_MALLOC |
| 3140 |
char *destination; |
| 3141 |
#endif |
| 3142 |
|
| 3143 |
register char *fastmap = bufp->fastmap; |
| 3144 |
unsigned char *pattern = bufp->buffer; |
| 3145 |
unsigned char *p = pattern; |
| 3146 |
register unsigned char *pend = pattern + bufp->used; |
| 3147 |
|
| 3148 |
#ifdef REL_ALLOC |
| 3149 |
/* This holds the pointer to the failure stack, when |
| 3150 |
it is allocated relocatably. */ |
| 3151 |
fail_stack_elt_t *failure_stack_ptr; |
| 3152 |
#endif |
| 3153 |
|
| 3154 |
/* Assume that each path through the pattern can be null until |
| 3155 |
proven otherwise. We set this false at the bottom of switch |
| 3156 |
statement, to which we get only if a particular path doesn't |
| 3157 |
match the empty string. */ |
| 3158 |
boolean path_can_be_null = true; |
| 3159 |
|
| 3160 |
/* We aren't doing a `succeed_n' to begin with. */ |
| 3161 |
boolean succeed_n_p = false; |
| 3162 |
|
| 3163 |
assert (fastmap != NULL && p != NULL); |
| 3164 |
|
| 3165 |
INIT_FAIL_STACK (); |
| 3166 |
bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ |
| 3167 |
bufp->fastmap_accurate = 1; /* It will be when we're done. */ |
| 3168 |
bufp->can_be_null = 0; |
| 3169 |
|
| 3170 |
while (1) |
| 3171 |
{ |
| 3172 |
if (p == pend || *p == succeed) |
| 3173 |
{ |
| 3174 |
/* We have reached the (effective) end of pattern. */ |
| 3175 |
if (!FAIL_STACK_EMPTY ()) |
| 3176 |
{ |
| 3177 |
bufp->can_be_null |= path_can_be_null; |
| 3178 |
|
| 3179 |
/* Reset for next path. */ |
| 3180 |
path_can_be_null = true; |
| 3181 |
|
| 3182 |
p = fail_stack.stack[--fail_stack.avail].pointer; |
| 3183 |
|
| 3184 |
continue; |
| 3185 |
} |
| 3186 |
else |
| 3187 |
break; |
| 3188 |
} |
| 3189 |
|
| 3190 |
/* We should never be about to go beyond the end of the pattern. */ |
| 3191 |
assert (p < pend); |
| 3192 |
|
| 3193 |
switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
| 3194 |
{ |
| 3195 |
|
| 3196 |
/* I guess the idea here is to simply not bother with a fastmap |
| 3197 |
if a backreference is used, since it's too hard to figure out |
| 3198 |
the fastmap for the corresponding group. Setting |
| 3199 |
`can_be_null' stops `re_search_2' from using the fastmap, so |
| 3200 |
that is all we do. */ |
| 3201 |
case duplicate: |
| 3202 |
bufp->can_be_null = 1; |
| 3203 |
goto done; |
| 3204 |
|
| 3205 |
|
| 3206 |
/* Following are the cases which match a character. These end |
| 3207 |
with `break'. */ |
| 3208 |
|
| 3209 |
case exactn: |
| 3210 |
fastmap[p[1]] = 1; |
| 3211 |
break; |
| 3212 |
|
| 3213 |
|
| 3214 |
case charset: |
| 3215 |
for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) |
| 3216 |
if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) |
| 3217 |
fastmap[j] = 1; |
| 3218 |
break; |
| 3219 |
|
| 3220 |
|
| 3221 |
case charset_not: |
| 3222 |
/* Chars beyond end of map must be allowed. */ |
| 3223 |
for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) |
| 3224 |
fastmap[j] = 1; |
| 3225 |
|
| 3226 |
for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) |
| 3227 |
if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) |
| 3228 |
fastmap[j] = 1; |
| 3229 |
break; |
| 3230 |
|
| 3231 |
|
| 3232 |
case wordchar: |
| 3233 |
for (j = 0; j < (1 << BYTEWIDTH); j++) |
| 3234 |
if (SYNTAX (j) == Sword) |
| 3235 |
fastmap[j] = 1; |
| 3236 |
break; |
| 3237 |
|
| 3238 |
|
| 3239 |
case notwordchar: |
| 3240 |
for (j = 0; j < (1 << BYTEWIDTH); j++) |
| 3241 |
if (SYNTAX (j) != Sword) |
| 3242 |
fastmap[j] = 1; |
| 3243 |
break; |
| 3244 |
|
| 3245 |
|
| 3246 |
case anychar: |
| 3247 |
{ |
| 3248 |
int fastmap_newline = fastmap['\n']; |
| 3249 |
|
| 3250 |
/* `.' matches anything ... */ |
| 3251 |
for (j = 0; j < (1 << BYTEWIDTH); j++) |
| 3252 |
fastmap[j] = 1; |
| 3253 |
|
| 3254 |
/* ... except perhaps newline. */ |
| 3255 |
if (!(bufp->syntax & RE_DOT_NEWLINE)) |
| 3256 |
fastmap['\n'] = fastmap_newline; |
| 3257 |
|
| 3258 |
/* Return if we have already set `can_be_null'; if we have, |
| 3259 |
then the fastmap is irrelevant. Something's wrong here. */ |
| 3260 |
else if (bufp->can_be_null) |
| 3261 |
goto done; |
| 3262 |
|
| 3263 |
/* Otherwise, have to check alternative paths. */ |
| 3264 |
break; |
| 3265 |
} |
| 3266 |
|
| 3267 |
#ifdef emacs |
| 3268 |
case syntaxspec: |
| 3269 |
k = *p++; |
| 3270 |
for (j = 0; j < (1 << BYTEWIDTH); j++) |
| 3271 |
if (SYNTAX (j) == (enum syntaxcode) k) |
| 3272 |
fastmap[j] = 1; |
| 3273 |
break; |
| 3274 |
|
| 3275 |
|
| 3276 |
case notsyntaxspec: |
| 3277 |
k = *p++; |
| 3278 |
for (j = 0; j < (1 << BYTEWIDTH); j++) |
| 3279 |
if (SYNTAX (j) != (enum syntaxcode) k) |
| 3280 |
fastmap[j] = 1; |
| 3281 |
break; |
| 3282 |
|
| 3283 |
|
| 3284 |
/* All cases after this match the empty string. These end with |
| 3285 |
`continue'. */ |
| 3286 |
|
| 3287 |
|
| 3288 |
case before_dot: |
| 3289 |
case at_dot: |
| 3290 |
case after_dot: |
| 3291 |
continue; |
| 3292 |
#endif /* emacs */ |
| 3293 |
|
| 3294 |
|
| 3295 |
case no_op: |
| 3296 |
case begline: |
| 3297 |
case endline: |
| 3298 |
case begbuf: |
| 3299 |
case endbuf: |
| 3300 |
case wordbound: |
| 3301 |
case notwordbound: |
| 3302 |
case wordbeg: |
| 3303 |
case wordend: |
| 3304 |
case push_dummy_failure: |
| 3305 |
continue; |
| 3306 |
|
| 3307 |
|
| 3308 |
case jump_n: |
| 3309 |
case pop_failure_jump: |
| 3310 |
case maybe_pop_jump: |
| 3311 |
case jump: |
| 3312 |
case jump_past_alt: |
| 3313 |
case dummy_failure_jump: |
| 3314 |
EXTRACT_NUMBER_AND_INCR (j, p); |
| 3315 |
p += j; |
| 3316 |
if (j > 0) |
| 3317 |
continue; |
| 3318 |
|
| 3319 |
/* Jump backward implies we just went through the body of a |
| 3320 |
loop and matched nothing. Opcode jumped to should be |
| 3321 |
`on_failure_jump' or `succeed_n'. Just treat it like an |
| 3322 |
ordinary jump. For a * loop, it has pushed its failure |
| 3323 |
point already; if so, discard that as redundant. */ |
| 3324 |
if ((re_opcode_t) *p != on_failure_jump |
| 3325 |
&& (re_opcode_t) *p != succeed_n) |
| 3326 |
continue; |
| 3327 |
|
| 3328 |
p++; |
| 3329 |
EXTRACT_NUMBER_AND_INCR (j, p); |
| 3330 |
p += j; |
| 3331 |
|
| 3332 |
/* If what's on the stack is where we are now, pop it. */ |
| 3333 |
if (!FAIL_STACK_EMPTY () |
| 3334 |
&& fail_stack.stack[fail_stack.avail - 1].pointer == p) |
| 3335 |
fail_stack.avail--; |
| 3336 |
|
| 3337 |
continue; |
| 3338 |
|
| 3339 |
|
| 3340 |
case on_failure_jump: |
| 3341 |
case on_failure_keep_string_jump: |
| 3342 |
handle_on_failure_jump: |
| 3343 |
EXTRACT_NUMBER_AND_INCR (j, p); |
| 3344 |
|
| 3345 |
/* For some patterns, e.g., `(a?)?', `p+j' here points to the |
| 3346 |
end of the pattern. We don't want to push such a point, |
| 3347 |
since when we restore it above, entering the switch will |
| 3348 |
increment `p' past the end of the pattern. We don't need |
| 3349 |
to push such a point since we obviously won't find any more |
| 3350 |
fastmap entries beyond `pend'. Such a pattern can match |
| 3351 |
the null string, though. */ |
| 3352 |
if (p + j < pend) |
| 3353 |
{ |
| 3354 |
if (!PUSH_PATTERN_OP (p + j, fail_stack)) |
| 3355 |
{ |
| 3356 |
RESET_FAIL_STACK (); |
| 3357 |
return -2; |
| 3358 |
} |
| 3359 |
} |
| 3360 |
else |
| 3361 |
bufp->can_be_null = 1; |
| 3362 |
|
| 3363 |
if (succeed_n_p) |
| 3364 |
{ |
| 3365 |
EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ |
| 3366 |
succeed_n_p = false; |
| 3367 |
} |
| 3368 |
|
| 3369 |
continue; |
| 3370 |
|
| 3371 |
|
| 3372 |
case succeed_n: |
| 3373 |
/* Get to the number of times to succeed. */ |
| 3374 |
p += 2; |
| 3375 |
|
| 3376 |
/* Increment p past the n for when k != 0. */ |
| 3377 |
EXTRACT_NUMBER_AND_INCR (k, p); |
| 3378 |
if (k == 0) |
| 3379 |
{ |
| 3380 |
p -= 4; |
| 3381 |
succeed_n_p = true; /* Spaghetti code alert. */ |
| 3382 |
goto handle_on_failure_jump; |
| 3383 |
} |
| 3384 |
continue; |
| 3385 |
|
| 3386 |
|
| 3387 |
case set_number_at: |
| 3388 |
p += 4; |
| 3389 |
continue; |
| 3390 |
|
| 3391 |
|
| 3392 |
case start_memory: |
| 3393 |
case stop_memory: |
| 3394 |
p += 2; |
| 3395 |
continue; |
| 3396 |
|
| 3397 |
|
| 3398 |
default: |
| 3399 |
abort (); /* We have listed all the cases. */ |
| 3400 |
} /* switch *p++ */ |
| 3401 |
|
| 3402 |
/* Getting here means we have found the possible starting |
| 3403 |
characters for one path of the pattern -- and that the empty |
| 3404 |
string does not match. We need not follow this path further. |
| 3405 |
Instead, look at the next alternative (remembered on the |
| 3406 |
stack), or quit if no more. The test at the top of the loop |
| 3407 |
does these things. */ |
| 3408 |
path_can_be_null = false; |
| 3409 |
p = pend; |
| 3410 |
} /* while p */ |
| 3411 |
|
| 3412 |
/* Set `can_be_null' for the last path (also the first path, if the |
| 3413 |
pattern is empty). */ |
| 3414 |
bufp->can_be_null |= path_can_be_null; |
| 3415 |
|
| 3416 |
done: |
| 3417 |
RESET_FAIL_STACK (); |
| 3418 |
return 0; |
| 3419 |
} /* re_compile_fastmap */ |
| 3420 |
#ifdef _LIBC |
| 3421 |
weak_alias (__re_compile_fastmap, re_compile_fastmap) |
| 3422 |
#endif |
| 3423 |
|
| 3424 |
/* Set REGS to hold NUM_REGS registers, storing them in STARTS and |
| 3425 |
ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use |
| 3426 |
this memory for recording register information. STARTS and ENDS |
| 3427 |
must be allocated using the malloc library routine, and must each |
| 3428 |
be at least NUM_REGS * sizeof (regoff_t) bytes long. |
| 3429 |
|
| 3430 |
If NUM_REGS == 0, then subsequent matches should allocate their own |
| 3431 |
register data. |
| 3432 |
|
| 3433 |
Unless this function is called, the first search or match using |
| 3434 |
PATTERN_BUFFER will allocate its own register data, without |
| 3435 |
freeing the old data. */ |
| 3436 |
|
| 3437 |
void |
| 3438 |
re_set_registers (bufp, regs, num_regs, starts, ends) |
| 3439 |
struct re_pattern_buffer *bufp; |
| 3440 |
struct re_registers *regs; |
| 3441 |
unsigned num_regs; |
| 3442 |
regoff_t *starts, *ends; |
| 3443 |
{ |
| 3444 |
if (num_regs) |
| 3445 |
{ |
| 3446 |
bufp->regs_allocated = REGS_REALLOCATE; |
| 3447 |
regs->num_regs = num_regs; |
| 3448 |
regs->start = starts; |
| 3449 |
regs->end = ends; |
| 3450 |
} |
| 3451 |
else |
| 3452 |
{ |
| 3453 |
bufp->regs_allocated = REGS_UNALLOCATED; |
| 3454 |
regs->num_regs = 0; |
| 3455 |
regs->start = regs->end = (regoff_t *) 0; |
| 3456 |
} |
| 3457 |
} |
| 3458 |
#ifdef _LIBC |
| 3459 |
weak_alias (__re_set_registers, re_set_registers) |
| 3460 |
#endif |
| 3461 |
|
| 3462 |
/* Searching routines. */ |
| 3463 |
|
| 3464 |
/* Like re_search_2, below, but only one string is specified, and |
| 3465 |
doesn't let you say where to stop matching. */ |
| 3466 |
|
| 3467 |
int |
| 3468 |
re_search (bufp, string, size, startpos, range, regs) |
| 3469 |
struct re_pattern_buffer *bufp; |
| 3470 |
const char *string; |
| 3471 |
int size, startpos, range; |
| 3472 |
struct re_registers *regs; |
| 3473 |
{ |
| 3474 |
return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
| 3475 |
regs, size); |
| 3476 |
} |
| 3477 |
#ifdef _LIBC |
| 3478 |
weak_alias (__re_search, re_search) |
| 3479 |
#endif |
| 3480 |
|
| 3481 |
|
| 3482 |
/* Using the compiled pattern in BUFP->buffer, first tries to match the |
| 3483 |
virtual concatenation of STRING1 and STRING2, starting first at index |
| 3484 |
STARTPOS, then at STARTPOS + 1, and so on. |
| 3485 |
|
| 3486 |
STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
| 3487 |
|
| 3488 |
RANGE is how far to scan while trying to match. RANGE = 0 means try |
| 3489 |
only at STARTPOS; in general, the last start tried is STARTPOS + |
| 3490 |
RANGE. |
| 3491 |
|
| 3492 |
In REGS, return the indices of the virtual concatenation of STRING1 |
| 3493 |
and STRING2 that matched the entire BUFP->buffer and its contained |
| 3494 |
subexpressions. |
| 3495 |
|
| 3496 |
Do not consider matching one past the index STOP in the virtual |
| 3497 |
concatenation of STRING1 and STRING2. |
| 3498 |
|
| 3499 |
We return either the position in the strings at which the match was |
| 3500 |
found, -1 if no match, or -2 if error (such as failure |
| 3501 |
stack overflow). */ |
| 3502 |
|
| 3503 |
int |
| 3504 |
re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop) |
| 3505 |
struct re_pattern_buffer *bufp; |
| 3506 |
const char *string1, *string2; |
| 3507 |
int size1, size2; |
| 3508 |
int startpos; |
| 3509 |
int range; |
| 3510 |
struct re_registers *regs; |
| 3511 |
int stop; |
| 3512 |
{ |
| 3513 |
int val; |
| 3514 |
register char *fastmap = bufp->fastmap; |
| 3515 |
register RE_TRANSLATE_TYPE translate = bufp->translate; |
| 3516 |
int total_size = size1 + size2; |
| 3517 |
int endpos = startpos + range; |
| 3518 |
|
| 3519 |
/* Check for out-of-range STARTPOS. */ |
| 3520 |
if (startpos < 0 || startpos > total_size) |
| 3521 |
return -1; |
| 3522 |
|
| 3523 |
/* Fix up RANGE if it might eventually take us outside |
| 3524 |
the virtual concatenation of STRING1 and STRING2. |
| 3525 |
Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
| 3526 |
if (endpos < 0) |
| 3527 |
range = 0 - startpos; |
| 3528 |
else if (endpos > total_size) |
| 3529 |
range = total_size - startpos; |
| 3530 |
|
| 3531 |
/* If the search isn't to be a backwards one, don't waste time in a |
| 3532 |
search for a pattern that must be anchored. */ |
| 3533 |
if (bufp->used > 0 && range > 0 |
| 3534 |
&& ((re_opcode_t) bufp->buffer[0] == begbuf |
| 3535 |
/* `begline' is like `begbuf' if it cannot match at newlines. */ |
| 3536 |
|| ((re_opcode_t) bufp->buffer[0] == begline |
| 3537 |
&& !bufp->newline_anchor))) |
| 3538 |
{ |
| 3539 |
if (startpos > 0) |
| 3540 |
return -1; |
| 3541 |
else |
| 3542 |
range = 1; |
| 3543 |
} |
| 3544 |
|
| 3545 |
#ifdef emacs |
| 3546 |
/* In a forward search for something that starts with \=. |
| 3547 |
don't keep searching past point. */ |
| 3548 |
if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) |
| 3549 |
{ |
| 3550 |
range = PT - startpos; |
| 3551 |
if (range <= 0) |
| 3552 |
return -1; |
| 3553 |
} |
| 3554 |
#endif /* emacs */ |
| 3555 |
|
| 3556 |
/* Update the fastmap now if not correct already. */ |
| 3557 |
if (fastmap && !bufp->fastmap_accurate) |
| 3558 |
if (re_compile_fastmap (bufp) == -2) |
| 3559 |
return -2; |
| 3560 |
|
| 3561 |
/* Loop through the string, looking for a place to start matching. */ |
| 3562 |
for (;;) |
| 3563 |
{ |
| 3564 |
/* If a fastmap is supplied, skip quickly over characters that |
| 3565 |
cannot be the start of a match. If the pattern can match the |
| 3566 |
null string, however, we don't need to skip characters; we want |
| 3567 |
the first null string. */ |
| 3568 |
if (fastmap && startpos < total_size && !bufp->can_be_null) |
| 3569 |
{ |
| 3570 |
if (range > 0) /* Searching forwards. */ |
| 3571 |
{ |
| 3572 |
register const char *d; |
| 3573 |
register int lim = 0; |
| 3574 |
int irange = range; |
| 3575 |
|
| 3576 |
if (startpos < size1 && startpos + range >= size1) |
| 3577 |
lim = range - (size1 - startpos); |
| 3578 |
|
| 3579 |
d = (startpos >= size1 ? string2 - size1 : string1) + startpos; |
| 3580 |
|
| 3581 |
/* Written out as an if-else to avoid testing `translate' |
| 3582 |
inside the loop. */ |
| 3583 |
if (translate) |
| 3584 |
while (range > lim |
| 3585 |
&& !fastmap[(unsigned char) |
| 3586 |
translate[(unsigned char) *d++]]) |
| 3587 |
range--; |
| 3588 |
else |
| 3589 |
while (range > lim && !fastmap[(unsigned char) *d++]) |
| 3590 |
range--; |
| 3591 |
|
| 3592 |
startpos += irange - range; |
| 3593 |
} |
| 3594 |
else /* Searching backwards. */ |
| 3595 |
{ |
| 3596 |
register char c = (size1 == 0 || startpos >= size1 |
| 3597 |
? string2[startpos - size1] |
| 3598 |
: string1[startpos]); |
| 3599 |
|
| 3600 |
if (!fastmap[(unsigned char) TRANSLATE (c)]) |
| 3601 |
goto advance; |
| 3602 |
} |
| 3603 |
} |
| 3604 |
|
| 3605 |
/* If can't match the null string, and that's all we have left, fail. */ |
| 3606 |
if (range >= 0 && startpos == total_size && fastmap |
| 3607 |
&& !bufp->can_be_null) |
| 3608 |
return -1; |
| 3609 |
|
| 3610 |
val = re_match_2_internal (bufp, string1, size1, string2, size2, |
| 3611 |
startpos, regs, stop); |
| 3612 |
#ifndef REGEX_MALLOC |
| 3613 |
# ifdef C_ALLOCA |
| 3614 |
alloca (0); |
| 3615 |
# endif |
| 3616 |
#endif |
| 3617 |
|
| 3618 |
if (val >= 0) |
| 3619 |
return startpos; |
| 3620 |
|
| 3621 |
if (val == -2) |
| 3622 |
return -2; |
| 3623 |
|
| 3624 |
advance: |
| 3625 |
if (!range) |
| 3626 |
break; |
| 3627 |
else if (range > 0) |
| 3628 |
{ |
| 3629 |
range--; |
| 3630 |
startpos++; |
| 3631 |
} |
| 3632 |
else |
| 3633 |
{ |
| 3634 |
range++; |
| 3635 |
startpos--; |
| 3636 |
} |
| 3637 |
} |
| 3638 |
return -1; |
| 3639 |
} /* re_search_2 */ |
| 3640 |
#ifdef _LIBC |
| 3641 |
weak_alias (__re_search_2, re_search_2) |
| 3642 |
#endif |
| 3643 |
|
| 3644 |
/* This converts PTR, a pointer into one of the search strings `string1' |
| 3645 |
and `string2' into an offset from the beginning of that string. */ |
| 3646 |
#define POINTER_TO_OFFSET(ptr) \ |
| 3647 |
(FIRST_STRING_P (ptr) \ |
| 3648 |
? ((regoff_t) ((ptr) - string1)) \ |
| 3649 |
: ((regoff_t) ((ptr) - string2 + size1))) |
| 3650 |
|
| 3651 |
/* Macros for dealing with the split strings in re_match_2. */ |
| 3652 |
|
| 3653 |
#define MATCHING_IN_FIRST_STRING (dend == end_match_1) |
| 3654 |
|
| 3655 |
/* Call before fetching a character with *d. This switches over to |
| 3656 |
string2 if necessary. */ |
| 3657 |
#define PREFETCH() \ |
| 3658 |
while (d == dend) \ |
| 3659 |
{ \ |
| 3660 |
/* End of string2 => fail. */ \ |
| 3661 |
if (dend == end_match_2) \ |
| 3662 |
goto fail; \ |
| 3663 |
/* End of string1 => advance to string2. */ \ |
| 3664 |
d = string2; \ |
| 3665 |
dend = end_match_2; \ |
| 3666 |
} |
| 3667 |
|
| 3668 |
|
| 3669 |
/* Test if at very beginning or at very end of the virtual concatenation |
| 3670 |
of `string1' and `string2'. If only one string, it's `string2'. */ |
| 3671 |
#define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
| 3672 |
#define AT_STRINGS_END(d) ((d) == end2) |
| 3673 |
|
| 3674 |
|
| 3675 |
/* Test if D points to a character which is word-constituent. We have |
| 3676 |
two special cases to check for: if past the end of string1, look at |
| 3677 |
the first character in string2; and if before the beginning of |
| 3678 |
string2, look at the last character in string1. */ |
| 3679 |
#define WORDCHAR_P(d) \ |
| 3680 |
(SYNTAX ((d) == end1 ? *string2 \ |
| 3681 |
: (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
| 3682 |
== Sword) |
| 3683 |
|
| 3684 |
/* Disabled due to a compiler bug -- see comment at case wordbound */ |
| 3685 |
#if 0 |
| 3686 |
/* Test if the character before D and the one at D differ with respect |
| 3687 |
to being word-constituent. */ |
| 3688 |
#define AT_WORD_BOUNDARY(d) \ |
| 3689 |
(AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ |
| 3690 |
|| WORDCHAR_P (d - 1) != WORDCHAR_P (d)) |
| 3691 |
#endif |
| 3692 |
|
| 3693 |
/* Free everything we malloc. */ |
| 3694 |
#ifdef MATCH_MAY_ALLOCATE |
| 3695 |
# define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL |
| 3696 |
# define FREE_VARIABLES() \ |
| 3697 |
do { \ |
| 3698 |
REGEX_FREE_STACK (fail_stack.stack); \ |
| 3699 |
FREE_VAR (regstart); \ |
| 3700 |
FREE_VAR (regend); \ |
| 3701 |
FREE_VAR (old_regstart); \ |
| 3702 |
FREE_VAR (old_regend); \ |
| 3703 |
FREE_VAR (best_regstart); \ |
| 3704 |
FREE_VAR (best_regend); \ |
| 3705 |
FREE_VAR (reg_info); \ |
| 3706 |
FREE_VAR (reg_dummy); \ |
| 3707 |
FREE_VAR (reg_info_dummy); \ |
| 3708 |
} while (0) |
| 3709 |
#else |
| 3710 |
# define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ |
| 3711 |
#endif /* not MATCH_MAY_ALLOCATE */ |
| 3712 |
|
| 3713 |
/* These values must meet several constraints. They must not be valid |
| 3714 |
register values; since we have a limit of 255 registers (because |
| 3715 |
we use only one byte in the pattern for the register number), we can |
| 3716 |
use numbers larger than 255. They must differ by 1, because of |
| 3717 |
NUM_FAILURE_ITEMS above. And the value for the lowest register must |
| 3718 |
be larger than the value for the highest register, so we do not try |
| 3719 |
to actually save any registers when none are active. */ |
| 3720 |
#define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) |
| 3721 |
#define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) |
| 3722 |
|
| 3723 |
/* Matching routines. */ |
| 3724 |
|
| 3725 |
#ifndef emacs /* Emacs never uses this. */ |
| 3726 |
/* re_match is like re_match_2 except it takes only a single string. */ |
| 3727 |
|
| 3728 |
int |
| 3729 |
re_match (bufp, string, size, pos, regs) |
| 3730 |
struct re_pattern_buffer *bufp; |
| 3731 |
const char *string; |
| 3732 |
int size, pos; |
| 3733 |
struct re_registers *regs; |
| 3734 |
{ |
| 3735 |
int result = re_match_2_internal (bufp, NULL, 0, string, size, |
| 3736 |
pos, regs, size); |
| 3737 |
# ifndef REGEX_MALLOC |
| 3738 |
# ifdef C_ALLOCA |
| 3739 |
alloca (0); |
| 3740 |
# endif |
| 3741 |
# endif |
| 3742 |
return result; |
| 3743 |
} |
| 3744 |
# ifdef _LIBC |
| 3745 |
weak_alias (__re_match, re_match) |
| 3746 |
# endif |
| 3747 |
#endif /* not emacs */ |
| 3748 |
|
| 3749 |
static boolean group_match_null_string_p _RE_ARGS ((unsigned char **p, |
| 3750 |
unsigned char *end, |
| 3751 |
register_info_type *reg_info)); |
| 3752 |
static boolean alt_match_null_string_p _RE_ARGS ((unsigned char *p, |
| 3753 |
unsigned char *end, |
| 3754 |
register_info_type *reg_info)); |
| 3755 |
static boolean common_op_match_null_string_p _RE_ARGS ((unsigned char **p, |
| 3756 |
unsigned char *end, |
| 3757 |
register_info_type *reg_info)); |
| 3758 |
static int bcmp_translate _RE_ARGS ((const char *s1, const char *s2, |
| 3759 |
int len, char *translate)); |
| 3760 |
|
| 3761 |
/* re_match_2 matches the compiled pattern in BUFP against the |
| 3762 |
the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 |
| 3763 |
and SIZE2, respectively). We start matching at POS, and stop |
| 3764 |
matching at STOP. |
| 3765 |
|
| 3766 |
If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
| 3767 |
store offsets for the substring each group matched in REGS. See the |
| 3768 |
documentation for exactly how many groups we fill. |
| 3769 |
|
| 3770 |
We return -1 if no match, -2 if an internal error (such as the |
| 3771 |
failure stack overflowing). Otherwise, we return the length of the |
| 3772 |
matched substring. */ |
| 3773 |
|
| 3774 |
int |
| 3775 |
re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) |
| 3776 |
struct re_pattern_buffer *bufp; |
| 3777 |
const char *string1, *string2; |
| 3778 |
int size1, size2; |
| 3779 |
int pos; |
| 3780 |
struct re_registers *regs; |
| 3781 |
int stop; |
| 3782 |
{ |
| 3783 |
int result = re_match_2_internal (bufp, string1, size1, string2, size2, |
| 3784 |
pos, regs, stop); |
| 3785 |
#ifndef REGEX_MALLOC |
| 3786 |
# ifdef C_ALLOCA |
| 3787 |
alloca (0); |
| 3788 |
# endif |
| 3789 |
#endif |
| 3790 |
return result; |
| 3791 |
} |
| 3792 |
#ifdef _LIBC |
| 3793 |
weak_alias (__re_match_2, re_match_2) |
| 3794 |
#endif |
| 3795 |
|
| 3796 |
/* This is a separate function so that we can force an alloca cleanup |
| 3797 |
afterwards. */ |
| 3798 |
static int |
| 3799 |
re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop) |
| 3800 |
struct re_pattern_buffer *bufp; |
| 3801 |
const char *string1, *string2; |
| 3802 |
int size1, size2; |
| 3803 |
int pos; |
| 3804 |
struct re_registers *regs; |
| 3805 |
int stop; |
| 3806 |
{ |
| 3807 |
/* General temporaries. */ |
| 3808 |
int mcnt; |
| 3809 |
unsigned char *p1; |
| 3810 |
|
| 3811 |
/* Just past the end of the corresponding string. */ |
| 3812 |
const char *end1, *end2; |
| 3813 |
|
| 3814 |
/* Pointers into string1 and string2, just past the last characters in |
| 3815 |
each to consider matching. */ |
| 3816 |
const char *end_match_1, *end_match_2; |
| 3817 |
|
| 3818 |
/* Where we are in the data, and the end of the current string. */ |
| 3819 |
const char *d, *dend; |
| 3820 |
|
| 3821 |
/* Where we are in the pattern, and the end of the pattern. */ |
| 3822 |
unsigned char *p = bufp->buffer; |
| 3823 |
register unsigned char *pend = p + bufp->used; |
| 3824 |
|
| 3825 |
/* Mark the opcode just after a start_memory, so we can test for an |
| 3826 |
empty subpattern when we get to the stop_memory. */ |
| 3827 |
unsigned char *just_past_start_mem = 0; |
| 3828 |
|
| 3829 |
/* We use this to map every character in the string. */ |
| 3830 |
RE_TRANSLATE_TYPE translate = bufp->translate; |
| 3831 |
|
| 3832 |
/* Failure point stack. Each place that can handle a failure further |
| 3833 |
down the line pushes a failure point on this stack. It consists of |
| 3834 |
restart, regend, and reg_info for all registers corresponding to |
| 3835 |
the subexpressions we're currently inside, plus the number of such |
| 3836 |
registers, and, finally, two char *'s. The first char * is where |
| 3837 |
to resume scanning the pattern; the second one is where to resume |
| 3838 |
scanning the strings. If the latter is zero, the failure point is |
| 3839 |
a ``dummy''; if a failure happens and the failure point is a dummy, |
| 3840 |
it gets discarded and the next next one is tried. */ |
| 3841 |
#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
| 3842 |
fail_stack_type fail_stack; |
| 3843 |
#endif |
| 3844 |
#ifdef DEBUG |
| 3845 |
static unsigned failure_id = 0; |
| 3846 |
unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
| 3847 |
#endif |
| 3848 |
|
| 3849 |
#ifdef REL_ALLOC |
| 3850 |
/* This holds the pointer to the failure stack, when |
| 3851 |
it is allocated relocatably. */ |
| 3852 |
fail_stack_elt_t *failure_stack_ptr; |
| 3853 |
#endif |
| 3854 |
|
| 3855 |
/* We fill all the registers internally, independent of what we |
| 3856 |
return, for use in backreferences. The number here includes |
| 3857 |
an element for register zero. */ |
| 3858 |
size_t num_regs = bufp->re_nsub + 1; |
| 3859 |
|
| 3860 |
/* The currently active registers. */ |
| 3861 |
active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
| 3862 |
active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG; |
| 3863 |
|
| 3864 |
/* Information on the contents of registers. These are pointers into |
| 3865 |
the input strings; they record just what was matched (on this |
| 3866 |
attempt) by a subexpression part of the pattern, that is, the |
| 3867 |
regnum-th regstart pointer points to where in the pattern we began |
| 3868 |
matching and the regnum-th regend points to right after where we |
| 3869 |
stopped matching the regnum-th subexpression. (The zeroth register |
| 3870 |
keeps track of what the whole pattern matches.) */ |
| 3871 |
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
| 3872 |
const char **regstart, **regend; |
| 3873 |
#endif |
| 3874 |
|
| 3875 |
/* If a group that's operated upon by a repetition operator fails to |
| 3876 |
match anything, then the register for its start will need to be |
| 3877 |
restored because it will have been set to wherever in the string we |
| 3878 |
are when we last see its open-group operator. Similarly for a |
| 3879 |
register's end. */ |
| 3880 |
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
| 3881 |
const char **old_regstart, **old_regend; |
| 3882 |
#endif |
| 3883 |
|
| 3884 |
/* The is_active field of reg_info helps us keep track of which (possibly |
| 3885 |
nested) subexpressions we are currently in. The matched_something |
| 3886 |
field of reg_info[reg_num] helps us tell whether or not we have |
| 3887 |
matched any of the pattern so far this time through the reg_num-th |
| 3888 |
subexpression. These two fields get reset each time through any |
| 3889 |
loop their register is in. */ |
| 3890 |
#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
| 3891 |
register_info_type *reg_info; |
| 3892 |
#endif |
| 3893 |
|
| 3894 |
/* The following record the register info as found in the above |
| 3895 |
variables when we find a match better than any we've seen before. |
| 3896 |
This happens as we backtrack through the failure points, which in |
| 3897 |
turn happens only if we have not yet matched the entire string. */ |
| 3898 |
unsigned best_regs_set = false; |
| 3899 |
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
| 3900 |
const char **best_regstart, **best_regend; |
| 3901 |
#endif |
| 3902 |
|
| 3903 |
/* Logically, this is `best_regend[0]'. But we don't want to have to |
| 3904 |
allocate space for that if we're not allocating space for anything |
| 3905 |
else (see below). Also, we never need info about register 0 for |
| 3906 |
any of the other register vectors, and it seems rather a kludge to |
| 3907 |
treat `best_regend' differently than the rest. So we keep track of |
| 3908 |
the end of the best match so far in a separate variable. We |
| 3909 |
initialize this to NULL so that when we backtrack the first time |
| 3910 |
and need to test it, it's not garbage. */ |
| 3911 |
const char *match_end = NULL; |
| 3912 |
|
| 3913 |
/* This helps SET_REGS_MATCHED avoid doing redundant work. */ |
| 3914 |
int set_regs_matched_done = 0; |
| 3915 |
|
| 3916 |
/* Used when we pop values we don't care about. */ |
| 3917 |
#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
| 3918 |
const char **reg_dummy; |
| 3919 |
register_info_type *reg_info_dummy; |
| 3920 |
#endif |
| 3921 |
|
| 3922 |
#ifdef DEBUG |
| 3923 |
/* Counts the total number of registers pushed. */ |
| 3924 |
unsigned num_regs_pushed = 0; |
| 3925 |
#endif |
| 3926 |
|
| 3927 |
DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); |
| 3928 |
|
| 3929 |
INIT_FAIL_STACK (); |
| 3930 |
|
| 3931 |
#ifdef MATCH_MAY_ALLOCATE |
| 3932 |
/* Do not bother to initialize all the register variables if there are |
| 3933 |
no groups in the pattern, as it takes a fair amount of time. If |
| 3934 |
there are groups, we include space for register 0 (the whole |
| 3935 |
pattern), even though we never use it, since it simplifies the |
| 3936 |
array indexing. We should fix this. */ |
| 3937 |
if (bufp->re_nsub) |
| 3938 |
{ |
| 3939 |
regstart = REGEX_TALLOC (num_regs, const char *); |
| 3940 |
regend = REGEX_TALLOC (num_regs, const char *); |
| 3941 |
old_regstart = REGEX_TALLOC (num_regs, const char *); |
| 3942 |
old_regend = REGEX_TALLOC (num_regs, const char *); |
| 3943 |
best_regstart = REGEX_TALLOC (num_regs, const char *); |
| 3944 |
best_regend = REGEX_TALLOC (num_regs, const char *); |
| 3945 |
reg_info = REGEX_TALLOC (num_regs, register_info_type); |
| 3946 |
reg_dummy = REGEX_TALLOC (num_regs, const char *); |
| 3947 |
reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type); |
| 3948 |
|
| 3949 |
if (!(regstart && regend && old_regstart && old_regend && reg_info |
| 3950 |
&& best_regstart && best_regend && reg_dummy && reg_info_dummy)) |
| 3951 |
{ |
| 3952 |
FREE_VARIABLES (); |
| 3953 |
return -2; |
| 3954 |
} |
| 3955 |
} |
| 3956 |
else |
| 3957 |
{ |
| 3958 |
/* We must initialize all our variables to NULL, so that |
| 3959 |
`FREE_VARIABLES' doesn't try to free them. */ |
| 3960 |
regstart = regend = old_regstart = old_regend = best_regstart |
| 3961 |
= best_regend = reg_dummy = NULL; |
| 3962 |
reg_info = reg_info_dummy = (register_info_type *) NULL; |
| 3963 |
} |
| 3964 |
#endif /* MATCH_MAY_ALLOCATE */ |
| 3965 |
|
| 3966 |
/* The starting position is bogus. */ |
| 3967 |
if (pos < 0 || pos > size1 + size2) |
| 3968 |
{ |
| 3969 |
FREE_VARIABLES (); |
| 3970 |
return -1; |
| 3971 |
} |
| 3972 |
|
| 3973 |
/* Initialize subexpression text positions to -1 to mark ones that no |
| 3974 |
start_memory/stop_memory has been seen for. Also initialize the |
| 3975 |
register information struct. */ |
| 3976 |
for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) |
| 3977 |
{ |
| 3978 |
regstart[mcnt] = regend[mcnt] |
| 3979 |
= old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; |
| 3980 |
|
| 3981 |
REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; |
| 3982 |
IS_ACTIVE (reg_info[mcnt]) = 0; |
| 3983 |
MATCHED_SOMETHING (reg_info[mcnt]) = 0; |
| 3984 |
EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; |
| 3985 |
} |
| 3986 |
|
| 3987 |
/* We move `string1' into `string2' if the latter's empty -- but not if |
| 3988 |
`string1' is null. */ |
| 3989 |
if (size2 == 0 && string1 != NULL) |
| 3990 |
{ |
| 3991 |
string2 = string1; |
| 3992 |
size2 = size1; |
| 3993 |
string1 = 0; |
| 3994 |
size1 = 0; |
| 3995 |
} |
| 3996 |
end1 = string1 + size1; |
| 3997 |
end2 = string2 + size2; |
| 3998 |
|
| 3999 |
/* Compute where to stop matching, within the two strings. */ |
| 4000 |
if (stop <= size1) |
| 4001 |
{ |
| 4002 |
end_match_1 = string1 + stop; |
| 4003 |
end_match_2 = string2; |
| 4004 |
} |
| 4005 |
else |
| 4006 |
{ |
| 4007 |
end_match_1 = end1; |
| 4008 |
end_match_2 = string2 + stop - size1; |
| 4009 |
} |
| 4010 |
|
| 4011 |
/* `p' scans through the pattern as `d' scans through the data. |
| 4012 |
`dend' is the end of the input string that `d' points within. `d' |
| 4013 |
is advanced into the following input string whenever necessary, but |
| 4014 |
this happens before fetching; therefore, at the beginning of the |
| 4015 |
loop, `d' can be pointing at the end of a string, but it cannot |
| 4016 |
equal `string2'. */ |
| 4017 |
if (size1 > 0 && pos <= size1) |
| 4018 |
{ |
| 4019 |
d = string1 + pos; |
| 4020 |
dend = end_match_1; |
| 4021 |
} |
| 4022 |
else |
| 4023 |
{ |
| 4024 |
d = string2 + pos - size1; |
| 4025 |
dend = end_match_2; |
| 4026 |
} |
| 4027 |
|
| 4028 |
DEBUG_PRINT1 ("The compiled pattern is:\n"); |
| 4029 |
DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); |
| 4030 |
DEBUG_PRINT1 ("The string to match is: `"); |
| 4031 |
DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); |
| 4032 |
DEBUG_PRINT1 ("'\n"); |
| 4033 |
|
| 4034 |
/* This loops over pattern commands. It exits by returning from the |
| 4035 |
function if the match is complete, or it drops through if the match |
| 4036 |
fails at this starting point in the input data. */ |
| 4037 |
for (;;) |
| 4038 |
{ |
| 4039 |
#ifdef _LIBC |
| 4040 |
DEBUG_PRINT2 ("\n%p: ", p); |
| 4041 |
#else |
| 4042 |
DEBUG_PRINT2 ("\n0x%x: ", p); |
| 4043 |
#endif |
| 4044 |
|
| 4045 |
if (p == pend) |
| 4046 |
{ /* End of pattern means we might have succeeded. */ |
| 4047 |
DEBUG_PRINT1 ("end of pattern ... "); |
| 4048 |
|
| 4049 |
/* If we haven't matched the entire string, and we want the |
| 4050 |
longest match, try backtracking. */ |
| 4051 |
if (d != end_match_2) |
| 4052 |
{ |
| 4053 |
/* 1 if this match ends in the same string (string1 or string2) |
| 4054 |
as the best previous match. */ |
| 4055 |
boolean same_str_p = (FIRST_STRING_P (match_end) |
| 4056 |
== MATCHING_IN_FIRST_STRING); |
| 4057 |
/* 1 if this match is the best seen so far. */ |
| 4058 |
boolean best_match_p; |
| 4059 |
|
| 4060 |
/* AIX compiler got confused when this was combined |
| 4061 |
with the previous declaration. */ |
| 4062 |
if (same_str_p) |
| 4063 |
best_match_p = d > match_end; |
| 4064 |
else |
| 4065 |
best_match_p = !MATCHING_IN_FIRST_STRING; |
| 4066 |
|
| 4067 |
DEBUG_PRINT1 ("backtracking.\n"); |
| 4068 |
|
| 4069 |
if (!FAIL_STACK_EMPTY ()) |
| 4070 |
{ /* More failure points to try. */ |
| 4071 |
|
| 4072 |
/* If exceeds best match so far, save it. */ |
| 4073 |
if (!best_regs_set || best_match_p) |
| 4074 |
{ |
| 4075 |
best_regs_set = true; |
| 4076 |
match_end = d; |
| 4077 |
|
| 4078 |
DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); |
| 4079 |
|
| 4080 |
for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) |
| 4081 |
{ |
| 4082 |
best_regstart[mcnt] = regstart[mcnt]; |
| 4083 |
best_regend[mcnt] = regend[mcnt]; |
| 4084 |
} |
| 4085 |
} |
| 4086 |
goto fail; |
| 4087 |
} |
| 4088 |
|
| 4089 |
/* If no failure points, don't restore garbage. And if |
| 4090 |
last match is real best match, don't restore second |
| 4091 |
best one. */ |
| 4092 |
else if (best_regs_set && !best_match_p) |
| 4093 |
{ |
| 4094 |
restore_best_regs: |
| 4095 |
/* Restore best match. It may happen that `dend == |
| 4096 |
end_match_1' while the restored d is in string2. |
| 4097 |
For example, the pattern `x.*y.*z' against the |
| 4098 |
strings `x-' and `y-z-', if the two strings are |
| 4099 |
not consecutive in memory. */ |
| 4100 |
DEBUG_PRINT1 ("Restoring best registers.\n"); |
| 4101 |
|
| 4102 |
d = match_end; |
| 4103 |
dend = ((d >= string1 && d <= end1) |
| 4104 |
? end_match_1 : end_match_2); |
| 4105 |
|
| 4106 |
for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) |
| 4107 |
{ |
| 4108 |
regstart[mcnt] = best_regstart[mcnt]; |
| 4109 |
regend[mcnt] = best_regend[mcnt]; |
| 4110 |
} |
| 4111 |
} |
| 4112 |
} /* d != end_match_2 */ |
| 4113 |
|
| 4114 |
succeed_label: |
| 4115 |
DEBUG_PRINT1 ("Accepting match.\n"); |
| 4116 |
|
| 4117 |
/* If caller wants register contents data back, do it. */ |
| 4118 |
if (regs && !bufp->no_sub) |
| 4119 |
{ |
| 4120 |
/* Have the register data arrays been allocated? */ |
| 4121 |
if (bufp->regs_allocated == REGS_UNALLOCATED) |
| 4122 |
{ /* No. So allocate them with malloc. We need one |
| 4123 |
extra element beyond `num_regs' for the `-1' marker |
| 4124 |
GNU code uses. */ |
| 4125 |
regs->num_regs = MAX (RE_NREGS, num_regs + 1); |
| 4126 |
regs->start = TALLOC (regs->num_regs, regoff_t); |
| 4127 |
regs->end = TALLOC (regs->num_regs, regoff_t); |
| 4128 |
if (regs->start == NULL || regs->end == NULL) |
| 4129 |
{ |
| 4130 |
FREE_VARIABLES (); |
| 4131 |
return -2; |
| 4132 |
} |
| 4133 |
bufp->regs_allocated = REGS_REALLOCATE; |
| 4134 |
} |
| 4135 |
else if (bufp->regs_allocated == REGS_REALLOCATE) |
| 4136 |
{ /* Yes. If we need more elements than were already |
| 4137 |
allocated, reallocate them. If we need fewer, just |
| 4138 |
leave it alone. */ |
| 4139 |
if (regs->num_regs < num_regs + 1) |
| 4140 |
{ |
| 4141 |
regs->num_regs = num_regs + 1; |
| 4142 |
RETALLOC (regs->start, regs->num_regs, regoff_t); |
| 4143 |
RETALLOC (regs->end, regs->num_regs, regoff_t); |
| 4144 |
if (regs->start == NULL || regs->end == NULL) |
| 4145 |
{ |
| 4146 |
FREE_VARIABLES (); |
| 4147 |
return -2; |
| 4148 |
} |
| 4149 |
} |
| 4150 |
} |
| 4151 |
else |
| 4152 |
{ |
| 4153 |
/* These braces fend off a "empty body in an else-statement" |
| 4154 |
warning under GCC when assert expands to nothing. */ |
| 4155 |
assert (bufp->regs_allocated == REGS_FIXED); |
| 4156 |
} |
| 4157 |
|
| 4158 |
/* Convert the pointer data in `regstart' and `regend' to |
| 4159 |
indices. Register zero has to be set differently, |
| 4160 |
since we haven't kept track of any info for it. */ |
| 4161 |
if (regs->num_regs > 0) |
| 4162 |
{ |
| 4163 |
regs->start[0] = pos; |
| 4164 |
regs->end[0] = (MATCHING_IN_FIRST_STRING |
| 4165 |
? ((regoff_t) (d - string1)) |
| 4166 |
: ((regoff_t) (d - string2 + size1))); |
| 4167 |
} |
| 4168 |
|
| 4169 |
/* Go through the first `min (num_regs, regs->num_regs)' |
| 4170 |
registers, since that is all we initialized. */ |
| 4171 |
for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs); |
| 4172 |
mcnt++) |
| 4173 |
{ |
| 4174 |
if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) |
| 4175 |
regs->start[mcnt] = regs->end[mcnt] = -1; |
| 4176 |
else |
| 4177 |
{ |
| 4178 |
regs->start[mcnt] |
| 4179 |
= (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); |
| 4180 |
regs->end[mcnt] |
| 4181 |
= (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); |
| 4182 |
} |
| 4183 |
} |
| 4184 |
|
| 4185 |
/* If the regs structure we return has more elements than |
| 4186 |
were in the pattern, set the extra elements to -1. If |
| 4187 |
we (re)allocated the registers, this is the case, |
| 4188 |
because we always allocate enough to have at least one |
| 4189 |
-1 at the end. */ |
| 4190 |
for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++) |
| 4191 |
regs->start[mcnt] = regs->end[mcnt] = -1; |
| 4192 |
} /* regs && !bufp->no_sub */ |
| 4193 |
|
| 4194 |
DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", |
| 4195 |
nfailure_points_pushed, nfailure_points_popped, |
| 4196 |
nfailure_points_pushed - nfailure_points_popped); |
| 4197 |
DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); |
| 4198 |
|
| 4199 |
mcnt = d - pos - (MATCHING_IN_FIRST_STRING |
| 4200 |
? string1 |
| 4201 |
: string2 - size1); |
| 4202 |
|
| 4203 |
DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); |
| 4204 |
|
| 4205 |
FREE_VARIABLES (); |
| 4206 |
return mcnt; |
| 4207 |
} |
| 4208 |
|
| 4209 |
/* Otherwise match next pattern command. */ |
| 4210 |
switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
| 4211 |
{ |
| 4212 |
/* Ignore these. Used to ignore the n of succeed_n's which |
| 4213 |
currently have n == 0. */ |
| 4214 |
case no_op: |
| 4215 |
DEBUG_PRINT1 ("EXECUTING no_op.\n"); |
| 4216 |
break; |
| 4217 |
|
| 4218 |
case succeed: |
| 4219 |
DEBUG_PRINT1 ("EXECUTING succeed.\n"); |
| 4220 |
goto succeed_label; |
| 4221 |
|
| 4222 |
/* Match the next n pattern characters exactly. The following |
| 4223 |
byte in the pattern defines n, and the n bytes after that |
| 4224 |
are the characters to match. */ |
| 4225 |
case exactn: |
| 4226 |
mcnt = *p++; |
| 4227 |
DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); |
| 4228 |
|
| 4229 |
/* This is written out as an if-else so we don't waste time |
| 4230 |
testing `translate' inside the loop. */ |
| 4231 |
if (translate) |
| 4232 |
{ |
| 4233 |
do |
| 4234 |
{ |
| 4235 |
PREFETCH (); |
| 4236 |
if ((unsigned char) translate[(unsigned char) *d++] |
| 4237 |
!= (unsigned char) *p++) |
| 4238 |
goto fail; |
| 4239 |
} |
| 4240 |
while (--mcnt); |
| 4241 |
} |
| 4242 |
else |
| 4243 |
{ |
| 4244 |
do |
| 4245 |
{ |
| 4246 |
PREFETCH (); |
| 4247 |
if (*d++ != (char) *p++) goto fail; |
| 4248 |
} |
| 4249 |
while (--mcnt); |
| 4250 |
} |
| 4251 |
SET_REGS_MATCHED (); |
| 4252 |
break; |
| 4253 |
|
| 4254 |
|
| 4255 |
/* Match any character except possibly a newline or a null. */ |
| 4256 |
case anychar: |
| 4257 |
DEBUG_PRINT1 ("EXECUTING anychar.\n"); |
| 4258 |
|
| 4259 |
PREFETCH (); |
| 4260 |
|
| 4261 |
if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') |
| 4262 |
|| (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) |
| 4263 |
goto fail; |
| 4264 |
|
| 4265 |
SET_REGS_MATCHED (); |
| 4266 |
DEBUG_PRINT2 (" Matched `%d'.\n", *d); |
| 4267 |
d++; |
| 4268 |
break; |
| 4269 |
|
| 4270 |
|
| 4271 |
case charset: |
| 4272 |
case charset_not: |
| 4273 |
{ |
| 4274 |
register unsigned char c; |
| 4275 |
boolean not = (re_opcode_t) *(p - 1) == charset_not; |
| 4276 |
|
| 4277 |
DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : ""); |
| 4278 |
|
| 4279 |
PREFETCH (); |
| 4280 |
c = TRANSLATE (*d); /* The character to match. */ |
| 4281 |
|
| 4282 |
/* Cast to `unsigned' instead of `unsigned char' in case the |
| 4283 |
bit list is a full 32 bytes long. */ |
| 4284 |
if (c < (unsigned) (*p * BYTEWIDTH) |
| 4285 |
&& p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
| 4286 |
not = !not; |
| 4287 |
|
| 4288 |
p += 1 + *p; |
| 4289 |
|
| 4290 |
if (!not) goto fail; |
| 4291 |
|
| 4292 |
SET_REGS_MATCHED (); |
| 4293 |
d++; |
| 4294 |
break; |
| 4295 |
} |
| 4296 |
|
| 4297 |
|
| 4298 |
/* The beginning of a group is represented by start_memory. |
| 4299 |
The arguments are the register number in the next byte, and the |
| 4300 |
number of groups inner to this one in the next. The text |
| 4301 |
matched within the group is recorded (in the internal |
| 4302 |
registers data structure) under the register number. */ |
| 4303 |
case start_memory: |
| 4304 |
DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]); |
| 4305 |
|
| 4306 |
/* Find out if this group can match the empty string. */ |
| 4307 |
p1 = p; /* To send to group_match_null_string_p. */ |
| 4308 |
|
| 4309 |
if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) |
| 4310 |
REG_MATCH_NULL_STRING_P (reg_info[*p]) |
| 4311 |
= group_match_null_string_p (&p1, pend, reg_info); |
| 4312 |
|
| 4313 |
/* Save the position in the string where we were the last time |
| 4314 |
we were at this open-group operator in case the group is |
| 4315 |
operated upon by a repetition operator, e.g., with `(a*)*b' |
| 4316 |
against `ab'; then we want to ignore where we are now in |
| 4317 |
the string in case this attempt to match fails. */ |
| 4318 |
old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) |
| 4319 |
? REG_UNSET (regstart[*p]) ? d : regstart[*p] |
| 4320 |
: regstart[*p]; |
| 4321 |
DEBUG_PRINT2 (" old_regstart: %d\n", |
| 4322 |
POINTER_TO_OFFSET (old_regstart[*p])); |
| 4323 |
|
| 4324 |
regstart[*p] = d; |
| 4325 |
DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); |
| 4326 |
|
| 4327 |
IS_ACTIVE (reg_info[*p]) = 1; |
| 4328 |
MATCHED_SOMETHING (reg_info[*p]) = 0; |
| 4329 |
|
| 4330 |
/* Clear this whenever we change the register activity status. */ |
| 4331 |
set_regs_matched_done = 0; |
| 4332 |
|
| 4333 |
/* This is the new highest active register. */ |
| 4334 |
highest_active_reg = *p; |
| 4335 |
|
| 4336 |
/* If nothing was active before, this is the new lowest active |
| 4337 |
register. */ |
| 4338 |
if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) |
| 4339 |
lowest_active_reg = *p; |
| 4340 |
|
| 4341 |
/* Move past the register number and inner group count. */ |
| 4342 |
p += 2; |
| 4343 |
just_past_start_mem = p; |
| 4344 |
|
| 4345 |
break; |
| 4346 |
|
| 4347 |
|
| 4348 |
/* The stop_memory opcode represents the end of a group. Its |
| 4349 |
arguments are the same as start_memory's: the register |
| 4350 |
number, and the number of inner groups. */ |
| 4351 |
case stop_memory: |
| 4352 |
DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]); |
| 4353 |
|
| 4354 |
/* We need to save the string position the last time we were at |
| 4355 |
this close-group operator in case the group is operated |
| 4356 |
upon by a repetition operator, e.g., with `((a*)*(b*)*)*' |
| 4357 |
against `aba'; then we want to ignore where we are now in |
| 4358 |
the string in case this attempt to match fails. */ |
| 4359 |
old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) |
| 4360 |
? REG_UNSET (regend[*p]) ? d : regend[*p] |
| 4361 |
: regend[*p]; |
| 4362 |
DEBUG_PRINT2 (" old_regend: %d\n", |
| 4363 |
POINTER_TO_OFFSET (old_regend[*p])); |
| 4364 |
|
| 4365 |
regend[*p] = d; |
| 4366 |
DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); |
| 4367 |
|
| 4368 |
/* This register isn't active anymore. */ |
| 4369 |
IS_ACTIVE (reg_info[*p]) = 0; |
| 4370 |
|
| 4371 |
/* Clear this whenever we change the register activity status. */ |
| 4372 |
set_regs_matched_done = 0; |
| 4373 |
|
| 4374 |
/* If this was the only register active, nothing is active |
| 4375 |
anymore. */ |
| 4376 |
if (lowest_active_reg == highest_active_reg) |
| 4377 |
{ |
| 4378 |
lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
| 4379 |
highest_active_reg = NO_HIGHEST_ACTIVE_REG; |
| 4380 |
} |
| 4381 |
else |
| 4382 |
{ /* We must scan for the new highest active register, since |
| 4383 |
it isn't necessarily one less than now: consider |
| 4384 |
(a(b)c(d(e)f)g). When group 3 ends, after the f), the |
| 4385 |
new highest active register is 1. */ |
| 4386 |
unsigned char r = *p - 1; |
| 4387 |
while (r > 0 && !IS_ACTIVE (reg_info[r])) |
| 4388 |
r--; |
| 4389 |
|
| 4390 |
/* If we end up at register zero, that means that we saved |
| 4391 |
the registers as the result of an `on_failure_jump', not |
| 4392 |
a `start_memory', and we jumped to past the innermost |
| 4393 |
`stop_memory'. For example, in ((.)*) we save |
| 4394 |
registers 1 and 2 as a result of the *, but when we pop |
| 4395 |
back to the second ), we are at the stop_memory 1. |
| 4396 |
Thus, nothing is active. */ |
| 4397 |
if (r == 0) |
| 4398 |
{ |
| 4399 |
lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
| 4400 |
highest_active_reg = NO_HIGHEST_ACTIVE_REG; |
| 4401 |
} |
| 4402 |
else |
| 4403 |
highest_active_reg = r; |
| 4404 |
} |
| 4405 |
|
| 4406 |
/* If just failed to match something this time around with a |
| 4407 |
group that's operated on by a repetition operator, try to |
| 4408 |
force exit from the ``loop'', and restore the register |
| 4409 |
information for this group that we had before trying this |
| 4410 |
last match. */ |
| 4411 |
if ((!MATCHED_SOMETHING (reg_info[*p]) |
| 4412 |
|| just_past_start_mem == p - 1) |
| 4413 |
&& (p + 2) < pend) |
| 4414 |
{ |
| 4415 |
boolean is_a_jump_n = false; |
| 4416 |
|
| 4417 |
p1 = p + 2; |
| 4418 |
mcnt = 0; |
| 4419 |
switch ((re_opcode_t) *p1++) |
| 4420 |
{ |
| 4421 |
case jump_n: |
| 4422 |
is_a_jump_n = true; |
| 4423 |
case pop_failure_jump: |
| 4424 |
case maybe_pop_jump: |
| 4425 |
case jump: |
| 4426 |
case dummy_failure_jump: |
| 4427 |
EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
| 4428 |
if (is_a_jump_n) |
| 4429 |
p1 += 2; |
| 4430 |
break; |
| 4431 |
|
| 4432 |
default: |
| 4433 |
/* do nothing */ ; |
| 4434 |
} |
| 4435 |
p1 += mcnt; |
| 4436 |
|
| 4437 |
/* If the next operation is a jump backwards in the pattern |
| 4438 |
to an on_failure_jump right before the start_memory |
| 4439 |
corresponding to this stop_memory, exit from the loop |
| 4440 |
by forcing a failure after pushing on the stack the |
| 4441 |
on_failure_jump's jump in the pattern, and d. */ |
| 4442 |
if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump |
| 4443 |
&& (re_opcode_t) p1[3] == start_memory && p1[4] == *p) |
| 4444 |
{ |
| 4445 |
/* If this group ever matched anything, then restore |
| 4446 |
what its registers were before trying this last |
| 4447 |
failed match, e.g., with `(a*)*b' against `ab' for |
| 4448 |
regstart[1], and, e.g., with `((a*)*(b*)*)*' |
| 4449 |
against `aba' for regend[3]. |
| 4450 |
|
| 4451 |
Also restore the registers for inner groups for, |
| 4452 |
e.g., `((a*)(b*))*' against `aba' (register 3 would |
| 4453 |
otherwise get trashed). */ |
| 4454 |
|
| 4455 |
if (EVER_MATCHED_SOMETHING (reg_info[*p])) |
| 4456 |
{ |
| 4457 |
unsigned r; |
| 4458 |
|
| 4459 |
EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; |
| 4460 |
|
| 4461 |
/* Restore this and inner groups' (if any) registers. */ |
| 4462 |
for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1); |
| 4463 |
r++) |
| 4464 |
{ |
| 4465 |
regstart[r] = old_regstart[r]; |
| 4466 |
|
| 4467 |
/* xx why this test? */ |
| 4468 |
if (old_regend[r] >= regstart[r]) |
| 4469 |
regend[r] = old_regend[r]; |
| 4470 |
} |
| 4471 |
} |
| 4472 |
p1++; |
| 4473 |
EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
| 4474 |
PUSH_FAILURE_POINT (p1 + mcnt, d, -2); |
| 4475 |
|
| 4476 |
goto fail; |
| 4477 |
} |
| 4478 |
} |
| 4479 |
|
| 4480 |
/* Move past the register number and the inner group count. */ |
| 4481 |
p += 2; |
| 4482 |
break; |
| 4483 |
|
| 4484 |
|
| 4485 |
/* \<digit> has been turned into a `duplicate' command which is |
| 4486 |
followed by the numeric value of <digit> as the register number. */ |
| 4487 |
case duplicate: |
| 4488 |
{ |
| 4489 |
register const char *d2, *dend2; |
| 4490 |
int regno = *p++; /* Get which register to match against. */ |
| 4491 |
DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); |
| 4492 |
|
| 4493 |
/* Can't back reference a group which we've never matched. */ |
| 4494 |
if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) |
| 4495 |
goto fail; |
| 4496 |
|
| 4497 |
/* Where in input to try to start matching. */ |
| 4498 |
d2 = regstart[regno]; |
| 4499 |
|
| 4500 |
/* Where to stop matching; if both the place to start and |
| 4501 |
the place to stop matching are in the same string, then |
| 4502 |
set to the place to stop, otherwise, for now have to use |
| 4503 |
the end of the first string. */ |
| 4504 |
|
| 4505 |
dend2 = ((FIRST_STRING_P (regstart[regno]) |
| 4506 |
== FIRST_STRING_P (regend[regno])) |
| 4507 |
? regend[regno] : end_match_1); |
| 4508 |
for (;;) |
| 4509 |
{ |
| 4510 |
/* If necessary, advance to next segment in register |
| 4511 |
contents. */ |
| 4512 |
while (d2 == dend2) |
| 4513 |
{ |
| 4514 |
if (dend2 == end_match_2) break; |
| 4515 |
if (dend2 == regend[regno]) break; |
| 4516 |
|
| 4517 |
/* End of string1 => advance to string2. */ |
| 4518 |
d2 = string2; |
| 4519 |
dend2 = regend[regno]; |
| 4520 |
} |
| 4521 |
/* At end of register contents => success */ |
| 4522 |
if (d2 == dend2) break; |
| 4523 |
|
| 4524 |
/* If necessary, advance to next segment in data. */ |
| 4525 |
PREFETCH (); |
| 4526 |
|
| 4527 |
/* How many characters left in this segment to match. */ |
| 4528 |
mcnt = dend - d; |
| 4529 |
|
| 4530 |
/* Want how many consecutive characters we can match in |
| 4531 |
one shot, so, if necessary, adjust the count. */ |
| 4532 |
if (mcnt > dend2 - d2) |
| 4533 |
mcnt = dend2 - d2; |
| 4534 |
|
| 4535 |
/* Compare that many; failure if mismatch, else move |
| 4536 |
past them. */ |
| 4537 |
if (translate |
| 4538 |
? bcmp_translate (d, d2, mcnt, translate) |
| 4539 |
: memcmp (d, d2, mcnt)) |
| 4540 |
goto fail; |
| 4541 |
d += mcnt, d2 += mcnt; |
| 4542 |
|
| 4543 |
/* Do this because we've match some characters. */ |
| 4544 |
SET_REGS_MATCHED (); |
| 4545 |
} |
| 4546 |
} |
| 4547 |
break; |
| 4548 |
|
| 4549 |
|
| 4550 |
/* begline matches the empty string at the beginning of the string |
| 4551 |
(unless `not_bol' is set in `bufp'), and, if |
| 4552 |
`newline_anchor' is set, after newlines. */ |
| 4553 |
case begline: |
| 4554 |
DEBUG_PRINT1 ("EXECUTING begline.\n"); |
| 4555 |
|
| 4556 |
if (AT_STRINGS_BEG (d)) |
| 4557 |
{ |
| 4558 |
if (!bufp->not_bol) break; |
| 4559 |
} |
| 4560 |
else if (d[-1] == '\n' && bufp->newline_anchor) |
| 4561 |
{ |
| 4562 |
break; |
| 4563 |
} |
| 4564 |
/* In all other cases, we fail. */ |
| 4565 |
goto fail; |
| 4566 |
|
| 4567 |
|
| 4568 |
/* endline is the dual of begline. */ |
| 4569 |
case endline: |
| 4570 |
DEBUG_PRINT1 ("EXECUTING endline.\n"); |
| 4571 |
|
| 4572 |
if (AT_STRINGS_END (d)) |
| 4573 |
{ |
| 4574 |
if (!bufp->not_eol) break; |
| 4575 |
} |
| 4576 |
|
| 4577 |
/* We have to ``prefetch'' the next character. */ |
| 4578 |
else if ((d == end1 ? *string2 : *d) == '\n' |
| 4579 |
&& bufp->newline_anchor) |
| 4580 |
{ |
| 4581 |
break; |
| 4582 |
} |
| 4583 |
goto fail; |
| 4584 |
|
| 4585 |
|
| 4586 |
/* Match at the very beginning of the data. */ |
| 4587 |
case begbuf: |
| 4588 |
DEBUG_PRINT1 ("EXECUTING begbuf.\n"); |
| 4589 |
if (AT_STRINGS_BEG (d)) |
| 4590 |
break; |
| 4591 |
goto fail; |
| 4592 |
|
| 4593 |
|
| 4594 |
/* Match at the very end of the data. */ |
| 4595 |
case endbuf: |
| 4596 |
DEBUG_PRINT1 ("EXECUTING endbuf.\n"); |
| 4597 |
if (AT_STRINGS_END (d)) |
| 4598 |
break; |
| 4599 |
goto fail; |
| 4600 |
|
| 4601 |
|
| 4602 |
/* on_failure_keep_string_jump is used to optimize `.*\n'. It |
| 4603 |
pushes NULL as the value for the string on the stack. Then |
| 4604 |
`pop_failure_point' will keep the current value for the |
| 4605 |
string, instead of restoring it. To see why, consider |
| 4606 |
matching `foo\nbar' against `.*\n'. The .* matches the foo; |
| 4607 |
then the . fails against the \n. But the next thing we want |
| 4608 |
to do is match the \n against the \n; if we restored the |
| 4609 |
string value, we would be back at the foo. |
| 4610 |
|
| 4611 |
Because this is used only in specific cases, we don't need to |
| 4612 |
check all the things that `on_failure_jump' does, to make |
| 4613 |
sure the right things get saved on the stack. Hence we don't |
| 4614 |
share its code. The only reason to push anything on the |
| 4615 |
stack at all is that otherwise we would have to change |
| 4616 |
`anychar's code to do something besides goto fail in this |
| 4617 |
case; that seems worse than this. */ |
| 4618 |
case on_failure_keep_string_jump: |
| 4619 |
DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); |
| 4620 |
|
| 4621 |
EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 4622 |
#ifdef _LIBC |
| 4623 |
DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt); |
| 4624 |
#else |
| 4625 |
DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); |
| 4626 |
#endif |
| 4627 |
|
| 4628 |
PUSH_FAILURE_POINT (p + mcnt, NULL, -2); |
| 4629 |
break; |
| 4630 |
|
| 4631 |
|
| 4632 |
/* Uses of on_failure_jump: |
| 4633 |
|
| 4634 |
Each alternative starts with an on_failure_jump that points |
| 4635 |
to the beginning of the next alternative. Each alternative |
| 4636 |
except the last ends with a jump that in effect jumps past |
| 4637 |
the rest of the alternatives. (They really jump to the |
| 4638 |
ending jump of the following alternative, because tensioning |
| 4639 |
these jumps is a hassle.) |
| 4640 |
|
| 4641 |
Repeats start with an on_failure_jump that points past both |
| 4642 |
the repetition text and either the following jump or |
| 4643 |
pop_failure_jump back to this on_failure_jump. */ |
| 4644 |
case on_failure_jump: |
| 4645 |
on_failure: |
| 4646 |
DEBUG_PRINT1 ("EXECUTING on_failure_jump"); |
| 4647 |
|
| 4648 |
EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 4649 |
#ifdef _LIBC |
| 4650 |
DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt); |
| 4651 |
#else |
| 4652 |
DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); |
| 4653 |
#endif |
| 4654 |
|
| 4655 |
/* If this on_failure_jump comes right before a group (i.e., |
| 4656 |
the original * applied to a group), save the information |
| 4657 |
for that group and all inner ones, so that if we fail back |
| 4658 |
to this point, the group's information will be correct. |
| 4659 |
For example, in \(a*\)*\1, we need the preceding group, |
| 4660 |
and in \(zz\(a*\)b*\)\2, we need the inner group. */ |
| 4661 |
|
| 4662 |
/* We can't use `p' to check ahead because we push |
| 4663 |
a failure point to `p + mcnt' after we do this. */ |
| 4664 |
p1 = p; |
| 4665 |
|
| 4666 |
/* We need to skip no_op's before we look for the |
| 4667 |
start_memory in case this on_failure_jump is happening as |
| 4668 |
the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 |
| 4669 |
against aba. */ |
| 4670 |
while (p1 < pend && (re_opcode_t) *p1 == no_op) |
| 4671 |
p1++; |
| 4672 |
|
| 4673 |
if (p1 < pend && (re_opcode_t) *p1 == start_memory) |
| 4674 |
{ |
| 4675 |
/* We have a new highest active register now. This will |
| 4676 |
get reset at the start_memory we are about to get to, |
| 4677 |
but we will have saved all the registers relevant to |
| 4678 |
this repetition op, as described above. */ |
| 4679 |
highest_active_reg = *(p1 + 1) + *(p1 + 2); |
| 4680 |
if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) |
| 4681 |
lowest_active_reg = *(p1 + 1); |
| 4682 |
} |
| 4683 |
|
| 4684 |
DEBUG_PRINT1 (":\n"); |
| 4685 |
PUSH_FAILURE_POINT (p + mcnt, d, -2); |
| 4686 |
break; |
| 4687 |
|
| 4688 |
|
| 4689 |
/* A smart repeat ends with `maybe_pop_jump'. |
| 4690 |
We change it to either `pop_failure_jump' or `jump'. */ |
| 4691 |
case maybe_pop_jump: |
| 4692 |
EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 4693 |
DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); |
| 4694 |
{ |
| 4695 |
register unsigned char *p2 = p; |
| 4696 |
|
| 4697 |
/* Compare the beginning of the repeat with what in the |
| 4698 |
pattern follows its end. If we can establish that there |
| 4699 |
is nothing that they would both match, i.e., that we |
| 4700 |
would have to backtrack because of (as in, e.g., `a*a') |
| 4701 |
then we can change to pop_failure_jump, because we'll |
| 4702 |
never have to backtrack. |
| 4703 |
|
| 4704 |
This is not true in the case of alternatives: in |
| 4705 |
`(a|ab)*' we do need to backtrack to the `ab' alternative |
| 4706 |
(e.g., if the string was `ab'). But instead of trying to |
| 4707 |
detect that here, the alternative has put on a dummy |
| 4708 |
failure point which is what we will end up popping. */ |
| 4709 |
|
| 4710 |
/* Skip over open/close-group commands. |
| 4711 |
If what follows this loop is a ...+ construct, |
| 4712 |
look at what begins its body, since we will have to |
| 4713 |
match at least one of that. */ |
| 4714 |
while (1) |
| 4715 |
{ |
| 4716 |
if (p2 + 2 < pend |
| 4717 |
&& ((re_opcode_t) *p2 == stop_memory |
| 4718 |
|| (re_opcode_t) *p2 == start_memory)) |
| 4719 |
p2 += 3; |
| 4720 |
else if (p2 + 6 < pend |
| 4721 |
&& (re_opcode_t) *p2 == dummy_failure_jump) |
| 4722 |
p2 += 6; |
| 4723 |
else |
| 4724 |
break; |
| 4725 |
} |
| 4726 |
|
| 4727 |
p1 = p + mcnt; |
| 4728 |
/* p1[0] ... p1[2] are the `on_failure_jump' corresponding |
| 4729 |
to the `maybe_finalize_jump' of this case. Examine what |
| 4730 |
follows. */ |
| 4731 |
|
| 4732 |
/* If we're at the end of the pattern, we can change. */ |
| 4733 |
if (p2 == pend) |
| 4734 |
{ |
| 4735 |
/* Consider what happens when matching ":\(.*\)" |
| 4736 |
against ":/". I don't really understand this code |
| 4737 |
yet. */ |
| 4738 |
p[-3] = (unsigned char) pop_failure_jump; |
| 4739 |
DEBUG_PRINT1 |
| 4740 |
(" End of pattern: change to `pop_failure_jump'.\n"); |
| 4741 |
} |
| 4742 |
|
| 4743 |
else if ((re_opcode_t) *p2 == exactn |
| 4744 |
|| (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) |
| 4745 |
{ |
| 4746 |
register unsigned char c |
| 4747 |
= *p2 == (unsigned char) endline ? '\n' : p2[2]; |
| 4748 |
|
| 4749 |
if ((re_opcode_t) p1[3] == exactn && p1[5] != c) |
| 4750 |
{ |
| 4751 |
p[-3] = (unsigned char) pop_failure_jump; |
| 4752 |
DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", |
| 4753 |
c, p1[5]); |
| 4754 |
} |
| 4755 |
|
| 4756 |
else if ((re_opcode_t) p1[3] == charset |
| 4757 |
|| (re_opcode_t) p1[3] == charset_not) |
| 4758 |
{ |
| 4759 |
int not = (re_opcode_t) p1[3] == charset_not; |
| 4760 |
|
| 4761 |
if (c < (unsigned char) (p1[4] * BYTEWIDTH) |
| 4762 |
&& p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
| 4763 |
not = !not; |
| 4764 |
|
| 4765 |
/* `not' is equal to 1 if c would match, which means |
| 4766 |
that we can't change to pop_failure_jump. */ |
| 4767 |
if (!not) |
| 4768 |
{ |
| 4769 |
p[-3] = (unsigned char) pop_failure_jump; |
| 4770 |
DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
| 4771 |
} |
| 4772 |
} |
| 4773 |
} |
| 4774 |
else if ((re_opcode_t) *p2 == charset) |
| 4775 |
{ |
| 4776 |
#ifdef DEBUG |
| 4777 |
register unsigned char c |
| 4778 |
= *p2 == (unsigned char) endline ? '\n' : p2[2]; |
| 4779 |
#endif |
| 4780 |
|
| 4781 |
#if 0 |
| 4782 |
if ((re_opcode_t) p1[3] == exactn |
| 4783 |
&& ! ((int) p2[1] * BYTEWIDTH > (int) p1[5] |
| 4784 |
&& (p2[2 + p1[5] / BYTEWIDTH] |
| 4785 |
& (1 << (p1[5] % BYTEWIDTH))))) |
| 4786 |
#else |
| 4787 |
if ((re_opcode_t) p1[3] == exactn |
| 4788 |
&& ! ((int) p2[1] * BYTEWIDTH > (int) p1[4] |
| 4789 |
&& (p2[2 + p1[4] / BYTEWIDTH] |
| 4790 |
& (1 << (p1[4] % BYTEWIDTH))))) |
| 4791 |
#endif |
| 4792 |
{ |
| 4793 |
p[-3] = (unsigned char) pop_failure_jump; |
| 4794 |
DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", |
| 4795 |
c, p1[5]); |
| 4796 |
} |
| 4797 |
|
| 4798 |
else if ((re_opcode_t) p1[3] == charset_not) |
| 4799 |
{ |
| 4800 |
int idx; |
| 4801 |
/* We win if the charset_not inside the loop |
| 4802 |
lists every character listed in the charset after. */ |
| 4803 |
for (idx = 0; idx < (int) p2[1]; idx++) |
| 4804 |
if (! (p2[2 + idx] == 0 |
| 4805 |
|| (idx < (int) p1[4] |
| 4806 |
&& ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) |
| 4807 |
break; |
| 4808 |
|
| 4809 |
if (idx == p2[1]) |
| 4810 |
{ |
| 4811 |
p[-3] = (unsigned char) pop_failure_jump; |
| 4812 |
DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
| 4813 |
} |
| 4814 |
} |
| 4815 |
else if ((re_opcode_t) p1[3] == charset) |
| 4816 |
{ |
| 4817 |
int idx; |
| 4818 |
/* We win if the charset inside the loop |
| 4819 |
has no overlap with the one after the loop. */ |
| 4820 |
for (idx = 0; |
| 4821 |
idx < (int) p2[1] && idx < (int) p1[4]; |
| 4822 |
idx++) |
| 4823 |
if ((p2[2 + idx] & p1[5 + idx]) != 0) |
| 4824 |
break; |
| 4825 |
|
| 4826 |
if (idx == p2[1] || idx == p1[4]) |
| 4827 |
{ |
| 4828 |
p[-3] = (unsigned char) pop_failure_jump; |
| 4829 |
DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); |
| 4830 |
} |
| 4831 |
} |
| 4832 |
} |
| 4833 |
} |
| 4834 |
p -= 2; /* Point at relative address again. */ |
| 4835 |
if ((re_opcode_t) p[-1] != pop_failure_jump) |
| 4836 |
{ |
| 4837 |
p[-1] = (unsigned char) jump; |
| 4838 |
DEBUG_PRINT1 (" Match => jump.\n"); |
| 4839 |
goto unconditional_jump; |
| 4840 |
} |
| 4841 |
/* Note fall through. */ |
| 4842 |
|
| 4843 |
|
| 4844 |
/* The end of a simple repeat has a pop_failure_jump back to |
| 4845 |
its matching on_failure_jump, where the latter will push a |
| 4846 |
failure point. The pop_failure_jump takes off failure |
| 4847 |
points put on by this pop_failure_jump's matching |
| 4848 |
on_failure_jump; we got through the pattern to here from the |
| 4849 |
matching on_failure_jump, so didn't fail. */ |
| 4850 |
case pop_failure_jump: |
| 4851 |
{ |
| 4852 |
/* We need to pass separate storage for the lowest and |
| 4853 |
highest registers, even though we don't care about the |
| 4854 |
actual values. Otherwise, we will restore only one |
| 4855 |
register from the stack, since lowest will == highest in |
| 4856 |
`pop_failure_point'. */ |
| 4857 |
active_reg_t dummy_low_reg, dummy_high_reg; |
| 4858 |
unsigned char *pdummy; |
| 4859 |
const char *sdummy; |
| 4860 |
|
| 4861 |
DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); |
| 4862 |
POP_FAILURE_POINT (sdummy, pdummy, |
| 4863 |
dummy_low_reg, dummy_high_reg, |
| 4864 |
reg_dummy, reg_dummy, reg_info_dummy); |
| 4865 |
} |
| 4866 |
/* Note fall through. */ |
| 4867 |
|
| 4868 |
unconditional_jump: |
| 4869 |
#ifdef _LIBC |
| 4870 |
DEBUG_PRINT2 ("\n%p: ", p); |
| 4871 |
#else |
| 4872 |
DEBUG_PRINT2 ("\n0x%x: ", p); |
| 4873 |
#endif |
| 4874 |
/* Note fall through. */ |
| 4875 |
|
| 4876 |
/* Unconditionally jump (without popping any failure points). */ |
| 4877 |
case jump: |
| 4878 |
EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
| 4879 |
DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); |
| 4880 |
p += mcnt; /* Do the jump. */ |
| 4881 |
#ifdef _LIBC |
| 4882 |
DEBUG_PRINT2 ("(to %p).\n", p); |
| 4883 |
#else |
| 4884 |
DEBUG_PRINT2 ("(to 0x%x).\n", p); |
| 4885 |
#endif |
| 4886 |
break; |
| 4887 |
|
| 4888 |
|
| 4889 |
/* We need this opcode so we can detect where alternatives end |
| 4890 |
in `group_match_null_string_p' et al. */ |
| 4891 |
case jump_past_alt: |
| 4892 |
DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); |
| 4893 |
goto unconditional_jump; |
| 4894 |
|
| 4895 |
|
| 4896 |
/* Normally, the on_failure_jump pushes a failure point, which |
| 4897 |
then gets popped at pop_failure_jump. We will end up at |
| 4898 |
pop_failure_jump, also, and with a pattern of, say, `a+', we |
| 4899 |
are skipping over the on_failure_jump, so we have to push |
| 4900 |
something meaningless for pop_failure_jump to pop. */ |
| 4901 |
case dummy_failure_jump: |
| 4902 |
DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); |
| 4903 |
/* It doesn't matter what we push for the string here. What |
| 4904 |
the code at `fail' tests is the value for the pattern. */ |
| 4905 |
PUSH_FAILURE_POINT (NULL, NULL, -2); |
| 4906 |
goto unconditional_jump; |
| 4907 |
|
| 4908 |
|
| 4909 |
/* At the end of an alternative, we need to push a dummy failure |
| 4910 |
point in case we are followed by a `pop_failure_jump', because |
| 4911 |
we don't want the failure point for the alternative to be |
| 4912 |
popped. For example, matching `(a|ab)*' against `aab' |
| 4913 |
requires that we match the `ab' alternative. */ |
| 4914 |
case push_dummy_failure: |
| 4915 |
DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); |
| 4916 |
/* See comments just above at `dummy_failure_jump' about the |
| 4917 |
two zeroes. */ |
| 4918 |
PUSH_FAILURE_POINT (NULL, NULL, -2); |
| 4919 |
break; |
| 4920 |
|
| 4921 |
/* Have to succeed matching what follows at least n times. |
| 4922 |
After that, handle like `on_failure_jump'. */ |
| 4923 |
case succeed_n: |
| 4924 |
EXTRACT_NUMBER (mcnt, p + 2); |
| 4925 |
DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); |
| 4926 |
|
| 4927 |
assert (mcnt >= 0); |
| 4928 |
/* Originally, this is how many times we HAVE to succeed. */ |
| 4929 |
if (mcnt > 0) |
| 4930 |
{ |
| 4931 |
mcnt--; |
| 4932 |
p += 2; |
| 4933 |
STORE_NUMBER_AND_INCR (p, mcnt); |
| 4934 |
#ifdef _LIBC |
| 4935 |
DEBUG_PRINT3 (" Setting %p to %d.\n", p - 2, mcnt); |
| 4936 |
#else |
| 4937 |
DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - 2, mcnt); |
| 4938 |
#endif |
| 4939 |
} |
| 4940 |
else if (mcnt == 0) |
| 4941 |
{ |
| 4942 |
#ifdef _LIBC |
| 4943 |
DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", p+2); |
| 4944 |
#else |
| 4945 |
DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2); |
| 4946 |
#endif |
| 4947 |
p[2] = (unsigned char) no_op; |
| 4948 |
p[3] = (unsigned char) no_op; |
| 4949 |
goto on_failure; |
| 4950 |
} |
| 4951 |
break; |
| 4952 |
|
| 4953 |
case jump_n: |
| 4954 |
EXTRACT_NUMBER (mcnt, p + 2); |
| 4955 |
DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); |
| 4956 |
|
| 4957 |
/* Originally, this is how many times we CAN jump. */ |
| 4958 |
if (mcnt) |
| 4959 |
{ |
| 4960 |
mcnt--; |
| 4961 |
STORE_NUMBER (p + 2, mcnt); |
| 4962 |
#ifdef _LIBC |
| 4963 |
DEBUG_PRINT3 (" Setting %p to %d.\n", p + 2, mcnt); |
| 4964 |
#else |
| 4965 |
DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + 2, mcnt); |
| 4966 |
#endif |
| 4967 |
goto unconditional_jump; |
| 4968 |
} |
| 4969 |
/* If don't have to jump any more, skip over the rest of command. */ |
| 4970 |
else |
| 4971 |
p += 4; |
| 4972 |
break; |
| 4973 |
|
| 4974 |
case set_number_at: |
| 4975 |
{ |
| 4976 |
DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); |
| 4977 |
|
| 4978 |
EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 4979 |
p1 = p + mcnt; |
| 4980 |
EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 4981 |
#ifdef _LIBC |
| 4982 |
DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt); |
| 4983 |
#else |
| 4984 |
DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); |
| 4985 |
#endif |
| 4986 |
STORE_NUMBER (p1, mcnt); |
| 4987 |
break; |
| 4988 |
} |
| 4989 |
|
| 4990 |
#if 0 |
| 4991 |
/* The DEC Alpha C compiler 3.x generates incorrect code for the |
| 4992 |
test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of |
| 4993 |
AT_WORD_BOUNDARY, so this code is disabled. Expanding the |
| 4994 |
macro and introducing temporary variables works around the bug. */ |
| 4995 |
|
| 4996 |
case wordbound: |
| 4997 |
DEBUG_PRINT1 ("EXECUTING wordbound.\n"); |
| 4998 |
if (AT_WORD_BOUNDARY (d)) |
| 4999 |
break; |
| 5000 |
goto fail; |
| 5001 |
|
| 5002 |
case notwordbound: |
| 5003 |
DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); |
| 5004 |
if (AT_WORD_BOUNDARY (d)) |
| 5005 |
goto fail; |
| 5006 |
break; |
| 5007 |
#else |
| 5008 |
case wordbound: |
| 5009 |
{ |
| 5010 |
boolean prevchar, thischar; |
| 5011 |
|
| 5012 |
DEBUG_PRINT1 ("EXECUTING wordbound.\n"); |
| 5013 |
if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
| 5014 |
break; |
| 5015 |
|
| 5016 |
prevchar = WORDCHAR_P (d - 1); |
| 5017 |
thischar = WORDCHAR_P (d); |
| 5018 |
if (prevchar != thischar) |
| 5019 |
break; |
| 5020 |
goto fail; |
| 5021 |
} |
| 5022 |
|
| 5023 |
case notwordbound: |
| 5024 |
{ |
| 5025 |
boolean prevchar, thischar; |
| 5026 |
|
| 5027 |
DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); |
| 5028 |
if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
| 5029 |
goto fail; |
| 5030 |
|
| 5031 |
prevchar = WORDCHAR_P (d - 1); |
| 5032 |
thischar = WORDCHAR_P (d); |
| 5033 |
if (prevchar != thischar) |
| 5034 |
goto fail; |
| 5035 |
break; |
| 5036 |
} |
| 5037 |
#endif |
| 5038 |
|
| 5039 |
case wordbeg: |
| 5040 |
DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); |
| 5041 |
if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) |
| 5042 |
break; |
| 5043 |
goto fail; |
| 5044 |
|
| 5045 |
case wordend: |
| 5046 |
DEBUG_PRINT1 ("EXECUTING wordend.\n"); |
| 5047 |
if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) |
| 5048 |
&& (!WORDCHAR_P (d) || AT_STRINGS_END (d))) |
| 5049 |
break; |
| 5050 |
goto fail; |
| 5051 |
|
| 5052 |
#ifdef emacs |
| 5053 |
case before_dot: |
| 5054 |
DEBUG_PRINT1 ("EXECUTING before_dot.\n"); |
| 5055 |
if (PTR_CHAR_POS ((unsigned char *) d) >= point) |
| 5056 |
goto fail; |
| 5057 |
break; |
| 5058 |
|
| 5059 |
case at_dot: |
| 5060 |
DEBUG_PRINT1 ("EXECUTING at_dot.\n"); |
| 5061 |
if (PTR_CHAR_POS ((unsigned char *) d) != point) |
| 5062 |
goto fail; |
| 5063 |
break; |
| 5064 |
|
| 5065 |
case after_dot: |
| 5066 |
DEBUG_PRINT1 ("EXECUTING after_dot.\n"); |
| 5067 |
if (PTR_CHAR_POS ((unsigned char *) d) <= point) |
| 5068 |
goto fail; |
| 5069 |
break; |
| 5070 |
|
| 5071 |
case syntaxspec: |
| 5072 |
DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); |
| 5073 |
mcnt = *p++; |
| 5074 |
goto matchsyntax; |
| 5075 |
|
| 5076 |
case wordchar: |
| 5077 |
DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); |
| 5078 |
mcnt = (int) Sword; |
| 5079 |
matchsyntax: |
| 5080 |
PREFETCH (); |
| 5081 |
/* Can't use *d++ here; SYNTAX may be an unsafe macro. */ |
| 5082 |
d++; |
| 5083 |
if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt) |
| 5084 |
goto fail; |
| 5085 |
SET_REGS_MATCHED (); |
| 5086 |
break; |
| 5087 |
|
| 5088 |
case notsyntaxspec: |
| 5089 |
DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); |
| 5090 |
mcnt = *p++; |
| 5091 |
goto matchnotsyntax; |
| 5092 |
|
| 5093 |
case notwordchar: |
| 5094 |
DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); |
| 5095 |
mcnt = (int) Sword; |
| 5096 |
matchnotsyntax: |
| 5097 |
PREFETCH (); |
| 5098 |
/* Can't use *d++ here; SYNTAX may be an unsafe macro. */ |
| 5099 |
d++; |
| 5100 |
if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt) |
| 5101 |
goto fail; |
| 5102 |
SET_REGS_MATCHED (); |
| 5103 |
break; |
| 5104 |
|
| 5105 |
#else /* not emacs */ |
| 5106 |
case wordchar: |
| 5107 |
DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); |
| 5108 |
PREFETCH (); |
| 5109 |
if (!WORDCHAR_P (d)) |
| 5110 |
goto fail; |
| 5111 |
SET_REGS_MATCHED (); |
| 5112 |
d++; |
| 5113 |
break; |
| 5114 |
|
| 5115 |
case notwordchar: |
| 5116 |
DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); |
| 5117 |
PREFETCH (); |
| 5118 |
if (WORDCHAR_P (d)) |
| 5119 |
goto fail; |
| 5120 |
SET_REGS_MATCHED (); |
| 5121 |
d++; |
| 5122 |
break; |
| 5123 |
#endif /* not emacs */ |
| 5124 |
|
| 5125 |
default: |
| 5126 |
abort (); |
| 5127 |
} |
| 5128 |
continue; /* Successfully executed one pattern command; keep going. */ |
| 5129 |
|
| 5130 |
|
| 5131 |
/* We goto here if a matching operation fails. */ |
| 5132 |
fail: |
| 5133 |
if (!FAIL_STACK_EMPTY ()) |
| 5134 |
{ /* A restart point is known. Restore to that state. */ |
| 5135 |
DEBUG_PRINT1 ("\nFAIL:\n"); |
| 5136 |
POP_FAILURE_POINT (d, p, |
| 5137 |
lowest_active_reg, highest_active_reg, |
| 5138 |
regstart, regend, reg_info); |
| 5139 |
|
| 5140 |
/* If this failure point is a dummy, try the next one. */ |
| 5141 |
if (!p) |
| 5142 |
goto fail; |
| 5143 |
|
| 5144 |
/* If we failed to the end of the pattern, don't examine *p. */ |
| 5145 |
assert (p <= pend); |
| 5146 |
if (p < pend) |
| 5147 |
{ |
| 5148 |
boolean is_a_jump_n = false; |
| 5149 |
|
| 5150 |
/* If failed to a backwards jump that's part of a repetition |
| 5151 |
loop, need to pop this failure point and use the next one. */ |
| 5152 |
switch ((re_opcode_t) *p) |
| 5153 |
{ |
| 5154 |
case jump_n: |
| 5155 |
is_a_jump_n = true; |
| 5156 |
case maybe_pop_jump: |
| 5157 |
case pop_failure_jump: |
| 5158 |
case jump: |
| 5159 |
p1 = p + 1; |
| 5160 |
EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
| 5161 |
p1 += mcnt; |
| 5162 |
|
| 5163 |
if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) |
| 5164 |
|| (!is_a_jump_n |
| 5165 |
&& (re_opcode_t) *p1 == on_failure_jump)) |
| 5166 |
goto fail; |
| 5167 |
break; |
| 5168 |
default: |
| 5169 |
/* do nothing */ ; |
| 5170 |
} |
| 5171 |
} |
| 5172 |
|
| 5173 |
if (d >= string1 && d <= end1) |
| 5174 |
dend = end_match_1; |
| 5175 |
} |
| 5176 |
else |
| 5177 |
break; /* Matching at this starting point really fails. */ |
| 5178 |
} /* for (;;) */ |
| 5179 |
|
| 5180 |
if (best_regs_set) |
| 5181 |
goto restore_best_regs; |
| 5182 |
|
| 5183 |
FREE_VARIABLES (); |
| 5184 |
|
| 5185 |
return -1; /* Failure to match. */ |
| 5186 |
} /* re_match_2 */ |
| 5187 |
|
| 5188 |
/* Subroutine definitions for re_match_2. */ |
| 5189 |
|
| 5190 |
|
| 5191 |
/* We are passed P pointing to a register number after a start_memory. |
| 5192 |
|
| 5193 |
Return true if the pattern up to the corresponding stop_memory can |
| 5194 |
match the empty string, and false otherwise. |
| 5195 |
|
| 5196 |
If we find the matching stop_memory, sets P to point to one past its number. |
| 5197 |
Otherwise, sets P to an undefined byte less than or equal to END. |
| 5198 |
|
| 5199 |
We don't handle duplicates properly (yet). */ |
| 5200 |
|
| 5201 |
static boolean |
| 5202 |
group_match_null_string_p (p, end, reg_info) |
| 5203 |
unsigned char **p, *end; |
| 5204 |
register_info_type *reg_info; |
| 5205 |
{ |
| 5206 |
int mcnt; |
| 5207 |
/* Point to after the args to the start_memory. */ |
| 5208 |
unsigned char *p1 = *p + 2; |
| 5209 |
|
| 5210 |
while (p1 < end) |
| 5211 |
{ |
| 5212 |
/* Skip over opcodes that can match nothing, and return true or |
| 5213 |
false, as appropriate, when we get to one that can't, or to the |
| 5214 |
matching stop_memory. */ |
| 5215 |
|
| 5216 |
switch ((re_opcode_t) *p1) |
| 5217 |
{ |
| 5218 |
/* Could be either a loop or a series of alternatives. */ |
| 5219 |
case on_failure_jump: |
| 5220 |
p1++; |
| 5221 |
EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
| 5222 |
|
| 5223 |
/* If the next operation is not a jump backwards in the |
| 5224 |
pattern. */ |
| 5225 |
|
| 5226 |
if (mcnt >= 0) |
| 5227 |
{ |
| 5228 |
/* Go through the on_failure_jumps of the alternatives, |
| 5229 |
seeing if any of the alternatives cannot match nothing. |
| 5230 |
The last alternative starts with only a jump, |
| 5231 |
whereas the rest start with on_failure_jump and end |
| 5232 |
with a jump, e.g., here is the pattern for `a|b|c': |
| 5233 |
|
| 5234 |
/on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 |
| 5235 |
/on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 |
| 5236 |
/exactn/1/c |
| 5237 |
|
| 5238 |
So, we have to first go through the first (n-1) |
| 5239 |
alternatives and then deal with the last one separately. */ |
| 5240 |
|
| 5241 |
|
| 5242 |
/* Deal with the first (n-1) alternatives, which start |
| 5243 |
with an on_failure_jump (see above) that jumps to right |
| 5244 |
past a jump_past_alt. */ |
| 5245 |
|
| 5246 |
while ((re_opcode_t) p1[mcnt-3] == jump_past_alt) |
| 5247 |
{ |
| 5248 |
/* `mcnt' holds how many bytes long the alternative |
| 5249 |
is, including the ending `jump_past_alt' and |
| 5250 |
its number. */ |
| 5251 |
|
| 5252 |
if (!alt_match_null_string_p (p1, p1 + mcnt - 3, |
| 5253 |
reg_info)) |
| 5254 |
return false; |
| 5255 |
|
| 5256 |
/* Move to right after this alternative, including the |
| 5257 |
jump_past_alt. */ |
| 5258 |
p1 += mcnt; |
| 5259 |
|
| 5260 |
/* Break if it's the beginning of an n-th alternative |
| 5261 |
that doesn't begin with an on_failure_jump. */ |
| 5262 |
if ((re_opcode_t) *p1 != on_failure_jump) |
| 5263 |
break; |
| 5264 |
|
| 5265 |
/* Still have to check that it's not an n-th |
| 5266 |
alternative that starts with an on_failure_jump. */ |
| 5267 |
p1++; |
| 5268 |
EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
| 5269 |
if ((re_opcode_t) p1[mcnt-3] != jump_past_alt) |
| 5270 |
{ |
| 5271 |
/* Get to the beginning of the n-th alternative. */ |
| 5272 |
p1 -= 3; |
| 5273 |
break; |
| 5274 |
} |
| 5275 |
} |
| 5276 |
|
| 5277 |
/* Deal with the last alternative: go back and get number |
| 5278 |
of the `jump_past_alt' just before it. `mcnt' contains |
| 5279 |
the length of the alternative. */ |
| 5280 |
EXTRACT_NUMBER (mcnt, p1 - 2); |
| 5281 |
|
| 5282 |
if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info)) |
| 5283 |
return false; |
| 5284 |
|
| 5285 |
p1 += mcnt; /* Get past the n-th alternative. */ |
| 5286 |
} /* if mcnt > 0 */ |
| 5287 |
break; |
| 5288 |
|
| 5289 |
|
| 5290 |
case stop_memory: |
| 5291 |
assert (p1[1] == **p); |
| 5292 |
*p = p1 + 2; |
| 5293 |
return true; |
| 5294 |
|
| 5295 |
|
| 5296 |
default: |
| 5297 |
if (!common_op_match_null_string_p (&p1, end, reg_info)) |
| 5298 |
return false; |
| 5299 |
} |
| 5300 |
} /* while p1 < end */ |
| 5301 |
|
| 5302 |
return false; |
| 5303 |
} /* group_match_null_string_p */ |
| 5304 |
|
| 5305 |
|
| 5306 |
/* Similar to group_match_null_string_p, but doesn't deal with alternatives: |
| 5307 |
It expects P to be the first byte of a single alternative and END one |
| 5308 |
byte past the last. The alternative can contain groups. */ |
| 5309 |
|
| 5310 |
static boolean |
| 5311 |
alt_match_null_string_p (p, end, reg_info) |
| 5312 |
unsigned char *p, *end; |
| 5313 |
register_info_type *reg_info; |
| 5314 |
{ |
| 5315 |
int mcnt; |
| 5316 |
unsigned char *p1 = p; |
| 5317 |
|
| 5318 |
while (p1 < end) |
| 5319 |
{ |
| 5320 |
/* Skip over opcodes that can match nothing, and break when we get |
| 5321 |
to one that can't. */ |
| 5322 |
|
| 5323 |
switch ((re_opcode_t) *p1) |
| 5324 |
{ |
| 5325 |
/* It's a loop. */ |
| 5326 |
case on_failure_jump: |
| 5327 |
p1++; |
| 5328 |
EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
| 5329 |
p1 += mcnt; |
| 5330 |
break; |
| 5331 |
|
| 5332 |
default: |
| 5333 |
if (!common_op_match_null_string_p (&p1, end, reg_info)) |
| 5334 |
return false; |
| 5335 |
} |
| 5336 |
} /* while p1 < end */ |
| 5337 |
|
| 5338 |
return true; |
| 5339 |
} /* alt_match_null_string_p */ |
| 5340 |
|
| 5341 |
|
| 5342 |
/* Deals with the ops common to group_match_null_string_p and |
| 5343 |
alt_match_null_string_p. |
| 5344 |
|
| 5345 |
Sets P to one after the op and its arguments, if any. */ |
| 5346 |
|
| 5347 |
static boolean |
| 5348 |
common_op_match_null_string_p (p, end, reg_info) |
| 5349 |
unsigned char **p, *end; |
| 5350 |
register_info_type *reg_info; |
| 5351 |
{ |
| 5352 |
int mcnt; |
| 5353 |
boolean ret; |
| 5354 |
int reg_no; |
| 5355 |
unsigned char *p1 = *p; |
| 5356 |
|
| 5357 |
switch ((re_opcode_t) *p1++) |
| 5358 |
{ |
| 5359 |
case no_op: |
| 5360 |
case begline: |
| 5361 |
case endline: |
| 5362 |
case begbuf: |
| 5363 |
case endbuf: |
| 5364 |
case wordbeg: |
| 5365 |
case wordend: |
| 5366 |
case wordbound: |
| 5367 |
case notwordbound: |
| 5368 |
#ifdef emacs |
| 5369 |
case before_dot: |
| 5370 |
case at_dot: |
| 5371 |
case after_dot: |
| 5372 |
#endif |
| 5373 |
break; |
| 5374 |
|
| 5375 |
case start_memory: |
| 5376 |
reg_no = *p1; |
| 5377 |
assert (reg_no > 0 && reg_no <= MAX_REGNUM); |
| 5378 |
ret = group_match_null_string_p (&p1, end, reg_info); |
| 5379 |
|
| 5380 |
/* Have to set this here in case we're checking a group which |
| 5381 |
contains a group and a back reference to it. */ |
| 5382 |
|
| 5383 |
if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) |
| 5384 |
REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; |
| 5385 |
|
| 5386 |
if (!ret) |
| 5387 |
return false; |
| 5388 |
break; |
| 5389 |
|
| 5390 |
/* If this is an optimized succeed_n for zero times, make the jump. */ |
| 5391 |
case jump: |
| 5392 |
EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
| 5393 |
if (mcnt >= 0) |
| 5394 |
p1 += mcnt; |
| 5395 |
else |
| 5396 |
return false; |
| 5397 |
break; |
| 5398 |
|
| 5399 |
case succeed_n: |
| 5400 |
/* Get to the number of times to succeed. */ |
| 5401 |
p1 += 2; |
| 5402 |
EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
| 5403 |
|
| 5404 |
if (mcnt == 0) |
| 5405 |
{ |
| 5406 |
p1 -= 4; |
| 5407 |
EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
| 5408 |
p1 += mcnt; |
| 5409 |
} |
| 5410 |
else |
| 5411 |
return false; |
| 5412 |
break; |
| 5413 |
|
| 5414 |
case duplicate: |
| 5415 |
if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) |
| 5416 |
return false; |
| 5417 |
break; |
| 5418 |
|
| 5419 |
case set_number_at: |
| 5420 |
p1 += 4; |
| 5421 |
|
| 5422 |
default: |
| 5423 |
/* All other opcodes mean we cannot match the empty string. */ |
| 5424 |
return false; |
| 5425 |
} |
| 5426 |
|
| 5427 |
*p = p1; |
| 5428 |
return true; |
| 5429 |
} /* common_op_match_null_string_p */ |
| 5430 |
|
| 5431 |
|
| 5432 |
/* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN |
| 5433 |
bytes; nonzero otherwise. */ |
| 5434 |
|
| 5435 |
static int |
| 5436 |
bcmp_translate (s1, s2, len, translate) |
| 5437 |
const char *s1, *s2; |
| 5438 |
register int len; |
| 5439 |
RE_TRANSLATE_TYPE translate; |
| 5440 |
{ |
| 5441 |
register const unsigned char *p1 = (const unsigned char *) s1; |
| 5442 |
register const unsigned char *p2 = (const unsigned char *) s2; |
| 5443 |
while (len) |
| 5444 |
{ |
| 5445 |
if (translate[*p1++] != translate[*p2++]) return 1; |
| 5446 |
len--; |
| 5447 |
} |
| 5448 |
return 0; |
| 5449 |
} |
| 5450 |
|
| 5451 |
/* Entry points for GNU code. */ |
| 5452 |
|
| 5453 |
/* re_compile_pattern is the GNU regular expression compiler: it |
| 5454 |
compiles PATTERN (of length SIZE) and puts the result in BUFP. |
| 5455 |
Returns 0 if the pattern was valid, otherwise an error string. |
| 5456 |
|
| 5457 |
Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
| 5458 |
are set in BUFP on entry. |
| 5459 |
|
| 5460 |
We call regex_compile to do the actual compilation. */ |
| 5461 |
|
| 5462 |
const char * |
| 5463 |
re_compile_pattern (pattern, length, bufp) |
| 5464 |
const char *pattern; |
| 5465 |
size_t length; |
| 5466 |
struct re_pattern_buffer *bufp; |
| 5467 |
{ |
| 5468 |
reg_errcode_t ret; |
| 5469 |
|
| 5470 |
/* GNU code is written to assume at least RE_NREGS registers will be set |
| 5471 |
(and at least one extra will be -1). */ |
| 5472 |
bufp->regs_allocated = REGS_UNALLOCATED; |
| 5473 |
|
| 5474 |
/* And GNU code determines whether or not to get register information |
| 5475 |
by passing null for the REGS argument to re_match, etc., not by |
| 5476 |
setting no_sub. */ |
| 5477 |
bufp->no_sub = 0; |
| 5478 |
|
| 5479 |
/* Match anchors at newline. */ |
| 5480 |
bufp->newline_anchor = 1; |
| 5481 |
|
| 5482 |
ret = regex_compile (pattern, length, re_syntax_options, bufp); |
| 5483 |
|
| 5484 |
if (!ret) |
| 5485 |
return NULL; |
| 5486 |
return gettext (re_error_msgid[(int) ret]); |
| 5487 |
} |
| 5488 |
#ifdef _LIBC |
| 5489 |
weak_alias (__re_compile_pattern, re_compile_pattern) |
| 5490 |
#endif |
| 5491 |
|
| 5492 |
/* Entry points compatible with 4.2 BSD regex library. We don't define |
| 5493 |
them unless specifically requested. */ |
| 5494 |
|
| 5495 |
#if defined _REGEX_RE_COMP || defined _LIBC |
| 5496 |
|
| 5497 |
/* BSD has one and only one pattern buffer. */ |
| 5498 |
static struct re_pattern_buffer re_comp_buf; |
| 5499 |
|
| 5500 |
char * |
| 5501 |
#ifdef _LIBC |
| 5502 |
/* Make these definitions weak in libc, so POSIX programs can redefine |
| 5503 |
these names if they don't use our functions, and still use |
| 5504 |
regcomp/regexec below without link errors. */ |
| 5505 |
weak_function |
| 5506 |
#endif |
| 5507 |
re_comp (s) |
| 5508 |
const char *s; |
| 5509 |
{ |
| 5510 |
reg_errcode_t ret; |
| 5511 |
|
| 5512 |
if (!s) |
| 5513 |
{ |
| 5514 |
if (!re_comp_buf.buffer) |
| 5515 |
return gettext ("No previous regular expression"); |
| 5516 |
return 0; |
| 5517 |
} |
| 5518 |
|
| 5519 |
if (!re_comp_buf.buffer) |
| 5520 |
{ |
| 5521 |
re_comp_buf.buffer = (unsigned char *) malloc (200); |
| 5522 |
if (re_comp_buf.buffer == NULL) |
| 5523 |
return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); |
| 5524 |
re_comp_buf.allocated = 200; |
| 5525 |
|
| 5526 |
re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); |
| 5527 |
if (re_comp_buf.fastmap == NULL) |
| 5528 |
return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); |
| 5529 |
} |
| 5530 |
|
| 5531 |
/* Since `re_exec' always passes NULL for the `regs' argument, we |
| 5532 |
don't need to initialize the pattern buffer fields which affect it. */ |
| 5533 |
|
| 5534 |
/* Match anchors at newlines. */ |
| 5535 |
re_comp_buf.newline_anchor = 1; |
| 5536 |
|
| 5537 |
ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); |
| 5538 |
|
| 5539 |
if (!ret) |
| 5540 |
return NULL; |
| 5541 |
|
| 5542 |
/* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
| 5543 |
return (char *) gettext (re_error_msgid[(int) ret]); |
| 5544 |
} |
| 5545 |
|
| 5546 |
|
| 5547 |
int |
| 5548 |
#ifdef _LIBC |
| 5549 |
weak_function |
| 5550 |
#endif |
| 5551 |
re_exec (s) |
| 5552 |
const char *s; |
| 5553 |
{ |
| 5554 |
const int len = strlen (s); |
| 5555 |
return |
| 5556 |
0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); |
| 5557 |
} |
| 5558 |
|
| 5559 |
#endif /* _REGEX_RE_COMP */ |
| 5560 |
|
| 5561 |
/* POSIX.2 functions. Don't define these for Emacs. */ |
| 5562 |
|
| 5563 |
#ifndef emacs |
| 5564 |
|
| 5565 |
/* regcomp takes a regular expression as a string and compiles it. |
| 5566 |
|
| 5567 |
PREG is a regex_t *. We do not expect any fields to be initialized, |
| 5568 |
since POSIX says we shouldn't. Thus, we set |
| 5569 |
|
| 5570 |
`buffer' to the compiled pattern; |
| 5571 |
`used' to the length of the compiled pattern; |
| 5572 |
`syntax' to RE_SYNTAX_POSIX_EXTENDED if the |
| 5573 |
REG_EXTENDED bit in CFLAGS is set; otherwise, to |
| 5574 |
RE_SYNTAX_POSIX_BASIC; |
| 5575 |
`newline_anchor' to REG_NEWLINE being set in CFLAGS; |
| 5576 |
`fastmap' to an allocated space for the fastmap; |
| 5577 |
`fastmap_accurate' to zero; |
| 5578 |
`re_nsub' to the number of subexpressions in PATTERN. |
| 5579 |
|
| 5580 |
PATTERN is the address of the pattern string. |
| 5581 |
|
| 5582 |
CFLAGS is a series of bits which affect compilation. |
| 5583 |
|
| 5584 |
If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we |
| 5585 |
use POSIX basic syntax. |
| 5586 |
|
| 5587 |
If REG_NEWLINE is set, then . and [^...] don't match newline. |
| 5588 |
Also, regexec will try a match beginning after every newline. |
| 5589 |
|
| 5590 |
If REG_ICASE is set, then we considers upper- and lowercase |
| 5591 |
versions of letters to be equivalent when matching. |
| 5592 |
|
| 5593 |
If REG_NOSUB is set, then when PREG is passed to regexec, that |
| 5594 |
routine will report only success or failure, and nothing about the |
| 5595 |
registers. |
| 5596 |
|
| 5597 |
It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
| 5598 |
the return codes and their meanings.) */ |
| 5599 |
|
| 5600 |
int |
| 5601 |
regcomp (preg, pattern, cflags) |
| 5602 |
regex_t *preg; |
| 5603 |
const char *pattern; |
| 5604 |
int cflags; |
| 5605 |
{ |
| 5606 |
reg_errcode_t ret; |
| 5607 |
reg_syntax_t syntax |
| 5608 |
= (cflags & REG_EXTENDED) ? |
| 5609 |
RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; |
| 5610 |
|
| 5611 |
/* regex_compile will allocate the space for the compiled pattern. */ |
| 5612 |
preg->buffer = 0; |
| 5613 |
preg->allocated = 0; |
| 5614 |
preg->used = 0; |
| 5615 |
|
| 5616 |
/* Try to allocate space for the fastmap. */ |
| 5617 |
preg->fastmap = (char *) malloc (1 << BYTEWIDTH); |
| 5618 |
|
| 5619 |
if (cflags & REG_ICASE) |
| 5620 |
{ |
| 5621 |
unsigned i; |
| 5622 |
|
| 5623 |
preg->translate |
| 5624 |
= (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE |
| 5625 |
* sizeof (*(RE_TRANSLATE_TYPE)0)); |
| 5626 |
if (preg->translate == NULL) |
| 5627 |
return (int) REG_ESPACE; |
| 5628 |
|
| 5629 |
/* Map uppercase characters to corresponding lowercase ones. */ |
| 5630 |
for (i = 0; i < CHAR_SET_SIZE; i++) |
| 5631 |
preg->translate[i] = ISUPPER (i) ? tolower (i) : i; |
| 5632 |
} |
| 5633 |
else |
| 5634 |
preg->translate = NULL; |
| 5635 |
|
| 5636 |
/* If REG_NEWLINE is set, newlines are treated differently. */ |
| 5637 |
if (cflags & REG_NEWLINE) |
| 5638 |
{ /* REG_NEWLINE implies neither . nor [^...] match newline. */ |
| 5639 |
syntax &= ~RE_DOT_NEWLINE; |
| 5640 |
syntax |= RE_HAT_LISTS_NOT_NEWLINE; |
| 5641 |
/* It also changes the matching behavior. */ |
| 5642 |
preg->newline_anchor = 1; |
| 5643 |
} |
| 5644 |
else |
| 5645 |
preg->newline_anchor = 0; |
| 5646 |
|
| 5647 |
preg->no_sub = !!(cflags & REG_NOSUB); |
| 5648 |
|
| 5649 |
/* POSIX says a null character in the pattern terminates it, so we |
| 5650 |
can use strlen here in compiling the pattern. */ |
| 5651 |
ret = regex_compile (pattern, strlen (pattern), syntax, preg); |
| 5652 |
|
| 5653 |
/* POSIX doesn't distinguish between an unmatched open-group and an |
| 5654 |
unmatched close-group: both are REG_EPAREN. */ |
| 5655 |
if (ret == REG_ERPAREN) ret = REG_EPAREN; |
| 5656 |
|
| 5657 |
if (ret == REG_NOERROR && preg->fastmap) |
| 5658 |
{ |
| 5659 |
/* Compute the fastmap now, since regexec cannot modify the pattern |
| 5660 |
buffer. */ |
| 5661 |
if (re_compile_fastmap (preg) == -2) |
| 5662 |
{ |
| 5663 |
/* Some error occured while computing the fastmap, just forget |
| 5664 |
about it. */ |
| 5665 |
free (preg->fastmap); |
| 5666 |
preg->fastmap = NULL; |
| 5667 |
} |
| 5668 |
} |
| 5669 |
|
| 5670 |
return (int) ret; |
| 5671 |
} |
| 5672 |
#ifdef _LIBC |
| 5673 |
weak_alias (__regcomp, regcomp) |
| 5674 |
#endif |
| 5675 |
|
| 5676 |
|
| 5677 |
/* regexec searches for a given pattern, specified by PREG, in the |
| 5678 |
string STRING. |
| 5679 |
|
| 5680 |
If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
| 5681 |
`regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
| 5682 |
least NMATCH elements, and we set them to the offsets of the |
| 5683 |
corresponding matched substrings. |
| 5684 |
|
| 5685 |
EFLAGS specifies `execution flags' which affect matching: if |
| 5686 |
REG_NOTBOL is set, then ^ does not match at the beginning of the |
| 5687 |
string; if REG_NOTEOL is set, then $ does not match at the end. |
| 5688 |
|
| 5689 |
We return 0 if we find a match and REG_NOMATCH if not. */ |
| 5690 |
|
| 5691 |
int |
| 5692 |
regexec (preg, string, nmatch, pmatch, eflags) |
| 5693 |
const regex_t *preg; |
| 5694 |
const char *string; |
| 5695 |
size_t nmatch; |
| 5696 |
regmatch_t pmatch[]; |
| 5697 |
int eflags; |
| 5698 |
{ |
| 5699 |
int ret; |
| 5700 |
struct re_registers regs; |
| 5701 |
regex_t private_preg; |
| 5702 |
int len = strlen (string); |
| 5703 |
boolean want_reg_info = !preg->no_sub && nmatch > 0; |
| 5704 |
|
| 5705 |
private_preg = *preg; |
| 5706 |
|
| 5707 |
private_preg.not_bol = !!(eflags & REG_NOTBOL); |
| 5708 |
private_preg.not_eol = !!(eflags & REG_NOTEOL); |
| 5709 |
|
| 5710 |
/* The user has told us exactly how many registers to return |
| 5711 |
information about, via `nmatch'. We have to pass that on to the |
| 5712 |
matching routines. */ |
| 5713 |
private_preg.regs_allocated = REGS_FIXED; |
| 5714 |
|
| 5715 |
if (want_reg_info) |
| 5716 |
{ |
| 5717 |
regs.num_regs = nmatch; |
| 5718 |
regs.start = TALLOC (nmatch * 2, regoff_t); |
| 5719 |
if (regs.start == NULL) |
| 5720 |
return (int) REG_NOMATCH; |
| 5721 |
regs.end = regs.start + nmatch; |
| 5722 |
} |
| 5723 |
|
| 5724 |
/* Perform the searching operation. */ |
| 5725 |
ret = re_search (&private_preg, string, len, |
| 5726 |
/* start: */ 0, /* range: */ len, |
| 5727 |
want_reg_info ? ®s : (struct re_registers *) 0); |
| 5728 |
|
| 5729 |
/* Copy the register information to the POSIX structure. */ |
| 5730 |
if (want_reg_info) |
| 5731 |
{ |
| 5732 |
if (ret >= 0) |
| 5733 |
{ |
| 5734 |
unsigned r; |
| 5735 |
|
| 5736 |
for (r = 0; r < nmatch; r++) |
| 5737 |
{ |
| 5738 |
pmatch[r].rm_so = regs.start[r]; |
| 5739 |
pmatch[r].rm_eo = regs.end[r]; |
| 5740 |
} |
| 5741 |
} |
| 5742 |
|
| 5743 |
/* If we needed the temporary register info, free the space now. */ |
| 5744 |
free (regs.start); |
| 5745 |
} |
| 5746 |
|
| 5747 |
/* We want zero return to mean success, unlike `re_search'. */ |
| 5748 |
return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; |
| 5749 |
} |
| 5750 |
#ifdef _LIBC |
| 5751 |
weak_alias (__regexec, regexec) |
| 5752 |
#endif |
| 5753 |
|
| 5754 |
|
| 5755 |
/* Returns a message corresponding to an error code, ERRCODE, returned |
| 5756 |
from either regcomp or regexec. We don't use PREG here. */ |
| 5757 |
|
| 5758 |
size_t |
| 5759 |
regerror (errcode, preg, errbuf, errbuf_size) |
| 5760 |
int errcode; |
| 5761 |
const regex_t *preg; |
| 5762 |
char *errbuf; |
| 5763 |
size_t errbuf_size; |
| 5764 |
{ |
| 5765 |
const char *msg; |
| 5766 |
size_t msg_size; |
| 5767 |
|
| 5768 |
if (errcode < 0 |
| 5769 |
|| errcode >= (int) (sizeof (re_error_msgid) |
| 5770 |
/ sizeof (re_error_msgid[0]))) |
| 5771 |
/* Only error codes returned by the rest of the code should be passed |
| 5772 |
to this routine. If we are given anything else, or if other regex |
| 5773 |
code generates an invalid error code, then the program has a bug. |
| 5774 |
Dump core so we can fix it. */ |
| 5775 |
abort (); |
| 5776 |
|
| 5777 |
msg = gettext (re_error_msgid[errcode]); |
| 5778 |
|
| 5779 |
msg_size = strlen (msg) + 1; /* Includes the null. */ |
| 5780 |
|
| 5781 |
if (errbuf_size != 0) |
| 5782 |
{ |
| 5783 |
if (msg_size > errbuf_size) |
| 5784 |
{ |
| 5785 |
#if defined HAVE_MEMPCPY || defined _LIBC |
| 5786 |
*((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0'; |
| 5787 |
#else |
| 5788 |
memcpy (errbuf, msg, errbuf_size - 1); |
| 5789 |
errbuf[errbuf_size - 1] = 0; |
| 5790 |
#endif |
| 5791 |
} |
| 5792 |
else |
| 5793 |
memcpy (errbuf, msg, msg_size); |
| 5794 |
} |
| 5795 |
|
| 5796 |
return msg_size; |
| 5797 |
} |
| 5798 |
#ifdef _LIBC |
| 5799 |
weak_alias (__regerror, regerror) |
| 5800 |
#endif |
| 5801 |
|
| 5802 |
|
| 5803 |
/* Free dynamically allocated space used by PREG. */ |
| 5804 |
|
| 5805 |
void |
| 5806 |
regfree (preg) |
| 5807 |
regex_t *preg; |
| 5808 |
{ |
| 5809 |
if (preg->buffer != NULL) |
| 5810 |
free (preg->buffer); |
| 5811 |
preg->buffer = NULL; |
| 5812 |
|
| 5813 |
preg->allocated = 0; |
| 5814 |
preg->used = 0; |
| 5815 |
|
| 5816 |
if (preg->fastmap != NULL) |
| 5817 |
free (preg->fastmap); |
| 5818 |
preg->fastmap = NULL; |
| 5819 |
preg->fastmap_accurate = 0; |
| 5820 |
|
| 5821 |
if (preg->translate != NULL) |
| 5822 |
free (preg->translate); |
| 5823 |
preg->translate = NULL; |
| 5824 |
} |
| 5825 |
#ifdef _LIBC |
| 5826 |
weak_alias (__regfree, regfree) |
| 5827 |
#endif |
| 5828 |
|
| 5829 |
#endif /* not emacs */ |