1 /* Extended regular expression matching and search library, 2 version 0.12. 3 (Implements POSIX draft P1003.2/D11.2, except for some of the 4 internationalization features.) 5 6 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 7 2002, 2005, 2010 Free Software Foundation, Inc. 8 This file is part of the GNU C Library. 9 10 The GNU C Library is free software; you can redistribute it and/or 11 modify it under the terms of the GNU Lesser General Public 12 License as published by the Free Software Foundation; either 13 version 2.1 of the License, or (at your option) any later version. 14 15 The GNU C Library is distributed in the hope that it will be useful, 16 but WITHOUT ANY WARRANTY; without even the implied warranty of 17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 18 Lesser General Public License for more details. 19 20 You should have received a copy of the GNU Lesser General Public 21 License along with the GNU C Library; if not, write to the Free 22 Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 23 02110-1301 USA. */ 24 25 /* This file has been modified for usage in libiberty. It includes "xregex.h" 26 instead of <regex.h>. The "xregex.h" header file renames all external 27 routines with an "x" prefix so they do not collide with the native regex 28 routines or with other components regex routines. */ 29 /* AIX requires this to be the first thing in the file. */ 30 #if defined _AIX && !defined __GNUC__ && !defined REGEX_MALLOC 31 #pragma alloca 32 #endif 33 34 #undef _GNU_SOURCE 35 #define _GNU_SOURCE 36 37 #ifndef INSIDE_RECURSION 38 # ifdef HAVE_CONFIG_H 39 # include <config.h> 40 # endif 41 #endif 42 43 #include <ansidecl.h> 44 45 #ifndef INSIDE_RECURSION 46 47 # if defined STDC_HEADERS && !defined emacs 48 # include <stddef.h> 49 # else 50 /* We need this for `regex.h', and perhaps for the Emacs include files. */ 51 # include <sys/types.h> 52 # endif 53 54 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC) 55 56 /* For platform which support the ISO C amendement 1 functionality we 57 support user defined character classes. */ 58 # if defined _LIBC || WIDE_CHAR_SUPPORT 59 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ 60 # include <wchar.h> 61 # include <wctype.h> 62 # endif 63 64 # ifdef _LIBC 65 /* We have to keep the namespace clean. */ 66 # define regfree(preg) __regfree (preg) 67 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) 68 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) 69 # define regerror(errcode, preg, errbuf, errbuf_size) \ 70 __regerror(errcode, preg, errbuf, errbuf_size) 71 # define re_set_registers(bu, re, nu, st, en) \ 72 __re_set_registers (bu, re, nu, st, en) 73 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ 74 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) 75 # define re_match(bufp, string, size, pos, regs) \ 76 __re_match (bufp, string, size, pos, regs) 77 # define re_search(bufp, string, size, startpos, range, regs) \ 78 __re_search (bufp, string, size, startpos, range, regs) 79 # define re_compile_pattern(pattern, length, bufp) \ 80 __re_compile_pattern (pattern, length, bufp) 81 # define re_set_syntax(syntax) __re_set_syntax (syntax) 82 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ 83 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) 84 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) 85 86 # define btowc __btowc 87 88 /* We are also using some library internals. */ 89 # include <locale/localeinfo.h> 90 # include <locale/elem-hash.h> 91 # include <langinfo.h> 92 # include <locale/coll-lookup.h> 93 # endif 94 95 /* This is for other GNU distributions with internationalized messages. */ 96 # if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC 97 # include <libintl.h> 98 # ifdef _LIBC 99 # undef gettext 100 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES) 101 # endif 102 # else 103 # define gettext(msgid) (msgid) 104 # endif 105 106 # ifndef gettext_noop 107 /* This define is so xgettext can find the internationalizable 108 strings. */ 109 # define gettext_noop(String) String 110 # endif 111 112 /* The `emacs' switch turns on certain matching commands 113 that make sense only in Emacs. */ 114 # ifdef emacs 115 116 # include "lisp.h" 117 # include "buffer.h" 118 # include "syntax.h" 119 120 # else /* not emacs */ 121 122 /* If we are not linking with Emacs proper, 123 we can't use the relocating allocator 124 even if config.h says that we can. */ 125 # undef REL_ALLOC 126 127 # if defined STDC_HEADERS || defined _LIBC 128 # include <stdlib.h> 129 # else 130 char *malloc (); 131 char *realloc (); 132 # endif 133 134 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. 135 If nothing else has been done, use the method below. */ 136 # ifdef INHIBIT_STRING_HEADER 137 # if !(defined HAVE_BZERO && defined HAVE_BCOPY) 138 # if !defined bzero && !defined bcopy 139 # undef INHIBIT_STRING_HEADER 140 # endif 141 # endif 142 # endif 143 144 /* This is the normal way of making sure we have a bcopy and a bzero. 145 This is used in most programs--a few other programs avoid this 146 by defining INHIBIT_STRING_HEADER. */ 147 # ifndef INHIBIT_STRING_HEADER 148 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC 149 # include <string.h> 150 # ifndef bzero 151 # ifndef _LIBC 152 # define bzero(s, n) (memset (s, '\0', n), (s)) 153 # else 154 # define bzero(s, n) __bzero (s, n) 155 # endif 156 # endif 157 # else 158 # include <strings.h> 159 # ifndef memcmp 160 # define memcmp(s1, s2, n) bcmp (s1, s2, n) 161 # endif 162 # ifndef memcpy 163 # define memcpy(d, s, n) (bcopy (s, d, n), (d)) 164 # endif 165 # endif 166 # endif 167 168 /* Define the syntax stuff for \<, \>, etc. */ 169 170 /* This must be nonzero for the wordchar and notwordchar pattern 171 commands in re_match_2. */ 172 # ifndef Sword 173 # define Sword 1 174 # endif 175 176 # ifdef SWITCH_ENUM_BUG 177 # define SWITCH_ENUM_CAST(x) ((int)(x)) 178 # else 179 # define SWITCH_ENUM_CAST(x) (x) 180 # endif 181 182 # endif /* not emacs */ 183 184 # if defined _LIBC || HAVE_LIMITS_H 185 # include <limits.h> 186 # endif 187 188 # ifndef MB_LEN_MAX 189 # define MB_LEN_MAX 1 190 # endif 191 192 /* Get the interface, including the syntax bits. */ 193 # include "xregex.h" /* change for libiberty */ 194 195 /* isalpha etc. are used for the character classes. */ 196 # include <ctype.h> 197 198 /* Jim Meyering writes: 199 200 "... Some ctype macros are valid only for character codes that 201 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when 202 using /bin/cc or gcc but without giving an ansi option). So, all 203 ctype uses should be through macros like ISPRINT... If 204 STDC_HEADERS is defined, then autoconf has verified that the ctype 205 macros don't need to be guarded with references to isascii. ... 206 Defining isascii to 1 should let any compiler worth its salt 207 eliminate the && through constant folding." 208 Solaris defines some of these symbols so we must undefine them first. */ 209 210 # undef ISASCII 211 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) 212 # define ISASCII(c) 1 213 # else 214 # define ISASCII(c) isascii(c) 215 # endif 216 217 # ifdef isblank 218 # define ISBLANK(c) (ISASCII (c) && isblank (c)) 219 # else 220 # define ISBLANK(c) ((c) == ' ' || (c) == '\t') 221 # endif 222 # ifdef isgraph 223 # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) 224 # else 225 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) 226 # endif 227 228 # undef ISPRINT 229 # define ISPRINT(c) (ISASCII (c) && isprint (c)) 230 # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) 231 # define ISALNUM(c) (ISASCII (c) && isalnum (c)) 232 # define ISALPHA(c) (ISASCII (c) && isalpha (c)) 233 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) 234 # define ISLOWER(c) (ISASCII (c) && islower (c)) 235 # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) 236 # define ISSPACE(c) (ISASCII (c) && isspace (c)) 237 # define ISUPPER(c) (ISASCII (c) && isupper (c)) 238 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) 239 240 # ifdef _tolower 241 # define TOLOWER(c) _tolower(c) 242 # else 243 # define TOLOWER(c) tolower(c) 244 # endif 245 246 # ifndef NULL 247 # define NULL (void *)0 248 # endif 249 250 /* We remove any previous definition of `SIGN_EXTEND_CHAR', 251 since ours (we hope) works properly with all combinations of 252 machines, compilers, `char' and `unsigned char' argument types. 253 (Per Bothner suggested the basic approach.) */ 254 # undef SIGN_EXTEND_CHAR 255 # if __STDC__ 256 # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) 257 # else /* not __STDC__ */ 258 /* As in Harbison and Steele. */ 259 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) 260 # endif 261 262 # ifndef emacs 263 /* How many characters in the character set. */ 264 # define CHAR_SET_SIZE 256 265 266 # ifdef SYNTAX_TABLE 267 268 extern char *re_syntax_table; 269 270 # else /* not SYNTAX_TABLE */ 271 272 static char re_syntax_table[CHAR_SET_SIZE]; 273 274 static void init_syntax_once (void); 275 276 static void 277 init_syntax_once (void) 278 { 279 register int c; 280 static int done = 0; 281 282 if (done) 283 return; 284 bzero (re_syntax_table, sizeof re_syntax_table); 285 286 for (c = 0; c < CHAR_SET_SIZE; ++c) 287 if (ISALNUM (c)) 288 re_syntax_table[c] = Sword; 289 290 re_syntax_table['_'] = Sword; 291 292 done = 1; 293 } 294 295 # endif /* not SYNTAX_TABLE */ 296 297 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)] 298 299 # endif /* emacs */ 300 301 /* Integer type for pointers. */ 302 # if !defined _LIBC && !defined HAVE_UINTPTR_T 303 typedef unsigned long int uintptr_t; 304 # endif 305 306 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we 307 use `alloca' instead of `malloc'. This is because using malloc in 308 re_search* or re_match* could cause memory leaks when C-g is used in 309 Emacs; also, malloc is slower and causes storage fragmentation. On 310 the other hand, malloc is more portable, and easier to debug. 311 312 Because we sometimes use alloca, some routines have to be macros, 313 not functions -- `alloca'-allocated space disappears at the end of the 314 function it is called in. */ 315 316 # ifdef REGEX_MALLOC 317 318 # define REGEX_ALLOCATE malloc 319 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) 320 # define REGEX_FREE free 321 322 # else /* not REGEX_MALLOC */ 323 324 /* Emacs already defines alloca, sometimes. */ 325 # ifndef alloca 326 327 /* Make alloca work the best possible way. */ 328 # ifdef __GNUC__ 329 # define alloca __builtin_alloca 330 # else /* not __GNUC__ */ 331 # if HAVE_ALLOCA_H 332 # include <alloca.h> 333 # endif /* HAVE_ALLOCA_H */ 334 # endif /* not __GNUC__ */ 335 336 # endif /* not alloca */ 337 338 # define REGEX_ALLOCATE alloca 339 340 /* Assumes a `char *destination' variable. */ 341 # define REGEX_REALLOCATE(source, osize, nsize) \ 342 (destination = (char *) alloca (nsize), \ 343 memcpy (destination, source, osize)) 344 345 /* No need to do anything to free, after alloca. */ 346 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ 347 348 # endif /* not REGEX_MALLOC */ 349 350 /* Define how to allocate the failure stack. */ 351 352 # if defined REL_ALLOC && defined REGEX_MALLOC 353 354 # define REGEX_ALLOCATE_STACK(size) \ 355 r_alloc (&failure_stack_ptr, (size)) 356 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ 357 r_re_alloc (&failure_stack_ptr, (nsize)) 358 # define REGEX_FREE_STACK(ptr) \ 359 r_alloc_free (&failure_stack_ptr) 360 361 # else /* not using relocating allocator */ 362 363 # ifdef REGEX_MALLOC 364 365 # define REGEX_ALLOCATE_STACK malloc 366 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) 367 # define REGEX_FREE_STACK free 368 369 # else /* not REGEX_MALLOC */ 370 371 # define REGEX_ALLOCATE_STACK alloca 372 373 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ 374 REGEX_REALLOCATE (source, osize, nsize) 375 /* No need to explicitly free anything. */ 376 # define REGEX_FREE_STACK(arg) 377 378 # endif /* not REGEX_MALLOC */ 379 # endif /* not using relocating allocator */ 380 381 382 /* True if `size1' is non-NULL and PTR is pointing anywhere inside 383 `string1' or just past its end. This works if PTR is NULL, which is 384 a good thing. */ 385 # define FIRST_STRING_P(ptr) \ 386 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) 387 388 /* (Re)Allocate N items of type T using malloc, or fail. */ 389 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) 390 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) 391 # define RETALLOC_IF(addr, n, t) \ 392 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) 393 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) 394 395 # define BYTEWIDTH 8 /* In bits. */ 396 397 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) 398 399 # undef MAX 400 # undef MIN 401 # define MAX(a, b) ((a) > (b) ? (a) : (b)) 402 # define MIN(a, b) ((a) < (b) ? (a) : (b)) 403 404 typedef char boolean; 405 # define false 0 406 # define true 1 407 408 static reg_errcode_t byte_regex_compile (const char *pattern, size_t size, 409 reg_syntax_t syntax, 410 struct re_pattern_buffer *bufp); 411 412 static int byte_re_match_2_internal (struct re_pattern_buffer *bufp, 413 const char *string1, int size1, 414 const char *string2, int size2, 415 int pos, 416 struct re_registers *regs, 417 int stop); 418 static int byte_re_search_2 (struct re_pattern_buffer *bufp, 419 const char *string1, int size1, 420 const char *string2, int size2, 421 int startpos, int range, 422 struct re_registers *regs, int stop); 423 static int byte_re_compile_fastmap (struct re_pattern_buffer *bufp); 424 425 #ifdef MBS_SUPPORT 426 static reg_errcode_t wcs_regex_compile (const char *pattern, size_t size, 427 reg_syntax_t syntax, 428 struct re_pattern_buffer *bufp); 429 430 431 static int wcs_re_match_2_internal (struct re_pattern_buffer *bufp, 432 const char *cstring1, int csize1, 433 const char *cstring2, int csize2, 434 int pos, 435 struct re_registers *regs, 436 int stop, 437 wchar_t *string1, int size1, 438 wchar_t *string2, int size2, 439 int *mbs_offset1, int *mbs_offset2); 440 static int wcs_re_search_2 (struct re_pattern_buffer *bufp, 441 const char *string1, int size1, 442 const char *string2, int size2, 443 int startpos, int range, 444 struct re_registers *regs, int stop); 445 static int wcs_re_compile_fastmap (struct re_pattern_buffer *bufp); 446 #endif 447 448 /* These are the command codes that appear in compiled regular 449 expressions. Some opcodes are followed by argument bytes. A 450 command code can specify any interpretation whatsoever for its 451 arguments. Zero bytes may appear in the compiled regular expression. */ 452 453 typedef enum 454 { 455 no_op = 0, 456 457 /* Succeed right away--no more backtracking. */ 458 succeed, 459 460 /* Followed by one byte giving n, then by n literal bytes. */ 461 exactn, 462 463 # ifdef MBS_SUPPORT 464 /* Same as exactn, but contains binary data. */ 465 exactn_bin, 466 # endif 467 468 /* Matches any (more or less) character. */ 469 anychar, 470 471 /* Matches any one char belonging to specified set. First 472 following byte is number of bitmap bytes. Then come bytes 473 for a bitmap saying which chars are in. Bits in each byte 474 are ordered low-bit-first. A character is in the set if its 475 bit is 1. A character too large to have a bit in the map is 476 automatically not in the set. */ 477 /* ifdef MBS_SUPPORT, following element is length of character 478 classes, length of collating symbols, length of equivalence 479 classes, length of character ranges, and length of characters. 480 Next, character class element, collating symbols elements, 481 equivalence class elements, range elements, and character 482 elements follow. 483 See regex_compile function. */ 484 charset, 485 486 /* Same parameters as charset, but match any character that is 487 not one of those specified. */ 488 charset_not, 489 490 /* Start remembering the text that is matched, for storing in a 491 register. Followed by one byte with the register number, in 492 the range 0 to one less than the pattern buffer's re_nsub 493 field. Then followed by one byte with the number of groups 494 inner to this one. (This last has to be part of the 495 start_memory only because we need it in the on_failure_jump 496 of re_match_2.) */ 497 start_memory, 498 499 /* Stop remembering the text that is matched and store it in a 500 memory register. Followed by one byte with the register 501 number, in the range 0 to one less than `re_nsub' in the 502 pattern buffer, and one byte with the number of inner groups, 503 just like `start_memory'. (We need the number of inner 504 groups here because we don't have any easy way of finding the 505 corresponding start_memory when we're at a stop_memory.) */ 506 stop_memory, 507 508 /* Match a duplicate of something remembered. Followed by one 509 byte containing the register number. */ 510 duplicate, 511 512 /* Fail unless at beginning of line. */ 513 begline, 514 515 /* Fail unless at end of line. */ 516 endline, 517 518 /* Succeeds if at beginning of buffer (if emacs) or at beginning 519 of string to be matched (if not). */ 520 begbuf, 521 522 /* Analogously, for end of buffer/string. */ 523 endbuf, 524 525 /* Followed by two byte relative address to which to jump. */ 526 jump, 527 528 /* Same as jump, but marks the end of an alternative. */ 529 jump_past_alt, 530 531 /* Followed by two-byte relative address of place to resume at 532 in case of failure. */ 533 /* ifdef MBS_SUPPORT, the size of address is 1. */ 534 on_failure_jump, 535 536 /* Like on_failure_jump, but pushes a placeholder instead of the 537 current string position when executed. */ 538 on_failure_keep_string_jump, 539 540 /* Throw away latest failure point and then jump to following 541 two-byte relative address. */ 542 /* ifdef MBS_SUPPORT, the size of address is 1. */ 543 pop_failure_jump, 544 545 /* Change to pop_failure_jump if know won't have to backtrack to 546 match; otherwise change to jump. This is used to jump 547 back to the beginning of a repeat. If what follows this jump 548 clearly won't match what the repeat does, such that we can be 549 sure that there is no use backtracking out of repetitions 550 already matched, then we change it to a pop_failure_jump. 551 Followed by two-byte address. */ 552 /* ifdef MBS_SUPPORT, the size of address is 1. */ 553 maybe_pop_jump, 554 555 /* Jump to following two-byte address, and push a dummy failure 556 point. This failure point will be thrown away if an attempt 557 is made to use it for a failure. A `+' construct makes this 558 before the first repeat. Also used as an intermediary kind 559 of jump when compiling an alternative. */ 560 /* ifdef MBS_SUPPORT, the size of address is 1. */ 561 dummy_failure_jump, 562 563 /* Push a dummy failure point and continue. Used at the end of 564 alternatives. */ 565 push_dummy_failure, 566 567 /* Followed by two-byte relative address and two-byte number n. 568 After matching N times, jump to the address upon failure. */ 569 /* ifdef MBS_SUPPORT, the size of address is 1. */ 570 succeed_n, 571 572 /* Followed by two-byte relative address, and two-byte number n. 573 Jump to the address N times, then fail. */ 574 /* ifdef MBS_SUPPORT, the size of address is 1. */ 575 jump_n, 576 577 /* Set the following two-byte relative address to the 578 subsequent two-byte number. The address *includes* the two 579 bytes of number. */ 580 /* ifdef MBS_SUPPORT, the size of address is 1. */ 581 set_number_at, 582 583 wordchar, /* Matches any word-constituent character. */ 584 notwordchar, /* Matches any char that is not a word-constituent. */ 585 586 wordbeg, /* Succeeds if at word beginning. */ 587 wordend, /* Succeeds if at word end. */ 588 589 wordbound, /* Succeeds if at a word boundary. */ 590 notwordbound /* Succeeds if not at a word boundary. */ 591 592 # ifdef emacs 593 ,before_dot, /* Succeeds if before point. */ 594 at_dot, /* Succeeds if at point. */ 595 after_dot, /* Succeeds if after point. */ 596 597 /* Matches any character whose syntax is specified. Followed by 598 a byte which contains a syntax code, e.g., Sword. */ 599 syntaxspec, 600 601 /* Matches any character whose syntax is not that specified. */ 602 notsyntaxspec 603 # endif /* emacs */ 604 } re_opcode_t; 605 #endif /* not INSIDE_RECURSION */ 606 607 608 #ifdef BYTE 609 # define CHAR_T char 610 # define UCHAR_T unsigned char 611 # define COMPILED_BUFFER_VAR bufp->buffer 612 # define OFFSET_ADDRESS_SIZE 2 613 # define PREFIX(name) byte_##name 614 # define ARG_PREFIX(name) name 615 # define PUT_CHAR(c) putchar (c) 616 #else 617 # ifdef WCHAR 618 # define CHAR_T wchar_t 619 # define UCHAR_T wchar_t 620 # define COMPILED_BUFFER_VAR wc_buffer 621 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */ 622 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1) 623 # define PREFIX(name) wcs_##name 624 # define ARG_PREFIX(name) c##name 625 /* Should we use wide stream?? */ 626 # define PUT_CHAR(c) printf ("%C", c); 627 # define TRUE 1 628 # define FALSE 0 629 # else 630 # ifdef MBS_SUPPORT 631 # define WCHAR 632 # define INSIDE_RECURSION 633 # include "regex.c" 634 # undef INSIDE_RECURSION 635 # endif 636 # define BYTE 637 # define INSIDE_RECURSION 638 # include "regex.c" 639 # undef INSIDE_RECURSION 640 # endif 641 #endif 642 643 #ifdef INSIDE_RECURSION 644 /* Common operations on the compiled pattern. */ 645 646 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ 647 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ 648 649 # ifdef WCHAR 650 # define STORE_NUMBER(destination, number) \ 651 do { \ 652 *(destination) = (UCHAR_T)(number); \ 653 } while (0) 654 # else /* BYTE */ 655 # define STORE_NUMBER(destination, number) \ 656 do { \ 657 (destination)[0] = (number) & 0377; \ 658 (destination)[1] = (number) >> 8; \ 659 } while (0) 660 # endif /* WCHAR */ 661 662 /* Same as STORE_NUMBER, except increment DESTINATION to 663 the byte after where the number is stored. Therefore, DESTINATION 664 must be an lvalue. */ 665 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ 666 667 # define STORE_NUMBER_AND_INCR(destination, number) \ 668 do { \ 669 STORE_NUMBER (destination, number); \ 670 (destination) += OFFSET_ADDRESS_SIZE; \ 671 } while (0) 672 673 /* Put into DESTINATION a number stored in two contiguous bytes starting 674 at SOURCE. */ 675 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ 676 677 # ifdef WCHAR 678 # define EXTRACT_NUMBER(destination, source) \ 679 do { \ 680 (destination) = *(source); \ 681 } while (0) 682 # else /* BYTE */ 683 # define EXTRACT_NUMBER(destination, source) \ 684 do { \ 685 (destination) = *(source) & 0377; \ 686 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ 687 } while (0) 688 # endif 689 690 # ifdef DEBUG 691 static void PREFIX(extract_number) (int *dest, UCHAR_T *source); 692 static void 693 PREFIX(extract_number) (int *dest, UCHAR_T *source) 694 { 695 # ifdef WCHAR 696 *dest = *source; 697 # else /* BYTE */ 698 int temp = SIGN_EXTEND_CHAR (*(source + 1)); 699 *dest = *source & 0377; 700 *dest += temp << 8; 701 # endif 702 } 703 704 # ifndef EXTRACT_MACROS /* To debug the macros. */ 705 # undef EXTRACT_NUMBER 706 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src) 707 # endif /* not EXTRACT_MACROS */ 708 709 # endif /* DEBUG */ 710 711 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. 712 SOURCE must be an lvalue. */ 713 714 # define EXTRACT_NUMBER_AND_INCR(destination, source) \ 715 do { \ 716 EXTRACT_NUMBER (destination, source); \ 717 (source) += OFFSET_ADDRESS_SIZE; \ 718 } while (0) 719 720 # ifdef DEBUG 721 static void PREFIX(extract_number_and_incr) (int *destination, 722 UCHAR_T **source); 723 static void 724 PREFIX(extract_number_and_incr) (int *destination, UCHAR_T **source) 725 { 726 PREFIX(extract_number) (destination, *source); 727 *source += OFFSET_ADDRESS_SIZE; 728 } 729 730 # ifndef EXTRACT_MACROS 731 # undef EXTRACT_NUMBER_AND_INCR 732 # define EXTRACT_NUMBER_AND_INCR(dest, src) \ 733 PREFIX(extract_number_and_incr) (&dest, &src) 734 # endif /* not EXTRACT_MACROS */ 735 736 # endif /* DEBUG */ 737 738 739 740 /* If DEBUG is defined, Regex prints many voluminous messages about what 741 it is doing (if the variable `debug' is nonzero). If linked with the 742 main program in `iregex.c', you can enter patterns and strings 743 interactively. And if linked with the main program in `main.c' and 744 the other test files, you can run the already-written tests. */ 745 746 # ifdef DEBUG 747 748 # ifndef DEFINED_ONCE 749 750 /* We use standard I/O for debugging. */ 751 # include <stdio.h> 752 753 /* It is useful to test things that ``must'' be true when debugging. */ 754 # include <assert.h> 755 756 static int debug; 757 758 # define DEBUG_STATEMENT(e) e 759 # define DEBUG_PRINT1(x) if (debug) printf (x) 760 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) 761 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) 762 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) 763 # endif /* not DEFINED_ONCE */ 764 765 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ 766 if (debug) PREFIX(print_partial_compiled_pattern) (s, e) 767 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ 768 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2) 769 770 771 /* Print the fastmap in human-readable form. */ 772 773 # ifndef DEFINED_ONCE 774 void 775 print_fastmap (char *fastmap) 776 { 777 unsigned was_a_range = 0; 778 unsigned i = 0; 779 780 while (i < (1 << BYTEWIDTH)) 781 { 782 if (fastmap[i++]) 783 { 784 was_a_range = 0; 785 putchar (i - 1); 786 while (i < (1 << BYTEWIDTH) && fastmap[i]) 787 { 788 was_a_range = 1; 789 i++; 790 } 791 if (was_a_range) 792 { 793 printf ("-"); 794 putchar (i - 1); 795 } 796 } 797 } 798 putchar ('\n'); 799 } 800 # endif /* not DEFINED_ONCE */ 801 802 803 /* Print a compiled pattern string in human-readable form, starting at 804 the START pointer into it and ending just before the pointer END. */ 805 806 void 807 PREFIX(print_partial_compiled_pattern) (UCHAR_T *start, UCHAR_T *end) 808 { 809 int mcnt, mcnt2; 810 UCHAR_T *p1; 811 UCHAR_T *p = start; 812 UCHAR_T *pend = end; 813 814 if (start == NULL) 815 { 816 printf ("(null)\n"); 817 return; 818 } 819 820 /* Loop over pattern commands. */ 821 while (p < pend) 822 { 823 # ifdef _LIBC 824 printf ("%td:\t", p - start); 825 # else 826 printf ("%ld:\t", (long int) (p - start)); 827 # endif 828 829 switch ((re_opcode_t) *p++) 830 { 831 case no_op: 832 printf ("/no_op"); 833 break; 834 835 case exactn: 836 mcnt = *p++; 837 printf ("/exactn/%d", mcnt); 838 do 839 { 840 putchar ('/'); 841 PUT_CHAR (*p++); 842 } 843 while (--mcnt); 844 break; 845 846 # ifdef MBS_SUPPORT 847 case exactn_bin: 848 mcnt = *p++; 849 printf ("/exactn_bin/%d", mcnt); 850 do 851 { 852 printf("/%lx", (long int) *p++); 853 } 854 while (--mcnt); 855 break; 856 # endif /* MBS_SUPPORT */ 857 858 case start_memory: 859 mcnt = *p++; 860 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++); 861 break; 862 863 case stop_memory: 864 mcnt = *p++; 865 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++); 866 break; 867 868 case duplicate: 869 printf ("/duplicate/%ld", (long int) *p++); 870 break; 871 872 case anychar: 873 printf ("/anychar"); 874 break; 875 876 case charset: 877 case charset_not: 878 { 879 # ifdef WCHAR 880 int i, length; 881 wchar_t *workp = p; 882 printf ("/charset [%s", 883 (re_opcode_t) *(workp - 1) == charset_not ? "^" : ""); 884 p += 5; 885 length = *workp++; /* the length of char_classes */ 886 for (i=0 ; i<length ; i++) 887 printf("[:%lx:]", (long int) *p++); 888 length = *workp++; /* the length of collating_symbol */ 889 for (i=0 ; i<length ;) 890 { 891 printf("[."); 892 while(*p != 0) 893 PUT_CHAR((i++,*p++)); 894 i++,p++; 895 printf(".]"); 896 } 897 length = *workp++; /* the length of equivalence_class */ 898 for (i=0 ; i<length ;) 899 { 900 printf("[="); 901 while(*p != 0) 902 PUT_CHAR((i++,*p++)); 903 i++,p++; 904 printf("=]"); 905 } 906 length = *workp++; /* the length of char_range */ 907 for (i=0 ; i<length ; i++) 908 { 909 wchar_t range_start = *p++; 910 wchar_t range_end = *p++; 911 printf("%C-%C", range_start, range_end); 912 } 913 length = *workp++; /* the length of char */ 914 for (i=0 ; i<length ; i++) 915 printf("%C", *p++); 916 putchar (']'); 917 # else 918 register int c, last = -100; 919 register int in_range = 0; 920 921 printf ("/charset [%s", 922 (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); 923 924 assert (p + *p < pend); 925 926 for (c = 0; c < 256; c++) 927 if (c / 8 < *p 928 && (p[1 + (c/8)] & (1 << (c % 8)))) 929 { 930 /* Are we starting a range? */ 931 if (last + 1 == c && ! in_range) 932 { 933 putchar ('-'); 934 in_range = 1; 935 } 936 /* Have we broken a range? */ 937 else if (last + 1 != c && in_range) 938 { 939 putchar (last); 940 in_range = 0; 941 } 942 943 if (! in_range) 944 putchar (c); 945 946 last = c; 947 } 948 949 if (in_range) 950 putchar (last); 951 952 putchar (']'); 953 954 p += 1 + *p; 955 # endif /* WCHAR */ 956 } 957 break; 958 959 case begline: 960 printf ("/begline"); 961 break; 962 963 case endline: 964 printf ("/endline"); 965 break; 966 967 case on_failure_jump: 968 PREFIX(extract_number_and_incr) (&mcnt, &p); 969 # ifdef _LIBC 970 printf ("/on_failure_jump to %td", p + mcnt - start); 971 # else 972 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start)); 973 # endif 974 break; 975 976 case on_failure_keep_string_jump: 977 PREFIX(extract_number_and_incr) (&mcnt, &p); 978 # ifdef _LIBC 979 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start); 980 # else 981 printf ("/on_failure_keep_string_jump to %ld", 982 (long int) (p + mcnt - start)); 983 # endif 984 break; 985 986 case dummy_failure_jump: 987 PREFIX(extract_number_and_incr) (&mcnt, &p); 988 # ifdef _LIBC 989 printf ("/dummy_failure_jump to %td", p + mcnt - start); 990 # else 991 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start)); 992 # endif 993 break; 994 995 case push_dummy_failure: 996 printf ("/push_dummy_failure"); 997 break; 998 999 case maybe_pop_jump: 1000 PREFIX(extract_number_and_incr) (&mcnt, &p); 1001 # ifdef _LIBC 1002 printf ("/maybe_pop_jump to %td", p + mcnt - start); 1003 # else 1004 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start)); 1005 # endif 1006 break; 1007 1008 case pop_failure_jump: 1009 PREFIX(extract_number_and_incr) (&mcnt, &p); 1010 # ifdef _LIBC 1011 printf ("/pop_failure_jump to %td", p + mcnt - start); 1012 # else 1013 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start)); 1014 # endif 1015 break; 1016 1017 case jump_past_alt: 1018 PREFIX(extract_number_and_incr) (&mcnt, &p); 1019 # ifdef _LIBC 1020 printf ("/jump_past_alt to %td", p + mcnt - start); 1021 # else 1022 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start)); 1023 # endif 1024 break; 1025 1026 case jump: 1027 PREFIX(extract_number_and_incr) (&mcnt, &p); 1028 # ifdef _LIBC 1029 printf ("/jump to %td", p + mcnt - start); 1030 # else 1031 printf ("/jump to %ld", (long int) (p + mcnt - start)); 1032 # endif 1033 break; 1034 1035 case succeed_n: 1036 PREFIX(extract_number_and_incr) (&mcnt, &p); 1037 p1 = p + mcnt; 1038 PREFIX(extract_number_and_incr) (&mcnt2, &p); 1039 # ifdef _LIBC 1040 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2); 1041 # else 1042 printf ("/succeed_n to %ld, %d times", 1043 (long int) (p1 - start), mcnt2); 1044 # endif 1045 break; 1046 1047 case jump_n: 1048 PREFIX(extract_number_and_incr) (&mcnt, &p); 1049 p1 = p + mcnt; 1050 PREFIX(extract_number_and_incr) (&mcnt2, &p); 1051 printf ("/jump_n to %d, %d times", p1 - start, mcnt2); 1052 break; 1053 1054 case set_number_at: 1055 PREFIX(extract_number_and_incr) (&mcnt, &p); 1056 p1 = p + mcnt; 1057 PREFIX(extract_number_and_incr) (&mcnt2, &p); 1058 # ifdef _LIBC 1059 printf ("/set_number_at location %td to %d", p1 - start, mcnt2); 1060 # else 1061 printf ("/set_number_at location %ld to %d", 1062 (long int) (p1 - start), mcnt2); 1063 # endif 1064 break; 1065 1066 case wordbound: 1067 printf ("/wordbound"); 1068 break; 1069 1070 case notwordbound: 1071 printf ("/notwordbound"); 1072 break; 1073 1074 case wordbeg: 1075 printf ("/wordbeg"); 1076 break; 1077 1078 case wordend: 1079 printf ("/wordend"); 1080 break; 1081 1082 # ifdef emacs 1083 case before_dot: 1084 printf ("/before_dot"); 1085 break; 1086 1087 case at_dot: 1088 printf ("/at_dot"); 1089 break; 1090 1091 case after_dot: 1092 printf ("/after_dot"); 1093 break; 1094 1095 case syntaxspec: 1096 printf ("/syntaxspec"); 1097 mcnt = *p++; 1098 printf ("/%d", mcnt); 1099 break; 1100 1101 case notsyntaxspec: 1102 printf ("/notsyntaxspec"); 1103 mcnt = *p++; 1104 printf ("/%d", mcnt); 1105 break; 1106 # endif /* emacs */ 1107 1108 case wordchar: 1109 printf ("/wordchar"); 1110 break; 1111 1112 case notwordchar: 1113 printf ("/notwordchar"); 1114 break; 1115 1116 case begbuf: 1117 printf ("/begbuf"); 1118 break; 1119 1120 case endbuf: 1121 printf ("/endbuf"); 1122 break; 1123 1124 default: 1125 printf ("?%ld", (long int) *(p-1)); 1126 } 1127 1128 putchar ('\n'); 1129 } 1130 1131 # ifdef _LIBC 1132 printf ("%td:\tend of pattern.\n", p - start); 1133 # else 1134 printf ("%ld:\tend of pattern.\n", (long int) (p - start)); 1135 # endif 1136 } 1137 1138 1139 void 1140 PREFIX(print_compiled_pattern) (struct re_pattern_buffer *bufp) 1141 { 1142 UCHAR_T *buffer = (UCHAR_T*) bufp->buffer; 1143 1144 PREFIX(print_partial_compiled_pattern) (buffer, buffer 1145 + bufp->used / sizeof(UCHAR_T)); 1146 printf ("%ld bytes used/%ld bytes allocated.\n", 1147 bufp->used, bufp->allocated); 1148 1149 if (bufp->fastmap_accurate && bufp->fastmap) 1150 { 1151 printf ("fastmap: "); 1152 print_fastmap (bufp->fastmap); 1153 } 1154 1155 # ifdef _LIBC 1156 printf ("re_nsub: %Zd\t", bufp->re_nsub); 1157 # else 1158 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub); 1159 # endif 1160 printf ("regs_alloc: %d\t", bufp->regs_allocated); 1161 printf ("can_be_null: %d\t", bufp->can_be_null); 1162 printf ("newline_anchor: %d\n", bufp->newline_anchor); 1163 printf ("no_sub: %d\t", bufp->no_sub); 1164 printf ("not_bol: %d\t", bufp->not_bol); 1165 printf ("not_eol: %d\t", bufp->not_eol); 1166 printf ("syntax: %lx\n", bufp->syntax); 1167 /* Perhaps we should print the translate table? */ 1168 } 1169 1170 1171 void 1172 PREFIX(print_double_string) (const CHAR_T *where, const CHAR_T *string1, 1173 int size1, const CHAR_T *string2, int size2) 1174 { 1175 int this_char; 1176 1177 if (where == NULL) 1178 printf ("(null)"); 1179 else 1180 { 1181 int cnt; 1182 1183 if (FIRST_STRING_P (where)) 1184 { 1185 for (this_char = where - string1; this_char < size1; this_char++) 1186 PUT_CHAR (string1[this_char]); 1187 1188 where = string2; 1189 } 1190 1191 cnt = 0; 1192 for (this_char = where - string2; this_char < size2; this_char++) 1193 { 1194 PUT_CHAR (string2[this_char]); 1195 if (++cnt > 100) 1196 { 1197 fputs ("...", stdout); 1198 break; 1199 } 1200 } 1201 } 1202 } 1203 1204 # ifndef DEFINED_ONCE 1205 void 1206 printchar (int c) 1207 { 1208 putc (c, stderr); 1209 } 1210 # endif 1211 1212 # else /* not DEBUG */ 1213 1214 # ifndef DEFINED_ONCE 1215 # undef assert 1216 # define assert(e) 1217 1218 # define DEBUG_STATEMENT(e) 1219 # define DEBUG_PRINT1(x) 1220 # define DEBUG_PRINT2(x1, x2) 1221 # define DEBUG_PRINT3(x1, x2, x3) 1222 # define DEBUG_PRINT4(x1, x2, x3, x4) 1223 # endif /* not DEFINED_ONCE */ 1224 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) 1225 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) 1226 1227 # endif /* not DEBUG */ 1228 1229 1230 1231 # ifdef WCHAR 1232 /* This convert a multibyte string to a wide character string. 1233 And write their correspondances to offset_buffer(see below) 1234 and write whether each wchar_t is binary data to is_binary. 1235 This assume invalid multibyte sequences as binary data. 1236 We assume offset_buffer and is_binary is already allocated 1237 enough space. */ 1238 1239 static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src, 1240 size_t len, int *offset_buffer, 1241 char *is_binary); 1242 static size_t 1243 convert_mbs_to_wcs (CHAR_T *dest, const unsigned char*src, size_t len, 1244 int *offset_buffer, char *is_binary) 1245 /* It hold correspondances between src(char string) and 1246 dest(wchar_t string) for optimization. 1247 e.g. src = "xxxyzz" 1248 dest = {'X', 'Y', 'Z'} 1249 (each "xxx", "y" and "zz" represent one multibyte character 1250 corresponding to 'X', 'Y' and 'Z'.) 1251 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")} 1252 = {0, 3, 4, 6} 1253 */ 1254 { 1255 wchar_t *pdest = dest; 1256 const unsigned char *psrc = src; 1257 size_t wc_count = 0; 1258 1259 mbstate_t mbs; 1260 int i, consumed; 1261 size_t mb_remain = len; 1262 size_t mb_count = 0; 1263 1264 /* Initialize the conversion state. */ 1265 memset (&mbs, 0, sizeof (mbstate_t)); 1266 1267 offset_buffer[0] = 0; 1268 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed, 1269 psrc += consumed) 1270 { 1271 #ifdef _LIBC 1272 consumed = __mbrtowc (pdest, psrc, mb_remain, &mbs); 1273 #else 1274 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs); 1275 #endif 1276 1277 if (consumed <= 0) 1278 /* failed to convert. maybe src contains binary data. 1279 So we consume 1 byte manualy. */ 1280 { 1281 *pdest = *psrc; 1282 consumed = 1; 1283 is_binary[wc_count] = TRUE; 1284 } 1285 else 1286 is_binary[wc_count] = FALSE; 1287 /* In sjis encoding, we use yen sign as escape character in 1288 place of reverse solidus. So we convert 0x5c(yen sign in 1289 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse 1290 solidus in UCS2). */ 1291 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5) 1292 *pdest = (wchar_t) *psrc; 1293 1294 offset_buffer[wc_count + 1] = mb_count += consumed; 1295 } 1296 1297 /* Fill remain of the buffer with sentinel. */ 1298 for (i = wc_count + 1 ; i <= len ; i++) 1299 offset_buffer[i] = mb_count + 1; 1300 1301 return wc_count; 1302 } 1303 1304 # endif /* WCHAR */ 1305 1306 #else /* not INSIDE_RECURSION */ 1307 1308 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can 1309 also be assigned to arbitrarily: each pattern buffer stores its own 1310 syntax, so it can be changed between regex compilations. */ 1311 /* This has no initializer because initialized variables in Emacs 1312 become read-only after dumping. */ 1313 reg_syntax_t re_syntax_options; 1314 1315 1316 /* Specify the precise syntax of regexps for compilation. This provides 1317 for compatibility for various utilities which historically have 1318 different, incompatible syntaxes. 1319 1320 The argument SYNTAX is a bit mask comprised of the various bits 1321 defined in regex.h. We return the old syntax. */ 1322 1323 reg_syntax_t 1324 re_set_syntax (reg_syntax_t syntax) 1325 { 1326 reg_syntax_t ret = re_syntax_options; 1327 1328 re_syntax_options = syntax; 1329 # ifdef DEBUG 1330 if (syntax & RE_DEBUG) 1331 debug = 1; 1332 else if (debug) /* was on but now is not */ 1333 debug = 0; 1334 # endif /* DEBUG */ 1335 return ret; 1336 } 1337 # ifdef _LIBC 1338 weak_alias (__re_set_syntax, re_set_syntax) 1339 # endif 1340 1341 /* This table gives an error message for each of the error codes listed 1342 in regex.h. Obviously the order here has to be same as there. 1343 POSIX doesn't require that we do anything for REG_NOERROR, 1344 but why not be nice? */ 1345 1346 static const char *re_error_msgid[] = 1347 { 1348 gettext_noop ("Success"), /* REG_NOERROR */ 1349 gettext_noop ("No match"), /* REG_NOMATCH */ 1350 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ 1351 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ 1352 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ 1353 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ 1354 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ 1355 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ 1356 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ 1357 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ 1358 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ 1359 gettext_noop ("Invalid range end"), /* REG_ERANGE */ 1360 gettext_noop ("Memory exhausted"), /* REG_ESPACE */ 1361 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ 1362 gettext_noop ("Premature end of regular expression"), /* REG_EEND */ 1363 gettext_noop ("Regular expression too big"), /* REG_ESIZE */ 1364 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */ 1365 }; 1366 1367 #endif /* INSIDE_RECURSION */ 1368 1369 #ifndef DEFINED_ONCE 1370 /* Avoiding alloca during matching, to placate r_alloc. */ 1371 1372 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the 1373 searching and matching functions should not call alloca. On some 1374 systems, alloca is implemented in terms of malloc, and if we're 1375 using the relocating allocator routines, then malloc could cause a 1376 relocation, which might (if the strings being searched are in the 1377 ralloc heap) shift the data out from underneath the regexp 1378 routines. 1379 1380 Here's another reason to avoid allocation: Emacs 1381 processes input from X in a signal handler; processing X input may 1382 call malloc; if input arrives while a matching routine is calling 1383 malloc, then we're scrod. But Emacs can't just block input while 1384 calling matching routines; then we don't notice interrupts when 1385 they come in. So, Emacs blocks input around all regexp calls 1386 except the matching calls, which it leaves unprotected, in the 1387 faith that they will not malloc. */ 1388 1389 /* Normally, this is fine. */ 1390 # define MATCH_MAY_ALLOCATE 1391 1392 /* When using GNU C, we are not REALLY using the C alloca, no matter 1393 what config.h may say. So don't take precautions for it. */ 1394 # ifdef __GNUC__ 1395 # undef C_ALLOCA 1396 # endif 1397 1398 /* The match routines may not allocate if (1) they would do it with malloc 1399 and (2) it's not safe for them to use malloc. 1400 Note that if REL_ALLOC is defined, matching would not use malloc for the 1401 failure stack, but we would still use it for the register vectors; 1402 so REL_ALLOC should not affect this. */ 1403 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs 1404 # undef MATCH_MAY_ALLOCATE 1405 # endif 1406 #endif /* not DEFINED_ONCE */ 1407 1408 #ifdef INSIDE_RECURSION 1409 /* Failure stack declarations and macros; both re_compile_fastmap and 1410 re_match_2 use a failure stack. These have to be macros because of 1411 REGEX_ALLOCATE_STACK. */ 1412 1413 1414 /* Number of failure points for which to initially allocate space 1415 when matching. If this number is exceeded, we allocate more 1416 space, so it is not a hard limit. */ 1417 # ifndef INIT_FAILURE_ALLOC 1418 # define INIT_FAILURE_ALLOC 5 1419 # endif 1420 1421 /* Roughly the maximum number of failure points on the stack. Would be 1422 exactly that if always used MAX_FAILURE_ITEMS items each time we failed. 1423 This is a variable only so users of regex can assign to it; we never 1424 change it ourselves. */ 1425 1426 # ifdef INT_IS_16BIT 1427 1428 # ifndef DEFINED_ONCE 1429 # if defined MATCH_MAY_ALLOCATE 1430 /* 4400 was enough to cause a crash on Alpha OSF/1, 1431 whose default stack limit is 2mb. */ 1432 long int re_max_failures = 4000; 1433 # else 1434 long int re_max_failures = 2000; 1435 # endif 1436 # endif 1437 1438 union PREFIX(fail_stack_elt) 1439 { 1440 UCHAR_T *pointer; 1441 long int integer; 1442 }; 1443 1444 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); 1445 1446 typedef struct 1447 { 1448 PREFIX(fail_stack_elt_t) *stack; 1449 unsigned long int size; 1450 unsigned long int avail; /* Offset of next open position. */ 1451 } PREFIX(fail_stack_type); 1452 1453 # else /* not INT_IS_16BIT */ 1454 1455 # ifndef DEFINED_ONCE 1456 # if defined MATCH_MAY_ALLOCATE 1457 /* 4400 was enough to cause a crash on Alpha OSF/1, 1458 whose default stack limit is 2mb. */ 1459 int re_max_failures = 4000; 1460 # else 1461 int re_max_failures = 2000; 1462 # endif 1463 # endif 1464 1465 union PREFIX(fail_stack_elt) 1466 { 1467 UCHAR_T *pointer; 1468 int integer; 1469 }; 1470 1471 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); 1472 1473 typedef struct 1474 { 1475 PREFIX(fail_stack_elt_t) *stack; 1476 unsigned size; 1477 unsigned avail; /* Offset of next open position. */ 1478 } PREFIX(fail_stack_type); 1479 1480 # endif /* INT_IS_16BIT */ 1481 1482 # ifndef DEFINED_ONCE 1483 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0) 1484 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) 1485 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) 1486 # endif 1487 1488 1489 /* Define macros to initialize and free the failure stack. 1490 Do `return -2' if the alloc fails. */ 1491 1492 # ifdef MATCH_MAY_ALLOCATE 1493 # define INIT_FAIL_STACK() \ 1494 do { \ 1495 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \ 1496 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \ 1497 \ 1498 if (fail_stack.stack == NULL) \ 1499 return -2; \ 1500 \ 1501 fail_stack.size = INIT_FAILURE_ALLOC; \ 1502 fail_stack.avail = 0; \ 1503 } while (0) 1504 1505 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) 1506 # else 1507 # define INIT_FAIL_STACK() \ 1508 do { \ 1509 fail_stack.avail = 0; \ 1510 } while (0) 1511 1512 # define RESET_FAIL_STACK() 1513 # endif 1514 1515 1516 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. 1517 1518 Return 1 if succeeds, and 0 if either ran out of memory 1519 allocating space for it or it was already too large. 1520 1521 REGEX_REALLOCATE_STACK requires `destination' be declared. */ 1522 1523 # define DOUBLE_FAIL_STACK(fail_stack) \ 1524 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \ 1525 ? 0 \ 1526 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \ 1527 REGEX_REALLOCATE_STACK ((fail_stack).stack, \ 1528 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \ 1529 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\ 1530 \ 1531 (fail_stack).stack == NULL \ 1532 ? 0 \ 1533 : ((fail_stack).size <<= 1, \ 1534 1))) 1535 1536 1537 /* Push pointer POINTER on FAIL_STACK. 1538 Return 1 if was able to do so and 0 if ran out of memory allocating 1539 space to do so. */ 1540 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ 1541 ((FAIL_STACK_FULL () \ 1542 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \ 1543 ? 0 \ 1544 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ 1545 1)) 1546 1547 /* Push a pointer value onto the failure stack. 1548 Assumes the variable `fail_stack'. Probably should only 1549 be called from within `PUSH_FAILURE_POINT'. */ 1550 # define PUSH_FAILURE_POINTER(item) \ 1551 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item) 1552 1553 /* This pushes an integer-valued item onto the failure stack. 1554 Assumes the variable `fail_stack'. Probably should only 1555 be called from within `PUSH_FAILURE_POINT'. */ 1556 # define PUSH_FAILURE_INT(item) \ 1557 fail_stack.stack[fail_stack.avail++].integer = (item) 1558 1559 /* Push a fail_stack_elt_t value onto the failure stack. 1560 Assumes the variable `fail_stack'. Probably should only 1561 be called from within `PUSH_FAILURE_POINT'. */ 1562 # define PUSH_FAILURE_ELT(item) \ 1563 fail_stack.stack[fail_stack.avail++] = (item) 1564 1565 /* These three POP... operations complement the three PUSH... operations. 1566 All assume that `fail_stack' is nonempty. */ 1567 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer 1568 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer 1569 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] 1570 1571 /* Used to omit pushing failure point id's when we're not debugging. */ 1572 # ifdef DEBUG 1573 # define DEBUG_PUSH PUSH_FAILURE_INT 1574 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () 1575 # else 1576 # define DEBUG_PUSH(item) 1577 # define DEBUG_POP(item_addr) 1578 # endif 1579 1580 1581 /* Push the information about the state we will need 1582 if we ever fail back to it. 1583 1584 Requires variables fail_stack, regstart, regend, reg_info, and 1585 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination' 1586 be declared. 1587 1588 Does `return FAILURE_CODE' if runs out of memory. */ 1589 1590 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ 1591 do { \ 1592 char *destination; \ 1593 /* Must be int, so when we don't save any registers, the arithmetic \ 1594 of 0 + -1 isn't done as unsigned. */ \ 1595 /* Can't be int, since there is not a shred of a guarantee that int \ 1596 is wide enough to hold a value of something to which pointer can \ 1597 be assigned */ \ 1598 active_reg_t this_reg; \ 1599 \ 1600 DEBUG_STATEMENT (failure_id++); \ 1601 DEBUG_STATEMENT (nfailure_points_pushed++); \ 1602 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ 1603 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ 1604 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ 1605 \ 1606 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \ 1607 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ 1608 \ 1609 /* Ensure we have enough space allocated for what we will push. */ \ 1610 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ 1611 { \ 1612 if (!DOUBLE_FAIL_STACK (fail_stack)) \ 1613 return failure_code; \ 1614 \ 1615 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ 1616 (fail_stack).size); \ 1617 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ 1618 } \ 1619 \ 1620 /* Push the info, starting with the registers. */ \ 1621 DEBUG_PRINT1 ("\n"); \ 1622 \ 1623 if (1) \ 1624 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ 1625 this_reg++) \ 1626 { \ 1627 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \ 1628 DEBUG_STATEMENT (num_regs_pushed++); \ 1629 \ 1630 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ 1631 PUSH_FAILURE_POINTER (regstart[this_reg]); \ 1632 \ 1633 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ 1634 PUSH_FAILURE_POINTER (regend[this_reg]); \ 1635 \ 1636 DEBUG_PRINT2 (" info: %p\n ", \ 1637 reg_info[this_reg].word.pointer); \ 1638 DEBUG_PRINT2 (" match_null=%d", \ 1639 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ 1640 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ 1641 DEBUG_PRINT2 (" matched_something=%d", \ 1642 MATCHED_SOMETHING (reg_info[this_reg])); \ 1643 DEBUG_PRINT2 (" ever_matched=%d", \ 1644 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ 1645 DEBUG_PRINT1 ("\n"); \ 1646 PUSH_FAILURE_ELT (reg_info[this_reg].word); \ 1647 } \ 1648 \ 1649 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\ 1650 PUSH_FAILURE_INT (lowest_active_reg); \ 1651 \ 1652 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\ 1653 PUSH_FAILURE_INT (highest_active_reg); \ 1654 \ 1655 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \ 1656 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ 1657 PUSH_FAILURE_POINTER (pattern_place); \ 1658 \ 1659 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \ 1660 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ 1661 size2); \ 1662 DEBUG_PRINT1 ("'\n"); \ 1663 PUSH_FAILURE_POINTER (string_place); \ 1664 \ 1665 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ 1666 DEBUG_PUSH (failure_id); \ 1667 } while (0) 1668 1669 # ifndef DEFINED_ONCE 1670 /* This is the number of items that are pushed and popped on the stack 1671 for each register. */ 1672 # define NUM_REG_ITEMS 3 1673 1674 /* Individual items aside from the registers. */ 1675 # ifdef DEBUG 1676 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ 1677 # else 1678 # define NUM_NONREG_ITEMS 4 1679 # endif 1680 1681 /* We push at most this many items on the stack. */ 1682 /* We used to use (num_regs - 1), which is the number of registers 1683 this regexp will save; but that was changed to 5 1684 to avoid stack overflow for a regexp with lots of parens. */ 1685 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS) 1686 1687 /* We actually push this many items. */ 1688 # define NUM_FAILURE_ITEMS \ 1689 (((0 \ 1690 ? 0 : highest_active_reg - lowest_active_reg + 1) \ 1691 * NUM_REG_ITEMS) \ 1692 + NUM_NONREG_ITEMS) 1693 1694 /* How many items can still be added to the stack without overflowing it. */ 1695 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) 1696 # endif /* not DEFINED_ONCE */ 1697 1698 1699 /* Pops what PUSH_FAIL_STACK pushes. 1700 1701 We restore into the parameters, all of which should be lvalues: 1702 STR -- the saved data position. 1703 PAT -- the saved pattern position. 1704 LOW_REG, HIGH_REG -- the highest and lowest active registers. 1705 REGSTART, REGEND -- arrays of string positions. 1706 REG_INFO -- array of information about each subexpression. 1707 1708 Also assumes the variables `fail_stack' and (if debugging), `bufp', 1709 `pend', `string1', `size1', `string2', and `size2'. */ 1710 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ 1711 { \ 1712 DEBUG_STATEMENT (unsigned failure_id;) \ 1713 active_reg_t this_reg; \ 1714 const UCHAR_T *string_temp; \ 1715 \ 1716 assert (!FAIL_STACK_EMPTY ()); \ 1717 \ 1718 /* Remove failure points and point to how many regs pushed. */ \ 1719 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ 1720 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ 1721 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ 1722 \ 1723 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ 1724 \ 1725 DEBUG_POP (&failure_id); \ 1726 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ 1727 \ 1728 /* If the saved string location is NULL, it came from an \ 1729 on_failure_keep_string_jump opcode, and we want to throw away the \ 1730 saved NULL, thus retaining our current position in the string. */ \ 1731 string_temp = POP_FAILURE_POINTER (); \ 1732 if (string_temp != NULL) \ 1733 str = (const CHAR_T *) string_temp; \ 1734 \ 1735 DEBUG_PRINT2 (" Popping string %p: `", str); \ 1736 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ 1737 DEBUG_PRINT1 ("'\n"); \ 1738 \ 1739 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \ 1740 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \ 1741 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ 1742 \ 1743 /* Restore register info. */ \ 1744 high_reg = (active_reg_t) POP_FAILURE_INT (); \ 1745 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \ 1746 \ 1747 low_reg = (active_reg_t) POP_FAILURE_INT (); \ 1748 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \ 1749 \ 1750 if (1) \ 1751 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ 1752 { \ 1753 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \ 1754 \ 1755 reg_info[this_reg].word = POP_FAILURE_ELT (); \ 1756 DEBUG_PRINT2 (" info: %p\n", \ 1757 reg_info[this_reg].word.pointer); \ 1758 \ 1759 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ 1760 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ 1761 \ 1762 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ 1763 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ 1764 } \ 1765 else \ 1766 { \ 1767 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \ 1768 { \ 1769 reg_info[this_reg].word.integer = 0; \ 1770 regend[this_reg] = 0; \ 1771 regstart[this_reg] = 0; \ 1772 } \ 1773 highest_active_reg = high_reg; \ 1774 } \ 1775 \ 1776 set_regs_matched_done = 0; \ 1777 DEBUG_STATEMENT (nfailure_points_popped++); \ 1778 } /* POP_FAILURE_POINT */ 1779 1780 /* Structure for per-register (a.k.a. per-group) information. 1781 Other register information, such as the 1782 starting and ending positions (which are addresses), and the list of 1783 inner groups (which is a bits list) are maintained in separate 1784 variables. 1785 1786 We are making a (strictly speaking) nonportable assumption here: that 1787 the compiler will pack our bit fields into something that fits into 1788 the type of `word', i.e., is something that fits into one item on the 1789 failure stack. */ 1790 1791 1792 /* Declarations and macros for re_match_2. */ 1793 1794 typedef union 1795 { 1796 PREFIX(fail_stack_elt_t) word; 1797 struct 1798 { 1799 /* This field is one if this group can match the empty string, 1800 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ 1801 # define MATCH_NULL_UNSET_VALUE 3 1802 unsigned match_null_string_p : 2; 1803 unsigned is_active : 1; 1804 unsigned matched_something : 1; 1805 unsigned ever_matched_something : 1; 1806 } bits; 1807 } PREFIX(register_info_type); 1808 1809 # ifndef DEFINED_ONCE 1810 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) 1811 # define IS_ACTIVE(R) ((R).bits.is_active) 1812 # define MATCHED_SOMETHING(R) ((R).bits.matched_something) 1813 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) 1814 1815 1816 /* Call this when have matched a real character; it sets `matched' flags 1817 for the subexpressions which we are currently inside. Also records 1818 that those subexprs have matched. */ 1819 # define SET_REGS_MATCHED() \ 1820 do \ 1821 { \ 1822 if (!set_regs_matched_done) \ 1823 { \ 1824 active_reg_t r; \ 1825 set_regs_matched_done = 1; \ 1826 for (r = lowest_active_reg; r <= highest_active_reg; r++) \ 1827 { \ 1828 MATCHED_SOMETHING (reg_info[r]) \ 1829 = EVER_MATCHED_SOMETHING (reg_info[r]) \ 1830 = 1; \ 1831 } \ 1832 } \ 1833 } \ 1834 while (0) 1835 # endif /* not DEFINED_ONCE */ 1836 1837 /* Registers are set to a sentinel when they haven't yet matched. */ 1838 static CHAR_T PREFIX(reg_unset_dummy); 1839 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy)) 1840 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE) 1841 1842 /* Subroutine declarations and macros for regex_compile. */ 1843 static void PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg); 1844 static void PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc, 1845 int arg1, int arg2); 1846 static void PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc, 1847 int arg, UCHAR_T *end); 1848 static void PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc, 1849 int arg1, int arg2, UCHAR_T *end); 1850 static boolean PREFIX(at_begline_loc_p) (const CHAR_T *pattern, 1851 const CHAR_T *p, 1852 reg_syntax_t syntax); 1853 static boolean PREFIX(at_endline_loc_p) (const CHAR_T *p, 1854 const CHAR_T *pend, 1855 reg_syntax_t syntax); 1856 # ifdef WCHAR 1857 static reg_errcode_t wcs_compile_range (CHAR_T range_start, 1858 const CHAR_T **p_ptr, 1859 const CHAR_T *pend, 1860 char *translate, 1861 reg_syntax_t syntax, 1862 UCHAR_T *b, 1863 CHAR_T *char_set); 1864 static void insert_space (int num, CHAR_T *loc, CHAR_T *end); 1865 # else /* BYTE */ 1866 static reg_errcode_t byte_compile_range (unsigned int range_start, 1867 const char **p_ptr, 1868 const char *pend, 1869 char *translate, 1870 reg_syntax_t syntax, 1871 unsigned char *b); 1872 # endif /* WCHAR */ 1873 1874 /* Fetch the next character in the uncompiled pattern---translating it 1875 if necessary. Also cast from a signed character in the constant 1876 string passed to us by the user to an unsigned char that we can use 1877 as an array index (in, e.g., `translate'). */ 1878 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, 1879 because it is impossible to allocate 4GB array for some encodings 1880 which have 4 byte character_set like UCS4. */ 1881 # ifndef PATFETCH 1882 # ifdef WCHAR 1883 # define PATFETCH(c) \ 1884 do {if (p == pend) return REG_EEND; \ 1885 c = (UCHAR_T) *p++; \ 1886 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \ 1887 } while (0) 1888 # else /* BYTE */ 1889 # define PATFETCH(c) \ 1890 do {if (p == pend) return REG_EEND; \ 1891 c = (unsigned char) *p++; \ 1892 if (translate) c = (unsigned char) translate[c]; \ 1893 } while (0) 1894 # endif /* WCHAR */ 1895 # endif 1896 1897 /* Fetch the next character in the uncompiled pattern, with no 1898 translation. */ 1899 # define PATFETCH_RAW(c) \ 1900 do {if (p == pend) return REG_EEND; \ 1901 c = (UCHAR_T) *p++; \ 1902 } while (0) 1903 1904 /* Go backwards one character in the pattern. */ 1905 # define PATUNFETCH p-- 1906 1907 1908 /* If `translate' is non-null, return translate[D], else just D. We 1909 cast the subscript to translate because some data is declared as 1910 `char *', to avoid warnings when a string constant is passed. But 1911 when we use a character as a subscript we must make it unsigned. */ 1912 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, 1913 because it is impossible to allocate 4GB array for some encodings 1914 which have 4 byte character_set like UCS4. */ 1915 1916 # ifndef TRANSLATE 1917 # ifdef WCHAR 1918 # define TRANSLATE(d) \ 1919 ((translate && ((UCHAR_T) (d)) <= 0xff) \ 1920 ? (char) translate[(unsigned char) (d)] : (d)) 1921 # else /* BYTE */ 1922 # define TRANSLATE(d) \ 1923 (translate ? (char) translate[(unsigned char) (d)] : (char) (d)) 1924 # endif /* WCHAR */ 1925 # endif 1926 1927 1928 /* Macros for outputting the compiled pattern into `buffer'. */ 1929 1930 /* If the buffer isn't allocated when it comes in, use this. */ 1931 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T)) 1932 1933 /* Make sure we have at least N more bytes of space in buffer. */ 1934 # ifdef WCHAR 1935 # define GET_BUFFER_SPACE(n) \ 1936 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \ 1937 + (n)*sizeof(CHAR_T)) > bufp->allocated) \ 1938 EXTEND_BUFFER () 1939 # else /* BYTE */ 1940 # define GET_BUFFER_SPACE(n) \ 1941 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \ 1942 EXTEND_BUFFER () 1943 # endif /* WCHAR */ 1944 1945 /* Make sure we have one more byte of buffer space and then add C to it. */ 1946 # define BUF_PUSH(c) \ 1947 do { \ 1948 GET_BUFFER_SPACE (1); \ 1949 *b++ = (UCHAR_T) (c); \ 1950 } while (0) 1951 1952 1953 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ 1954 # define BUF_PUSH_2(c1, c2) \ 1955 do { \ 1956 GET_BUFFER_SPACE (2); \ 1957 *b++ = (UCHAR_T) (c1); \ 1958 *b++ = (UCHAR_T) (c2); \ 1959 } while (0) 1960 1961 1962 /* As with BUF_PUSH_2, except for three bytes. */ 1963 # define BUF_PUSH_3(c1, c2, c3) \ 1964 do { \ 1965 GET_BUFFER_SPACE (3); \ 1966 *b++ = (UCHAR_T) (c1); \ 1967 *b++ = (UCHAR_T) (c2); \ 1968 *b++ = (UCHAR_T) (c3); \ 1969 } while (0) 1970 1971 /* Store a jump with opcode OP at LOC to location TO. We store a 1972 relative address offset by the three bytes the jump itself occupies. */ 1973 # define STORE_JUMP(op, loc, to) \ 1974 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE))) 1975 1976 /* Likewise, for a two-argument jump. */ 1977 # define STORE_JUMP2(op, loc, to, arg) \ 1978 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg) 1979 1980 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ 1981 # define INSERT_JUMP(op, loc, to) \ 1982 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b) 1983 1984 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ 1985 # define INSERT_JUMP2(op, loc, to, arg) \ 1986 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\ 1987 arg, b) 1988 1989 /* This is not an arbitrary limit: the arguments which represent offsets 1990 into the pattern are two bytes long. So if 2^16 bytes turns out to 1991 be too small, many things would have to change. */ 1992 /* Any other compiler which, like MSC, has allocation limit below 2^16 1993 bytes will have to use approach similar to what was done below for 1994 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up 1995 reallocating to 0 bytes. Such thing is not going to work too well. 1996 You have been warned!! */ 1997 # ifndef DEFINED_ONCE 1998 # if defined _MSC_VER && !defined WIN32 1999 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. 2000 The REALLOC define eliminates a flurry of conversion warnings, 2001 but is not required. */ 2002 # define MAX_BUF_SIZE 65500L 2003 # define REALLOC(p,s) realloc ((p), (size_t) (s)) 2004 # else 2005 # define MAX_BUF_SIZE (1L << 16) 2006 # define REALLOC(p,s) realloc ((p), (s)) 2007 # endif 2008 2009 /* Extend the buffer by twice its current size via realloc and 2010 reset the pointers that pointed into the old block to point to the 2011 correct places in the new one. If extending the buffer results in it 2012 being larger than MAX_BUF_SIZE, then flag memory exhausted. */ 2013 # if __BOUNDED_POINTERS__ 2014 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) 2015 # define MOVE_BUFFER_POINTER(P) \ 2016 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) 2017 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ 2018 else \ 2019 { \ 2020 SET_HIGH_BOUND (b); \ 2021 SET_HIGH_BOUND (begalt); \ 2022 if (fixup_alt_jump) \ 2023 SET_HIGH_BOUND (fixup_alt_jump); \ 2024 if (laststart) \ 2025 SET_HIGH_BOUND (laststart); \ 2026 if (pending_exact) \ 2027 SET_HIGH_BOUND (pending_exact); \ 2028 } 2029 # else 2030 # define MOVE_BUFFER_POINTER(P) (P) += incr 2031 # define ELSE_EXTEND_BUFFER_HIGH_BOUND 2032 # endif 2033 # endif /* not DEFINED_ONCE */ 2034 2035 # ifdef WCHAR 2036 # define EXTEND_BUFFER() \ 2037 do { \ 2038 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ 2039 int wchar_count; \ 2040 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \ 2041 return REG_ESIZE; \ 2042 bufp->allocated <<= 1; \ 2043 if (bufp->allocated > MAX_BUF_SIZE) \ 2044 bufp->allocated = MAX_BUF_SIZE; \ 2045 /* How many characters the new buffer can have? */ \ 2046 wchar_count = bufp->allocated / sizeof(UCHAR_T); \ 2047 if (wchar_count == 0) wchar_count = 1; \ 2048 /* Truncate the buffer to CHAR_T align. */ \ 2049 bufp->allocated = wchar_count * sizeof(UCHAR_T); \ 2050 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \ 2051 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \ 2052 if (COMPILED_BUFFER_VAR == NULL) \ 2053 return REG_ESPACE; \ 2054 /* If the buffer moved, move all the pointers into it. */ \ 2055 if (old_buffer != COMPILED_BUFFER_VAR) \ 2056 { \ 2057 int incr = COMPILED_BUFFER_VAR - old_buffer; \ 2058 MOVE_BUFFER_POINTER (b); \ 2059 MOVE_BUFFER_POINTER (begalt); \ 2060 if (fixup_alt_jump) \ 2061 MOVE_BUFFER_POINTER (fixup_alt_jump); \ 2062 if (laststart) \ 2063 MOVE_BUFFER_POINTER (laststart); \ 2064 if (pending_exact) \ 2065 MOVE_BUFFER_POINTER (pending_exact); \ 2066 } \ 2067 ELSE_EXTEND_BUFFER_HIGH_BOUND \ 2068 } while (0) 2069 # else /* BYTE */ 2070 # define EXTEND_BUFFER() \ 2071 do { \ 2072 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ 2073 if (bufp->allocated == MAX_BUF_SIZE) \ 2074 return REG_ESIZE; \ 2075 bufp->allocated <<= 1; \ 2076 if (bufp->allocated > MAX_BUF_SIZE) \ 2077 bufp->allocated = MAX_BUF_SIZE; \ 2078 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \ 2079 bufp->allocated); \ 2080 if (COMPILED_BUFFER_VAR == NULL) \ 2081 return REG_ESPACE; \ 2082 /* If the buffer moved, move all the pointers into it. */ \ 2083 if (old_buffer != COMPILED_BUFFER_VAR) \ 2084 { \ 2085 int incr = COMPILED_BUFFER_VAR - old_buffer; \ 2086 MOVE_BUFFER_POINTER (b); \ 2087 MOVE_BUFFER_POINTER (begalt); \ 2088 if (fixup_alt_jump) \ 2089 MOVE_BUFFER_POINTER (fixup_alt_jump); \ 2090 if (laststart) \ 2091 MOVE_BUFFER_POINTER (laststart); \ 2092 if (pending_exact) \ 2093 MOVE_BUFFER_POINTER (pending_exact); \ 2094 } \ 2095 ELSE_EXTEND_BUFFER_HIGH_BOUND \ 2096 } while (0) 2097 # endif /* WCHAR */ 2098 2099 # ifndef DEFINED_ONCE 2100 /* Since we have one byte reserved for the register number argument to 2101 {start,stop}_memory, the maximum number of groups we can report 2102 things about is what fits in that byte. */ 2103 # define MAX_REGNUM 255 2104 2105 /* But patterns can have more than `MAX_REGNUM' registers. We just 2106 ignore the excess. */ 2107 typedef unsigned regnum_t; 2108 2109 2110 /* Macros for the compile stack. */ 2111 2112 /* Since offsets can go either forwards or backwards, this type needs to 2113 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ 2114 /* int may be not enough when sizeof(int) == 2. */ 2115 typedef long pattern_offset_t; 2116 2117 typedef struct 2118 { 2119 pattern_offset_t begalt_offset; 2120 pattern_offset_t fixup_alt_jump; 2121 pattern_offset_t inner_group_offset; 2122 pattern_offset_t laststart_offset; 2123 regnum_t regnum; 2124 } compile_stack_elt_t; 2125 2126 2127 typedef struct 2128 { 2129 compile_stack_elt_t *stack; 2130 unsigned size; 2131 unsigned avail; /* Offset of next open position. */ 2132 } compile_stack_type; 2133 2134 2135 # define INIT_COMPILE_STACK_SIZE 32 2136 2137 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0) 2138 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) 2139 2140 /* The next available element. */ 2141 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) 2142 2143 # endif /* not DEFINED_ONCE */ 2144 2145 /* Set the bit for character C in a list. */ 2146 # ifndef DEFINED_ONCE 2147 # define SET_LIST_BIT(c) \ 2148 (b[((unsigned char) (c)) / BYTEWIDTH] \ 2149 |= 1 << (((unsigned char) c) % BYTEWIDTH)) 2150 # endif /* DEFINED_ONCE */ 2151 2152 /* Get the next unsigned number in the uncompiled pattern. */ 2153 # define GET_UNSIGNED_NUMBER(num) \ 2154 { \ 2155 while (p != pend) \ 2156 { \ 2157 PATFETCH (c); \ 2158 if (c < '0' || c > '9') \ 2159 break; \ 2160 if (num <= RE_DUP_MAX) \ 2161 { \ 2162 if (num < 0) \ 2163 num = 0; \ 2164 num = num * 10 + c - '0'; \ 2165 } \ 2166 } \ 2167 } 2168 2169 # ifndef DEFINED_ONCE 2170 # if defined _LIBC || WIDE_CHAR_SUPPORT 2171 /* The GNU C library provides support for user-defined character classes 2172 and the functions from ISO C amendement 1. */ 2173 # ifdef CHARCLASS_NAME_MAX 2174 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX 2175 # else 2176 /* This shouldn't happen but some implementation might still have this 2177 problem. Use a reasonable default value. */ 2178 # define CHAR_CLASS_MAX_LENGTH 256 2179 # endif 2180 2181 # ifdef _LIBC 2182 # define IS_CHAR_CLASS(string) __wctype (string) 2183 # else 2184 # define IS_CHAR_CLASS(string) wctype (string) 2185 # endif 2186 # else 2187 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ 2188 2189 # define IS_CHAR_CLASS(string) \ 2190 (STREQ (string, "alpha") || STREQ (string, "upper") \ 2191 || STREQ (string, "lower") || STREQ (string, "digit") \ 2192 || STREQ (string, "alnum") || STREQ (string, "xdigit") \ 2193 || STREQ (string, "space") || STREQ (string, "print") \ 2194 || STREQ (string, "punct") || STREQ (string, "graph") \ 2195 || STREQ (string, "cntrl") || STREQ (string, "blank")) 2196 # endif 2197 # endif /* DEFINED_ONCE */ 2198 2199 # ifndef MATCH_MAY_ALLOCATE 2200 2201 /* If we cannot allocate large objects within re_match_2_internal, 2202 we make the fail stack and register vectors global. 2203 The fail stack, we grow to the maximum size when a regexp 2204 is compiled. 2205 The register vectors, we adjust in size each time we 2206 compile a regexp, according to the number of registers it needs. */ 2207 2208 static PREFIX(fail_stack_type) fail_stack; 2209 2210 /* Size with which the following vectors are currently allocated. 2211 That is so we can make them bigger as needed, 2212 but never make them smaller. */ 2213 # ifdef DEFINED_ONCE 2214 static int regs_allocated_size; 2215 2216 static const char ** regstart, ** regend; 2217 static const char ** old_regstart, ** old_regend; 2218 static const char **best_regstart, **best_regend; 2219 static const char **reg_dummy; 2220 # endif /* DEFINED_ONCE */ 2221 2222 static PREFIX(register_info_type) *PREFIX(reg_info); 2223 static PREFIX(register_info_type) *PREFIX(reg_info_dummy); 2224 2225 /* Make the register vectors big enough for NUM_REGS registers, 2226 but don't make them smaller. */ 2227 2228 static void 2229 PREFIX(regex_grow_registers) (int num_regs) 2230 { 2231 if (num_regs > regs_allocated_size) 2232 { 2233 RETALLOC_IF (regstart, num_regs, const char *); 2234 RETALLOC_IF (regend, num_regs, const char *); 2235 RETALLOC_IF (old_regstart, num_regs, const char *); 2236 RETALLOC_IF (old_regend, num_regs, const char *); 2237 RETALLOC_IF (best_regstart, num_regs, const char *); 2238 RETALLOC_IF (best_regend, num_regs, const char *); 2239 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type)); 2240 RETALLOC_IF (reg_dummy, num_regs, const char *); 2241 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type)); 2242 2243 regs_allocated_size = num_regs; 2244 } 2245 } 2246 2247 # endif /* not MATCH_MAY_ALLOCATE */ 2248 2249 # ifndef DEFINED_ONCE 2250 static boolean group_in_compile_stack (compile_stack_type compile_stack, 2251 regnum_t regnum); 2252 # endif /* not DEFINED_ONCE */ 2253 2254 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. 2255 Returns one of error codes defined in `regex.h', or zero for success. 2256 2257 Assumes the `allocated' (and perhaps `buffer') and `translate' 2258 fields are set in BUFP on entry. 2259 2260 If it succeeds, results are put in BUFP (if it returns an error, the 2261 contents of BUFP are undefined): 2262 `buffer' is the compiled pattern; 2263 `syntax' is set to SYNTAX; 2264 `used' is set to the length of the compiled pattern; 2265 `fastmap_accurate' is zero; 2266 `re_nsub' is the number of subexpressions in PATTERN; 2267 `not_bol' and `not_eol' are zero; 2268 2269 The `fastmap' and `newline_anchor' fields are neither 2270 examined nor set. */ 2271 2272 /* Return, freeing storage we allocated. */ 2273 # ifdef WCHAR 2274 # define FREE_STACK_RETURN(value) \ 2275 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value) 2276 # else 2277 # define FREE_STACK_RETURN(value) \ 2278 return (free (compile_stack.stack), value) 2279 # endif /* WCHAR */ 2280 2281 static reg_errcode_t 2282 PREFIX(regex_compile) (const char *ARG_PREFIX(pattern), 2283 size_t ARG_PREFIX(size), reg_syntax_t syntax, 2284 struct re_pattern_buffer *bufp) 2285 { 2286 /* We fetch characters from PATTERN here. Even though PATTERN is 2287 `char *' (i.e., signed), we declare these variables as unsigned, so 2288 they can be reliably used as array indices. */ 2289 register UCHAR_T c, c1; 2290 2291 #ifdef WCHAR 2292 /* A temporary space to keep wchar_t pattern and compiled pattern. */ 2293 CHAR_T *pattern, *COMPILED_BUFFER_VAR; 2294 size_t size; 2295 /* offset buffer for optimization. See convert_mbs_to_wc. */ 2296 int *mbs_offset = NULL; 2297 /* It hold whether each wchar_t is binary data or not. */ 2298 char *is_binary = NULL; 2299 /* A flag whether exactn is handling binary data or not. */ 2300 char is_exactn_bin = FALSE; 2301 #endif /* WCHAR */ 2302 2303 /* A random temporary spot in PATTERN. */ 2304 const CHAR_T *p1; 2305 2306 /* Points to the end of the buffer, where we should append. */ 2307 register UCHAR_T *b; 2308 2309 /* Keeps track of unclosed groups. */ 2310 compile_stack_type compile_stack; 2311 2312 /* Points to the current (ending) position in the pattern. */ 2313 #ifdef WCHAR 2314 const CHAR_T *p; 2315 const CHAR_T *pend; 2316 #else /* BYTE */ 2317 const CHAR_T *p = pattern; 2318 const CHAR_T *pend = pattern + size; 2319 #endif /* WCHAR */ 2320 2321 /* How to translate the characters in the pattern. */ 2322 RE_TRANSLATE_TYPE translate = bufp->translate; 2323 2324 /* Address of the count-byte of the most recently inserted `exactn' 2325 command. This makes it possible to tell if a new exact-match 2326 character can be added to that command or if the character requires 2327 a new `exactn' command. */ 2328 UCHAR_T *pending_exact = 0; 2329 2330 /* Address of start of the most recently finished expression. 2331 This tells, e.g., postfix * where to find the start of its 2332 operand. Reset at the beginning of groups and alternatives. */ 2333 UCHAR_T *laststart = 0; 2334 2335 /* Address of beginning of regexp, or inside of last group. */ 2336 UCHAR_T *begalt; 2337 2338 /* Address of the place where a forward jump should go to the end of 2339 the containing expression. Each alternative of an `or' -- except the 2340 last -- ends with a forward jump of this sort. */ 2341 UCHAR_T *fixup_alt_jump = 0; 2342 2343 /* Counts open-groups as they are encountered. Remembered for the 2344 matching close-group on the compile stack, so the same register 2345 number is put in the stop_memory as the start_memory. */ 2346 regnum_t regnum = 0; 2347 2348 #ifdef WCHAR 2349 /* Initialize the wchar_t PATTERN and offset_buffer. */ 2350 p = pend = pattern = TALLOC(csize + 1, CHAR_T); 2351 mbs_offset = TALLOC(csize + 1, int); 2352 is_binary = TALLOC(csize + 1, char); 2353 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL) 2354 { 2355 free(pattern); 2356 free(mbs_offset); 2357 free(is_binary); 2358 return REG_ESPACE; 2359 } 2360 pattern[csize] = L'\0'; /* sentinel */ 2361 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary); 2362 pend = p + size; 2363 if (size < 0) 2364 { 2365 free(pattern); 2366 free(mbs_offset); 2367 free(is_binary); 2368 return REG_BADPAT; 2369 } 2370 #endif 2371 2372 #ifdef DEBUG 2373 DEBUG_PRINT1 ("\nCompiling pattern: "); 2374 if (debug) 2375 { 2376 unsigned debug_count; 2377 2378 for (debug_count = 0; debug_count < size; debug_count++) 2379 PUT_CHAR (pattern[debug_count]); 2380 putchar ('\n'); 2381 } 2382 #endif /* DEBUG */ 2383 2384 /* Initialize the compile stack. */ 2385 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); 2386 if (compile_stack.stack == NULL) 2387 { 2388 #ifdef WCHAR 2389 free(pattern); 2390 free(mbs_offset); 2391 free(is_binary); 2392 #endif 2393 return REG_ESPACE; 2394 } 2395 2396 compile_stack.size = INIT_COMPILE_STACK_SIZE; 2397 compile_stack.avail = 0; 2398 2399 /* Initialize the pattern buffer. */ 2400 bufp->syntax = syntax; 2401 bufp->fastmap_accurate = 0; 2402 bufp->not_bol = bufp->not_eol = 0; 2403 2404 /* Set `used' to zero, so that if we return an error, the pattern 2405 printer (for debugging) will think there's no pattern. We reset it 2406 at the end. */ 2407 bufp->used = 0; 2408 2409 /* Always count groups, whether or not bufp->no_sub is set. */ 2410 bufp->re_nsub = 0; 2411 2412 #if !defined emacs && !defined SYNTAX_TABLE 2413 /* Initialize the syntax table. */ 2414 init_syntax_once (); 2415 #endif 2416 2417 if (bufp->allocated == 0) 2418 { 2419 if (bufp->buffer) 2420 { /* If zero allocated, but buffer is non-null, try to realloc 2421 enough space. This loses if buffer's address is bogus, but 2422 that is the user's responsibility. */ 2423 #ifdef WCHAR 2424 /* Free bufp->buffer and allocate an array for wchar_t pattern 2425 buffer. */ 2426 free(bufp->buffer); 2427 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T), 2428 UCHAR_T); 2429 #else 2430 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T); 2431 #endif /* WCHAR */ 2432 } 2433 else 2434 { /* Caller did not allocate a buffer. Do it for them. */ 2435 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T), 2436 UCHAR_T); 2437 } 2438 2439 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE); 2440 #ifdef WCHAR 2441 bufp->buffer = (char*)COMPILED_BUFFER_VAR; 2442 #endif /* WCHAR */ 2443 bufp->allocated = INIT_BUF_SIZE; 2444 } 2445 #ifdef WCHAR 2446 else 2447 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer; 2448 #endif 2449 2450 begalt = b = COMPILED_BUFFER_VAR; 2451 2452 /* Loop through the uncompiled pattern until we're at the end. */ 2453 while (p != pend) 2454 { 2455 PATFETCH (c); 2456 2457 switch (c) 2458 { 2459 case '^': 2460 { 2461 if ( /* If at start of pattern, it's an operator. */ 2462 p == pattern + 1 2463 /* If context independent, it's an operator. */ 2464 || syntax & RE_CONTEXT_INDEP_ANCHORS 2465 /* Otherwise, depends on what's come before. */ 2466 || PREFIX(at_begline_loc_p) (pattern, p, syntax)) 2467 BUF_PUSH (begline); 2468 else 2469 goto normal_char; 2470 } 2471 break; 2472 2473 2474 case '$': 2475 { 2476 if ( /* If at end of pattern, it's an operator. */ 2477 p == pend 2478 /* If context independent, it's an operator. */ 2479 || syntax & RE_CONTEXT_INDEP_ANCHORS 2480 /* Otherwise, depends on what's next. */ 2481 || PREFIX(at_endline_loc_p) (p, pend, syntax)) 2482 BUF_PUSH (endline); 2483 else 2484 goto normal_char; 2485 } 2486 break; 2487 2488 2489 case '+': 2490 case '?': 2491 if ((syntax & RE_BK_PLUS_QM) 2492 || (syntax & RE_LIMITED_OPS)) 2493 goto normal_char; 2494 handle_plus: 2495 case '*': 2496 /* If there is no previous pattern... */ 2497 if (!laststart) 2498 { 2499 if (syntax & RE_CONTEXT_INVALID_OPS) 2500 FREE_STACK_RETURN (REG_BADRPT); 2501 else if (!(syntax & RE_CONTEXT_INDEP_OPS)) 2502 goto normal_char; 2503 } 2504 2505 { 2506 /* Are we optimizing this jump? */ 2507 boolean keep_string_p = false; 2508 2509 /* 1 means zero (many) matches is allowed. */ 2510 char zero_times_ok = 0, many_times_ok = 0; 2511 2512 /* If there is a sequence of repetition chars, collapse it 2513 down to just one (the right one). We can't combine 2514 interval operators with these because of, e.g., `a{2}*', 2515 which should only match an even number of `a's. */ 2516 2517 for (;;) 2518 { 2519 zero_times_ok |= c != '+'; 2520 many_times_ok |= c != '?'; 2521 2522 if (p == pend) 2523 break; 2524 2525 PATFETCH (c); 2526 2527 if (c == '*' 2528 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) 2529 ; 2530 2531 else if (syntax & RE_BK_PLUS_QM && c == '\\') 2532 { 2533 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); 2534 2535 PATFETCH (c1); 2536 if (!(c1 == '+' || c1 == '?')) 2537 { 2538 PATUNFETCH; 2539 PATUNFETCH; 2540 break; 2541 } 2542 2543 c = c1; 2544 } 2545 else 2546 { 2547 PATUNFETCH; 2548 break; 2549 } 2550 2551 /* If we get here, we found another repeat character. */ 2552 } 2553 2554 /* Star, etc. applied to an empty pattern is equivalent 2555 to an empty pattern. */ 2556 if (!laststart) 2557 break; 2558 2559 /* Now we know whether or not zero matches is allowed 2560 and also whether or not two or more matches is allowed. */ 2561 if (many_times_ok) 2562 { /* More than one repetition is allowed, so put in at the 2563 end a backward relative jump from `b' to before the next 2564 jump we're going to put in below (which jumps from 2565 laststart to after this jump). 2566 2567 But if we are at the `*' in the exact sequence `.*\n', 2568 insert an unconditional jump backwards to the ., 2569 instead of the beginning of the loop. This way we only 2570 push a failure point once, instead of every time 2571 through the loop. */ 2572 assert (p - 1 > pattern); 2573 2574 /* Allocate the space for the jump. */ 2575 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 2576 2577 /* We know we are not at the first character of the pattern, 2578 because laststart was nonzero. And we've already 2579 incremented `p', by the way, to be the character after 2580 the `*'. Do we have to do something analogous here 2581 for null bytes, because of RE_DOT_NOT_NULL? */ 2582 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') 2583 && zero_times_ok 2584 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') 2585 && !(syntax & RE_DOT_NEWLINE)) 2586 { /* We have .*\n. */ 2587 STORE_JUMP (jump, b, laststart); 2588 keep_string_p = true; 2589 } 2590 else 2591 /* Anything else. */ 2592 STORE_JUMP (maybe_pop_jump, b, laststart - 2593 (1 + OFFSET_ADDRESS_SIZE)); 2594 2595 /* We've added more stuff to the buffer. */ 2596 b += 1 + OFFSET_ADDRESS_SIZE; 2597 } 2598 2599 /* On failure, jump from laststart to b + 3, which will be the 2600 end of the buffer after this jump is inserted. */ 2601 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of 2602 'b + 3'. */ 2603 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 2604 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump 2605 : on_failure_jump, 2606 laststart, b + 1 + OFFSET_ADDRESS_SIZE); 2607 pending_exact = 0; 2608 b += 1 + OFFSET_ADDRESS_SIZE; 2609 2610 if (!zero_times_ok) 2611 { 2612 /* At least one repetition is required, so insert a 2613 `dummy_failure_jump' before the initial 2614 `on_failure_jump' instruction of the loop. This 2615 effects a skip over that instruction the first time 2616 we hit that loop. */ 2617 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 2618 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 2619 2 + 2 * OFFSET_ADDRESS_SIZE); 2620 b += 1 + OFFSET_ADDRESS_SIZE; 2621 } 2622 } 2623 break; 2624 2625 2626 case '.': 2627 laststart = b; 2628 BUF_PUSH (anychar); 2629 break; 2630 2631 2632 case '[': 2633 { 2634 boolean had_char_class = false; 2635 #ifdef WCHAR 2636 CHAR_T range_start = 0xffffffff; 2637 #else 2638 unsigned int range_start = 0xffffffff; 2639 #endif 2640 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2641 2642 #ifdef WCHAR 2643 /* We assume a charset(_not) structure as a wchar_t array. 2644 charset[0] = (re_opcode_t) charset(_not) 2645 charset[1] = l (= length of char_classes) 2646 charset[2] = m (= length of collating_symbols) 2647 charset[3] = n (= length of equivalence_classes) 2648 charset[4] = o (= length of char_ranges) 2649 charset[5] = p (= length of chars) 2650 2651 charset[6] = char_class (wctype_t) 2652 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t) 2653 ... 2654 charset[l+5] = char_class (wctype_t) 2655 2656 charset[l+6] = collating_symbol (wchar_t) 2657 ... 2658 charset[l+m+5] = collating_symbol (wchar_t) 2659 ifdef _LIBC we use the index if 2660 _NL_COLLATE_SYMB_EXTRAMB instead of 2661 wchar_t string. 2662 2663 charset[l+m+6] = equivalence_classes (wchar_t) 2664 ... 2665 charset[l+m+n+5] = equivalence_classes (wchar_t) 2666 ifdef _LIBC we use the index in 2667 _NL_COLLATE_WEIGHT instead of 2668 wchar_t string. 2669 2670 charset[l+m+n+6] = range_start 2671 charset[l+m+n+7] = range_end 2672 ... 2673 charset[l+m+n+2o+4] = range_start 2674 charset[l+m+n+2o+5] = range_end 2675 ifdef _LIBC we use the value looked up 2676 in _NL_COLLATE_COLLSEQ instead of 2677 wchar_t character. 2678 2679 charset[l+m+n+2o+6] = char 2680 ... 2681 charset[l+m+n+2o+p+5] = char 2682 2683 */ 2684 2685 /* We need at least 6 spaces: the opcode, the length of 2686 char_classes, the length of collating_symbols, the length of 2687 equivalence_classes, the length of char_ranges, the length of 2688 chars. */ 2689 GET_BUFFER_SPACE (6); 2690 2691 /* Save b as laststart. And We use laststart as the pointer 2692 to the first element of the charset here. 2693 In other words, laststart[i] indicates charset[i]. */ 2694 laststart = b; 2695 2696 /* We test `*p == '^' twice, instead of using an if 2697 statement, so we only need one BUF_PUSH. */ 2698 BUF_PUSH (*p == '^' ? charset_not : charset); 2699 if (*p == '^') 2700 p++; 2701 2702 /* Push the length of char_classes, the length of 2703 collating_symbols, the length of equivalence_classes, the 2704 length of char_ranges and the length of chars. */ 2705 BUF_PUSH_3 (0, 0, 0); 2706 BUF_PUSH_2 (0, 0); 2707 2708 /* Remember the first position in the bracket expression. */ 2709 p1 = p; 2710 2711 /* charset_not matches newline according to a syntax bit. */ 2712 if ((re_opcode_t) b[-6] == charset_not 2713 && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) 2714 { 2715 BUF_PUSH('\n'); 2716 laststart[5]++; /* Update the length of characters */ 2717 } 2718 2719 /* Read in characters and ranges, setting map bits. */ 2720 for (;;) 2721 { 2722 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2723 2724 PATFETCH (c); 2725 2726 /* \ might escape characters inside [...] and [^...]. */ 2727 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') 2728 { 2729 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); 2730 2731 PATFETCH (c1); 2732 BUF_PUSH(c1); 2733 laststart[5]++; /* Update the length of chars */ 2734 range_start = c1; 2735 continue; 2736 } 2737 2738 /* Could be the end of the bracket expression. If it's 2739 not (i.e., when the bracket expression is `[]' so 2740 far), the ']' character bit gets set way below. */ 2741 if (c == ']' && p != p1 + 1) 2742 break; 2743 2744 /* Look ahead to see if it's a range when the last thing 2745 was a character class. */ 2746 if (had_char_class && c == '-' && *p != ']') 2747 FREE_STACK_RETURN (REG_ERANGE); 2748 2749 /* Look ahead to see if it's a range when the last thing 2750 was a character: if this is a hyphen not at the 2751 beginning or the end of a list, then it's the range 2752 operator. */ 2753 if (c == '-' 2754 && !(p - 2 >= pattern && p[-2] == '[') 2755 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') 2756 && *p != ']') 2757 { 2758 reg_errcode_t ret; 2759 /* Allocate the space for range_start and range_end. */ 2760 GET_BUFFER_SPACE (2); 2761 /* Update the pointer to indicate end of buffer. */ 2762 b += 2; 2763 ret = wcs_compile_range (range_start, &p, pend, translate, 2764 syntax, b, laststart); 2765 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); 2766 range_start = 0xffffffff; 2767 } 2768 else if (p[0] == '-' && p[1] != ']') 2769 { /* This handles ranges made up of characters only. */ 2770 reg_errcode_t ret; 2771 2772 /* Move past the `-'. */ 2773 PATFETCH (c1); 2774 /* Allocate the space for range_start and range_end. */ 2775 GET_BUFFER_SPACE (2); 2776 /* Update the pointer to indicate end of buffer. */ 2777 b += 2; 2778 ret = wcs_compile_range (c, &p, pend, translate, syntax, b, 2779 laststart); 2780 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); 2781 range_start = 0xffffffff; 2782 } 2783 2784 /* See if we're at the beginning of a possible character 2785 class. */ 2786 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') 2787 { /* Leave room for the null. */ 2788 char str[CHAR_CLASS_MAX_LENGTH + 1]; 2789 2790 PATFETCH (c); 2791 c1 = 0; 2792 2793 /* If pattern is `[[:'. */ 2794 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2795 2796 for (;;) 2797 { 2798 PATFETCH (c); 2799 if ((c == ':' && *p == ']') || p == pend) 2800 break; 2801 if (c1 < CHAR_CLASS_MAX_LENGTH) 2802 str[c1++] = c; 2803 else 2804 /* This is in any case an invalid class name. */ 2805 str[0] = '\0'; 2806 } 2807 str[c1] = '\0'; 2808 2809 /* If isn't a word bracketed by `[:' and `:]': 2810 undo the ending character, the letters, and leave 2811 the leading `:' and `[' (but store them as character). */ 2812 if (c == ':' && *p == ']') 2813 { 2814 wctype_t wt; 2815 uintptr_t alignedp; 2816 2817 /* Query the character class as wctype_t. */ 2818 wt = IS_CHAR_CLASS (str); 2819 if (wt == 0) 2820 FREE_STACK_RETURN (REG_ECTYPE); 2821 2822 /* Throw away the ] at the end of the character 2823 class. */ 2824 PATFETCH (c); 2825 2826 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2827 2828 /* Allocate the space for character class. */ 2829 GET_BUFFER_SPACE(CHAR_CLASS_SIZE); 2830 /* Update the pointer to indicate end of buffer. */ 2831 b += CHAR_CLASS_SIZE; 2832 /* Move data which follow character classes 2833 not to violate the data. */ 2834 insert_space(CHAR_CLASS_SIZE, 2835 laststart + 6 + laststart[1], 2836 b - 1); 2837 alignedp = ((uintptr_t)(laststart + 6 + laststart[1]) 2838 + __alignof__(wctype_t) - 1) 2839 & ~(uintptr_t)(__alignof__(wctype_t) - 1); 2840 /* Store the character class. */ 2841 *((wctype_t*)alignedp) = wt; 2842 /* Update length of char_classes */ 2843 laststart[1] += CHAR_CLASS_SIZE; 2844 2845 had_char_class = true; 2846 } 2847 else 2848 { 2849 c1++; 2850 while (c1--) 2851 PATUNFETCH; 2852 BUF_PUSH ('['); 2853 BUF_PUSH (':'); 2854 laststart[5] += 2; /* Update the length of characters */ 2855 range_start = ':'; 2856 had_char_class = false; 2857 } 2858 } 2859 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '=' 2860 || *p == '.')) 2861 { 2862 CHAR_T str[128]; /* Should be large enough. */ 2863 CHAR_T delim = *p; /* '=' or '.' */ 2864 # ifdef _LIBC 2865 uint32_t nrules = 2866 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 2867 # endif 2868 PATFETCH (c); 2869 c1 = 0; 2870 2871 /* If pattern is `[[=' or '[[.'. */ 2872 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2873 2874 for (;;) 2875 { 2876 PATFETCH (c); 2877 if ((c == delim && *p == ']') || p == pend) 2878 break; 2879 if (c1 < sizeof (str) - 1) 2880 str[c1++] = c; 2881 else 2882 /* This is in any case an invalid class name. */ 2883 str[0] = '\0'; 2884 } 2885 str[c1] = '\0'; 2886 2887 if (c == delim && *p == ']' && str[0] != '\0') 2888 { 2889 unsigned int i, offset; 2890 /* If we have no collation data we use the default 2891 collation in which each character is in a class 2892 by itself. It also means that ASCII is the 2893 character set and therefore we cannot have character 2894 with more than one byte in the multibyte 2895 representation. */ 2896 2897 /* If not defined _LIBC, we push the name and 2898 `\0' for the sake of matching performance. */ 2899 int datasize = c1 + 1; 2900 2901 # ifdef _LIBC 2902 int32_t idx = 0; 2903 if (nrules == 0) 2904 # endif 2905 { 2906 if (c1 != 1) 2907 FREE_STACK_RETURN (REG_ECOLLATE); 2908 } 2909 # ifdef _LIBC 2910 else 2911 { 2912 const int32_t *table; 2913 const int32_t *weights; 2914 const int32_t *extra; 2915 const int32_t *indirect; 2916 wint_t *cp; 2917 2918 /* This #include defines a local function! */ 2919 # include <locale/weightwc.h> 2920 2921 if(delim == '=') 2922 { 2923 /* We push the index for equivalence class. */ 2924 cp = (wint_t*)str; 2925 2926 table = (const int32_t *) 2927 _NL_CURRENT (LC_COLLATE, 2928 _NL_COLLATE_TABLEWC); 2929 weights = (const int32_t *) 2930 _NL_CURRENT (LC_COLLATE, 2931 _NL_COLLATE_WEIGHTWC); 2932 extra = (const int32_t *) 2933 _NL_CURRENT (LC_COLLATE, 2934 _NL_COLLATE_EXTRAWC); 2935 indirect = (const int32_t *) 2936 _NL_CURRENT (LC_COLLATE, 2937 _NL_COLLATE_INDIRECTWC); 2938 2939 idx = findidx ((const wint_t**)&cp); 2940 if (idx == 0 || cp < (wint_t*) str + c1) 2941 /* This is no valid character. */ 2942 FREE_STACK_RETURN (REG_ECOLLATE); 2943 2944 str[0] = (wchar_t)idx; 2945 } 2946 else /* delim == '.' */ 2947 { 2948 /* We push collation sequence value 2949 for collating symbol. */ 2950 int32_t table_size; 2951 const int32_t *symb_table; 2952 const unsigned char *extra; 2953 int32_t idx; 2954 int32_t elem; 2955 int32_t second; 2956 int32_t hash; 2957 char char_str[c1]; 2958 2959 /* We have to convert the name to a single-byte 2960 string. This is possible since the names 2961 consist of ASCII characters and the internal 2962 representation is UCS4. */ 2963 for (i = 0; i < c1; ++i) 2964 char_str[i] = str[i]; 2965 2966 table_size = 2967 _NL_CURRENT_WORD (LC_COLLATE, 2968 _NL_COLLATE_SYMB_HASH_SIZEMB); 2969 symb_table = (const int32_t *) 2970 _NL_CURRENT (LC_COLLATE, 2971 _NL_COLLATE_SYMB_TABLEMB); 2972 extra = (const unsigned char *) 2973 _NL_CURRENT (LC_COLLATE, 2974 _NL_COLLATE_SYMB_EXTRAMB); 2975 2976 /* Locate the character in the hashing table. */ 2977 hash = elem_hash (char_str, c1); 2978 2979 idx = 0; 2980 elem = hash % table_size; 2981 second = hash % (table_size - 2); 2982 while (symb_table[2 * elem] != 0) 2983 { 2984 /* First compare the hashing value. */ 2985 if (symb_table[2 * elem] == hash 2986 && c1 == extra[symb_table[2 * elem + 1]] 2987 && memcmp (char_str, 2988 &extra[symb_table[2 * elem + 1] 2989 + 1], c1) == 0) 2990 { 2991 /* Yep, this is the entry. */ 2992 idx = symb_table[2 * elem + 1]; 2993 idx += 1 + extra[idx]; 2994 break; 2995 } 2996 2997 /* Next entry. */ 2998 elem += second; 2999 } 3000 3001 if (symb_table[2 * elem] != 0) 3002 { 3003 /* Compute the index of the byte sequence 3004 in the table. */ 3005 idx += 1 + extra[idx]; 3006 /* Adjust for the alignment. */ 3007 idx = (idx + 3) & ~3; 3008 3009 str[0] = (wchar_t) idx + 4; 3010 } 3011 else if (symb_table[2 * elem] == 0 && c1 == 1) 3012 { 3013 /* No valid character. Match it as a 3014 single byte character. */ 3015 had_char_class = false; 3016 BUF_PUSH(str[0]); 3017 /* Update the length of characters */ 3018 laststart[5]++; 3019 range_start = str[0]; 3020 3021 /* Throw away the ] at the end of the 3022 collating symbol. */ 3023 PATFETCH (c); 3024 /* exit from the switch block. */ 3025 continue; 3026 } 3027 else 3028 FREE_STACK_RETURN (REG_ECOLLATE); 3029 } 3030 datasize = 1; 3031 } 3032 # endif 3033 /* Throw away the ] at the end of the equivalence 3034 class (or collating symbol). */ 3035 PATFETCH (c); 3036 3037 /* Allocate the space for the equivalence class 3038 (or collating symbol) (and '\0' if needed). */ 3039 GET_BUFFER_SPACE(datasize); 3040 /* Update the pointer to indicate end of buffer. */ 3041 b += datasize; 3042 3043 if (delim == '=') 3044 { /* equivalence class */ 3045 /* Calculate the offset of char_ranges, 3046 which is next to equivalence_classes. */ 3047 offset = laststart[1] + laststart[2] 3048 + laststart[3] +6; 3049 /* Insert space. */ 3050 insert_space(datasize, laststart + offset, b - 1); 3051 3052 /* Write the equivalence_class and \0. */ 3053 for (i = 0 ; i < datasize ; i++) 3054 laststart[offset + i] = str[i]; 3055 3056 /* Update the length of equivalence_classes. */ 3057 laststart[3] += datasize; 3058 had_char_class = true; 3059 } 3060 else /* delim == '.' */ 3061 { /* collating symbol */ 3062 /* Calculate the offset of the equivalence_classes, 3063 which is next to collating_symbols. */ 3064 offset = laststart[1] + laststart[2] + 6; 3065 /* Insert space and write the collationg_symbol 3066 and \0. */ 3067 insert_space(datasize, laststart + offset, b-1); 3068 for (i = 0 ; i < datasize ; i++) 3069 laststart[offset + i] = str[i]; 3070 3071 /* In re_match_2_internal if range_start < -1, we 3072 assume -range_start is the offset of the 3073 collating symbol which is specified as 3074 the character of the range start. So we assign 3075 -(laststart[1] + laststart[2] + 6) to 3076 range_start. */ 3077 range_start = -(laststart[1] + laststart[2] + 6); 3078 /* Update the length of collating_symbol. */ 3079 laststart[2] += datasize; 3080 had_char_class = false; 3081 } 3082 } 3083 else 3084 { 3085 c1++; 3086 while (c1--) 3087 PATUNFETCH; 3088 BUF_PUSH ('['); 3089 BUF_PUSH (delim); 3090 laststart[5] += 2; /* Update the length of characters */ 3091 range_start = delim; 3092 had_char_class = false; 3093 } 3094 } 3095 else 3096 { 3097 had_char_class = false; 3098 BUF_PUSH(c); 3099 laststart[5]++; /* Update the length of characters */ 3100 range_start = c; 3101 } 3102 } 3103 3104 #else /* BYTE */ 3105 /* Ensure that we have enough space to push a charset: the 3106 opcode, the length count, and the bitset; 34 bytes in all. */ 3107 GET_BUFFER_SPACE (34); 3108 3109 laststart = b; 3110 3111 /* We test `*p == '^' twice, instead of using an if 3112 statement, so we only need one BUF_PUSH. */ 3113 BUF_PUSH (*p == '^' ? charset_not : charset); 3114 if (*p == '^') 3115 p++; 3116 3117 /* Remember the first position in the bracket expression. */ 3118 p1 = p; 3119 3120 /* Push the number of bytes in the bitmap. */ 3121 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); 3122 3123 /* Clear the whole map. */ 3124 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); 3125 3126 /* charset_not matches newline according to a syntax bit. */ 3127 if ((re_opcode_t) b[-2] == charset_not 3128 && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) 3129 SET_LIST_BIT ('\n'); 3130 3131 /* Read in characters and ranges, setting map bits. */ 3132 for (;;) 3133 { 3134 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3135 3136 PATFETCH (c); 3137 3138 /* \ might escape characters inside [...] and [^...]. */ 3139 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') 3140 { 3141 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); 3142 3143 PATFETCH (c1); 3144 SET_LIST_BIT (c1); 3145 range_start = c1; 3146 continue; 3147 } 3148 3149 /* Could be the end of the bracket expression. If it's 3150 not (i.e., when the bracket expression is `[]' so 3151 far), the ']' character bit gets set way below. */ 3152 if (c == ']' && p != p1 + 1) 3153 break; 3154 3155 /* Look ahead to see if it's a range when the last thing 3156 was a character class. */ 3157 if (had_char_class && c == '-' && *p != ']') 3158 FREE_STACK_RETURN (REG_ERANGE); 3159 3160 /* Look ahead to see if it's a range when the last thing 3161 was a character: if this is a hyphen not at the 3162 beginning or the end of a list, then it's the range 3163 operator. */ 3164 if (c == '-' 3165 && !(p - 2 >= pattern && p[-2] == '[') 3166 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') 3167 && *p != ']') 3168 { 3169 reg_errcode_t ret 3170 = byte_compile_range (range_start, &p, pend, translate, 3171 syntax, b); 3172 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); 3173 range_start = 0xffffffff; 3174 } 3175 3176 else if (p[0] == '-' && p[1] != ']') 3177 { /* This handles ranges made up of characters only. */ 3178 reg_errcode_t ret; 3179 3180 /* Move past the `-'. */ 3181 PATFETCH (c1); 3182 3183 ret = byte_compile_range (c, &p, pend, translate, syntax, b); 3184 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); 3185 range_start = 0xffffffff; 3186 } 3187 3188 /* See if we're at the beginning of a possible character 3189 class. */ 3190 3191 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') 3192 { /* Leave room for the null. */ 3193 char str[CHAR_CLASS_MAX_LENGTH + 1]; 3194 3195 PATFETCH (c); 3196 c1 = 0; 3197 3198 /* If pattern is `[[:'. */ 3199 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3200 3201 for (;;) 3202 { 3203 PATFETCH (c); 3204 if ((c == ':' && *p == ']') || p == pend) 3205 break; 3206 if (c1 < CHAR_CLASS_MAX_LENGTH) 3207 str[c1++] = c; 3208 else 3209 /* This is in any case an invalid class name. */ 3210 str[0] = '\0'; 3211 } 3212 str[c1] = '\0'; 3213 3214 /* If isn't a word bracketed by `[:' and `:]': 3215 undo the ending character, the letters, and leave 3216 the leading `:' and `[' (but set bits for them). */ 3217 if (c == ':' && *p == ']') 3218 { 3219 # if defined _LIBC || WIDE_CHAR_SUPPORT 3220 boolean is_lower = STREQ (str, "lower"); 3221 boolean is_upper = STREQ (str, "upper"); 3222 wctype_t wt; 3223 int ch; 3224 3225 wt = IS_CHAR_CLASS (str); 3226 if (wt == 0) 3227 FREE_STACK_RETURN (REG_ECTYPE); 3228 3229 /* Throw away the ] at the end of the character 3230 class. */ 3231 PATFETCH (c); 3232 3233 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3234 3235 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) 3236 { 3237 # ifdef _LIBC 3238 if (__iswctype (__btowc (ch), wt)) 3239 SET_LIST_BIT (ch); 3240 # else 3241 if (iswctype (btowc (ch), wt)) 3242 SET_LIST_BIT (ch); 3243 # endif 3244 3245 if (translate && (is_upper || is_lower) 3246 && (ISUPPER (ch) || ISLOWER (ch))) 3247 SET_LIST_BIT (ch); 3248 } 3249 3250 had_char_class = true; 3251 # else 3252 int ch; 3253 boolean is_alnum = STREQ (str, "alnum"); 3254 boolean is_alpha = STREQ (str, "alpha"); 3255 boolean is_blank = STREQ (str, "blank"); 3256 boolean is_cntrl = STREQ (str, "cntrl"); 3257 boolean is_digit = STREQ (str, "digit"); 3258 boolean is_graph = STREQ (str, "graph"); 3259 boolean is_lower = STREQ (str, "lower"); 3260 boolean is_print = STREQ (str, "print"); 3261 boolean is_punct = STREQ (str, "punct"); 3262 boolean is_space = STREQ (str, "space"); 3263 boolean is_upper = STREQ (str, "upper"); 3264 boolean is_xdigit = STREQ (str, "xdigit"); 3265 3266 if (!IS_CHAR_CLASS (str)) 3267 FREE_STACK_RETURN (REG_ECTYPE); 3268 3269 /* Throw away the ] at the end of the character 3270 class. */ 3271 PATFETCH (c); 3272 3273 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3274 3275 for (ch = 0; ch < 1 << BYTEWIDTH; ch++) 3276 { 3277 /* This was split into 3 if's to 3278 avoid an arbitrary limit in some compiler. */ 3279 if ( (is_alnum && ISALNUM (ch)) 3280 || (is_alpha && ISALPHA (ch)) 3281 || (is_blank && ISBLANK (ch)) 3282 || (is_cntrl && ISCNTRL (ch))) 3283 SET_LIST_BIT (ch); 3284 if ( (is_digit && ISDIGIT (ch)) 3285 || (is_graph && ISGRAPH (ch)) 3286 || (is_lower && ISLOWER (ch)) 3287 || (is_print && ISPRINT (ch))) 3288 SET_LIST_BIT (ch); 3289 if ( (is_punct && ISPUNCT (ch)) 3290 || (is_space && ISSPACE (ch)) 3291 || (is_upper && ISUPPER (ch)) 3292 || (is_xdigit && ISXDIGIT (ch))) 3293 SET_LIST_BIT (ch); 3294 if ( translate && (is_upper || is_lower) 3295 && (ISUPPER (ch) || ISLOWER (ch))) 3296 SET_LIST_BIT (ch); 3297 } 3298 had_char_class = true; 3299 # endif /* libc || wctype.h */ 3300 } 3301 else 3302 { 3303 c1++; 3304 while (c1--) 3305 PATUNFETCH; 3306 SET_LIST_BIT ('['); 3307 SET_LIST_BIT (':'); 3308 range_start = ':'; 3309 had_char_class = false; 3310 } 3311 } 3312 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=') 3313 { 3314 unsigned char str[MB_LEN_MAX + 1]; 3315 # ifdef _LIBC 3316 uint32_t nrules = 3317 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 3318 # endif 3319 3320 PATFETCH (c); 3321 c1 = 0; 3322 3323 /* If pattern is `[[='. */ 3324 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3325 3326 for (;;) 3327 { 3328 PATFETCH (c); 3329 if ((c == '=' && *p == ']') || p == pend) 3330 break; 3331 if (c1 < MB_LEN_MAX) 3332 str[c1++] = c; 3333 else 3334 /* This is in any case an invalid class name. */ 3335 str[0] = '\0'; 3336 } 3337 str[c1] = '\0'; 3338 3339 if (c == '=' && *p == ']' && str[0] != '\0') 3340 { 3341 /* If we have no collation data we use the default 3342 collation in which each character is in a class 3343 by itself. It also means that ASCII is the 3344 character set and therefore we cannot have character 3345 with more than one byte in the multibyte 3346 representation. */ 3347 # ifdef _LIBC 3348 if (nrules == 0) 3349 # endif 3350 { 3351 if (c1 != 1) 3352 FREE_STACK_RETURN (REG_ECOLLATE); 3353 3354 /* Throw away the ] at the end of the equivalence 3355 class. */ 3356 PATFETCH (c); 3357 3358 /* Set the bit for the character. */ 3359 SET_LIST_BIT (str[0]); 3360 } 3361 # ifdef _LIBC 3362 else 3363 { 3364 /* Try to match the byte sequence in `str' against 3365 those known to the collate implementation. 3366 First find out whether the bytes in `str' are 3367 actually from exactly one character. */ 3368 const int32_t *table; 3369 const unsigned char *weights; 3370 const unsigned char *extra; 3371 const int32_t *indirect; 3372 int32_t idx; 3373 const unsigned char *cp = str; 3374 int ch; 3375 3376 /* This #include defines a local function! */ 3377 # include <locale/weight.h> 3378 3379 table = (const int32_t *) 3380 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB); 3381 weights = (const unsigned char *) 3382 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB); 3383 extra = (const unsigned char *) 3384 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB); 3385 indirect = (const int32_t *) 3386 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB); 3387 3388 idx = findidx (&cp); 3389 if (idx == 0 || cp < str + c1) 3390 /* This is no valid character. */ 3391 FREE_STACK_RETURN (REG_ECOLLATE); 3392 3393 /* Throw away the ] at the end of the equivalence 3394 class. */ 3395 PATFETCH (c); 3396 3397 /* Now we have to go throught the whole table 3398 and find all characters which have the same 3399 first level weight. 3400 3401 XXX Note that this is not entirely correct. 3402 we would have to match multibyte sequences 3403 but this is not possible with the current 3404 implementation. */ 3405 for (ch = 1; ch < 256; ++ch) 3406 /* XXX This test would have to be changed if we 3407 would allow matching multibyte sequences. */ 3408 if (table[ch] > 0) 3409 { 3410 int32_t idx2 = table[ch]; 3411 size_t len = weights[idx2]; 3412 3413 /* Test whether the lenghts match. */ 3414 if (weights[idx] == len) 3415 { 3416 /* They do. New compare the bytes of 3417 the weight. */ 3418 size_t cnt = 0; 3419 3420 while (cnt < len 3421 && (weights[idx + 1 + cnt] 3422 == weights[idx2 + 1 + cnt])) 3423 ++cnt; 3424 3425 if (cnt == len) 3426 /* They match. Mark the character as 3427 acceptable. */ 3428 SET_LIST_BIT (ch); 3429 } 3430 } 3431 } 3432 # endif 3433 had_char_class = true; 3434 } 3435 else 3436 { 3437 c1++; 3438 while (c1--) 3439 PATUNFETCH; 3440 SET_LIST_BIT ('['); 3441 SET_LIST_BIT ('='); 3442 range_start = '='; 3443 had_char_class = false; 3444 } 3445 } 3446 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.') 3447 { 3448 unsigned char str[128]; /* Should be large enough. */ 3449 # ifdef _LIBC 3450 uint32_t nrules = 3451 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 3452 # endif 3453 3454 PATFETCH (c); 3455 c1 = 0; 3456 3457 /* If pattern is `[[.'. */ 3458 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3459 3460 for (;;) 3461 { 3462 PATFETCH (c); 3463 if ((c == '.' && *p == ']') || p == pend) 3464 break; 3465 if (c1 < sizeof (str)) 3466 str[c1++] = c; 3467 else 3468 /* This is in any case an invalid class name. */ 3469 str[0] = '\0'; 3470 } 3471 str[c1] = '\0'; 3472 3473 if (c == '.' && *p == ']' && str[0] != '\0') 3474 { 3475 /* If we have no collation data we use the default 3476 collation in which each character is the name 3477 for its own class which contains only the one 3478 character. It also means that ASCII is the 3479 character set and therefore we cannot have character 3480 with more than one byte in the multibyte 3481 representation. */ 3482 # ifdef _LIBC 3483 if (nrules == 0) 3484 # endif 3485 { 3486 if (c1 != 1) 3487 FREE_STACK_RETURN (REG_ECOLLATE); 3488 3489 /* Throw away the ] at the end of the equivalence 3490 class. */ 3491 PATFETCH (c); 3492 3493 /* Set the bit for the character. */ 3494 SET_LIST_BIT (str[0]); 3495 range_start = ((const unsigned char *) str)[0]; 3496 } 3497 # ifdef _LIBC 3498 else 3499 { 3500 /* Try to match the byte sequence in `str' against 3501 those known to the collate implementation. 3502 First find out whether the bytes in `str' are 3503 actually from exactly one character. */ 3504 int32_t table_size; 3505 const int32_t *symb_table; 3506 const unsigned char *extra; 3507 int32_t idx; 3508 int32_t elem; 3509 int32_t second; 3510 int32_t hash; 3511 3512 table_size = 3513 _NL_CURRENT_WORD (LC_COLLATE, 3514 _NL_COLLATE_SYMB_HASH_SIZEMB); 3515 symb_table = (const int32_t *) 3516 _NL_CURRENT (LC_COLLATE, 3517 _NL_COLLATE_SYMB_TABLEMB); 3518 extra = (const unsigned char *) 3519 _NL_CURRENT (LC_COLLATE, 3520 _NL_COLLATE_SYMB_EXTRAMB); 3521 3522 /* Locate the character in the hashing table. */ 3523 hash = elem_hash (str, c1); 3524 3525 idx = 0; 3526 elem = hash % table_size; 3527 second = hash % (table_size - 2); 3528 while (symb_table[2 * elem] != 0) 3529 { 3530 /* First compare the hashing value. */ 3531 if (symb_table[2 * elem] == hash 3532 && c1 == extra[symb_table[2 * elem + 1]] 3533 && memcmp (str, 3534 &extra[symb_table[2 * elem + 1] 3535 + 1], 3536 c1) == 0) 3537 { 3538 /* Yep, this is the entry. */ 3539 idx = symb_table[2 * elem + 1]; 3540 idx += 1 + extra[idx]; 3541 break; 3542 } 3543 3544 /* Next entry. */ 3545 elem += second; 3546 } 3547 3548 if (symb_table[2 * elem] == 0) 3549 /* This is no valid character. */ 3550 FREE_STACK_RETURN (REG_ECOLLATE); 3551 3552 /* Throw away the ] at the end of the equivalence 3553 class. */ 3554 PATFETCH (c); 3555 3556 /* Now add the multibyte character(s) we found 3557 to the accept list. 3558 3559 XXX Note that this is not entirely correct. 3560 we would have to match multibyte sequences 3561 but this is not possible with the current 3562 implementation. Also, we have to match 3563 collating symbols, which expand to more than 3564 one file, as a whole and not allow the 3565 individual bytes. */ 3566 c1 = extra[idx++]; 3567 if (c1 == 1) 3568 range_start = extra[idx]; 3569 while (c1-- > 0) 3570 { 3571 SET_LIST_BIT (extra[idx]); 3572 ++idx; 3573 } 3574 } 3575 # endif 3576 had_char_class = false; 3577 } 3578 else 3579 { 3580 c1++; 3581 while (c1--) 3582 PATUNFETCH; 3583 SET_LIST_BIT ('['); 3584 SET_LIST_BIT ('.'); 3585 range_start = '.'; 3586 had_char_class = false; 3587 } 3588 } 3589 else 3590 { 3591 had_char_class = false; 3592 SET_LIST_BIT (c); 3593 range_start = c; 3594 } 3595 } 3596 3597 /* Discard any (non)matching list bytes that are all 0 at the 3598 end of the map. Decrease the map-length byte too. */ 3599 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) 3600 b[-1]--; 3601 b += b[-1]; 3602 #endif /* WCHAR */ 3603 } 3604 break; 3605 3606 3607 case '(': 3608 if (syntax & RE_NO_BK_PARENS) 3609 goto handle_open; 3610 else 3611 goto normal_char; 3612 3613 3614 case ')': 3615 if (syntax & RE_NO_BK_PARENS) 3616 goto handle_close; 3617 else 3618 goto normal_char; 3619 3620 3621 case '\n': 3622 if (syntax & RE_NEWLINE_ALT) 3623 goto handle_alt; 3624 else 3625 goto normal_char; 3626 3627 3628 case '|': 3629 if (syntax & RE_NO_BK_VBAR) 3630 goto handle_alt; 3631 else 3632 goto normal_char; 3633 3634 3635 case '{': 3636 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) 3637 goto handle_interval; 3638 else 3639 goto normal_char; 3640 3641 3642 case '\\': 3643 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); 3644 3645 /* Do not translate the character after the \, so that we can 3646 distinguish, e.g., \B from \b, even if we normally would 3647 translate, e.g., B to b. */ 3648 PATFETCH_RAW (c); 3649 3650 switch (c) 3651 { 3652 case '(': 3653 if (syntax & RE_NO_BK_PARENS) 3654 goto normal_backslash; 3655 3656 handle_open: 3657 bufp->re_nsub++; 3658 regnum++; 3659 3660 if (COMPILE_STACK_FULL) 3661 { 3662 RETALLOC (compile_stack.stack, compile_stack.size << 1, 3663 compile_stack_elt_t); 3664 if (compile_stack.stack == NULL) return REG_ESPACE; 3665 3666 compile_stack.size <<= 1; 3667 } 3668 3669 /* These are the values to restore when we hit end of this 3670 group. They are all relative offsets, so that if the 3671 whole pattern moves because of realloc, they will still 3672 be valid. */ 3673 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR; 3674 COMPILE_STACK_TOP.fixup_alt_jump 3675 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0; 3676 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR; 3677 COMPILE_STACK_TOP.regnum = regnum; 3678 3679 /* We will eventually replace the 0 with the number of 3680 groups inner to this one. But do not push a 3681 start_memory for groups beyond the last one we can 3682 represent in the compiled pattern. */ 3683 if (regnum <= MAX_REGNUM) 3684 { 3685 COMPILE_STACK_TOP.inner_group_offset = b 3686 - COMPILED_BUFFER_VAR + 2; 3687 BUF_PUSH_3 (start_memory, regnum, 0); 3688 } 3689 3690 compile_stack.avail++; 3691 3692 fixup_alt_jump = 0; 3693 laststart = 0; 3694 begalt = b; 3695 /* If we've reached MAX_REGNUM groups, then this open 3696 won't actually generate any code, so we'll have to 3697 clear pending_exact explicitly. */ 3698 pending_exact = 0; 3699 break; 3700 3701 3702 case ')': 3703 if (syntax & RE_NO_BK_PARENS) goto normal_backslash; 3704 3705 if (COMPILE_STACK_EMPTY) 3706 { 3707 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) 3708 goto normal_backslash; 3709 else 3710 FREE_STACK_RETURN (REG_ERPAREN); 3711 } 3712 3713 handle_close: 3714 if (fixup_alt_jump) 3715 { /* Push a dummy failure point at the end of the 3716 alternative for a possible future 3717 `pop_failure_jump' to pop. See comments at 3718 `push_dummy_failure' in `re_match_2'. */ 3719 BUF_PUSH (push_dummy_failure); 3720 3721 /* We allocated space for this jump when we assigned 3722 to `fixup_alt_jump', in the `handle_alt' case below. */ 3723 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); 3724 } 3725 3726 /* See similar code for backslashed left paren above. */ 3727 if (COMPILE_STACK_EMPTY) 3728 { 3729 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) 3730 goto normal_char; 3731 else 3732 FREE_STACK_RETURN (REG_ERPAREN); 3733 } 3734 3735 /* Since we just checked for an empty stack above, this 3736 ``can't happen''. */ 3737 assert (compile_stack.avail != 0); 3738 { 3739 /* We don't just want to restore into `regnum', because 3740 later groups should continue to be numbered higher, 3741 as in `(ab)c(de)' -- the second group is #2. */ 3742 regnum_t this_group_regnum; 3743 3744 compile_stack.avail--; 3745 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset; 3746 fixup_alt_jump 3747 = COMPILE_STACK_TOP.fixup_alt_jump 3748 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1 3749 : 0; 3750 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset; 3751 this_group_regnum = COMPILE_STACK_TOP.regnum; 3752 /* If we've reached MAX_REGNUM groups, then this open 3753 won't actually generate any code, so we'll have to 3754 clear pending_exact explicitly. */ 3755 pending_exact = 0; 3756 3757 /* We're at the end of the group, so now we know how many 3758 groups were inside this one. */ 3759 if (this_group_regnum <= MAX_REGNUM) 3760 { 3761 UCHAR_T *inner_group_loc 3762 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset; 3763 3764 *inner_group_loc = regnum - this_group_regnum; 3765 BUF_PUSH_3 (stop_memory, this_group_regnum, 3766 regnum - this_group_regnum); 3767 } 3768 } 3769 break; 3770 3771 3772 case '|': /* `\|'. */ 3773 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) 3774 goto normal_backslash; 3775 handle_alt: 3776 if (syntax & RE_LIMITED_OPS) 3777 goto normal_char; 3778 3779 /* Insert before the previous alternative a jump which 3780 jumps to this alternative if the former fails. */ 3781 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 3782 INSERT_JUMP (on_failure_jump, begalt, 3783 b + 2 + 2 * OFFSET_ADDRESS_SIZE); 3784 pending_exact = 0; 3785 b += 1 + OFFSET_ADDRESS_SIZE; 3786 3787 /* The alternative before this one has a jump after it 3788 which gets executed if it gets matched. Adjust that 3789 jump so it will jump to this alternative's analogous 3790 jump (put in below, which in turn will jump to the next 3791 (if any) alternative's such jump, etc.). The last such 3792 jump jumps to the correct final destination. A picture: 3793 _____ _____ 3794 | | | | 3795 | v | v 3796 a | b | c 3797 3798 If we are at `b', then fixup_alt_jump right now points to a 3799 three-byte space after `a'. We'll put in the jump, set 3800 fixup_alt_jump to right after `b', and leave behind three 3801 bytes which we'll fill in when we get to after `c'. */ 3802 3803 if (fixup_alt_jump) 3804 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); 3805 3806 /* Mark and leave space for a jump after this alternative, 3807 to be filled in later either by next alternative or 3808 when know we're at the end of a series of alternatives. */ 3809 fixup_alt_jump = b; 3810 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 3811 b += 1 + OFFSET_ADDRESS_SIZE; 3812 3813 laststart = 0; 3814 begalt = b; 3815 break; 3816 3817 3818 case '{': 3819 /* If \{ is a literal. */ 3820 if (!(syntax & RE_INTERVALS) 3821 /* If we're at `\{' and it's not the open-interval 3822 operator. */ 3823 || (syntax & RE_NO_BK_BRACES)) 3824 goto normal_backslash; 3825 3826 handle_interval: 3827 { 3828 /* If got here, then the syntax allows intervals. */ 3829 3830 /* At least (most) this many matches must be made. */ 3831 int lower_bound = -1, upper_bound = -1; 3832 3833 /* Place in the uncompiled pattern (i.e., just after 3834 the '{') to go back to if the interval is invalid. */ 3835 const CHAR_T *beg_interval = p; 3836 3837 if (p == pend) 3838 goto invalid_interval; 3839 3840 GET_UNSIGNED_NUMBER (lower_bound); 3841 3842 if (c == ',') 3843 { 3844 GET_UNSIGNED_NUMBER (upper_bound); 3845 if (upper_bound < 0) 3846 upper_bound = RE_DUP_MAX; 3847 } 3848 else 3849 /* Interval such as `{1}' => match exactly once. */ 3850 upper_bound = lower_bound; 3851 3852 if (! (0 <= lower_bound && lower_bound <= upper_bound)) 3853 goto invalid_interval; 3854 3855 if (!(syntax & RE_NO_BK_BRACES)) 3856 { 3857 if (c != '\\' || p == pend) 3858 goto invalid_interval; 3859 PATFETCH (c); 3860 } 3861 3862 if (c != '}') 3863 goto invalid_interval; 3864 3865 /* If it's invalid to have no preceding re. */ 3866 if (!laststart) 3867 { 3868 if (syntax & RE_CONTEXT_INVALID_OPS 3869 && !(syntax & RE_INVALID_INTERVAL_ORD)) 3870 FREE_STACK_RETURN (REG_BADRPT); 3871 else if (syntax & RE_CONTEXT_INDEP_OPS) 3872 laststart = b; 3873 else 3874 goto unfetch_interval; 3875 } 3876 3877 /* We just parsed a valid interval. */ 3878 3879 if (RE_DUP_MAX < upper_bound) 3880 FREE_STACK_RETURN (REG_BADBR); 3881 3882 /* If the upper bound is zero, don't want to succeed at 3883 all; jump from `laststart' to `b + 3', which will be 3884 the end of the buffer after we insert the jump. */ 3885 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' 3886 instead of 'b + 3'. */ 3887 if (upper_bound == 0) 3888 { 3889 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 3890 INSERT_JUMP (jump, laststart, b + 1 3891 + OFFSET_ADDRESS_SIZE); 3892 b += 1 + OFFSET_ADDRESS_SIZE; 3893 } 3894 3895 /* Otherwise, we have a nontrivial interval. When 3896 we're all done, the pattern will look like: 3897 set_number_at <jump count> <upper bound> 3898 set_number_at <succeed_n count> <lower bound> 3899 succeed_n <after jump addr> <succeed_n count> 3900 <body of loop> 3901 jump_n <succeed_n addr> <jump count> 3902 (The upper bound and `jump_n' are omitted if 3903 `upper_bound' is 1, though.) */ 3904 else 3905 { /* If the upper bound is > 1, we need to insert 3906 more at the end of the loop. */ 3907 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE + 3908 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE); 3909 3910 GET_BUFFER_SPACE (nbytes); 3911 3912 /* Initialize lower bound of the `succeed_n', even 3913 though it will be set during matching by its 3914 attendant `set_number_at' (inserted next), 3915 because `re_compile_fastmap' needs to know. 3916 Jump to the `jump_n' we might insert below. */ 3917 INSERT_JUMP2 (succeed_n, laststart, 3918 b + 1 + 2 * OFFSET_ADDRESS_SIZE 3919 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE) 3920 , lower_bound); 3921 b += 1 + 2 * OFFSET_ADDRESS_SIZE; 3922 3923 /* Code to initialize the lower bound. Insert 3924 before the `succeed_n'. The `5' is the last two 3925 bytes of this `set_number_at', plus 3 bytes of 3926 the following `succeed_n'. */ 3927 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE' 3928 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE' 3929 of the following `succeed_n'. */ 3930 PREFIX(insert_op2) (set_number_at, laststart, 1 3931 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b); 3932 b += 1 + 2 * OFFSET_ADDRESS_SIZE; 3933 3934 if (upper_bound > 1) 3935 { /* More than one repetition is allowed, so 3936 append a backward jump to the `succeed_n' 3937 that starts this interval. 3938 3939 When we've reached this during matching, 3940 we'll have matched the interval once, so 3941 jump back only `upper_bound - 1' times. */ 3942 STORE_JUMP2 (jump_n, b, laststart 3943 + 2 * OFFSET_ADDRESS_SIZE + 1, 3944 upper_bound - 1); 3945 b += 1 + 2 * OFFSET_ADDRESS_SIZE; 3946 3947 /* The location we want to set is the second 3948 parameter of the `jump_n'; that is `b-2' as 3949 an absolute address. `laststart' will be 3950 the `set_number_at' we're about to insert; 3951 `laststart+3' the number to set, the source 3952 for the relative address. But we are 3953 inserting into the middle of the pattern -- 3954 so everything is getting moved up by 5. 3955 Conclusion: (b - 2) - (laststart + 3) + 5, 3956 i.e., b - laststart. 3957 3958 We insert this at the beginning of the loop 3959 so that if we fail during matching, we'll 3960 reinitialize the bounds. */ 3961 PREFIX(insert_op2) (set_number_at, laststart, 3962 b - laststart, 3963 upper_bound - 1, b); 3964 b += 1 + 2 * OFFSET_ADDRESS_SIZE; 3965 } 3966 } 3967 pending_exact = 0; 3968 break; 3969 3970 invalid_interval: 3971 if (!(syntax & RE_INVALID_INTERVAL_ORD)) 3972 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR); 3973 unfetch_interval: 3974 /* Match the characters as literals. */ 3975 p = beg_interval; 3976 c = '{'; 3977 if (syntax & RE_NO_BK_BRACES) 3978 goto normal_char; 3979 else 3980 goto normal_backslash; 3981 } 3982 3983 #ifdef emacs 3984 /* There is no way to specify the before_dot and after_dot 3985 operators. rms says this is ok. --karl */ 3986 case '=': 3987 BUF_PUSH (at_dot); 3988 break; 3989 3990 case 's': 3991 laststart = b; 3992 PATFETCH (c); 3993 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); 3994 break; 3995 3996 case 'S': 3997 laststart = b; 3998 PATFETCH (c); 3999 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); 4000 break; 4001 #endif /* emacs */ 4002 4003 4004 case 'w': 4005 if (syntax & RE_NO_GNU_OPS) 4006 goto normal_char; 4007 laststart = b; 4008 BUF_PUSH (wordchar); 4009 break; 4010 4011 4012 case 'W': 4013 if (syntax & RE_NO_GNU_OPS) 4014 goto normal_char; 4015 laststart = b; 4016 BUF_PUSH (notwordchar); 4017 break; 4018 4019 4020 case '<': 4021 if (syntax & RE_NO_GNU_OPS) 4022 goto normal_char; 4023 BUF_PUSH (wordbeg); 4024 break; 4025 4026 case '>': 4027 if (syntax & RE_NO_GNU_OPS) 4028 goto normal_char; 4029 BUF_PUSH (wordend); 4030 break; 4031 4032 case 'b': 4033 if (syntax & RE_NO_GNU_OPS) 4034 goto normal_char; 4035 BUF_PUSH (wordbound); 4036 break; 4037 4038 case 'B': 4039 if (syntax & RE_NO_GNU_OPS) 4040 goto normal_char; 4041 BUF_PUSH (notwordbound); 4042 break; 4043 4044 case '`': 4045 if (syntax & RE_NO_GNU_OPS) 4046 goto normal_char; 4047 BUF_PUSH (begbuf); 4048 break; 4049 4050 case '\'': 4051 if (syntax & RE_NO_GNU_OPS) 4052 goto normal_char; 4053 BUF_PUSH (endbuf); 4054 break; 4055 4056 case '1': case '2': case '3': case '4': case '5': 4057 case '6': case '7': case '8': case '9': 4058 if (syntax & RE_NO_BK_REFS) 4059 goto normal_char; 4060 4061 c1 = c - '0'; 4062 4063 if (c1 > regnum) 4064 FREE_STACK_RETURN (REG_ESUBREG); 4065 4066 /* Can't back reference to a subexpression if inside of it. */ 4067 if (group_in_compile_stack (compile_stack, (regnum_t) c1)) 4068 goto normal_char; 4069 4070 laststart = b; 4071 BUF_PUSH_2 (duplicate, c1); 4072 break; 4073 4074 4075 case '+': 4076 case '?': 4077 if (syntax & RE_BK_PLUS_QM) 4078 goto handle_plus; 4079 else 4080 goto normal_backslash; 4081 4082 default: 4083 normal_backslash: 4084 /* You might think it would be useful for \ to mean 4085 not to translate; but if we don't translate it 4086 it will never match anything. */ 4087 c = TRANSLATE (c); 4088 goto normal_char; 4089 } 4090 break; 4091 4092 4093 default: 4094 /* Expects the character in `c'. */ 4095 normal_char: 4096 /* If no exactn currently being built. */ 4097 if (!pending_exact 4098 #ifdef WCHAR 4099 /* If last exactn handle binary(or character) and 4100 new exactn handle character(or binary). */ 4101 || is_exactn_bin != is_binary[p - 1 - pattern] 4102 #endif /* WCHAR */ 4103 4104 /* If last exactn not at current position. */ 4105 || pending_exact + *pending_exact + 1 != b 4106 4107 /* We have only one byte following the exactn for the count. */ 4108 || *pending_exact == (1 << BYTEWIDTH) - 1 4109 4110 /* If followed by a repetition operator. */ 4111 || *p == '*' || *p == '^' 4112 || ((syntax & RE_BK_PLUS_QM) 4113 ? *p == '\\' && (p[1] == '+' || p[1] == '?') 4114 : (*p == '+' || *p == '?')) 4115 || ((syntax & RE_INTERVALS) 4116 && ((syntax & RE_NO_BK_BRACES) 4117 ? *p == '{' 4118 : (p[0] == '\\' && p[1] == '{')))) 4119 { 4120 /* Start building a new exactn. */ 4121 4122 laststart = b; 4123 4124 #ifdef WCHAR 4125 /* Is this exactn binary data or character? */ 4126 is_exactn_bin = is_binary[p - 1 - pattern]; 4127 if (is_exactn_bin) 4128 BUF_PUSH_2 (exactn_bin, 0); 4129 else 4130 BUF_PUSH_2 (exactn, 0); 4131 #else 4132 BUF_PUSH_2 (exactn, 0); 4133 #endif /* WCHAR */ 4134 pending_exact = b - 1; 4135 } 4136 4137 BUF_PUSH (c); 4138 (*pending_exact)++; 4139 break; 4140 } /* switch (c) */ 4141 } /* while p != pend */ 4142 4143 4144 /* Through the pattern now. */ 4145 4146 if (fixup_alt_jump) 4147 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); 4148 4149 if (!COMPILE_STACK_EMPTY) 4150 FREE_STACK_RETURN (REG_EPAREN); 4151 4152 /* If we don't want backtracking, force success 4153 the first time we reach the end of the compiled pattern. */ 4154 if (syntax & RE_NO_POSIX_BACKTRACKING) 4155 BUF_PUSH (succeed); 4156 4157 #ifdef WCHAR 4158 free (pattern); 4159 free (mbs_offset); 4160 free (is_binary); 4161 #endif 4162 free (compile_stack.stack); 4163 4164 /* We have succeeded; set the length of the buffer. */ 4165 #ifdef WCHAR 4166 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR; 4167 #else 4168 bufp->used = b - bufp->buffer; 4169 #endif 4170 4171 #ifdef DEBUG 4172 if (debug) 4173 { 4174 DEBUG_PRINT1 ("\nCompiled pattern: \n"); 4175 PREFIX(print_compiled_pattern) (bufp); 4176 } 4177 #endif /* DEBUG */ 4178 4179 #ifndef MATCH_MAY_ALLOCATE 4180 /* Initialize the failure stack to the largest possible stack. This 4181 isn't necessary unless we're trying to avoid calling alloca in 4182 the search and match routines. */ 4183 { 4184 int num_regs = bufp->re_nsub + 1; 4185 4186 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size 4187 is strictly greater than re_max_failures, the largest possible stack 4188 is 2 * re_max_failures failure points. */ 4189 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) 4190 { 4191 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); 4192 4193 # ifdef emacs 4194 if (! fail_stack.stack) 4195 fail_stack.stack 4196 = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size 4197 * sizeof (PREFIX(fail_stack_elt_t))); 4198 else 4199 fail_stack.stack 4200 = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack, 4201 (fail_stack.size 4202 * sizeof (PREFIX(fail_stack_elt_t)))); 4203 # else /* not emacs */ 4204 if (! fail_stack.stack) 4205 fail_stack.stack 4206 = (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size 4207 * sizeof (PREFIX(fail_stack_elt_t))); 4208 else 4209 fail_stack.stack 4210 = (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack, 4211 (fail_stack.size 4212 * sizeof (PREFIX(fail_stack_elt_t)))); 4213 # endif /* not emacs */ 4214 } 4215 4216 PREFIX(regex_grow_registers) (num_regs); 4217 } 4218 #endif /* not MATCH_MAY_ALLOCATE */ 4219 4220 return REG_NOERROR; 4221 } /* regex_compile */ 4222 4223 /* Subroutines for `regex_compile'. */ 4224 4225 /* Store OP at LOC followed by two-byte integer parameter ARG. */ 4226 /* ifdef WCHAR, integer parameter is 1 wchar_t. */ 4227 4228 static void 4229 PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg) 4230 { 4231 *loc = (UCHAR_T) op; 4232 STORE_NUMBER (loc + 1, arg); 4233 } 4234 4235 4236 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ 4237 /* ifdef WCHAR, integer parameter is 1 wchar_t. */ 4238 4239 static void 4240 PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, int arg2) 4241 { 4242 *loc = (UCHAR_T) op; 4243 STORE_NUMBER (loc + 1, arg1); 4244 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2); 4245 } 4246 4247 4248 /* Copy the bytes from LOC to END to open up three bytes of space at LOC 4249 for OP followed by two-byte integer parameter ARG. */ 4250 /* ifdef WCHAR, integer parameter is 1 wchar_t. */ 4251 4252 static void 4253 PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc, int arg, UCHAR_T *end) 4254 { 4255 register UCHAR_T *pfrom = end; 4256 register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE; 4257 4258 while (pfrom != loc) 4259 *--pto = *--pfrom; 4260 4261 PREFIX(store_op1) (op, loc, arg); 4262 } 4263 4264 4265 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ 4266 /* ifdef WCHAR, integer parameter is 1 wchar_t. */ 4267 4268 static void 4269 PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, 4270 int arg2, UCHAR_T *end) 4271 { 4272 register UCHAR_T *pfrom = end; 4273 register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE; 4274 4275 while (pfrom != loc) 4276 *--pto = *--pfrom; 4277 4278 PREFIX(store_op2) (op, loc, arg1, arg2); 4279 } 4280 4281 4282 /* P points to just after a ^ in PATTERN. Return true if that ^ comes 4283 after an alternative or a begin-subexpression. We assume there is at 4284 least one character before the ^. */ 4285 4286 static boolean 4287 PREFIX(at_begline_loc_p) (const CHAR_T *pattern, const CHAR_T *p, 4288 reg_syntax_t syntax) 4289 { 4290 const CHAR_T *prev = p - 2; 4291 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; 4292 4293 return 4294 /* After a subexpression? */ 4295 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) 4296 /* After an alternative? */ 4297 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); 4298 } 4299 4300 4301 /* The dual of at_begline_loc_p. This one is for $. We assume there is 4302 at least one character after the $, i.e., `P < PEND'. */ 4303 4304 static boolean 4305 PREFIX(at_endline_loc_p) (const CHAR_T *p, const CHAR_T *pend, 4306 reg_syntax_t syntax) 4307 { 4308 const CHAR_T *next = p; 4309 boolean next_backslash = *next == '\\'; 4310 const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0; 4311 4312 return 4313 /* Before a subexpression? */ 4314 (syntax & RE_NO_BK_PARENS ? *next == ')' 4315 : next_backslash && next_next && *next_next == ')') 4316 /* Before an alternative? */ 4317 || (syntax & RE_NO_BK_VBAR ? *next == '|' 4318 : next_backslash && next_next && *next_next == '|'); 4319 } 4320 4321 #else /* not INSIDE_RECURSION */ 4322 4323 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and 4324 false if it's not. */ 4325 4326 static boolean 4327 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum) 4328 { 4329 int this_element; 4330 4331 for (this_element = compile_stack.avail - 1; 4332 this_element >= 0; 4333 this_element--) 4334 if (compile_stack.stack[this_element].regnum == regnum) 4335 return true; 4336 4337 return false; 4338 } 4339 #endif /* not INSIDE_RECURSION */ 4340 4341 #ifdef INSIDE_RECURSION 4342 4343 #ifdef WCHAR 4344 /* This insert space, which size is "num", into the pattern at "loc". 4345 "end" must point the end of the allocated buffer. */ 4346 static void 4347 insert_space (int num, CHAR_T *loc, CHAR_T *end) 4348 { 4349 register CHAR_T *pto = end; 4350 register CHAR_T *pfrom = end - num; 4351 4352 while (pfrom >= loc) 4353 *pto-- = *pfrom--; 4354 } 4355 #endif /* WCHAR */ 4356 4357 #ifdef WCHAR 4358 static reg_errcode_t 4359 wcs_compile_range (CHAR_T range_start_char, const CHAR_T **p_ptr, 4360 const CHAR_T *pend, RE_TRANSLATE_TYPE translate, 4361 reg_syntax_t syntax, CHAR_T *b, CHAR_T *char_set) 4362 { 4363 const CHAR_T *p = *p_ptr; 4364 CHAR_T range_start, range_end; 4365 reg_errcode_t ret; 4366 # ifdef _LIBC 4367 uint32_t nrules; 4368 uint32_t start_val, end_val; 4369 # endif 4370 if (p == pend) 4371 return REG_ERANGE; 4372 4373 # ifdef _LIBC 4374 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 4375 if (nrules != 0) 4376 { 4377 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE, 4378 _NL_COLLATE_COLLSEQWC); 4379 const unsigned char *extra = (const unsigned char *) 4380 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); 4381 4382 if (range_start_char < -1) 4383 { 4384 /* range_start is a collating symbol. */ 4385 int32_t *wextra; 4386 /* Retreive the index and get collation sequence value. */ 4387 wextra = (int32_t*)(extra + char_set[-range_start_char]); 4388 start_val = wextra[1 + *wextra]; 4389 } 4390 else 4391 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char)); 4392 4393 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0])); 4394 4395 /* Report an error if the range is empty and the syntax prohibits 4396 this. */ 4397 ret = ((syntax & RE_NO_EMPTY_RANGES) 4398 && (start_val > end_val))? REG_ERANGE : REG_NOERROR; 4399 4400 /* Insert space to the end of the char_ranges. */ 4401 insert_space(2, b - char_set[5] - 2, b - 1); 4402 *(b - char_set[5] - 2) = (wchar_t)start_val; 4403 *(b - char_set[5] - 1) = (wchar_t)end_val; 4404 char_set[4]++; /* ranges_index */ 4405 } 4406 else 4407 # endif 4408 { 4409 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char): 4410 range_start_char; 4411 range_end = TRANSLATE (p[0]); 4412 /* Report an error if the range is empty and the syntax prohibits 4413 this. */ 4414 ret = ((syntax & RE_NO_EMPTY_RANGES) 4415 && (range_start > range_end))? REG_ERANGE : REG_NOERROR; 4416 4417 /* Insert space to the end of the char_ranges. */ 4418 insert_space(2, b - char_set[5] - 2, b - 1); 4419 *(b - char_set[5] - 2) = range_start; 4420 *(b - char_set[5] - 1) = range_end; 4421 char_set[4]++; /* ranges_index */ 4422 } 4423 /* Have to increment the pointer into the pattern string, so the 4424 caller isn't still at the ending character. */ 4425 (*p_ptr)++; 4426 4427 return ret; 4428 } 4429 #else /* BYTE */ 4430 /* Read the ending character of a range (in a bracket expression) from the 4431 uncompiled pattern *P_PTR (which ends at PEND). We assume the 4432 starting character is in `P[-2]'. (`P[-1]' is the character `-'.) 4433 Then we set the translation of all bits between the starting and 4434 ending characters (inclusive) in the compiled pattern B. 4435 4436 Return an error code. 4437 4438 We use these short variable names so we can use the same macros as 4439 `regex_compile' itself. */ 4440 4441 static reg_errcode_t 4442 byte_compile_range (unsigned int range_start_char, const char **p_ptr, 4443 const char *pend, RE_TRANSLATE_TYPE translate, 4444 reg_syntax_t syntax, unsigned char *b) 4445 { 4446 unsigned this_char; 4447 const char *p = *p_ptr; 4448 reg_errcode_t ret; 4449 # if _LIBC 4450 const unsigned char *collseq; 4451 unsigned int start_colseq; 4452 unsigned int end_colseq; 4453 # else 4454 unsigned end_char; 4455 # endif 4456 4457 if (p == pend) 4458 return REG_ERANGE; 4459 4460 /* Have to increment the pointer into the pattern string, so the 4461 caller isn't still at the ending character. */ 4462 (*p_ptr)++; 4463 4464 /* Report an error if the range is empty and the syntax prohibits this. */ 4465 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; 4466 4467 # if _LIBC 4468 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE, 4469 _NL_COLLATE_COLLSEQMB); 4470 4471 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)]; 4472 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])]; 4473 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char) 4474 { 4475 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)]; 4476 4477 if (start_colseq <= this_colseq && this_colseq <= end_colseq) 4478 { 4479 SET_LIST_BIT (TRANSLATE (this_char)); 4480 ret = REG_NOERROR; 4481 } 4482 } 4483 # else 4484 /* Here we see why `this_char' has to be larger than an `unsigned 4485 char' -- we would otherwise go into an infinite loop, since all 4486 characters <= 0xff. */ 4487 range_start_char = TRANSLATE (range_start_char); 4488 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE, 4489 and some compilers cast it to int implicitly, so following for_loop 4490 may fall to (almost) infinite loop. 4491 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff. 4492 To avoid this, we cast p[0] to unsigned int and truncate it. */ 4493 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1)); 4494 4495 for (this_char = range_start_char; this_char <= end_char; ++this_char) 4496 { 4497 SET_LIST_BIT (TRANSLATE (this_char)); 4498 ret = REG_NOERROR; 4499 } 4500 # endif 4501 4502 return ret; 4503 } 4504 #endif /* WCHAR */ 4505 4506 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in 4507 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible 4508 characters can start a string that matches the pattern. This fastmap 4509 is used by re_search to skip quickly over impossible starting points. 4510 4511 The caller must supply the address of a (1 << BYTEWIDTH)-byte data 4512 area as BUFP->fastmap. 4513 4514 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in 4515 the pattern buffer. 4516 4517 Returns 0 if we succeed, -2 if an internal error. */ 4518 4519 #ifdef WCHAR 4520 /* local function for re_compile_fastmap. 4521 truncate wchar_t character to char. */ 4522 static unsigned char truncate_wchar (CHAR_T c); 4523 4524 static unsigned char 4525 truncate_wchar (CHAR_T c) 4526 { 4527 unsigned char buf[MB_CUR_MAX]; 4528 mbstate_t state; 4529 int retval; 4530 memset (&state, '\0', sizeof (state)); 4531 # ifdef _LIBC 4532 retval = __wcrtomb (buf, c, &state); 4533 # else 4534 retval = wcrtomb (buf, c, &state); 4535 # endif 4536 return retval > 0 ? buf[0] : (unsigned char) c; 4537 } 4538 #endif /* WCHAR */ 4539 4540 static int 4541 PREFIX(re_compile_fastmap) (struct re_pattern_buffer *bufp) 4542 { 4543 int j, k; 4544 #ifdef MATCH_MAY_ALLOCATE 4545 PREFIX(fail_stack_type) fail_stack; 4546 #endif 4547 #ifndef REGEX_MALLOC 4548 char *destination; 4549 #endif 4550 4551 register char *fastmap = bufp->fastmap; 4552 4553 #ifdef WCHAR 4554 /* We need to cast pattern to (wchar_t*), because we casted this compiled 4555 pattern to (char*) in regex_compile. */ 4556 UCHAR_T *pattern = (UCHAR_T*)bufp->buffer; 4557 register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used); 4558 #else /* BYTE */ 4559 UCHAR_T *pattern = bufp->buffer; 4560 register UCHAR_T *pend = pattern + bufp->used; 4561 #endif /* WCHAR */ 4562 UCHAR_T *p = pattern; 4563 4564 #ifdef REL_ALLOC 4565 /* This holds the pointer to the failure stack, when 4566 it is allocated relocatably. */ 4567 fail_stack_elt_t *failure_stack_ptr; 4568 #endif 4569 4570 /* Assume that each path through the pattern can be null until 4571 proven otherwise. We set this false at the bottom of switch 4572 statement, to which we get only if a particular path doesn't 4573 match the empty string. */ 4574 boolean path_can_be_null = true; 4575 4576 /* We aren't doing a `succeed_n' to begin with. */ 4577 boolean succeed_n_p = false; 4578 4579 assert (fastmap != NULL && p != NULL); 4580 4581 INIT_FAIL_STACK (); 4582 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ 4583 bufp->fastmap_accurate = 1; /* It will be when we're done. */ 4584 bufp->can_be_null = 0; 4585 4586 while (1) 4587 { 4588 if (p == pend || *p == (UCHAR_T) succeed) 4589 { 4590 /* We have reached the (effective) end of pattern. */ 4591 if (!FAIL_STACK_EMPTY ()) 4592 { 4593 bufp->can_be_null |= path_can_be_null; 4594 4595 /* Reset for next path. */ 4596 path_can_be_null = true; 4597 4598 p = fail_stack.stack[--fail_stack.avail].pointer; 4599 4600 continue; 4601 } 4602 else 4603 break; 4604 } 4605 4606 /* We should never be about to go beyond the end of the pattern. */ 4607 assert (p < pend); 4608 4609 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) 4610 { 4611 4612 /* I guess the idea here is to simply not bother with a fastmap 4613 if a backreference is used, since it's too hard to figure out 4614 the fastmap for the corresponding group. Setting 4615 `can_be_null' stops `re_search_2' from using the fastmap, so 4616 that is all we do. */ 4617 case duplicate: 4618 bufp->can_be_null = 1; 4619 goto done; 4620 4621 4622 /* Following are the cases which match a character. These end 4623 with `break'. */ 4624 4625 #ifdef WCHAR 4626 case exactn: 4627 fastmap[truncate_wchar(p[1])] = 1; 4628 break; 4629 #else /* BYTE */ 4630 case exactn: 4631 fastmap[p[1]] = 1; 4632 break; 4633 #endif /* WCHAR */ 4634 #ifdef MBS_SUPPORT 4635 case exactn_bin: 4636 fastmap[p[1]] = 1; 4637 break; 4638 #endif 4639 4640 #ifdef WCHAR 4641 /* It is hard to distinguish fastmap from (multi byte) characters 4642 which depends on current locale. */ 4643 case charset: 4644 case charset_not: 4645 case wordchar: 4646 case notwordchar: 4647 bufp->can_be_null = 1; 4648 goto done; 4649 #else /* BYTE */ 4650 case charset: 4651 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) 4652 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) 4653 fastmap[j] = 1; 4654 break; 4655 4656 4657 case charset_not: 4658 /* Chars beyond end of map must be allowed. */ 4659 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) 4660 fastmap[j] = 1; 4661 4662 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) 4663 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) 4664 fastmap[j] = 1; 4665 break; 4666 4667 4668 case wordchar: 4669 for (j = 0; j < (1 << BYTEWIDTH); j++) 4670 if (SYNTAX (j) == Sword) 4671 fastmap[j] = 1; 4672 break; 4673 4674 4675 case notwordchar: 4676 for (j = 0; j < (1 << BYTEWIDTH); j++) 4677 if (SYNTAX (j) != Sword) 4678 fastmap[j] = 1; 4679 break; 4680 #endif /* WCHAR */ 4681 4682 case anychar: 4683 { 4684 int fastmap_newline = fastmap['\n']; 4685 4686 /* `.' matches anything ... */ 4687 for (j = 0; j < (1 << BYTEWIDTH); j++) 4688 fastmap[j] = 1; 4689 4690 /* ... except perhaps newline. */ 4691 if (!(bufp->syntax & RE_DOT_NEWLINE)) 4692 fastmap['\n'] = fastmap_newline; 4693 4694 /* Return if we have already set `can_be_null'; if we have, 4695 then the fastmap is irrelevant. Something's wrong here. */ 4696 else if (bufp->can_be_null) 4697 goto done; 4698 4699 /* Otherwise, have to check alternative paths. */ 4700 break; 4701 } 4702 4703 #ifdef emacs 4704 case syntaxspec: 4705 k = *p++; 4706 for (j = 0; j < (1 << BYTEWIDTH); j++) 4707 if (SYNTAX (j) == (enum syntaxcode) k) 4708 fastmap[j] = 1; 4709 break; 4710 4711 4712 case notsyntaxspec: 4713 k = *p++; 4714 for (j = 0; j < (1 << BYTEWIDTH); j++) 4715 if (SYNTAX (j) != (enum syntaxcode) k) 4716 fastmap[j] = 1; 4717 break; 4718 4719 4720 /* All cases after this match the empty string. These end with 4721 `continue'. */ 4722 4723 4724 case before_dot: 4725 case at_dot: 4726 case after_dot: 4727 continue; 4728 #endif /* emacs */ 4729 4730 4731 case no_op: 4732 case begline: 4733 case endline: 4734 case begbuf: 4735 case endbuf: 4736 case wordbound: 4737 case notwordbound: 4738 case wordbeg: 4739 case wordend: 4740 case push_dummy_failure: 4741 continue; 4742 4743 4744 case jump_n: 4745 case pop_failure_jump: 4746 case maybe_pop_jump: 4747 case jump: 4748 case jump_past_alt: 4749 case dummy_failure_jump: 4750 EXTRACT_NUMBER_AND_INCR (j, p); 4751 p += j; 4752 if (j > 0) 4753 continue; 4754 4755 /* Jump backward implies we just went through the body of a 4756 loop and matched nothing. Opcode jumped to should be 4757 `on_failure_jump' or `succeed_n'. Just treat it like an 4758 ordinary jump. For a * loop, it has pushed its failure 4759 point already; if so, discard that as redundant. */ 4760 if ((re_opcode_t) *p != on_failure_jump 4761 && (re_opcode_t) *p != succeed_n) 4762 continue; 4763 4764 p++; 4765 EXTRACT_NUMBER_AND_INCR (j, p); 4766 p += j; 4767 4768 /* If what's on the stack is where we are now, pop it. */ 4769 if (!FAIL_STACK_EMPTY () 4770 && fail_stack.stack[fail_stack.avail - 1].pointer == p) 4771 fail_stack.avail--; 4772 4773 continue; 4774 4775 4776 case on_failure_jump: 4777 case on_failure_keep_string_jump: 4778 handle_on_failure_jump: 4779 EXTRACT_NUMBER_AND_INCR (j, p); 4780 4781 /* For some patterns, e.g., `(a?)?', `p+j' here points to the 4782 end of the pattern. We don't want to push such a point, 4783 since when we restore it above, entering the switch will 4784 increment `p' past the end of the pattern. We don't need 4785 to push such a point since we obviously won't find any more 4786 fastmap entries beyond `pend'. Such a pattern can match 4787 the null string, though. */ 4788 if (p + j < pend) 4789 { 4790 if (!PUSH_PATTERN_OP (p + j, fail_stack)) 4791 { 4792 RESET_FAIL_STACK (); 4793 return -2; 4794 } 4795 } 4796 else 4797 bufp->can_be_null = 1; 4798 4799 if (succeed_n_p) 4800 { 4801 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ 4802 succeed_n_p = false; 4803 } 4804 4805 continue; 4806 4807 4808 case succeed_n: 4809 /* Get to the number of times to succeed. */ 4810 p += OFFSET_ADDRESS_SIZE; 4811 4812 /* Increment p past the n for when k != 0. */ 4813 EXTRACT_NUMBER_AND_INCR (k, p); 4814 if (k == 0) 4815 { 4816 p -= 2 * OFFSET_ADDRESS_SIZE; 4817 succeed_n_p = true; /* Spaghetti code alert. */ 4818 goto handle_on_failure_jump; 4819 } 4820 continue; 4821 4822 4823 case set_number_at: 4824 p += 2 * OFFSET_ADDRESS_SIZE; 4825 continue; 4826 4827 4828 case start_memory: 4829 case stop_memory: 4830 p += 2; 4831 continue; 4832 4833 4834 default: 4835 abort (); /* We have listed all the cases. */ 4836 } /* switch *p++ */ 4837 4838 /* Getting here means we have found the possible starting 4839 characters for one path of the pattern -- and that the empty 4840 string does not match. We need not follow this path further. 4841 Instead, look at the next alternative (remembered on the 4842 stack), or quit if no more. The test at the top of the loop 4843 does these things. */ 4844 path_can_be_null = false; 4845 p = pend; 4846 } /* while p */ 4847 4848 /* Set `can_be_null' for the last path (also the first path, if the 4849 pattern is empty). */ 4850 bufp->can_be_null |= path_can_be_null; 4851 4852 done: 4853 RESET_FAIL_STACK (); 4854 return 0; 4855 } 4856 4857 #else /* not INSIDE_RECURSION */ 4858 4859 int 4860 re_compile_fastmap (struct re_pattern_buffer *bufp) 4861 { 4862 # ifdef MBS_SUPPORT 4863 if (MB_CUR_MAX != 1) 4864 return wcs_re_compile_fastmap(bufp); 4865 else 4866 # endif 4867 return byte_re_compile_fastmap(bufp); 4868 } /* re_compile_fastmap */ 4869 #ifdef _LIBC 4870 weak_alias (__re_compile_fastmap, re_compile_fastmap) 4871 #endif 4872 4873 4874 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and 4875 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use 4876 this memory for recording register information. STARTS and ENDS 4877 must be allocated using the malloc library routine, and must each 4878 be at least NUM_REGS * sizeof (regoff_t) bytes long. 4879 4880 If NUM_REGS == 0, then subsequent matches should allocate their own 4881 register data. 4882 4883 Unless this function is called, the first search or match using 4884 PATTERN_BUFFER will allocate its own register data, without 4885 freeing the old data. */ 4886 4887 void 4888 re_set_registers (struct re_pattern_buffer *bufp, 4889 struct re_registers *regs, unsigned num_regs, 4890 regoff_t *starts, regoff_t *ends) 4891 { 4892 if (num_regs) 4893 { 4894 bufp->regs_allocated = REGS_REALLOCATE; 4895 regs->num_regs = num_regs; 4896 regs->start = starts; 4897 regs->end = ends; 4898 } 4899 else 4900 { 4901 bufp->regs_allocated = REGS_UNALLOCATED; 4902 regs->num_regs = 0; 4903 regs->start = regs->end = (regoff_t *) 0; 4904 } 4905 } 4906 #ifdef _LIBC 4907 weak_alias (__re_set_registers, re_set_registers) 4908 #endif 4909 4910 /* Searching routines. */ 4911 4912 /* Like re_search_2, below, but only one string is specified, and 4913 doesn't let you say where to stop matching. */ 4914 4915 int 4916 re_search (struct re_pattern_buffer *bufp, const char *string, int size, 4917 int startpos, int range, struct re_registers *regs) 4918 { 4919 return re_search_2 (bufp, NULL, 0, string, size, startpos, range, 4920 regs, size); 4921 } 4922 #ifdef _LIBC 4923 weak_alias (__re_search, re_search) 4924 #endif 4925 4926 4927 /* Using the compiled pattern in BUFP->buffer, first tries to match the 4928 virtual concatenation of STRING1 and STRING2, starting first at index 4929 STARTPOS, then at STARTPOS + 1, and so on. 4930 4931 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. 4932 4933 RANGE is how far to scan while trying to match. RANGE = 0 means try 4934 only at STARTPOS; in general, the last start tried is STARTPOS + 4935 RANGE. 4936 4937 In REGS, return the indices of the virtual concatenation of STRING1 4938 and STRING2 that matched the entire BUFP->buffer and its contained 4939 subexpressions. 4940 4941 Do not consider matching one past the index STOP in the virtual 4942 concatenation of STRING1 and STRING2. 4943 4944 We return either the position in the strings at which the match was 4945 found, -1 if no match, or -2 if error (such as failure 4946 stack overflow). */ 4947 4948 int 4949 re_search_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, 4950 const char *string2, int size2, int startpos, int range, 4951 struct re_registers *regs, int stop) 4952 { 4953 # ifdef MBS_SUPPORT 4954 if (MB_CUR_MAX != 1) 4955 return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos, 4956 range, regs, stop); 4957 else 4958 # endif 4959 return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos, 4960 range, regs, stop); 4961 } /* re_search_2 */ 4962 #ifdef _LIBC 4963 weak_alias (__re_search_2, re_search_2) 4964 #endif 4965 4966 #endif /* not INSIDE_RECURSION */ 4967 4968 #ifdef INSIDE_RECURSION 4969 4970 #ifdef MATCH_MAY_ALLOCATE 4971 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL 4972 #else 4973 # define FREE_VAR(var) if (var) free (var); var = NULL 4974 #endif 4975 4976 #ifdef WCHAR 4977 # define MAX_ALLOCA_SIZE 2000 4978 4979 # define FREE_WCS_BUFFERS() \ 4980 do { \ 4981 if (size1 > MAX_ALLOCA_SIZE) \ 4982 { \ 4983 free (wcs_string1); \ 4984 free (mbs_offset1); \ 4985 } \ 4986 else \ 4987 { \ 4988 FREE_VAR (wcs_string1); \ 4989 FREE_VAR (mbs_offset1); \ 4990 } \ 4991 if (size2 > MAX_ALLOCA_SIZE) \ 4992 { \ 4993 free (wcs_string2); \ 4994 free (mbs_offset2); \ 4995 } \ 4996 else \ 4997 { \ 4998 FREE_VAR (wcs_string2); \ 4999 FREE_VAR (mbs_offset2); \ 5000 } \ 5001 } while (0) 5002 5003 #endif 5004 5005 5006 static int 5007 PREFIX(re_search_2) (struct re_pattern_buffer *bufp, const char *string1, 5008 int size1, const char *string2, int size2, 5009 int startpos, int range, 5010 struct re_registers *regs, int stop) 5011 { 5012 int val; 5013 register char *fastmap = bufp->fastmap; 5014 register RE_TRANSLATE_TYPE translate = bufp->translate; 5015 int total_size = size1 + size2; 5016 int endpos = startpos + range; 5017 #ifdef WCHAR 5018 /* We need wchar_t* buffers correspond to cstring1, cstring2. */ 5019 wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL; 5020 /* We need the size of wchar_t buffers correspond to csize1, csize2. */ 5021 int wcs_size1 = 0, wcs_size2 = 0; 5022 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ 5023 int *mbs_offset1 = NULL, *mbs_offset2 = NULL; 5024 /* They hold whether each wchar_t is binary data or not. */ 5025 char *is_binary = NULL; 5026 #endif /* WCHAR */ 5027 5028 /* Check for out-of-range STARTPOS. */ 5029 if (startpos < 0 || startpos > total_size) 5030 return -1; 5031 5032 /* Fix up RANGE if it might eventually take us outside 5033 the virtual concatenation of STRING1 and STRING2. 5034 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ 5035 if (endpos < 0) 5036 range = 0 - startpos; 5037 else if (endpos > total_size) 5038 range = total_size - startpos; 5039 5040 /* If the search isn't to be a backwards one, don't waste time in a 5041 search for a pattern that must be anchored. */ 5042 if (bufp->used > 0 && range > 0 5043 && ((re_opcode_t) bufp->buffer[0] == begbuf 5044 /* `begline' is like `begbuf' if it cannot match at newlines. */ 5045 || ((re_opcode_t) bufp->buffer[0] == begline 5046 && !bufp->newline_anchor))) 5047 { 5048 if (startpos > 0) 5049 return -1; 5050 else 5051 range = 1; 5052 } 5053 5054 #ifdef emacs 5055 /* In a forward search for something that starts with \=. 5056 don't keep searching past point. */ 5057 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) 5058 { 5059 range = PT - startpos; 5060 if (range <= 0) 5061 return -1; 5062 } 5063 #endif /* emacs */ 5064 5065 /* Update the fastmap now if not correct already. */ 5066 if (fastmap && !bufp->fastmap_accurate) 5067 if (re_compile_fastmap (bufp) == -2) 5068 return -2; 5069 5070 #ifdef WCHAR 5071 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and 5072 fill them with converted string. */ 5073 if (size1 != 0) 5074 { 5075 if (size1 > MAX_ALLOCA_SIZE) 5076 { 5077 wcs_string1 = TALLOC (size1 + 1, CHAR_T); 5078 mbs_offset1 = TALLOC (size1 + 1, int); 5079 is_binary = TALLOC (size1 + 1, char); 5080 } 5081 else 5082 { 5083 wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T); 5084 mbs_offset1 = REGEX_TALLOC (size1 + 1, int); 5085 is_binary = REGEX_TALLOC (size1 + 1, char); 5086 } 5087 if (!wcs_string1 || !mbs_offset1 || !is_binary) 5088 { 5089 if (size1 > MAX_ALLOCA_SIZE) 5090 { 5091 free (wcs_string1); 5092 free (mbs_offset1); 5093 free (is_binary); 5094 } 5095 else 5096 { 5097 FREE_VAR (wcs_string1); 5098 FREE_VAR (mbs_offset1); 5099 FREE_VAR (is_binary); 5100 } 5101 return -2; 5102 } 5103 wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1, 5104 mbs_offset1, is_binary); 5105 wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */ 5106 if (size1 > MAX_ALLOCA_SIZE) 5107 free (is_binary); 5108 else 5109 FREE_VAR (is_binary); 5110 } 5111 if (size2 != 0) 5112 { 5113 if (size2 > MAX_ALLOCA_SIZE) 5114 { 5115 wcs_string2 = TALLOC (size2 + 1, CHAR_T); 5116 mbs_offset2 = TALLOC (size2 + 1, int); 5117 is_binary = TALLOC (size2 + 1, char); 5118 } 5119 else 5120 { 5121 wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T); 5122 mbs_offset2 = REGEX_TALLOC (size2 + 1, int); 5123 is_binary = REGEX_TALLOC (size2 + 1, char); 5124 } 5125 if (!wcs_string2 || !mbs_offset2 || !is_binary) 5126 { 5127 FREE_WCS_BUFFERS (); 5128 if (size2 > MAX_ALLOCA_SIZE) 5129 free (is_binary); 5130 else 5131 FREE_VAR (is_binary); 5132 return -2; 5133 } 5134 wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2, 5135 mbs_offset2, is_binary); 5136 wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */ 5137 if (size2 > MAX_ALLOCA_SIZE) 5138 free (is_binary); 5139 else 5140 FREE_VAR (is_binary); 5141 } 5142 #endif /* WCHAR */ 5143 5144 5145 /* Loop through the string, looking for a place to start matching. */ 5146 for (;;) 5147 { 5148 /* If a fastmap is supplied, skip quickly over characters that 5149 cannot be the start of a match. If the pattern can match the 5150 null string, however, we don't need to skip characters; we want 5151 the first null string. */ 5152 if (fastmap && startpos < total_size && !bufp->can_be_null) 5153 { 5154 if (range > 0) /* Searching forwards. */ 5155 { 5156 register const char *d; 5157 register int lim = 0; 5158 int irange = range; 5159 5160 if (startpos < size1 && startpos + range >= size1) 5161 lim = range - (size1 - startpos); 5162 5163 d = (startpos >= size1 ? string2 - size1 : string1) + startpos; 5164 5165 /* Written out as an if-else to avoid testing `translate' 5166 inside the loop. */ 5167 if (translate) 5168 while (range > lim 5169 && !fastmap[(unsigned char) 5170 translate[(unsigned char) *d++]]) 5171 range--; 5172 else 5173 while (range > lim && !fastmap[(unsigned char) *d++]) 5174 range--; 5175 5176 startpos += irange - range; 5177 } 5178 else /* Searching backwards. */ 5179 { 5180 register CHAR_T c = (size1 == 0 || startpos >= size1 5181 ? string2[startpos - size1] 5182 : string1[startpos]); 5183 5184 if (!fastmap[(unsigned char) TRANSLATE (c)]) 5185 goto advance; 5186 } 5187 } 5188 5189 /* If can't match the null string, and that's all we have left, fail. */ 5190 if (range >= 0 && startpos == total_size && fastmap 5191 && !bufp->can_be_null) 5192 { 5193 #ifdef WCHAR 5194 FREE_WCS_BUFFERS (); 5195 #endif 5196 return -1; 5197 } 5198 5199 #ifdef WCHAR 5200 val = wcs_re_match_2_internal (bufp, string1, size1, string2, 5201 size2, startpos, regs, stop, 5202 wcs_string1, wcs_size1, 5203 wcs_string2, wcs_size2, 5204 mbs_offset1, mbs_offset2); 5205 #else /* BYTE */ 5206 val = byte_re_match_2_internal (bufp, string1, size1, string2, 5207 size2, startpos, regs, stop); 5208 #endif /* BYTE */ 5209 5210 #ifndef REGEX_MALLOC 5211 # ifdef C_ALLOCA 5212 alloca (0); 5213 # endif 5214 #endif 5215 5216 if (val >= 0) 5217 { 5218 #ifdef WCHAR 5219 FREE_WCS_BUFFERS (); 5220 #endif 5221 return startpos; 5222 } 5223 5224 if (val == -2) 5225 { 5226 #ifdef WCHAR 5227 FREE_WCS_BUFFERS (); 5228 #endif 5229 return -2; 5230 } 5231 5232 advance: 5233 if (!range) 5234 break; 5235 else if (range > 0) 5236 { 5237 range--; 5238 startpos++; 5239 } 5240 else 5241 { 5242 range++; 5243 startpos--; 5244 } 5245 } 5246 #ifdef WCHAR 5247 FREE_WCS_BUFFERS (); 5248 #endif 5249 return -1; 5250 } 5251 5252 #ifdef WCHAR 5253 /* This converts PTR, a pointer into one of the search wchar_t strings 5254 `string1' and `string2' into an multibyte string offset from the 5255 beginning of that string. We use mbs_offset to optimize. 5256 See convert_mbs_to_wcs. */ 5257 # define POINTER_TO_OFFSET(ptr) \ 5258 (FIRST_STRING_P (ptr) \ 5259 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \ 5260 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \ 5261 + csize1))) 5262 #else /* BYTE */ 5263 /* This converts PTR, a pointer into one of the search strings `string1' 5264 and `string2' into an offset from the beginning of that string. */ 5265 # define POINTER_TO_OFFSET(ptr) \ 5266 (FIRST_STRING_P (ptr) \ 5267 ? ((regoff_t) ((ptr) - string1)) \ 5268 : ((regoff_t) ((ptr) - string2 + size1))) 5269 #endif /* WCHAR */ 5270 5271 /* Macros for dealing with the split strings in re_match_2. */ 5272 5273 #define MATCHING_IN_FIRST_STRING (dend == end_match_1) 5274 5275 /* Call before fetching a character with *d. This switches over to 5276 string2 if necessary. */ 5277 #define PREFETCH() \ 5278 while (d == dend) \ 5279 { \ 5280 /* End of string2 => fail. */ \ 5281 if (dend == end_match_2) \ 5282 goto fail; \ 5283 /* End of string1 => advance to string2. */ \ 5284 d = string2; \ 5285 dend = end_match_2; \ 5286 } 5287 5288 /* Test if at very beginning or at very end of the virtual concatenation 5289 of `string1' and `string2'. If only one string, it's `string2'. */ 5290 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) 5291 #define AT_STRINGS_END(d) ((d) == end2) 5292 5293 5294 /* Test if D points to a character which is word-constituent. We have 5295 two special cases to check for: if past the end of string1, look at 5296 the first character in string2; and if before the beginning of 5297 string2, look at the last character in string1. */ 5298 #ifdef WCHAR 5299 /* Use internationalized API instead of SYNTAX. */ 5300 # define WORDCHAR_P(d) \ 5301 (iswalnum ((wint_t)((d) == end1 ? *string2 \ 5302 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \ 5303 || ((d) == end1 ? *string2 \ 5304 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_') 5305 #else /* BYTE */ 5306 # define WORDCHAR_P(d) \ 5307 (SYNTAX ((d) == end1 ? *string2 \ 5308 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ 5309 == Sword) 5310 #endif /* WCHAR */ 5311 5312 /* Disabled due to a compiler bug -- see comment at case wordbound */ 5313 #if 0 5314 /* Test if the character before D and the one at D differ with respect 5315 to being word-constituent. */ 5316 #define AT_WORD_BOUNDARY(d) \ 5317 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ 5318 || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) 5319 #endif 5320 5321 /* Free everything we malloc. */ 5322 #ifdef MATCH_MAY_ALLOCATE 5323 # ifdef WCHAR 5324 # define FREE_VARIABLES() \ 5325 do { \ 5326 REGEX_FREE_STACK (fail_stack.stack); \ 5327 FREE_VAR (regstart); \ 5328 FREE_VAR (regend); \ 5329 FREE_VAR (old_regstart); \ 5330 FREE_VAR (old_regend); \ 5331 FREE_VAR (best_regstart); \ 5332 FREE_VAR (best_regend); \ 5333 FREE_VAR (reg_info); \ 5334 FREE_VAR (reg_dummy); \ 5335 FREE_VAR (reg_info_dummy); \ 5336 if (!cant_free_wcs_buf) \ 5337 { \ 5338 FREE_VAR (string1); \ 5339 FREE_VAR (string2); \ 5340 FREE_VAR (mbs_offset1); \ 5341 FREE_VAR (mbs_offset2); \ 5342 } \ 5343 } while (0) 5344 # else /* BYTE */ 5345 # define FREE_VARIABLES() \ 5346 do { \ 5347 REGEX_FREE_STACK (fail_stack.stack); \ 5348 FREE_VAR (regstart); \ 5349 FREE_VAR (regend); \ 5350 FREE_VAR (old_regstart); \ 5351 FREE_VAR (old_regend); \ 5352 FREE_VAR (best_regstart); \ 5353 FREE_VAR (best_regend); \ 5354 FREE_VAR (reg_info); \ 5355 FREE_VAR (reg_dummy); \ 5356 FREE_VAR (reg_info_dummy); \ 5357 } while (0) 5358 # endif /* WCHAR */ 5359 #else 5360 # ifdef WCHAR 5361 # define FREE_VARIABLES() \ 5362 do { \ 5363 if (!cant_free_wcs_buf) \ 5364 { \ 5365 FREE_VAR (string1); \ 5366 FREE_VAR (string2); \ 5367 FREE_VAR (mbs_offset1); \ 5368 FREE_VAR (mbs_offset2); \ 5369 } \ 5370 } while (0) 5371 # else /* BYTE */ 5372 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ 5373 # endif /* WCHAR */ 5374 #endif /* not MATCH_MAY_ALLOCATE */ 5375 5376 /* These values must meet several constraints. They must not be valid 5377 register values; since we have a limit of 255 registers (because 5378 we use only one byte in the pattern for the register number), we can 5379 use numbers larger than 255. They must differ by 1, because of 5380 NUM_FAILURE_ITEMS above. And the value for the lowest register must 5381 be larger than the value for the highest register, so we do not try 5382 to actually save any registers when none are active. */ 5383 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) 5384 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) 5385 5386 #else /* not INSIDE_RECURSION */ 5387 /* Matching routines. */ 5388 5389 #ifndef emacs /* Emacs never uses this. */ 5390 /* re_match is like re_match_2 except it takes only a single string. */ 5391 5392 int 5393 re_match (struct re_pattern_buffer *bufp, const char *string, 5394 int size, int pos, struct re_registers *regs) 5395 { 5396 int result; 5397 # ifdef MBS_SUPPORT 5398 if (MB_CUR_MAX != 1) 5399 result = wcs_re_match_2_internal (bufp, NULL, 0, string, size, 5400 pos, regs, size, 5401 NULL, 0, NULL, 0, NULL, NULL); 5402 else 5403 # endif 5404 result = byte_re_match_2_internal (bufp, NULL, 0, string, size, 5405 pos, regs, size); 5406 # ifndef REGEX_MALLOC 5407 # ifdef C_ALLOCA 5408 alloca (0); 5409 # endif 5410 # endif 5411 return result; 5412 } 5413 # ifdef _LIBC 5414 weak_alias (__re_match, re_match) 5415 # endif 5416 #endif /* not emacs */ 5417 5418 #endif /* not INSIDE_RECURSION */ 5419 5420 #ifdef INSIDE_RECURSION 5421 static boolean PREFIX(group_match_null_string_p) (UCHAR_T **p, 5422 UCHAR_T *end, 5423 PREFIX(register_info_type) *reg_info); 5424 static boolean PREFIX(alt_match_null_string_p) (UCHAR_T *p, 5425 UCHAR_T *end, 5426 PREFIX(register_info_type) *reg_info); 5427 static boolean PREFIX(common_op_match_null_string_p) (UCHAR_T **p, 5428 UCHAR_T *end, 5429 PREFIX(register_info_type) *reg_info); 5430 static int PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2, 5431 int len, char *translate); 5432 #else /* not INSIDE_RECURSION */ 5433 5434 /* re_match_2 matches the compiled pattern in BUFP against the 5435 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 5436 and SIZE2, respectively). We start matching at POS, and stop 5437 matching at STOP. 5438 5439 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we 5440 store offsets for the substring each group matched in REGS. See the 5441 documentation for exactly how many groups we fill. 5442 5443 We return -1 if no match, -2 if an internal error (such as the 5444 failure stack overflowing). Otherwise, we return the length of the 5445 matched substring. */ 5446 5447 int 5448 re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, 5449 const char *string2, int size2, int pos, 5450 struct re_registers *regs, int stop) 5451 { 5452 int result; 5453 # ifdef MBS_SUPPORT 5454 if (MB_CUR_MAX != 1) 5455 result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2, 5456 pos, regs, stop, 5457 NULL, 0, NULL, 0, NULL, NULL); 5458 else 5459 # endif 5460 result = byte_re_match_2_internal (bufp, string1, size1, string2, size2, 5461 pos, regs, stop); 5462 5463 #ifndef REGEX_MALLOC 5464 # ifdef C_ALLOCA 5465 alloca (0); 5466 # endif 5467 #endif 5468 return result; 5469 } 5470 #ifdef _LIBC 5471 weak_alias (__re_match_2, re_match_2) 5472 #endif 5473 5474 #endif /* not INSIDE_RECURSION */ 5475 5476 #ifdef INSIDE_RECURSION 5477 5478 #ifdef WCHAR 5479 static int count_mbs_length (int *, int); 5480 5481 /* This check the substring (from 0, to length) of the multibyte string, 5482 to which offset_buffer correspond. And count how many wchar_t_characters 5483 the substring occupy. We use offset_buffer to optimization. 5484 See convert_mbs_to_wcs. */ 5485 5486 static int 5487 count_mbs_length(int *offset_buffer, int length) 5488 { 5489 int upper, lower; 5490 5491 /* Check whether the size is valid. */ 5492 if (length < 0) 5493 return -1; 5494 5495 if (offset_buffer == NULL) 5496 return 0; 5497 5498 /* If there are no multibyte character, offset_buffer[i] == i. 5499 Optmize for this case. */ 5500 if (offset_buffer[length] == length) 5501 return length; 5502 5503 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */ 5504 upper = length; 5505 lower = 0; 5506 5507 while (true) 5508 { 5509 int middle = (lower + upper) / 2; 5510 if (middle == lower || middle == upper) 5511 break; 5512 if (offset_buffer[middle] > length) 5513 upper = middle; 5514 else if (offset_buffer[middle] < length) 5515 lower = middle; 5516 else 5517 return middle; 5518 } 5519 5520 return -1; 5521 } 5522 #endif /* WCHAR */ 5523 5524 /* This is a separate function so that we can force an alloca cleanup 5525 afterwards. */ 5526 #ifdef WCHAR 5527 static int 5528 wcs_re_match_2_internal (struct re_pattern_buffer *bufp, 5529 const char *cstring1, int csize1, 5530 const char *cstring2, int csize2, 5531 int pos, 5532 struct re_registers *regs, 5533 int stop, 5534 /* string1 == string2 == NULL means string1/2, size1/2 and 5535 mbs_offset1/2 need seting up in this function. */ 5536 /* We need wchar_t* buffers correspond to cstring1, cstring2. */ 5537 wchar_t *string1, int size1, 5538 wchar_t *string2, int size2, 5539 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ 5540 int *mbs_offset1, int *mbs_offset2) 5541 #else /* BYTE */ 5542 static int 5543 byte_re_match_2_internal (struct re_pattern_buffer *bufp, 5544 const char *string1, int size1, 5545 const char *string2, int size2, 5546 int pos, 5547 struct re_registers *regs, int stop) 5548 #endif /* BYTE */ 5549 { 5550 /* General temporaries. */ 5551 int mcnt; 5552 UCHAR_T *p1; 5553 #ifdef WCHAR 5554 /* They hold whether each wchar_t is binary data or not. */ 5555 char *is_binary = NULL; 5556 /* If true, we can't free string1/2, mbs_offset1/2. */ 5557 int cant_free_wcs_buf = 1; 5558 #endif /* WCHAR */ 5559 5560 /* Just past the end of the corresponding string. */ 5561 const CHAR_T *end1, *end2; 5562 5563 /* Pointers into string1 and string2, just past the last characters in 5564 each to consider matching. */ 5565 const CHAR_T *end_match_1, *end_match_2; 5566 5567 /* Where we are in the data, and the end of the current string. */ 5568 const CHAR_T *d, *dend; 5569 5570 /* Where we are in the pattern, and the end of the pattern. */ 5571 #ifdef WCHAR 5572 UCHAR_T *pattern, *p; 5573 register UCHAR_T *pend; 5574 #else /* BYTE */ 5575 UCHAR_T *p = bufp->buffer; 5576 register UCHAR_T *pend = p + bufp->used; 5577 #endif /* WCHAR */ 5578 5579 /* Mark the opcode just after a start_memory, so we can test for an 5580 empty subpattern when we get to the stop_memory. */ 5581 UCHAR_T *just_past_start_mem = 0; 5582 5583 /* We use this to map every character in the string. */ 5584 RE_TRANSLATE_TYPE translate = bufp->translate; 5585 5586 /* Failure point stack. Each place that can handle a failure further 5587 down the line pushes a failure point on this stack. It consists of 5588 restart, regend, and reg_info for all registers corresponding to 5589 the subexpressions we're currently inside, plus the number of such 5590 registers, and, finally, two char *'s. The first char * is where 5591 to resume scanning the pattern; the second one is where to resume 5592 scanning the strings. If the latter is zero, the failure point is 5593 a ``dummy''; if a failure happens and the failure point is a dummy, 5594 it gets discarded and the next next one is tried. */ 5595 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ 5596 PREFIX(fail_stack_type) fail_stack; 5597 #endif 5598 #ifdef DEBUG 5599 static unsigned failure_id; 5600 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; 5601 #endif 5602 5603 #ifdef REL_ALLOC 5604 /* This holds the pointer to the failure stack, when 5605 it is allocated relocatably. */ 5606 fail_stack_elt_t *failure_stack_ptr; 5607 #endif 5608 5609 /* We fill all the registers internally, independent of what we 5610 return, for use in backreferences. The number here includes 5611 an element for register zero. */ 5612 size_t num_regs = bufp->re_nsub + 1; 5613 5614 /* The currently active registers. */ 5615 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG; 5616 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG; 5617 5618 /* Information on the contents of registers. These are pointers into 5619 the input strings; they record just what was matched (on this 5620 attempt) by a subexpression part of the pattern, that is, the 5621 regnum-th regstart pointer points to where in the pattern we began 5622 matching and the regnum-th regend points to right after where we 5623 stopped matching the regnum-th subexpression. (The zeroth register 5624 keeps track of what the whole pattern matches.) */ 5625 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ 5626 const CHAR_T **regstart, **regend; 5627 #endif 5628 5629 /* If a group that's operated upon by a repetition operator fails to 5630 match anything, then the register for its start will need to be 5631 restored because it will have been set to wherever in the string we 5632 are when we last see its open-group operator. Similarly for a 5633 register's end. */ 5634 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ 5635 const CHAR_T **old_regstart, **old_regend; 5636 #endif 5637 5638 /* The is_active field of reg_info helps us keep track of which (possibly 5639 nested) subexpressions we are currently in. The matched_something 5640 field of reg_info[reg_num] helps us tell whether or not we have 5641 matched any of the pattern so far this time through the reg_num-th 5642 subexpression. These two fields get reset each time through any 5643 loop their register is in. */ 5644 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ 5645 PREFIX(register_info_type) *reg_info; 5646 #endif 5647 5648 /* The following record the register info as found in the above 5649 variables when we find a match better than any we've seen before. 5650 This happens as we backtrack through the failure points, which in 5651 turn happens only if we have not yet matched the entire string. */ 5652 unsigned best_regs_set = false; 5653 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ 5654 const CHAR_T **best_regstart, **best_regend; 5655 #endif 5656 5657 /* Logically, this is `best_regend[0]'. But we don't want to have to 5658 allocate space for that if we're not allocating space for anything 5659 else (see below). Also, we never need info about register 0 for 5660 any of the other register vectors, and it seems rather a kludge to 5661 treat `best_regend' differently than the rest. So we keep track of 5662 the end of the best match so far in a separate variable. We 5663 initialize this to NULL so that when we backtrack the first time 5664 and need to test it, it's not garbage. */ 5665 const CHAR_T *match_end = NULL; 5666 5667 /* This helps SET_REGS_MATCHED avoid doing redundant work. */ 5668 int set_regs_matched_done = 0; 5669 5670 /* Used when we pop values we don't care about. */ 5671 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ 5672 const CHAR_T **reg_dummy; 5673 PREFIX(register_info_type) *reg_info_dummy; 5674 #endif 5675 5676 #ifdef DEBUG 5677 /* Counts the total number of registers pushed. */ 5678 unsigned num_regs_pushed = 0; 5679 #endif 5680 5681 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); 5682 5683 INIT_FAIL_STACK (); 5684 5685 #ifdef MATCH_MAY_ALLOCATE 5686 /* Do not bother to initialize all the register variables if there are 5687 no groups in the pattern, as it takes a fair amount of time. If 5688 there are groups, we include space for register 0 (the whole 5689 pattern), even though we never use it, since it simplifies the 5690 array indexing. We should fix this. */ 5691 if (bufp->re_nsub) 5692 { 5693 regstart = REGEX_TALLOC (num_regs, const CHAR_T *); 5694 regend = REGEX_TALLOC (num_regs, const CHAR_T *); 5695 old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *); 5696 old_regend = REGEX_TALLOC (num_regs, const CHAR_T *); 5697 best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *); 5698 best_regend = REGEX_TALLOC (num_regs, const CHAR_T *); 5699 reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type)); 5700 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *); 5701 reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type)); 5702 5703 if (!(regstart && regend && old_regstart && old_regend && reg_info 5704 && best_regstart && best_regend && reg_dummy && reg_info_dummy)) 5705 { 5706 FREE_VARIABLES (); 5707 return -2; 5708 } 5709 } 5710 else 5711 { 5712 /* We must initialize all our variables to NULL, so that 5713 `FREE_VARIABLES' doesn't try to free them. */ 5714 regstart = regend = old_regstart = old_regend = best_regstart 5715 = best_regend = reg_dummy = NULL; 5716 reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL; 5717 } 5718 #endif /* MATCH_MAY_ALLOCATE */ 5719 5720 /* The starting position is bogus. */ 5721 #ifdef WCHAR 5722 if (pos < 0 || pos > csize1 + csize2) 5723 #else /* BYTE */ 5724 if (pos < 0 || pos > size1 + size2) 5725 #endif 5726 { 5727 FREE_VARIABLES (); 5728 return -1; 5729 } 5730 5731 #ifdef WCHAR 5732 /* Allocate wchar_t array for string1 and string2 and 5733 fill them with converted string. */ 5734 if (string1 == NULL && string2 == NULL) 5735 { 5736 /* We need seting up buffers here. */ 5737 5738 /* We must free wcs buffers in this function. */ 5739 cant_free_wcs_buf = 0; 5740 5741 if (csize1 != 0) 5742 { 5743 string1 = REGEX_TALLOC (csize1 + 1, CHAR_T); 5744 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int); 5745 is_binary = REGEX_TALLOC (csize1 + 1, char); 5746 if (!string1 || !mbs_offset1 || !is_binary) 5747 { 5748 FREE_VAR (string1); 5749 FREE_VAR (mbs_offset1); 5750 FREE_VAR (is_binary); 5751 return -2; 5752 } 5753 } 5754 if (csize2 != 0) 5755 { 5756 string2 = REGEX_TALLOC (csize2 + 1, CHAR_T); 5757 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int); 5758 is_binary = REGEX_TALLOC (csize2 + 1, char); 5759 if (!string2 || !mbs_offset2 || !is_binary) 5760 { 5761 FREE_VAR (string1); 5762 FREE_VAR (mbs_offset1); 5763 FREE_VAR (string2); 5764 FREE_VAR (mbs_offset2); 5765 FREE_VAR (is_binary); 5766 return -2; 5767 } 5768 size2 = convert_mbs_to_wcs(string2, cstring2, csize2, 5769 mbs_offset2, is_binary); 5770 string2[size2] = L'\0'; /* for a sentinel */ 5771 FREE_VAR (is_binary); 5772 } 5773 } 5774 5775 /* We need to cast pattern to (wchar_t*), because we casted this compiled 5776 pattern to (char*) in regex_compile. */ 5777 p = pattern = (CHAR_T*)bufp->buffer; 5778 pend = (CHAR_T*)(bufp->buffer + bufp->used); 5779 5780 #endif /* WCHAR */ 5781 5782 /* Initialize subexpression text positions to -1 to mark ones that no 5783 start_memory/stop_memory has been seen for. Also initialize the 5784 register information struct. */ 5785 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) 5786 { 5787 regstart[mcnt] = regend[mcnt] 5788 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; 5789 5790 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; 5791 IS_ACTIVE (reg_info[mcnt]) = 0; 5792 MATCHED_SOMETHING (reg_info[mcnt]) = 0; 5793 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; 5794 } 5795 5796 /* We move `string1' into `string2' if the latter's empty -- but not if 5797 `string1' is null. */ 5798 if (size2 == 0 && string1 != NULL) 5799 { 5800 string2 = string1; 5801 size2 = size1; 5802 string1 = 0; 5803 size1 = 0; 5804 #ifdef WCHAR 5805 mbs_offset2 = mbs_offset1; 5806 csize2 = csize1; 5807 mbs_offset1 = NULL; 5808 csize1 = 0; 5809 #endif 5810 } 5811 end1 = string1 + size1; 5812 end2 = string2 + size2; 5813 5814 /* Compute where to stop matching, within the two strings. */ 5815 #ifdef WCHAR 5816 if (stop <= csize1) 5817 { 5818 mcnt = count_mbs_length(mbs_offset1, stop); 5819 end_match_1 = string1 + mcnt; 5820 end_match_2 = string2; 5821 } 5822 else 5823 { 5824 if (stop > csize1 + csize2) 5825 stop = csize1 + csize2; 5826 end_match_1 = end1; 5827 mcnt = count_mbs_length(mbs_offset2, stop-csize1); 5828 end_match_2 = string2 + mcnt; 5829 } 5830 if (mcnt < 0) 5831 { /* count_mbs_length return error. */ 5832 FREE_VARIABLES (); 5833 return -1; 5834 } 5835 #else 5836 if (stop <= size1) 5837 { 5838 end_match_1 = string1 + stop; 5839 end_match_2 = string2; 5840 } 5841 else 5842 { 5843 end_match_1 = end1; 5844 end_match_2 = string2 + stop - size1; 5845 } 5846 #endif /* WCHAR */ 5847 5848 /* `p' scans through the pattern as `d' scans through the data. 5849 `dend' is the end of the input string that `d' points within. `d' 5850 is advanced into the following input string whenever necessary, but 5851 this happens before fetching; therefore, at the beginning of the 5852 loop, `d' can be pointing at the end of a string, but it cannot 5853 equal `string2'. */ 5854 #ifdef WCHAR 5855 if (size1 > 0 && pos <= csize1) 5856 { 5857 mcnt = count_mbs_length(mbs_offset1, pos); 5858 d = string1 + mcnt; 5859 dend = end_match_1; 5860 } 5861 else 5862 { 5863 mcnt = count_mbs_length(mbs_offset2, pos-csize1); 5864 d = string2 + mcnt; 5865 dend = end_match_2; 5866 } 5867 5868 if (mcnt < 0) 5869 { /* count_mbs_length return error. */ 5870 FREE_VARIABLES (); 5871 return -1; 5872 } 5873 #else 5874 if (size1 > 0 && pos <= size1) 5875 { 5876 d = string1 + pos; 5877 dend = end_match_1; 5878 } 5879 else 5880 { 5881 d = string2 + pos - size1; 5882 dend = end_match_2; 5883 } 5884 #endif /* WCHAR */ 5885 5886 DEBUG_PRINT1 ("The compiled pattern is:\n"); 5887 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); 5888 DEBUG_PRINT1 ("The string to match is: `"); 5889 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); 5890 DEBUG_PRINT1 ("'\n"); 5891 5892 /* This loops over pattern commands. It exits by returning from the 5893 function if the match is complete, or it drops through if the match 5894 fails at this starting point in the input data. */ 5895 for (;;) 5896 { 5897 #ifdef _LIBC 5898 DEBUG_PRINT2 ("\n%p: ", p); 5899 #else 5900 DEBUG_PRINT2 ("\n0x%x: ", p); 5901 #endif 5902 5903 if (p == pend) 5904 { /* End of pattern means we might have succeeded. */ 5905 DEBUG_PRINT1 ("end of pattern ... "); 5906 5907 /* If we haven't matched the entire string, and we want the 5908 longest match, try backtracking. */ 5909 if (d != end_match_2) 5910 { 5911 /* 1 if this match ends in the same string (string1 or string2) 5912 as the best previous match. */ 5913 boolean same_str_p; 5914 5915 /* 1 if this match is the best seen so far. */ 5916 boolean best_match_p; 5917 5918 same_str_p = (FIRST_STRING_P (match_end) 5919 == MATCHING_IN_FIRST_STRING); 5920 5921 /* AIX compiler got confused when this was combined 5922 with the previous declaration. */ 5923 if (same_str_p) 5924 best_match_p = d > match_end; 5925 else 5926 best_match_p = !MATCHING_IN_FIRST_STRING; 5927 5928 DEBUG_PRINT1 ("backtracking.\n"); 5929 5930 if (!FAIL_STACK_EMPTY ()) 5931 { /* More failure points to try. */ 5932 5933 /* If exceeds best match so far, save it. */ 5934 if (!best_regs_set || best_match_p) 5935 { 5936 best_regs_set = true; 5937 match_end = d; 5938 5939 DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); 5940 5941 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) 5942 { 5943 best_regstart[mcnt] = regstart[mcnt]; 5944 best_regend[mcnt] = regend[mcnt]; 5945 } 5946 } 5947 goto fail; 5948 } 5949 5950 /* If no failure points, don't restore garbage. And if 5951 last match is real best match, don't restore second 5952 best one. */ 5953 else if (best_regs_set && !best_match_p) 5954 { 5955 restore_best_regs: 5956 /* Restore best match. It may happen that `dend == 5957 end_match_1' while the restored d is in string2. 5958 For example, the pattern `x.*y.*z' against the 5959 strings `x-' and `y-z-', if the two strings are 5960 not consecutive in memory. */ 5961 DEBUG_PRINT1 ("Restoring best registers.\n"); 5962 5963 d = match_end; 5964 dend = ((d >= string1 && d <= end1) 5965 ? end_match_1 : end_match_2); 5966 5967 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) 5968 { 5969 regstart[mcnt] = best_regstart[mcnt]; 5970 regend[mcnt] = best_regend[mcnt]; 5971 } 5972 } 5973 } /* d != end_match_2 */ 5974 5975 succeed_label: 5976 DEBUG_PRINT1 ("Accepting match.\n"); 5977 /* If caller wants register contents data back, do it. */ 5978 if (regs && !bufp->no_sub) 5979 { 5980 /* Have the register data arrays been allocated? */ 5981 if (bufp->regs_allocated == REGS_UNALLOCATED) 5982 { /* No. So allocate them with malloc. We need one 5983 extra element beyond `num_regs' for the `-1' marker 5984 GNU code uses. */ 5985 regs->num_regs = MAX (RE_NREGS, num_regs + 1); 5986 regs->start = TALLOC (regs->num_regs, regoff_t); 5987 regs->end = TALLOC (regs->num_regs, regoff_t); 5988 if (regs->start == NULL || regs->end == NULL) 5989 { 5990 FREE_VARIABLES (); 5991 return -2; 5992 } 5993 bufp->regs_allocated = REGS_REALLOCATE; 5994 } 5995 else if (bufp->regs_allocated == REGS_REALLOCATE) 5996 { /* Yes. If we need more elements than were already 5997 allocated, reallocate them. If we need fewer, just 5998 leave it alone. */ 5999 if (regs->num_regs < num_regs + 1) 6000 { 6001 regs->num_regs = num_regs + 1; 6002 RETALLOC (regs->start, regs->num_regs, regoff_t); 6003 RETALLOC (regs->end, regs->num_regs, regoff_t); 6004 if (regs->start == NULL || regs->end == NULL) 6005 { 6006 FREE_VARIABLES (); 6007 return -2; 6008 } 6009 } 6010 } 6011 else 6012 { 6013 /* These braces fend off a "empty body in an else-statement" 6014 warning under GCC when assert expands to nothing. */ 6015 assert (bufp->regs_allocated == REGS_FIXED); 6016 } 6017 6018 /* Convert the pointer data in `regstart' and `regend' to 6019 indices. Register zero has to be set differently, 6020 since we haven't kept track of any info for it. */ 6021 if (regs->num_regs > 0) 6022 { 6023 regs->start[0] = pos; 6024 #ifdef WCHAR 6025 if (MATCHING_IN_FIRST_STRING) 6026 regs->end[0] = mbs_offset1 != NULL ? 6027 mbs_offset1[d-string1] : 0; 6028 else 6029 regs->end[0] = csize1 + (mbs_offset2 != NULL ? 6030 mbs_offset2[d-string2] : 0); 6031 #else 6032 regs->end[0] = (MATCHING_IN_FIRST_STRING 6033 ? ((regoff_t) (d - string1)) 6034 : ((regoff_t) (d - string2 + size1))); 6035 #endif /* WCHAR */ 6036 } 6037 6038 /* Go through the first `min (num_regs, regs->num_regs)' 6039 registers, since that is all we initialized. */ 6040 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs); 6041 mcnt++) 6042 { 6043 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) 6044 regs->start[mcnt] = regs->end[mcnt] = -1; 6045 else 6046 { 6047 regs->start[mcnt] 6048 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); 6049 regs->end[mcnt] 6050 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); 6051 } 6052 } 6053 6054 /* If the regs structure we return has more elements than 6055 were in the pattern, set the extra elements to -1. If 6056 we (re)allocated the registers, this is the case, 6057 because we always allocate enough to have at least one 6058 -1 at the end. */ 6059 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++) 6060 regs->start[mcnt] = regs->end[mcnt] = -1; 6061 } /* regs && !bufp->no_sub */ 6062 6063 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", 6064 nfailure_points_pushed, nfailure_points_popped, 6065 nfailure_points_pushed - nfailure_points_popped); 6066 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); 6067 6068 #ifdef WCHAR 6069 if (MATCHING_IN_FIRST_STRING) 6070 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0; 6071 else 6072 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) + 6073 csize1; 6074 mcnt -= pos; 6075 #else 6076 mcnt = d - pos - (MATCHING_IN_FIRST_STRING 6077 ? string1 6078 : string2 - size1); 6079 #endif /* WCHAR */ 6080 6081 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); 6082 6083 FREE_VARIABLES (); 6084 return mcnt; 6085 } 6086 6087 /* Otherwise match next pattern command. */ 6088 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) 6089 { 6090 /* Ignore these. Used to ignore the n of succeed_n's which 6091 currently have n == 0. */ 6092 case no_op: 6093 DEBUG_PRINT1 ("EXECUTING no_op.\n"); 6094 break; 6095 6096 case succeed: 6097 DEBUG_PRINT1 ("EXECUTING succeed.\n"); 6098 goto succeed_label; 6099 6100 /* Match the next n pattern characters exactly. The following 6101 byte in the pattern defines n, and the n bytes after that 6102 are the characters to match. */ 6103 case exactn: 6104 #ifdef MBS_SUPPORT 6105 case exactn_bin: 6106 #endif 6107 mcnt = *p++; 6108 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); 6109 6110 /* This is written out as an if-else so we don't waste time 6111 testing `translate' inside the loop. */ 6112 if (translate) 6113 { 6114 do 6115 { 6116 PREFETCH (); 6117 #ifdef WCHAR 6118 if (*d <= 0xff) 6119 { 6120 if ((UCHAR_T) translate[(unsigned char) *d++] 6121 != (UCHAR_T) *p++) 6122 goto fail; 6123 } 6124 else 6125 { 6126 if (*d++ != (CHAR_T) *p++) 6127 goto fail; 6128 } 6129 #else 6130 if ((UCHAR_T) translate[(unsigned char) *d++] 6131 != (UCHAR_T) *p++) 6132 goto fail; 6133 #endif /* WCHAR */ 6134 } 6135 while (--mcnt); 6136 } 6137 else 6138 { 6139 do 6140 { 6141 PREFETCH (); 6142 if (*d++ != (CHAR_T) *p++) goto fail; 6143 } 6144 while (--mcnt); 6145 } 6146 SET_REGS_MATCHED (); 6147 break; 6148 6149 6150 /* Match any character except possibly a newline or a null. */ 6151 case anychar: 6152 DEBUG_PRINT1 ("EXECUTING anychar.\n"); 6153 6154 PREFETCH (); 6155 6156 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') 6157 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) 6158 goto fail; 6159 6160 SET_REGS_MATCHED (); 6161 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d); 6162 d++; 6163 break; 6164 6165 6166 case charset: 6167 case charset_not: 6168 { 6169 register UCHAR_T c; 6170 #ifdef WCHAR 6171 unsigned int i, char_class_length, coll_symbol_length, 6172 equiv_class_length, ranges_length, chars_length, length; 6173 CHAR_T *workp, *workp2, *charset_top; 6174 #define WORK_BUFFER_SIZE 128 6175 CHAR_T str_buf[WORK_BUFFER_SIZE]; 6176 # ifdef _LIBC 6177 uint32_t nrules; 6178 # endif /* _LIBC */ 6179 #endif /* WCHAR */ 6180 boolean negate = (re_opcode_t) *(p - 1) == charset_not; 6181 6182 DEBUG_PRINT2 ("EXECUTING charset%s.\n", negate ? "_not" : ""); 6183 PREFETCH (); 6184 c = TRANSLATE (*d); /* The character to match. */ 6185 #ifdef WCHAR 6186 # ifdef _LIBC 6187 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 6188 # endif /* _LIBC */ 6189 charset_top = p - 1; 6190 char_class_length = *p++; 6191 coll_symbol_length = *p++; 6192 equiv_class_length = *p++; 6193 ranges_length = *p++; 6194 chars_length = *p++; 6195 /* p points charset[6], so the address of the next instruction 6196 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'], 6197 where l=length of char_classes, m=length of collating_symbol, 6198 n=equivalence_class, o=length of char_range, 6199 p'=length of character. */ 6200 workp = p; 6201 /* Update p to indicate the next instruction. */ 6202 p += char_class_length + coll_symbol_length+ equiv_class_length + 6203 2*ranges_length + chars_length; 6204 6205 /* match with char_class? */ 6206 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE) 6207 { 6208 wctype_t wctype; 6209 uintptr_t alignedp = ((uintptr_t)workp 6210 + __alignof__(wctype_t) - 1) 6211 & ~(uintptr_t)(__alignof__(wctype_t) - 1); 6212 wctype = *((wctype_t*)alignedp); 6213 workp += CHAR_CLASS_SIZE; 6214 # ifdef _LIBC 6215 if (__iswctype((wint_t)c, wctype)) 6216 goto char_set_matched; 6217 # else 6218 if (iswctype((wint_t)c, wctype)) 6219 goto char_set_matched; 6220 # endif 6221 } 6222 6223 /* match with collating_symbol? */ 6224 # ifdef _LIBC 6225 if (nrules != 0) 6226 { 6227 const unsigned char *extra = (const unsigned char *) 6228 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); 6229 6230 for (workp2 = workp + coll_symbol_length ; workp < workp2 ; 6231 workp++) 6232 { 6233 int32_t *wextra; 6234 wextra = (int32_t*)(extra + *workp++); 6235 for (i = 0; i < *wextra; ++i) 6236 if (TRANSLATE(d[i]) != wextra[1 + i]) 6237 break; 6238 6239 if (i == *wextra) 6240 { 6241 /* Update d, however d will be incremented at 6242 char_set_matched:, we decrement d here. */ 6243 d += i - 1; 6244 goto char_set_matched; 6245 } 6246 } 6247 } 6248 else /* (nrules == 0) */ 6249 # endif 6250 /* If we can't look up collation data, we use wcscoll 6251 instead. */ 6252 { 6253 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;) 6254 { 6255 const CHAR_T *backup_d = d, *backup_dend = dend; 6256 # ifdef _LIBC 6257 length = __wcslen (workp); 6258 # else 6259 length = wcslen (workp); 6260 # endif 6261 6262 /* If wcscoll(the collating symbol, whole string) > 0, 6263 any substring of the string never match with the 6264 collating symbol. */ 6265 # ifdef _LIBC 6266 if (__wcscoll (workp, d) > 0) 6267 # else 6268 if (wcscoll (workp, d) > 0) 6269 # endif 6270 { 6271 workp += length + 1; 6272 continue; 6273 } 6274 6275 /* First, we compare the collating symbol with 6276 the first character of the string. 6277 If it don't match, we add the next character to 6278 the compare buffer in turn. */ 6279 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++) 6280 { 6281 int match; 6282 if (d == dend) 6283 { 6284 if (dend == end_match_2) 6285 break; 6286 d = string2; 6287 dend = end_match_2; 6288 } 6289 6290 /* add next character to the compare buffer. */ 6291 str_buf[i] = TRANSLATE(*d); 6292 str_buf[i+1] = '\0'; 6293 6294 # ifdef _LIBC 6295 match = __wcscoll (workp, str_buf); 6296 # else 6297 match = wcscoll (workp, str_buf); 6298 # endif 6299 if (match == 0) 6300 goto char_set_matched; 6301 6302 if (match < 0) 6303 /* (str_buf > workp) indicate (str_buf + X > workp), 6304 because for all X (str_buf + X > str_buf). 6305 So we don't need continue this loop. */ 6306 break; 6307 6308 /* Otherwise(str_buf < workp), 6309 (str_buf+next_character) may equals (workp). 6310 So we continue this loop. */ 6311 } 6312 /* not matched */ 6313 d = backup_d; 6314 dend = backup_dend; 6315 workp += length + 1; 6316 } 6317 } 6318 /* match with equivalence_class? */ 6319 # ifdef _LIBC 6320 if (nrules != 0) 6321 { 6322 const CHAR_T *backup_d = d, *backup_dend = dend; 6323 /* Try to match the equivalence class against 6324 those known to the collate implementation. */ 6325 const int32_t *table; 6326 const int32_t *weights; 6327 const int32_t *extra; 6328 const int32_t *indirect; 6329 int32_t idx, idx2; 6330 wint_t *cp; 6331 size_t len; 6332 6333 /* This #include defines a local function! */ 6334 # include <locale/weightwc.h> 6335 6336 table = (const int32_t *) 6337 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC); 6338 weights = (const wint_t *) 6339 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC); 6340 extra = (const wint_t *) 6341 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC); 6342 indirect = (const int32_t *) 6343 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC); 6344 6345 /* Write 1 collating element to str_buf, and 6346 get its index. */ 6347 idx2 = 0; 6348 6349 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++) 6350 { 6351 cp = (wint_t*)str_buf; 6352 if (d == dend) 6353 { 6354 if (dend == end_match_2) 6355 break; 6356 d = string2; 6357 dend = end_match_2; 6358 } 6359 str_buf[i] = TRANSLATE(*(d+i)); 6360 str_buf[i+1] = '\0'; /* sentinel */ 6361 idx2 = findidx ((const wint_t**)&cp); 6362 } 6363 6364 /* Update d, however d will be incremented at 6365 char_set_matched:, we decrement d here. */ 6366 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1); 6367 if (d >= dend) 6368 { 6369 if (dend == end_match_2) 6370 d = dend; 6371 else 6372 { 6373 d = string2; 6374 dend = end_match_2; 6375 } 6376 } 6377 6378 len = weights[idx2]; 6379 6380 for (workp2 = workp + equiv_class_length ; workp < workp2 ; 6381 workp++) 6382 { 6383 idx = (int32_t)*workp; 6384 /* We already checked idx != 0 in regex_compile. */ 6385 6386 if (idx2 != 0 && len == weights[idx]) 6387 { 6388 int cnt = 0; 6389 while (cnt < len && (weights[idx + 1 + cnt] 6390 == weights[idx2 + 1 + cnt])) 6391 ++cnt; 6392 6393 if (cnt == len) 6394 goto char_set_matched; 6395 } 6396 } 6397 /* not matched */ 6398 d = backup_d; 6399 dend = backup_dend; 6400 } 6401 else /* (nrules == 0) */ 6402 # endif 6403 /* If we can't look up collation data, we use wcscoll 6404 instead. */ 6405 { 6406 for (workp2 = workp + equiv_class_length ; workp < workp2 ;) 6407 { 6408 const CHAR_T *backup_d = d, *backup_dend = dend; 6409 # ifdef _LIBC 6410 length = __wcslen (workp); 6411 # else 6412 length = wcslen (workp); 6413 # endif 6414 6415 /* If wcscoll(the collating symbol, whole string) > 0, 6416 any substring of the string never match with the 6417 collating symbol. */ 6418 # ifdef _LIBC 6419 if (__wcscoll (workp, d) > 0) 6420 # else 6421 if (wcscoll (workp, d) > 0) 6422 # endif 6423 { 6424 workp += length + 1; 6425 break; 6426 } 6427 6428 /* First, we compare the equivalence class with 6429 the first character of the string. 6430 If it don't match, we add the next character to 6431 the compare buffer in turn. */ 6432 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++) 6433 { 6434 int match; 6435 if (d == dend) 6436 { 6437 if (dend == end_match_2) 6438 break; 6439 d = string2; 6440 dend = end_match_2; 6441 } 6442 6443 /* add next character to the compare buffer. */ 6444 str_buf[i] = TRANSLATE(*d); 6445 str_buf[i+1] = '\0'; 6446 6447 # ifdef _LIBC 6448 match = __wcscoll (workp, str_buf); 6449 # else 6450 match = wcscoll (workp, str_buf); 6451 # endif 6452 6453 if (match == 0) 6454 goto char_set_matched; 6455 6456 if (match < 0) 6457 /* (str_buf > workp) indicate (str_buf + X > workp), 6458 because for all X (str_buf + X > str_buf). 6459 So we don't need continue this loop. */ 6460 break; 6461 6462 /* Otherwise(str_buf < workp), 6463 (str_buf+next_character) may equals (workp). 6464 So we continue this loop. */ 6465 } 6466 /* not matched */ 6467 d = backup_d; 6468 dend = backup_dend; 6469 workp += length + 1; 6470 } 6471 } 6472 6473 /* match with char_range? */ 6474 # ifdef _LIBC 6475 if (nrules != 0) 6476 { 6477 uint32_t collseqval; 6478 const char *collseq = (const char *) 6479 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC); 6480 6481 collseqval = collseq_table_lookup (collseq, c); 6482 6483 for (; workp < p - chars_length ;) 6484 { 6485 uint32_t start_val, end_val; 6486 6487 /* We already compute the collation sequence value 6488 of the characters (or collating symbols). */ 6489 start_val = (uint32_t) *workp++; /* range_start */ 6490 end_val = (uint32_t) *workp++; /* range_end */ 6491 6492 if (start_val <= collseqval && collseqval <= end_val) 6493 goto char_set_matched; 6494 } 6495 } 6496 else 6497 # endif 6498 { 6499 /* We set range_start_char at str_buf[0], range_end_char 6500 at str_buf[4], and compared char at str_buf[2]. */ 6501 str_buf[1] = 0; 6502 str_buf[2] = c; 6503 str_buf[3] = 0; 6504 str_buf[5] = 0; 6505 for (; workp < p - chars_length ;) 6506 { 6507 wchar_t *range_start_char, *range_end_char; 6508 6509 /* match if (range_start_char <= c <= range_end_char). */ 6510 6511 /* If range_start(or end) < 0, we assume -range_start(end) 6512 is the offset of the collating symbol which is specified 6513 as the character of the range start(end). */ 6514 6515 /* range_start */ 6516 if (*workp < 0) 6517 range_start_char = charset_top - (*workp++); 6518 else 6519 { 6520 str_buf[0] = *workp++; 6521 range_start_char = str_buf; 6522 } 6523 6524 /* range_end */ 6525 if (*workp < 0) 6526 range_end_char = charset_top - (*workp++); 6527 else 6528 { 6529 str_buf[4] = *workp++; 6530 range_end_char = str_buf + 4; 6531 } 6532 6533 # ifdef _LIBC 6534 if (__wcscoll (range_start_char, str_buf+2) <= 0 6535 && __wcscoll (str_buf+2, range_end_char) <= 0) 6536 # else 6537 if (wcscoll (range_start_char, str_buf+2) <= 0 6538 && wcscoll (str_buf+2, range_end_char) <= 0) 6539 # endif 6540 goto char_set_matched; 6541 } 6542 } 6543 6544 /* match with char? */ 6545 for (; workp < p ; workp++) 6546 if (c == *workp) 6547 goto char_set_matched; 6548 6549 negate = !negate; 6550 6551 char_set_matched: 6552 if (negate) goto fail; 6553 #else 6554 /* Cast to `unsigned' instead of `unsigned char' in case the 6555 bit list is a full 32 bytes long. */ 6556 if (c < (unsigned) (*p * BYTEWIDTH) 6557 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) 6558 negate = !negate; 6559 6560 p += 1 + *p; 6561 6562 if (!negate) goto fail; 6563 #undef WORK_BUFFER_SIZE 6564 #endif /* WCHAR */ 6565 SET_REGS_MATCHED (); 6566 d++; 6567 break; 6568 } 6569 6570 6571 /* The beginning of a group is represented by start_memory. 6572 The arguments are the register number in the next byte, and the 6573 number of groups inner to this one in the next. The text 6574 matched within the group is recorded (in the internal 6575 registers data structure) under the register number. */ 6576 case start_memory: 6577 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n", 6578 (long int) *p, (long int) p[1]); 6579 6580 /* Find out if this group can match the empty string. */ 6581 p1 = p; /* To send to group_match_null_string_p. */ 6582 6583 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) 6584 REG_MATCH_NULL_STRING_P (reg_info[*p]) 6585 = PREFIX(group_match_null_string_p) (&p1, pend, reg_info); 6586 6587 /* Save the position in the string where we were the last time 6588 we were at this open-group operator in case the group is 6589 operated upon by a repetition operator, e.g., with `(a*)*b' 6590 against `ab'; then we want to ignore where we are now in 6591 the string in case this attempt to match fails. */ 6592 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) 6593 ? REG_UNSET (regstart[*p]) ? d : regstart[*p] 6594 : regstart[*p]; 6595 DEBUG_PRINT2 (" old_regstart: %d\n", 6596 POINTER_TO_OFFSET (old_regstart[*p])); 6597 6598 regstart[*p] = d; 6599 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); 6600 6601 IS_ACTIVE (reg_info[*p]) = 1; 6602 MATCHED_SOMETHING (reg_info[*p]) = 0; 6603 6604 /* Clear this whenever we change the register activity status. */ 6605 set_regs_matched_done = 0; 6606 6607 /* This is the new highest active register. */ 6608 highest_active_reg = *p; 6609 6610 /* If nothing was active before, this is the new lowest active 6611 register. */ 6612 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) 6613 lowest_active_reg = *p; 6614 6615 /* Move past the register number and inner group count. */ 6616 p += 2; 6617 just_past_start_mem = p; 6618 6619 break; 6620 6621 6622 /* The stop_memory opcode represents the end of a group. Its 6623 arguments are the same as start_memory's: the register 6624 number, and the number of inner groups. */ 6625 case stop_memory: 6626 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n", 6627 (long int) *p, (long int) p[1]); 6628 6629 /* We need to save the string position the last time we were at 6630 this close-group operator in case the group is operated 6631 upon by a repetition operator, e.g., with `((a*)*(b*)*)*' 6632 against `aba'; then we want to ignore where we are now in 6633 the string in case this attempt to match fails. */ 6634 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) 6635 ? REG_UNSET (regend[*p]) ? d : regend[*p] 6636 : regend[*p]; 6637 DEBUG_PRINT2 (" old_regend: %d\n", 6638 POINTER_TO_OFFSET (old_regend[*p])); 6639 6640 regend[*p] = d; 6641 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); 6642 6643 /* This register isn't active anymore. */ 6644 IS_ACTIVE (reg_info[*p]) = 0; 6645 6646 /* Clear this whenever we change the register activity status. */ 6647 set_regs_matched_done = 0; 6648 6649 /* If this was the only register active, nothing is active 6650 anymore. */ 6651 if (lowest_active_reg == highest_active_reg) 6652 { 6653 lowest_active_reg = NO_LOWEST_ACTIVE_REG; 6654 highest_active_reg = NO_HIGHEST_ACTIVE_REG; 6655 } 6656 else 6657 { /* We must scan for the new highest active register, since 6658 it isn't necessarily one less than now: consider 6659 (a(b)c(d(e)f)g). When group 3 ends, after the f), the 6660 new highest active register is 1. */ 6661 UCHAR_T r = *p - 1; 6662 while (r > 0 && !IS_ACTIVE (reg_info[r])) 6663 r--; 6664 6665 /* If we end up at register zero, that means that we saved 6666 the registers as the result of an `on_failure_jump', not 6667 a `start_memory', and we jumped to past the innermost 6668 `stop_memory'. For example, in ((.)*) we save 6669 registers 1 and 2 as a result of the *, but when we pop 6670 back to the second ), we are at the stop_memory 1. 6671 Thus, nothing is active. */ 6672 if (r == 0) 6673 { 6674 lowest_active_reg = NO_LOWEST_ACTIVE_REG; 6675 highest_active_reg = NO_HIGHEST_ACTIVE_REG; 6676 } 6677 else 6678 highest_active_reg = r; 6679 } 6680 6681 /* If just failed to match something this time around with a 6682 group that's operated on by a repetition operator, try to 6683 force exit from the ``loop'', and restore the register 6684 information for this group that we had before trying this 6685 last match. */ 6686 if ((!MATCHED_SOMETHING (reg_info[*p]) 6687 || just_past_start_mem == p - 1) 6688 && (p + 2) < pend) 6689 { 6690 boolean is_a_jump_n = false; 6691 6692 p1 = p + 2; 6693 mcnt = 0; 6694 switch ((re_opcode_t) *p1++) 6695 { 6696 case jump_n: 6697 is_a_jump_n = true; 6698 case pop_failure_jump: 6699 case maybe_pop_jump: 6700 case jump: 6701 case dummy_failure_jump: 6702 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 6703 if (is_a_jump_n) 6704 p1 += OFFSET_ADDRESS_SIZE; 6705 break; 6706 6707 default: 6708 /* do nothing */ ; 6709 } 6710 p1 += mcnt; 6711 6712 /* If the next operation is a jump backwards in the pattern 6713 to an on_failure_jump right before the start_memory 6714 corresponding to this stop_memory, exit from the loop 6715 by forcing a failure after pushing on the stack the 6716 on_failure_jump's jump in the pattern, and d. */ 6717 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump 6718 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory 6719 && p1[2+OFFSET_ADDRESS_SIZE] == *p) 6720 { 6721 /* If this group ever matched anything, then restore 6722 what its registers were before trying this last 6723 failed match, e.g., with `(a*)*b' against `ab' for 6724 regstart[1], and, e.g., with `((a*)*(b*)*)*' 6725 against `aba' for regend[3]. 6726 6727 Also restore the registers for inner groups for, 6728 e.g., `((a*)(b*))*' against `aba' (register 3 would 6729 otherwise get trashed). */ 6730 6731 if (EVER_MATCHED_SOMETHING (reg_info[*p])) 6732 { 6733 unsigned r; 6734 6735 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; 6736 6737 /* Restore this and inner groups' (if any) registers. */ 6738 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1); 6739 r++) 6740 { 6741 regstart[r] = old_regstart[r]; 6742 6743 /* xx why this test? */ 6744 if (old_regend[r] >= regstart[r]) 6745 regend[r] = old_regend[r]; 6746 } 6747 } 6748 p1++; 6749 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 6750 PUSH_FAILURE_POINT (p1 + mcnt, d, -2); 6751 6752 goto fail; 6753 } 6754 } 6755 6756 /* Move past the register number and the inner group count. */ 6757 p += 2; 6758 break; 6759 6760 6761 /* \<digit> has been turned into a `duplicate' command which is 6762 followed by the numeric value of <digit> as the register number. */ 6763 case duplicate: 6764 { 6765 register const CHAR_T *d2, *dend2; 6766 int regno = *p++; /* Get which register to match against. */ 6767 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); 6768 6769 /* Can't back reference a group which we've never matched. */ 6770 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) 6771 goto fail; 6772 6773 /* Where in input to try to start matching. */ 6774 d2 = regstart[regno]; 6775 6776 /* Where to stop matching; if both the place to start and 6777 the place to stop matching are in the same string, then 6778 set to the place to stop, otherwise, for now have to use 6779 the end of the first string. */ 6780 6781 dend2 = ((FIRST_STRING_P (regstart[regno]) 6782 == FIRST_STRING_P (regend[regno])) 6783 ? regend[regno] : end_match_1); 6784 for (;;) 6785 { 6786 /* If necessary, advance to next segment in register 6787 contents. */ 6788 while (d2 == dend2) 6789 { 6790 if (dend2 == end_match_2) break; 6791 if (dend2 == regend[regno]) break; 6792 6793 /* End of string1 => advance to string2. */ 6794 d2 = string2; 6795 dend2 = regend[regno]; 6796 } 6797 /* At end of register contents => success */ 6798 if (d2 == dend2) break; 6799 6800 /* If necessary, advance to next segment in data. */ 6801 PREFETCH (); 6802 6803 /* How many characters left in this segment to match. */ 6804 mcnt = dend - d; 6805 6806 /* Want how many consecutive characters we can match in 6807 one shot, so, if necessary, adjust the count. */ 6808 if (mcnt > dend2 - d2) 6809 mcnt = dend2 - d2; 6810 6811 /* Compare that many; failure if mismatch, else move 6812 past them. */ 6813 if (translate 6814 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate) 6815 : memcmp (d, d2, mcnt*sizeof(UCHAR_T))) 6816 goto fail; 6817 d += mcnt, d2 += mcnt; 6818 6819 /* Do this because we've match some characters. */ 6820 SET_REGS_MATCHED (); 6821 } 6822 } 6823 break; 6824 6825 6826 /* begline matches the empty string at the beginning of the string 6827 (unless `not_bol' is set in `bufp'), and, if 6828 `newline_anchor' is set, after newlines. */ 6829 case begline: 6830 DEBUG_PRINT1 ("EXECUTING begline.\n"); 6831 6832 if (AT_STRINGS_BEG (d)) 6833 { 6834 if (!bufp->not_bol) break; 6835 } 6836 else if (d[-1] == '\n' && bufp->newline_anchor) 6837 { 6838 break; 6839 } 6840 /* In all other cases, we fail. */ 6841 goto fail; 6842 6843 6844 /* endline is the dual of begline. */ 6845 case endline: 6846 DEBUG_PRINT1 ("EXECUTING endline.\n"); 6847 6848 if (AT_STRINGS_END (d)) 6849 { 6850 if (!bufp->not_eol) break; 6851 } 6852 6853 /* We have to ``prefetch'' the next character. */ 6854 else if ((d == end1 ? *string2 : *d) == '\n' 6855 && bufp->newline_anchor) 6856 { 6857 break; 6858 } 6859 goto fail; 6860 6861 6862 /* Match at the very beginning of the data. */ 6863 case begbuf: 6864 DEBUG_PRINT1 ("EXECUTING begbuf.\n"); 6865 if (AT_STRINGS_BEG (d)) 6866 break; 6867 goto fail; 6868 6869 6870 /* Match at the very end of the data. */ 6871 case endbuf: 6872 DEBUG_PRINT1 ("EXECUTING endbuf.\n"); 6873 if (AT_STRINGS_END (d)) 6874 break; 6875 goto fail; 6876 6877 6878 /* on_failure_keep_string_jump is used to optimize `.*\n'. It 6879 pushes NULL as the value for the string on the stack. Then 6880 `pop_failure_point' will keep the current value for the 6881 string, instead of restoring it. To see why, consider 6882 matching `foo\nbar' against `.*\n'. The .* matches the foo; 6883 then the . fails against the \n. But the next thing we want 6884 to do is match the \n against the \n; if we restored the 6885 string value, we would be back at the foo. 6886 6887 Because this is used only in specific cases, we don't need to 6888 check all the things that `on_failure_jump' does, to make 6889 sure the right things get saved on the stack. Hence we don't 6890 share its code. The only reason to push anything on the 6891 stack at all is that otherwise we would have to change 6892 `anychar's code to do something besides goto fail in this 6893 case; that seems worse than this. */ 6894 case on_failure_keep_string_jump: 6895 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); 6896 6897 EXTRACT_NUMBER_AND_INCR (mcnt, p); 6898 #ifdef _LIBC 6899 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt); 6900 #else 6901 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); 6902 #endif 6903 6904 PUSH_FAILURE_POINT (p + mcnt, NULL, -2); 6905 break; 6906 6907 6908 /* Uses of on_failure_jump: 6909 6910 Each alternative starts with an on_failure_jump that points 6911 to the beginning of the next alternative. Each alternative 6912 except the last ends with a jump that in effect jumps past 6913 the rest of the alternatives. (They really jump to the 6914 ending jump of the following alternative, because tensioning 6915 these jumps is a hassle.) 6916 6917 Repeats start with an on_failure_jump that points past both 6918 the repetition text and either the following jump or 6919 pop_failure_jump back to this on_failure_jump. */ 6920 case on_failure_jump: 6921 on_failure: 6922 DEBUG_PRINT1 ("EXECUTING on_failure_jump"); 6923 6924 EXTRACT_NUMBER_AND_INCR (mcnt, p); 6925 #ifdef _LIBC 6926 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt); 6927 #else 6928 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); 6929 #endif 6930 6931 /* If this on_failure_jump comes right before a group (i.e., 6932 the original * applied to a group), save the information 6933 for that group and all inner ones, so that if we fail back 6934 to this point, the group's information will be correct. 6935 For example, in \(a*\)*\1, we need the preceding group, 6936 and in \(zz\(a*\)b*\)\2, we need the inner group. */ 6937 6938 /* We can't use `p' to check ahead because we push 6939 a failure point to `p + mcnt' after we do this. */ 6940 p1 = p; 6941 6942 /* We need to skip no_op's before we look for the 6943 start_memory in case this on_failure_jump is happening as 6944 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 6945 against aba. */ 6946 while (p1 < pend && (re_opcode_t) *p1 == no_op) 6947 p1++; 6948 6949 if (p1 < pend && (re_opcode_t) *p1 == start_memory) 6950 { 6951 /* We have a new highest active register now. This will 6952 get reset at the start_memory we are about to get to, 6953 but we will have saved all the registers relevant to 6954 this repetition op, as described above. */ 6955 highest_active_reg = *(p1 + 1) + *(p1 + 2); 6956 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) 6957 lowest_active_reg = *(p1 + 1); 6958 } 6959 6960 DEBUG_PRINT1 (":\n"); 6961 PUSH_FAILURE_POINT (p + mcnt, d, -2); 6962 break; 6963 6964 6965 /* A smart repeat ends with `maybe_pop_jump'. 6966 We change it to either `pop_failure_jump' or `jump'. */ 6967 case maybe_pop_jump: 6968 EXTRACT_NUMBER_AND_INCR (mcnt, p); 6969 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); 6970 { 6971 register UCHAR_T *p2 = p; 6972 6973 /* Compare the beginning of the repeat with what in the 6974 pattern follows its end. If we can establish that there 6975 is nothing that they would both match, i.e., that we 6976 would have to backtrack because of (as in, e.g., `a*a') 6977 then we can change to pop_failure_jump, because we'll 6978 never have to backtrack. 6979 6980 This is not true in the case of alternatives: in 6981 `(a|ab)*' we do need to backtrack to the `ab' alternative 6982 (e.g., if the string was `ab'). But instead of trying to 6983 detect that here, the alternative has put on a dummy 6984 failure point which is what we will end up popping. */ 6985 6986 /* Skip over open/close-group commands. 6987 If what follows this loop is a ...+ construct, 6988 look at what begins its body, since we will have to 6989 match at least one of that. */ 6990 while (1) 6991 { 6992 if (p2 + 2 < pend 6993 && ((re_opcode_t) *p2 == stop_memory 6994 || (re_opcode_t) *p2 == start_memory)) 6995 p2 += 3; 6996 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend 6997 && (re_opcode_t) *p2 == dummy_failure_jump) 6998 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE; 6999 else 7000 break; 7001 } 7002 7003 p1 = p + mcnt; 7004 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding 7005 to the `maybe_finalize_jump' of this case. Examine what 7006 follows. */ 7007 7008 /* If we're at the end of the pattern, we can change. */ 7009 if (p2 == pend) 7010 { 7011 /* Consider what happens when matching ":\(.*\)" 7012 against ":/". I don't really understand this code 7013 yet. */ 7014 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) 7015 pop_failure_jump; 7016 DEBUG_PRINT1 7017 (" End of pattern: change to `pop_failure_jump'.\n"); 7018 } 7019 7020 else if ((re_opcode_t) *p2 == exactn 7021 #ifdef MBS_SUPPORT 7022 || (re_opcode_t) *p2 == exactn_bin 7023 #endif 7024 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) 7025 { 7026 register UCHAR_T c 7027 = *p2 == (UCHAR_T) endline ? '\n' : p2[2]; 7028 7029 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn 7030 #ifdef MBS_SUPPORT 7031 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin 7032 #endif 7033 ) && p1[3+OFFSET_ADDRESS_SIZE] != c) 7034 { 7035 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) 7036 pop_failure_jump; 7037 #ifdef WCHAR 7038 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n", 7039 (wint_t) c, 7040 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]); 7041 #else 7042 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", 7043 (char) c, 7044 (char) p1[3+OFFSET_ADDRESS_SIZE]); 7045 #endif 7046 } 7047 7048 #ifndef WCHAR 7049 else if ((re_opcode_t) p1[3] == charset 7050 || (re_opcode_t) p1[3] == charset_not) 7051 { 7052 int negate = (re_opcode_t) p1[3] == charset_not; 7053 7054 if (c < (unsigned) (p1[4] * BYTEWIDTH) 7055 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) 7056 negate = !negate; 7057 7058 /* `negate' is equal to 1 if c would match, which means 7059 that we can't change to pop_failure_jump. */ 7060 if (!negate) 7061 { 7062 p[-3] = (unsigned char) pop_failure_jump; 7063 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7064 } 7065 } 7066 #endif /* not WCHAR */ 7067 } 7068 #ifndef WCHAR 7069 else if ((re_opcode_t) *p2 == charset) 7070 { 7071 /* We win if the first character of the loop is not part 7072 of the charset. */ 7073 if ((re_opcode_t) p1[3] == exactn 7074 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5] 7075 && (p2[2 + p1[5] / BYTEWIDTH] 7076 & (1 << (p1[5] % BYTEWIDTH))))) 7077 { 7078 p[-3] = (unsigned char) pop_failure_jump; 7079 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7080 } 7081 7082 else if ((re_opcode_t) p1[3] == charset_not) 7083 { 7084 int idx; 7085 /* We win if the charset_not inside the loop 7086 lists every character listed in the charset after. */ 7087 for (idx = 0; idx < (int) p2[1]; idx++) 7088 if (! (p2[2 + idx] == 0 7089 || (idx < (int) p1[4] 7090 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) 7091 break; 7092 7093 if (idx == p2[1]) 7094 { 7095 p[-3] = (unsigned char) pop_failure_jump; 7096 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7097 } 7098 } 7099 else if ((re_opcode_t) p1[3] == charset) 7100 { 7101 int idx; 7102 /* We win if the charset inside the loop 7103 has no overlap with the one after the loop. */ 7104 for (idx = 0; 7105 idx < (int) p2[1] && idx < (int) p1[4]; 7106 idx++) 7107 if ((p2[2 + idx] & p1[5 + idx]) != 0) 7108 break; 7109 7110 if (idx == p2[1] || idx == p1[4]) 7111 { 7112 p[-3] = (unsigned char) pop_failure_jump; 7113 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7114 } 7115 } 7116 } 7117 #endif /* not WCHAR */ 7118 } 7119 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */ 7120 if ((re_opcode_t) p[-1] != pop_failure_jump) 7121 { 7122 p[-1] = (UCHAR_T) jump; 7123 DEBUG_PRINT1 (" Match => jump.\n"); 7124 goto unconditional_jump; 7125 } 7126 /* Note fall through. */ 7127 7128 7129 /* The end of a simple repeat has a pop_failure_jump back to 7130 its matching on_failure_jump, where the latter will push a 7131 failure point. The pop_failure_jump takes off failure 7132 points put on by this pop_failure_jump's matching 7133 on_failure_jump; we got through the pattern to here from the 7134 matching on_failure_jump, so didn't fail. */ 7135 case pop_failure_jump: 7136 { 7137 /* We need to pass separate storage for the lowest and 7138 highest registers, even though we don't care about the 7139 actual values. Otherwise, we will restore only one 7140 register from the stack, since lowest will == highest in 7141 `pop_failure_point'. */ 7142 active_reg_t dummy_low_reg, dummy_high_reg; 7143 UCHAR_T *pdummy ATTRIBUTE_UNUSED = NULL; 7144 const CHAR_T *sdummy ATTRIBUTE_UNUSED = NULL; 7145 7146 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); 7147 POP_FAILURE_POINT (sdummy, pdummy, 7148 dummy_low_reg, dummy_high_reg, 7149 reg_dummy, reg_dummy, reg_info_dummy); 7150 } 7151 /* Note fall through. */ 7152 7153 unconditional_jump: 7154 #ifdef _LIBC 7155 DEBUG_PRINT2 ("\n%p: ", p); 7156 #else 7157 DEBUG_PRINT2 ("\n0x%x: ", p); 7158 #endif 7159 /* Note fall through. */ 7160 7161 /* Unconditionally jump (without popping any failure points). */ 7162 case jump: 7163 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ 7164 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); 7165 p += mcnt; /* Do the jump. */ 7166 #ifdef _LIBC 7167 DEBUG_PRINT2 ("(to %p).\n", p); 7168 #else 7169 DEBUG_PRINT2 ("(to 0x%x).\n", p); 7170 #endif 7171 break; 7172 7173 7174 /* We need this opcode so we can detect where alternatives end 7175 in `group_match_null_string_p' et al. */ 7176 case jump_past_alt: 7177 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); 7178 goto unconditional_jump; 7179 7180 7181 /* Normally, the on_failure_jump pushes a failure point, which 7182 then gets popped at pop_failure_jump. We will end up at 7183 pop_failure_jump, also, and with a pattern of, say, `a+', we 7184 are skipping over the on_failure_jump, so we have to push 7185 something meaningless for pop_failure_jump to pop. */ 7186 case dummy_failure_jump: 7187 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); 7188 /* It doesn't matter what we push for the string here. What 7189 the code at `fail' tests is the value for the pattern. */ 7190 PUSH_FAILURE_POINT (NULL, NULL, -2); 7191 goto unconditional_jump; 7192 7193 7194 /* At the end of an alternative, we need to push a dummy failure 7195 point in case we are followed by a `pop_failure_jump', because 7196 we don't want the failure point for the alternative to be 7197 popped. For example, matching `(a|ab)*' against `aab' 7198 requires that we match the `ab' alternative. */ 7199 case push_dummy_failure: 7200 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); 7201 /* See comments just above at `dummy_failure_jump' about the 7202 two zeroes. */ 7203 PUSH_FAILURE_POINT (NULL, NULL, -2); 7204 break; 7205 7206 /* Have to succeed matching what follows at least n times. 7207 After that, handle like `on_failure_jump'. */ 7208 case succeed_n: 7209 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); 7210 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); 7211 7212 assert (mcnt >= 0); 7213 /* Originally, this is how many times we HAVE to succeed. */ 7214 if (mcnt > 0) 7215 { 7216 mcnt--; 7217 p += OFFSET_ADDRESS_SIZE; 7218 STORE_NUMBER_AND_INCR (p, mcnt); 7219 #ifdef _LIBC 7220 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE 7221 , mcnt); 7222 #else 7223 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE 7224 , mcnt); 7225 #endif 7226 } 7227 else if (mcnt == 0) 7228 { 7229 #ifdef _LIBC 7230 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", 7231 p + OFFSET_ADDRESS_SIZE); 7232 #else 7233 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", 7234 p + OFFSET_ADDRESS_SIZE); 7235 #endif /* _LIBC */ 7236 7237 #ifdef WCHAR 7238 p[1] = (UCHAR_T) no_op; 7239 #else 7240 p[2] = (UCHAR_T) no_op; 7241 p[3] = (UCHAR_T) no_op; 7242 #endif /* WCHAR */ 7243 goto on_failure; 7244 } 7245 break; 7246 7247 case jump_n: 7248 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); 7249 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); 7250 7251 /* Originally, this is how many times we CAN jump. */ 7252 if (mcnt) 7253 { 7254 mcnt--; 7255 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt); 7256 7257 #ifdef _LIBC 7258 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE, 7259 mcnt); 7260 #else 7261 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE, 7262 mcnt); 7263 #endif /* _LIBC */ 7264 goto unconditional_jump; 7265 } 7266 /* If don't have to jump any more, skip over the rest of command. */ 7267 else 7268 p += 2 * OFFSET_ADDRESS_SIZE; 7269 break; 7270 7271 case set_number_at: 7272 { 7273 DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); 7274 7275 EXTRACT_NUMBER_AND_INCR (mcnt, p); 7276 p1 = p + mcnt; 7277 EXTRACT_NUMBER_AND_INCR (mcnt, p); 7278 #ifdef _LIBC 7279 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt); 7280 #else 7281 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); 7282 #endif 7283 STORE_NUMBER (p1, mcnt); 7284 break; 7285 } 7286 7287 #if 0 7288 /* The DEC Alpha C compiler 3.x generates incorrect code for the 7289 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of 7290 AT_WORD_BOUNDARY, so this code is disabled. Expanding the 7291 macro and introducing temporary variables works around the bug. */ 7292 7293 case wordbound: 7294 DEBUG_PRINT1 ("EXECUTING wordbound.\n"); 7295 if (AT_WORD_BOUNDARY (d)) 7296 break; 7297 goto fail; 7298 7299 case notwordbound: 7300 DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); 7301 if (AT_WORD_BOUNDARY (d)) 7302 goto fail; 7303 break; 7304 #else 7305 case wordbound: 7306 { 7307 boolean prevchar, thischar; 7308 7309 DEBUG_PRINT1 ("EXECUTING wordbound.\n"); 7310 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) 7311 break; 7312 7313 prevchar = WORDCHAR_P (d - 1); 7314 thischar = WORDCHAR_P (d); 7315 if (prevchar != thischar) 7316 break; 7317 goto fail; 7318 } 7319 7320 case notwordbound: 7321 { 7322 boolean prevchar, thischar; 7323 7324 DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); 7325 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) 7326 goto fail; 7327 7328 prevchar = WORDCHAR_P (d - 1); 7329 thischar = WORDCHAR_P (d); 7330 if (prevchar != thischar) 7331 goto fail; 7332 break; 7333 } 7334 #endif 7335 7336 case wordbeg: 7337 DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); 7338 if (!AT_STRINGS_END (d) && WORDCHAR_P (d) 7339 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) 7340 break; 7341 goto fail; 7342 7343 case wordend: 7344 DEBUG_PRINT1 ("EXECUTING wordend.\n"); 7345 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) 7346 && (AT_STRINGS_END (d) || !WORDCHAR_P (d))) 7347 break; 7348 goto fail; 7349 7350 #ifdef emacs 7351 case before_dot: 7352 DEBUG_PRINT1 ("EXECUTING before_dot.\n"); 7353 if (PTR_CHAR_POS ((unsigned char *) d) >= point) 7354 goto fail; 7355 break; 7356 7357 case at_dot: 7358 DEBUG_PRINT1 ("EXECUTING at_dot.\n"); 7359 if (PTR_CHAR_POS ((unsigned char *) d) != point) 7360 goto fail; 7361 break; 7362 7363 case after_dot: 7364 DEBUG_PRINT1 ("EXECUTING after_dot.\n"); 7365 if (PTR_CHAR_POS ((unsigned char *) d) <= point) 7366 goto fail; 7367 break; 7368 7369 case syntaxspec: 7370 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); 7371 mcnt = *p++; 7372 goto matchsyntax; 7373 7374 case wordchar: 7375 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); 7376 mcnt = (int) Sword; 7377 matchsyntax: 7378 PREFETCH (); 7379 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ 7380 d++; 7381 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt) 7382 goto fail; 7383 SET_REGS_MATCHED (); 7384 break; 7385 7386 case notsyntaxspec: 7387 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); 7388 mcnt = *p++; 7389 goto matchnotsyntax; 7390 7391 case notwordchar: 7392 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); 7393 mcnt = (int) Sword; 7394 matchnotsyntax: 7395 PREFETCH (); 7396 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ 7397 d++; 7398 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt) 7399 goto fail; 7400 SET_REGS_MATCHED (); 7401 break; 7402 7403 #else /* not emacs */ 7404 case wordchar: 7405 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); 7406 PREFETCH (); 7407 if (!WORDCHAR_P (d)) 7408 goto fail; 7409 SET_REGS_MATCHED (); 7410 d++; 7411 break; 7412 7413 case notwordchar: 7414 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); 7415 PREFETCH (); 7416 if (WORDCHAR_P (d)) 7417 goto fail; 7418 SET_REGS_MATCHED (); 7419 d++; 7420 break; 7421 #endif /* not emacs */ 7422 7423 default: 7424 abort (); 7425 } 7426 continue; /* Successfully executed one pattern command; keep going. */ 7427 7428 7429 /* We goto here if a matching operation fails. */ 7430 fail: 7431 if (!FAIL_STACK_EMPTY ()) 7432 { /* A restart point is known. Restore to that state. */ 7433 DEBUG_PRINT1 ("\nFAIL:\n"); 7434 POP_FAILURE_POINT (d, p, 7435 lowest_active_reg, highest_active_reg, 7436 regstart, regend, reg_info); 7437 7438 /* If this failure point is a dummy, try the next one. */ 7439 if (!p) 7440 goto fail; 7441 7442 /* If we failed to the end of the pattern, don't examine *p. */ 7443 assert (p <= pend); 7444 if (p < pend) 7445 { 7446 boolean is_a_jump_n = false; 7447 7448 /* If failed to a backwards jump that's part of a repetition 7449 loop, need to pop this failure point and use the next one. */ 7450 switch ((re_opcode_t) *p) 7451 { 7452 case jump_n: 7453 is_a_jump_n = true; 7454 case maybe_pop_jump: 7455 case pop_failure_jump: 7456 case jump: 7457 p1 = p + 1; 7458 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7459 p1 += mcnt; 7460 7461 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) 7462 || (!is_a_jump_n 7463 && (re_opcode_t) *p1 == on_failure_jump)) 7464 goto fail; 7465 break; 7466 default: 7467 /* do nothing */ ; 7468 } 7469 } 7470 7471 if (d >= string1 && d <= end1) 7472 dend = end_match_1; 7473 } 7474 else 7475 break; /* Matching at this starting point really fails. */ 7476 } /* for (;;) */ 7477 7478 if (best_regs_set) 7479 goto restore_best_regs; 7480 7481 FREE_VARIABLES (); 7482 7483 return -1; /* Failure to match. */ 7484 } /* re_match_2 */ 7485 7486 /* Subroutine definitions for re_match_2. */ 7487 7488 7489 /* We are passed P pointing to a register number after a start_memory. 7490 7491 Return true if the pattern up to the corresponding stop_memory can 7492 match the empty string, and false otherwise. 7493 7494 If we find the matching stop_memory, sets P to point to one past its number. 7495 Otherwise, sets P to an undefined byte less than or equal to END. 7496 7497 We don't handle duplicates properly (yet). */ 7498 7499 static boolean 7500 PREFIX(group_match_null_string_p) (UCHAR_T **p, UCHAR_T *end, 7501 PREFIX(register_info_type) *reg_info) 7502 { 7503 int mcnt; 7504 /* Point to after the args to the start_memory. */ 7505 UCHAR_T *p1 = *p + 2; 7506 7507 while (p1 < end) 7508 { 7509 /* Skip over opcodes that can match nothing, and return true or 7510 false, as appropriate, when we get to one that can't, or to the 7511 matching stop_memory. */ 7512 7513 switch ((re_opcode_t) *p1) 7514 { 7515 /* Could be either a loop or a series of alternatives. */ 7516 case on_failure_jump: 7517 p1++; 7518 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7519 7520 /* If the next operation is not a jump backwards in the 7521 pattern. */ 7522 7523 if (mcnt >= 0) 7524 { 7525 /* Go through the on_failure_jumps of the alternatives, 7526 seeing if any of the alternatives cannot match nothing. 7527 The last alternative starts with only a jump, 7528 whereas the rest start with on_failure_jump and end 7529 with a jump, e.g., here is the pattern for `a|b|c': 7530 7531 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 7532 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 7533 /exactn/1/c 7534 7535 So, we have to first go through the first (n-1) 7536 alternatives and then deal with the last one separately. */ 7537 7538 7539 /* Deal with the first (n-1) alternatives, which start 7540 with an on_failure_jump (see above) that jumps to right 7541 past a jump_past_alt. */ 7542 7543 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] == 7544 jump_past_alt) 7545 { 7546 /* `mcnt' holds how many bytes long the alternative 7547 is, including the ending `jump_past_alt' and 7548 its number. */ 7549 7550 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt - 7551 (1 + OFFSET_ADDRESS_SIZE), 7552 reg_info)) 7553 return false; 7554 7555 /* Move to right after this alternative, including the 7556 jump_past_alt. */ 7557 p1 += mcnt; 7558 7559 /* Break if it's the beginning of an n-th alternative 7560 that doesn't begin with an on_failure_jump. */ 7561 if ((re_opcode_t) *p1 != on_failure_jump) 7562 break; 7563 7564 /* Still have to check that it's not an n-th 7565 alternative that starts with an on_failure_jump. */ 7566 p1++; 7567 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7568 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] != 7569 jump_past_alt) 7570 { 7571 /* Get to the beginning of the n-th alternative. */ 7572 p1 -= 1 + OFFSET_ADDRESS_SIZE; 7573 break; 7574 } 7575 } 7576 7577 /* Deal with the last alternative: go back and get number 7578 of the `jump_past_alt' just before it. `mcnt' contains 7579 the length of the alternative. */ 7580 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE); 7581 7582 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info)) 7583 return false; 7584 7585 p1 += mcnt; /* Get past the n-th alternative. */ 7586 } /* if mcnt > 0 */ 7587 break; 7588 7589 7590 case stop_memory: 7591 assert (p1[1] == **p); 7592 *p = p1 + 2; 7593 return true; 7594 7595 7596 default: 7597 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) 7598 return false; 7599 } 7600 } /* while p1 < end */ 7601 7602 return false; 7603 } /* group_match_null_string_p */ 7604 7605 7606 /* Similar to group_match_null_string_p, but doesn't deal with alternatives: 7607 It expects P to be the first byte of a single alternative and END one 7608 byte past the last. The alternative can contain groups. */ 7609 7610 static boolean 7611 PREFIX(alt_match_null_string_p) (UCHAR_T *p, UCHAR_T *end, 7612 PREFIX(register_info_type) *reg_info) 7613 { 7614 int mcnt; 7615 UCHAR_T *p1 = p; 7616 7617 while (p1 < end) 7618 { 7619 /* Skip over opcodes that can match nothing, and break when we get 7620 to one that can't. */ 7621 7622 switch ((re_opcode_t) *p1) 7623 { 7624 /* It's a loop. */ 7625 case on_failure_jump: 7626 p1++; 7627 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7628 p1 += mcnt; 7629 break; 7630 7631 default: 7632 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) 7633 return false; 7634 } 7635 } /* while p1 < end */ 7636 7637 return true; 7638 } /* alt_match_null_string_p */ 7639 7640 7641 /* Deals with the ops common to group_match_null_string_p and 7642 alt_match_null_string_p. 7643 7644 Sets P to one after the op and its arguments, if any. */ 7645 7646 static boolean 7647 PREFIX(common_op_match_null_string_p) (UCHAR_T **p, UCHAR_T *end, 7648 PREFIX(register_info_type) *reg_info) 7649 { 7650 int mcnt; 7651 boolean ret; 7652 int reg_no; 7653 UCHAR_T *p1 = *p; 7654 7655 switch ((re_opcode_t) *p1++) 7656 { 7657 case no_op: 7658 case begline: 7659 case endline: 7660 case begbuf: 7661 case endbuf: 7662 case wordbeg: 7663 case wordend: 7664 case wordbound: 7665 case notwordbound: 7666 #ifdef emacs 7667 case before_dot: 7668 case at_dot: 7669 case after_dot: 7670 #endif 7671 break; 7672 7673 case start_memory: 7674 reg_no = *p1; 7675 assert (reg_no > 0 && reg_no <= MAX_REGNUM); 7676 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info); 7677 7678 /* Have to set this here in case we're checking a group which 7679 contains a group and a back reference to it. */ 7680 7681 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) 7682 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; 7683 7684 if (!ret) 7685 return false; 7686 break; 7687 7688 /* If this is an optimized succeed_n for zero times, make the jump. */ 7689 case jump: 7690 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7691 if (mcnt >= 0) 7692 p1 += mcnt; 7693 else 7694 return false; 7695 break; 7696 7697 case succeed_n: 7698 /* Get to the number of times to succeed. */ 7699 p1 += OFFSET_ADDRESS_SIZE; 7700 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7701 7702 if (mcnt == 0) 7703 { 7704 p1 -= 2 * OFFSET_ADDRESS_SIZE; 7705 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7706 p1 += mcnt; 7707 } 7708 else 7709 return false; 7710 break; 7711 7712 case duplicate: 7713 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) 7714 return false; 7715 break; 7716 7717 case set_number_at: 7718 p1 += 2 * OFFSET_ADDRESS_SIZE; 7719 7720 default: 7721 /* All other opcodes mean we cannot match the empty string. */ 7722 return false; 7723 } 7724 7725 *p = p1; 7726 return true; 7727 } /* common_op_match_null_string_p */ 7728 7729 7730 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN 7731 bytes; nonzero otherwise. */ 7732 7733 static int 7734 PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2, register int len, 7735 RE_TRANSLATE_TYPE translate) 7736 { 7737 register const UCHAR_T *p1 = (const UCHAR_T *) s1; 7738 register const UCHAR_T *p2 = (const UCHAR_T *) s2; 7739 while (len) 7740 { 7741 #ifdef WCHAR 7742 if (((*p1<=0xff)?translate[*p1++]:*p1++) 7743 != ((*p2<=0xff)?translate[*p2++]:*p2++)) 7744 return 1; 7745 #else /* BYTE */ 7746 if (translate[*p1++] != translate[*p2++]) return 1; 7747 #endif /* WCHAR */ 7748 len--; 7749 } 7750 return 0; 7751 } 7752 7753 7754 #else /* not INSIDE_RECURSION */ 7755 7756 /* Entry points for GNU code. */ 7757 7758 /* re_compile_pattern is the GNU regular expression compiler: it 7759 compiles PATTERN (of length SIZE) and puts the result in BUFP. 7760 Returns 0 if the pattern was valid, otherwise an error string. 7761 7762 Assumes the `allocated' (and perhaps `buffer') and `translate' fields 7763 are set in BUFP on entry. 7764 7765 We call regex_compile to do the actual compilation. */ 7766 7767 const char * 7768 re_compile_pattern (const char *pattern, size_t length, 7769 struct re_pattern_buffer *bufp) 7770 { 7771 reg_errcode_t ret; 7772 7773 /* GNU code is written to assume at least RE_NREGS registers will be set 7774 (and at least one extra will be -1). */ 7775 bufp->regs_allocated = REGS_UNALLOCATED; 7776 7777 /* And GNU code determines whether or not to get register information 7778 by passing null for the REGS argument to re_match, etc., not by 7779 setting no_sub. */ 7780 bufp->no_sub = 0; 7781 7782 /* Match anchors at newline. */ 7783 bufp->newline_anchor = 1; 7784 7785 # ifdef MBS_SUPPORT 7786 if (MB_CUR_MAX != 1) 7787 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp); 7788 else 7789 # endif 7790 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp); 7791 7792 if (!ret) 7793 return NULL; 7794 return gettext (re_error_msgid[(int) ret]); 7795 } 7796 #ifdef _LIBC 7797 weak_alias (__re_compile_pattern, re_compile_pattern) 7798 #endif 7799 7800 /* Entry points compatible with 4.2 BSD regex library. We don't define 7801 them unless specifically requested. */ 7802 7803 #if defined _REGEX_RE_COMP || defined _LIBC 7804 7805 /* BSD has one and only one pattern buffer. */ 7806 static struct re_pattern_buffer re_comp_buf; 7807 7808 char * 7809 #ifdef _LIBC 7810 /* Make these definitions weak in libc, so POSIX programs can redefine 7811 these names if they don't use our functions, and still use 7812 regcomp/regexec below without link errors. */ 7813 weak_function 7814 #endif 7815 re_comp (const char *s) 7816 { 7817 reg_errcode_t ret; 7818 7819 if (!s) 7820 { 7821 if (!re_comp_buf.buffer) 7822 return (char *) gettext ("No previous regular expression"); 7823 return 0; 7824 } 7825 7826 if (!re_comp_buf.buffer) 7827 { 7828 re_comp_buf.buffer = (unsigned char *) malloc (200); 7829 if (re_comp_buf.buffer == NULL) 7830 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); 7831 re_comp_buf.allocated = 200; 7832 7833 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); 7834 if (re_comp_buf.fastmap == NULL) 7835 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); 7836 } 7837 7838 /* Since `re_exec' always passes NULL for the `regs' argument, we 7839 don't need to initialize the pattern buffer fields which affect it. */ 7840 7841 /* Match anchors at newlines. */ 7842 re_comp_buf.newline_anchor = 1; 7843 7844 # ifdef MBS_SUPPORT 7845 if (MB_CUR_MAX != 1) 7846 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); 7847 else 7848 # endif 7849 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); 7850 7851 if (!ret) 7852 return NULL; 7853 7854 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ 7855 return (char *) gettext (re_error_msgid[(int) ret]); 7856 } 7857 7858 7859 int 7860 #ifdef _LIBC 7861 weak_function 7862 #endif 7863 re_exec (const char *s) 7864 { 7865 const int len = strlen (s); 7866 return 7867 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); 7868 } 7869 7870 #endif /* _REGEX_RE_COMP */ 7871 7872 /* POSIX.2 functions. Don't define these for Emacs. */ 7873 7874 #ifndef emacs 7875 7876 /* regcomp takes a regular expression as a string and compiles it. 7877 7878 PREG is a regex_t *. We do not expect any fields to be initialized, 7879 since POSIX says we shouldn't. Thus, we set 7880 7881 `buffer' to the compiled pattern; 7882 `used' to the length of the compiled pattern; 7883 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the 7884 REG_EXTENDED bit in CFLAGS is set; otherwise, to 7885 RE_SYNTAX_POSIX_BASIC; 7886 `newline_anchor' to REG_NEWLINE being set in CFLAGS; 7887 `fastmap' to an allocated space for the fastmap; 7888 `fastmap_accurate' to zero; 7889 `re_nsub' to the number of subexpressions in PATTERN. 7890 7891 PATTERN is the address of the pattern string. 7892 7893 CFLAGS is a series of bits which affect compilation. 7894 7895 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we 7896 use POSIX basic syntax. 7897 7898 If REG_NEWLINE is set, then . and [^...] don't match newline. 7899 Also, regexec will try a match beginning after every newline. 7900 7901 If REG_ICASE is set, then we considers upper- and lowercase 7902 versions of letters to be equivalent when matching. 7903 7904 If REG_NOSUB is set, then when PREG is passed to regexec, that 7905 routine will report only success or failure, and nothing about the 7906 registers. 7907 7908 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for 7909 the return codes and their meanings.) */ 7910 7911 int 7912 regcomp (regex_t *preg, const char *pattern, int cflags) 7913 { 7914 reg_errcode_t ret; 7915 reg_syntax_t syntax 7916 = (cflags & REG_EXTENDED) ? 7917 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; 7918 7919 /* regex_compile will allocate the space for the compiled pattern. */ 7920 preg->buffer = 0; 7921 preg->allocated = 0; 7922 preg->used = 0; 7923 7924 /* Try to allocate space for the fastmap. */ 7925 preg->fastmap = (char *) malloc (1 << BYTEWIDTH); 7926 7927 if (cflags & REG_ICASE) 7928 { 7929 int i; 7930 7931 preg->translate 7932 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE 7933 * sizeof (*(RE_TRANSLATE_TYPE)0)); 7934 if (preg->translate == NULL) 7935 return (int) REG_ESPACE; 7936 7937 /* Map uppercase characters to corresponding lowercase ones. */ 7938 for (i = 0; i < CHAR_SET_SIZE; i++) 7939 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; 7940 } 7941 else 7942 preg->translate = NULL; 7943 7944 /* If REG_NEWLINE is set, newlines are treated differently. */ 7945 if (cflags & REG_NEWLINE) 7946 { /* REG_NEWLINE implies neither . nor [^...] match newline. */ 7947 syntax &= ~RE_DOT_NEWLINE; 7948 syntax |= RE_HAT_LISTS_NOT_NEWLINE; 7949 /* It also changes the matching behavior. */ 7950 preg->newline_anchor = 1; 7951 } 7952 else 7953 preg->newline_anchor = 0; 7954 7955 preg->no_sub = !!(cflags & REG_NOSUB); 7956 7957 /* POSIX says a null character in the pattern terminates it, so we 7958 can use strlen here in compiling the pattern. */ 7959 # ifdef MBS_SUPPORT 7960 if (MB_CUR_MAX != 1) 7961 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg); 7962 else 7963 # endif 7964 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg); 7965 7966 /* POSIX doesn't distinguish between an unmatched open-group and an 7967 unmatched close-group: both are REG_EPAREN. */ 7968 if (ret == REG_ERPAREN) ret = REG_EPAREN; 7969 7970 if (ret == REG_NOERROR && preg->fastmap) 7971 { 7972 /* Compute the fastmap now, since regexec cannot modify the pattern 7973 buffer. */ 7974 if (re_compile_fastmap (preg) == -2) 7975 { 7976 /* Some error occurred while computing the fastmap, just forget 7977 about it. */ 7978 free (preg->fastmap); 7979 preg->fastmap = NULL; 7980 } 7981 } 7982 7983 return (int) ret; 7984 } 7985 #ifdef _LIBC 7986 weak_alias (__regcomp, regcomp) 7987 #endif 7988 7989 7990 /* regexec searches for a given pattern, specified by PREG, in the 7991 string STRING. 7992 7993 If NMATCH is zero or REG_NOSUB was set in the cflags argument to 7994 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at 7995 least NMATCH elements, and we set them to the offsets of the 7996 corresponding matched substrings. 7997 7998 EFLAGS specifies `execution flags' which affect matching: if 7999 REG_NOTBOL is set, then ^ does not match at the beginning of the 8000 string; if REG_NOTEOL is set, then $ does not match at the end. 8001 8002 We return 0 if we find a match and REG_NOMATCH if not. */ 8003 8004 int 8005 regexec (const regex_t *preg, const char *string, size_t nmatch, 8006 regmatch_t pmatch[], int eflags) 8007 { 8008 int ret; 8009 struct re_registers regs; 8010 regex_t private_preg; 8011 int len = strlen (string); 8012 boolean want_reg_info = !preg->no_sub && nmatch > 0; 8013 8014 private_preg = *preg; 8015 8016 private_preg.not_bol = !!(eflags & REG_NOTBOL); 8017 private_preg.not_eol = !!(eflags & REG_NOTEOL); 8018 8019 /* The user has told us exactly how many registers to return 8020 information about, via `nmatch'. We have to pass that on to the 8021 matching routines. */ 8022 private_preg.regs_allocated = REGS_FIXED; 8023 8024 if (want_reg_info) 8025 { 8026 regs.num_regs = nmatch; 8027 regs.start = TALLOC (nmatch * 2, regoff_t); 8028 if (regs.start == NULL) 8029 return (int) REG_NOMATCH; 8030 regs.end = regs.start + nmatch; 8031 } 8032 8033 /* Perform the searching operation. */ 8034 ret = re_search (&private_preg, string, len, 8035 /* start: */ 0, /* range: */ len, 8036 want_reg_info ? ®s : (struct re_registers *) 0); 8037 8038 /* Copy the register information to the POSIX structure. */ 8039 if (want_reg_info) 8040 { 8041 if (ret >= 0) 8042 { 8043 unsigned r; 8044 8045 for (r = 0; r < nmatch; r++) 8046 { 8047 pmatch[r].rm_so = regs.start[r]; 8048 pmatch[r].rm_eo = regs.end[r]; 8049 } 8050 } 8051 8052 /* If we needed the temporary register info, free the space now. */ 8053 free (regs.start); 8054 } 8055 8056 /* We want zero return to mean success, unlike `re_search'. */ 8057 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; 8058 } 8059 #ifdef _LIBC 8060 weak_alias (__regexec, regexec) 8061 #endif 8062 8063 8064 /* Returns a message corresponding to an error code, ERRCODE, returned 8065 from either regcomp or regexec. We don't use PREG here. */ 8066 8067 size_t 8068 regerror (int errcode, const regex_t *preg ATTRIBUTE_UNUSED, 8069 char *errbuf, size_t errbuf_size) 8070 { 8071 const char *msg; 8072 size_t msg_size; 8073 8074 if (errcode < 0 8075 || errcode >= (int) (sizeof (re_error_msgid) 8076 / sizeof (re_error_msgid[0]))) 8077 /* Only error codes returned by the rest of the code should be passed 8078 to this routine. If we are given anything else, or if other regex 8079 code generates an invalid error code, then the program has a bug. 8080 Dump core so we can fix it. */ 8081 abort (); 8082 8083 msg = gettext (re_error_msgid[errcode]); 8084 8085 msg_size = strlen (msg) + 1; /* Includes the null. */ 8086 8087 if (errbuf_size != 0) 8088 { 8089 if (msg_size > errbuf_size) 8090 { 8091 #if defined HAVE_MEMPCPY || defined _LIBC 8092 *((char *) mempcpy (errbuf, msg, errbuf_size - 1)) = '\0'; 8093 #else 8094 memcpy (errbuf, msg, errbuf_size - 1); 8095 errbuf[errbuf_size - 1] = 0; 8096 #endif 8097 } 8098 else 8099 memcpy (errbuf, msg, msg_size); 8100 } 8101 8102 return msg_size; 8103 } 8104 #ifdef _LIBC 8105 weak_alias (__regerror, regerror) 8106 #endif 8107 8108 8109 /* Free dynamically allocated space used by PREG. */ 8110 8111 void 8112 regfree (regex_t *preg) 8113 { 8114 if (preg->buffer != NULL) 8115 free (preg->buffer); 8116 preg->buffer = NULL; 8117 8118 preg->allocated = 0; 8119 preg->used = 0; 8120 8121 if (preg->fastmap != NULL) 8122 free (preg->fastmap); 8123 preg->fastmap = NULL; 8124 preg->fastmap_accurate = 0; 8125 8126 if (preg->translate != NULL) 8127 free (preg->translate); 8128 preg->translate = NULL; 8129 } 8130 #ifdef _LIBC 8131 weak_alias (__regfree, regfree) 8132 #endif 8133 8134 #endif /* not emacs */ 8135 8136 #endif /* not INSIDE_RECURSION */ 8137 8138 8139 #undef STORE_NUMBER 8140 #undef STORE_NUMBER_AND_INCR 8141 #undef EXTRACT_NUMBER 8142 #undef EXTRACT_NUMBER_AND_INCR 8143 8144 #undef DEBUG_PRINT_COMPILED_PATTERN 8145 #undef DEBUG_PRINT_DOUBLE_STRING 8146 8147 #undef INIT_FAIL_STACK 8148 #undef RESET_FAIL_STACK 8149 #undef DOUBLE_FAIL_STACK 8150 #undef PUSH_PATTERN_OP 8151 #undef PUSH_FAILURE_POINTER 8152 #undef PUSH_FAILURE_INT 8153 #undef PUSH_FAILURE_ELT 8154 #undef POP_FAILURE_POINTER 8155 #undef POP_FAILURE_INT 8156 #undef POP_FAILURE_ELT 8157 #undef DEBUG_PUSH 8158 #undef DEBUG_POP 8159 #undef PUSH_FAILURE_POINT 8160 #undef POP_FAILURE_POINT 8161 8162 #undef REG_UNSET_VALUE 8163 #undef REG_UNSET 8164 8165 #undef PATFETCH 8166 #undef PATFETCH_RAW 8167 #undef PATUNFETCH 8168 #undef TRANSLATE 8169 8170 #undef INIT_BUF_SIZE 8171 #undef GET_BUFFER_SPACE 8172 #undef BUF_PUSH 8173 #undef BUF_PUSH_2 8174 #undef BUF_PUSH_3 8175 #undef STORE_JUMP 8176 #undef STORE_JUMP2 8177 #undef INSERT_JUMP 8178 #undef INSERT_JUMP2 8179 #undef EXTEND_BUFFER 8180 #undef GET_UNSIGNED_NUMBER 8181 #undef FREE_STACK_RETURN 8182 8183 # undef POINTER_TO_OFFSET 8184 # undef MATCHING_IN_FRST_STRING 8185 # undef PREFETCH 8186 # undef AT_STRINGS_BEG 8187 # undef AT_STRINGS_END 8188 # undef WORDCHAR_P 8189 # undef FREE_VAR 8190 # undef FREE_VARIABLES 8191 # undef NO_HIGHEST_ACTIVE_REG 8192 # undef NO_LOWEST_ACTIVE_REG 8193 8194 # undef CHAR_T 8195 # undef UCHAR_T 8196 # undef COMPILED_BUFFER_VAR 8197 # undef OFFSET_ADDRESS_SIZE 8198 # undef CHAR_CLASS_SIZE 8199 # undef PREFIX 8200 # undef ARG_PREFIX 8201 # undef PUT_CHAR 8202 # undef BYTE 8203 # undef WCHAR 8204 8205 # define DEFINED_ONCE 8206