| 1 | /* Extended regular expression matching and search library, version |
| 2 | 0.12. (Implements POSIX draft P1003.2/D11.2, except for some of the |
| 3 | internationalization features.) |
| 4 | |
| 5 | Copyright (C) 1993-2013 Free Software Foundation, Inc. |
| 6 | |
| 7 | This program is free software; you can redistribute it and/or modify |
| 8 | it under the terms of the GNU General Public License as published by |
| 9 | the Free Software Foundation; either version 3, or (at your option) |
| 10 | any later version. |
| 11 | |
| 12 | This program 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 |
| 15 | GNU General Public License for more details. |
| 16 | |
| 17 | You should have received a copy of the GNU General Public License |
| 18 | along with this program. If not, see <http://www.gnu.org/licenses/>. */ |
| 19 | |
| 20 | /* TODO: |
| 21 | - structure the opcode space into opcode+flag. |
| 22 | - merge with glibc's regex.[ch]. |
| 23 | - replace (succeed_n + jump_n + set_number_at) with something that doesn't |
| 24 | need to modify the compiled regexp so that re_match can be reentrant. |
| 25 | - get rid of on_failure_jump_smart by doing the optimization in re_comp |
| 26 | rather than at run-time, so that re_match can be reentrant. |
| 27 | */ |
| 28 | |
| 29 | /* AIX requires this to be the first thing in the file. */ |
| 30 | #if defined _AIX && !defined REGEX_MALLOC |
| 31 | #pragma alloca |
| 32 | #endif |
| 33 | |
| 34 | /* Ignore some GCC warnings for now. This section should go away |
| 35 | once the Emacs and Gnulib regex code is merged. */ |
| 36 | #if 4 < __GNUC__ + (5 <= __GNUC_MINOR__) || defined __clang__ |
| 37 | # pragma GCC diagnostic ignored "-Wstrict-overflow" |
| 38 | # ifndef emacs |
| 39 | # pragma GCC diagnostic ignored "-Wunused-function" |
| 40 | # pragma GCC diagnostic ignored "-Wunused-macros" |
| 41 | # pragma GCC diagnostic ignored "-Wunused-result" |
| 42 | # pragma GCC diagnostic ignored "-Wunused-variable" |
| 43 | # endif |
| 44 | #endif |
| 45 | |
| 46 | #if 4 < __GNUC__ + (5 <= __GNUC_MINOR__) && ! defined __clang__ |
| 47 | # pragma GCC diagnostic ignored "-Wunused-but-set-variable" |
| 48 | #endif |
| 49 | |
| 50 | #include <config.h> |
| 51 | |
| 52 | #include <stddef.h> |
| 53 | |
| 54 | #ifdef emacs |
| 55 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ |
| 56 | # include <sys/types.h> |
| 57 | #endif |
| 58 | |
| 59 | /* Whether to use ISO C Amendment 1 wide char functions. |
| 60 | Those should not be used for Emacs since it uses its own. */ |
| 61 | #if defined _LIBC |
| 62 | #define WIDE_CHAR_SUPPORT 1 |
| 63 | #else |
| 64 | #define WIDE_CHAR_SUPPORT \ |
| 65 | (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC && !emacs) |
| 66 | #endif |
| 67 | |
| 68 | /* For platform which support the ISO C amendment 1 functionality we |
| 69 | support user defined character classes. */ |
| 70 | #if WIDE_CHAR_SUPPORT |
| 71 | /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ |
| 72 | # include <wchar.h> |
| 73 | # include <wctype.h> |
| 74 | #endif |
| 75 | |
| 76 | #ifdef _LIBC |
| 77 | /* We have to keep the namespace clean. */ |
| 78 | # define regfree(preg) __regfree (preg) |
| 79 | # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) |
| 80 | # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) |
| 81 | # define regerror(err_code, preg, errbuf, errbuf_size) \ |
| 82 | __regerror (err_code, preg, errbuf, errbuf_size) |
| 83 | # define re_set_registers(bu, re, nu, st, en) \ |
| 84 | __re_set_registers (bu, re, nu, st, en) |
| 85 | # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ |
| 86 | __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) |
| 87 | # define re_match(bufp, string, size, pos, regs) \ |
| 88 | __re_match (bufp, string, size, pos, regs) |
| 89 | # define re_search(bufp, string, size, startpos, range, regs) \ |
| 90 | __re_search (bufp, string, size, startpos, range, regs) |
| 91 | # define re_compile_pattern(pattern, length, bufp) \ |
| 92 | __re_compile_pattern (pattern, length, bufp) |
| 93 | # define re_set_syntax(syntax) __re_set_syntax (syntax) |
| 94 | # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ |
| 95 | __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) |
| 96 | # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) |
| 97 | |
| 98 | /* Make sure we call libc's function even if the user overrides them. */ |
| 99 | # define btowc __btowc |
| 100 | # define iswctype __iswctype |
| 101 | # define wctype __wctype |
| 102 | |
| 103 | # define WEAK_ALIAS(a,b) weak_alias (a, b) |
| 104 | |
| 105 | /* We are also using some library internals. */ |
| 106 | # include <locale/localeinfo.h> |
| 107 | # include <locale/elem-hash.h> |
| 108 | # include <langinfo.h> |
| 109 | #else |
| 110 | # define WEAK_ALIAS(a,b) |
| 111 | #endif |
| 112 | |
| 113 | /* This is for other GNU distributions with internationalized messages. */ |
| 114 | #if HAVE_LIBINTL_H || defined _LIBC |
| 115 | # include <libintl.h> |
| 116 | #else |
| 117 | # define gettext(msgid) (msgid) |
| 118 | #endif |
| 119 | |
| 120 | #ifndef gettext_noop |
| 121 | /* This define is so xgettext can find the internationalizable |
| 122 | strings. */ |
| 123 | # define gettext_noop(String) String |
| 124 | #endif |
| 125 | |
| 126 | /* The `emacs' switch turns on certain matching commands |
| 127 | that make sense only in Emacs. */ |
| 128 | #ifdef emacs |
| 129 | |
| 130 | # include "lisp.h" |
| 131 | # include "character.h" |
| 132 | # include "buffer.h" |
| 133 | |
| 134 | /* Make syntax table lookup grant data in gl_state. */ |
| 135 | # define SYNTAX_ENTRY_VIA_PROPERTY |
| 136 | |
| 137 | # include "syntax.h" |
| 138 | # include "category.h" |
| 139 | |
| 140 | # ifdef malloc |
| 141 | # undef malloc |
| 142 | # endif |
| 143 | # define malloc xmalloc |
| 144 | # ifdef realloc |
| 145 | # undef realloc |
| 146 | # endif |
| 147 | # define realloc xrealloc |
| 148 | # ifdef free |
| 149 | # undef free |
| 150 | # endif |
| 151 | # define free xfree |
| 152 | |
| 153 | /* Converts the pointer to the char to BEG-based offset from the start. */ |
| 154 | # define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d)) |
| 155 | # define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object))) |
| 156 | |
| 157 | # define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte) |
| 158 | # define RE_TARGET_MULTIBYTE_P(bufp) ((bufp)->target_multibyte) |
| 159 | # define RE_STRING_CHAR(p, multibyte) \ |
| 160 | (multibyte ? (STRING_CHAR (p)) : (*(p))) |
| 161 | # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) \ |
| 162 | (multibyte ? (STRING_CHAR_AND_LENGTH (p, len)) : ((len) = 1, *(p))) |
| 163 | |
| 164 | # define RE_CHAR_TO_MULTIBYTE(c) UNIBYTE_TO_CHAR (c) |
| 165 | |
| 166 | # define RE_CHAR_TO_UNIBYTE(c) CHAR_TO_BYTE_SAFE (c) |
| 167 | |
| 168 | /* Set C a (possibly converted to multibyte) character before P. P |
| 169 | points into a string which is the virtual concatenation of STR1 |
| 170 | (which ends at END1) or STR2 (which ends at END2). */ |
| 171 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ |
| 172 | do { \ |
| 173 | if (target_multibyte) \ |
| 174 | { \ |
| 175 | re_char *dtemp = (p) == (str2) ? (end1) : (p); \ |
| 176 | re_char *dlimit = ((p) > (str2) && (p) <= (end2)) ? (str2) : (str1); \ |
| 177 | while (dtemp-- > dlimit && !CHAR_HEAD_P (*dtemp)); \ |
| 178 | c = STRING_CHAR (dtemp); \ |
| 179 | } \ |
| 180 | else \ |
| 181 | { \ |
| 182 | (c = ((p) == (str2) ? (end1) : (p))[-1]); \ |
| 183 | (c) = RE_CHAR_TO_MULTIBYTE (c); \ |
| 184 | } \ |
| 185 | } while (0) |
| 186 | |
| 187 | /* Set C a (possibly converted to multibyte) character at P, and set |
| 188 | LEN to the byte length of that character. */ |
| 189 | # define GET_CHAR_AFTER(c, p, len) \ |
| 190 | do { \ |
| 191 | if (target_multibyte) \ |
| 192 | (c) = STRING_CHAR_AND_LENGTH (p, len); \ |
| 193 | else \ |
| 194 | { \ |
| 195 | (c) = *p; \ |
| 196 | len = 1; \ |
| 197 | (c) = RE_CHAR_TO_MULTIBYTE (c); \ |
| 198 | } \ |
| 199 | } while (0) |
| 200 | |
| 201 | #else /* not emacs */ |
| 202 | |
| 203 | /* If we are not linking with Emacs proper, |
| 204 | we can't use the relocating allocator |
| 205 | even if config.h says that we can. */ |
| 206 | # undef REL_ALLOC |
| 207 | |
| 208 | # include <unistd.h> |
| 209 | |
| 210 | /* When used in Emacs's lib-src, we need xmalloc and xrealloc. */ |
| 211 | |
| 212 | static void * |
| 213 | xmalloc (size_t size) |
| 214 | { |
| 215 | void *val = malloc (size); |
| 216 | if (!val && size) |
| 217 | { |
| 218 | write (2, "virtual memory exhausted\n", 25); |
| 219 | exit (1); |
| 220 | } |
| 221 | return val; |
| 222 | } |
| 223 | |
| 224 | static void * |
| 225 | xrealloc (void *block, size_t size) |
| 226 | { |
| 227 | void *val; |
| 228 | /* We must call malloc explicitly when BLOCK is 0, since some |
| 229 | reallocs don't do this. */ |
| 230 | if (! block) |
| 231 | val = malloc (size); |
| 232 | else |
| 233 | val = realloc (block, size); |
| 234 | if (!val && size) |
| 235 | { |
| 236 | write (2, "virtual memory exhausted\n", 25); |
| 237 | exit (1); |
| 238 | } |
| 239 | return val; |
| 240 | } |
| 241 | |
| 242 | # ifdef malloc |
| 243 | # undef malloc |
| 244 | # endif |
| 245 | # define malloc xmalloc |
| 246 | # ifdef realloc |
| 247 | # undef realloc |
| 248 | # endif |
| 249 | # define realloc xrealloc |
| 250 | |
| 251 | # include <stdbool.h> |
| 252 | # include <string.h> |
| 253 | |
| 254 | /* Define the syntax stuff for \<, \>, etc. */ |
| 255 | |
| 256 | /* Sword must be nonzero for the wordchar pattern commands in re_match_2. */ |
| 257 | enum syntaxcode { Swhitespace = 0, Sword = 1, Ssymbol = 2 }; |
| 258 | |
| 259 | /* Dummy macros for non-Emacs environments. */ |
| 260 | # define CHAR_CHARSET(c) 0 |
| 261 | # define CHARSET_LEADING_CODE_BASE(c) 0 |
| 262 | # define MAX_MULTIBYTE_LENGTH 1 |
| 263 | # define RE_MULTIBYTE_P(x) 0 |
| 264 | # define RE_TARGET_MULTIBYTE_P(x) 0 |
| 265 | # define WORD_BOUNDARY_P(c1, c2) (0) |
| 266 | # define CHAR_HEAD_P(p) (1) |
| 267 | # define SINGLE_BYTE_CHAR_P(c) (1) |
| 268 | # define SAME_CHARSET_P(c1, c2) (1) |
| 269 | # define BYTES_BY_CHAR_HEAD(p) (1) |
| 270 | # define PREV_CHAR_BOUNDARY(p, limit) ((p)--) |
| 271 | # define STRING_CHAR(p) (*(p)) |
| 272 | # define RE_STRING_CHAR(p, multibyte) STRING_CHAR (p) |
| 273 | # define CHAR_STRING(c, s) (*(s) = (c), 1) |
| 274 | # define STRING_CHAR_AND_LENGTH(p, actual_len) ((actual_len) = 1, *(p)) |
| 275 | # define RE_STRING_CHAR_AND_LENGTH(p, len, multibyte) STRING_CHAR_AND_LENGTH (p, len) |
| 276 | # define RE_CHAR_TO_MULTIBYTE(c) (c) |
| 277 | # define RE_CHAR_TO_UNIBYTE(c) (c) |
| 278 | # define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \ |
| 279 | (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1))) |
| 280 | # define GET_CHAR_AFTER(c, p, len) \ |
| 281 | (c = *p, len = 1) |
| 282 | # define MAKE_CHAR(charset, c1, c2) (c1) |
| 283 | # define BYTE8_TO_CHAR(c) (c) |
| 284 | # define CHAR_BYTE8_P(c) (0) |
| 285 | # define CHAR_LEADING_CODE(c) (c) |
| 286 | |
| 287 | #endif /* not emacs */ |
| 288 | |
| 289 | #ifndef RE_TRANSLATE |
| 290 | # define RE_TRANSLATE(TBL, C) ((unsigned char)(TBL)[C]) |
| 291 | # define RE_TRANSLATE_P(TBL) (TBL) |
| 292 | #endif |
| 293 | \f |
| 294 | /* Get the interface, including the syntax bits. */ |
| 295 | #include "regex.h" |
| 296 | |
| 297 | /* isalpha etc. are used for the character classes. */ |
| 298 | #include <ctype.h> |
| 299 | |
| 300 | #ifdef emacs |
| 301 | |
| 302 | /* 1 if C is an ASCII character. */ |
| 303 | # define IS_REAL_ASCII(c) ((c) < 0200) |
| 304 | |
| 305 | /* 1 if C is a unibyte character. */ |
| 306 | # define ISUNIBYTE(c) (SINGLE_BYTE_CHAR_P ((c))) |
| 307 | |
| 308 | /* The Emacs definitions should not be directly affected by locales. */ |
| 309 | |
| 310 | /* In Emacs, these are only used for single-byte characters. */ |
| 311 | # define ISDIGIT(c) ((c) >= '0' && (c) <= '9') |
| 312 | # define ISCNTRL(c) ((c) < ' ') |
| 313 | # define ISXDIGIT(c) (((c) >= '0' && (c) <= '9') \ |
| 314 | || ((c) >= 'a' && (c) <= 'f') \ |
| 315 | || ((c) >= 'A' && (c) <= 'F')) |
| 316 | |
| 317 | /* This is only used for single-byte characters. */ |
| 318 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
| 319 | |
| 320 | /* The rest must handle multibyte characters. */ |
| 321 | |
| 322 | # define ISGRAPH(c) (SINGLE_BYTE_CHAR_P (c) \ |
| 323 | ? (c) > ' ' && !((c) >= 0177 && (c) <= 0237) \ |
| 324 | : 1) |
| 325 | |
| 326 | # define ISPRINT(c) (SINGLE_BYTE_CHAR_P (c) \ |
| 327 | ? (c) >= ' ' && !((c) >= 0177 && (c) <= 0237) \ |
| 328 | : 1) |
| 329 | |
| 330 | # define ISALNUM(c) (IS_REAL_ASCII (c) \ |
| 331 | ? (((c) >= 'a' && (c) <= 'z') \ |
| 332 | || ((c) >= 'A' && (c) <= 'Z') \ |
| 333 | || ((c) >= '0' && (c) <= '9')) \ |
| 334 | : SYNTAX (c) == Sword) |
| 335 | |
| 336 | # define ISALPHA(c) (IS_REAL_ASCII (c) \ |
| 337 | ? (((c) >= 'a' && (c) <= 'z') \ |
| 338 | || ((c) >= 'A' && (c) <= 'Z')) \ |
| 339 | : SYNTAX (c) == Sword) |
| 340 | |
| 341 | # define ISLOWER(c) lowercasep (c) |
| 342 | |
| 343 | # define ISPUNCT(c) (IS_REAL_ASCII (c) \ |
| 344 | ? ((c) > ' ' && (c) < 0177 \ |
| 345 | && !(((c) >= 'a' && (c) <= 'z') \ |
| 346 | || ((c) >= 'A' && (c) <= 'Z') \ |
| 347 | || ((c) >= '0' && (c) <= '9'))) \ |
| 348 | : SYNTAX (c) != Sword) |
| 349 | |
| 350 | # define ISSPACE(c) (SYNTAX (c) == Swhitespace) |
| 351 | |
| 352 | # define ISUPPER(c) uppercasep (c) |
| 353 | |
| 354 | # define ISWORD(c) (SYNTAX (c) == Sword) |
| 355 | |
| 356 | #else /* not emacs */ |
| 357 | |
| 358 | /* 1 if C is an ASCII character. */ |
| 359 | # define IS_REAL_ASCII(c) ((c) < 0200) |
| 360 | |
| 361 | /* This distinction is not meaningful, except in Emacs. */ |
| 362 | # define ISUNIBYTE(c) 1 |
| 363 | |
| 364 | # ifdef isblank |
| 365 | # define ISBLANK(c) isblank (c) |
| 366 | # else |
| 367 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
| 368 | # endif |
| 369 | # ifdef isgraph |
| 370 | # define ISGRAPH(c) isgraph (c) |
| 371 | # else |
| 372 | # define ISGRAPH(c) (isprint (c) && !isspace (c)) |
| 373 | # endif |
| 374 | |
| 375 | /* Solaris defines ISPRINT so we must undefine it first. */ |
| 376 | # undef ISPRINT |
| 377 | # define ISPRINT(c) isprint (c) |
| 378 | # define ISDIGIT(c) isdigit (c) |
| 379 | # define ISALNUM(c) isalnum (c) |
| 380 | # define ISALPHA(c) isalpha (c) |
| 381 | # define ISCNTRL(c) iscntrl (c) |
| 382 | # define ISLOWER(c) islower (c) |
| 383 | # define ISPUNCT(c) ispunct (c) |
| 384 | # define ISSPACE(c) isspace (c) |
| 385 | # define ISUPPER(c) isupper (c) |
| 386 | # define ISXDIGIT(c) isxdigit (c) |
| 387 | |
| 388 | # define ISWORD(c) ISALPHA (c) |
| 389 | |
| 390 | # ifdef _tolower |
| 391 | # define TOLOWER(c) _tolower (c) |
| 392 | # else |
| 393 | # define TOLOWER(c) tolower (c) |
| 394 | # endif |
| 395 | |
| 396 | /* How many characters in the character set. */ |
| 397 | # define CHAR_SET_SIZE 256 |
| 398 | |
| 399 | # ifdef SYNTAX_TABLE |
| 400 | |
| 401 | extern char *re_syntax_table; |
| 402 | |
| 403 | # else /* not SYNTAX_TABLE */ |
| 404 | |
| 405 | static char re_syntax_table[CHAR_SET_SIZE]; |
| 406 | |
| 407 | static void |
| 408 | init_syntax_once (void) |
| 409 | { |
| 410 | register int c; |
| 411 | static int done = 0; |
| 412 | |
| 413 | if (done) |
| 414 | return; |
| 415 | |
| 416 | memset (re_syntax_table, 0, sizeof re_syntax_table); |
| 417 | |
| 418 | for (c = 0; c < CHAR_SET_SIZE; ++c) |
| 419 | if (ISALNUM (c)) |
| 420 | re_syntax_table[c] = Sword; |
| 421 | |
| 422 | re_syntax_table['_'] = Ssymbol; |
| 423 | |
| 424 | done = 1; |
| 425 | } |
| 426 | |
| 427 | # endif /* not SYNTAX_TABLE */ |
| 428 | |
| 429 | # define SYNTAX(c) re_syntax_table[(c)] |
| 430 | |
| 431 | #endif /* not emacs */ |
| 432 | \f |
| 433 | #define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
| 434 | \f |
| 435 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we |
| 436 | use `alloca' instead of `malloc'. This is because using malloc in |
| 437 | re_search* or re_match* could cause memory leaks when C-g is used in |
| 438 | Emacs; also, malloc is slower and causes storage fragmentation. On |
| 439 | the other hand, malloc is more portable, and easier to debug. |
| 440 | |
| 441 | Because we sometimes use alloca, some routines have to be macros, |
| 442 | not functions -- `alloca'-allocated space disappears at the end of the |
| 443 | function it is called in. */ |
| 444 | |
| 445 | #ifdef REGEX_MALLOC |
| 446 | |
| 447 | # define REGEX_ALLOCATE malloc |
| 448 | # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) |
| 449 | # define REGEX_FREE free |
| 450 | |
| 451 | #else /* not REGEX_MALLOC */ |
| 452 | |
| 453 | /* Emacs already defines alloca, sometimes. */ |
| 454 | # ifndef alloca |
| 455 | |
| 456 | /* Make alloca work the best possible way. */ |
| 457 | # ifdef __GNUC__ |
| 458 | # define alloca __builtin_alloca |
| 459 | # else /* not __GNUC__ */ |
| 460 | # ifdef HAVE_ALLOCA_H |
| 461 | # include <alloca.h> |
| 462 | # endif /* HAVE_ALLOCA_H */ |
| 463 | # endif /* not __GNUC__ */ |
| 464 | |
| 465 | # endif /* not alloca */ |
| 466 | |
| 467 | # define REGEX_ALLOCATE alloca |
| 468 | |
| 469 | /* Assumes a `char *destination' variable. */ |
| 470 | # define REGEX_REALLOCATE(source, osize, nsize) \ |
| 471 | (destination = (char *) alloca (nsize), \ |
| 472 | memcpy (destination, source, osize)) |
| 473 | |
| 474 | /* No need to do anything to free, after alloca. */ |
| 475 | # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
| 476 | |
| 477 | #endif /* not REGEX_MALLOC */ |
| 478 | |
| 479 | /* Define how to allocate the failure stack. */ |
| 480 | |
| 481 | #if defined REL_ALLOC && defined REGEX_MALLOC |
| 482 | |
| 483 | # define REGEX_ALLOCATE_STACK(size) \ |
| 484 | r_alloc (&failure_stack_ptr, (size)) |
| 485 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
| 486 | r_re_alloc (&failure_stack_ptr, (nsize)) |
| 487 | # define REGEX_FREE_STACK(ptr) \ |
| 488 | r_alloc_free (&failure_stack_ptr) |
| 489 | |
| 490 | #else /* not using relocating allocator */ |
| 491 | |
| 492 | # ifdef REGEX_MALLOC |
| 493 | |
| 494 | # define REGEX_ALLOCATE_STACK malloc |
| 495 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) |
| 496 | # define REGEX_FREE_STACK free |
| 497 | |
| 498 | # else /* not REGEX_MALLOC */ |
| 499 | |
| 500 | # define REGEX_ALLOCATE_STACK alloca |
| 501 | |
| 502 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
| 503 | REGEX_REALLOCATE (source, osize, nsize) |
| 504 | /* No need to explicitly free anything. */ |
| 505 | # define REGEX_FREE_STACK(arg) ((void)0) |
| 506 | |
| 507 | # endif /* not REGEX_MALLOC */ |
| 508 | #endif /* not using relocating allocator */ |
| 509 | |
| 510 | |
| 511 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside |
| 512 | `string1' or just past its end. This works if PTR is NULL, which is |
| 513 | a good thing. */ |
| 514 | #define FIRST_STRING_P(ptr) \ |
| 515 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
| 516 | |
| 517 | /* (Re)Allocate N items of type T using malloc, or fail. */ |
| 518 | #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) |
| 519 | #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) |
| 520 | #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) |
| 521 | |
| 522 | #define BYTEWIDTH 8 /* In bits. */ |
| 523 | |
| 524 | #define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) |
| 525 | |
| 526 | #undef MAX |
| 527 | #undef MIN |
| 528 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) |
| 529 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) |
| 530 | |
| 531 | /* Type of source-pattern and string chars. */ |
| 532 | #ifdef _MSC_VER |
| 533 | typedef unsigned char re_char; |
| 534 | #else |
| 535 | typedef const unsigned char re_char; |
| 536 | #endif |
| 537 | |
| 538 | typedef char boolean; |
| 539 | |
| 540 | static regoff_t re_match_2_internal (struct re_pattern_buffer *bufp, |
| 541 | re_char *string1, size_t size1, |
| 542 | re_char *string2, size_t size2, |
| 543 | ssize_t pos, |
| 544 | struct re_registers *regs, |
| 545 | ssize_t stop); |
| 546 | \f |
| 547 | /* These are the command codes that appear in compiled regular |
| 548 | expressions. Some opcodes are followed by argument bytes. A |
| 549 | command code can specify any interpretation whatsoever for its |
| 550 | arguments. Zero bytes may appear in the compiled regular expression. */ |
| 551 | |
| 552 | typedef enum |
| 553 | { |
| 554 | no_op = 0, |
| 555 | |
| 556 | /* Succeed right away--no more backtracking. */ |
| 557 | succeed, |
| 558 | |
| 559 | /* Followed by one byte giving n, then by n literal bytes. */ |
| 560 | exactn, |
| 561 | |
| 562 | /* Matches any (more or less) character. */ |
| 563 | anychar, |
| 564 | |
| 565 | /* Matches any one char belonging to specified set. First |
| 566 | following byte is number of bitmap bytes. Then come bytes |
| 567 | for a bitmap saying which chars are in. Bits in each byte |
| 568 | are ordered low-bit-first. A character is in the set if its |
| 569 | bit is 1. A character too large to have a bit in the map is |
| 570 | automatically not in the set. |
| 571 | |
| 572 | If the length byte has the 0x80 bit set, then that stuff |
| 573 | is followed by a range table: |
| 574 | 2 bytes of flags for character sets (low 8 bits, high 8 bits) |
| 575 | See RANGE_TABLE_WORK_BITS below. |
| 576 | 2 bytes, the number of pairs that follow (upto 32767) |
| 577 | pairs, each 2 multibyte characters, |
| 578 | each multibyte character represented as 3 bytes. */ |
| 579 | charset, |
| 580 | |
| 581 | /* Same parameters as charset, but match any character that is |
| 582 | not one of those specified. */ |
| 583 | charset_not, |
| 584 | |
| 585 | /* Start remembering the text that is matched, for storing in a |
| 586 | register. Followed by one byte with the register number, in |
| 587 | the range 0 to one less than the pattern buffer's re_nsub |
| 588 | field. */ |
| 589 | start_memory, |
| 590 | |
| 591 | /* Stop remembering the text that is matched and store it in a |
| 592 | memory register. Followed by one byte with the register |
| 593 | number, in the range 0 to one less than `re_nsub' in the |
| 594 | pattern buffer. */ |
| 595 | stop_memory, |
| 596 | |
| 597 | /* Match a duplicate of something remembered. Followed by one |
| 598 | byte containing the register number. */ |
| 599 | duplicate, |
| 600 | |
| 601 | /* Fail unless at beginning of line. */ |
| 602 | begline, |
| 603 | |
| 604 | /* Fail unless at end of line. */ |
| 605 | endline, |
| 606 | |
| 607 | /* Succeeds if at beginning of buffer (if emacs) or at beginning |
| 608 | of string to be matched (if not). */ |
| 609 | begbuf, |
| 610 | |
| 611 | /* Analogously, for end of buffer/string. */ |
| 612 | endbuf, |
| 613 | |
| 614 | /* Followed by two byte relative address to which to jump. */ |
| 615 | jump, |
| 616 | |
| 617 | /* Followed by two-byte relative address of place to resume at |
| 618 | in case of failure. */ |
| 619 | on_failure_jump, |
| 620 | |
| 621 | /* Like on_failure_jump, but pushes a placeholder instead of the |
| 622 | current string position when executed. */ |
| 623 | on_failure_keep_string_jump, |
| 624 | |
| 625 | /* Just like `on_failure_jump', except that it checks that we |
| 626 | don't get stuck in an infinite loop (matching an empty string |
| 627 | indefinitely). */ |
| 628 | on_failure_jump_loop, |
| 629 | |
| 630 | /* Just like `on_failure_jump_loop', except that it checks for |
| 631 | a different kind of loop (the kind that shows up with non-greedy |
| 632 | operators). This operation has to be immediately preceded |
| 633 | by a `no_op'. */ |
| 634 | on_failure_jump_nastyloop, |
| 635 | |
| 636 | /* A smart `on_failure_jump' used for greedy * and + operators. |
| 637 | It analyzes the loop before which it is put and if the |
| 638 | loop does not require backtracking, it changes itself to |
| 639 | `on_failure_keep_string_jump' and short-circuits the loop, |
| 640 | else it just defaults to changing itself into `on_failure_jump'. |
| 641 | It assumes that it is pointing to just past a `jump'. */ |
| 642 | on_failure_jump_smart, |
| 643 | |
| 644 | /* Followed by two-byte relative address and two-byte number n. |
| 645 | After matching N times, jump to the address upon failure. |
| 646 | Does not work if N starts at 0: use on_failure_jump_loop |
| 647 | instead. */ |
| 648 | succeed_n, |
| 649 | |
| 650 | /* Followed by two-byte relative address, and two-byte number n. |
| 651 | Jump to the address N times, then fail. */ |
| 652 | jump_n, |
| 653 | |
| 654 | /* Set the following two-byte relative address to the |
| 655 | subsequent two-byte number. The address *includes* the two |
| 656 | bytes of number. */ |
| 657 | set_number_at, |
| 658 | |
| 659 | wordbeg, /* Succeeds if at word beginning. */ |
| 660 | wordend, /* Succeeds if at word end. */ |
| 661 | |
| 662 | wordbound, /* Succeeds if at a word boundary. */ |
| 663 | notwordbound, /* Succeeds if not at a word boundary. */ |
| 664 | |
| 665 | symbeg, /* Succeeds if at symbol beginning. */ |
| 666 | symend, /* Succeeds if at symbol end. */ |
| 667 | |
| 668 | /* Matches any character whose syntax is specified. Followed by |
| 669 | a byte which contains a syntax code, e.g., Sword. */ |
| 670 | syntaxspec, |
| 671 | |
| 672 | /* Matches any character whose syntax is not that specified. */ |
| 673 | notsyntaxspec |
| 674 | |
| 675 | #ifdef emacs |
| 676 | ,before_dot, /* Succeeds if before point. */ |
| 677 | at_dot, /* Succeeds if at point. */ |
| 678 | after_dot, /* Succeeds if after point. */ |
| 679 | |
| 680 | /* Matches any character whose category-set contains the specified |
| 681 | category. The operator is followed by a byte which contains a |
| 682 | category code (mnemonic ASCII character). */ |
| 683 | categoryspec, |
| 684 | |
| 685 | /* Matches any character whose category-set does not contain the |
| 686 | specified category. The operator is followed by a byte which |
| 687 | contains the category code (mnemonic ASCII character). */ |
| 688 | notcategoryspec |
| 689 | #endif /* emacs */ |
| 690 | } re_opcode_t; |
| 691 | \f |
| 692 | /* Common operations on the compiled pattern. */ |
| 693 | |
| 694 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ |
| 695 | |
| 696 | #define STORE_NUMBER(destination, number) \ |
| 697 | do { \ |
| 698 | (destination)[0] = (number) & 0377; \ |
| 699 | (destination)[1] = (number) >> 8; \ |
| 700 | } while (0) |
| 701 | |
| 702 | /* Same as STORE_NUMBER, except increment DESTINATION to |
| 703 | the byte after where the number is stored. Therefore, DESTINATION |
| 704 | must be an lvalue. */ |
| 705 | |
| 706 | #define STORE_NUMBER_AND_INCR(destination, number) \ |
| 707 | do { \ |
| 708 | STORE_NUMBER (destination, number); \ |
| 709 | (destination) += 2; \ |
| 710 | } while (0) |
| 711 | |
| 712 | /* Put into DESTINATION a number stored in two contiguous bytes starting |
| 713 | at SOURCE. */ |
| 714 | |
| 715 | #define EXTRACT_NUMBER(destination, source) \ |
| 716 | ((destination) = extract_number (source)) |
| 717 | |
| 718 | static int |
| 719 | extract_number (re_char *source) |
| 720 | { |
| 721 | return (SIGN_EXTEND_CHAR (source[1]) << 8) + source[0]; |
| 722 | } |
| 723 | |
| 724 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. |
| 725 | SOURCE must be an lvalue. */ |
| 726 | |
| 727 | #define EXTRACT_NUMBER_AND_INCR(destination, source) \ |
| 728 | ((destination) = extract_number_and_incr (&source)) |
| 729 | |
| 730 | static int |
| 731 | extract_number_and_incr (re_char **source) |
| 732 | { |
| 733 | int num = extract_number (*source); |
| 734 | *source += 2; |
| 735 | return num; |
| 736 | } |
| 737 | \f |
| 738 | /* Store a multibyte character in three contiguous bytes starting |
| 739 | DESTINATION, and increment DESTINATION to the byte after where the |
| 740 | character is stored. Therefore, DESTINATION must be an lvalue. */ |
| 741 | |
| 742 | #define STORE_CHARACTER_AND_INCR(destination, character) \ |
| 743 | do { \ |
| 744 | (destination)[0] = (character) & 0377; \ |
| 745 | (destination)[1] = ((character) >> 8) & 0377; \ |
| 746 | (destination)[2] = (character) >> 16; \ |
| 747 | (destination) += 3; \ |
| 748 | } while (0) |
| 749 | |
| 750 | /* Put into DESTINATION a character stored in three contiguous bytes |
| 751 | starting at SOURCE. */ |
| 752 | |
| 753 | #define EXTRACT_CHARACTER(destination, source) \ |
| 754 | do { \ |
| 755 | (destination) = ((source)[0] \ |
| 756 | | ((source)[1] << 8) \ |
| 757 | | ((source)[2] << 16)); \ |
| 758 | } while (0) |
| 759 | |
| 760 | |
| 761 | /* Macros for charset. */ |
| 762 | |
| 763 | /* Size of bitmap of charset P in bytes. P is a start of charset, |
| 764 | i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */ |
| 765 | #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F) |
| 766 | |
| 767 | /* Nonzero if charset P has range table. */ |
| 768 | #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80) |
| 769 | |
| 770 | /* Return the address of range table of charset P. But not the start |
| 771 | of table itself, but the before where the number of ranges is |
| 772 | stored. `2 +' means to skip re_opcode_t and size of bitmap, |
| 773 | and the 2 bytes of flags at the start of the range table. */ |
| 774 | #define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)]) |
| 775 | |
| 776 | /* Extract the bit flags that start a range table. */ |
| 777 | #define CHARSET_RANGE_TABLE_BITS(p) \ |
| 778 | ((p)[2 + CHARSET_BITMAP_SIZE (p)] \ |
| 779 | + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100) |
| 780 | |
| 781 | /* Return the address of end of RANGE_TABLE. COUNT is number of |
| 782 | ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2' |
| 783 | is start of range and end of range. `* 3' is size of each start |
| 784 | and end. */ |
| 785 | #define CHARSET_RANGE_TABLE_END(range_table, count) \ |
| 786 | ((range_table) + (count) * 2 * 3) |
| 787 | |
| 788 | /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in. |
| 789 | COUNT is number of ranges in RANGE_TABLE. */ |
| 790 | #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \ |
| 791 | do \ |
| 792 | { \ |
| 793 | re_wchar_t range_start, range_end; \ |
| 794 | re_char *rtp; \ |
| 795 | re_char *range_table_end \ |
| 796 | = CHARSET_RANGE_TABLE_END ((range_table), (count)); \ |
| 797 | \ |
| 798 | for (rtp = (range_table); rtp < range_table_end; rtp += 2 * 3) \ |
| 799 | { \ |
| 800 | EXTRACT_CHARACTER (range_start, rtp); \ |
| 801 | EXTRACT_CHARACTER (range_end, rtp + 3); \ |
| 802 | \ |
| 803 | if (range_start <= (c) && (c) <= range_end) \ |
| 804 | { \ |
| 805 | (not) = !(not); \ |
| 806 | break; \ |
| 807 | } \ |
| 808 | } \ |
| 809 | } \ |
| 810 | while (0) |
| 811 | |
| 812 | /* Test if C is in range table of CHARSET. The flag NOT is negated if |
| 813 | C is listed in it. */ |
| 814 | #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \ |
| 815 | do \ |
| 816 | { \ |
| 817 | /* Number of ranges in range table. */ \ |
| 818 | int count; \ |
| 819 | re_char *range_table = CHARSET_RANGE_TABLE (charset); \ |
| 820 | \ |
| 821 | EXTRACT_NUMBER_AND_INCR (count, range_table); \ |
| 822 | CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \ |
| 823 | } \ |
| 824 | while (0) |
| 825 | \f |
| 826 | /* If DEBUG is defined, Regex prints many voluminous messages about what |
| 827 | it is doing (if the variable `debug' is nonzero). If linked with the |
| 828 | main program in `iregex.c', you can enter patterns and strings |
| 829 | interactively. And if linked with the main program in `main.c' and |
| 830 | the other test files, you can run the already-written tests. */ |
| 831 | |
| 832 | #ifdef DEBUG |
| 833 | |
| 834 | /* We use standard I/O for debugging. */ |
| 835 | # include <stdio.h> |
| 836 | |
| 837 | /* It is useful to test things that ``must'' be true when debugging. */ |
| 838 | # include <assert.h> |
| 839 | |
| 840 | static int debug = -100000; |
| 841 | |
| 842 | # define DEBUG_STATEMENT(e) e |
| 843 | # define DEBUG_PRINT(...) if (debug > 0) printf (__VA_ARGS__) |
| 844 | # define DEBUG_COMPILES_ARGUMENTS |
| 845 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
| 846 | if (debug > 0) print_partial_compiled_pattern (s, e) |
| 847 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
| 848 | if (debug > 0) print_double_string (w, s1, sz1, s2, sz2) |
| 849 | |
| 850 | |
| 851 | /* Print the fastmap in human-readable form. */ |
| 852 | |
| 853 | static void |
| 854 | print_fastmap (char *fastmap) |
| 855 | { |
| 856 | unsigned was_a_range = 0; |
| 857 | unsigned i = 0; |
| 858 | |
| 859 | while (i < (1 << BYTEWIDTH)) |
| 860 | { |
| 861 | if (fastmap[i++]) |
| 862 | { |
| 863 | was_a_range = 0; |
| 864 | putchar (i - 1); |
| 865 | while (i < (1 << BYTEWIDTH) && fastmap[i]) |
| 866 | { |
| 867 | was_a_range = 1; |
| 868 | i++; |
| 869 | } |
| 870 | if (was_a_range) |
| 871 | { |
| 872 | printf ("-"); |
| 873 | putchar (i - 1); |
| 874 | } |
| 875 | } |
| 876 | } |
| 877 | putchar ('\n'); |
| 878 | } |
| 879 | |
| 880 | |
| 881 | /* Print a compiled pattern string in human-readable form, starting at |
| 882 | the START pointer into it and ending just before the pointer END. */ |
| 883 | |
| 884 | static void |
| 885 | print_partial_compiled_pattern (re_char *start, re_char *end) |
| 886 | { |
| 887 | int mcnt, mcnt2; |
| 888 | re_char *p = start; |
| 889 | re_char *pend = end; |
| 890 | |
| 891 | if (start == NULL) |
| 892 | { |
| 893 | fprintf (stderr, "(null)\n"); |
| 894 | return; |
| 895 | } |
| 896 | |
| 897 | /* Loop over pattern commands. */ |
| 898 | while (p < pend) |
| 899 | { |
| 900 | fprintf (stderr, "%td:\t", p - start); |
| 901 | |
| 902 | switch ((re_opcode_t) *p++) |
| 903 | { |
| 904 | case no_op: |
| 905 | fprintf (stderr, "/no_op"); |
| 906 | break; |
| 907 | |
| 908 | case succeed: |
| 909 | fprintf (stderr, "/succeed"); |
| 910 | break; |
| 911 | |
| 912 | case exactn: |
| 913 | mcnt = *p++; |
| 914 | fprintf (stderr, "/exactn/%d", mcnt); |
| 915 | do |
| 916 | { |
| 917 | fprintf (stderr, "/%c", *p++); |
| 918 | } |
| 919 | while (--mcnt); |
| 920 | break; |
| 921 | |
| 922 | case start_memory: |
| 923 | fprintf (stderr, "/start_memory/%d", *p++); |
| 924 | break; |
| 925 | |
| 926 | case stop_memory: |
| 927 | fprintf (stderr, "/stop_memory/%d", *p++); |
| 928 | break; |
| 929 | |
| 930 | case duplicate: |
| 931 | fprintf (stderr, "/duplicate/%d", *p++); |
| 932 | break; |
| 933 | |
| 934 | case anychar: |
| 935 | fprintf (stderr, "/anychar"); |
| 936 | break; |
| 937 | |
| 938 | case charset: |
| 939 | case charset_not: |
| 940 | { |
| 941 | register int c, last = -100; |
| 942 | register int in_range = 0; |
| 943 | int length = CHARSET_BITMAP_SIZE (p - 1); |
| 944 | int has_range_table = CHARSET_RANGE_TABLE_EXISTS_P (p - 1); |
| 945 | |
| 946 | fprintf (stderr, "/charset [%s", |
| 947 | (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); |
| 948 | |
| 949 | if (p + *p >= pend) |
| 950 | fprintf (stderr, " !extends past end of pattern! "); |
| 951 | |
| 952 | for (c = 0; c < 256; c++) |
| 953 | if (c / 8 < length |
| 954 | && (p[1 + (c/8)] & (1 << (c % 8)))) |
| 955 | { |
| 956 | /* Are we starting a range? */ |
| 957 | if (last + 1 == c && ! in_range) |
| 958 | { |
| 959 | fprintf (stderr, "-"); |
| 960 | in_range = 1; |
| 961 | } |
| 962 | /* Have we broken a range? */ |
| 963 | else if (last + 1 != c && in_range) |
| 964 | { |
| 965 | fprintf (stderr, "%c", last); |
| 966 | in_range = 0; |
| 967 | } |
| 968 | |
| 969 | if (! in_range) |
| 970 | fprintf (stderr, "%c", c); |
| 971 | |
| 972 | last = c; |
| 973 | } |
| 974 | |
| 975 | if (in_range) |
| 976 | fprintf (stderr, "%c", last); |
| 977 | |
| 978 | fprintf (stderr, "]"); |
| 979 | |
| 980 | p += 1 + length; |
| 981 | |
| 982 | if (has_range_table) |
| 983 | { |
| 984 | int count; |
| 985 | fprintf (stderr, "has-range-table"); |
| 986 | |
| 987 | /* ??? Should print the range table; for now, just skip it. */ |
| 988 | p += 2; /* skip range table bits */ |
| 989 | EXTRACT_NUMBER_AND_INCR (count, p); |
| 990 | p = CHARSET_RANGE_TABLE_END (p, count); |
| 991 | } |
| 992 | } |
| 993 | break; |
| 994 | |
| 995 | case begline: |
| 996 | fprintf (stderr, "/begline"); |
| 997 | break; |
| 998 | |
| 999 | case endline: |
| 1000 | fprintf (stderr, "/endline"); |
| 1001 | break; |
| 1002 | |
| 1003 | case on_failure_jump: |
| 1004 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 1005 | fprintf (stderr, "/on_failure_jump to %td", p + mcnt - start); |
| 1006 | break; |
| 1007 | |
| 1008 | case on_failure_keep_string_jump: |
| 1009 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 1010 | fprintf (stderr, "/on_failure_keep_string_jump to %td", |
| 1011 | p + mcnt - start); |
| 1012 | break; |
| 1013 | |
| 1014 | case on_failure_jump_nastyloop: |
| 1015 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 1016 | fprintf (stderr, "/on_failure_jump_nastyloop to %td", |
| 1017 | p + mcnt - start); |
| 1018 | break; |
| 1019 | |
| 1020 | case on_failure_jump_loop: |
| 1021 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 1022 | fprintf (stderr, "/on_failure_jump_loop to %td", |
| 1023 | p + mcnt - start); |
| 1024 | break; |
| 1025 | |
| 1026 | case on_failure_jump_smart: |
| 1027 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 1028 | fprintf (stderr, "/on_failure_jump_smart to %td", |
| 1029 | p + mcnt - start); |
| 1030 | break; |
| 1031 | |
| 1032 | case jump: |
| 1033 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 1034 | fprintf (stderr, "/jump to %td", p + mcnt - start); |
| 1035 | break; |
| 1036 | |
| 1037 | case succeed_n: |
| 1038 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 1039 | EXTRACT_NUMBER_AND_INCR (mcnt2, p); |
| 1040 | fprintf (stderr, "/succeed_n to %td, %d times", |
| 1041 | p - 2 + mcnt - start, mcnt2); |
| 1042 | break; |
| 1043 | |
| 1044 | case jump_n: |
| 1045 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 1046 | EXTRACT_NUMBER_AND_INCR (mcnt2, p); |
| 1047 | fprintf (stderr, "/jump_n to %td, %d times", |
| 1048 | p - 2 + mcnt - start, mcnt2); |
| 1049 | break; |
| 1050 | |
| 1051 | case set_number_at: |
| 1052 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 1053 | EXTRACT_NUMBER_AND_INCR (mcnt2, p); |
| 1054 | fprintf (stderr, "/set_number_at location %td to %d", |
| 1055 | p - 2 + mcnt - start, mcnt2); |
| 1056 | break; |
| 1057 | |
| 1058 | case wordbound: |
| 1059 | fprintf (stderr, "/wordbound"); |
| 1060 | break; |
| 1061 | |
| 1062 | case notwordbound: |
| 1063 | fprintf (stderr, "/notwordbound"); |
| 1064 | break; |
| 1065 | |
| 1066 | case wordbeg: |
| 1067 | fprintf (stderr, "/wordbeg"); |
| 1068 | break; |
| 1069 | |
| 1070 | case wordend: |
| 1071 | fprintf (stderr, "/wordend"); |
| 1072 | break; |
| 1073 | |
| 1074 | case symbeg: |
| 1075 | fprintf (stderr, "/symbeg"); |
| 1076 | break; |
| 1077 | |
| 1078 | case symend: |
| 1079 | fprintf (stderr, "/symend"); |
| 1080 | break; |
| 1081 | |
| 1082 | case syntaxspec: |
| 1083 | fprintf (stderr, "/syntaxspec"); |
| 1084 | mcnt = *p++; |
| 1085 | fprintf (stderr, "/%d", mcnt); |
| 1086 | break; |
| 1087 | |
| 1088 | case notsyntaxspec: |
| 1089 | fprintf (stderr, "/notsyntaxspec"); |
| 1090 | mcnt = *p++; |
| 1091 | fprintf (stderr, "/%d", mcnt); |
| 1092 | break; |
| 1093 | |
| 1094 | # ifdef emacs |
| 1095 | case before_dot: |
| 1096 | fprintf (stderr, "/before_dot"); |
| 1097 | break; |
| 1098 | |
| 1099 | case at_dot: |
| 1100 | fprintf (stderr, "/at_dot"); |
| 1101 | break; |
| 1102 | |
| 1103 | case after_dot: |
| 1104 | fprintf (stderr, "/after_dot"); |
| 1105 | break; |
| 1106 | |
| 1107 | case categoryspec: |
| 1108 | fprintf (stderr, "/categoryspec"); |
| 1109 | mcnt = *p++; |
| 1110 | fprintf (stderr, "/%d", mcnt); |
| 1111 | break; |
| 1112 | |
| 1113 | case notcategoryspec: |
| 1114 | fprintf (stderr, "/notcategoryspec"); |
| 1115 | mcnt = *p++; |
| 1116 | fprintf (stderr, "/%d", mcnt); |
| 1117 | break; |
| 1118 | # endif /* emacs */ |
| 1119 | |
| 1120 | case begbuf: |
| 1121 | fprintf (stderr, "/begbuf"); |
| 1122 | break; |
| 1123 | |
| 1124 | case endbuf: |
| 1125 | fprintf (stderr, "/endbuf"); |
| 1126 | break; |
| 1127 | |
| 1128 | default: |
| 1129 | fprintf (stderr, "?%d", *(p-1)); |
| 1130 | } |
| 1131 | |
| 1132 | fprintf (stderr, "\n"); |
| 1133 | } |
| 1134 | |
| 1135 | fprintf (stderr, "%td:\tend of pattern.\n", p - start); |
| 1136 | } |
| 1137 | |
| 1138 | |
| 1139 | static void |
| 1140 | print_compiled_pattern (struct re_pattern_buffer *bufp) |
| 1141 | { |
| 1142 | re_char *buffer = bufp->buffer; |
| 1143 | |
| 1144 | print_partial_compiled_pattern (buffer, buffer + bufp->used); |
| 1145 | printf ("%ld bytes used/%ld bytes allocated.\n", |
| 1146 | bufp->used, bufp->allocated); |
| 1147 | |
| 1148 | if (bufp->fastmap_accurate && bufp->fastmap) |
| 1149 | { |
| 1150 | printf ("fastmap: "); |
| 1151 | print_fastmap (bufp->fastmap); |
| 1152 | } |
| 1153 | |
| 1154 | printf ("re_nsub: %zu\t", bufp->re_nsub); |
| 1155 | printf ("regs_alloc: %d\t", bufp->regs_allocated); |
| 1156 | printf ("can_be_null: %d\t", bufp->can_be_null); |
| 1157 | printf ("no_sub: %d\t", bufp->no_sub); |
| 1158 | printf ("not_bol: %d\t", bufp->not_bol); |
| 1159 | printf ("not_eol: %d\t", bufp->not_eol); |
| 1160 | printf ("syntax: %lx\n", bufp->syntax); |
| 1161 | fflush (stdout); |
| 1162 | /* Perhaps we should print the translate table? */ |
| 1163 | } |
| 1164 | |
| 1165 | |
| 1166 | static void |
| 1167 | print_double_string (re_char *where, re_char *string1, ssize_t size1, |
| 1168 | re_char *string2, ssize_t size2) |
| 1169 | { |
| 1170 | ssize_t this_char; |
| 1171 | |
| 1172 | if (where == NULL) |
| 1173 | printf ("(null)"); |
| 1174 | else |
| 1175 | { |
| 1176 | if (FIRST_STRING_P (where)) |
| 1177 | { |
| 1178 | for (this_char = where - string1; this_char < size1; this_char++) |
| 1179 | putchar (string1[this_char]); |
| 1180 | |
| 1181 | where = string2; |
| 1182 | } |
| 1183 | |
| 1184 | for (this_char = where - string2; this_char < size2; this_char++) |
| 1185 | putchar (string2[this_char]); |
| 1186 | } |
| 1187 | } |
| 1188 | |
| 1189 | #else /* not DEBUG */ |
| 1190 | |
| 1191 | # undef assert |
| 1192 | # define assert(e) |
| 1193 | |
| 1194 | # define DEBUG_STATEMENT(e) |
| 1195 | # if __STDC_VERSION__ < 199901L |
| 1196 | # define DEBUG_COMPILES_ARGUMENTS |
| 1197 | # define DEBUG_PRINT /* 'DEBUG_PRINT (x, y)' discards X and Y. */ (void) |
| 1198 | # else |
| 1199 | # define DEBUG_PRINT(...) |
| 1200 | # endif |
| 1201 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) |
| 1202 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) |
| 1203 | |
| 1204 | #endif /* not DEBUG */ |
| 1205 | \f |
| 1206 | /* Use this to suppress gcc's `...may be used before initialized' warnings. */ |
| 1207 | #ifdef lint |
| 1208 | # define IF_LINT(Code) Code |
| 1209 | #else |
| 1210 | # define IF_LINT(Code) /* empty */ |
| 1211 | #endif |
| 1212 | \f |
| 1213 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can |
| 1214 | also be assigned to arbitrarily: each pattern buffer stores its own |
| 1215 | syntax, so it can be changed between regex compilations. */ |
| 1216 | /* This has no initializer because initialized variables in Emacs |
| 1217 | become read-only after dumping. */ |
| 1218 | reg_syntax_t re_syntax_options; |
| 1219 | |
| 1220 | |
| 1221 | /* Specify the precise syntax of regexps for compilation. This provides |
| 1222 | for compatibility for various utilities which historically have |
| 1223 | different, incompatible syntaxes. |
| 1224 | |
| 1225 | The argument SYNTAX is a bit mask comprised of the various bits |
| 1226 | defined in regex.h. We return the old syntax. */ |
| 1227 | |
| 1228 | reg_syntax_t |
| 1229 | re_set_syntax (reg_syntax_t syntax) |
| 1230 | { |
| 1231 | reg_syntax_t ret = re_syntax_options; |
| 1232 | |
| 1233 | re_syntax_options = syntax; |
| 1234 | return ret; |
| 1235 | } |
| 1236 | WEAK_ALIAS (__re_set_syntax, re_set_syntax) |
| 1237 | |
| 1238 | /* Regexp to use to replace spaces, or NULL meaning don't. */ |
| 1239 | static re_char *whitespace_regexp; |
| 1240 | |
| 1241 | void |
| 1242 | re_set_whitespace_regexp (const char *regexp) |
| 1243 | { |
| 1244 | whitespace_regexp = (re_char *) regexp; |
| 1245 | } |
| 1246 | WEAK_ALIAS (__re_set_syntax, re_set_syntax) |
| 1247 | \f |
| 1248 | /* This table gives an error message for each of the error codes listed |
| 1249 | in regex.h. Obviously the order here has to be same as there. |
| 1250 | POSIX doesn't require that we do anything for REG_NOERROR, |
| 1251 | but why not be nice? */ |
| 1252 | |
| 1253 | static const char *re_error_msgid[] = |
| 1254 | { |
| 1255 | gettext_noop ("Success"), /* REG_NOERROR */ |
| 1256 | gettext_noop ("No match"), /* REG_NOMATCH */ |
| 1257 | gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ |
| 1258 | gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ |
| 1259 | gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ |
| 1260 | gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ |
| 1261 | gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ |
| 1262 | gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ |
| 1263 | gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ |
| 1264 | gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ |
| 1265 | gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ |
| 1266 | gettext_noop ("Invalid range end"), /* REG_ERANGE */ |
| 1267 | gettext_noop ("Memory exhausted"), /* REG_ESPACE */ |
| 1268 | gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ |
| 1269 | gettext_noop ("Premature end of regular expression"), /* REG_EEND */ |
| 1270 | gettext_noop ("Regular expression too big"), /* REG_ESIZE */ |
| 1271 | gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */ |
| 1272 | gettext_noop ("Range striding over charsets") /* REG_ERANGEX */ |
| 1273 | }; |
| 1274 | \f |
| 1275 | /* Avoiding alloca during matching, to placate r_alloc. */ |
| 1276 | |
| 1277 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the |
| 1278 | searching and matching functions should not call alloca. On some |
| 1279 | systems, alloca is implemented in terms of malloc, and if we're |
| 1280 | using the relocating allocator routines, then malloc could cause a |
| 1281 | relocation, which might (if the strings being searched are in the |
| 1282 | ralloc heap) shift the data out from underneath the regexp |
| 1283 | routines. |
| 1284 | |
| 1285 | Here's another reason to avoid allocation: Emacs |
| 1286 | processes input from X in a signal handler; processing X input may |
| 1287 | call malloc; if input arrives while a matching routine is calling |
| 1288 | malloc, then we're scrod. But Emacs can't just block input while |
| 1289 | calling matching routines; then we don't notice interrupts when |
| 1290 | they come in. So, Emacs blocks input around all regexp calls |
| 1291 | except the matching calls, which it leaves unprotected, in the |
| 1292 | faith that they will not malloc. */ |
| 1293 | |
| 1294 | /* Normally, this is fine. */ |
| 1295 | #define MATCH_MAY_ALLOCATE |
| 1296 | |
| 1297 | /* The match routines may not allocate if (1) they would do it with malloc |
| 1298 | and (2) it's not safe for them to use malloc. |
| 1299 | Note that if REL_ALLOC is defined, matching would not use malloc for the |
| 1300 | failure stack, but we would still use it for the register vectors; |
| 1301 | so REL_ALLOC should not affect this. */ |
| 1302 | #if defined REGEX_MALLOC && defined emacs |
| 1303 | # undef MATCH_MAY_ALLOCATE |
| 1304 | #endif |
| 1305 | |
| 1306 | \f |
| 1307 | /* Failure stack declarations and macros; both re_compile_fastmap and |
| 1308 | re_match_2 use a failure stack. These have to be macros because of |
| 1309 | REGEX_ALLOCATE_STACK. */ |
| 1310 | |
| 1311 | |
| 1312 | /* Approximate number of failure points for which to initially allocate space |
| 1313 | when matching. If this number is exceeded, we allocate more |
| 1314 | space, so it is not a hard limit. */ |
| 1315 | #ifndef INIT_FAILURE_ALLOC |
| 1316 | # define INIT_FAILURE_ALLOC 20 |
| 1317 | #endif |
| 1318 | |
| 1319 | /* Roughly the maximum number of failure points on the stack. Would be |
| 1320 | exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed. |
| 1321 | This is a variable only so users of regex can assign to it; we never |
| 1322 | change it ourselves. We always multiply it by TYPICAL_FAILURE_SIZE |
| 1323 | before using it, so it should probably be a byte-count instead. */ |
| 1324 | # if defined MATCH_MAY_ALLOCATE |
| 1325 | /* Note that 4400 was enough to cause a crash on Alpha OSF/1, |
| 1326 | whose default stack limit is 2mb. In order for a larger |
| 1327 | value to work reliably, you have to try to make it accord |
| 1328 | with the process stack limit. */ |
| 1329 | size_t re_max_failures = 40000; |
| 1330 | # else |
| 1331 | size_t re_max_failures = 4000; |
| 1332 | # endif |
| 1333 | |
| 1334 | union fail_stack_elt |
| 1335 | { |
| 1336 | re_char *pointer; |
| 1337 | /* This should be the biggest `int' that's no bigger than a pointer. */ |
| 1338 | long integer; |
| 1339 | }; |
| 1340 | |
| 1341 | typedef union fail_stack_elt fail_stack_elt_t; |
| 1342 | |
| 1343 | typedef struct |
| 1344 | { |
| 1345 | fail_stack_elt_t *stack; |
| 1346 | size_t size; |
| 1347 | size_t avail; /* Offset of next open position. */ |
| 1348 | size_t frame; /* Offset of the cur constructed frame. */ |
| 1349 | } fail_stack_type; |
| 1350 | |
| 1351 | #define FAIL_STACK_EMPTY() (fail_stack.frame == 0) |
| 1352 | |
| 1353 | |
| 1354 | /* Define macros to initialize and free the failure stack. |
| 1355 | Do `return -2' if the alloc fails. */ |
| 1356 | |
| 1357 | #ifdef MATCH_MAY_ALLOCATE |
| 1358 | # define INIT_FAIL_STACK() \ |
| 1359 | do { \ |
| 1360 | fail_stack.stack = \ |
| 1361 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \ |
| 1362 | * sizeof (fail_stack_elt_t)); \ |
| 1363 | \ |
| 1364 | if (fail_stack.stack == NULL) \ |
| 1365 | return -2; \ |
| 1366 | \ |
| 1367 | fail_stack.size = INIT_FAILURE_ALLOC; \ |
| 1368 | fail_stack.avail = 0; \ |
| 1369 | fail_stack.frame = 0; \ |
| 1370 | } while (0) |
| 1371 | #else |
| 1372 | # define INIT_FAIL_STACK() \ |
| 1373 | do { \ |
| 1374 | fail_stack.avail = 0; \ |
| 1375 | fail_stack.frame = 0; \ |
| 1376 | } while (0) |
| 1377 | |
| 1378 | # define RETALLOC_IF(addr, n, t) \ |
| 1379 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) |
| 1380 | #endif |
| 1381 | |
| 1382 | |
| 1383 | /* Double the size of FAIL_STACK, up to a limit |
| 1384 | which allows approximately `re_max_failures' items. |
| 1385 | |
| 1386 | Return 1 if succeeds, and 0 if either ran out of memory |
| 1387 | allocating space for it or it was already too large. |
| 1388 | |
| 1389 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
| 1390 | |
| 1391 | /* Factor to increase the failure stack size by |
| 1392 | when we increase it. |
| 1393 | This used to be 2, but 2 was too wasteful |
| 1394 | because the old discarded stacks added up to as much space |
| 1395 | were as ultimate, maximum-size stack. */ |
| 1396 | #define FAIL_STACK_GROWTH_FACTOR 4 |
| 1397 | |
| 1398 | #define GROW_FAIL_STACK(fail_stack) \ |
| 1399 | (((fail_stack).size * sizeof (fail_stack_elt_t) \ |
| 1400 | >= re_max_failures * TYPICAL_FAILURE_SIZE) \ |
| 1401 | ? 0 \ |
| 1402 | : ((fail_stack).stack \ |
| 1403 | = REGEX_REALLOCATE_STACK ((fail_stack).stack, \ |
| 1404 | (fail_stack).size * sizeof (fail_stack_elt_t), \ |
| 1405 | MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ |
| 1406 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ |
| 1407 | * FAIL_STACK_GROWTH_FACTOR))), \ |
| 1408 | \ |
| 1409 | (fail_stack).stack == NULL \ |
| 1410 | ? 0 \ |
| 1411 | : ((fail_stack).size \ |
| 1412 | = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \ |
| 1413 | ((fail_stack).size * sizeof (fail_stack_elt_t) \ |
| 1414 | * FAIL_STACK_GROWTH_FACTOR)) \ |
| 1415 | / sizeof (fail_stack_elt_t)), \ |
| 1416 | 1))) |
| 1417 | |
| 1418 | |
| 1419 | /* Push a pointer value onto the failure stack. |
| 1420 | Assumes the variable `fail_stack'. Probably should only |
| 1421 | be called from within `PUSH_FAILURE_POINT'. */ |
| 1422 | #define PUSH_FAILURE_POINTER(item) \ |
| 1423 | fail_stack.stack[fail_stack.avail++].pointer = (item) |
| 1424 | |
| 1425 | /* This pushes an integer-valued item onto the failure stack. |
| 1426 | Assumes the variable `fail_stack'. Probably should only |
| 1427 | be called from within `PUSH_FAILURE_POINT'. */ |
| 1428 | #define PUSH_FAILURE_INT(item) \ |
| 1429 | fail_stack.stack[fail_stack.avail++].integer = (item) |
| 1430 | |
| 1431 | /* These POP... operations complement the PUSH... operations. |
| 1432 | All assume that `fail_stack' is nonempty. */ |
| 1433 | #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer |
| 1434 | #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer |
| 1435 | |
| 1436 | /* Individual items aside from the registers. */ |
| 1437 | #define NUM_NONREG_ITEMS 3 |
| 1438 | |
| 1439 | /* Used to examine the stack (to detect infinite loops). */ |
| 1440 | #define FAILURE_PAT(h) fail_stack.stack[(h) - 1].pointer |
| 1441 | #define FAILURE_STR(h) (fail_stack.stack[(h) - 2].pointer) |
| 1442 | #define NEXT_FAILURE_HANDLE(h) fail_stack.stack[(h) - 3].integer |
| 1443 | #define TOP_FAILURE_HANDLE() fail_stack.frame |
| 1444 | |
| 1445 | |
| 1446 | #define ENSURE_FAIL_STACK(space) \ |
| 1447 | while (REMAINING_AVAIL_SLOTS <= space) { \ |
| 1448 | if (!GROW_FAIL_STACK (fail_stack)) \ |
| 1449 | return -2; \ |
| 1450 | DEBUG_PRINT ("\n Doubled stack; size now: %zd\n", (fail_stack).size);\ |
| 1451 | DEBUG_PRINT (" slots available: %zd\n", REMAINING_AVAIL_SLOTS);\ |
| 1452 | } |
| 1453 | |
| 1454 | /* Push register NUM onto the stack. */ |
| 1455 | #define PUSH_FAILURE_REG(num) \ |
| 1456 | do { \ |
| 1457 | char *destination; \ |
| 1458 | long n = num; \ |
| 1459 | ENSURE_FAIL_STACK(3); \ |
| 1460 | DEBUG_PRINT (" Push reg %ld (spanning %p -> %p)\n", \ |
| 1461 | n, regstart[n], regend[n]); \ |
| 1462 | PUSH_FAILURE_POINTER (regstart[n]); \ |
| 1463 | PUSH_FAILURE_POINTER (regend[n]); \ |
| 1464 | PUSH_FAILURE_INT (n); \ |
| 1465 | } while (0) |
| 1466 | |
| 1467 | /* Change the counter's value to VAL, but make sure that it will |
| 1468 | be reset when backtracking. */ |
| 1469 | #define PUSH_NUMBER(ptr,val) \ |
| 1470 | do { \ |
| 1471 | char *destination; \ |
| 1472 | int c; \ |
| 1473 | ENSURE_FAIL_STACK(3); \ |
| 1474 | EXTRACT_NUMBER (c, ptr); \ |
| 1475 | DEBUG_PRINT (" Push number %p = %d -> %d\n", ptr, c, val); \ |
| 1476 | PUSH_FAILURE_INT (c); \ |
| 1477 | PUSH_FAILURE_POINTER (ptr); \ |
| 1478 | PUSH_FAILURE_INT (-1); \ |
| 1479 | STORE_NUMBER (ptr, val); \ |
| 1480 | } while (0) |
| 1481 | |
| 1482 | /* Pop a saved register off the stack. */ |
| 1483 | #define POP_FAILURE_REG_OR_COUNT() \ |
| 1484 | do { \ |
| 1485 | long pfreg = POP_FAILURE_INT (); \ |
| 1486 | if (pfreg == -1) \ |
| 1487 | { \ |
| 1488 | /* It's a counter. */ \ |
| 1489 | /* Here, we discard `const', making re_match non-reentrant. */ \ |
| 1490 | unsigned char *ptr = (unsigned char*) POP_FAILURE_POINTER (); \ |
| 1491 | pfreg = POP_FAILURE_INT (); \ |
| 1492 | STORE_NUMBER (ptr, pfreg); \ |
| 1493 | DEBUG_PRINT (" Pop counter %p = %ld\n", ptr, pfreg); \ |
| 1494 | } \ |
| 1495 | else \ |
| 1496 | { \ |
| 1497 | regend[pfreg] = POP_FAILURE_POINTER (); \ |
| 1498 | regstart[pfreg] = POP_FAILURE_POINTER (); \ |
| 1499 | DEBUG_PRINT (" Pop reg %ld (spanning %p -> %p)\n", \ |
| 1500 | pfreg, regstart[pfreg], regend[pfreg]); \ |
| 1501 | } \ |
| 1502 | } while (0) |
| 1503 | |
| 1504 | /* Check that we are not stuck in an infinite loop. */ |
| 1505 | #define CHECK_INFINITE_LOOP(pat_cur, string_place) \ |
| 1506 | do { \ |
| 1507 | ssize_t failure = TOP_FAILURE_HANDLE (); \ |
| 1508 | /* Check for infinite matching loops */ \ |
| 1509 | while (failure > 0 \ |
| 1510 | && (FAILURE_STR (failure) == string_place \ |
| 1511 | || FAILURE_STR (failure) == NULL)) \ |
| 1512 | { \ |
| 1513 | assert (FAILURE_PAT (failure) >= bufp->buffer \ |
| 1514 | && FAILURE_PAT (failure) <= bufp->buffer + bufp->used); \ |
| 1515 | if (FAILURE_PAT (failure) == pat_cur) \ |
| 1516 | { \ |
| 1517 | cycle = 1; \ |
| 1518 | break; \ |
| 1519 | } \ |
| 1520 | DEBUG_PRINT (" Other pattern: %p\n", FAILURE_PAT (failure)); \ |
| 1521 | failure = NEXT_FAILURE_HANDLE(failure); \ |
| 1522 | } \ |
| 1523 | DEBUG_PRINT (" Other string: %p\n", FAILURE_STR (failure)); \ |
| 1524 | } while (0) |
| 1525 | |
| 1526 | /* Push the information about the state we will need |
| 1527 | if we ever fail back to it. |
| 1528 | |
| 1529 | Requires variables fail_stack, regstart, regend and |
| 1530 | num_regs be declared. GROW_FAIL_STACK requires `destination' be |
| 1531 | declared. |
| 1532 | |
| 1533 | Does `return FAILURE_CODE' if runs out of memory. */ |
| 1534 | |
| 1535 | #define PUSH_FAILURE_POINT(pattern, string_place) \ |
| 1536 | do { \ |
| 1537 | char *destination; \ |
| 1538 | /* Must be int, so when we don't save any registers, the arithmetic \ |
| 1539 | of 0 + -1 isn't done as unsigned. */ \ |
| 1540 | \ |
| 1541 | DEBUG_STATEMENT (nfailure_points_pushed++); \ |
| 1542 | DEBUG_PRINT ("\nPUSH_FAILURE_POINT:\n"); \ |
| 1543 | DEBUG_PRINT (" Before push, next avail: %zd\n", (fail_stack).avail); \ |
| 1544 | DEBUG_PRINT (" size: %zd\n", (fail_stack).size);\ |
| 1545 | \ |
| 1546 | ENSURE_FAIL_STACK (NUM_NONREG_ITEMS); \ |
| 1547 | \ |
| 1548 | DEBUG_PRINT ("\n"); \ |
| 1549 | \ |
| 1550 | DEBUG_PRINT (" Push frame index: %zd\n", fail_stack.frame); \ |
| 1551 | PUSH_FAILURE_INT (fail_stack.frame); \ |
| 1552 | \ |
| 1553 | DEBUG_PRINT (" Push string %p: `", string_place); \ |
| 1554 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, size2);\ |
| 1555 | DEBUG_PRINT ("'\n"); \ |
| 1556 | PUSH_FAILURE_POINTER (string_place); \ |
| 1557 | \ |
| 1558 | DEBUG_PRINT (" Push pattern %p: ", pattern); \ |
| 1559 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern, pend); \ |
| 1560 | PUSH_FAILURE_POINTER (pattern); \ |
| 1561 | \ |
| 1562 | /* Close the frame by moving the frame pointer past it. */ \ |
| 1563 | fail_stack.frame = fail_stack.avail; \ |
| 1564 | } while (0) |
| 1565 | |
| 1566 | /* Estimate the size of data pushed by a typical failure stack entry. |
| 1567 | An estimate is all we need, because all we use this for |
| 1568 | is to choose a limit for how big to make the failure stack. */ |
| 1569 | /* BEWARE, the value `20' is hard-coded in emacs.c:main(). */ |
| 1570 | #define TYPICAL_FAILURE_SIZE 20 |
| 1571 | |
| 1572 | /* How many items can still be added to the stack without overflowing it. */ |
| 1573 | #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) |
| 1574 | |
| 1575 | |
| 1576 | /* Pops what PUSH_FAIL_STACK pushes. |
| 1577 | |
| 1578 | We restore into the parameters, all of which should be lvalues: |
| 1579 | STR -- the saved data position. |
| 1580 | PAT -- the saved pattern position. |
| 1581 | REGSTART, REGEND -- arrays of string positions. |
| 1582 | |
| 1583 | Also assumes the variables `fail_stack' and (if debugging), `bufp', |
| 1584 | `pend', `string1', `size1', `string2', and `size2'. */ |
| 1585 | |
| 1586 | #define POP_FAILURE_POINT(str, pat) \ |
| 1587 | do { \ |
| 1588 | assert (!FAIL_STACK_EMPTY ()); \ |
| 1589 | \ |
| 1590 | /* Remove failure points and point to how many regs pushed. */ \ |
| 1591 | DEBUG_PRINT ("POP_FAILURE_POINT:\n"); \ |
| 1592 | DEBUG_PRINT (" Before pop, next avail: %zd\n", fail_stack.avail); \ |
| 1593 | DEBUG_PRINT (" size: %zd\n", fail_stack.size); \ |
| 1594 | \ |
| 1595 | /* Pop the saved registers. */ \ |
| 1596 | while (fail_stack.frame < fail_stack.avail) \ |
| 1597 | POP_FAILURE_REG_OR_COUNT (); \ |
| 1598 | \ |
| 1599 | pat = POP_FAILURE_POINTER (); \ |
| 1600 | DEBUG_PRINT (" Popping pattern %p: ", pat); \ |
| 1601 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ |
| 1602 | \ |
| 1603 | /* If the saved string location is NULL, it came from an \ |
| 1604 | on_failure_keep_string_jump opcode, and we want to throw away the \ |
| 1605 | saved NULL, thus retaining our current position in the string. */ \ |
| 1606 | str = POP_FAILURE_POINTER (); \ |
| 1607 | DEBUG_PRINT (" Popping string %p: `", str); \ |
| 1608 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
| 1609 | DEBUG_PRINT ("'\n"); \ |
| 1610 | \ |
| 1611 | fail_stack.frame = POP_FAILURE_INT (); \ |
| 1612 | DEBUG_PRINT (" Popping frame index: %zd\n", fail_stack.frame); \ |
| 1613 | \ |
| 1614 | assert (fail_stack.avail >= 0); \ |
| 1615 | assert (fail_stack.frame <= fail_stack.avail); \ |
| 1616 | \ |
| 1617 | DEBUG_STATEMENT (nfailure_points_popped++); \ |
| 1618 | } while (0) /* POP_FAILURE_POINT */ |
| 1619 | |
| 1620 | |
| 1621 | \f |
| 1622 | /* Registers are set to a sentinel when they haven't yet matched. */ |
| 1623 | #define REG_UNSET(e) ((e) == NULL) |
| 1624 | \f |
| 1625 | /* Subroutine declarations and macros for regex_compile. */ |
| 1626 | |
| 1627 | static reg_errcode_t regex_compile (re_char *pattern, size_t size, |
| 1628 | reg_syntax_t syntax, |
| 1629 | struct re_pattern_buffer *bufp); |
| 1630 | static void store_op1 (re_opcode_t op, unsigned char *loc, int arg); |
| 1631 | static void store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2); |
| 1632 | static void insert_op1 (re_opcode_t op, unsigned char *loc, |
| 1633 | int arg, unsigned char *end); |
| 1634 | static void insert_op2 (re_opcode_t op, unsigned char *loc, |
| 1635 | int arg1, int arg2, unsigned char *end); |
| 1636 | static boolean at_begline_loc_p (re_char *pattern, re_char *p, |
| 1637 | reg_syntax_t syntax); |
| 1638 | static boolean at_endline_loc_p (re_char *p, re_char *pend, |
| 1639 | reg_syntax_t syntax); |
| 1640 | static re_char *skip_one_char (re_char *p); |
| 1641 | static int analyse_first (re_char *p, re_char *pend, |
| 1642 | char *fastmap, const int multibyte); |
| 1643 | |
| 1644 | /* Fetch the next character in the uncompiled pattern, with no |
| 1645 | translation. */ |
| 1646 | #define PATFETCH(c) \ |
| 1647 | do { \ |
| 1648 | int len; \ |
| 1649 | if (p == pend) return REG_EEND; \ |
| 1650 | c = RE_STRING_CHAR_AND_LENGTH (p, len, multibyte); \ |
| 1651 | p += len; \ |
| 1652 | } while (0) |
| 1653 | |
| 1654 | |
| 1655 | /* If `translate' is non-null, return translate[D], else just D. We |
| 1656 | cast the subscript to translate because some data is declared as |
| 1657 | `char *', to avoid warnings when a string constant is passed. But |
| 1658 | when we use a character as a subscript we must make it unsigned. */ |
| 1659 | #ifndef TRANSLATE |
| 1660 | # define TRANSLATE(d) \ |
| 1661 | (RE_TRANSLATE_P (translate) ? RE_TRANSLATE (translate, (d)) : (d)) |
| 1662 | #endif |
| 1663 | |
| 1664 | |
| 1665 | /* Macros for outputting the compiled pattern into `buffer'. */ |
| 1666 | |
| 1667 | /* If the buffer isn't allocated when it comes in, use this. */ |
| 1668 | #define INIT_BUF_SIZE 32 |
| 1669 | |
| 1670 | /* Make sure we have at least N more bytes of space in buffer. */ |
| 1671 | #define GET_BUFFER_SPACE(n) \ |
| 1672 | while ((size_t) (b - bufp->buffer + (n)) > bufp->allocated) \ |
| 1673 | EXTEND_BUFFER () |
| 1674 | |
| 1675 | /* Make sure we have one more byte of buffer space and then add C to it. */ |
| 1676 | #define BUF_PUSH(c) \ |
| 1677 | do { \ |
| 1678 | GET_BUFFER_SPACE (1); \ |
| 1679 | *b++ = (unsigned char) (c); \ |
| 1680 | } while (0) |
| 1681 | |
| 1682 | |
| 1683 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ |
| 1684 | #define BUF_PUSH_2(c1, c2) \ |
| 1685 | do { \ |
| 1686 | GET_BUFFER_SPACE (2); \ |
| 1687 | *b++ = (unsigned char) (c1); \ |
| 1688 | *b++ = (unsigned char) (c2); \ |
| 1689 | } while (0) |
| 1690 | |
| 1691 | |
| 1692 | /* Store a jump with opcode OP at LOC to location TO. We store a |
| 1693 | relative address offset by the three bytes the jump itself occupies. */ |
| 1694 | #define STORE_JUMP(op, loc, to) \ |
| 1695 | store_op1 (op, loc, (to) - (loc) - 3) |
| 1696 | |
| 1697 | /* Likewise, for a two-argument jump. */ |
| 1698 | #define STORE_JUMP2(op, loc, to, arg) \ |
| 1699 | store_op2 (op, loc, (to) - (loc) - 3, arg) |
| 1700 | |
| 1701 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
| 1702 | #define INSERT_JUMP(op, loc, to) \ |
| 1703 | insert_op1 (op, loc, (to) - (loc) - 3, b) |
| 1704 | |
| 1705 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ |
| 1706 | #define INSERT_JUMP2(op, loc, to, arg) \ |
| 1707 | insert_op2 (op, loc, (to) - (loc) - 3, arg, b) |
| 1708 | |
| 1709 | |
| 1710 | /* This is not an arbitrary limit: the arguments which represent offsets |
| 1711 | into the pattern are two bytes long. So if 2^15 bytes turns out to |
| 1712 | be too small, many things would have to change. */ |
| 1713 | # define MAX_BUF_SIZE (1L << 15) |
| 1714 | |
| 1715 | /* Extend the buffer by twice its current size via realloc and |
| 1716 | reset the pointers that pointed into the old block to point to the |
| 1717 | correct places in the new one. If extending the buffer results in it |
| 1718 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
| 1719 | #if __BOUNDED_POINTERS__ |
| 1720 | # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) |
| 1721 | # define MOVE_BUFFER_POINTER(P) \ |
| 1722 | (__ptrlow (P) = new_buffer + (__ptrlow (P) - old_buffer), \ |
| 1723 | SET_HIGH_BOUND (P), \ |
| 1724 | __ptrvalue (P) = new_buffer + (__ptrvalue (P) - old_buffer)) |
| 1725 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
| 1726 | else \ |
| 1727 | { \ |
| 1728 | SET_HIGH_BOUND (b); \ |
| 1729 | SET_HIGH_BOUND (begalt); \ |
| 1730 | if (fixup_alt_jump) \ |
| 1731 | SET_HIGH_BOUND (fixup_alt_jump); \ |
| 1732 | if (laststart) \ |
| 1733 | SET_HIGH_BOUND (laststart); \ |
| 1734 | if (pending_exact) \ |
| 1735 | SET_HIGH_BOUND (pending_exact); \ |
| 1736 | } |
| 1737 | #else |
| 1738 | # define MOVE_BUFFER_POINTER(P) ((P) = new_buffer + ((P) - old_buffer)) |
| 1739 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND |
| 1740 | #endif |
| 1741 | #define EXTEND_BUFFER() \ |
| 1742 | do { \ |
| 1743 | unsigned char *old_buffer = bufp->buffer; \ |
| 1744 | if (bufp->allocated == MAX_BUF_SIZE) \ |
| 1745 | return REG_ESIZE; \ |
| 1746 | bufp->allocated <<= 1; \ |
| 1747 | if (bufp->allocated > MAX_BUF_SIZE) \ |
| 1748 | bufp->allocated = MAX_BUF_SIZE; \ |
| 1749 | RETALLOC (bufp->buffer, bufp->allocated, unsigned char); \ |
| 1750 | if (bufp->buffer == NULL) \ |
| 1751 | return REG_ESPACE; \ |
| 1752 | /* If the buffer moved, move all the pointers into it. */ \ |
| 1753 | if (old_buffer != bufp->buffer) \ |
| 1754 | { \ |
| 1755 | unsigned char *new_buffer = bufp->buffer; \ |
| 1756 | MOVE_BUFFER_POINTER (b); \ |
| 1757 | MOVE_BUFFER_POINTER (begalt); \ |
| 1758 | if (fixup_alt_jump) \ |
| 1759 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ |
| 1760 | if (laststart) \ |
| 1761 | MOVE_BUFFER_POINTER (laststart); \ |
| 1762 | if (pending_exact) \ |
| 1763 | MOVE_BUFFER_POINTER (pending_exact); \ |
| 1764 | } \ |
| 1765 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
| 1766 | } while (0) |
| 1767 | |
| 1768 | |
| 1769 | /* Since we have one byte reserved for the register number argument to |
| 1770 | {start,stop}_memory, the maximum number of groups we can report |
| 1771 | things about is what fits in that byte. */ |
| 1772 | #define MAX_REGNUM 255 |
| 1773 | |
| 1774 | /* But patterns can have more than `MAX_REGNUM' registers. We just |
| 1775 | ignore the excess. */ |
| 1776 | typedef int regnum_t; |
| 1777 | |
| 1778 | |
| 1779 | /* Macros for the compile stack. */ |
| 1780 | |
| 1781 | /* Since offsets can go either forwards or backwards, this type needs to |
| 1782 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
| 1783 | /* int may be not enough when sizeof(int) == 2. */ |
| 1784 | typedef long pattern_offset_t; |
| 1785 | |
| 1786 | typedef struct |
| 1787 | { |
| 1788 | pattern_offset_t begalt_offset; |
| 1789 | pattern_offset_t fixup_alt_jump; |
| 1790 | pattern_offset_t laststart_offset; |
| 1791 | regnum_t regnum; |
| 1792 | } compile_stack_elt_t; |
| 1793 | |
| 1794 | |
| 1795 | typedef struct |
| 1796 | { |
| 1797 | compile_stack_elt_t *stack; |
| 1798 | size_t size; |
| 1799 | size_t avail; /* Offset of next open position. */ |
| 1800 | } compile_stack_type; |
| 1801 | |
| 1802 | |
| 1803 | #define INIT_COMPILE_STACK_SIZE 32 |
| 1804 | |
| 1805 | #define COMPILE_STACK_EMPTY (compile_stack.avail == 0) |
| 1806 | #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) |
| 1807 | |
| 1808 | /* The next available element. */ |
| 1809 | #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
| 1810 | |
| 1811 | /* Explicit quit checking is needed for Emacs, which uses polling to |
| 1812 | process input events. */ |
| 1813 | #ifdef emacs |
| 1814 | # define IMMEDIATE_QUIT_CHECK \ |
| 1815 | do { \ |
| 1816 | if (immediate_quit) QUIT; \ |
| 1817 | } while (0) |
| 1818 | #else |
| 1819 | # define IMMEDIATE_QUIT_CHECK ((void)0) |
| 1820 | #endif |
| 1821 | \f |
| 1822 | /* Structure to manage work area for range table. */ |
| 1823 | struct range_table_work_area |
| 1824 | { |
| 1825 | int *table; /* actual work area. */ |
| 1826 | int allocated; /* allocated size for work area in bytes. */ |
| 1827 | int used; /* actually used size in words. */ |
| 1828 | int bits; /* flag to record character classes */ |
| 1829 | }; |
| 1830 | |
| 1831 | /* Make sure that WORK_AREA can hold more N multibyte characters. |
| 1832 | This is used only in set_image_of_range and set_image_of_range_1. |
| 1833 | It expects WORK_AREA to be a pointer. |
| 1834 | If it can't get the space, it returns from the surrounding function. */ |
| 1835 | |
| 1836 | #define EXTEND_RANGE_TABLE(work_area, n) \ |
| 1837 | do { \ |
| 1838 | if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \ |
| 1839 | { \ |
| 1840 | extend_range_table_work_area (&work_area); \ |
| 1841 | if ((work_area).table == 0) \ |
| 1842 | return (REG_ESPACE); \ |
| 1843 | } \ |
| 1844 | } while (0) |
| 1845 | |
| 1846 | #define SET_RANGE_TABLE_WORK_AREA_BIT(work_area, bit) \ |
| 1847 | (work_area).bits |= (bit) |
| 1848 | |
| 1849 | /* Bits used to implement the multibyte-part of the various character classes |
| 1850 | such as [:alnum:] in a charset's range table. */ |
| 1851 | #define BIT_WORD 0x1 |
| 1852 | #define BIT_LOWER 0x2 |
| 1853 | #define BIT_PUNCT 0x4 |
| 1854 | #define BIT_SPACE 0x8 |
| 1855 | #define BIT_UPPER 0x10 |
| 1856 | #define BIT_MULTIBYTE 0x20 |
| 1857 | |
| 1858 | /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */ |
| 1859 | #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \ |
| 1860 | do { \ |
| 1861 | EXTEND_RANGE_TABLE ((work_area), 2); \ |
| 1862 | (work_area).table[(work_area).used++] = (range_start); \ |
| 1863 | (work_area).table[(work_area).used++] = (range_end); \ |
| 1864 | } while (0) |
| 1865 | |
| 1866 | /* Free allocated memory for WORK_AREA. */ |
| 1867 | #define FREE_RANGE_TABLE_WORK_AREA(work_area) \ |
| 1868 | do { \ |
| 1869 | if ((work_area).table) \ |
| 1870 | free ((work_area).table); \ |
| 1871 | } while (0) |
| 1872 | |
| 1873 | #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0, (work_area).bits = 0) |
| 1874 | #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used) |
| 1875 | #define RANGE_TABLE_WORK_BITS(work_area) ((work_area).bits) |
| 1876 | #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i]) |
| 1877 | \f |
| 1878 | |
| 1879 | /* Set the bit for character C in a list. */ |
| 1880 | #define SET_LIST_BIT(c) (b[((c)) / BYTEWIDTH] |= 1 << ((c) % BYTEWIDTH)) |
| 1881 | |
| 1882 | |
| 1883 | #ifdef emacs |
| 1884 | |
| 1885 | /* Store characters in the range FROM to TO in the bitmap at B (for |
| 1886 | ASCII and unibyte characters) and WORK_AREA (for multibyte |
| 1887 | characters) while translating them and paying attention to the |
| 1888 | continuity of translated characters. |
| 1889 | |
| 1890 | Implementation note: It is better to implement these fairly big |
| 1891 | macros by a function, but it's not that easy because macros called |
| 1892 | in this macro assume various local variables already declared. */ |
| 1893 | |
| 1894 | /* Both FROM and TO are ASCII characters. */ |
| 1895 | |
| 1896 | #define SETUP_ASCII_RANGE(work_area, FROM, TO) \ |
| 1897 | do { \ |
| 1898 | int C0, C1; \ |
| 1899 | \ |
| 1900 | for (C0 = (FROM); C0 <= (TO); C0++) \ |
| 1901 | { \ |
| 1902 | C1 = TRANSLATE (C0); \ |
| 1903 | if (! ASCII_CHAR_P (C1)) \ |
| 1904 | { \ |
| 1905 | SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \ |
| 1906 | if ((C1 = RE_CHAR_TO_UNIBYTE (C1)) < 0) \ |
| 1907 | C1 = C0; \ |
| 1908 | } \ |
| 1909 | SET_LIST_BIT (C1); \ |
| 1910 | } \ |
| 1911 | } while (0) |
| 1912 | |
| 1913 | |
| 1914 | /* Both FROM and TO are unibyte characters (0x80..0xFF). */ |
| 1915 | |
| 1916 | #define SETUP_UNIBYTE_RANGE(work_area, FROM, TO) \ |
| 1917 | do { \ |
| 1918 | int C0, C1, C2, I; \ |
| 1919 | int USED = RANGE_TABLE_WORK_USED (work_area); \ |
| 1920 | \ |
| 1921 | for (C0 = (FROM); C0 <= (TO); C0++) \ |
| 1922 | { \ |
| 1923 | C1 = RE_CHAR_TO_MULTIBYTE (C0); \ |
| 1924 | if (CHAR_BYTE8_P (C1)) \ |
| 1925 | SET_LIST_BIT (C0); \ |
| 1926 | else \ |
| 1927 | { \ |
| 1928 | C2 = TRANSLATE (C1); \ |
| 1929 | if (C2 == C1 \ |
| 1930 | || (C1 = RE_CHAR_TO_UNIBYTE (C2)) < 0) \ |
| 1931 | C1 = C0; \ |
| 1932 | SET_LIST_BIT (C1); \ |
| 1933 | for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \ |
| 1934 | { \ |
| 1935 | int from = RANGE_TABLE_WORK_ELT (work_area, I); \ |
| 1936 | int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \ |
| 1937 | \ |
| 1938 | if (C2 >= from - 1 && C2 <= to + 1) \ |
| 1939 | { \ |
| 1940 | if (C2 == from - 1) \ |
| 1941 | RANGE_TABLE_WORK_ELT (work_area, I)--; \ |
| 1942 | else if (C2 == to + 1) \ |
| 1943 | RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \ |
| 1944 | break; \ |
| 1945 | } \ |
| 1946 | } \ |
| 1947 | if (I < USED) \ |
| 1948 | SET_RANGE_TABLE_WORK_AREA ((work_area), C2, C2); \ |
| 1949 | } \ |
| 1950 | } \ |
| 1951 | } while (0) |
| 1952 | |
| 1953 | |
| 1954 | /* Both FROM and TO are multibyte characters. */ |
| 1955 | |
| 1956 | #define SETUP_MULTIBYTE_RANGE(work_area, FROM, TO) \ |
| 1957 | do { \ |
| 1958 | int C0, C1, C2, I, USED = RANGE_TABLE_WORK_USED (work_area); \ |
| 1959 | \ |
| 1960 | SET_RANGE_TABLE_WORK_AREA ((work_area), (FROM), (TO)); \ |
| 1961 | for (C0 = (FROM); C0 <= (TO); C0++) \ |
| 1962 | { \ |
| 1963 | C1 = TRANSLATE (C0); \ |
| 1964 | if ((C2 = RE_CHAR_TO_UNIBYTE (C1)) >= 0 \ |
| 1965 | || (C1 != C0 && (C2 = RE_CHAR_TO_UNIBYTE (C0)) >= 0)) \ |
| 1966 | SET_LIST_BIT (C2); \ |
| 1967 | if (C1 >= (FROM) && C1 <= (TO)) \ |
| 1968 | continue; \ |
| 1969 | for (I = RANGE_TABLE_WORK_USED (work_area) - 2; I >= USED; I -= 2) \ |
| 1970 | { \ |
| 1971 | int from = RANGE_TABLE_WORK_ELT (work_area, I); \ |
| 1972 | int to = RANGE_TABLE_WORK_ELT (work_area, I + 1); \ |
| 1973 | \ |
| 1974 | if (C1 >= from - 1 && C1 <= to + 1) \ |
| 1975 | { \ |
| 1976 | if (C1 == from - 1) \ |
| 1977 | RANGE_TABLE_WORK_ELT (work_area, I)--; \ |
| 1978 | else if (C1 == to + 1) \ |
| 1979 | RANGE_TABLE_WORK_ELT (work_area, I + 1)++; \ |
| 1980 | break; \ |
| 1981 | } \ |
| 1982 | } \ |
| 1983 | if (I < USED) \ |
| 1984 | SET_RANGE_TABLE_WORK_AREA ((work_area), C1, C1); \ |
| 1985 | } \ |
| 1986 | } while (0) |
| 1987 | |
| 1988 | #endif /* emacs */ |
| 1989 | |
| 1990 | /* Get the next unsigned number in the uncompiled pattern. */ |
| 1991 | #define GET_UNSIGNED_NUMBER(num) \ |
| 1992 | do { \ |
| 1993 | if (p == pend) \ |
| 1994 | FREE_STACK_RETURN (REG_EBRACE); \ |
| 1995 | else \ |
| 1996 | { \ |
| 1997 | PATFETCH (c); \ |
| 1998 | while ('0' <= c && c <= '9') \ |
| 1999 | { \ |
| 2000 | int prev; \ |
| 2001 | if (num < 0) \ |
| 2002 | num = 0; \ |
| 2003 | prev = num; \ |
| 2004 | num = num * 10 + c - '0'; \ |
| 2005 | if (num / 10 != prev) \ |
| 2006 | FREE_STACK_RETURN (REG_BADBR); \ |
| 2007 | if (p == pend) \ |
| 2008 | FREE_STACK_RETURN (REG_EBRACE); \ |
| 2009 | PATFETCH (c); \ |
| 2010 | } \ |
| 2011 | } \ |
| 2012 | } while (0) |
| 2013 | \f |
| 2014 | #if ! WIDE_CHAR_SUPPORT |
| 2015 | |
| 2016 | /* Map a string to the char class it names (if any). */ |
| 2017 | re_wctype_t |
| 2018 | re_wctype (const re_char *str) |
| 2019 | { |
| 2020 | const char *string = (const char *) str; |
| 2021 | if (STREQ (string, "alnum")) return RECC_ALNUM; |
| 2022 | else if (STREQ (string, "alpha")) return RECC_ALPHA; |
| 2023 | else if (STREQ (string, "word")) return RECC_WORD; |
| 2024 | else if (STREQ (string, "ascii")) return RECC_ASCII; |
| 2025 | else if (STREQ (string, "nonascii")) return RECC_NONASCII; |
| 2026 | else if (STREQ (string, "graph")) return RECC_GRAPH; |
| 2027 | else if (STREQ (string, "lower")) return RECC_LOWER; |
| 2028 | else if (STREQ (string, "print")) return RECC_PRINT; |
| 2029 | else if (STREQ (string, "punct")) return RECC_PUNCT; |
| 2030 | else if (STREQ (string, "space")) return RECC_SPACE; |
| 2031 | else if (STREQ (string, "upper")) return RECC_UPPER; |
| 2032 | else if (STREQ (string, "unibyte")) return RECC_UNIBYTE; |
| 2033 | else if (STREQ (string, "multibyte")) return RECC_MULTIBYTE; |
| 2034 | else if (STREQ (string, "digit")) return RECC_DIGIT; |
| 2035 | else if (STREQ (string, "xdigit")) return RECC_XDIGIT; |
| 2036 | else if (STREQ (string, "cntrl")) return RECC_CNTRL; |
| 2037 | else if (STREQ (string, "blank")) return RECC_BLANK; |
| 2038 | else return 0; |
| 2039 | } |
| 2040 | |
| 2041 | /* True if CH is in the char class CC. */ |
| 2042 | boolean |
| 2043 | re_iswctype (int ch, re_wctype_t cc) |
| 2044 | { |
| 2045 | switch (cc) |
| 2046 | { |
| 2047 | case RECC_ALNUM: return ISALNUM (ch) != 0; |
| 2048 | case RECC_ALPHA: return ISALPHA (ch) != 0; |
| 2049 | case RECC_BLANK: return ISBLANK (ch) != 0; |
| 2050 | case RECC_CNTRL: return ISCNTRL (ch) != 0; |
| 2051 | case RECC_DIGIT: return ISDIGIT (ch) != 0; |
| 2052 | case RECC_GRAPH: return ISGRAPH (ch) != 0; |
| 2053 | case RECC_LOWER: return ISLOWER (ch) != 0; |
| 2054 | case RECC_PRINT: return ISPRINT (ch) != 0; |
| 2055 | case RECC_PUNCT: return ISPUNCT (ch) != 0; |
| 2056 | case RECC_SPACE: return ISSPACE (ch) != 0; |
| 2057 | case RECC_UPPER: return ISUPPER (ch) != 0; |
| 2058 | case RECC_XDIGIT: return ISXDIGIT (ch) != 0; |
| 2059 | case RECC_ASCII: return IS_REAL_ASCII (ch) != 0; |
| 2060 | case RECC_NONASCII: return !IS_REAL_ASCII (ch); |
| 2061 | case RECC_UNIBYTE: return ISUNIBYTE (ch) != 0; |
| 2062 | case RECC_MULTIBYTE: return !ISUNIBYTE (ch); |
| 2063 | case RECC_WORD: return ISWORD (ch) != 0; |
| 2064 | case RECC_ERROR: return false; |
| 2065 | default: |
| 2066 | abort (); |
| 2067 | } |
| 2068 | } |
| 2069 | |
| 2070 | /* Return a bit-pattern to use in the range-table bits to match multibyte |
| 2071 | chars of class CC. */ |
| 2072 | static int |
| 2073 | re_wctype_to_bit (re_wctype_t cc) |
| 2074 | { |
| 2075 | switch (cc) |
| 2076 | { |
| 2077 | case RECC_NONASCII: case RECC_PRINT: case RECC_GRAPH: |
| 2078 | case RECC_MULTIBYTE: return BIT_MULTIBYTE; |
| 2079 | case RECC_ALPHA: case RECC_ALNUM: case RECC_WORD: return BIT_WORD; |
| 2080 | case RECC_LOWER: return BIT_LOWER; |
| 2081 | case RECC_UPPER: return BIT_UPPER; |
| 2082 | case RECC_PUNCT: return BIT_PUNCT; |
| 2083 | case RECC_SPACE: return BIT_SPACE; |
| 2084 | case RECC_ASCII: case RECC_DIGIT: case RECC_XDIGIT: case RECC_CNTRL: |
| 2085 | case RECC_BLANK: case RECC_UNIBYTE: case RECC_ERROR: return 0; |
| 2086 | default: |
| 2087 | abort (); |
| 2088 | } |
| 2089 | } |
| 2090 | #endif |
| 2091 | \f |
| 2092 | /* Filling in the work area of a range. */ |
| 2093 | |
| 2094 | /* Actually extend the space in WORK_AREA. */ |
| 2095 | |
| 2096 | static void |
| 2097 | extend_range_table_work_area (struct range_table_work_area *work_area) |
| 2098 | { |
| 2099 | work_area->allocated += 16 * sizeof (int); |
| 2100 | work_area->table = realloc (work_area->table, work_area->allocated); |
| 2101 | } |
| 2102 | |
| 2103 | #if 0 |
| 2104 | #ifdef emacs |
| 2105 | |
| 2106 | /* Carefully find the ranges of codes that are equivalent |
| 2107 | under case conversion to the range start..end when passed through |
| 2108 | TRANSLATE. Handle the case where non-letters can come in between |
| 2109 | two upper-case letters (which happens in Latin-1). |
| 2110 | Also handle the case of groups of more than 2 case-equivalent chars. |
| 2111 | |
| 2112 | The basic method is to look at consecutive characters and see |
| 2113 | if they can form a run that can be handled as one. |
| 2114 | |
| 2115 | Returns -1 if successful, REG_ESPACE if ran out of space. */ |
| 2116 | |
| 2117 | static int |
| 2118 | set_image_of_range_1 (struct range_table_work_area *work_area, |
| 2119 | re_wchar_t start, re_wchar_t end, |
| 2120 | RE_TRANSLATE_TYPE translate) |
| 2121 | { |
| 2122 | /* `one_case' indicates a character, or a run of characters, |
| 2123 | each of which is an isolate (no case-equivalents). |
| 2124 | This includes all ASCII non-letters. |
| 2125 | |
| 2126 | `two_case' indicates a character, or a run of characters, |
| 2127 | each of which has two case-equivalent forms. |
| 2128 | This includes all ASCII letters. |
| 2129 | |
| 2130 | `strange' indicates a character that has more than one |
| 2131 | case-equivalent. */ |
| 2132 | |
| 2133 | enum case_type {one_case, two_case, strange}; |
| 2134 | |
| 2135 | /* Describe the run that is in progress, |
| 2136 | which the next character can try to extend. |
| 2137 | If run_type is strange, that means there really is no run. |
| 2138 | If run_type is one_case, then run_start...run_end is the run. |
| 2139 | If run_type is two_case, then the run is run_start...run_end, |
| 2140 | and the case-equivalents end at run_eqv_end. */ |
| 2141 | |
| 2142 | enum case_type run_type = strange; |
| 2143 | int run_start, run_end, run_eqv_end; |
| 2144 | |
| 2145 | Lisp_Object eqv_table; |
| 2146 | |
| 2147 | if (!RE_TRANSLATE_P (translate)) |
| 2148 | { |
| 2149 | EXTEND_RANGE_TABLE (work_area, 2); |
| 2150 | work_area->table[work_area->used++] = (start); |
| 2151 | work_area->table[work_area->used++] = (end); |
| 2152 | return -1; |
| 2153 | } |
| 2154 | |
| 2155 | eqv_table = XCHAR_TABLE (translate)->extras[2]; |
| 2156 | |
| 2157 | for (; start <= end; start++) |
| 2158 | { |
| 2159 | enum case_type this_type; |
| 2160 | int eqv = RE_TRANSLATE (eqv_table, start); |
| 2161 | int minchar, maxchar; |
| 2162 | |
| 2163 | /* Classify this character */ |
| 2164 | if (eqv == start) |
| 2165 | this_type = one_case; |
| 2166 | else if (RE_TRANSLATE (eqv_table, eqv) == start) |
| 2167 | this_type = two_case; |
| 2168 | else |
| 2169 | this_type = strange; |
| 2170 | |
| 2171 | if (start < eqv) |
| 2172 | minchar = start, maxchar = eqv; |
| 2173 | else |
| 2174 | minchar = eqv, maxchar = start; |
| 2175 | |
| 2176 | /* Can this character extend the run in progress? */ |
| 2177 | if (this_type == strange || this_type != run_type |
| 2178 | || !(minchar == run_end + 1 |
| 2179 | && (run_type == two_case |
| 2180 | ? maxchar == run_eqv_end + 1 : 1))) |
| 2181 | { |
| 2182 | /* No, end the run. |
| 2183 | Record each of its equivalent ranges. */ |
| 2184 | if (run_type == one_case) |
| 2185 | { |
| 2186 | EXTEND_RANGE_TABLE (work_area, 2); |
| 2187 | work_area->table[work_area->used++] = run_start; |
| 2188 | work_area->table[work_area->used++] = run_end; |
| 2189 | } |
| 2190 | else if (run_type == two_case) |
| 2191 | { |
| 2192 | EXTEND_RANGE_TABLE (work_area, 4); |
| 2193 | work_area->table[work_area->used++] = run_start; |
| 2194 | work_area->table[work_area->used++] = run_end; |
| 2195 | work_area->table[work_area->used++] |
| 2196 | = RE_TRANSLATE (eqv_table, run_start); |
| 2197 | work_area->table[work_area->used++] |
| 2198 | = RE_TRANSLATE (eqv_table, run_end); |
| 2199 | } |
| 2200 | run_type = strange; |
| 2201 | } |
| 2202 | |
| 2203 | if (this_type == strange) |
| 2204 | { |
| 2205 | /* For a strange character, add each of its equivalents, one |
| 2206 | by one. Don't start a range. */ |
| 2207 | do |
| 2208 | { |
| 2209 | EXTEND_RANGE_TABLE (work_area, 2); |
| 2210 | work_area->table[work_area->used++] = eqv; |
| 2211 | work_area->table[work_area->used++] = eqv; |
| 2212 | eqv = RE_TRANSLATE (eqv_table, eqv); |
| 2213 | } |
| 2214 | while (eqv != start); |
| 2215 | } |
| 2216 | |
| 2217 | /* Add this char to the run, or start a new run. */ |
| 2218 | else if (run_type == strange) |
| 2219 | { |
| 2220 | /* Initialize a new range. */ |
| 2221 | run_type = this_type; |
| 2222 | run_start = start; |
| 2223 | run_end = start; |
| 2224 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); |
| 2225 | } |
| 2226 | else |
| 2227 | { |
| 2228 | /* Extend a running range. */ |
| 2229 | run_end = minchar; |
| 2230 | run_eqv_end = RE_TRANSLATE (eqv_table, run_end); |
| 2231 | } |
| 2232 | } |
| 2233 | |
| 2234 | /* If a run is still in progress at the end, finish it now |
| 2235 | by recording its equivalent ranges. */ |
| 2236 | if (run_type == one_case) |
| 2237 | { |
| 2238 | EXTEND_RANGE_TABLE (work_area, 2); |
| 2239 | work_area->table[work_area->used++] = run_start; |
| 2240 | work_area->table[work_area->used++] = run_end; |
| 2241 | } |
| 2242 | else if (run_type == two_case) |
| 2243 | { |
| 2244 | EXTEND_RANGE_TABLE (work_area, 4); |
| 2245 | work_area->table[work_area->used++] = run_start; |
| 2246 | work_area->table[work_area->used++] = run_end; |
| 2247 | work_area->table[work_area->used++] |
| 2248 | = RE_TRANSLATE (eqv_table, run_start); |
| 2249 | work_area->table[work_area->used++] |
| 2250 | = RE_TRANSLATE (eqv_table, run_end); |
| 2251 | } |
| 2252 | |
| 2253 | return -1; |
| 2254 | } |
| 2255 | |
| 2256 | #endif /* emacs */ |
| 2257 | |
| 2258 | /* Record the image of the range start..end when passed through |
| 2259 | TRANSLATE. This is not necessarily TRANSLATE(start)..TRANSLATE(end) |
| 2260 | and is not even necessarily contiguous. |
| 2261 | Normally we approximate it with the smallest contiguous range that contains |
| 2262 | all the chars we need. However, for Latin-1 we go to extra effort |
| 2263 | to do a better job. |
| 2264 | |
| 2265 | This function is not called for ASCII ranges. |
| 2266 | |
| 2267 | Returns -1 if successful, REG_ESPACE if ran out of space. */ |
| 2268 | |
| 2269 | static int |
| 2270 | set_image_of_range (struct range_table_work_area *work_area, |
| 2271 | re_wchar_t start, re_wchar_t end, |
| 2272 | RE_TRANSLATE_TYPE translate) |
| 2273 | { |
| 2274 | re_wchar_t cmin, cmax; |
| 2275 | |
| 2276 | #ifdef emacs |
| 2277 | /* For Latin-1 ranges, use set_image_of_range_1 |
| 2278 | to get proper handling of ranges that include letters and nonletters. |
| 2279 | For a range that includes the whole of Latin-1, this is not necessary. |
| 2280 | For other character sets, we don't bother to get this right. */ |
| 2281 | if (RE_TRANSLATE_P (translate) && start < 04400 |
| 2282 | && !(start < 04200 && end >= 04377)) |
| 2283 | { |
| 2284 | int newend; |
| 2285 | int tem; |
| 2286 | newend = end; |
| 2287 | if (newend > 04377) |
| 2288 | newend = 04377; |
| 2289 | tem = set_image_of_range_1 (work_area, start, newend, translate); |
| 2290 | if (tem > 0) |
| 2291 | return tem; |
| 2292 | |
| 2293 | start = 04400; |
| 2294 | if (end < 04400) |
| 2295 | return -1; |
| 2296 | } |
| 2297 | #endif |
| 2298 | |
| 2299 | EXTEND_RANGE_TABLE (work_area, 2); |
| 2300 | work_area->table[work_area->used++] = (start); |
| 2301 | work_area->table[work_area->used++] = (end); |
| 2302 | |
| 2303 | cmin = -1, cmax = -1; |
| 2304 | |
| 2305 | if (RE_TRANSLATE_P (translate)) |
| 2306 | { |
| 2307 | int ch; |
| 2308 | |
| 2309 | for (ch = start; ch <= end; ch++) |
| 2310 | { |
| 2311 | re_wchar_t c = TRANSLATE (ch); |
| 2312 | if (! (start <= c && c <= end)) |
| 2313 | { |
| 2314 | if (cmin == -1) |
| 2315 | cmin = c, cmax = c; |
| 2316 | else |
| 2317 | { |
| 2318 | cmin = MIN (cmin, c); |
| 2319 | cmax = MAX (cmax, c); |
| 2320 | } |
| 2321 | } |
| 2322 | } |
| 2323 | |
| 2324 | if (cmin != -1) |
| 2325 | { |
| 2326 | EXTEND_RANGE_TABLE (work_area, 2); |
| 2327 | work_area->table[work_area->used++] = (cmin); |
| 2328 | work_area->table[work_area->used++] = (cmax); |
| 2329 | } |
| 2330 | } |
| 2331 | |
| 2332 | return -1; |
| 2333 | } |
| 2334 | #endif /* 0 */ |
| 2335 | \f |
| 2336 | #ifndef MATCH_MAY_ALLOCATE |
| 2337 | |
| 2338 | /* If we cannot allocate large objects within re_match_2_internal, |
| 2339 | we make the fail stack and register vectors global. |
| 2340 | The fail stack, we grow to the maximum size when a regexp |
| 2341 | is compiled. |
| 2342 | The register vectors, we adjust in size each time we |
| 2343 | compile a regexp, according to the number of registers it needs. */ |
| 2344 | |
| 2345 | static fail_stack_type fail_stack; |
| 2346 | |
| 2347 | /* Size with which the following vectors are currently allocated. |
| 2348 | That is so we can make them bigger as needed, |
| 2349 | but never make them smaller. */ |
| 2350 | static int regs_allocated_size; |
| 2351 | |
| 2352 | static re_char ** regstart, ** regend; |
| 2353 | static re_char **best_regstart, **best_regend; |
| 2354 | |
| 2355 | /* Make the register vectors big enough for NUM_REGS registers, |
| 2356 | but don't make them smaller. */ |
| 2357 | |
| 2358 | static |
| 2359 | regex_grow_registers (int num_regs) |
| 2360 | { |
| 2361 | if (num_regs > regs_allocated_size) |
| 2362 | { |
| 2363 | RETALLOC_IF (regstart, num_regs, re_char *); |
| 2364 | RETALLOC_IF (regend, num_regs, re_char *); |
| 2365 | RETALLOC_IF (best_regstart, num_regs, re_char *); |
| 2366 | RETALLOC_IF (best_regend, num_regs, re_char *); |
| 2367 | |
| 2368 | regs_allocated_size = num_regs; |
| 2369 | } |
| 2370 | } |
| 2371 | |
| 2372 | #endif /* not MATCH_MAY_ALLOCATE */ |
| 2373 | \f |
| 2374 | static boolean group_in_compile_stack (compile_stack_type compile_stack, |
| 2375 | regnum_t regnum); |
| 2376 | |
| 2377 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
| 2378 | Returns one of error codes defined in `regex.h', or zero for success. |
| 2379 | |
| 2380 | Assumes the `allocated' (and perhaps `buffer') and `translate' |
| 2381 | fields are set in BUFP on entry. |
| 2382 | |
| 2383 | If it succeeds, results are put in BUFP (if it returns an error, the |
| 2384 | contents of BUFP are undefined): |
| 2385 | `buffer' is the compiled pattern; |
| 2386 | `syntax' is set to SYNTAX; |
| 2387 | `used' is set to the length of the compiled pattern; |
| 2388 | `fastmap_accurate' is zero; |
| 2389 | `re_nsub' is the number of subexpressions in PATTERN; |
| 2390 | `not_bol' and `not_eol' are zero; |
| 2391 | |
| 2392 | The `fastmap' field is neither examined nor set. */ |
| 2393 | |
| 2394 | /* Insert the `jump' from the end of last alternative to "here". |
| 2395 | The space for the jump has already been allocated. */ |
| 2396 | #define FIXUP_ALT_JUMP() \ |
| 2397 | do { \ |
| 2398 | if (fixup_alt_jump) \ |
| 2399 | STORE_JUMP (jump, fixup_alt_jump, b); \ |
| 2400 | } while (0) |
| 2401 | |
| 2402 | |
| 2403 | /* Return, freeing storage we allocated. */ |
| 2404 | #define FREE_STACK_RETURN(value) \ |
| 2405 | do { \ |
| 2406 | FREE_RANGE_TABLE_WORK_AREA (range_table_work); \ |
| 2407 | free (compile_stack.stack); \ |
| 2408 | return value; \ |
| 2409 | } while (0) |
| 2410 | |
| 2411 | static reg_errcode_t |
| 2412 | regex_compile (const re_char *pattern, size_t size, reg_syntax_t syntax, struct re_pattern_buffer *bufp) |
| 2413 | { |
| 2414 | /* We fetch characters from PATTERN here. */ |
| 2415 | register re_wchar_t c, c1; |
| 2416 | |
| 2417 | /* Points to the end of the buffer, where we should append. */ |
| 2418 | register unsigned char *b; |
| 2419 | |
| 2420 | /* Keeps track of unclosed groups. */ |
| 2421 | compile_stack_type compile_stack; |
| 2422 | |
| 2423 | /* Points to the current (ending) position in the pattern. */ |
| 2424 | #ifdef AIX |
| 2425 | /* `const' makes AIX compiler fail. */ |
| 2426 | unsigned char *p = pattern; |
| 2427 | #else |
| 2428 | re_char *p = pattern; |
| 2429 | #endif |
| 2430 | re_char *pend = pattern + size; |
| 2431 | |
| 2432 | /* How to translate the characters in the pattern. */ |
| 2433 | RE_TRANSLATE_TYPE translate = bufp->translate; |
| 2434 | |
| 2435 | /* Address of the count-byte of the most recently inserted `exactn' |
| 2436 | command. This makes it possible to tell if a new exact-match |
| 2437 | character can be added to that command or if the character requires |
| 2438 | a new `exactn' command. */ |
| 2439 | unsigned char *pending_exact = 0; |
| 2440 | |
| 2441 | /* Address of start of the most recently finished expression. |
| 2442 | This tells, e.g., postfix * where to find the start of its |
| 2443 | operand. Reset at the beginning of groups and alternatives. */ |
| 2444 | unsigned char *laststart = 0; |
| 2445 | |
| 2446 | /* Address of beginning of regexp, or inside of last group. */ |
| 2447 | unsigned char *begalt; |
| 2448 | |
| 2449 | /* Place in the uncompiled pattern (i.e., the {) to |
| 2450 | which to go back if the interval is invalid. */ |
| 2451 | re_char *beg_interval; |
| 2452 | |
| 2453 | /* Address of the place where a forward jump should go to the end of |
| 2454 | the containing expression. Each alternative of an `or' -- except the |
| 2455 | last -- ends with a forward jump of this sort. */ |
| 2456 | unsigned char *fixup_alt_jump = 0; |
| 2457 | |
| 2458 | /* Work area for range table of charset. */ |
| 2459 | struct range_table_work_area range_table_work; |
| 2460 | |
| 2461 | /* If the object matched can contain multibyte characters. */ |
| 2462 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
| 2463 | |
| 2464 | /* Nonzero if we have pushed down into a subpattern. */ |
| 2465 | int in_subpattern = 0; |
| 2466 | |
| 2467 | /* These hold the values of p, pattern, and pend from the main |
| 2468 | pattern when we have pushed into a subpattern. */ |
| 2469 | re_char *main_p IF_LINT (= NULL); |
| 2470 | re_char *main_pattern IF_LINT (= NULL); |
| 2471 | re_char *main_pend IF_LINT (= NULL); |
| 2472 | |
| 2473 | #ifdef DEBUG |
| 2474 | debug++; |
| 2475 | DEBUG_PRINT ("\nCompiling pattern: "); |
| 2476 | if (debug > 0) |
| 2477 | { |
| 2478 | unsigned debug_count; |
| 2479 | |
| 2480 | for (debug_count = 0; debug_count < size; debug_count++) |
| 2481 | putchar (pattern[debug_count]); |
| 2482 | putchar ('\n'); |
| 2483 | } |
| 2484 | #endif /* DEBUG */ |
| 2485 | |
| 2486 | /* Initialize the compile stack. */ |
| 2487 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); |
| 2488 | if (compile_stack.stack == NULL) |
| 2489 | return REG_ESPACE; |
| 2490 | |
| 2491 | compile_stack.size = INIT_COMPILE_STACK_SIZE; |
| 2492 | compile_stack.avail = 0; |
| 2493 | |
| 2494 | range_table_work.table = 0; |
| 2495 | range_table_work.allocated = 0; |
| 2496 | |
| 2497 | /* Initialize the pattern buffer. */ |
| 2498 | bufp->syntax = syntax; |
| 2499 | bufp->fastmap_accurate = 0; |
| 2500 | bufp->not_bol = bufp->not_eol = 0; |
| 2501 | bufp->used_syntax = 0; |
| 2502 | |
| 2503 | /* Set `used' to zero, so that if we return an error, the pattern |
| 2504 | printer (for debugging) will think there's no pattern. We reset it |
| 2505 | at the end. */ |
| 2506 | bufp->used = 0; |
| 2507 | |
| 2508 | /* Always count groups, whether or not bufp->no_sub is set. */ |
| 2509 | bufp->re_nsub = 0; |
| 2510 | |
| 2511 | #if !defined emacs && !defined SYNTAX_TABLE |
| 2512 | /* Initialize the syntax table. */ |
| 2513 | init_syntax_once (); |
| 2514 | #endif |
| 2515 | |
| 2516 | if (bufp->allocated == 0) |
| 2517 | { |
| 2518 | if (bufp->buffer) |
| 2519 | { /* If zero allocated, but buffer is non-null, try to realloc |
| 2520 | enough space. This loses if buffer's address is bogus, but |
| 2521 | that is the user's responsibility. */ |
| 2522 | RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char); |
| 2523 | } |
| 2524 | else |
| 2525 | { /* Caller did not allocate a buffer. Do it for them. */ |
| 2526 | bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char); |
| 2527 | } |
| 2528 | if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE); |
| 2529 | |
| 2530 | bufp->allocated = INIT_BUF_SIZE; |
| 2531 | } |
| 2532 | |
| 2533 | begalt = b = bufp->buffer; |
| 2534 | |
| 2535 | /* Loop through the uncompiled pattern until we're at the end. */ |
| 2536 | while (1) |
| 2537 | { |
| 2538 | if (p == pend) |
| 2539 | { |
| 2540 | /* If this is the end of an included regexp, |
| 2541 | pop back to the main regexp and try again. */ |
| 2542 | if (in_subpattern) |
| 2543 | { |
| 2544 | in_subpattern = 0; |
| 2545 | pattern = main_pattern; |
| 2546 | p = main_p; |
| 2547 | pend = main_pend; |
| 2548 | continue; |
| 2549 | } |
| 2550 | /* If this is the end of the main regexp, we are done. */ |
| 2551 | break; |
| 2552 | } |
| 2553 | |
| 2554 | PATFETCH (c); |
| 2555 | |
| 2556 | switch (c) |
| 2557 | { |
| 2558 | case ' ': |
| 2559 | { |
| 2560 | re_char *p1 = p; |
| 2561 | |
| 2562 | /* If there's no special whitespace regexp, treat |
| 2563 | spaces normally. And don't try to do this recursively. */ |
| 2564 | if (!whitespace_regexp || in_subpattern) |
| 2565 | goto normal_char; |
| 2566 | |
| 2567 | /* Peek past following spaces. */ |
| 2568 | while (p1 != pend) |
| 2569 | { |
| 2570 | if (*p1 != ' ') |
| 2571 | break; |
| 2572 | p1++; |
| 2573 | } |
| 2574 | /* If the spaces are followed by a repetition op, |
| 2575 | treat them normally. */ |
| 2576 | if (p1 != pend |
| 2577 | && (*p1 == '*' || *p1 == '+' || *p1 == '?' |
| 2578 | || (*p1 == '\\' && p1 + 1 != pend && p1[1] == '{'))) |
| 2579 | goto normal_char; |
| 2580 | |
| 2581 | /* Replace the spaces with the whitespace regexp. */ |
| 2582 | in_subpattern = 1; |
| 2583 | main_p = p1; |
| 2584 | main_pend = pend; |
| 2585 | main_pattern = pattern; |
| 2586 | p = pattern = whitespace_regexp; |
| 2587 | pend = p + strlen ((const char *) p); |
| 2588 | break; |
| 2589 | } |
| 2590 | |
| 2591 | case '^': |
| 2592 | { |
| 2593 | if ( /* If at start of pattern, it's an operator. */ |
| 2594 | p == pattern + 1 |
| 2595 | /* If context independent, it's an operator. */ |
| 2596 | || syntax & RE_CONTEXT_INDEP_ANCHORS |
| 2597 | /* Otherwise, depends on what's come before. */ |
| 2598 | || at_begline_loc_p (pattern, p, syntax)) |
| 2599 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline); |
| 2600 | else |
| 2601 | goto normal_char; |
| 2602 | } |
| 2603 | break; |
| 2604 | |
| 2605 | |
| 2606 | case '$': |
| 2607 | { |
| 2608 | if ( /* If at end of pattern, it's an operator. */ |
| 2609 | p == pend |
| 2610 | /* If context independent, it's an operator. */ |
| 2611 | || syntax & RE_CONTEXT_INDEP_ANCHORS |
| 2612 | /* Otherwise, depends on what's next. */ |
| 2613 | || at_endline_loc_p (p, pend, syntax)) |
| 2614 | BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline); |
| 2615 | else |
| 2616 | goto normal_char; |
| 2617 | } |
| 2618 | break; |
| 2619 | |
| 2620 | |
| 2621 | case '+': |
| 2622 | case '?': |
| 2623 | if ((syntax & RE_BK_PLUS_QM) |
| 2624 | || (syntax & RE_LIMITED_OPS)) |
| 2625 | goto normal_char; |
| 2626 | handle_plus: |
| 2627 | case '*': |
| 2628 | /* If there is no previous pattern... */ |
| 2629 | if (!laststart) |
| 2630 | { |
| 2631 | if (syntax & RE_CONTEXT_INVALID_OPS) |
| 2632 | FREE_STACK_RETURN (REG_BADRPT); |
| 2633 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) |
| 2634 | goto normal_char; |
| 2635 | } |
| 2636 | |
| 2637 | { |
| 2638 | /* 1 means zero (many) matches is allowed. */ |
| 2639 | boolean zero_times_ok = 0, many_times_ok = 0; |
| 2640 | boolean greedy = 1; |
| 2641 | |
| 2642 | /* If there is a sequence of repetition chars, collapse it |
| 2643 | down to just one (the right one). We can't combine |
| 2644 | interval operators with these because of, e.g., `a{2}*', |
| 2645 | which should only match an even number of `a's. */ |
| 2646 | |
| 2647 | for (;;) |
| 2648 | { |
| 2649 | if ((syntax & RE_FRUGAL) |
| 2650 | && c == '?' && (zero_times_ok || many_times_ok)) |
| 2651 | greedy = 0; |
| 2652 | else |
| 2653 | { |
| 2654 | zero_times_ok |= c != '+'; |
| 2655 | many_times_ok |= c != '?'; |
| 2656 | } |
| 2657 | |
| 2658 | if (p == pend) |
| 2659 | break; |
| 2660 | else if (*p == '*' |
| 2661 | || (!(syntax & RE_BK_PLUS_QM) |
| 2662 | && (*p == '+' || *p == '?'))) |
| 2663 | ; |
| 2664 | else if (syntax & RE_BK_PLUS_QM && *p == '\\') |
| 2665 | { |
| 2666 | if (p+1 == pend) |
| 2667 | FREE_STACK_RETURN (REG_EESCAPE); |
| 2668 | if (p[1] == '+' || p[1] == '?') |
| 2669 | PATFETCH (c); /* Gobble up the backslash. */ |
| 2670 | else |
| 2671 | break; |
| 2672 | } |
| 2673 | else |
| 2674 | break; |
| 2675 | /* If we get here, we found another repeat character. */ |
| 2676 | PATFETCH (c); |
| 2677 | } |
| 2678 | |
| 2679 | /* Star, etc. applied to an empty pattern is equivalent |
| 2680 | to an empty pattern. */ |
| 2681 | if (!laststart || laststart == b) |
| 2682 | break; |
| 2683 | |
| 2684 | /* Now we know whether or not zero matches is allowed |
| 2685 | and also whether or not two or more matches is allowed. */ |
| 2686 | if (greedy) |
| 2687 | { |
| 2688 | if (many_times_ok) |
| 2689 | { |
| 2690 | boolean simple = skip_one_char (laststart) == b; |
| 2691 | size_t startoffset = 0; |
| 2692 | re_opcode_t ofj = |
| 2693 | /* Check if the loop can match the empty string. */ |
| 2694 | (simple || !analyse_first (laststart, b, NULL, 0)) |
| 2695 | ? on_failure_jump : on_failure_jump_loop; |
| 2696 | assert (skip_one_char (laststart) <= b); |
| 2697 | |
| 2698 | if (!zero_times_ok && simple) |
| 2699 | { /* Since simple * loops can be made faster by using |
| 2700 | on_failure_keep_string_jump, we turn simple P+ |
| 2701 | into PP* if P is simple. */ |
| 2702 | unsigned char *p1, *p2; |
| 2703 | startoffset = b - laststart; |
| 2704 | GET_BUFFER_SPACE (startoffset); |
| 2705 | p1 = b; p2 = laststart; |
| 2706 | while (p2 < p1) |
| 2707 | *b++ = *p2++; |
| 2708 | zero_times_ok = 1; |
| 2709 | } |
| 2710 | |
| 2711 | GET_BUFFER_SPACE (6); |
| 2712 | if (!zero_times_ok) |
| 2713 | /* A + loop. */ |
| 2714 | STORE_JUMP (ofj, b, b + 6); |
| 2715 | else |
| 2716 | /* Simple * loops can use on_failure_keep_string_jump |
| 2717 | depending on what follows. But since we don't know |
| 2718 | that yet, we leave the decision up to |
| 2719 | on_failure_jump_smart. */ |
| 2720 | INSERT_JUMP (simple ? on_failure_jump_smart : ofj, |
| 2721 | laststart + startoffset, b + 6); |
| 2722 | b += 3; |
| 2723 | STORE_JUMP (jump, b, laststart + startoffset); |
| 2724 | b += 3; |
| 2725 | } |
| 2726 | else |
| 2727 | { |
| 2728 | /* A simple ? pattern. */ |
| 2729 | assert (zero_times_ok); |
| 2730 | GET_BUFFER_SPACE (3); |
| 2731 | INSERT_JUMP (on_failure_jump, laststart, b + 3); |
| 2732 | b += 3; |
| 2733 | } |
| 2734 | } |
| 2735 | else /* not greedy */ |
| 2736 | { /* I wish the greedy and non-greedy cases could be merged. */ |
| 2737 | |
| 2738 | GET_BUFFER_SPACE (7); /* We might use less. */ |
| 2739 | if (many_times_ok) |
| 2740 | { |
| 2741 | boolean emptyp = analyse_first (laststart, b, NULL, 0); |
| 2742 | |
| 2743 | /* The non-greedy multiple match looks like |
| 2744 | a repeat..until: we only need a conditional jump |
| 2745 | at the end of the loop. */ |
| 2746 | if (emptyp) BUF_PUSH (no_op); |
| 2747 | STORE_JUMP (emptyp ? on_failure_jump_nastyloop |
| 2748 | : on_failure_jump, b, laststart); |
| 2749 | b += 3; |
| 2750 | if (zero_times_ok) |
| 2751 | { |
| 2752 | /* The repeat...until naturally matches one or more. |
| 2753 | To also match zero times, we need to first jump to |
| 2754 | the end of the loop (its conditional jump). */ |
| 2755 | INSERT_JUMP (jump, laststart, b); |
| 2756 | b += 3; |
| 2757 | } |
| 2758 | } |
| 2759 | else |
| 2760 | { |
| 2761 | /* non-greedy a?? */ |
| 2762 | INSERT_JUMP (jump, laststart, b + 3); |
| 2763 | b += 3; |
| 2764 | INSERT_JUMP (on_failure_jump, laststart, laststart + 6); |
| 2765 | b += 3; |
| 2766 | } |
| 2767 | } |
| 2768 | } |
| 2769 | pending_exact = 0; |
| 2770 | break; |
| 2771 | |
| 2772 | |
| 2773 | case '.': |
| 2774 | laststart = b; |
| 2775 | BUF_PUSH (anychar); |
| 2776 | break; |
| 2777 | |
| 2778 | |
| 2779 | case '[': |
| 2780 | { |
| 2781 | re_char *p1; |
| 2782 | |
| 2783 | CLEAR_RANGE_TABLE_WORK_USED (range_table_work); |
| 2784 | |
| 2785 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
| 2786 | |
| 2787 | /* Ensure that we have enough space to push a charset: the |
| 2788 | opcode, the length count, and the bitset; 34 bytes in all. */ |
| 2789 | GET_BUFFER_SPACE (34); |
| 2790 | |
| 2791 | laststart = b; |
| 2792 | |
| 2793 | /* We test `*p == '^' twice, instead of using an if |
| 2794 | statement, so we only need one BUF_PUSH. */ |
| 2795 | BUF_PUSH (*p == '^' ? charset_not : charset); |
| 2796 | if (*p == '^') |
| 2797 | p++; |
| 2798 | |
| 2799 | /* Remember the first position in the bracket expression. */ |
| 2800 | p1 = p; |
| 2801 | |
| 2802 | /* Push the number of bytes in the bitmap. */ |
| 2803 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); |
| 2804 | |
| 2805 | /* Clear the whole map. */ |
| 2806 | memset (b, 0, (1 << BYTEWIDTH) / BYTEWIDTH); |
| 2807 | |
| 2808 | /* charset_not matches newline according to a syntax bit. */ |
| 2809 | if ((re_opcode_t) b[-2] == charset_not |
| 2810 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) |
| 2811 | SET_LIST_BIT ('\n'); |
| 2812 | |
| 2813 | /* Read in characters and ranges, setting map bits. */ |
| 2814 | for (;;) |
| 2815 | { |
| 2816 | boolean escaped_char = false; |
| 2817 | const unsigned char *p2 = p; |
| 2818 | re_wchar_t ch; |
| 2819 | |
| 2820 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
| 2821 | |
| 2822 | /* Don't translate yet. The range TRANSLATE(X..Y) cannot |
| 2823 | always be determined from TRANSLATE(X) and TRANSLATE(Y) |
| 2824 | So the translation is done later in a loop. Example: |
| 2825 | (let ((case-fold-search t)) (string-match "[A-_]" "A")) */ |
| 2826 | PATFETCH (c); |
| 2827 | |
| 2828 | /* \ might escape characters inside [...] and [^...]. */ |
| 2829 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') |
| 2830 | { |
| 2831 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
| 2832 | |
| 2833 | PATFETCH (c); |
| 2834 | escaped_char = true; |
| 2835 | } |
| 2836 | else |
| 2837 | { |
| 2838 | /* Could be the end of the bracket expression. If it's |
| 2839 | not (i.e., when the bracket expression is `[]' so |
| 2840 | far), the ']' character bit gets set way below. */ |
| 2841 | if (c == ']' && p2 != p1) |
| 2842 | break; |
| 2843 | } |
| 2844 | |
| 2845 | /* See if we're at the beginning of a possible character |
| 2846 | class. */ |
| 2847 | |
| 2848 | if (!escaped_char && |
| 2849 | syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') |
| 2850 | { |
| 2851 | /* Leave room for the null. */ |
| 2852 | unsigned char str[CHAR_CLASS_MAX_LENGTH + 1]; |
| 2853 | const unsigned char *class_beg; |
| 2854 | |
| 2855 | PATFETCH (c); |
| 2856 | c1 = 0; |
| 2857 | class_beg = p; |
| 2858 | |
| 2859 | /* If pattern is `[[:'. */ |
| 2860 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
| 2861 | |
| 2862 | for (;;) |
| 2863 | { |
| 2864 | PATFETCH (c); |
| 2865 | if ((c == ':' && *p == ']') || p == pend) |
| 2866 | break; |
| 2867 | if (c1 < CHAR_CLASS_MAX_LENGTH) |
| 2868 | str[c1++] = c; |
| 2869 | else |
| 2870 | /* This is in any case an invalid class name. */ |
| 2871 | str[0] = '\0'; |
| 2872 | } |
| 2873 | str[c1] = '\0'; |
| 2874 | |
| 2875 | /* If isn't a word bracketed by `[:' and `:]': |
| 2876 | undo the ending character, the letters, and |
| 2877 | leave the leading `:' and `[' (but set bits for |
| 2878 | them). */ |
| 2879 | if (c == ':' && *p == ']') |
| 2880 | { |
| 2881 | re_wctype_t cc = re_wctype (str); |
| 2882 | |
| 2883 | if (cc == 0) |
| 2884 | FREE_STACK_RETURN (REG_ECTYPE); |
| 2885 | |
| 2886 | /* Throw away the ] at the end of the character |
| 2887 | class. */ |
| 2888 | PATFETCH (c); |
| 2889 | |
| 2890 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
| 2891 | |
| 2892 | #ifndef emacs |
| 2893 | for (ch = 0; ch < (1 << BYTEWIDTH); ++ch) |
| 2894 | if (re_iswctype (btowc (ch), cc)) |
| 2895 | { |
| 2896 | c = TRANSLATE (ch); |
| 2897 | if (c < (1 << BYTEWIDTH)) |
| 2898 | SET_LIST_BIT (c); |
| 2899 | } |
| 2900 | #else /* emacs */ |
| 2901 | /* Most character classes in a multibyte match |
| 2902 | just set a flag. Exceptions are is_blank, |
| 2903 | is_digit, is_cntrl, and is_xdigit, since |
| 2904 | they can only match ASCII characters. We |
| 2905 | don't need to handle them for multibyte. |
| 2906 | They are distinguished by a negative wctype. */ |
| 2907 | |
| 2908 | /* Setup the gl_state object to its buffer-defined |
| 2909 | value. This hardcodes the buffer-global |
| 2910 | syntax-table for ASCII chars, while the other chars |
| 2911 | will obey syntax-table properties. It's not ideal, |
| 2912 | but it's the way it's been done until now. */ |
| 2913 | SETUP_BUFFER_SYNTAX_TABLE (); |
| 2914 | |
| 2915 | for (ch = 0; ch < 256; ++ch) |
| 2916 | { |
| 2917 | c = RE_CHAR_TO_MULTIBYTE (ch); |
| 2918 | if (! CHAR_BYTE8_P (c) |
| 2919 | && re_iswctype (c, cc)) |
| 2920 | { |
| 2921 | SET_LIST_BIT (ch); |
| 2922 | c1 = TRANSLATE (c); |
| 2923 | if (c1 == c) |
| 2924 | continue; |
| 2925 | if (ASCII_CHAR_P (c1)) |
| 2926 | SET_LIST_BIT (c1); |
| 2927 | else if ((c1 = RE_CHAR_TO_UNIBYTE (c1)) >= 0) |
| 2928 | SET_LIST_BIT (c1); |
| 2929 | } |
| 2930 | } |
| 2931 | SET_RANGE_TABLE_WORK_AREA_BIT |
| 2932 | (range_table_work, re_wctype_to_bit (cc)); |
| 2933 | #endif /* emacs */ |
| 2934 | /* In most cases the matching rule for char classes |
| 2935 | only uses the syntax table for multibyte chars, |
| 2936 | so that the content of the syntax-table it is not |
| 2937 | hardcoded in the range_table. SPACE and WORD are |
| 2938 | the two exceptions. */ |
| 2939 | if ((1 << cc) & ((1 << RECC_SPACE) | (1 << RECC_WORD))) |
| 2940 | bufp->used_syntax = 1; |
| 2941 | |
| 2942 | /* Repeat the loop. */ |
| 2943 | continue; |
| 2944 | } |
| 2945 | else |
| 2946 | { |
| 2947 | /* Go back to right after the "[:". */ |
| 2948 | p = class_beg; |
| 2949 | SET_LIST_BIT ('['); |
| 2950 | |
| 2951 | /* Because the `:' may starts the range, we |
| 2952 | can't simply set bit and repeat the loop. |
| 2953 | Instead, just set it to C and handle below. */ |
| 2954 | c = ':'; |
| 2955 | } |
| 2956 | } |
| 2957 | |
| 2958 | if (p < pend && p[0] == '-' && p[1] != ']') |
| 2959 | { |
| 2960 | |
| 2961 | /* Discard the `-'. */ |
| 2962 | PATFETCH (c1); |
| 2963 | |
| 2964 | /* Fetch the character which ends the range. */ |
| 2965 | PATFETCH (c1); |
| 2966 | #ifdef emacs |
| 2967 | if (CHAR_BYTE8_P (c1) |
| 2968 | && ! ASCII_CHAR_P (c) && ! CHAR_BYTE8_P (c)) |
| 2969 | /* Treat the range from a multibyte character to |
| 2970 | raw-byte character as empty. */ |
| 2971 | c = c1 + 1; |
| 2972 | #endif /* emacs */ |
| 2973 | } |
| 2974 | else |
| 2975 | /* Range from C to C. */ |
| 2976 | c1 = c; |
| 2977 | |
| 2978 | if (c > c1) |
| 2979 | { |
| 2980 | if (syntax & RE_NO_EMPTY_RANGES) |
| 2981 | FREE_STACK_RETURN (REG_ERANGEX); |
| 2982 | /* Else, repeat the loop. */ |
| 2983 | } |
| 2984 | else |
| 2985 | { |
| 2986 | #ifndef emacs |
| 2987 | /* Set the range into bitmap */ |
| 2988 | for (; c <= c1; c++) |
| 2989 | { |
| 2990 | ch = TRANSLATE (c); |
| 2991 | if (ch < (1 << BYTEWIDTH)) |
| 2992 | SET_LIST_BIT (ch); |
| 2993 | } |
| 2994 | #else /* emacs */ |
| 2995 | if (c < 128) |
| 2996 | { |
| 2997 | ch = MIN (127, c1); |
| 2998 | SETUP_ASCII_RANGE (range_table_work, c, ch); |
| 2999 | c = ch + 1; |
| 3000 | if (CHAR_BYTE8_P (c1)) |
| 3001 | c = BYTE8_TO_CHAR (128); |
| 3002 | } |
| 3003 | if (c <= c1) |
| 3004 | { |
| 3005 | if (CHAR_BYTE8_P (c)) |
| 3006 | { |
| 3007 | c = CHAR_TO_BYTE8 (c); |
| 3008 | c1 = CHAR_TO_BYTE8 (c1); |
| 3009 | for (; c <= c1; c++) |
| 3010 | SET_LIST_BIT (c); |
| 3011 | } |
| 3012 | else if (multibyte) |
| 3013 | { |
| 3014 | SETUP_MULTIBYTE_RANGE (range_table_work, c, c1); |
| 3015 | } |
| 3016 | else |
| 3017 | { |
| 3018 | SETUP_UNIBYTE_RANGE (range_table_work, c, c1); |
| 3019 | } |
| 3020 | } |
| 3021 | #endif /* emacs */ |
| 3022 | } |
| 3023 | } |
| 3024 | |
| 3025 | /* Discard any (non)matching list bytes that are all 0 at the |
| 3026 | end of the map. Decrease the map-length byte too. */ |
| 3027 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) |
| 3028 | b[-1]--; |
| 3029 | b += b[-1]; |
| 3030 | |
| 3031 | /* Build real range table from work area. */ |
| 3032 | if (RANGE_TABLE_WORK_USED (range_table_work) |
| 3033 | || RANGE_TABLE_WORK_BITS (range_table_work)) |
| 3034 | { |
| 3035 | int i; |
| 3036 | int used = RANGE_TABLE_WORK_USED (range_table_work); |
| 3037 | |
| 3038 | /* Allocate space for COUNT + RANGE_TABLE. Needs two |
| 3039 | bytes for flags, two for COUNT, and three bytes for |
| 3040 | each character. */ |
| 3041 | GET_BUFFER_SPACE (4 + used * 3); |
| 3042 | |
| 3043 | /* Indicate the existence of range table. */ |
| 3044 | laststart[1] |= 0x80; |
| 3045 | |
| 3046 | /* Store the character class flag bits into the range table. |
| 3047 | If not in emacs, these flag bits are always 0. */ |
| 3048 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff; |
| 3049 | *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8; |
| 3050 | |
| 3051 | STORE_NUMBER_AND_INCR (b, used / 2); |
| 3052 | for (i = 0; i < used; i++) |
| 3053 | STORE_CHARACTER_AND_INCR |
| 3054 | (b, RANGE_TABLE_WORK_ELT (range_table_work, i)); |
| 3055 | } |
| 3056 | } |
| 3057 | break; |
| 3058 | |
| 3059 | |
| 3060 | case '(': |
| 3061 | if (syntax & RE_NO_BK_PARENS) |
| 3062 | goto handle_open; |
| 3063 | else |
| 3064 | goto normal_char; |
| 3065 | |
| 3066 | |
| 3067 | case ')': |
| 3068 | if (syntax & RE_NO_BK_PARENS) |
| 3069 | goto handle_close; |
| 3070 | else |
| 3071 | goto normal_char; |
| 3072 | |
| 3073 | |
| 3074 | case '\n': |
| 3075 | if (syntax & RE_NEWLINE_ALT) |
| 3076 | goto handle_alt; |
| 3077 | else |
| 3078 | goto normal_char; |
| 3079 | |
| 3080 | |
| 3081 | case '|': |
| 3082 | if (syntax & RE_NO_BK_VBAR) |
| 3083 | goto handle_alt; |
| 3084 | else |
| 3085 | goto normal_char; |
| 3086 | |
| 3087 | |
| 3088 | case '{': |
| 3089 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) |
| 3090 | goto handle_interval; |
| 3091 | else |
| 3092 | goto normal_char; |
| 3093 | |
| 3094 | |
| 3095 | case '\\': |
| 3096 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
| 3097 | |
| 3098 | /* Do not translate the character after the \, so that we can |
| 3099 | distinguish, e.g., \B from \b, even if we normally would |
| 3100 | translate, e.g., B to b. */ |
| 3101 | PATFETCH (c); |
| 3102 | |
| 3103 | switch (c) |
| 3104 | { |
| 3105 | case '(': |
| 3106 | if (syntax & RE_NO_BK_PARENS) |
| 3107 | goto normal_backslash; |
| 3108 | |
| 3109 | handle_open: |
| 3110 | { |
| 3111 | int shy = 0; |
| 3112 | regnum_t regnum = 0; |
| 3113 | if (p+1 < pend) |
| 3114 | { |
| 3115 | /* Look for a special (?...) construct */ |
| 3116 | if ((syntax & RE_SHY_GROUPS) && *p == '?') |
| 3117 | { |
| 3118 | PATFETCH (c); /* Gobble up the '?'. */ |
| 3119 | while (!shy) |
| 3120 | { |
| 3121 | PATFETCH (c); |
| 3122 | switch (c) |
| 3123 | { |
| 3124 | case ':': shy = 1; break; |
| 3125 | case '0': |
| 3126 | /* An explicitly specified regnum must start |
| 3127 | with non-0. */ |
| 3128 | if (regnum == 0) |
| 3129 | FREE_STACK_RETURN (REG_BADPAT); |
| 3130 | case '1': case '2': case '3': case '4': |
| 3131 | case '5': case '6': case '7': case '8': case '9': |
| 3132 | regnum = 10*regnum + (c - '0'); break; |
| 3133 | default: |
| 3134 | /* Only (?:...) is supported right now. */ |
| 3135 | FREE_STACK_RETURN (REG_BADPAT); |
| 3136 | } |
| 3137 | } |
| 3138 | } |
| 3139 | } |
| 3140 | |
| 3141 | if (!shy) |
| 3142 | regnum = ++bufp->re_nsub; |
| 3143 | else if (regnum) |
| 3144 | { /* It's actually not shy, but explicitly numbered. */ |
| 3145 | shy = 0; |
| 3146 | if (regnum > bufp->re_nsub) |
| 3147 | bufp->re_nsub = regnum; |
| 3148 | else if (regnum > bufp->re_nsub |
| 3149 | /* Ideally, we'd want to check that the specified |
| 3150 | group can't have matched (i.e. all subgroups |
| 3151 | using the same regnum are in other branches of |
| 3152 | OR patterns), but we don't currently keep track |
| 3153 | of enough info to do that easily. */ |
| 3154 | || group_in_compile_stack (compile_stack, regnum)) |
| 3155 | FREE_STACK_RETURN (REG_BADPAT); |
| 3156 | } |
| 3157 | else |
| 3158 | /* It's really shy. */ |
| 3159 | regnum = - bufp->re_nsub; |
| 3160 | |
| 3161 | if (COMPILE_STACK_FULL) |
| 3162 | { |
| 3163 | RETALLOC (compile_stack.stack, compile_stack.size << 1, |
| 3164 | compile_stack_elt_t); |
| 3165 | if (compile_stack.stack == NULL) return REG_ESPACE; |
| 3166 | |
| 3167 | compile_stack.size <<= 1; |
| 3168 | } |
| 3169 | |
| 3170 | /* These are the values to restore when we hit end of this |
| 3171 | group. They are all relative offsets, so that if the |
| 3172 | whole pattern moves because of realloc, they will still |
| 3173 | be valid. */ |
| 3174 | COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; |
| 3175 | COMPILE_STACK_TOP.fixup_alt_jump |
| 3176 | = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; |
| 3177 | COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; |
| 3178 | COMPILE_STACK_TOP.regnum = regnum; |
| 3179 | |
| 3180 | /* Do not push a start_memory for groups beyond the last one |
| 3181 | we can represent in the compiled pattern. */ |
| 3182 | if (regnum <= MAX_REGNUM && regnum > 0) |
| 3183 | BUF_PUSH_2 (start_memory, regnum); |
| 3184 | |
| 3185 | compile_stack.avail++; |
| 3186 | |
| 3187 | fixup_alt_jump = 0; |
| 3188 | laststart = 0; |
| 3189 | begalt = b; |
| 3190 | /* If we've reached MAX_REGNUM groups, then this open |
| 3191 | won't actually generate any code, so we'll have to |
| 3192 | clear pending_exact explicitly. */ |
| 3193 | pending_exact = 0; |
| 3194 | break; |
| 3195 | } |
| 3196 | |
| 3197 | case ')': |
| 3198 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; |
| 3199 | |
| 3200 | if (COMPILE_STACK_EMPTY) |
| 3201 | { |
| 3202 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) |
| 3203 | goto normal_backslash; |
| 3204 | else |
| 3205 | FREE_STACK_RETURN (REG_ERPAREN); |
| 3206 | } |
| 3207 | |
| 3208 | handle_close: |
| 3209 | FIXUP_ALT_JUMP (); |
| 3210 | |
| 3211 | /* See similar code for backslashed left paren above. */ |
| 3212 | if (COMPILE_STACK_EMPTY) |
| 3213 | { |
| 3214 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) |
| 3215 | goto normal_char; |
| 3216 | else |
| 3217 | FREE_STACK_RETURN (REG_ERPAREN); |
| 3218 | } |
| 3219 | |
| 3220 | /* Since we just checked for an empty stack above, this |
| 3221 | ``can't happen''. */ |
| 3222 | assert (compile_stack.avail != 0); |
| 3223 | { |
| 3224 | /* We don't just want to restore into `regnum', because |
| 3225 | later groups should continue to be numbered higher, |
| 3226 | as in `(ab)c(de)' -- the second group is #2. */ |
| 3227 | regnum_t regnum; |
| 3228 | |
| 3229 | compile_stack.avail--; |
| 3230 | begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset; |
| 3231 | fixup_alt_jump |
| 3232 | = COMPILE_STACK_TOP.fixup_alt_jump |
| 3233 | ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1 |
| 3234 | : 0; |
| 3235 | laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset; |
| 3236 | regnum = COMPILE_STACK_TOP.regnum; |
| 3237 | /* If we've reached MAX_REGNUM groups, then this open |
| 3238 | won't actually generate any code, so we'll have to |
| 3239 | clear pending_exact explicitly. */ |
| 3240 | pending_exact = 0; |
| 3241 | |
| 3242 | /* We're at the end of the group, so now we know how many |
| 3243 | groups were inside this one. */ |
| 3244 | if (regnum <= MAX_REGNUM && regnum > 0) |
| 3245 | BUF_PUSH_2 (stop_memory, regnum); |
| 3246 | } |
| 3247 | break; |
| 3248 | |
| 3249 | |
| 3250 | case '|': /* `\|'. */ |
| 3251 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) |
| 3252 | goto normal_backslash; |
| 3253 | handle_alt: |
| 3254 | if (syntax & RE_LIMITED_OPS) |
| 3255 | goto normal_char; |
| 3256 | |
| 3257 | /* Insert before the previous alternative a jump which |
| 3258 | jumps to this alternative if the former fails. */ |
| 3259 | GET_BUFFER_SPACE (3); |
| 3260 | INSERT_JUMP (on_failure_jump, begalt, b + 6); |
| 3261 | pending_exact = 0; |
| 3262 | b += 3; |
| 3263 | |
| 3264 | /* The alternative before this one has a jump after it |
| 3265 | which gets executed if it gets matched. Adjust that |
| 3266 | jump so it will jump to this alternative's analogous |
| 3267 | jump (put in below, which in turn will jump to the next |
| 3268 | (if any) alternative's such jump, etc.). The last such |
| 3269 | jump jumps to the correct final destination. A picture: |
| 3270 | _____ _____ |
| 3271 | | | | | |
| 3272 | | v | v |
| 3273 | a | b | c |
| 3274 | |
| 3275 | If we are at `b', then fixup_alt_jump right now points to a |
| 3276 | three-byte space after `a'. We'll put in the jump, set |
| 3277 | fixup_alt_jump to right after `b', and leave behind three |
| 3278 | bytes which we'll fill in when we get to after `c'. */ |
| 3279 | |
| 3280 | FIXUP_ALT_JUMP (); |
| 3281 | |
| 3282 | /* Mark and leave space for a jump after this alternative, |
| 3283 | to be filled in later either by next alternative or |
| 3284 | when know we're at the end of a series of alternatives. */ |
| 3285 | fixup_alt_jump = b; |
| 3286 | GET_BUFFER_SPACE (3); |
| 3287 | b += 3; |
| 3288 | |
| 3289 | laststart = 0; |
| 3290 | begalt = b; |
| 3291 | break; |
| 3292 | |
| 3293 | |
| 3294 | case '{': |
| 3295 | /* If \{ is a literal. */ |
| 3296 | if (!(syntax & RE_INTERVALS) |
| 3297 | /* If we're at `\{' and it's not the open-interval |
| 3298 | operator. */ |
| 3299 | || (syntax & RE_NO_BK_BRACES)) |
| 3300 | goto normal_backslash; |
| 3301 | |
| 3302 | handle_interval: |
| 3303 | { |
| 3304 | /* If got here, then the syntax allows intervals. */ |
| 3305 | |
| 3306 | /* At least (most) this many matches must be made. */ |
| 3307 | int lower_bound = 0, upper_bound = -1; |
| 3308 | |
| 3309 | beg_interval = p; |
| 3310 | |
| 3311 | GET_UNSIGNED_NUMBER (lower_bound); |
| 3312 | |
| 3313 | if (c == ',') |
| 3314 | GET_UNSIGNED_NUMBER (upper_bound); |
| 3315 | else |
| 3316 | /* Interval such as `{1}' => match exactly once. */ |
| 3317 | upper_bound = lower_bound; |
| 3318 | |
| 3319 | if (lower_bound < 0 || upper_bound > RE_DUP_MAX |
| 3320 | || (upper_bound >= 0 && lower_bound > upper_bound)) |
| 3321 | FREE_STACK_RETURN (REG_BADBR); |
| 3322 | |
| 3323 | if (!(syntax & RE_NO_BK_BRACES)) |
| 3324 | { |
| 3325 | if (c != '\\') |
| 3326 | FREE_STACK_RETURN (REG_BADBR); |
| 3327 | if (p == pend) |
| 3328 | FREE_STACK_RETURN (REG_EESCAPE); |
| 3329 | PATFETCH (c); |
| 3330 | } |
| 3331 | |
| 3332 | if (c != '}') |
| 3333 | FREE_STACK_RETURN (REG_BADBR); |
| 3334 | |
| 3335 | /* We just parsed a valid interval. */ |
| 3336 | |
| 3337 | /* If it's invalid to have no preceding re. */ |
| 3338 | if (!laststart) |
| 3339 | { |
| 3340 | if (syntax & RE_CONTEXT_INVALID_OPS) |
| 3341 | FREE_STACK_RETURN (REG_BADRPT); |
| 3342 | else if (syntax & RE_CONTEXT_INDEP_OPS) |
| 3343 | laststart = b; |
| 3344 | else |
| 3345 | goto unfetch_interval; |
| 3346 | } |
| 3347 | |
| 3348 | if (upper_bound == 0) |
| 3349 | /* If the upper bound is zero, just drop the sub pattern |
| 3350 | altogether. */ |
| 3351 | b = laststart; |
| 3352 | else if (lower_bound == 1 && upper_bound == 1) |
| 3353 | /* Just match it once: nothing to do here. */ |
| 3354 | ; |
| 3355 | |
| 3356 | /* Otherwise, we have a nontrivial interval. When |
| 3357 | we're all done, the pattern will look like: |
| 3358 | set_number_at <jump count> <upper bound> |
| 3359 | set_number_at <succeed_n count> <lower bound> |
| 3360 | succeed_n <after jump addr> <succeed_n count> |
| 3361 | <body of loop> |
| 3362 | jump_n <succeed_n addr> <jump count> |
| 3363 | (The upper bound and `jump_n' are omitted if |
| 3364 | `upper_bound' is 1, though.) */ |
| 3365 | else |
| 3366 | { /* If the upper bound is > 1, we need to insert |
| 3367 | more at the end of the loop. */ |
| 3368 | unsigned int nbytes = (upper_bound < 0 ? 3 |
| 3369 | : upper_bound > 1 ? 5 : 0); |
| 3370 | unsigned int startoffset = 0; |
| 3371 | |
| 3372 | GET_BUFFER_SPACE (20); /* We might use less. */ |
| 3373 | |
| 3374 | if (lower_bound == 0) |
| 3375 | { |
| 3376 | /* A succeed_n that starts with 0 is really a |
| 3377 | a simple on_failure_jump_loop. */ |
| 3378 | INSERT_JUMP (on_failure_jump_loop, laststart, |
| 3379 | b + 3 + nbytes); |
| 3380 | b += 3; |
| 3381 | } |
| 3382 | else |
| 3383 | { |
| 3384 | /* Initialize lower bound of the `succeed_n', even |
| 3385 | though it will be set during matching by its |
| 3386 | attendant `set_number_at' (inserted next), |
| 3387 | because `re_compile_fastmap' needs to know. |
| 3388 | Jump to the `jump_n' we might insert below. */ |
| 3389 | INSERT_JUMP2 (succeed_n, laststart, |
| 3390 | b + 5 + nbytes, |
| 3391 | lower_bound); |
| 3392 | b += 5; |
| 3393 | |
| 3394 | /* Code to initialize the lower bound. Insert |
| 3395 | before the `succeed_n'. The `5' is the last two |
| 3396 | bytes of this `set_number_at', plus 3 bytes of |
| 3397 | the following `succeed_n'. */ |
| 3398 | insert_op2 (set_number_at, laststart, 5, lower_bound, b); |
| 3399 | b += 5; |
| 3400 | startoffset += 5; |
| 3401 | } |
| 3402 | |
| 3403 | if (upper_bound < 0) |
| 3404 | { |
| 3405 | /* A negative upper bound stands for infinity, |
| 3406 | in which case it degenerates to a plain jump. */ |
| 3407 | STORE_JUMP (jump, b, laststart + startoffset); |
| 3408 | b += 3; |
| 3409 | } |
| 3410 | else if (upper_bound > 1) |
| 3411 | { /* More than one repetition is allowed, so |
| 3412 | append a backward jump to the `succeed_n' |
| 3413 | that starts this interval. |
| 3414 | |
| 3415 | When we've reached this during matching, |
| 3416 | we'll have matched the interval once, so |
| 3417 | jump back only `upper_bound - 1' times. */ |
| 3418 | STORE_JUMP2 (jump_n, b, laststart + startoffset, |
| 3419 | upper_bound - 1); |
| 3420 | b += 5; |
| 3421 | |
| 3422 | /* The location we want to set is the second |
| 3423 | parameter of the `jump_n'; that is `b-2' as |
| 3424 | an absolute address. `laststart' will be |
| 3425 | the `set_number_at' we're about to insert; |
| 3426 | `laststart+3' the number to set, the source |
| 3427 | for the relative address. But we are |
| 3428 | inserting into the middle of the pattern -- |
| 3429 | so everything is getting moved up by 5. |
| 3430 | Conclusion: (b - 2) - (laststart + 3) + 5, |
| 3431 | i.e., b - laststart. |
| 3432 | |
| 3433 | We insert this at the beginning of the loop |
| 3434 | so that if we fail during matching, we'll |
| 3435 | reinitialize the bounds. */ |
| 3436 | insert_op2 (set_number_at, laststart, b - laststart, |
| 3437 | upper_bound - 1, b); |
| 3438 | b += 5; |
| 3439 | } |
| 3440 | } |
| 3441 | pending_exact = 0; |
| 3442 | beg_interval = NULL; |
| 3443 | } |
| 3444 | break; |
| 3445 | |
| 3446 | unfetch_interval: |
| 3447 | /* If an invalid interval, match the characters as literals. */ |
| 3448 | assert (beg_interval); |
| 3449 | p = beg_interval; |
| 3450 | beg_interval = NULL; |
| 3451 | |
| 3452 | /* normal_char and normal_backslash need `c'. */ |
| 3453 | c = '{'; |
| 3454 | |
| 3455 | if (!(syntax & RE_NO_BK_BRACES)) |
| 3456 | { |
| 3457 | assert (p > pattern && p[-1] == '\\'); |
| 3458 | goto normal_backslash; |
| 3459 | } |
| 3460 | else |
| 3461 | goto normal_char; |
| 3462 | |
| 3463 | #ifdef emacs |
| 3464 | /* There is no way to specify the before_dot and after_dot |
| 3465 | operators. rms says this is ok. --karl */ |
| 3466 | case '=': |
| 3467 | laststart = b; |
| 3468 | BUF_PUSH (at_dot); |
| 3469 | break; |
| 3470 | |
| 3471 | case 's': |
| 3472 | laststart = b; |
| 3473 | PATFETCH (c); |
| 3474 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); |
| 3475 | break; |
| 3476 | |
| 3477 | case 'S': |
| 3478 | laststart = b; |
| 3479 | PATFETCH (c); |
| 3480 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); |
| 3481 | break; |
| 3482 | |
| 3483 | case 'c': |
| 3484 | laststart = b; |
| 3485 | PATFETCH (c); |
| 3486 | BUF_PUSH_2 (categoryspec, c); |
| 3487 | break; |
| 3488 | |
| 3489 | case 'C': |
| 3490 | laststart = b; |
| 3491 | PATFETCH (c); |
| 3492 | BUF_PUSH_2 (notcategoryspec, c); |
| 3493 | break; |
| 3494 | #endif /* emacs */ |
| 3495 | |
| 3496 | |
| 3497 | case 'w': |
| 3498 | if (syntax & RE_NO_GNU_OPS) |
| 3499 | goto normal_char; |
| 3500 | laststart = b; |
| 3501 | BUF_PUSH_2 (syntaxspec, Sword); |
| 3502 | break; |
| 3503 | |
| 3504 | |
| 3505 | case 'W': |
| 3506 | if (syntax & RE_NO_GNU_OPS) |
| 3507 | goto normal_char; |
| 3508 | laststart = b; |
| 3509 | BUF_PUSH_2 (notsyntaxspec, Sword); |
| 3510 | break; |
| 3511 | |
| 3512 | |
| 3513 | case '<': |
| 3514 | if (syntax & RE_NO_GNU_OPS) |
| 3515 | goto normal_char; |
| 3516 | laststart = b; |
| 3517 | BUF_PUSH (wordbeg); |
| 3518 | break; |
| 3519 | |
| 3520 | case '>': |
| 3521 | if (syntax & RE_NO_GNU_OPS) |
| 3522 | goto normal_char; |
| 3523 | laststart = b; |
| 3524 | BUF_PUSH (wordend); |
| 3525 | break; |
| 3526 | |
| 3527 | case '_': |
| 3528 | if (syntax & RE_NO_GNU_OPS) |
| 3529 | goto normal_char; |
| 3530 | laststart = b; |
| 3531 | PATFETCH (c); |
| 3532 | if (c == '<') |
| 3533 | BUF_PUSH (symbeg); |
| 3534 | else if (c == '>') |
| 3535 | BUF_PUSH (symend); |
| 3536 | else |
| 3537 | FREE_STACK_RETURN (REG_BADPAT); |
| 3538 | break; |
| 3539 | |
| 3540 | case 'b': |
| 3541 | if (syntax & RE_NO_GNU_OPS) |
| 3542 | goto normal_char; |
| 3543 | BUF_PUSH (wordbound); |
| 3544 | break; |
| 3545 | |
| 3546 | case 'B': |
| 3547 | if (syntax & RE_NO_GNU_OPS) |
| 3548 | goto normal_char; |
| 3549 | BUF_PUSH (notwordbound); |
| 3550 | break; |
| 3551 | |
| 3552 | case '`': |
| 3553 | if (syntax & RE_NO_GNU_OPS) |
| 3554 | goto normal_char; |
| 3555 | BUF_PUSH (begbuf); |
| 3556 | break; |
| 3557 | |
| 3558 | case '\'': |
| 3559 | if (syntax & RE_NO_GNU_OPS) |
| 3560 | goto normal_char; |
| 3561 | BUF_PUSH (endbuf); |
| 3562 | break; |
| 3563 | |
| 3564 | case '1': case '2': case '3': case '4': case '5': |
| 3565 | case '6': case '7': case '8': case '9': |
| 3566 | { |
| 3567 | regnum_t reg; |
| 3568 | |
| 3569 | if (syntax & RE_NO_BK_REFS) |
| 3570 | goto normal_backslash; |
| 3571 | |
| 3572 | reg = c - '0'; |
| 3573 | |
| 3574 | if (reg > bufp->re_nsub || reg < 1 |
| 3575 | /* Can't back reference to a subexp before its end. */ |
| 3576 | || group_in_compile_stack (compile_stack, reg)) |
| 3577 | FREE_STACK_RETURN (REG_ESUBREG); |
| 3578 | |
| 3579 | laststart = b; |
| 3580 | BUF_PUSH_2 (duplicate, reg); |
| 3581 | } |
| 3582 | break; |
| 3583 | |
| 3584 | |
| 3585 | case '+': |
| 3586 | case '?': |
| 3587 | if (syntax & RE_BK_PLUS_QM) |
| 3588 | goto handle_plus; |
| 3589 | else |
| 3590 | goto normal_backslash; |
| 3591 | |
| 3592 | default: |
| 3593 | normal_backslash: |
| 3594 | /* You might think it would be useful for \ to mean |
| 3595 | not to translate; but if we don't translate it |
| 3596 | it will never match anything. */ |
| 3597 | goto normal_char; |
| 3598 | } |
| 3599 | break; |
| 3600 | |
| 3601 | |
| 3602 | default: |
| 3603 | /* Expects the character in `c'. */ |
| 3604 | normal_char: |
| 3605 | /* If no exactn currently being built. */ |
| 3606 | if (!pending_exact |
| 3607 | |
| 3608 | /* If last exactn not at current position. */ |
| 3609 | || pending_exact + *pending_exact + 1 != b |
| 3610 | |
| 3611 | /* We have only one byte following the exactn for the count. */ |
| 3612 | || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH |
| 3613 | |
| 3614 | /* If followed by a repetition operator. */ |
| 3615 | || (p != pend && (*p == '*' || *p == '^')) |
| 3616 | || ((syntax & RE_BK_PLUS_QM) |
| 3617 | ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?') |
| 3618 | : p != pend && (*p == '+' || *p == '?')) |
| 3619 | || ((syntax & RE_INTERVALS) |
| 3620 | && ((syntax & RE_NO_BK_BRACES) |
| 3621 | ? p != pend && *p == '{' |
| 3622 | : p + 1 < pend && p[0] == '\\' && p[1] == '{'))) |
| 3623 | { |
| 3624 | /* Start building a new exactn. */ |
| 3625 | |
| 3626 | laststart = b; |
| 3627 | |
| 3628 | BUF_PUSH_2 (exactn, 0); |
| 3629 | pending_exact = b - 1; |
| 3630 | } |
| 3631 | |
| 3632 | GET_BUFFER_SPACE (MAX_MULTIBYTE_LENGTH); |
| 3633 | { |
| 3634 | int len; |
| 3635 | |
| 3636 | if (multibyte) |
| 3637 | { |
| 3638 | c = TRANSLATE (c); |
| 3639 | len = CHAR_STRING (c, b); |
| 3640 | b += len; |
| 3641 | } |
| 3642 | else |
| 3643 | { |
| 3644 | c1 = RE_CHAR_TO_MULTIBYTE (c); |
| 3645 | if (! CHAR_BYTE8_P (c1)) |
| 3646 | { |
| 3647 | re_wchar_t c2 = TRANSLATE (c1); |
| 3648 | |
| 3649 | if (c1 != c2 && (c1 = RE_CHAR_TO_UNIBYTE (c2)) >= 0) |
| 3650 | c = c1; |
| 3651 | } |
| 3652 | *b++ = c; |
| 3653 | len = 1; |
| 3654 | } |
| 3655 | (*pending_exact) += len; |
| 3656 | } |
| 3657 | |
| 3658 | break; |
| 3659 | } /* switch (c) */ |
| 3660 | } /* while p != pend */ |
| 3661 | |
| 3662 | |
| 3663 | /* Through the pattern now. */ |
| 3664 | |
| 3665 | FIXUP_ALT_JUMP (); |
| 3666 | |
| 3667 | if (!COMPILE_STACK_EMPTY) |
| 3668 | FREE_STACK_RETURN (REG_EPAREN); |
| 3669 | |
| 3670 | /* If we don't want backtracking, force success |
| 3671 | the first time we reach the end of the compiled pattern. */ |
| 3672 | if (syntax & RE_NO_POSIX_BACKTRACKING) |
| 3673 | BUF_PUSH (succeed); |
| 3674 | |
| 3675 | /* We have succeeded; set the length of the buffer. */ |
| 3676 | bufp->used = b - bufp->buffer; |
| 3677 | |
| 3678 | #ifdef DEBUG |
| 3679 | if (debug > 0) |
| 3680 | { |
| 3681 | re_compile_fastmap (bufp); |
| 3682 | DEBUG_PRINT ("\nCompiled pattern: \n"); |
| 3683 | print_compiled_pattern (bufp); |
| 3684 | } |
| 3685 | debug--; |
| 3686 | #endif /* DEBUG */ |
| 3687 | |
| 3688 | #ifndef MATCH_MAY_ALLOCATE |
| 3689 | /* Initialize the failure stack to the largest possible stack. This |
| 3690 | isn't necessary unless we're trying to avoid calling alloca in |
| 3691 | the search and match routines. */ |
| 3692 | { |
| 3693 | int num_regs = bufp->re_nsub + 1; |
| 3694 | |
| 3695 | if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE) |
| 3696 | { |
| 3697 | fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE; |
| 3698 | falk_stack.stack = realloc (fail_stack.stack, |
| 3699 | fail_stack.size * sizeof *falk_stack.stack); |
| 3700 | } |
| 3701 | |
| 3702 | regex_grow_registers (num_regs); |
| 3703 | } |
| 3704 | #endif /* not MATCH_MAY_ALLOCATE */ |
| 3705 | |
| 3706 | FREE_STACK_RETURN (REG_NOERROR); |
| 3707 | } /* regex_compile */ |
| 3708 | \f |
| 3709 | /* Subroutines for `regex_compile'. */ |
| 3710 | |
| 3711 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
| 3712 | |
| 3713 | static void |
| 3714 | store_op1 (re_opcode_t op, unsigned char *loc, int arg) |
| 3715 | { |
| 3716 | *loc = (unsigned char) op; |
| 3717 | STORE_NUMBER (loc + 1, arg); |
| 3718 | } |
| 3719 | |
| 3720 | |
| 3721 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ |
| 3722 | |
| 3723 | static void |
| 3724 | store_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2) |
| 3725 | { |
| 3726 | *loc = (unsigned char) op; |
| 3727 | STORE_NUMBER (loc + 1, arg1); |
| 3728 | STORE_NUMBER (loc + 3, arg2); |
| 3729 | } |
| 3730 | |
| 3731 | |
| 3732 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC |
| 3733 | for OP followed by two-byte integer parameter ARG. */ |
| 3734 | |
| 3735 | static void |
| 3736 | insert_op1 (re_opcode_t op, unsigned char *loc, int arg, unsigned char *end) |
| 3737 | { |
| 3738 | register unsigned char *pfrom = end; |
| 3739 | register unsigned char *pto = end + 3; |
| 3740 | |
| 3741 | while (pfrom != loc) |
| 3742 | *--pto = *--pfrom; |
| 3743 | |
| 3744 | store_op1 (op, loc, arg); |
| 3745 | } |
| 3746 | |
| 3747 | |
| 3748 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ |
| 3749 | |
| 3750 | static void |
| 3751 | insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2, unsigned char *end) |
| 3752 | { |
| 3753 | register unsigned char *pfrom = end; |
| 3754 | register unsigned char *pto = end + 5; |
| 3755 | |
| 3756 | while (pfrom != loc) |
| 3757 | *--pto = *--pfrom; |
| 3758 | |
| 3759 | store_op2 (op, loc, arg1, arg2); |
| 3760 | } |
| 3761 | |
| 3762 | |
| 3763 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes |
| 3764 | after an alternative or a begin-subexpression. We assume there is at |
| 3765 | least one character before the ^. */ |
| 3766 | |
| 3767 | static boolean |
| 3768 | at_begline_loc_p (const re_char *pattern, const re_char *p, reg_syntax_t syntax) |
| 3769 | { |
| 3770 | re_char *prev = p - 2; |
| 3771 | boolean odd_backslashes; |
| 3772 | |
| 3773 | /* After a subexpression? */ |
| 3774 | if (*prev == '(') |
| 3775 | odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0; |
| 3776 | |
| 3777 | /* After an alternative? */ |
| 3778 | else if (*prev == '|') |
| 3779 | odd_backslashes = (syntax & RE_NO_BK_VBAR) == 0; |
| 3780 | |
| 3781 | /* After a shy subexpression? */ |
| 3782 | else if (*prev == ':' && (syntax & RE_SHY_GROUPS)) |
| 3783 | { |
| 3784 | /* Skip over optional regnum. */ |
| 3785 | while (prev - 1 >= pattern && prev[-1] >= '0' && prev[-1] <= '9') |
| 3786 | --prev; |
| 3787 | |
| 3788 | if (!(prev - 2 >= pattern |
| 3789 | && prev[-1] == '?' && prev[-2] == '(')) |
| 3790 | return false; |
| 3791 | prev -= 2; |
| 3792 | odd_backslashes = (syntax & RE_NO_BK_PARENS) == 0; |
| 3793 | } |
| 3794 | else |
| 3795 | return false; |
| 3796 | |
| 3797 | /* Count the number of preceding backslashes. */ |
| 3798 | p = prev; |
| 3799 | while (prev - 1 >= pattern && prev[-1] == '\\') |
| 3800 | --prev; |
| 3801 | return (p - prev) & odd_backslashes; |
| 3802 | } |
| 3803 | |
| 3804 | |
| 3805 | /* The dual of at_begline_loc_p. This one is for $. We assume there is |
| 3806 | at least one character after the $, i.e., `P < PEND'. */ |
| 3807 | |
| 3808 | static boolean |
| 3809 | at_endline_loc_p (const re_char *p, const re_char *pend, reg_syntax_t syntax) |
| 3810 | { |
| 3811 | re_char *next = p; |
| 3812 | boolean next_backslash = *next == '\\'; |
| 3813 | re_char *next_next = p + 1 < pend ? p + 1 : 0; |
| 3814 | |
| 3815 | return |
| 3816 | /* Before a subexpression? */ |
| 3817 | (syntax & RE_NO_BK_PARENS ? *next == ')' |
| 3818 | : next_backslash && next_next && *next_next == ')') |
| 3819 | /* Before an alternative? */ |
| 3820 | || (syntax & RE_NO_BK_VBAR ? *next == '|' |
| 3821 | : next_backslash && next_next && *next_next == '|'); |
| 3822 | } |
| 3823 | |
| 3824 | |
| 3825 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
| 3826 | false if it's not. */ |
| 3827 | |
| 3828 | static boolean |
| 3829 | group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum) |
| 3830 | { |
| 3831 | ssize_t this_element; |
| 3832 | |
| 3833 | for (this_element = compile_stack.avail - 1; |
| 3834 | this_element >= 0; |
| 3835 | this_element--) |
| 3836 | if (compile_stack.stack[this_element].regnum == regnum) |
| 3837 | return true; |
| 3838 | |
| 3839 | return false; |
| 3840 | } |
| 3841 | \f |
| 3842 | /* analyse_first. |
| 3843 | If fastmap is non-NULL, go through the pattern and fill fastmap |
| 3844 | with all the possible leading chars. If fastmap is NULL, don't |
| 3845 | bother filling it up (obviously) and only return whether the |
| 3846 | pattern could potentially match the empty string. |
| 3847 | |
| 3848 | Return 1 if p..pend might match the empty string. |
| 3849 | Return 0 if p..pend matches at least one char. |
| 3850 | Return -1 if fastmap was not updated accurately. */ |
| 3851 | |
| 3852 | static int |
| 3853 | analyse_first (const re_char *p, const re_char *pend, char *fastmap, const int multibyte) |
| 3854 | { |
| 3855 | int j, k; |
| 3856 | boolean not; |
| 3857 | |
| 3858 | /* If all elements for base leading-codes in fastmap is set, this |
| 3859 | flag is set true. */ |
| 3860 | boolean match_any_multibyte_characters = false; |
| 3861 | |
| 3862 | assert (p); |
| 3863 | |
| 3864 | /* The loop below works as follows: |
| 3865 | - It has a working-list kept in the PATTERN_STACK and which basically |
| 3866 | starts by only containing a pointer to the first operation. |
| 3867 | - If the opcode we're looking at is a match against some set of |
| 3868 | chars, then we add those chars to the fastmap and go on to the |
| 3869 | next work element from the worklist (done via `break'). |
| 3870 | - If the opcode is a control operator on the other hand, we either |
| 3871 | ignore it (if it's meaningless at this point, such as `start_memory') |
| 3872 | or execute it (if it's a jump). If the jump has several destinations |
| 3873 | (i.e. `on_failure_jump'), then we push the other destination onto the |
| 3874 | worklist. |
| 3875 | We guarantee termination by ignoring backward jumps (more or less), |
| 3876 | so that `p' is monotonically increasing. More to the point, we |
| 3877 | never set `p' (or push) anything `<= p1'. */ |
| 3878 | |
| 3879 | while (p < pend) |
| 3880 | { |
| 3881 | /* `p1' is used as a marker of how far back a `on_failure_jump' |
| 3882 | can go without being ignored. It is normally equal to `p' |
| 3883 | (which prevents any backward `on_failure_jump') except right |
| 3884 | after a plain `jump', to allow patterns such as: |
| 3885 | 0: jump 10 |
| 3886 | 3..9: <body> |
| 3887 | 10: on_failure_jump 3 |
| 3888 | as used for the *? operator. */ |
| 3889 | re_char *p1 = p; |
| 3890 | |
| 3891 | switch (*p++) |
| 3892 | { |
| 3893 | case succeed: |
| 3894 | return 1; |
| 3895 | |
| 3896 | case duplicate: |
| 3897 | /* If the first character has to match a backreference, that means |
| 3898 | that the group was empty (since it already matched). Since this |
| 3899 | is the only case that interests us here, we can assume that the |
| 3900 | backreference must match the empty string. */ |
| 3901 | p++; |
| 3902 | continue; |
| 3903 | |
| 3904 | |
| 3905 | /* Following are the cases which match a character. These end |
| 3906 | with `break'. */ |
| 3907 | |
| 3908 | case exactn: |
| 3909 | if (fastmap) |
| 3910 | { |
| 3911 | /* If multibyte is nonzero, the first byte of each |
| 3912 | character is an ASCII or a leading code. Otherwise, |
| 3913 | each byte is a character. Thus, this works in both |
| 3914 | cases. */ |
| 3915 | fastmap[p[1]] = 1; |
| 3916 | if (! multibyte) |
| 3917 | { |
| 3918 | /* For the case of matching this unibyte regex |
| 3919 | against multibyte, we must set a leading code of |
| 3920 | the corresponding multibyte character. */ |
| 3921 | int c = RE_CHAR_TO_MULTIBYTE (p[1]); |
| 3922 | |
| 3923 | fastmap[CHAR_LEADING_CODE (c)] = 1; |
| 3924 | } |
| 3925 | } |
| 3926 | break; |
| 3927 | |
| 3928 | |
| 3929 | case anychar: |
| 3930 | /* We could put all the chars except for \n (and maybe \0) |
| 3931 | but we don't bother since it is generally not worth it. */ |
| 3932 | if (!fastmap) break; |
| 3933 | return -1; |
| 3934 | |
| 3935 | |
| 3936 | case charset_not: |
| 3937 | if (!fastmap) break; |
| 3938 | { |
| 3939 | /* Chars beyond end of bitmap are possible matches. */ |
| 3940 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH; |
| 3941 | j < (1 << BYTEWIDTH); j++) |
| 3942 | fastmap[j] = 1; |
| 3943 | } |
| 3944 | |
| 3945 | /* Fallthrough */ |
| 3946 | case charset: |
| 3947 | if (!fastmap) break; |
| 3948 | not = (re_opcode_t) *(p - 1) == charset_not; |
| 3949 | for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++; |
| 3950 | j >= 0; j--) |
| 3951 | if (!!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) ^ not) |
| 3952 | fastmap[j] = 1; |
| 3953 | |
| 3954 | #ifdef emacs |
| 3955 | if (/* Any leading code can possibly start a character |
| 3956 | which doesn't match the specified set of characters. */ |
| 3957 | not |
| 3958 | || |
| 3959 | /* If we can match a character class, we can match any |
| 3960 | multibyte characters. */ |
| 3961 | (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) |
| 3962 | && CHARSET_RANGE_TABLE_BITS (&p[-2]) != 0)) |
| 3963 | |
| 3964 | { |
| 3965 | if (match_any_multibyte_characters == false) |
| 3966 | { |
| 3967 | for (j = MIN_MULTIBYTE_LEADING_CODE; |
| 3968 | j <= MAX_MULTIBYTE_LEADING_CODE; j++) |
| 3969 | fastmap[j] = 1; |
| 3970 | match_any_multibyte_characters = true; |
| 3971 | } |
| 3972 | } |
| 3973 | |
| 3974 | else if (!not && CHARSET_RANGE_TABLE_EXISTS_P (&p[-2]) |
| 3975 | && match_any_multibyte_characters == false) |
| 3976 | { |
| 3977 | /* Set fastmap[I] to 1 where I is a leading code of each |
| 3978 | multibyte character in the range table. */ |
| 3979 | int c, count; |
| 3980 | unsigned char lc1, lc2; |
| 3981 | |
| 3982 | /* Make P points the range table. `+ 2' is to skip flag |
| 3983 | bits for a character class. */ |
| 3984 | p += CHARSET_BITMAP_SIZE (&p[-2]) + 2; |
| 3985 | |
| 3986 | /* Extract the number of ranges in range table into COUNT. */ |
| 3987 | EXTRACT_NUMBER_AND_INCR (count, p); |
| 3988 | for (; count > 0; count--, p += 3) |
| 3989 | { |
| 3990 | /* Extract the start and end of each range. */ |
| 3991 | EXTRACT_CHARACTER (c, p); |
| 3992 | lc1 = CHAR_LEADING_CODE (c); |
| 3993 | p += 3; |
| 3994 | EXTRACT_CHARACTER (c, p); |
| 3995 | lc2 = CHAR_LEADING_CODE (c); |
| 3996 | for (j = lc1; j <= lc2; j++) |
| 3997 | fastmap[j] = 1; |
| 3998 | } |
| 3999 | } |
| 4000 | #endif |
| 4001 | break; |
| 4002 | |
| 4003 | case syntaxspec: |
| 4004 | case notsyntaxspec: |
| 4005 | if (!fastmap) break; |
| 4006 | #ifndef emacs |
| 4007 | not = (re_opcode_t)p[-1] == notsyntaxspec; |
| 4008 | k = *p++; |
| 4009 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
| 4010 | if ((SYNTAX (j) == (enum syntaxcode) k) ^ not) |
| 4011 | fastmap[j] = 1; |
| 4012 | break; |
| 4013 | #else /* emacs */ |
| 4014 | /* This match depends on text properties. These end with |
| 4015 | aborting optimizations. */ |
| 4016 | return -1; |
| 4017 | |
| 4018 | case categoryspec: |
| 4019 | case notcategoryspec: |
| 4020 | if (!fastmap) break; |
| 4021 | not = (re_opcode_t)p[-1] == notcategoryspec; |
| 4022 | k = *p++; |
| 4023 | for (j = (1 << BYTEWIDTH); j >= 0; j--) |
| 4024 | if ((CHAR_HAS_CATEGORY (j, k)) ^ not) |
| 4025 | fastmap[j] = 1; |
| 4026 | |
| 4027 | /* Any leading code can possibly start a character which |
| 4028 | has or doesn't has the specified category. */ |
| 4029 | if (match_any_multibyte_characters == false) |
| 4030 | { |
| 4031 | for (j = MIN_MULTIBYTE_LEADING_CODE; |
| 4032 | j <= MAX_MULTIBYTE_LEADING_CODE; j++) |
| 4033 | fastmap[j] = 1; |
| 4034 | match_any_multibyte_characters = true; |
| 4035 | } |
| 4036 | break; |
| 4037 | |
| 4038 | /* All cases after this match the empty string. These end with |
| 4039 | `continue'. */ |
| 4040 | |
| 4041 | case before_dot: |
| 4042 | case at_dot: |
| 4043 | case after_dot: |
| 4044 | #endif /* !emacs */ |
| 4045 | case no_op: |
| 4046 | case begline: |
| 4047 | case endline: |
| 4048 | case begbuf: |
| 4049 | case endbuf: |
| 4050 | case wordbound: |
| 4051 | case notwordbound: |
| 4052 | case wordbeg: |
| 4053 | case wordend: |
| 4054 | case symbeg: |
| 4055 | case symend: |
| 4056 | continue; |
| 4057 | |
| 4058 | |
| 4059 | case jump: |
| 4060 | EXTRACT_NUMBER_AND_INCR (j, p); |
| 4061 | if (j < 0) |
| 4062 | /* Backward jumps can only go back to code that we've already |
| 4063 | visited. `re_compile' should make sure this is true. */ |
| 4064 | break; |
| 4065 | p += j; |
| 4066 | switch (*p) |
| 4067 | { |
| 4068 | case on_failure_jump: |
| 4069 | case on_failure_keep_string_jump: |
| 4070 | case on_failure_jump_loop: |
| 4071 | case on_failure_jump_nastyloop: |
| 4072 | case on_failure_jump_smart: |
| 4073 | p++; |
| 4074 | break; |
| 4075 | default: |
| 4076 | continue; |
| 4077 | }; |
| 4078 | /* Keep `p1' to allow the `on_failure_jump' we are jumping to |
| 4079 | to jump back to "just after here". */ |
| 4080 | /* Fallthrough */ |
| 4081 | |
| 4082 | case on_failure_jump: |
| 4083 | case on_failure_keep_string_jump: |
| 4084 | case on_failure_jump_nastyloop: |
| 4085 | case on_failure_jump_loop: |
| 4086 | case on_failure_jump_smart: |
| 4087 | EXTRACT_NUMBER_AND_INCR (j, p); |
| 4088 | if (p + j <= p1) |
| 4089 | ; /* Backward jump to be ignored. */ |
| 4090 | else |
| 4091 | { /* We have to look down both arms. |
| 4092 | We first go down the "straight" path so as to minimize |
| 4093 | stack usage when going through alternatives. */ |
| 4094 | int r = analyse_first (p, pend, fastmap, multibyte); |
| 4095 | if (r) return r; |
| 4096 | p += j; |
| 4097 | } |
| 4098 | continue; |
| 4099 | |
| 4100 | |
| 4101 | case jump_n: |
| 4102 | /* This code simply does not properly handle forward jump_n. */ |
| 4103 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p); assert (j < 0)); |
| 4104 | p += 4; |
| 4105 | /* jump_n can either jump or fall through. The (backward) jump |
| 4106 | case has already been handled, so we only need to look at the |
| 4107 | fallthrough case. */ |
| 4108 | continue; |
| 4109 | |
| 4110 | case succeed_n: |
| 4111 | /* If N == 0, it should be an on_failure_jump_loop instead. */ |
| 4112 | DEBUG_STATEMENT (EXTRACT_NUMBER (j, p + 2); assert (j > 0)); |
| 4113 | p += 4; |
| 4114 | /* We only care about one iteration of the loop, so we don't |
| 4115 | need to consider the case where this behaves like an |
| 4116 | on_failure_jump. */ |
| 4117 | continue; |
| 4118 | |
| 4119 | |
| 4120 | case set_number_at: |
| 4121 | p += 4; |
| 4122 | continue; |
| 4123 | |
| 4124 | |
| 4125 | case start_memory: |
| 4126 | case stop_memory: |
| 4127 | p += 1; |
| 4128 | continue; |
| 4129 | |
| 4130 | |
| 4131 | default: |
| 4132 | abort (); /* We have listed all the cases. */ |
| 4133 | } /* switch *p++ */ |
| 4134 | |
| 4135 | /* Getting here means we have found the possible starting |
| 4136 | characters for one path of the pattern -- and that the empty |
| 4137 | string does not match. We need not follow this path further. */ |
| 4138 | return 0; |
| 4139 | } /* while p */ |
| 4140 | |
| 4141 | /* We reached the end without matching anything. */ |
| 4142 | return 1; |
| 4143 | |
| 4144 | } /* analyse_first */ |
| 4145 | \f |
| 4146 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in |
| 4147 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible |
| 4148 | characters can start a string that matches the pattern. This fastmap |
| 4149 | is used by re_search to skip quickly over impossible starting points. |
| 4150 | |
| 4151 | Character codes above (1 << BYTEWIDTH) are not represented in the |
| 4152 | fastmap, but the leading codes are represented. Thus, the fastmap |
| 4153 | indicates which character sets could start a match. |
| 4154 | |
| 4155 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data |
| 4156 | area as BUFP->fastmap. |
| 4157 | |
| 4158 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in |
| 4159 | the pattern buffer. |
| 4160 | |
| 4161 | Returns 0 if we succeed, -2 if an internal error. */ |
| 4162 | |
| 4163 | int |
| 4164 | re_compile_fastmap (struct re_pattern_buffer *bufp) |
| 4165 | { |
| 4166 | char *fastmap = bufp->fastmap; |
| 4167 | int analysis; |
| 4168 | |
| 4169 | assert (fastmap && bufp->buffer); |
| 4170 | |
| 4171 | memset (fastmap, 0, 1 << BYTEWIDTH); /* Assume nothing's valid. */ |
| 4172 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ |
| 4173 | |
| 4174 | analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used, |
| 4175 | fastmap, RE_MULTIBYTE_P (bufp)); |
| 4176 | bufp->can_be_null = (analysis != 0); |
| 4177 | return 0; |
| 4178 | } /* re_compile_fastmap */ |
| 4179 | \f |
| 4180 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and |
| 4181 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use |
| 4182 | this memory for recording register information. STARTS and ENDS |
| 4183 | must be allocated using the malloc library routine, and must each |
| 4184 | be at least NUM_REGS * sizeof (regoff_t) bytes long. |
| 4185 | |
| 4186 | If NUM_REGS == 0, then subsequent matches should allocate their own |
| 4187 | register data. |
| 4188 | |
| 4189 | Unless this function is called, the first search or match using |
| 4190 | PATTERN_BUFFER will allocate its own register data, without |
| 4191 | freeing the old data. */ |
| 4192 | |
| 4193 | void |
| 4194 | re_set_registers (struct re_pattern_buffer *bufp, struct re_registers *regs, unsigned int num_regs, regoff_t *starts, regoff_t *ends) |
| 4195 | { |
| 4196 | if (num_regs) |
| 4197 | { |
| 4198 | bufp->regs_allocated = REGS_REALLOCATE; |
| 4199 | regs->num_regs = num_regs; |
| 4200 | regs->start = starts; |
| 4201 | regs->end = ends; |
| 4202 | } |
| 4203 | else |
| 4204 | { |
| 4205 | bufp->regs_allocated = REGS_UNALLOCATED; |
| 4206 | regs->num_regs = 0; |
| 4207 | regs->start = regs->end = (regoff_t *) 0; |
| 4208 | } |
| 4209 | } |
| 4210 | WEAK_ALIAS (__re_set_registers, re_set_registers) |
| 4211 | \f |
| 4212 | /* Searching routines. */ |
| 4213 | |
| 4214 | /* Like re_search_2, below, but only one string is specified, and |
| 4215 | doesn't let you say where to stop matching. */ |
| 4216 | |
| 4217 | regoff_t |
| 4218 | re_search (struct re_pattern_buffer *bufp, const char *string, size_t size, |
| 4219 | ssize_t startpos, ssize_t range, struct re_registers *regs) |
| 4220 | { |
| 4221 | return re_search_2 (bufp, NULL, 0, string, size, startpos, range, |
| 4222 | regs, size); |
| 4223 | } |
| 4224 | WEAK_ALIAS (__re_search, re_search) |
| 4225 | |
| 4226 | /* Head address of virtual concatenation of string. */ |
| 4227 | #define HEAD_ADDR_VSTRING(P) \ |
| 4228 | (((P) >= size1 ? string2 : string1)) |
| 4229 | |
| 4230 | /* Address of POS in the concatenation of virtual string. */ |
| 4231 | #define POS_ADDR_VSTRING(POS) \ |
| 4232 | (((POS) >= size1 ? string2 - size1 : string1) + (POS)) |
| 4233 | |
| 4234 | /* Using the compiled pattern in BUFP->buffer, first tries to match the |
| 4235 | virtual concatenation of STRING1 and STRING2, starting first at index |
| 4236 | STARTPOS, then at STARTPOS + 1, and so on. |
| 4237 | |
| 4238 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
| 4239 | |
| 4240 | RANGE is how far to scan while trying to match. RANGE = 0 means try |
| 4241 | only at STARTPOS; in general, the last start tried is STARTPOS + |
| 4242 | RANGE. |
| 4243 | |
| 4244 | In REGS, return the indices of the virtual concatenation of STRING1 |
| 4245 | and STRING2 that matched the entire BUFP->buffer and its contained |
| 4246 | subexpressions. |
| 4247 | |
| 4248 | Do not consider matching one past the index STOP in the virtual |
| 4249 | concatenation of STRING1 and STRING2. |
| 4250 | |
| 4251 | We return either the position in the strings at which the match was |
| 4252 | found, -1 if no match, or -2 if error (such as failure |
| 4253 | stack overflow). */ |
| 4254 | |
| 4255 | regoff_t |
| 4256 | re_search_2 (struct re_pattern_buffer *bufp, const char *str1, size_t size1, |
| 4257 | const char *str2, size_t size2, ssize_t startpos, ssize_t range, |
| 4258 | struct re_registers *regs, ssize_t stop) |
| 4259 | { |
| 4260 | regoff_t val; |
| 4261 | re_char *string1 = (re_char*) str1; |
| 4262 | re_char *string2 = (re_char*) str2; |
| 4263 | register char *fastmap = bufp->fastmap; |
| 4264 | register RE_TRANSLATE_TYPE translate = bufp->translate; |
| 4265 | size_t total_size = size1 + size2; |
| 4266 | ssize_t endpos = startpos + range; |
| 4267 | boolean anchored_start; |
| 4268 | /* Nonzero if we are searching multibyte string. */ |
| 4269 | const boolean multibyte = RE_TARGET_MULTIBYTE_P (bufp); |
| 4270 | |
| 4271 | /* Check for out-of-range STARTPOS. */ |
| 4272 | if (startpos < 0 || startpos > total_size) |
| 4273 | return -1; |
| 4274 | |
| 4275 | /* Fix up RANGE if it might eventually take us outside |
| 4276 | the virtual concatenation of STRING1 and STRING2. |
| 4277 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
| 4278 | if (endpos < 0) |
| 4279 | range = 0 - startpos; |
| 4280 | else if (endpos > total_size) |
| 4281 | range = total_size - startpos; |
| 4282 | |
| 4283 | /* If the search isn't to be a backwards one, don't waste time in a |
| 4284 | search for a pattern anchored at beginning of buffer. */ |
| 4285 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0) |
| 4286 | { |
| 4287 | if (startpos > 0) |
| 4288 | return -1; |
| 4289 | else |
| 4290 | range = 0; |
| 4291 | } |
| 4292 | |
| 4293 | #ifdef emacs |
| 4294 | /* In a forward search for something that starts with \=. |
| 4295 | don't keep searching past point. */ |
| 4296 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) |
| 4297 | { |
| 4298 | range = PT_BYTE - BEGV_BYTE - startpos; |
| 4299 | if (range < 0) |
| 4300 | return -1; |
| 4301 | } |
| 4302 | #endif /* emacs */ |
| 4303 | |
| 4304 | /* Update the fastmap now if not correct already. */ |
| 4305 | if (fastmap && !bufp->fastmap_accurate) |
| 4306 | re_compile_fastmap (bufp); |
| 4307 | |
| 4308 | /* See whether the pattern is anchored. */ |
| 4309 | anchored_start = (bufp->buffer[0] == begline); |
| 4310 | |
| 4311 | #ifdef emacs |
| 4312 | gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */ |
| 4313 | { |
| 4314 | ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos)); |
| 4315 | |
| 4316 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); |
| 4317 | } |
| 4318 | #endif |
| 4319 | |
| 4320 | /* Loop through the string, looking for a place to start matching. */ |
| 4321 | for (;;) |
| 4322 | { |
| 4323 | /* If the pattern is anchored, |
| 4324 | skip quickly past places we cannot match. |
| 4325 | We don't bother to treat startpos == 0 specially |
| 4326 | because that case doesn't repeat. */ |
| 4327 | if (anchored_start && startpos > 0) |
| 4328 | { |
| 4329 | if (! ((startpos <= size1 ? string1[startpos - 1] |
| 4330 | : string2[startpos - size1 - 1]) |
| 4331 | == '\n')) |
| 4332 | goto advance; |
| 4333 | } |
| 4334 | |
| 4335 | /* If a fastmap is supplied, skip quickly over characters that |
| 4336 | cannot be the start of a match. If the pattern can match the |
| 4337 | null string, however, we don't need to skip characters; we want |
| 4338 | the first null string. */ |
| 4339 | if (fastmap && startpos < total_size && !bufp->can_be_null) |
| 4340 | { |
| 4341 | register re_char *d; |
| 4342 | register re_wchar_t buf_ch; |
| 4343 | |
| 4344 | d = POS_ADDR_VSTRING (startpos); |
| 4345 | |
| 4346 | if (range > 0) /* Searching forwards. */ |
| 4347 | { |
| 4348 | register int lim = 0; |
| 4349 | ssize_t irange = range; |
| 4350 | |
| 4351 | if (startpos < size1 && startpos + range >= size1) |
| 4352 | lim = range - (size1 - startpos); |
| 4353 | |
| 4354 | /* Written out as an if-else to avoid testing `translate' |
| 4355 | inside the loop. */ |
| 4356 | if (RE_TRANSLATE_P (translate)) |
| 4357 | { |
| 4358 | if (multibyte) |
| 4359 | while (range > lim) |
| 4360 | { |
| 4361 | int buf_charlen; |
| 4362 | |
| 4363 | buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen); |
| 4364 | buf_ch = RE_TRANSLATE (translate, buf_ch); |
| 4365 | if (fastmap[CHAR_LEADING_CODE (buf_ch)]) |
| 4366 | break; |
| 4367 | |
| 4368 | range -= buf_charlen; |
| 4369 | d += buf_charlen; |
| 4370 | } |
| 4371 | else |
| 4372 | while (range > lim) |
| 4373 | { |
| 4374 | register re_wchar_t ch, translated; |
| 4375 | |
| 4376 | buf_ch = *d; |
| 4377 | ch = RE_CHAR_TO_MULTIBYTE (buf_ch); |
| 4378 | translated = RE_TRANSLATE (translate, ch); |
| 4379 | if (translated != ch |
| 4380 | && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0) |
| 4381 | buf_ch = ch; |
| 4382 | if (fastmap[buf_ch]) |
| 4383 | break; |
| 4384 | d++; |
| 4385 | range--; |
| 4386 | } |
| 4387 | } |
| 4388 | else |
| 4389 | { |
| 4390 | if (multibyte) |
| 4391 | while (range > lim) |
| 4392 | { |
| 4393 | int buf_charlen; |
| 4394 | |
| 4395 | buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen); |
| 4396 | if (fastmap[CHAR_LEADING_CODE (buf_ch)]) |
| 4397 | break; |
| 4398 | range -= buf_charlen; |
| 4399 | d += buf_charlen; |
| 4400 | } |
| 4401 | else |
| 4402 | while (range > lim && !fastmap[*d]) |
| 4403 | { |
| 4404 | d++; |
| 4405 | range--; |
| 4406 | } |
| 4407 | } |
| 4408 | startpos += irange - range; |
| 4409 | } |
| 4410 | else /* Searching backwards. */ |
| 4411 | { |
| 4412 | if (multibyte) |
| 4413 | { |
| 4414 | buf_ch = STRING_CHAR (d); |
| 4415 | buf_ch = TRANSLATE (buf_ch); |
| 4416 | if (! fastmap[CHAR_LEADING_CODE (buf_ch)]) |
| 4417 | goto advance; |
| 4418 | } |
| 4419 | else |
| 4420 | { |
| 4421 | register re_wchar_t ch, translated; |
| 4422 | |
| 4423 | buf_ch = *d; |
| 4424 | ch = RE_CHAR_TO_MULTIBYTE (buf_ch); |
| 4425 | translated = TRANSLATE (ch); |
| 4426 | if (translated != ch |
| 4427 | && (ch = RE_CHAR_TO_UNIBYTE (translated)) >= 0) |
| 4428 | buf_ch = ch; |
| 4429 | if (! fastmap[TRANSLATE (buf_ch)]) |
| 4430 | goto advance; |
| 4431 | } |
| 4432 | } |
| 4433 | } |
| 4434 | |
| 4435 | /* If can't match the null string, and that's all we have left, fail. */ |
| 4436 | if (range >= 0 && startpos == total_size && fastmap |
| 4437 | && !bufp->can_be_null) |
| 4438 | return -1; |
| 4439 | |
| 4440 | val = re_match_2_internal (bufp, string1, size1, string2, size2, |
| 4441 | startpos, regs, stop); |
| 4442 | |
| 4443 | if (val >= 0) |
| 4444 | return startpos; |
| 4445 | |
| 4446 | if (val == -2) |
| 4447 | return -2; |
| 4448 | |
| 4449 | advance: |
| 4450 | if (!range) |
| 4451 | break; |
| 4452 | else if (range > 0) |
| 4453 | { |
| 4454 | /* Update STARTPOS to the next character boundary. */ |
| 4455 | if (multibyte) |
| 4456 | { |
| 4457 | re_char *p = POS_ADDR_VSTRING (startpos); |
| 4458 | int len = BYTES_BY_CHAR_HEAD (*p); |
| 4459 | |
| 4460 | range -= len; |
| 4461 | if (range < 0) |
| 4462 | break; |
| 4463 | startpos += len; |
| 4464 | } |
| 4465 | else |
| 4466 | { |
| 4467 | range--; |
| 4468 | startpos++; |
| 4469 | } |
| 4470 | } |
| 4471 | else |
| 4472 | { |
| 4473 | range++; |
| 4474 | startpos--; |
| 4475 | |
| 4476 | /* Update STARTPOS to the previous character boundary. */ |
| 4477 | if (multibyte) |
| 4478 | { |
| 4479 | re_char *p = POS_ADDR_VSTRING (startpos) + 1; |
| 4480 | re_char *p0 = p; |
| 4481 | re_char *phead = HEAD_ADDR_VSTRING (startpos); |
| 4482 | |
| 4483 | /* Find the head of multibyte form. */ |
| 4484 | PREV_CHAR_BOUNDARY (p, phead); |
| 4485 | range += p0 - 1 - p; |
| 4486 | if (range > 0) |
| 4487 | break; |
| 4488 | |
| 4489 | startpos -= p0 - 1 - p; |
| 4490 | } |
| 4491 | } |
| 4492 | } |
| 4493 | return -1; |
| 4494 | } /* re_search_2 */ |
| 4495 | WEAK_ALIAS (__re_search_2, re_search_2) |
| 4496 | \f |
| 4497 | /* Declarations and macros for re_match_2. */ |
| 4498 | |
| 4499 | static int bcmp_translate (re_char *s1, re_char *s2, |
| 4500 | register ssize_t len, |
| 4501 | RE_TRANSLATE_TYPE translate, |
| 4502 | const int multibyte); |
| 4503 | |
| 4504 | /* This converts PTR, a pointer into one of the search strings `string1' |
| 4505 | and `string2' into an offset from the beginning of that string. */ |
| 4506 | #define POINTER_TO_OFFSET(ptr) \ |
| 4507 | (FIRST_STRING_P (ptr) \ |
| 4508 | ? (ptr) - string1 \ |
| 4509 | : (ptr) - string2 + (ptrdiff_t) size1) |
| 4510 | |
| 4511 | /* Call before fetching a character with *d. This switches over to |
| 4512 | string2 if necessary. |
| 4513 | Check re_match_2_internal for a discussion of why end_match_2 might |
| 4514 | not be within string2 (but be equal to end_match_1 instead). */ |
| 4515 | #define PREFETCH() \ |
| 4516 | while (d == dend) \ |
| 4517 | { \ |
| 4518 | /* End of string2 => fail. */ \ |
| 4519 | if (dend == end_match_2) \ |
| 4520 | goto fail; \ |
| 4521 | /* End of string1 => advance to string2. */ \ |
| 4522 | d = string2; \ |
| 4523 | dend = end_match_2; \ |
| 4524 | } |
| 4525 | |
| 4526 | /* Call before fetching a char with *d if you already checked other limits. |
| 4527 | This is meant for use in lookahead operations like wordend, etc.. |
| 4528 | where we might need to look at parts of the string that might be |
| 4529 | outside of the LIMITs (i.e past `stop'). */ |
| 4530 | #define PREFETCH_NOLIMIT() \ |
| 4531 | if (d == end1) \ |
| 4532 | { \ |
| 4533 | d = string2; \ |
| 4534 | dend = end_match_2; \ |
| 4535 | } \ |
| 4536 | |
| 4537 | /* Test if at very beginning or at very end of the virtual concatenation |
| 4538 | of `string1' and `string2'. If only one string, it's `string2'. */ |
| 4539 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
| 4540 | #define AT_STRINGS_END(d) ((d) == end2) |
| 4541 | |
| 4542 | /* Disabled due to a compiler bug -- see comment at case wordbound */ |
| 4543 | |
| 4544 | /* The comment at case wordbound is following one, but we don't use |
| 4545 | AT_WORD_BOUNDARY anymore to support multibyte form. |
| 4546 | |
| 4547 | The DEC Alpha C compiler 3.x generates incorrect code for the |
| 4548 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of |
| 4549 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the |
| 4550 | macro and introducing temporary variables works around the bug. */ |
| 4551 | |
| 4552 | #if 0 |
| 4553 | /* Test if D points to a character which is word-constituent. We have |
| 4554 | two special cases to check for: if past the end of string1, look at |
| 4555 | the first character in string2; and if before the beginning of |
| 4556 | string2, look at the last character in string1. */ |
| 4557 | #define WORDCHAR_P(d) \ |
| 4558 | (SYNTAX ((d) == end1 ? *string2 \ |
| 4559 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
| 4560 | == Sword) |
| 4561 | |
| 4562 | /* Test if the character before D and the one at D differ with respect |
| 4563 | to being word-constituent. */ |
| 4564 | #define AT_WORD_BOUNDARY(d) \ |
| 4565 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ |
| 4566 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) |
| 4567 | #endif |
| 4568 | |
| 4569 | /* Free everything we malloc. */ |
| 4570 | #ifdef MATCH_MAY_ALLOCATE |
| 4571 | # define FREE_VAR(var) \ |
| 4572 | do { \ |
| 4573 | if (var) \ |
| 4574 | { \ |
| 4575 | REGEX_FREE (var); \ |
| 4576 | var = NULL; \ |
| 4577 | } \ |
| 4578 | } while (0) |
| 4579 | # define FREE_VARIABLES() \ |
| 4580 | do { \ |
| 4581 | REGEX_FREE_STACK (fail_stack.stack); \ |
| 4582 | FREE_VAR (regstart); \ |
| 4583 | FREE_VAR (regend); \ |
| 4584 | FREE_VAR (best_regstart); \ |
| 4585 | FREE_VAR (best_regend); \ |
| 4586 | } while (0) |
| 4587 | #else |
| 4588 | # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ |
| 4589 | #endif /* not MATCH_MAY_ALLOCATE */ |
| 4590 | |
| 4591 | \f |
| 4592 | /* Optimization routines. */ |
| 4593 | |
| 4594 | /* If the operation is a match against one or more chars, |
| 4595 | return a pointer to the next operation, else return NULL. */ |
| 4596 | static re_char * |
| 4597 | skip_one_char (const re_char *p) |
| 4598 | { |
| 4599 | switch (*p++) |
| 4600 | { |
| 4601 | case anychar: |
| 4602 | break; |
| 4603 | |
| 4604 | case exactn: |
| 4605 | p += *p + 1; |
| 4606 | break; |
| 4607 | |
| 4608 | case charset_not: |
| 4609 | case charset: |
| 4610 | if (CHARSET_RANGE_TABLE_EXISTS_P (p - 1)) |
| 4611 | { |
| 4612 | int mcnt; |
| 4613 | p = CHARSET_RANGE_TABLE (p - 1); |
| 4614 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 4615 | p = CHARSET_RANGE_TABLE_END (p, mcnt); |
| 4616 | } |
| 4617 | else |
| 4618 | p += 1 + CHARSET_BITMAP_SIZE (p - 1); |
| 4619 | break; |
| 4620 | |
| 4621 | case syntaxspec: |
| 4622 | case notsyntaxspec: |
| 4623 | #ifdef emacs |
| 4624 | case categoryspec: |
| 4625 | case notcategoryspec: |
| 4626 | #endif /* emacs */ |
| 4627 | p++; |
| 4628 | break; |
| 4629 | |
| 4630 | default: |
| 4631 | p = NULL; |
| 4632 | } |
| 4633 | return p; |
| 4634 | } |
| 4635 | |
| 4636 | |
| 4637 | /* Jump over non-matching operations. */ |
| 4638 | static re_char * |
| 4639 | skip_noops (const re_char *p, const re_char *pend) |
| 4640 | { |
| 4641 | int mcnt; |
| 4642 | while (p < pend) |
| 4643 | { |
| 4644 | switch (*p) |
| 4645 | { |
| 4646 | case start_memory: |
| 4647 | case stop_memory: |
| 4648 | p += 2; break; |
| 4649 | case no_op: |
| 4650 | p += 1; break; |
| 4651 | case jump: |
| 4652 | p += 1; |
| 4653 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 4654 | p += mcnt; |
| 4655 | break; |
| 4656 | default: |
| 4657 | return p; |
| 4658 | } |
| 4659 | } |
| 4660 | assert (p == pend); |
| 4661 | return p; |
| 4662 | } |
| 4663 | |
| 4664 | /* Non-zero if "p1 matches something" implies "p2 fails". */ |
| 4665 | static int |
| 4666 | mutually_exclusive_p (struct re_pattern_buffer *bufp, const re_char *p1, const re_char *p2) |
| 4667 | { |
| 4668 | re_opcode_t op2; |
| 4669 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
| 4670 | unsigned char *pend = bufp->buffer + bufp->used; |
| 4671 | |
| 4672 | assert (p1 >= bufp->buffer && p1 < pend |
| 4673 | && p2 >= bufp->buffer && p2 <= pend); |
| 4674 | |
| 4675 | /* Skip over open/close-group commands. |
| 4676 | If what follows this loop is a ...+ construct, |
| 4677 | look at what begins its body, since we will have to |
| 4678 | match at least one of that. */ |
| 4679 | p2 = skip_noops (p2, pend); |
| 4680 | /* The same skip can be done for p1, except that this function |
| 4681 | is only used in the case where p1 is a simple match operator. */ |
| 4682 | /* p1 = skip_noops (p1, pend); */ |
| 4683 | |
| 4684 | assert (p1 >= bufp->buffer && p1 < pend |
| 4685 | && p2 >= bufp->buffer && p2 <= pend); |
| 4686 | |
| 4687 | op2 = p2 == pend ? succeed : *p2; |
| 4688 | |
| 4689 | switch (op2) |
| 4690 | { |
| 4691 | case succeed: |
| 4692 | case endbuf: |
| 4693 | /* If we're at the end of the pattern, we can change. */ |
| 4694 | if (skip_one_char (p1)) |
| 4695 | { |
| 4696 | DEBUG_PRINT (" End of pattern: fast loop.\n"); |
| 4697 | return 1; |
| 4698 | } |
| 4699 | break; |
| 4700 | |
| 4701 | case endline: |
| 4702 | case exactn: |
| 4703 | { |
| 4704 | register re_wchar_t c |
| 4705 | = (re_opcode_t) *p2 == endline ? '\n' |
| 4706 | : RE_STRING_CHAR (p2 + 2, multibyte); |
| 4707 | |
| 4708 | if ((re_opcode_t) *p1 == exactn) |
| 4709 | { |
| 4710 | if (c != RE_STRING_CHAR (p1 + 2, multibyte)) |
| 4711 | { |
| 4712 | DEBUG_PRINT (" '%c' != '%c' => fast loop.\n", c, p1[2]); |
| 4713 | return 1; |
| 4714 | } |
| 4715 | } |
| 4716 | |
| 4717 | else if ((re_opcode_t) *p1 == charset |
| 4718 | || (re_opcode_t) *p1 == charset_not) |
| 4719 | { |
| 4720 | int not = (re_opcode_t) *p1 == charset_not; |
| 4721 | |
| 4722 | /* Test if C is listed in charset (or charset_not) |
| 4723 | at `p1'. */ |
| 4724 | if (! multibyte || IS_REAL_ASCII (c)) |
| 4725 | { |
| 4726 | if (c < CHARSET_BITMAP_SIZE (p1) * BYTEWIDTH |
| 4727 | && p1[2 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
| 4728 | not = !not; |
| 4729 | } |
| 4730 | else if (CHARSET_RANGE_TABLE_EXISTS_P (p1)) |
| 4731 | CHARSET_LOOKUP_RANGE_TABLE (not, c, p1); |
| 4732 | |
| 4733 | /* `not' is equal to 1 if c would match, which means |
| 4734 | that we can't change to pop_failure_jump. */ |
| 4735 | if (!not) |
| 4736 | { |
| 4737 | DEBUG_PRINT (" No match => fast loop.\n"); |
| 4738 | return 1; |
| 4739 | } |
| 4740 | } |
| 4741 | else if ((re_opcode_t) *p1 == anychar |
| 4742 | && c == '\n') |
| 4743 | { |
| 4744 | DEBUG_PRINT (" . != \\n => fast loop.\n"); |
| 4745 | return 1; |
| 4746 | } |
| 4747 | } |
| 4748 | break; |
| 4749 | |
| 4750 | case charset: |
| 4751 | { |
| 4752 | if ((re_opcode_t) *p1 == exactn) |
| 4753 | /* Reuse the code above. */ |
| 4754 | return mutually_exclusive_p (bufp, p2, p1); |
| 4755 | |
| 4756 | /* It is hard to list up all the character in charset |
| 4757 | P2 if it includes multibyte character. Give up in |
| 4758 | such case. */ |
| 4759 | else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2)) |
| 4760 | { |
| 4761 | /* Now, we are sure that P2 has no range table. |
| 4762 | So, for the size of bitmap in P2, `p2[1]' is |
| 4763 | enough. But P1 may have range table, so the |
| 4764 | size of bitmap table of P1 is extracted by |
| 4765 | using macro `CHARSET_BITMAP_SIZE'. |
| 4766 | |
| 4767 | In a multibyte case, we know that all the character |
| 4768 | listed in P2 is ASCII. In a unibyte case, P1 has only a |
| 4769 | bitmap table. So, in both cases, it is enough to test |
| 4770 | only the bitmap table of P1. */ |
| 4771 | |
| 4772 | if ((re_opcode_t) *p1 == charset) |
| 4773 | { |
| 4774 | int idx; |
| 4775 | /* We win if the charset inside the loop |
| 4776 | has no overlap with the one after the loop. */ |
| 4777 | for (idx = 0; |
| 4778 | (idx < (int) p2[1] |
| 4779 | && idx < CHARSET_BITMAP_SIZE (p1)); |
| 4780 | idx++) |
| 4781 | if ((p2[2 + idx] & p1[2 + idx]) != 0) |
| 4782 | break; |
| 4783 | |
| 4784 | if (idx == p2[1] |
| 4785 | || idx == CHARSET_BITMAP_SIZE (p1)) |
| 4786 | { |
| 4787 | DEBUG_PRINT (" No match => fast loop.\n"); |
| 4788 | return 1; |
| 4789 | } |
| 4790 | } |
| 4791 | else if ((re_opcode_t) *p1 == charset_not) |
| 4792 | { |
| 4793 | int idx; |
| 4794 | /* We win if the charset_not inside the loop lists |
| 4795 | every character listed in the charset after. */ |
| 4796 | for (idx = 0; idx < (int) p2[1]; idx++) |
| 4797 | if (! (p2[2 + idx] == 0 |
| 4798 | || (idx < CHARSET_BITMAP_SIZE (p1) |
| 4799 | && ((p2[2 + idx] & ~ p1[2 + idx]) == 0)))) |
| 4800 | break; |
| 4801 | |
| 4802 | if (idx == p2[1]) |
| 4803 | { |
| 4804 | DEBUG_PRINT (" No match => fast loop.\n"); |
| 4805 | return 1; |
| 4806 | } |
| 4807 | } |
| 4808 | } |
| 4809 | } |
| 4810 | break; |
| 4811 | |
| 4812 | case charset_not: |
| 4813 | switch (*p1) |
| 4814 | { |
| 4815 | case exactn: |
| 4816 | case charset: |
| 4817 | /* Reuse the code above. */ |
| 4818 | return mutually_exclusive_p (bufp, p2, p1); |
| 4819 | case charset_not: |
| 4820 | /* When we have two charset_not, it's very unlikely that |
| 4821 | they don't overlap. The union of the two sets of excluded |
| 4822 | chars should cover all possible chars, which, as a matter of |
| 4823 | fact, is virtually impossible in multibyte buffers. */ |
| 4824 | break; |
| 4825 | } |
| 4826 | break; |
| 4827 | |
| 4828 | case wordend: |
| 4829 | return ((re_opcode_t) *p1 == syntaxspec && p1[1] == Sword); |
| 4830 | case symend: |
| 4831 | return ((re_opcode_t) *p1 == syntaxspec |
| 4832 | && (p1[1] == Ssymbol || p1[1] == Sword)); |
| 4833 | case notsyntaxspec: |
| 4834 | return ((re_opcode_t) *p1 == syntaxspec && p1[1] == p2[1]); |
| 4835 | |
| 4836 | case wordbeg: |
| 4837 | return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == Sword); |
| 4838 | case symbeg: |
| 4839 | return ((re_opcode_t) *p1 == notsyntaxspec |
| 4840 | && (p1[1] == Ssymbol || p1[1] == Sword)); |
| 4841 | case syntaxspec: |
| 4842 | return ((re_opcode_t) *p1 == notsyntaxspec && p1[1] == p2[1]); |
| 4843 | |
| 4844 | case wordbound: |
| 4845 | return (((re_opcode_t) *p1 == notsyntaxspec |
| 4846 | || (re_opcode_t) *p1 == syntaxspec) |
| 4847 | && p1[1] == Sword); |
| 4848 | |
| 4849 | #ifdef emacs |
| 4850 | case categoryspec: |
| 4851 | return ((re_opcode_t) *p1 == notcategoryspec && p1[1] == p2[1]); |
| 4852 | case notcategoryspec: |
| 4853 | return ((re_opcode_t) *p1 == categoryspec && p1[1] == p2[1]); |
| 4854 | #endif /* emacs */ |
| 4855 | |
| 4856 | default: |
| 4857 | ; |
| 4858 | } |
| 4859 | |
| 4860 | /* Safe default. */ |
| 4861 | return 0; |
| 4862 | } |
| 4863 | |
| 4864 | \f |
| 4865 | /* Matching routines. */ |
| 4866 | |
| 4867 | #ifndef emacs /* Emacs never uses this. */ |
| 4868 | /* re_match is like re_match_2 except it takes only a single string. */ |
| 4869 | |
| 4870 | regoff_t |
| 4871 | re_match (struct re_pattern_buffer *bufp, const char *string, |
| 4872 | size_t size, ssize_t pos, struct re_registers *regs) |
| 4873 | { |
| 4874 | regoff_t result = re_match_2_internal (bufp, NULL, 0, (re_char*) string, |
| 4875 | size, pos, regs, size); |
| 4876 | return result; |
| 4877 | } |
| 4878 | WEAK_ALIAS (__re_match, re_match) |
| 4879 | #endif /* not emacs */ |
| 4880 | |
| 4881 | #ifdef emacs |
| 4882 | /* In Emacs, this is the string or buffer in which we |
| 4883 | are matching. It is used for looking up syntax properties. */ |
| 4884 | Lisp_Object re_match_object; |
| 4885 | #endif |
| 4886 | |
| 4887 | /* re_match_2 matches the compiled pattern in BUFP against the |
| 4888 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 |
| 4889 | and SIZE2, respectively). We start matching at POS, and stop |
| 4890 | matching at STOP. |
| 4891 | |
| 4892 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
| 4893 | store offsets for the substring each group matched in REGS. See the |
| 4894 | documentation for exactly how many groups we fill. |
| 4895 | |
| 4896 | We return -1 if no match, -2 if an internal error (such as the |
| 4897 | failure stack overflowing). Otherwise, we return the length of the |
| 4898 | matched substring. */ |
| 4899 | |
| 4900 | regoff_t |
| 4901 | re_match_2 (struct re_pattern_buffer *bufp, const char *string1, |
| 4902 | size_t size1, const char *string2, size_t size2, ssize_t pos, |
| 4903 | struct re_registers *regs, ssize_t stop) |
| 4904 | { |
| 4905 | regoff_t result; |
| 4906 | |
| 4907 | #ifdef emacs |
| 4908 | ssize_t charpos; |
| 4909 | gl_state.object = re_match_object; /* Used by SYNTAX_TABLE_BYTE_TO_CHAR. */ |
| 4910 | charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (pos)); |
| 4911 | SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object, charpos, 1); |
| 4912 | #endif |
| 4913 | |
| 4914 | result = re_match_2_internal (bufp, (re_char*) string1, size1, |
| 4915 | (re_char*) string2, size2, |
| 4916 | pos, regs, stop); |
| 4917 | return result; |
| 4918 | } |
| 4919 | WEAK_ALIAS (__re_match_2, re_match_2) |
| 4920 | |
| 4921 | |
| 4922 | /* This is a separate function so that we can force an alloca cleanup |
| 4923 | afterwards. */ |
| 4924 | static regoff_t |
| 4925 | re_match_2_internal (struct re_pattern_buffer *bufp, const re_char *string1, |
| 4926 | size_t size1, const re_char *string2, size_t size2, |
| 4927 | ssize_t pos, struct re_registers *regs, ssize_t stop) |
| 4928 | { |
| 4929 | /* General temporaries. */ |
| 4930 | int mcnt; |
| 4931 | size_t reg; |
| 4932 | |
| 4933 | /* Just past the end of the corresponding string. */ |
| 4934 | re_char *end1, *end2; |
| 4935 | |
| 4936 | /* Pointers into string1 and string2, just past the last characters in |
| 4937 | each to consider matching. */ |
| 4938 | re_char *end_match_1, *end_match_2; |
| 4939 | |
| 4940 | /* Where we are in the data, and the end of the current string. */ |
| 4941 | re_char *d, *dend; |
| 4942 | |
| 4943 | /* Used sometimes to remember where we were before starting matching |
| 4944 | an operator so that we can go back in case of failure. This "atomic" |
| 4945 | behavior of matching opcodes is indispensable to the correctness |
| 4946 | of the on_failure_keep_string_jump optimization. */ |
| 4947 | re_char *dfail; |
| 4948 | |
| 4949 | /* Where we are in the pattern, and the end of the pattern. */ |
| 4950 | re_char *p = bufp->buffer; |
| 4951 | re_char *pend = p + bufp->used; |
| 4952 | |
| 4953 | /* We use this to map every character in the string. */ |
| 4954 | RE_TRANSLATE_TYPE translate = bufp->translate; |
| 4955 | |
| 4956 | /* Nonzero if BUFP is setup from a multibyte regex. */ |
| 4957 | const boolean multibyte = RE_MULTIBYTE_P (bufp); |
| 4958 | |
| 4959 | /* Nonzero if STRING1/STRING2 are multibyte. */ |
| 4960 | const boolean target_multibyte = RE_TARGET_MULTIBYTE_P (bufp); |
| 4961 | |
| 4962 | /* Failure point stack. Each place that can handle a failure further |
| 4963 | down the line pushes a failure point on this stack. It consists of |
| 4964 | regstart, and regend for all registers corresponding to |
| 4965 | the subexpressions we're currently inside, plus the number of such |
| 4966 | registers, and, finally, two char *'s. The first char * is where |
| 4967 | to resume scanning the pattern; the second one is where to resume |
| 4968 | scanning the strings. */ |
| 4969 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
| 4970 | fail_stack_type fail_stack; |
| 4971 | #endif |
| 4972 | #ifdef DEBUG_COMPILES_ARGUMENTS |
| 4973 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
| 4974 | #endif |
| 4975 | |
| 4976 | #if defined REL_ALLOC && defined REGEX_MALLOC |
| 4977 | /* This holds the pointer to the failure stack, when |
| 4978 | it is allocated relocatably. */ |
| 4979 | fail_stack_elt_t *failure_stack_ptr; |
| 4980 | #endif |
| 4981 | |
| 4982 | /* We fill all the registers internally, independent of what we |
| 4983 | return, for use in backreferences. The number here includes |
| 4984 | an element for register zero. */ |
| 4985 | size_t num_regs = bufp->re_nsub + 1; |
| 4986 | |
| 4987 | /* Information on the contents of registers. These are pointers into |
| 4988 | the input strings; they record just what was matched (on this |
| 4989 | attempt) by a subexpression part of the pattern, that is, the |
| 4990 | regnum-th regstart pointer points to where in the pattern we began |
| 4991 | matching and the regnum-th regend points to right after where we |
| 4992 | stopped matching the regnum-th subexpression. (The zeroth register |
| 4993 | keeps track of what the whole pattern matches.) */ |
| 4994 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
| 4995 | re_char **regstart, **regend; |
| 4996 | #endif |
| 4997 | |
| 4998 | /* The following record the register info as found in the above |
| 4999 | variables when we find a match better than any we've seen before. |
| 5000 | This happens as we backtrack through the failure points, which in |
| 5001 | turn happens only if we have not yet matched the entire string. */ |
| 5002 | unsigned best_regs_set = false; |
| 5003 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
| 5004 | re_char **best_regstart, **best_regend; |
| 5005 | #endif |
| 5006 | |
| 5007 | /* Logically, this is `best_regend[0]'. But we don't want to have to |
| 5008 | allocate space for that if we're not allocating space for anything |
| 5009 | else (see below). Also, we never need info about register 0 for |
| 5010 | any of the other register vectors, and it seems rather a kludge to |
| 5011 | treat `best_regend' differently than the rest. So we keep track of |
| 5012 | the end of the best match so far in a separate variable. We |
| 5013 | initialize this to NULL so that when we backtrack the first time |
| 5014 | and need to test it, it's not garbage. */ |
| 5015 | re_char *match_end = NULL; |
| 5016 | |
| 5017 | #ifdef DEBUG_COMPILES_ARGUMENTS |
| 5018 | /* Counts the total number of registers pushed. */ |
| 5019 | unsigned num_regs_pushed = 0; |
| 5020 | #endif |
| 5021 | |
| 5022 | DEBUG_PRINT ("\n\nEntering re_match_2.\n"); |
| 5023 | |
| 5024 | INIT_FAIL_STACK (); |
| 5025 | |
| 5026 | #ifdef MATCH_MAY_ALLOCATE |
| 5027 | /* Do not bother to initialize all the register variables if there are |
| 5028 | no groups in the pattern, as it takes a fair amount of time. If |
| 5029 | there are groups, we include space for register 0 (the whole |
| 5030 | pattern), even though we never use it, since it simplifies the |
| 5031 | array indexing. We should fix this. */ |
| 5032 | if (bufp->re_nsub) |
| 5033 | { |
| 5034 | regstart = REGEX_TALLOC (num_regs, re_char *); |
| 5035 | regend = REGEX_TALLOC (num_regs, re_char *); |
| 5036 | best_regstart = REGEX_TALLOC (num_regs, re_char *); |
| 5037 | best_regend = REGEX_TALLOC (num_regs, re_char *); |
| 5038 | |
| 5039 | if (!(regstart && regend && best_regstart && best_regend)) |
| 5040 | { |
| 5041 | FREE_VARIABLES (); |
| 5042 | return -2; |
| 5043 | } |
| 5044 | } |
| 5045 | else |
| 5046 | { |
| 5047 | /* We must initialize all our variables to NULL, so that |
| 5048 | `FREE_VARIABLES' doesn't try to free them. */ |
| 5049 | regstart = regend = best_regstart = best_regend = NULL; |
| 5050 | } |
| 5051 | #endif /* MATCH_MAY_ALLOCATE */ |
| 5052 | |
| 5053 | /* The starting position is bogus. */ |
| 5054 | if (pos < 0 || pos > size1 + size2) |
| 5055 | { |
| 5056 | FREE_VARIABLES (); |
| 5057 | return -1; |
| 5058 | } |
| 5059 | |
| 5060 | /* Initialize subexpression text positions to -1 to mark ones that no |
| 5061 | start_memory/stop_memory has been seen for. Also initialize the |
| 5062 | register information struct. */ |
| 5063 | for (reg = 1; reg < num_regs; reg++) |
| 5064 | regstart[reg] = regend[reg] = NULL; |
| 5065 | |
| 5066 | /* We move `string1' into `string2' if the latter's empty -- but not if |
| 5067 | `string1' is null. */ |
| 5068 | if (size2 == 0 && string1 != NULL) |
| 5069 | { |
| 5070 | string2 = string1; |
| 5071 | size2 = size1; |
| 5072 | string1 = 0; |
| 5073 | size1 = 0; |
| 5074 | } |
| 5075 | end1 = string1 + size1; |
| 5076 | end2 = string2 + size2; |
| 5077 | |
| 5078 | /* `p' scans through the pattern as `d' scans through the data. |
| 5079 | `dend' is the end of the input string that `d' points within. `d' |
| 5080 | is advanced into the following input string whenever necessary, but |
| 5081 | this happens before fetching; therefore, at the beginning of the |
| 5082 | loop, `d' can be pointing at the end of a string, but it cannot |
| 5083 | equal `string2'. */ |
| 5084 | if (pos >= size1) |
| 5085 | { |
| 5086 | /* Only match within string2. */ |
| 5087 | d = string2 + pos - size1; |
| 5088 | dend = end_match_2 = string2 + stop - size1; |
| 5089 | end_match_1 = end1; /* Just to give it a value. */ |
| 5090 | } |
| 5091 | else |
| 5092 | { |
| 5093 | if (stop < size1) |
| 5094 | { |
| 5095 | /* Only match within string1. */ |
| 5096 | end_match_1 = string1 + stop; |
| 5097 | /* BEWARE! |
| 5098 | When we reach end_match_1, PREFETCH normally switches to string2. |
| 5099 | But in the present case, this means that just doing a PREFETCH |
| 5100 | makes us jump from `stop' to `gap' within the string. |
| 5101 | What we really want here is for the search to stop as |
| 5102 | soon as we hit end_match_1. That's why we set end_match_2 |
| 5103 | to end_match_1 (since PREFETCH fails as soon as we hit |
| 5104 | end_match_2). */ |
| 5105 | end_match_2 = end_match_1; |
| 5106 | } |
| 5107 | else |
| 5108 | { /* It's important to use this code when stop == size so that |
| 5109 | moving `d' from end1 to string2 will not prevent the d == dend |
| 5110 | check from catching the end of string. */ |
| 5111 | end_match_1 = end1; |
| 5112 | end_match_2 = string2 + stop - size1; |
| 5113 | } |
| 5114 | d = string1 + pos; |
| 5115 | dend = end_match_1; |
| 5116 | } |
| 5117 | |
| 5118 | DEBUG_PRINT ("The compiled pattern is: "); |
| 5119 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); |
| 5120 | DEBUG_PRINT ("The string to match is: `"); |
| 5121 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); |
| 5122 | DEBUG_PRINT ("'\n"); |
| 5123 | |
| 5124 | /* This loops over pattern commands. It exits by returning from the |
| 5125 | function if the match is complete, or it drops through if the match |
| 5126 | fails at this starting point in the input data. */ |
| 5127 | for (;;) |
| 5128 | { |
| 5129 | DEBUG_PRINT ("\n%p: ", p); |
| 5130 | |
| 5131 | if (p == pend) |
| 5132 | { |
| 5133 | ptrdiff_t dcnt; |
| 5134 | |
| 5135 | /* End of pattern means we might have succeeded. */ |
| 5136 | DEBUG_PRINT ("end of pattern ... "); |
| 5137 | |
| 5138 | /* If we haven't matched the entire string, and we want the |
| 5139 | longest match, try backtracking. */ |
| 5140 | if (d != end_match_2) |
| 5141 | { |
| 5142 | /* 1 if this match ends in the same string (string1 or string2) |
| 5143 | as the best previous match. */ |
| 5144 | boolean same_str_p = (FIRST_STRING_P (match_end) |
| 5145 | == FIRST_STRING_P (d)); |
| 5146 | /* 1 if this match is the best seen so far. */ |
| 5147 | boolean best_match_p; |
| 5148 | |
| 5149 | /* AIX compiler got confused when this was combined |
| 5150 | with the previous declaration. */ |
| 5151 | if (same_str_p) |
| 5152 | best_match_p = d > match_end; |
| 5153 | else |
| 5154 | best_match_p = !FIRST_STRING_P (d); |
| 5155 | |
| 5156 | DEBUG_PRINT ("backtracking.\n"); |
| 5157 | |
| 5158 | if (!FAIL_STACK_EMPTY ()) |
| 5159 | { /* More failure points to try. */ |
| 5160 | |
| 5161 | /* If exceeds best match so far, save it. */ |
| 5162 | if (!best_regs_set || best_match_p) |
| 5163 | { |
| 5164 | best_regs_set = true; |
| 5165 | match_end = d; |
| 5166 | |
| 5167 | DEBUG_PRINT ("\nSAVING match as best so far.\n"); |
| 5168 | |
| 5169 | for (reg = 1; reg < num_regs; reg++) |
| 5170 | { |
| 5171 | best_regstart[reg] = regstart[reg]; |
| 5172 | best_regend[reg] = regend[reg]; |
| 5173 | } |
| 5174 | } |
| 5175 | goto fail; |
| 5176 | } |
| 5177 | |
| 5178 | /* If no failure points, don't restore garbage. And if |
| 5179 | last match is real best match, don't restore second |
| 5180 | best one. */ |
| 5181 | else if (best_regs_set && !best_match_p) |
| 5182 | { |
| 5183 | restore_best_regs: |
| 5184 | /* Restore best match. It may happen that `dend == |
| 5185 | end_match_1' while the restored d is in string2. |
| 5186 | For example, the pattern `x.*y.*z' against the |
| 5187 | strings `x-' and `y-z-', if the two strings are |
| 5188 | not consecutive in memory. */ |
| 5189 | DEBUG_PRINT ("Restoring best registers.\n"); |
| 5190 | |
| 5191 | d = match_end; |
| 5192 | dend = ((d >= string1 && d <= end1) |
| 5193 | ? end_match_1 : end_match_2); |
| 5194 | |
| 5195 | for (reg = 1; reg < num_regs; reg++) |
| 5196 | { |
| 5197 | regstart[reg] = best_regstart[reg]; |
| 5198 | regend[reg] = best_regend[reg]; |
| 5199 | } |
| 5200 | } |
| 5201 | } /* d != end_match_2 */ |
| 5202 | |
| 5203 | succeed_label: |
| 5204 | DEBUG_PRINT ("Accepting match.\n"); |
| 5205 | |
| 5206 | /* If caller wants register contents data back, do it. */ |
| 5207 | if (regs && !bufp->no_sub) |
| 5208 | { |
| 5209 | /* Have the register data arrays been allocated? */ |
| 5210 | if (bufp->regs_allocated == REGS_UNALLOCATED) |
| 5211 | { /* No. So allocate them with malloc. We need one |
| 5212 | extra element beyond `num_regs' for the `-1' marker |
| 5213 | GNU code uses. */ |
| 5214 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); |
| 5215 | regs->start = TALLOC (regs->num_regs, regoff_t); |
| 5216 | regs->end = TALLOC (regs->num_regs, regoff_t); |
| 5217 | if (regs->start == NULL || regs->end == NULL) |
| 5218 | { |
| 5219 | FREE_VARIABLES (); |
| 5220 | return -2; |
| 5221 | } |
| 5222 | bufp->regs_allocated = REGS_REALLOCATE; |
| 5223 | } |
| 5224 | else if (bufp->regs_allocated == REGS_REALLOCATE) |
| 5225 | { /* Yes. If we need more elements than were already |
| 5226 | allocated, reallocate them. If we need fewer, just |
| 5227 | leave it alone. */ |
| 5228 | if (regs->num_regs < num_regs + 1) |
| 5229 | { |
| 5230 | regs->num_regs = num_regs + 1; |
| 5231 | RETALLOC (regs->start, regs->num_regs, regoff_t); |
| 5232 | RETALLOC (regs->end, regs->num_regs, regoff_t); |
| 5233 | if (regs->start == NULL || regs->end == NULL) |
| 5234 | { |
| 5235 | FREE_VARIABLES (); |
| 5236 | return -2; |
| 5237 | } |
| 5238 | } |
| 5239 | } |
| 5240 | else |
| 5241 | { |
| 5242 | /* These braces fend off a "empty body in an else-statement" |
| 5243 | warning under GCC when assert expands to nothing. */ |
| 5244 | assert (bufp->regs_allocated == REGS_FIXED); |
| 5245 | } |
| 5246 | |
| 5247 | /* Convert the pointer data in `regstart' and `regend' to |
| 5248 | indices. Register zero has to be set differently, |
| 5249 | since we haven't kept track of any info for it. */ |
| 5250 | if (regs->num_regs > 0) |
| 5251 | { |
| 5252 | regs->start[0] = pos; |
| 5253 | regs->end[0] = POINTER_TO_OFFSET (d); |
| 5254 | } |
| 5255 | |
| 5256 | /* Go through the first `min (num_regs, regs->num_regs)' |
| 5257 | registers, since that is all we initialized. */ |
| 5258 | for (reg = 1; reg < MIN (num_regs, regs->num_regs); reg++) |
| 5259 | { |
| 5260 | if (REG_UNSET (regstart[reg]) || REG_UNSET (regend[reg])) |
| 5261 | regs->start[reg] = regs->end[reg] = -1; |
| 5262 | else |
| 5263 | { |
| 5264 | regs->start[reg] = POINTER_TO_OFFSET (regstart[reg]); |
| 5265 | regs->end[reg] = POINTER_TO_OFFSET (regend[reg]); |
| 5266 | } |
| 5267 | } |
| 5268 | |
| 5269 | /* If the regs structure we return has more elements than |
| 5270 | were in the pattern, set the extra elements to -1. If |
| 5271 | we (re)allocated the registers, this is the case, |
| 5272 | because we always allocate enough to have at least one |
| 5273 | -1 at the end. */ |
| 5274 | for (reg = num_regs; reg < regs->num_regs; reg++) |
| 5275 | regs->start[reg] = regs->end[reg] = -1; |
| 5276 | } /* regs && !bufp->no_sub */ |
| 5277 | |
| 5278 | DEBUG_PRINT ("%u failure points pushed, %u popped (%u remain).\n", |
| 5279 | nfailure_points_pushed, nfailure_points_popped, |
| 5280 | nfailure_points_pushed - nfailure_points_popped); |
| 5281 | DEBUG_PRINT ("%u registers pushed.\n", num_regs_pushed); |
| 5282 | |
| 5283 | dcnt = POINTER_TO_OFFSET (d) - pos; |
| 5284 | |
| 5285 | DEBUG_PRINT ("Returning %td from re_match_2.\n", dcnt); |
| 5286 | |
| 5287 | FREE_VARIABLES (); |
| 5288 | return dcnt; |
| 5289 | } |
| 5290 | |
| 5291 | /* Otherwise match next pattern command. */ |
| 5292 | switch (*p++) |
| 5293 | { |
| 5294 | /* Ignore these. Used to ignore the n of succeed_n's which |
| 5295 | currently have n == 0. */ |
| 5296 | case no_op: |
| 5297 | DEBUG_PRINT ("EXECUTING no_op.\n"); |
| 5298 | break; |
| 5299 | |
| 5300 | case succeed: |
| 5301 | DEBUG_PRINT ("EXECUTING succeed.\n"); |
| 5302 | goto succeed_label; |
| 5303 | |
| 5304 | /* Match the next n pattern characters exactly. The following |
| 5305 | byte in the pattern defines n, and the n bytes after that |
| 5306 | are the characters to match. */ |
| 5307 | case exactn: |
| 5308 | mcnt = *p++; |
| 5309 | DEBUG_PRINT ("EXECUTING exactn %d.\n", mcnt); |
| 5310 | |
| 5311 | /* Remember the start point to rollback upon failure. */ |
| 5312 | dfail = d; |
| 5313 | |
| 5314 | #ifndef emacs |
| 5315 | /* This is written out as an if-else so we don't waste time |
| 5316 | testing `translate' inside the loop. */ |
| 5317 | if (RE_TRANSLATE_P (translate)) |
| 5318 | do |
| 5319 | { |
| 5320 | PREFETCH (); |
| 5321 | if (RE_TRANSLATE (translate, *d) != *p++) |
| 5322 | { |
| 5323 | d = dfail; |
| 5324 | goto fail; |
| 5325 | } |
| 5326 | d++; |
| 5327 | } |
| 5328 | while (--mcnt); |
| 5329 | else |
| 5330 | do |
| 5331 | { |
| 5332 | PREFETCH (); |
| 5333 | if (*d++ != *p++) |
| 5334 | { |
| 5335 | d = dfail; |
| 5336 | goto fail; |
| 5337 | } |
| 5338 | } |
| 5339 | while (--mcnt); |
| 5340 | #else /* emacs */ |
| 5341 | /* The cost of testing `translate' is comparatively small. */ |
| 5342 | if (target_multibyte) |
| 5343 | do |
| 5344 | { |
| 5345 | int pat_charlen, buf_charlen; |
| 5346 | int pat_ch, buf_ch; |
| 5347 | |
| 5348 | PREFETCH (); |
| 5349 | if (multibyte) |
| 5350 | pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen); |
| 5351 | else |
| 5352 | { |
| 5353 | pat_ch = RE_CHAR_TO_MULTIBYTE (*p); |
| 5354 | pat_charlen = 1; |
| 5355 | } |
| 5356 | buf_ch = STRING_CHAR_AND_LENGTH (d, buf_charlen); |
| 5357 | |
| 5358 | if (TRANSLATE (buf_ch) != pat_ch) |
| 5359 | { |
| 5360 | d = dfail; |
| 5361 | goto fail; |
| 5362 | } |
| 5363 | |
| 5364 | p += pat_charlen; |
| 5365 | d += buf_charlen; |
| 5366 | mcnt -= pat_charlen; |
| 5367 | } |
| 5368 | while (mcnt > 0); |
| 5369 | else |
| 5370 | do |
| 5371 | { |
| 5372 | int pat_charlen; |
| 5373 | int pat_ch, buf_ch; |
| 5374 | |
| 5375 | PREFETCH (); |
| 5376 | if (multibyte) |
| 5377 | { |
| 5378 | pat_ch = STRING_CHAR_AND_LENGTH (p, pat_charlen); |
| 5379 | pat_ch = RE_CHAR_TO_UNIBYTE (pat_ch); |
| 5380 | } |
| 5381 | else |
| 5382 | { |
| 5383 | pat_ch = *p; |
| 5384 | pat_charlen = 1; |
| 5385 | } |
| 5386 | buf_ch = RE_CHAR_TO_MULTIBYTE (*d); |
| 5387 | if (! CHAR_BYTE8_P (buf_ch)) |
| 5388 | { |
| 5389 | buf_ch = TRANSLATE (buf_ch); |
| 5390 | buf_ch = RE_CHAR_TO_UNIBYTE (buf_ch); |
| 5391 | if (buf_ch < 0) |
| 5392 | buf_ch = *d; |
| 5393 | } |
| 5394 | else |
| 5395 | buf_ch = *d; |
| 5396 | if (buf_ch != pat_ch) |
| 5397 | { |
| 5398 | d = dfail; |
| 5399 | goto fail; |
| 5400 | } |
| 5401 | p += pat_charlen; |
| 5402 | d++; |
| 5403 | } |
| 5404 | while (--mcnt); |
| 5405 | #endif |
| 5406 | break; |
| 5407 | |
| 5408 | |
| 5409 | /* Match any character except possibly a newline or a null. */ |
| 5410 | case anychar: |
| 5411 | { |
| 5412 | int buf_charlen; |
| 5413 | re_wchar_t buf_ch; |
| 5414 | |
| 5415 | DEBUG_PRINT ("EXECUTING anychar.\n"); |
| 5416 | |
| 5417 | PREFETCH (); |
| 5418 | buf_ch = RE_STRING_CHAR_AND_LENGTH (d, buf_charlen, |
| 5419 | target_multibyte); |
| 5420 | buf_ch = TRANSLATE (buf_ch); |
| 5421 | |
| 5422 | if ((!(bufp->syntax & RE_DOT_NEWLINE) |
| 5423 | && buf_ch == '\n') |
| 5424 | || ((bufp->syntax & RE_DOT_NOT_NULL) |
| 5425 | && buf_ch == '\000')) |
| 5426 | goto fail; |
| 5427 | |
| 5428 | DEBUG_PRINT (" Matched `%d'.\n", *d); |
| 5429 | d += buf_charlen; |
| 5430 | } |
| 5431 | break; |
| 5432 | |
| 5433 | |
| 5434 | case charset: |
| 5435 | case charset_not: |
| 5436 | { |
| 5437 | register unsigned int c; |
| 5438 | boolean not = (re_opcode_t) *(p - 1) == charset_not; |
| 5439 | int len; |
| 5440 | |
| 5441 | /* Start of actual range_table, or end of bitmap if there is no |
| 5442 | range table. */ |
| 5443 | re_char *range_table IF_LINT (= NULL); |
| 5444 | |
| 5445 | /* Nonzero if there is a range table. */ |
| 5446 | int range_table_exists; |
| 5447 | |
| 5448 | /* Number of ranges of range table. This is not included |
| 5449 | in the initial byte-length of the command. */ |
| 5450 | int count = 0; |
| 5451 | |
| 5452 | /* Whether matching against a unibyte character. */ |
| 5453 | boolean unibyte_char = false; |
| 5454 | |
| 5455 | DEBUG_PRINT ("EXECUTING charset%s.\n", not ? "_not" : ""); |
| 5456 | |
| 5457 | range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]); |
| 5458 | |
| 5459 | if (range_table_exists) |
| 5460 | { |
| 5461 | range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */ |
| 5462 | EXTRACT_NUMBER_AND_INCR (count, range_table); |
| 5463 | } |
| 5464 | |
| 5465 | PREFETCH (); |
| 5466 | c = RE_STRING_CHAR_AND_LENGTH (d, len, target_multibyte); |
| 5467 | if (target_multibyte) |
| 5468 | { |
| 5469 | int c1; |
| 5470 | |
| 5471 | c = TRANSLATE (c); |
| 5472 | c1 = RE_CHAR_TO_UNIBYTE (c); |
| 5473 | if (c1 >= 0) |
| 5474 | { |
| 5475 | unibyte_char = true; |
| 5476 | c = c1; |
| 5477 | } |
| 5478 | } |
| 5479 | else |
| 5480 | { |
| 5481 | int c1 = RE_CHAR_TO_MULTIBYTE (c); |
| 5482 | |
| 5483 | if (! CHAR_BYTE8_P (c1)) |
| 5484 | { |
| 5485 | c1 = TRANSLATE (c1); |
| 5486 | c1 = RE_CHAR_TO_UNIBYTE (c1); |
| 5487 | if (c1 >= 0) |
| 5488 | { |
| 5489 | unibyte_char = true; |
| 5490 | c = c1; |
| 5491 | } |
| 5492 | } |
| 5493 | else |
| 5494 | unibyte_char = true; |
| 5495 | } |
| 5496 | |
| 5497 | if (unibyte_char && c < (1 << BYTEWIDTH)) |
| 5498 | { /* Lookup bitmap. */ |
| 5499 | /* Cast to `unsigned' instead of `unsigned char' in |
| 5500 | case the bit list is a full 32 bytes long. */ |
| 5501 | if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH) |
| 5502 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
| 5503 | not = !not; |
| 5504 | } |
| 5505 | #ifdef emacs |
| 5506 | else if (range_table_exists) |
| 5507 | { |
| 5508 | int class_bits = CHARSET_RANGE_TABLE_BITS (&p[-1]); |
| 5509 | |
| 5510 | if ( (class_bits & BIT_LOWER && ISLOWER (c)) |
| 5511 | | (class_bits & BIT_MULTIBYTE) |
| 5512 | | (class_bits & BIT_PUNCT && ISPUNCT (c)) |
| 5513 | | (class_bits & BIT_SPACE && ISSPACE (c)) |
| 5514 | | (class_bits & BIT_UPPER && ISUPPER (c)) |
| 5515 | | (class_bits & BIT_WORD && ISWORD (c))) |
| 5516 | not = !not; |
| 5517 | else |
| 5518 | CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count); |
| 5519 | } |
| 5520 | #endif /* emacs */ |
| 5521 | |
| 5522 | if (range_table_exists) |
| 5523 | p = CHARSET_RANGE_TABLE_END (range_table, count); |
| 5524 | else |
| 5525 | p += CHARSET_BITMAP_SIZE (&p[-1]) + 1; |
| 5526 | |
| 5527 | if (!not) goto fail; |
| 5528 | |
| 5529 | d += len; |
| 5530 | } |
| 5531 | break; |
| 5532 | |
| 5533 | |
| 5534 | /* The beginning of a group is represented by start_memory. |
| 5535 | The argument is the register number. The text |
| 5536 | matched within the group is recorded (in the internal |
| 5537 | registers data structure) under the register number. */ |
| 5538 | case start_memory: |
| 5539 | DEBUG_PRINT ("EXECUTING start_memory %d:\n", *p); |
| 5540 | |
| 5541 | /* In case we need to undo this operation (via backtracking). */ |
| 5542 | PUSH_FAILURE_REG (*p); |
| 5543 | |
| 5544 | regstart[*p] = d; |
| 5545 | regend[*p] = NULL; /* probably unnecessary. -sm */ |
| 5546 | DEBUG_PRINT (" regstart: %td\n", POINTER_TO_OFFSET (regstart[*p])); |
| 5547 | |
| 5548 | /* Move past the register number and inner group count. */ |
| 5549 | p += 1; |
| 5550 | break; |
| 5551 | |
| 5552 | |
| 5553 | /* The stop_memory opcode represents the end of a group. Its |
| 5554 | argument is the same as start_memory's: the register number. */ |
| 5555 | case stop_memory: |
| 5556 | DEBUG_PRINT ("EXECUTING stop_memory %d:\n", *p); |
| 5557 | |
| 5558 | assert (!REG_UNSET (regstart[*p])); |
| 5559 | /* Strictly speaking, there should be code such as: |
| 5560 | |
| 5561 | assert (REG_UNSET (regend[*p])); |
| 5562 | PUSH_FAILURE_REGSTOP ((unsigned int)*p); |
| 5563 | |
| 5564 | But the only info to be pushed is regend[*p] and it is known to |
| 5565 | be UNSET, so there really isn't anything to push. |
| 5566 | Not pushing anything, on the other hand deprives us from the |
| 5567 | guarantee that regend[*p] is UNSET since undoing this operation |
| 5568 | will not reset its value properly. This is not important since |
| 5569 | the value will only be read on the next start_memory or at |
| 5570 | the very end and both events can only happen if this stop_memory |
| 5571 | is *not* undone. */ |
| 5572 | |
| 5573 | regend[*p] = d; |
| 5574 | DEBUG_PRINT (" regend: %td\n", POINTER_TO_OFFSET (regend[*p])); |
| 5575 | |
| 5576 | /* Move past the register number and the inner group count. */ |
| 5577 | p += 1; |
| 5578 | break; |
| 5579 | |
| 5580 | |
| 5581 | /* \<digit> has been turned into a `duplicate' command which is |
| 5582 | followed by the numeric value of <digit> as the register number. */ |
| 5583 | case duplicate: |
| 5584 | { |
| 5585 | register re_char *d2, *dend2; |
| 5586 | int regno = *p++; /* Get which register to match against. */ |
| 5587 | DEBUG_PRINT ("EXECUTING duplicate %d.\n", regno); |
| 5588 | |
| 5589 | /* Can't back reference a group which we've never matched. */ |
| 5590 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) |
| 5591 | goto fail; |
| 5592 | |
| 5593 | /* Where in input to try to start matching. */ |
| 5594 | d2 = regstart[regno]; |
| 5595 | |
| 5596 | /* Remember the start point to rollback upon failure. */ |
| 5597 | dfail = d; |
| 5598 | |
| 5599 | /* Where to stop matching; if both the place to start and |
| 5600 | the place to stop matching are in the same string, then |
| 5601 | set to the place to stop, otherwise, for now have to use |
| 5602 | the end of the first string. */ |
| 5603 | |
| 5604 | dend2 = ((FIRST_STRING_P (regstart[regno]) |
| 5605 | == FIRST_STRING_P (regend[regno])) |
| 5606 | ? regend[regno] : end_match_1); |
| 5607 | for (;;) |
| 5608 | { |
| 5609 | ptrdiff_t dcnt; |
| 5610 | |
| 5611 | /* If necessary, advance to next segment in register |
| 5612 | contents. */ |
| 5613 | while (d2 == dend2) |
| 5614 | { |
| 5615 | if (dend2 == end_match_2) break; |
| 5616 | if (dend2 == regend[regno]) break; |
| 5617 | |
| 5618 | /* End of string1 => advance to string2. */ |
| 5619 | d2 = string2; |
| 5620 | dend2 = regend[regno]; |
| 5621 | } |
| 5622 | /* At end of register contents => success */ |
| 5623 | if (d2 == dend2) break; |
| 5624 | |
| 5625 | /* If necessary, advance to next segment in data. */ |
| 5626 | PREFETCH (); |
| 5627 | |
| 5628 | /* How many characters left in this segment to match. */ |
| 5629 | dcnt = dend - d; |
| 5630 | |
| 5631 | /* Want how many consecutive characters we can match in |
| 5632 | one shot, so, if necessary, adjust the count. */ |
| 5633 | if (dcnt > dend2 - d2) |
| 5634 | dcnt = dend2 - d2; |
| 5635 | |
| 5636 | /* Compare that many; failure if mismatch, else move |
| 5637 | past them. */ |
| 5638 | if (RE_TRANSLATE_P (translate) |
| 5639 | ? bcmp_translate (d, d2, dcnt, translate, target_multibyte) |
| 5640 | : memcmp (d, d2, dcnt)) |
| 5641 | { |
| 5642 | d = dfail; |
| 5643 | goto fail; |
| 5644 | } |
| 5645 | d += dcnt, d2 += dcnt; |
| 5646 | } |
| 5647 | } |
| 5648 | break; |
| 5649 | |
| 5650 | |
| 5651 | /* begline matches the empty string at the beginning of the string |
| 5652 | (unless `not_bol' is set in `bufp'), and after newlines. */ |
| 5653 | case begline: |
| 5654 | DEBUG_PRINT ("EXECUTING begline.\n"); |
| 5655 | |
| 5656 | if (AT_STRINGS_BEG (d)) |
| 5657 | { |
| 5658 | if (!bufp->not_bol) break; |
| 5659 | } |
| 5660 | else |
| 5661 | { |
| 5662 | unsigned c; |
| 5663 | GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2); |
| 5664 | if (c == '\n') |
| 5665 | break; |
| 5666 | } |
| 5667 | /* In all other cases, we fail. */ |
| 5668 | goto fail; |
| 5669 | |
| 5670 | |
| 5671 | /* endline is the dual of begline. */ |
| 5672 | case endline: |
| 5673 | DEBUG_PRINT ("EXECUTING endline.\n"); |
| 5674 | |
| 5675 | if (AT_STRINGS_END (d)) |
| 5676 | { |
| 5677 | if (!bufp->not_eol) break; |
| 5678 | } |
| 5679 | else |
| 5680 | { |
| 5681 | PREFETCH_NOLIMIT (); |
| 5682 | if (*d == '\n') |
| 5683 | break; |
| 5684 | } |
| 5685 | goto fail; |
| 5686 | |
| 5687 | |
| 5688 | /* Match at the very beginning of the data. */ |
| 5689 | case begbuf: |
| 5690 | DEBUG_PRINT ("EXECUTING begbuf.\n"); |
| 5691 | if (AT_STRINGS_BEG (d)) |
| 5692 | break; |
| 5693 | goto fail; |
| 5694 | |
| 5695 | |
| 5696 | /* Match at the very end of the data. */ |
| 5697 | case endbuf: |
| 5698 | DEBUG_PRINT ("EXECUTING endbuf.\n"); |
| 5699 | if (AT_STRINGS_END (d)) |
| 5700 | break; |
| 5701 | goto fail; |
| 5702 | |
| 5703 | |
| 5704 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It |
| 5705 | pushes NULL as the value for the string on the stack. Then |
| 5706 | `POP_FAILURE_POINT' will keep the current value for the |
| 5707 | string, instead of restoring it. To see why, consider |
| 5708 | matching `foo\nbar' against `.*\n'. The .* matches the foo; |
| 5709 | then the . fails against the \n. But the next thing we want |
| 5710 | to do is match the \n against the \n; if we restored the |
| 5711 | string value, we would be back at the foo. |
| 5712 | |
| 5713 | Because this is used only in specific cases, we don't need to |
| 5714 | check all the things that `on_failure_jump' does, to make |
| 5715 | sure the right things get saved on the stack. Hence we don't |
| 5716 | share its code. The only reason to push anything on the |
| 5717 | stack at all is that otherwise we would have to change |
| 5718 | `anychar's code to do something besides goto fail in this |
| 5719 | case; that seems worse than this. */ |
| 5720 | case on_failure_keep_string_jump: |
| 5721 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 5722 | DEBUG_PRINT ("EXECUTING on_failure_keep_string_jump %d (to %p):\n", |
| 5723 | mcnt, p + mcnt); |
| 5724 | |
| 5725 | PUSH_FAILURE_POINT (p - 3, NULL); |
| 5726 | break; |
| 5727 | |
| 5728 | /* A nasty loop is introduced by the non-greedy *? and +?. |
| 5729 | With such loops, the stack only ever contains one failure point |
| 5730 | at a time, so that a plain on_failure_jump_loop kind of |
| 5731 | cycle detection cannot work. Worse yet, such a detection |
| 5732 | can not only fail to detect a cycle, but it can also wrongly |
| 5733 | detect a cycle (between different instantiations of the same |
| 5734 | loop). |
| 5735 | So the method used for those nasty loops is a little different: |
| 5736 | We use a special cycle-detection-stack-frame which is pushed |
| 5737 | when the on_failure_jump_nastyloop failure-point is *popped*. |
| 5738 | This special frame thus marks the beginning of one iteration |
| 5739 | through the loop and we can hence easily check right here |
| 5740 | whether something matched between the beginning and the end of |
| 5741 | the loop. */ |
| 5742 | case on_failure_jump_nastyloop: |
| 5743 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 5744 | DEBUG_PRINT ("EXECUTING on_failure_jump_nastyloop %d (to %p):\n", |
| 5745 | mcnt, p + mcnt); |
| 5746 | |
| 5747 | assert ((re_opcode_t)p[-4] == no_op); |
| 5748 | { |
| 5749 | int cycle = 0; |
| 5750 | CHECK_INFINITE_LOOP (p - 4, d); |
| 5751 | if (!cycle) |
| 5752 | /* If there's a cycle, just continue without pushing |
| 5753 | this failure point. The failure point is the "try again" |
| 5754 | option, which shouldn't be tried. |
| 5755 | We want (x?)*?y\1z to match both xxyz and xxyxz. */ |
| 5756 | PUSH_FAILURE_POINT (p - 3, d); |
| 5757 | } |
| 5758 | break; |
| 5759 | |
| 5760 | /* Simple loop detecting on_failure_jump: just check on the |
| 5761 | failure stack if the same spot was already hit earlier. */ |
| 5762 | case on_failure_jump_loop: |
| 5763 | on_failure: |
| 5764 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 5765 | DEBUG_PRINT ("EXECUTING on_failure_jump_loop %d (to %p):\n", |
| 5766 | mcnt, p + mcnt); |
| 5767 | { |
| 5768 | int cycle = 0; |
| 5769 | CHECK_INFINITE_LOOP (p - 3, d); |
| 5770 | if (cycle) |
| 5771 | /* If there's a cycle, get out of the loop, as if the matching |
| 5772 | had failed. We used to just `goto fail' here, but that was |
| 5773 | aborting the search a bit too early: we want to keep the |
| 5774 | empty-loop-match and keep matching after the loop. |
| 5775 | We want (x?)*y\1z to match both xxyz and xxyxz. */ |
| 5776 | p += mcnt; |
| 5777 | else |
| 5778 | PUSH_FAILURE_POINT (p - 3, d); |
| 5779 | } |
| 5780 | break; |
| 5781 | |
| 5782 | |
| 5783 | /* Uses of on_failure_jump: |
| 5784 | |
| 5785 | Each alternative starts with an on_failure_jump that points |
| 5786 | to the beginning of the next alternative. Each alternative |
| 5787 | except the last ends with a jump that in effect jumps past |
| 5788 | the rest of the alternatives. (They really jump to the |
| 5789 | ending jump of the following alternative, because tensioning |
| 5790 | these jumps is a hassle.) |
| 5791 | |
| 5792 | Repeats start with an on_failure_jump that points past both |
| 5793 | the repetition text and either the following jump or |
| 5794 | pop_failure_jump back to this on_failure_jump. */ |
| 5795 | case on_failure_jump: |
| 5796 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 5797 | DEBUG_PRINT ("EXECUTING on_failure_jump %d (to %p):\n", |
| 5798 | mcnt, p + mcnt); |
| 5799 | |
| 5800 | PUSH_FAILURE_POINT (p -3, d); |
| 5801 | break; |
| 5802 | |
| 5803 | /* This operation is used for greedy *. |
| 5804 | Compare the beginning of the repeat with what in the |
| 5805 | pattern follows its end. If we can establish that there |
| 5806 | is nothing that they would both match, i.e., that we |
| 5807 | would have to backtrack because of (as in, e.g., `a*a') |
| 5808 | then we can use a non-backtracking loop based on |
| 5809 | on_failure_keep_string_jump instead of on_failure_jump. */ |
| 5810 | case on_failure_jump_smart: |
| 5811 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 5812 | DEBUG_PRINT ("EXECUTING on_failure_jump_smart %d (to %p).\n", |
| 5813 | mcnt, p + mcnt); |
| 5814 | { |
| 5815 | re_char *p1 = p; /* Next operation. */ |
| 5816 | /* Here, we discard `const', making re_match non-reentrant. */ |
| 5817 | unsigned char *p2 = (unsigned char*) p + mcnt; /* Jump dest. */ |
| 5818 | unsigned char *p3 = (unsigned char*) p - 3; /* opcode location. */ |
| 5819 | |
| 5820 | p -= 3; /* Reset so that we will re-execute the |
| 5821 | instruction once it's been changed. */ |
| 5822 | |
| 5823 | EXTRACT_NUMBER (mcnt, p2 - 2); |
| 5824 | |
| 5825 | /* Ensure this is a indeed the trivial kind of loop |
| 5826 | we are expecting. */ |
| 5827 | assert (skip_one_char (p1) == p2 - 3); |
| 5828 | assert ((re_opcode_t) p2[-3] == jump && p2 + mcnt == p); |
| 5829 | DEBUG_STATEMENT (debug += 2); |
| 5830 | if (mutually_exclusive_p (bufp, p1, p2)) |
| 5831 | { |
| 5832 | /* Use a fast `on_failure_keep_string_jump' loop. */ |
| 5833 | DEBUG_PRINT (" smart exclusive => fast loop.\n"); |
| 5834 | *p3 = (unsigned char) on_failure_keep_string_jump; |
| 5835 | STORE_NUMBER (p2 - 2, mcnt + 3); |
| 5836 | } |
| 5837 | else |
| 5838 | { |
| 5839 | /* Default to a safe `on_failure_jump' loop. */ |
| 5840 | DEBUG_PRINT (" smart default => slow loop.\n"); |
| 5841 | *p3 = (unsigned char) on_failure_jump; |
| 5842 | } |
| 5843 | DEBUG_STATEMENT (debug -= 2); |
| 5844 | } |
| 5845 | break; |
| 5846 | |
| 5847 | /* Unconditionally jump (without popping any failure points). */ |
| 5848 | case jump: |
| 5849 | unconditional_jump: |
| 5850 | IMMEDIATE_QUIT_CHECK; |
| 5851 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
| 5852 | DEBUG_PRINT ("EXECUTING jump %d ", mcnt); |
| 5853 | p += mcnt; /* Do the jump. */ |
| 5854 | DEBUG_PRINT ("(to %p).\n", p); |
| 5855 | break; |
| 5856 | |
| 5857 | |
| 5858 | /* Have to succeed matching what follows at least n times. |
| 5859 | After that, handle like `on_failure_jump'. */ |
| 5860 | case succeed_n: |
| 5861 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
| 5862 | EXTRACT_NUMBER (mcnt, p + 2); |
| 5863 | DEBUG_PRINT ("EXECUTING succeed_n %d.\n", mcnt); |
| 5864 | |
| 5865 | /* Originally, mcnt is how many times we HAVE to succeed. */ |
| 5866 | if (mcnt != 0) |
| 5867 | { |
| 5868 | /* Here, we discard `const', making re_match non-reentrant. */ |
| 5869 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ |
| 5870 | mcnt--; |
| 5871 | p += 4; |
| 5872 | PUSH_NUMBER (p2, mcnt); |
| 5873 | } |
| 5874 | else |
| 5875 | /* The two bytes encoding mcnt == 0 are two no_op opcodes. */ |
| 5876 | goto on_failure; |
| 5877 | break; |
| 5878 | |
| 5879 | case jump_n: |
| 5880 | /* Signedness doesn't matter since we only compare MCNT to 0. */ |
| 5881 | EXTRACT_NUMBER (mcnt, p + 2); |
| 5882 | DEBUG_PRINT ("EXECUTING jump_n %d.\n", mcnt); |
| 5883 | |
| 5884 | /* Originally, this is how many times we CAN jump. */ |
| 5885 | if (mcnt != 0) |
| 5886 | { |
| 5887 | /* Here, we discard `const', making re_match non-reentrant. */ |
| 5888 | unsigned char *p2 = (unsigned char*) p + 2; /* counter loc. */ |
| 5889 | mcnt--; |
| 5890 | PUSH_NUMBER (p2, mcnt); |
| 5891 | goto unconditional_jump; |
| 5892 | } |
| 5893 | /* If don't have to jump any more, skip over the rest of command. */ |
| 5894 | else |
| 5895 | p += 4; |
| 5896 | break; |
| 5897 | |
| 5898 | case set_number_at: |
| 5899 | { |
| 5900 | unsigned char *p2; /* Location of the counter. */ |
| 5901 | DEBUG_PRINT ("EXECUTING set_number_at.\n"); |
| 5902 | |
| 5903 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 5904 | /* Here, we discard `const', making re_match non-reentrant. */ |
| 5905 | p2 = (unsigned char*) p + mcnt; |
| 5906 | /* Signedness doesn't matter since we only copy MCNT's bits . */ |
| 5907 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
| 5908 | DEBUG_PRINT (" Setting %p to %d.\n", p2, mcnt); |
| 5909 | PUSH_NUMBER (p2, mcnt); |
| 5910 | break; |
| 5911 | } |
| 5912 | |
| 5913 | case wordbound: |
| 5914 | case notwordbound: |
| 5915 | { |
| 5916 | boolean not = (re_opcode_t) *(p - 1) == notwordbound; |
| 5917 | DEBUG_PRINT ("EXECUTING %swordbound.\n", not ? "not" : ""); |
| 5918 | |
| 5919 | /* We SUCCEED (or FAIL) in one of the following cases: */ |
| 5920 | |
| 5921 | /* Case 1: D is at the beginning or the end of string. */ |
| 5922 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
| 5923 | not = !not; |
| 5924 | else |
| 5925 | { |
| 5926 | /* C1 is the character before D, S1 is the syntax of C1, C2 |
| 5927 | is the character at D, and S2 is the syntax of C2. */ |
| 5928 | re_wchar_t c1, c2; |
| 5929 | int s1, s2; |
| 5930 | int dummy; |
| 5931 | #ifdef emacs |
| 5932 | ssize_t offset = PTR_TO_OFFSET (d - 1); |
| 5933 | ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); |
| 5934 | UPDATE_SYNTAX_TABLE (charpos); |
| 5935 | #endif |
| 5936 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
| 5937 | s1 = SYNTAX (c1); |
| 5938 | #ifdef emacs |
| 5939 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
| 5940 | #endif |
| 5941 | PREFETCH_NOLIMIT (); |
| 5942 | GET_CHAR_AFTER (c2, d, dummy); |
| 5943 | s2 = SYNTAX (c2); |
| 5944 | |
| 5945 | if (/* Case 2: Only one of S1 and S2 is Sword. */ |
| 5946 | ((s1 == Sword) != (s2 == Sword)) |
| 5947 | /* Case 3: Both of S1 and S2 are Sword, and macro |
| 5948 | WORD_BOUNDARY_P (C1, C2) returns nonzero. */ |
| 5949 | || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2))) |
| 5950 | not = !not; |
| 5951 | } |
| 5952 | if (not) |
| 5953 | break; |
| 5954 | else |
| 5955 | goto fail; |
| 5956 | } |
| 5957 | |
| 5958 | case wordbeg: |
| 5959 | DEBUG_PRINT ("EXECUTING wordbeg.\n"); |
| 5960 | |
| 5961 | /* We FAIL in one of the following cases: */ |
| 5962 | |
| 5963 | /* Case 1: D is at the end of string. */ |
| 5964 | if (AT_STRINGS_END (d)) |
| 5965 | goto fail; |
| 5966 | else |
| 5967 | { |
| 5968 | /* C1 is the character before D, S1 is the syntax of C1, C2 |
| 5969 | is the character at D, and S2 is the syntax of C2. */ |
| 5970 | re_wchar_t c1, c2; |
| 5971 | int s1, s2; |
| 5972 | int dummy; |
| 5973 | #ifdef emacs |
| 5974 | ssize_t offset = PTR_TO_OFFSET (d); |
| 5975 | ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); |
| 5976 | UPDATE_SYNTAX_TABLE (charpos); |
| 5977 | #endif |
| 5978 | PREFETCH (); |
| 5979 | GET_CHAR_AFTER (c2, d, dummy); |
| 5980 | s2 = SYNTAX (c2); |
| 5981 | |
| 5982 | /* Case 2: S2 is not Sword. */ |
| 5983 | if (s2 != Sword) |
| 5984 | goto fail; |
| 5985 | |
| 5986 | /* Case 3: D is not at the beginning of string ... */ |
| 5987 | if (!AT_STRINGS_BEG (d)) |
| 5988 | { |
| 5989 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
| 5990 | #ifdef emacs |
| 5991 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); |
| 5992 | #endif |
| 5993 | s1 = SYNTAX (c1); |
| 5994 | |
| 5995 | /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2) |
| 5996 | returns 0. */ |
| 5997 | if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
| 5998 | goto fail; |
| 5999 | } |
| 6000 | } |
| 6001 | break; |
| 6002 | |
| 6003 | case wordend: |
| 6004 | DEBUG_PRINT ("EXECUTING wordend.\n"); |
| 6005 | |
| 6006 | /* We FAIL in one of the following cases: */ |
| 6007 | |
| 6008 | /* Case 1: D is at the beginning of string. */ |
| 6009 | if (AT_STRINGS_BEG (d)) |
| 6010 | goto fail; |
| 6011 | else |
| 6012 | { |
| 6013 | /* C1 is the character before D, S1 is the syntax of C1, C2 |
| 6014 | is the character at D, and S2 is the syntax of C2. */ |
| 6015 | re_wchar_t c1, c2; |
| 6016 | int s1, s2; |
| 6017 | int dummy; |
| 6018 | #ifdef emacs |
| 6019 | ssize_t offset = PTR_TO_OFFSET (d) - 1; |
| 6020 | ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); |
| 6021 | UPDATE_SYNTAX_TABLE (charpos); |
| 6022 | #endif |
| 6023 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
| 6024 | s1 = SYNTAX (c1); |
| 6025 | |
| 6026 | /* Case 2: S1 is not Sword. */ |
| 6027 | if (s1 != Sword) |
| 6028 | goto fail; |
| 6029 | |
| 6030 | /* Case 3: D is not at the end of string ... */ |
| 6031 | if (!AT_STRINGS_END (d)) |
| 6032 | { |
| 6033 | PREFETCH_NOLIMIT (); |
| 6034 | GET_CHAR_AFTER (c2, d, dummy); |
| 6035 | #ifdef emacs |
| 6036 | UPDATE_SYNTAX_TABLE_FORWARD (charpos); |
| 6037 | #endif |
| 6038 | s2 = SYNTAX (c2); |
| 6039 | |
| 6040 | /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2) |
| 6041 | returns 0. */ |
| 6042 | if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2)) |
| 6043 | goto fail; |
| 6044 | } |
| 6045 | } |
| 6046 | break; |
| 6047 | |
| 6048 | case symbeg: |
| 6049 | DEBUG_PRINT ("EXECUTING symbeg.\n"); |
| 6050 | |
| 6051 | /* We FAIL in one of the following cases: */ |
| 6052 | |
| 6053 | /* Case 1: D is at the end of string. */ |
| 6054 | if (AT_STRINGS_END (d)) |
| 6055 | goto fail; |
| 6056 | else |
| 6057 | { |
| 6058 | /* C1 is the character before D, S1 is the syntax of C1, C2 |
| 6059 | is the character at D, and S2 is the syntax of C2. */ |
| 6060 | re_wchar_t c1, c2; |
| 6061 | int s1, s2; |
| 6062 | #ifdef emacs |
| 6063 | ssize_t offset = PTR_TO_OFFSET (d); |
| 6064 | ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); |
| 6065 | UPDATE_SYNTAX_TABLE (charpos); |
| 6066 | #endif |
| 6067 | PREFETCH (); |
| 6068 | c2 = RE_STRING_CHAR (d, target_multibyte); |
| 6069 | s2 = SYNTAX (c2); |
| 6070 | |
| 6071 | /* Case 2: S2 is neither Sword nor Ssymbol. */ |
| 6072 | if (s2 != Sword && s2 != Ssymbol) |
| 6073 | goto fail; |
| 6074 | |
| 6075 | /* Case 3: D is not at the beginning of string ... */ |
| 6076 | if (!AT_STRINGS_BEG (d)) |
| 6077 | { |
| 6078 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
| 6079 | #ifdef emacs |
| 6080 | UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1); |
| 6081 | #endif |
| 6082 | s1 = SYNTAX (c1); |
| 6083 | |
| 6084 | /* ... and S1 is Sword or Ssymbol. */ |
| 6085 | if (s1 == Sword || s1 == Ssymbol) |
| 6086 | goto fail; |
| 6087 | } |
| 6088 | } |
| 6089 | break; |
| 6090 | |
| 6091 | case symend: |
| 6092 | DEBUG_PRINT ("EXECUTING symend.\n"); |
| 6093 | |
| 6094 | /* We FAIL in one of the following cases: */ |
| 6095 | |
| 6096 | /* Case 1: D is at the beginning of string. */ |
| 6097 | if (AT_STRINGS_BEG (d)) |
| 6098 | goto fail; |
| 6099 | else |
| 6100 | { |
| 6101 | /* C1 is the character before D, S1 is the syntax of C1, C2 |
| 6102 | is the character at D, and S2 is the syntax of C2. */ |
| 6103 | re_wchar_t c1, c2; |
| 6104 | int s1, s2; |
| 6105 | #ifdef emacs |
| 6106 | ssize_t offset = PTR_TO_OFFSET (d) - 1; |
| 6107 | ssize_t charpos = SYNTAX_TABLE_BYTE_TO_CHAR (offset); |
| 6108 | UPDATE_SYNTAX_TABLE (charpos); |
| 6109 | #endif |
| 6110 | GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2); |
| 6111 | s1 = SYNTAX (c1); |
| 6112 | |
| 6113 | /* Case 2: S1 is neither Ssymbol nor Sword. */ |
| 6114 | if (s1 != Sword && s1 != Ssymbol) |
| 6115 | goto fail; |
| 6116 | |
| 6117 | /* Case 3: D is not at the end of string ... */ |
| 6118 | if (!AT_STRINGS_END (d)) |
| 6119 | { |
| 6120 | PREFETCH_NOLIMIT (); |
| 6121 | c2 = RE_STRING_CHAR (d, target_multibyte); |
| 6122 | #ifdef emacs |
| 6123 | UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1); |
| 6124 | #endif |
| 6125 | s2 = SYNTAX (c2); |
| 6126 | |
| 6127 | /* ... and S2 is Sword or Ssymbol. */ |
| 6128 | if (s2 == Sword || s2 == Ssymbol) |
| 6129 | goto fail; |
| 6130 | } |
| 6131 | } |
| 6132 | break; |
| 6133 | |
| 6134 | case syntaxspec: |
| 6135 | case notsyntaxspec: |
| 6136 | { |
| 6137 | boolean not = (re_opcode_t) *(p - 1) == notsyntaxspec; |
| 6138 | mcnt = *p++; |
| 6139 | DEBUG_PRINT ("EXECUTING %ssyntaxspec %d.\n", not ? "not" : "", |
| 6140 | mcnt); |
| 6141 | PREFETCH (); |
| 6142 | #ifdef emacs |
| 6143 | { |
| 6144 | ssize_t offset = PTR_TO_OFFSET (d); |
| 6145 | ssize_t pos1 = SYNTAX_TABLE_BYTE_TO_CHAR (offset); |
| 6146 | UPDATE_SYNTAX_TABLE (pos1); |
| 6147 | } |
| 6148 | #endif |
| 6149 | { |
| 6150 | int len; |
| 6151 | re_wchar_t c; |
| 6152 | |
| 6153 | GET_CHAR_AFTER (c, d, len); |
| 6154 | if ((SYNTAX (c) != (enum syntaxcode) mcnt) ^ not) |
| 6155 | goto fail; |
| 6156 | d += len; |
| 6157 | } |
| 6158 | } |
| 6159 | break; |
| 6160 | |
| 6161 | #ifdef emacs |
| 6162 | case before_dot: |
| 6163 | DEBUG_PRINT ("EXECUTING before_dot.\n"); |
| 6164 | if (PTR_BYTE_POS (d) >= PT_BYTE) |
| 6165 | goto fail; |
| 6166 | break; |
| 6167 | |
| 6168 | case at_dot: |
| 6169 | DEBUG_PRINT ("EXECUTING at_dot.\n"); |
| 6170 | if (PTR_BYTE_POS (d) != PT_BYTE) |
| 6171 | goto fail; |
| 6172 | break; |
| 6173 | |
| 6174 | case after_dot: |
| 6175 | DEBUG_PRINT ("EXECUTING after_dot.\n"); |
| 6176 | if (PTR_BYTE_POS (d) <= PT_BYTE) |
| 6177 | goto fail; |
| 6178 | break; |
| 6179 | |
| 6180 | case categoryspec: |
| 6181 | case notcategoryspec: |
| 6182 | { |
| 6183 | boolean not = (re_opcode_t) *(p - 1) == notcategoryspec; |
| 6184 | mcnt = *p++; |
| 6185 | DEBUG_PRINT ("EXECUTING %scategoryspec %d.\n", |
| 6186 | not ? "not" : "", mcnt); |
| 6187 | PREFETCH (); |
| 6188 | |
| 6189 | { |
| 6190 | int len; |
| 6191 | re_wchar_t c; |
| 6192 | GET_CHAR_AFTER (c, d, len); |
| 6193 | if ((!CHAR_HAS_CATEGORY (c, mcnt)) ^ not) |
| 6194 | goto fail; |
| 6195 | d += len; |
| 6196 | } |
| 6197 | } |
| 6198 | break; |
| 6199 | |
| 6200 | #endif /* emacs */ |
| 6201 | |
| 6202 | default: |
| 6203 | abort (); |
| 6204 | } |
| 6205 | continue; /* Successfully executed one pattern command; keep going. */ |
| 6206 | |
| 6207 | |
| 6208 | /* We goto here if a matching operation fails. */ |
| 6209 | fail: |
| 6210 | IMMEDIATE_QUIT_CHECK; |
| 6211 | if (!FAIL_STACK_EMPTY ()) |
| 6212 | { |
| 6213 | re_char *str, *pat; |
| 6214 | /* A restart point is known. Restore to that state. */ |
| 6215 | DEBUG_PRINT ("\nFAIL:\n"); |
| 6216 | POP_FAILURE_POINT (str, pat); |
| 6217 | switch (*pat++) |
| 6218 | { |
| 6219 | case on_failure_keep_string_jump: |
| 6220 | assert (str == NULL); |
| 6221 | goto continue_failure_jump; |
| 6222 | |
| 6223 | case on_failure_jump_nastyloop: |
| 6224 | assert ((re_opcode_t)pat[-2] == no_op); |
| 6225 | PUSH_FAILURE_POINT (pat - 2, str); |
| 6226 | /* Fallthrough */ |
| 6227 | |
| 6228 | case on_failure_jump_loop: |
| 6229 | case on_failure_jump: |
| 6230 | case succeed_n: |
| 6231 | d = str; |
| 6232 | continue_failure_jump: |
| 6233 | EXTRACT_NUMBER_AND_INCR (mcnt, pat); |
| 6234 | p = pat + mcnt; |
| 6235 | break; |
| 6236 | |
| 6237 | case no_op: |
| 6238 | /* A special frame used for nastyloops. */ |
| 6239 | goto fail; |
| 6240 | |
| 6241 | default: |
| 6242 | abort (); |
| 6243 | } |
| 6244 | |
| 6245 | assert (p >= bufp->buffer && p <= pend); |
| 6246 | |
| 6247 | if (d >= string1 && d <= end1) |
| 6248 | dend = end_match_1; |
| 6249 | } |
| 6250 | else |
| 6251 | break; /* Matching at this starting point really fails. */ |
| 6252 | } /* for (;;) */ |
| 6253 | |
| 6254 | if (best_regs_set) |
| 6255 | goto restore_best_regs; |
| 6256 | |
| 6257 | FREE_VARIABLES (); |
| 6258 | |
| 6259 | return -1; /* Failure to match. */ |
| 6260 | } |
| 6261 | \f |
| 6262 | /* Subroutine definitions for re_match_2. */ |
| 6263 | |
| 6264 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN |
| 6265 | bytes; nonzero otherwise. */ |
| 6266 | |
| 6267 | static int |
| 6268 | bcmp_translate (const re_char *s1, const re_char *s2, register ssize_t len, |
| 6269 | RE_TRANSLATE_TYPE translate, const int target_multibyte) |
| 6270 | { |
| 6271 | register re_char *p1 = s1, *p2 = s2; |
| 6272 | re_char *p1_end = s1 + len; |
| 6273 | re_char *p2_end = s2 + len; |
| 6274 | |
| 6275 | /* FIXME: Checking both p1 and p2 presumes that the two strings might have |
| 6276 | different lengths, but relying on a single `len' would break this. -sm */ |
| 6277 | while (p1 < p1_end && p2 < p2_end) |
| 6278 | { |
| 6279 | int p1_charlen, p2_charlen; |
| 6280 | re_wchar_t p1_ch, p2_ch; |
| 6281 | |
| 6282 | GET_CHAR_AFTER (p1_ch, p1, p1_charlen); |
| 6283 | GET_CHAR_AFTER (p2_ch, p2, p2_charlen); |
| 6284 | |
| 6285 | if (RE_TRANSLATE (translate, p1_ch) |
| 6286 | != RE_TRANSLATE (translate, p2_ch)) |
| 6287 | return 1; |
| 6288 | |
| 6289 | p1 += p1_charlen, p2 += p2_charlen; |
| 6290 | } |
| 6291 | |
| 6292 | if (p1 != p1_end || p2 != p2_end) |
| 6293 | return 1; |
| 6294 | |
| 6295 | return 0; |
| 6296 | } |
| 6297 | \f |
| 6298 | /* Entry points for GNU code. */ |
| 6299 | |
| 6300 | /* re_compile_pattern is the GNU regular expression compiler: it |
| 6301 | compiles PATTERN (of length SIZE) and puts the result in BUFP. |
| 6302 | Returns 0 if the pattern was valid, otherwise an error string. |
| 6303 | |
| 6304 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
| 6305 | are set in BUFP on entry. |
| 6306 | |
| 6307 | We call regex_compile to do the actual compilation. */ |
| 6308 | |
| 6309 | const char * |
| 6310 | re_compile_pattern (const char *pattern, size_t length, |
| 6311 | struct re_pattern_buffer *bufp) |
| 6312 | { |
| 6313 | reg_errcode_t ret; |
| 6314 | |
| 6315 | /* GNU code is written to assume at least RE_NREGS registers will be set |
| 6316 | (and at least one extra will be -1). */ |
| 6317 | bufp->regs_allocated = REGS_UNALLOCATED; |
| 6318 | |
| 6319 | /* And GNU code determines whether or not to get register information |
| 6320 | by passing null for the REGS argument to re_match, etc., not by |
| 6321 | setting no_sub. */ |
| 6322 | bufp->no_sub = 0; |
| 6323 | |
| 6324 | ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp); |
| 6325 | |
| 6326 | if (!ret) |
| 6327 | return NULL; |
| 6328 | return gettext (re_error_msgid[(int) ret]); |
| 6329 | } |
| 6330 | WEAK_ALIAS (__re_compile_pattern, re_compile_pattern) |
| 6331 | \f |
| 6332 | /* Entry points compatible with 4.2 BSD regex library. We don't define |
| 6333 | them unless specifically requested. */ |
| 6334 | |
| 6335 | #if defined _REGEX_RE_COMP || defined _LIBC |
| 6336 | |
| 6337 | /* BSD has one and only one pattern buffer. */ |
| 6338 | static struct re_pattern_buffer re_comp_buf; |
| 6339 | |
| 6340 | char * |
| 6341 | # ifdef _LIBC |
| 6342 | /* Make these definitions weak in libc, so POSIX programs can redefine |
| 6343 | these names if they don't use our functions, and still use |
| 6344 | regcomp/regexec below without link errors. */ |
| 6345 | weak_function |
| 6346 | # endif |
| 6347 | re_comp (const char *s) |
| 6348 | { |
| 6349 | reg_errcode_t ret; |
| 6350 | |
| 6351 | if (!s) |
| 6352 | { |
| 6353 | if (!re_comp_buf.buffer) |
| 6354 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
| 6355 | return (char *) gettext ("No previous regular expression"); |
| 6356 | return 0; |
| 6357 | } |
| 6358 | |
| 6359 | if (!re_comp_buf.buffer) |
| 6360 | { |
| 6361 | re_comp_buf.buffer = malloc (200); |
| 6362 | if (re_comp_buf.buffer == NULL) |
| 6363 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
| 6364 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); |
| 6365 | re_comp_buf.allocated = 200; |
| 6366 | |
| 6367 | re_comp_buf.fastmap = malloc (1 << BYTEWIDTH); |
| 6368 | if (re_comp_buf.fastmap == NULL) |
| 6369 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
| 6370 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); |
| 6371 | } |
| 6372 | |
| 6373 | /* Since `re_exec' always passes NULL for the `regs' argument, we |
| 6374 | don't need to initialize the pattern buffer fields which affect it. */ |
| 6375 | |
| 6376 | ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); |
| 6377 | |
| 6378 | if (!ret) |
| 6379 | return NULL; |
| 6380 | |
| 6381 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
| 6382 | return (char *) gettext (re_error_msgid[(int) ret]); |
| 6383 | } |
| 6384 | |
| 6385 | |
| 6386 | int |
| 6387 | # ifdef _LIBC |
| 6388 | weak_function |
| 6389 | # endif |
| 6390 | re_exec (const char *s) |
| 6391 | { |
| 6392 | const size_t len = strlen (s); |
| 6393 | return (re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0) |
| 6394 | >= 0); |
| 6395 | } |
| 6396 | #endif /* _REGEX_RE_COMP */ |
| 6397 | \f |
| 6398 | /* POSIX.2 functions. Don't define these for Emacs. */ |
| 6399 | |
| 6400 | #ifndef emacs |
| 6401 | |
| 6402 | /* regcomp takes a regular expression as a string and compiles it. |
| 6403 | |
| 6404 | PREG is a regex_t *. We do not expect any fields to be initialized, |
| 6405 | since POSIX says we shouldn't. Thus, we set |
| 6406 | |
| 6407 | `buffer' to the compiled pattern; |
| 6408 | `used' to the length of the compiled pattern; |
| 6409 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the |
| 6410 | REG_EXTENDED bit in CFLAGS is set; otherwise, to |
| 6411 | RE_SYNTAX_POSIX_BASIC; |
| 6412 | `fastmap' to an allocated space for the fastmap; |
| 6413 | `fastmap_accurate' to zero; |
| 6414 | `re_nsub' to the number of subexpressions in PATTERN. |
| 6415 | |
| 6416 | PATTERN is the address of the pattern string. |
| 6417 | |
| 6418 | CFLAGS is a series of bits which affect compilation. |
| 6419 | |
| 6420 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we |
| 6421 | use POSIX basic syntax. |
| 6422 | |
| 6423 | If REG_NEWLINE is set, then . and [^...] don't match newline. |
| 6424 | Also, regexec will try a match beginning after every newline. |
| 6425 | |
| 6426 | If REG_ICASE is set, then we considers upper- and lowercase |
| 6427 | versions of letters to be equivalent when matching. |
| 6428 | |
| 6429 | If REG_NOSUB is set, then when PREG is passed to regexec, that |
| 6430 | routine will report only success or failure, and nothing about the |
| 6431 | registers. |
| 6432 | |
| 6433 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
| 6434 | the return codes and their meanings.) */ |
| 6435 | |
| 6436 | reg_errcode_t |
| 6437 | regcomp (regex_t *__restrict preg, const char *__restrict pattern, |
| 6438 | int cflags) |
| 6439 | { |
| 6440 | reg_errcode_t ret; |
| 6441 | reg_syntax_t syntax |
| 6442 | = (cflags & REG_EXTENDED) ? |
| 6443 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; |
| 6444 | |
| 6445 | /* regex_compile will allocate the space for the compiled pattern. */ |
| 6446 | preg->buffer = 0; |
| 6447 | preg->allocated = 0; |
| 6448 | preg->used = 0; |
| 6449 | |
| 6450 | /* Try to allocate space for the fastmap. */ |
| 6451 | preg->fastmap = malloc (1 << BYTEWIDTH); |
| 6452 | |
| 6453 | if (cflags & REG_ICASE) |
| 6454 | { |
| 6455 | unsigned i; |
| 6456 | |
| 6457 | preg->translate = malloc (CHAR_SET_SIZE * sizeof *preg->translate); |
| 6458 | if (preg->translate == NULL) |
| 6459 | return (int) REG_ESPACE; |
| 6460 | |
| 6461 | /* Map uppercase characters to corresponding lowercase ones. */ |
| 6462 | for (i = 0; i < CHAR_SET_SIZE; i++) |
| 6463 | preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; |
| 6464 | } |
| 6465 | else |
| 6466 | preg->translate = NULL; |
| 6467 | |
| 6468 | /* If REG_NEWLINE is set, newlines are treated differently. */ |
| 6469 | if (cflags & REG_NEWLINE) |
| 6470 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ |
| 6471 | syntax &= ~RE_DOT_NEWLINE; |
| 6472 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; |
| 6473 | } |
| 6474 | else |
| 6475 | syntax |= RE_NO_NEWLINE_ANCHOR; |
| 6476 | |
| 6477 | preg->no_sub = !!(cflags & REG_NOSUB); |
| 6478 | |
| 6479 | /* POSIX says a null character in the pattern terminates it, so we |
| 6480 | can use strlen here in compiling the pattern. */ |
| 6481 | ret = regex_compile ((re_char*) pattern, strlen (pattern), syntax, preg); |
| 6482 | |
| 6483 | /* POSIX doesn't distinguish between an unmatched open-group and an |
| 6484 | unmatched close-group: both are REG_EPAREN. */ |
| 6485 | if (ret == REG_ERPAREN) |
| 6486 | ret = REG_EPAREN; |
| 6487 | |
| 6488 | if (ret == REG_NOERROR && preg->fastmap) |
| 6489 | { /* Compute the fastmap now, since regexec cannot modify the pattern |
| 6490 | buffer. */ |
| 6491 | re_compile_fastmap (preg); |
| 6492 | if (preg->can_be_null) |
| 6493 | { /* The fastmap can't be used anyway. */ |
| 6494 | free (preg->fastmap); |
| 6495 | preg->fastmap = NULL; |
| 6496 | } |
| 6497 | } |
| 6498 | return ret; |
| 6499 | } |
| 6500 | WEAK_ALIAS (__regcomp, regcomp) |
| 6501 | |
| 6502 | |
| 6503 | /* regexec searches for a given pattern, specified by PREG, in the |
| 6504 | string STRING. |
| 6505 | |
| 6506 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
| 6507 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
| 6508 | least NMATCH elements, and we set them to the offsets of the |
| 6509 | corresponding matched substrings. |
| 6510 | |
| 6511 | EFLAGS specifies `execution flags' which affect matching: if |
| 6512 | REG_NOTBOL is set, then ^ does not match at the beginning of the |
| 6513 | string; if REG_NOTEOL is set, then $ does not match at the end. |
| 6514 | |
| 6515 | We return 0 if we find a match and REG_NOMATCH if not. */ |
| 6516 | |
| 6517 | reg_errcode_t |
| 6518 | regexec (const regex_t *__restrict preg, const char *__restrict string, |
| 6519 | size_t nmatch, regmatch_t pmatch[__restrict_arr], int eflags) |
| 6520 | { |
| 6521 | regoff_t ret; |
| 6522 | struct re_registers regs; |
| 6523 | regex_t private_preg; |
| 6524 | size_t len = strlen (string); |
| 6525 | boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch; |
| 6526 | |
| 6527 | private_preg = *preg; |
| 6528 | |
| 6529 | private_preg.not_bol = !!(eflags & REG_NOTBOL); |
| 6530 | private_preg.not_eol = !!(eflags & REG_NOTEOL); |
| 6531 | |
| 6532 | /* The user has told us exactly how many registers to return |
| 6533 | information about, via `nmatch'. We have to pass that on to the |
| 6534 | matching routines. */ |
| 6535 | private_preg.regs_allocated = REGS_FIXED; |
| 6536 | |
| 6537 | if (want_reg_info) |
| 6538 | { |
| 6539 | regs.num_regs = nmatch; |
| 6540 | regs.start = TALLOC (nmatch * 2, regoff_t); |
| 6541 | if (regs.start == NULL) |
| 6542 | return REG_NOMATCH; |
| 6543 | regs.end = regs.start + nmatch; |
| 6544 | } |
| 6545 | |
| 6546 | /* Instead of using not_eol to implement REG_NOTEOL, we could simply |
| 6547 | pass (&private_preg, string, len + 1, 0, len, ...) pretending the string |
| 6548 | was a little bit longer but still only matching the real part. |
| 6549 | This works because the `endline' will check for a '\n' and will find a |
| 6550 | '\0', correctly deciding that this is not the end of a line. |
| 6551 | But it doesn't work out so nicely for REG_NOTBOL, since we don't have |
| 6552 | a convenient '\0' there. For all we know, the string could be preceded |
| 6553 | by '\n' which would throw things off. */ |
| 6554 | |
| 6555 | /* Perform the searching operation. */ |
| 6556 | ret = re_search (&private_preg, string, len, |
| 6557 | /* start: */ 0, /* range: */ len, |
| 6558 | want_reg_info ? ®s : (struct re_registers *) 0); |
| 6559 | |
| 6560 | /* Copy the register information to the POSIX structure. */ |
| 6561 | if (want_reg_info) |
| 6562 | { |
| 6563 | if (ret >= 0) |
| 6564 | { |
| 6565 | unsigned r; |
| 6566 | |
| 6567 | for (r = 0; r < nmatch; r++) |
| 6568 | { |
| 6569 | pmatch[r].rm_so = regs.start[r]; |
| 6570 | pmatch[r].rm_eo = regs.end[r]; |
| 6571 | } |
| 6572 | } |
| 6573 | |
| 6574 | /* If we needed the temporary register info, free the space now. */ |
| 6575 | free (regs.start); |
| 6576 | } |
| 6577 | |
| 6578 | /* We want zero return to mean success, unlike `re_search'. */ |
| 6579 | return ret >= 0 ? REG_NOERROR : REG_NOMATCH; |
| 6580 | } |
| 6581 | WEAK_ALIAS (__regexec, regexec) |
| 6582 | |
| 6583 | |
| 6584 | /* Returns a message corresponding to an error code, ERR_CODE, returned |
| 6585 | from either regcomp or regexec. We don't use PREG here. |
| 6586 | |
| 6587 | ERR_CODE was previously called ERRCODE, but that name causes an |
| 6588 | error with msvc8 compiler. */ |
| 6589 | |
| 6590 | size_t |
| 6591 | regerror (int err_code, const regex_t *preg, char *errbuf, size_t errbuf_size) |
| 6592 | { |
| 6593 | const char *msg; |
| 6594 | size_t msg_size; |
| 6595 | |
| 6596 | if (err_code < 0 |
| 6597 | || err_code >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0]))) |
| 6598 | /* Only error codes returned by the rest of the code should be passed |
| 6599 | to this routine. If we are given anything else, or if other regex |
| 6600 | code generates an invalid error code, then the program has a bug. |
| 6601 | Dump core so we can fix it. */ |
| 6602 | abort (); |
| 6603 | |
| 6604 | msg = gettext (re_error_msgid[err_code]); |
| 6605 | |
| 6606 | msg_size = strlen (msg) + 1; /* Includes the null. */ |
| 6607 | |
| 6608 | if (errbuf_size != 0) |
| 6609 | { |
| 6610 | if (msg_size > errbuf_size) |
| 6611 | { |
| 6612 | memcpy (errbuf, msg, errbuf_size - 1); |
| 6613 | errbuf[errbuf_size - 1] = 0; |
| 6614 | } |
| 6615 | else |
| 6616 | strcpy (errbuf, msg); |
| 6617 | } |
| 6618 | |
| 6619 | return msg_size; |
| 6620 | } |
| 6621 | WEAK_ALIAS (__regerror, regerror) |
| 6622 | |
| 6623 | |
| 6624 | /* Free dynamically allocated space used by PREG. */ |
| 6625 | |
| 6626 | void |
| 6627 | regfree (regex_t *preg) |
| 6628 | { |
| 6629 | free (preg->buffer); |
| 6630 | preg->buffer = NULL; |
| 6631 | |
| 6632 | preg->allocated = 0; |
| 6633 | preg->used = 0; |
| 6634 | |
| 6635 | free (preg->fastmap); |
| 6636 | preg->fastmap = NULL; |
| 6637 | preg->fastmap_accurate = 0; |
| 6638 | |
| 6639 | free (preg->translate); |
| 6640 | preg->translate = NULL; |
| 6641 | } |
| 6642 | WEAK_ALIAS (__regfree, regfree) |
| 6643 | |
| 6644 | #endif /* not emacs */ |