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) 1993-2018 Free Software Foundation, Inc. 7 This file is part of the GNU C Library. 8 9 The GNU C Library is free software; you can redistribute it and/or 10 modify it under the terms of the GNU Lesser General Public 11 License as published by the Free Software Foundation; either 12 version 2.1 of the License, or (at your option) any later version. 13 14 The GNU C Library is distributed in the hope that it will be useful, 15 but WITHOUT ANY WARRANTY; without even the implied warranty of 16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 17 Lesser General Public License for more details. 18 19 You should have received a copy of the GNU Lesser General Public 20 License along with the GNU C Library; if not, write to the Free 21 Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 22 02110-1301 USA. */ 23 24 /* This file has been modified for usage in libiberty. It includes "xregex.h" 25 instead of <regex.h>. The "xregex.h" header file renames all external 26 routines with an "x" prefix so they do not collide with the native regex 27 routines or with other components regex routines. */ 28 /* AIX requires this to be the first thing in the file. */ 29 #if defined _AIX && !defined __GNUC__ && !defined REGEX_MALLOC 30 #pragma alloca 31 #endif 32 33 #undef _GNU_SOURCE 34 #define _GNU_SOURCE 35 36 #ifndef INSIDE_RECURSION 37 # ifdef HAVE_CONFIG_H 38 # include <config.h> 39 # endif 40 #endif 41 42 #include <ansidecl.h> 43 44 #ifndef INSIDE_RECURSION 45 46 # if defined STDC_HEADERS && !defined emacs 47 # include <stddef.h> 48 # define PTR_INT_TYPE ptrdiff_t 49 # else 50 /* We need this for `regex.h', and perhaps for the Emacs include files. */ 51 # include <sys/types.h> 52 # define PTR_INT_TYPE long 53 # endif 54 55 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC) 56 57 /* For platform which support the ISO C amendement 1 functionality we 58 support user defined character classes. */ 59 # if defined _LIBC || WIDE_CHAR_SUPPORT 60 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ 61 # include <wchar.h> 62 # include <wctype.h> 63 # endif 64 65 # ifdef _LIBC 66 /* We have to keep the namespace clean. */ 67 # define regfree(preg) __regfree (preg) 68 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) 69 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) 70 # define regerror(errcode, preg, errbuf, errbuf_size) \ 71 __regerror(errcode, preg, errbuf, errbuf_size) 72 # define re_set_registers(bu, re, nu, st, en) \ 73 __re_set_registers (bu, re, nu, st, en) 74 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ 75 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) 76 # define re_match(bufp, string, size, pos, regs) \ 77 __re_match (bufp, string, size, pos, regs) 78 # define re_search(bufp, string, size, startpos, range, regs) \ 79 __re_search (bufp, string, size, startpos, range, regs) 80 # define re_compile_pattern(pattern, length, bufp) \ 81 __re_compile_pattern (pattern, length, bufp) 82 # define re_set_syntax(syntax) __re_set_syntax (syntax) 83 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ 84 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) 85 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) 86 87 # define btowc __btowc 88 89 /* We are also using some library internals. */ 90 # include <locale/localeinfo.h> 91 # include <locale/elem-hash.h> 92 # include <langinfo.h> 93 # include <locale/coll-lookup.h> 94 # endif 95 96 /* This is for other GNU distributions with internationalized messages. */ 97 # if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC 98 # include <libintl.h> 99 # ifdef _LIBC 100 # undef gettext 101 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES) 102 # endif 103 # else 104 # define gettext(msgid) (msgid) 105 # endif 106 107 # ifndef gettext_noop 108 /* This define is so xgettext can find the internationalizable 109 strings. */ 110 # define gettext_noop(String) String 111 # endif 112 113 /* The `emacs' switch turns on certain matching commands 114 that make sense only in Emacs. */ 115 # ifdef emacs 116 117 # include "lisp.h" 118 # include "buffer.h" 119 # include "syntax.h" 120 121 # else /* not emacs */ 122 123 /* If we are not linking with Emacs proper, 124 we can't use the relocating allocator 125 even if config.h says that we can. */ 126 # undef REL_ALLOC 127 128 # if defined STDC_HEADERS || defined _LIBC 129 # include <stdlib.h> 130 # else 131 char *malloc (); 132 char *realloc (); 133 # endif 134 135 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. 136 If nothing else has been done, use the method below. */ 137 # ifdef INHIBIT_STRING_HEADER 138 # if !(defined HAVE_BZERO && defined HAVE_BCOPY) 139 # if !defined bzero && !defined bcopy 140 # undef INHIBIT_STRING_HEADER 141 # endif 142 # endif 143 # endif 144 145 /* This is the normal way of making sure we have a bcopy and a bzero. 146 This is used in most programs--a few other programs avoid this 147 by defining INHIBIT_STRING_HEADER. */ 148 # ifndef INHIBIT_STRING_HEADER 149 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC 150 # include <string.h> 151 # ifndef bzero 152 # ifndef _LIBC 153 # define bzero(s, n) ((void) memset (s, '\0', n)) 154 # else 155 # define bzero(s, n) __bzero (s, n) 156 # endif 157 # endif 158 # else 159 # include <strings.h> 160 # ifndef memcmp 161 # define memcmp(s1, s2, n) bcmp (s1, s2, n) 162 # endif 163 # ifndef memcpy 164 # define memcpy(d, s, n) (bcopy (s, d, n), (d)) 165 # endif 166 # endif 167 # endif 168 169 /* Define the syntax stuff for \<, \>, etc. */ 170 171 /* This must be nonzero for the wordchar and notwordchar pattern 172 commands in re_match_2. */ 173 # ifndef Sword 174 # define Sword 1 175 # endif 176 177 # ifdef SWITCH_ENUM_BUG 178 # define SWITCH_ENUM_CAST(x) ((int)(x)) 179 # else 180 # define SWITCH_ENUM_CAST(x) (x) 181 # endif 182 183 # endif /* not emacs */ 184 185 # if defined _LIBC || HAVE_LIMITS_H 186 # include <limits.h> 187 # endif 188 189 # ifndef MB_LEN_MAX 190 # define MB_LEN_MAX 1 191 # endif 192 193 /* Get the interface, including the syntax bits. */ 194 # include "xregex.h" /* change for libiberty */ 195 196 /* isalpha etc. are used for the character classes. */ 197 # include <ctype.h> 198 199 /* Jim Meyering writes: 200 201 "... Some ctype macros are valid only for character codes that 202 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when 203 using /bin/cc or gcc but without giving an ansi option). So, all 204 ctype uses should be through macros like ISPRINT... If 205 STDC_HEADERS is defined, then autoconf has verified that the ctype 206 macros don't need to be guarded with references to isascii. ... 207 Defining isascii to 1 should let any compiler worth its salt 208 eliminate the && through constant folding." 209 Solaris defines some of these symbols so we must undefine them first. */ 210 211 # undef ISASCII 212 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) 213 # define ISASCII(c) 1 214 # else 215 # define ISASCII(c) isascii(c) 216 # endif 217 218 # ifdef isblank 219 # define ISBLANK(c) (ISASCII (c) && isblank (c)) 220 # else 221 # define ISBLANK(c) ((c) == ' ' || (c) == '\t') 222 # endif 223 # ifdef isgraph 224 # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) 225 # else 226 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) 227 # endif 228 229 # undef ISPRINT 230 # define ISPRINT(c) (ISASCII (c) && isprint (c)) 231 # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) 232 # define ISALNUM(c) (ISASCII (c) && isalnum (c)) 233 # define ISALPHA(c) (ISASCII (c) && isalpha (c)) 234 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) 235 # define ISLOWER(c) (ISASCII (c) && islower (c)) 236 # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) 237 # define ISSPACE(c) (ISASCII (c) && isspace (c)) 238 # define ISUPPER(c) (ISASCII (c) && isupper (c)) 239 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) 240 241 # ifdef _tolower 242 # define TOLOWER(c) _tolower(c) 243 # else 244 # define TOLOWER(c) tolower(c) 245 # endif 246 247 # ifndef NULL 248 # define NULL (void *)0 249 # endif 250 251 /* We remove any previous definition of `SIGN_EXTEND_CHAR', 252 since ours (we hope) works properly with all combinations of 253 machines, compilers, `char' and `unsigned char' argument types. 254 (Per Bothner suggested the basic approach.) */ 255 # undef SIGN_EXTEND_CHAR 256 # if __STDC__ 257 # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) 258 # else /* not __STDC__ */ 259 /* As in Harbison and Steele. */ 260 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) 261 # endif 262 263 # ifndef emacs 264 /* How many characters in the character set. */ 265 # define CHAR_SET_SIZE 256 266 267 # ifdef SYNTAX_TABLE 268 269 extern char *re_syntax_table; 270 271 # else /* not SYNTAX_TABLE */ 272 273 static char re_syntax_table[CHAR_SET_SIZE]; 274 275 static void init_syntax_once (void); 276 277 static void 278 init_syntax_once (void) 279 { 280 register int c; 281 static int done = 0; 282 283 if (done) 284 return; 285 bzero (re_syntax_table, sizeof re_syntax_table); 286 287 for (c = 0; c < CHAR_SET_SIZE; ++c) 288 if (ISALNUM (c)) 289 re_syntax_table[c] = Sword; 290 291 re_syntax_table['_'] = Sword; 292 293 done = 1; 294 } 295 296 # endif /* not SYNTAX_TABLE */ 297 298 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)] 299 300 # endif /* emacs */ 301 302 /* Integer type for pointers. */ 303 # if !defined _LIBC && !defined HAVE_UINTPTR_T 304 typedef unsigned long int uintptr_t; 305 # endif 306 307 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we 308 use `alloca' instead of `malloc'. This is because using malloc in 309 re_search* or re_match* could cause memory leaks when C-g is used in 310 Emacs; also, malloc is slower and causes storage fragmentation. On 311 the other hand, malloc is more portable, and easier to debug. 312 313 Because we sometimes use alloca, some routines have to be macros, 314 not functions -- `alloca'-allocated space disappears at the end of the 315 function it is called in. */ 316 317 # ifdef REGEX_MALLOC 318 319 # define REGEX_ALLOCATE malloc 320 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) 321 # define REGEX_FREE free 322 323 # else /* not REGEX_MALLOC */ 324 325 /* Emacs already defines alloca, sometimes. */ 326 # ifndef alloca 327 328 /* Make alloca work the best possible way. */ 329 # ifdef __GNUC__ 330 # define alloca __builtin_alloca 331 # else /* not __GNUC__ */ 332 # if HAVE_ALLOCA_H 333 # include <alloca.h> 334 # endif /* HAVE_ALLOCA_H */ 335 # endif /* not __GNUC__ */ 336 337 # endif /* not alloca */ 338 339 # define REGEX_ALLOCATE alloca 340 341 /* Assumes a `char *destination' variable. */ 342 # define REGEX_REALLOCATE(source, osize, nsize) \ 343 (destination = (char *) alloca (nsize), \ 344 memcpy (destination, source, osize)) 345 346 /* No need to do anything to free, after alloca. */ 347 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ 348 349 # endif /* not REGEX_MALLOC */ 350 351 /* Define how to allocate the failure stack. */ 352 353 # if defined REL_ALLOC && defined REGEX_MALLOC 354 355 # define REGEX_ALLOCATE_STACK(size) \ 356 r_alloc (&failure_stack_ptr, (size)) 357 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ 358 r_re_alloc (&failure_stack_ptr, (nsize)) 359 # define REGEX_FREE_STACK(ptr) \ 360 r_alloc_free (&failure_stack_ptr) 361 362 # else /* not using relocating allocator */ 363 364 # ifdef REGEX_MALLOC 365 366 # define REGEX_ALLOCATE_STACK malloc 367 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) 368 # define REGEX_FREE_STACK free 369 370 # else /* not REGEX_MALLOC */ 371 372 # define REGEX_ALLOCATE_STACK alloca 373 374 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ 375 REGEX_REALLOCATE (source, osize, nsize) 376 /* No need to explicitly free anything. */ 377 # define REGEX_FREE_STACK(arg) 378 379 # endif /* not REGEX_MALLOC */ 380 # endif /* not using relocating allocator */ 381 382 383 /* True if `size1' is non-NULL and PTR is pointing anywhere inside 384 `string1' or just past its end. This works if PTR is NULL, which is 385 a good thing. */ 386 # define FIRST_STRING_P(ptr) \ 387 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) 388 389 /* (Re)Allocate N items of type T using malloc, or fail. */ 390 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) 391 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) 392 # define RETALLOC_IF(addr, n, t) \ 393 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) 394 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) 395 396 # define BYTEWIDTH 8 /* In bits. */ 397 398 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) 399 400 # undef MAX 401 # undef MIN 402 # define MAX(a, b) ((a) > (b) ? (a) : (b)) 403 # define MIN(a, b) ((a) < (b) ? (a) : (b)) 404 405 typedef char boolean; 406 # define false 0 407 # define true 1 408 409 static reg_errcode_t byte_regex_compile (const char *pattern, size_t size, 410 reg_syntax_t syntax, 411 struct re_pattern_buffer *bufp); 412 413 static int byte_re_match_2_internal (struct re_pattern_buffer *bufp, 414 const char *string1, int size1, 415 const char *string2, int size2, 416 int pos, 417 struct re_registers *regs, 418 int stop); 419 static int byte_re_search_2 (struct re_pattern_buffer *bufp, 420 const char *string1, int size1, 421 const char *string2, int size2, 422 int startpos, int range, 423 struct re_registers *regs, int stop); 424 static int byte_re_compile_fastmap (struct re_pattern_buffer *bufp); 425 426 #ifdef MBS_SUPPORT 427 static reg_errcode_t wcs_regex_compile (const char *pattern, size_t size, 428 reg_syntax_t syntax, 429 struct re_pattern_buffer *bufp); 430 431 432 static int wcs_re_match_2_internal (struct re_pattern_buffer *bufp, 433 const char *cstring1, int csize1, 434 const char *cstring2, int csize2, 435 int pos, 436 struct re_registers *regs, 437 int stop, 438 wchar_t *string1, int size1, 439 wchar_t *string2, int size2, 440 int *mbs_offset1, int *mbs_offset2); 441 static int wcs_re_search_2 (struct re_pattern_buffer *bufp, 442 const char *string1, int size1, 443 const char *string2, int size2, 444 int startpos, int range, 445 struct re_registers *regs, int stop); 446 static int wcs_re_compile_fastmap (struct re_pattern_buffer *bufp); 447 #endif 448 449 /* These are the command codes that appear in compiled regular 450 expressions. Some opcodes are followed by argument bytes. A 451 command code can specify any interpretation whatsoever for its 452 arguments. Zero bytes may appear in the compiled regular expression. */ 453 454 typedef enum 455 { 456 no_op = 0, 457 458 /* Succeed right away--no more backtracking. */ 459 succeed, 460 461 /* Followed by one byte giving n, then by n literal bytes. */ 462 exactn, 463 464 # ifdef MBS_SUPPORT 465 /* Same as exactn, but contains binary data. */ 466 exactn_bin, 467 # endif 468 469 /* Matches any (more or less) character. */ 470 anychar, 471 472 /* Matches any one char belonging to specified set. First 473 following byte is number of bitmap bytes. Then come bytes 474 for a bitmap saying which chars are in. Bits in each byte 475 are ordered low-bit-first. A character is in the set if its 476 bit is 1. A character too large to have a bit in the map is 477 automatically not in the set. */ 478 /* ifdef MBS_SUPPORT, following element is length of character 479 classes, length of collating symbols, length of equivalence 480 classes, length of character ranges, and length of characters. 481 Next, character class element, collating symbols elements, 482 equivalence class elements, range elements, and character 483 elements follow. 484 See regex_compile function. */ 485 charset, 486 487 /* Same parameters as charset, but match any character that is 488 not one of those specified. */ 489 charset_not, 490 491 /* Start remembering the text that is matched, for storing in a 492 register. Followed by one byte with the register number, in 493 the range 0 to one less than the pattern buffer's re_nsub 494 field. Then followed by one byte with the number of groups 495 inner to this one. (This last has to be part of the 496 start_memory only because we need it in the on_failure_jump 497 of re_match_2.) */ 498 start_memory, 499 500 /* Stop remembering the text that is matched and store it in a 501 memory register. Followed by one byte with the register 502 number, in the range 0 to one less than `re_nsub' in the 503 pattern buffer, and one byte with the number of inner groups, 504 just like `start_memory'. (We need the number of inner 505 groups here because we don't have any easy way of finding the 506 corresponding start_memory when we're at a stop_memory.) */ 507 stop_memory, 508 509 /* Match a duplicate of something remembered. Followed by one 510 byte containing the register number. */ 511 duplicate, 512 513 /* Fail unless at beginning of line. */ 514 begline, 515 516 /* Fail unless at end of line. */ 517 endline, 518 519 /* Succeeds if at beginning of buffer (if emacs) or at beginning 520 of string to be matched (if not). */ 521 begbuf, 522 523 /* Analogously, for end of buffer/string. */ 524 endbuf, 525 526 /* Followed by two byte relative address to which to jump. */ 527 jump, 528 529 /* Same as jump, but marks the end of an alternative. */ 530 jump_past_alt, 531 532 /* Followed by two-byte relative address of place to resume at 533 in case of failure. */ 534 /* ifdef MBS_SUPPORT, the size of address is 1. */ 535 on_failure_jump, 536 537 /* Like on_failure_jump, but pushes a placeholder instead of the 538 current string position when executed. */ 539 on_failure_keep_string_jump, 540 541 /* Throw away latest failure point and then jump to following 542 two-byte relative address. */ 543 /* ifdef MBS_SUPPORT, the size of address is 1. */ 544 pop_failure_jump, 545 546 /* Change to pop_failure_jump if know won't have to backtrack to 547 match; otherwise change to jump. This is used to jump 548 back to the beginning of a repeat. If what follows this jump 549 clearly won't match what the repeat does, such that we can be 550 sure that there is no use backtracking out of repetitions 551 already matched, then we change it to a pop_failure_jump. 552 Followed by two-byte address. */ 553 /* ifdef MBS_SUPPORT, the size of address is 1. */ 554 maybe_pop_jump, 555 556 /* Jump to following two-byte address, and push a dummy failure 557 point. This failure point will be thrown away if an attempt 558 is made to use it for a failure. A `+' construct makes this 559 before the first repeat. Also used as an intermediary kind 560 of jump when compiling an alternative. */ 561 /* ifdef MBS_SUPPORT, the size of address is 1. */ 562 dummy_failure_jump, 563 564 /* Push a dummy failure point and continue. Used at the end of 565 alternatives. */ 566 push_dummy_failure, 567 568 /* Followed by two-byte relative address and two-byte number n. 569 After matching N times, jump to the address upon failure. */ 570 /* ifdef MBS_SUPPORT, the size of address is 1. */ 571 succeed_n, 572 573 /* Followed by two-byte relative address, and two-byte number n. 574 Jump to the address N times, then fail. */ 575 /* ifdef MBS_SUPPORT, the size of address is 1. */ 576 jump_n, 577 578 /* Set the following two-byte relative address to the 579 subsequent two-byte number. The address *includes* the two 580 bytes of number. */ 581 /* ifdef MBS_SUPPORT, the size of address is 1. */ 582 set_number_at, 583 584 wordchar, /* Matches any word-constituent character. */ 585 notwordchar, /* Matches any char that is not a word-constituent. */ 586 587 wordbeg, /* Succeeds if at word beginning. */ 588 wordend, /* Succeeds if at word end. */ 589 590 wordbound, /* Succeeds if at a word boundary. */ 591 notwordbound /* Succeeds if not at a word boundary. */ 592 593 # ifdef emacs 594 ,before_dot, /* Succeeds if before point. */ 595 at_dot, /* Succeeds if at point. */ 596 after_dot, /* Succeeds if after point. */ 597 598 /* Matches any character whose syntax is specified. Followed by 599 a byte which contains a syntax code, e.g., Sword. */ 600 syntaxspec, 601 602 /* Matches any character whose syntax is not that specified. */ 603 notsyntaxspec 604 # endif /* emacs */ 605 } re_opcode_t; 606 #endif /* not INSIDE_RECURSION */ 607 608 609 #ifdef BYTE 610 # define CHAR_T char 611 # define UCHAR_T unsigned char 612 # define COMPILED_BUFFER_VAR bufp->buffer 613 # define OFFSET_ADDRESS_SIZE 2 614 # define PREFIX(name) byte_##name 615 # define ARG_PREFIX(name) name 616 # define PUT_CHAR(c) putchar (c) 617 #else 618 # ifdef WCHAR 619 # define CHAR_T wchar_t 620 # define UCHAR_T wchar_t 621 # define COMPILED_BUFFER_VAR wc_buffer 622 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */ 623 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1) 624 # define PREFIX(name) wcs_##name 625 # define ARG_PREFIX(name) c##name 626 /* Should we use wide stream?? */ 627 # define PUT_CHAR(c) printf ("%C", c); 628 # define TRUE 1 629 # define FALSE 0 630 # else 631 # ifdef MBS_SUPPORT 632 # define WCHAR 633 # define INSIDE_RECURSION 634 # include "regex.c" 635 # undef INSIDE_RECURSION 636 # endif 637 # define BYTE 638 # define INSIDE_RECURSION 639 # include "regex.c" 640 # undef INSIDE_RECURSION 641 # endif 642 #endif 643 644 #ifdef INSIDE_RECURSION 645 /* Common operations on the compiled pattern. */ 646 647 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ 648 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ 649 650 # ifdef WCHAR 651 # define STORE_NUMBER(destination, number) \ 652 do { \ 653 *(destination) = (UCHAR_T)(number); \ 654 } while (0) 655 # else /* BYTE */ 656 # define STORE_NUMBER(destination, number) \ 657 do { \ 658 (destination)[0] = (number) & 0377; \ 659 (destination)[1] = (number) >> 8; \ 660 } while (0) 661 # endif /* WCHAR */ 662 663 /* Same as STORE_NUMBER, except increment DESTINATION to 664 the byte after where the number is stored. Therefore, DESTINATION 665 must be an lvalue. */ 666 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ 667 668 # define STORE_NUMBER_AND_INCR(destination, number) \ 669 do { \ 670 STORE_NUMBER (destination, number); \ 671 (destination) += OFFSET_ADDRESS_SIZE; \ 672 } while (0) 673 674 /* Put into DESTINATION a number stored in two contiguous bytes starting 675 at SOURCE. */ 676 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ 677 678 # ifdef WCHAR 679 # define EXTRACT_NUMBER(destination, source) \ 680 do { \ 681 (destination) = *(source); \ 682 } while (0) 683 # else /* BYTE */ 684 # define EXTRACT_NUMBER(destination, source) \ 685 do { \ 686 (destination) = *(source) & 0377; \ 687 (destination) += ((unsigned) SIGN_EXTEND_CHAR (*((source) + 1))) << 8; \ 688 } while (0) 689 # endif 690 691 # ifdef DEBUG 692 static void PREFIX(extract_number) (int *dest, UCHAR_T *source); 693 static void 694 PREFIX(extract_number) (int *dest, UCHAR_T *source) 695 { 696 # ifdef WCHAR 697 *dest = *source; 698 # else /* BYTE */ 699 int temp = SIGN_EXTEND_CHAR (*(source + 1)); 700 *dest = *source & 0377; 701 *dest += temp << 8; 702 # endif 703 } 704 705 # ifndef EXTRACT_MACROS /* To debug the macros. */ 706 # undef EXTRACT_NUMBER 707 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src) 708 # endif /* not EXTRACT_MACROS */ 709 710 # endif /* DEBUG */ 711 712 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. 713 SOURCE must be an lvalue. */ 714 715 # define EXTRACT_NUMBER_AND_INCR(destination, source) \ 716 do { \ 717 EXTRACT_NUMBER (destination, source); \ 718 (source) += OFFSET_ADDRESS_SIZE; \ 719 } while (0) 720 721 # ifdef DEBUG 722 static void PREFIX(extract_number_and_incr) (int *destination, 723 UCHAR_T **source); 724 static void 725 PREFIX(extract_number_and_incr) (int *destination, UCHAR_T **source) 726 { 727 PREFIX(extract_number) (destination, *source); 728 *source += OFFSET_ADDRESS_SIZE; 729 } 730 731 # ifndef EXTRACT_MACROS 732 # undef EXTRACT_NUMBER_AND_INCR 733 # define EXTRACT_NUMBER_AND_INCR(dest, src) \ 734 PREFIX(extract_number_and_incr) (&dest, &src) 735 # endif /* not EXTRACT_MACROS */ 736 737 # endif /* DEBUG */ 738 739 740 741 /* If DEBUG is defined, Regex prints many voluminous messages about what 742 it is doing (if the variable `debug' is nonzero). If linked with the 743 main program in `iregex.c', you can enter patterns and strings 744 interactively. And if linked with the main program in `main.c' and 745 the other test files, you can run the already-written tests. */ 746 747 # ifdef DEBUG 748 749 # ifndef DEFINED_ONCE 750 751 /* We use standard I/O for debugging. */ 752 # include <stdio.h> 753 754 /* It is useful to test things that ``must'' be true when debugging. */ 755 # include <assert.h> 756 757 static int debug; 758 759 # define DEBUG_STATEMENT(e) e 760 # define DEBUG_PRINT1(x) if (debug) printf (x) 761 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) 762 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) 763 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) 764 # endif /* not DEFINED_ONCE */ 765 766 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ 767 if (debug) PREFIX(print_partial_compiled_pattern) (s, e) 768 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ 769 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2) 770 771 772 /* Print the fastmap in human-readable form. */ 773 774 # ifndef DEFINED_ONCE 775 void 776 print_fastmap (char *fastmap) 777 { 778 unsigned was_a_range = 0; 779 unsigned i = 0; 780 781 while (i < (1 << BYTEWIDTH)) 782 { 783 if (fastmap[i++]) 784 { 785 was_a_range = 0; 786 putchar (i - 1); 787 while (i < (1 << BYTEWIDTH) && fastmap[i]) 788 { 789 was_a_range = 1; 790 i++; 791 } 792 if (was_a_range) 793 { 794 printf ("-"); 795 putchar (i - 1); 796 } 797 } 798 } 799 putchar ('\n'); 800 } 801 # endif /* not DEFINED_ONCE */ 802 803 804 /* Print a compiled pattern string in human-readable form, starting at 805 the START pointer into it and ending just before the pointer END. */ 806 807 void 808 PREFIX(print_partial_compiled_pattern) (UCHAR_T *start, UCHAR_T *end) 809 { 810 int mcnt, mcnt2; 811 UCHAR_T *p1; 812 UCHAR_T *p = start; 813 UCHAR_T *pend = end; 814 815 if (start == NULL) 816 { 817 printf ("(null)\n"); 818 return; 819 } 820 821 /* Loop over pattern commands. */ 822 while (p < pend) 823 { 824 # ifdef _LIBC 825 printf ("%td:\t", p - start); 826 # else 827 printf ("%ld:\t", (long int) (p - start)); 828 # endif 829 830 switch ((re_opcode_t) *p++) 831 { 832 case no_op: 833 printf ("/no_op"); 834 break; 835 836 case exactn: 837 mcnt = *p++; 838 printf ("/exactn/%d", mcnt); 839 do 840 { 841 putchar ('/'); 842 PUT_CHAR (*p++); 843 } 844 while (--mcnt); 845 break; 846 847 # ifdef MBS_SUPPORT 848 case exactn_bin: 849 mcnt = *p++; 850 printf ("/exactn_bin/%d", mcnt); 851 do 852 { 853 printf("/%lx", (long int) *p++); 854 } 855 while (--mcnt); 856 break; 857 # endif /* MBS_SUPPORT */ 858 859 case start_memory: 860 mcnt = *p++; 861 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++); 862 break; 863 864 case stop_memory: 865 mcnt = *p++; 866 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++); 867 break; 868 869 case duplicate: 870 printf ("/duplicate/%ld", (long int) *p++); 871 break; 872 873 case anychar: 874 printf ("/anychar"); 875 break; 876 877 case charset: 878 case charset_not: 879 { 880 # ifdef WCHAR 881 int i, length; 882 wchar_t *workp = p; 883 printf ("/charset [%s", 884 (re_opcode_t) *(workp - 1) == charset_not ? "^" : ""); 885 p += 5; 886 length = *workp++; /* the length of char_classes */ 887 for (i=0 ; i<length ; i++) 888 printf("[:%lx:]", (long int) *p++); 889 length = *workp++; /* the length of collating_symbol */ 890 for (i=0 ; i<length ;) 891 { 892 printf("[."); 893 while(*p != 0) 894 PUT_CHAR((i++,*p++)); 895 i++,p++; 896 printf(".]"); 897 } 898 length = *workp++; /* the length of equivalence_class */ 899 for (i=0 ; i<length ;) 900 { 901 printf("[="); 902 while(*p != 0) 903 PUT_CHAR((i++,*p++)); 904 i++,p++; 905 printf("=]"); 906 } 907 length = *workp++; /* the length of char_range */ 908 for (i=0 ; i<length ; i++) 909 { 910 wchar_t range_start = *p++; 911 wchar_t range_end = *p++; 912 printf("%C-%C", range_start, range_end); 913 } 914 length = *workp++; /* the length of char */ 915 for (i=0 ; i<length ; i++) 916 printf("%C", *p++); 917 putchar (']'); 918 # else 919 register int c, last = -100; 920 register int in_range = 0; 921 922 printf ("/charset [%s", 923 (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); 924 925 assert (p + *p < pend); 926 927 for (c = 0; c < 256; c++) 928 if (c / 8 < *p 929 && (p[1 + (c/8)] & (1 << (c % 8)))) 930 { 931 /* Are we starting a range? */ 932 if (last + 1 == c && ! in_range) 933 { 934 putchar ('-'); 935 in_range = 1; 936 } 937 /* Have we broken a range? */ 938 else if (last + 1 != c && in_range) 939 { 940 putchar (last); 941 in_range = 0; 942 } 943 944 if (! in_range) 945 putchar (c); 946 947 last = c; 948 } 949 950 if (in_range) 951 putchar (last); 952 953 putchar (']'); 954 955 p += 1 + *p; 956 # endif /* WCHAR */ 957 } 958 break; 959 960 case begline: 961 printf ("/begline"); 962 break; 963 964 case endline: 965 printf ("/endline"); 966 break; 967 968 case on_failure_jump: 969 PREFIX(extract_number_and_incr) (&mcnt, &p); 970 # ifdef _LIBC 971 printf ("/on_failure_jump to %td", p + mcnt - start); 972 # else 973 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start)); 974 # endif 975 break; 976 977 case on_failure_keep_string_jump: 978 PREFIX(extract_number_and_incr) (&mcnt, &p); 979 # ifdef _LIBC 980 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start); 981 # else 982 printf ("/on_failure_keep_string_jump to %ld", 983 (long int) (p + mcnt - start)); 984 # endif 985 break; 986 987 case dummy_failure_jump: 988 PREFIX(extract_number_and_incr) (&mcnt, &p); 989 # ifdef _LIBC 990 printf ("/dummy_failure_jump to %td", p + mcnt - start); 991 # else 992 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start)); 993 # endif 994 break; 995 996 case push_dummy_failure: 997 printf ("/push_dummy_failure"); 998 break; 999 1000 case maybe_pop_jump: 1001 PREFIX(extract_number_and_incr) (&mcnt, &p); 1002 # ifdef _LIBC 1003 printf ("/maybe_pop_jump to %td", p + mcnt - start); 1004 # else 1005 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start)); 1006 # endif 1007 break; 1008 1009 case pop_failure_jump: 1010 PREFIX(extract_number_and_incr) (&mcnt, &p); 1011 # ifdef _LIBC 1012 printf ("/pop_failure_jump to %td", p + mcnt - start); 1013 # else 1014 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start)); 1015 # endif 1016 break; 1017 1018 case jump_past_alt: 1019 PREFIX(extract_number_and_incr) (&mcnt, &p); 1020 # ifdef _LIBC 1021 printf ("/jump_past_alt to %td", p + mcnt - start); 1022 # else 1023 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start)); 1024 # endif 1025 break; 1026 1027 case jump: 1028 PREFIX(extract_number_and_incr) (&mcnt, &p); 1029 # ifdef _LIBC 1030 printf ("/jump to %td", p + mcnt - start); 1031 # else 1032 printf ("/jump to %ld", (long int) (p + mcnt - start)); 1033 # endif 1034 break; 1035 1036 case succeed_n: 1037 PREFIX(extract_number_and_incr) (&mcnt, &p); 1038 p1 = p + mcnt; 1039 PREFIX(extract_number_and_incr) (&mcnt2, &p); 1040 # ifdef _LIBC 1041 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2); 1042 # else 1043 printf ("/succeed_n to %ld, %d times", 1044 (long int) (p1 - start), mcnt2); 1045 # endif 1046 break; 1047 1048 case jump_n: 1049 PREFIX(extract_number_and_incr) (&mcnt, &p); 1050 p1 = p + mcnt; 1051 PREFIX(extract_number_and_incr) (&mcnt2, &p); 1052 printf ("/jump_n to %d, %d times", p1 - start, mcnt2); 1053 break; 1054 1055 case set_number_at: 1056 PREFIX(extract_number_and_incr) (&mcnt, &p); 1057 p1 = p + mcnt; 1058 PREFIX(extract_number_and_incr) (&mcnt2, &p); 1059 # ifdef _LIBC 1060 printf ("/set_number_at location %td to %d", p1 - start, mcnt2); 1061 # else 1062 printf ("/set_number_at location %ld to %d", 1063 (long int) (p1 - start), mcnt2); 1064 # endif 1065 break; 1066 1067 case wordbound: 1068 printf ("/wordbound"); 1069 break; 1070 1071 case notwordbound: 1072 printf ("/notwordbound"); 1073 break; 1074 1075 case wordbeg: 1076 printf ("/wordbeg"); 1077 break; 1078 1079 case wordend: 1080 printf ("/wordend"); 1081 break; 1082 1083 # ifdef emacs 1084 case before_dot: 1085 printf ("/before_dot"); 1086 break; 1087 1088 case at_dot: 1089 printf ("/at_dot"); 1090 break; 1091 1092 case after_dot: 1093 printf ("/after_dot"); 1094 break; 1095 1096 case syntaxspec: 1097 printf ("/syntaxspec"); 1098 mcnt = *p++; 1099 printf ("/%d", mcnt); 1100 break; 1101 1102 case notsyntaxspec: 1103 printf ("/notsyntaxspec"); 1104 mcnt = *p++; 1105 printf ("/%d", mcnt); 1106 break; 1107 # endif /* emacs */ 1108 1109 case wordchar: 1110 printf ("/wordchar"); 1111 break; 1112 1113 case notwordchar: 1114 printf ("/notwordchar"); 1115 break; 1116 1117 case begbuf: 1118 printf ("/begbuf"); 1119 break; 1120 1121 case endbuf: 1122 printf ("/endbuf"); 1123 break; 1124 1125 default: 1126 printf ("?%ld", (long int) *(p-1)); 1127 } 1128 1129 putchar ('\n'); 1130 } 1131 1132 # ifdef _LIBC 1133 printf ("%td:\tend of pattern.\n", p - start); 1134 # else 1135 printf ("%ld:\tend of pattern.\n", (long int) (p - start)); 1136 # endif 1137 } 1138 1139 1140 void 1141 PREFIX(print_compiled_pattern) (struct re_pattern_buffer *bufp) 1142 { 1143 UCHAR_T *buffer = (UCHAR_T*) bufp->buffer; 1144 1145 PREFIX(print_partial_compiled_pattern) (buffer, buffer 1146 + bufp->used / sizeof(UCHAR_T)); 1147 printf ("%ld bytes used/%ld bytes allocated.\n", 1148 bufp->used, bufp->allocated); 1149 1150 if (bufp->fastmap_accurate && bufp->fastmap) 1151 { 1152 printf ("fastmap: "); 1153 print_fastmap (bufp->fastmap); 1154 } 1155 1156 # ifdef _LIBC 1157 printf ("re_nsub: %Zd\t", bufp->re_nsub); 1158 # else 1159 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub); 1160 # endif 1161 printf ("regs_alloc: %d\t", bufp->regs_allocated); 1162 printf ("can_be_null: %d\t", bufp->can_be_null); 1163 printf ("newline_anchor: %d\n", bufp->newline_anchor); 1164 printf ("no_sub: %d\t", bufp->no_sub); 1165 printf ("not_bol: %d\t", bufp->not_bol); 1166 printf ("not_eol: %d\t", bufp->not_eol); 1167 printf ("syntax: %lx\n", bufp->syntax); 1168 /* Perhaps we should print the translate table? */ 1169 } 1170 1171 1172 void 1173 PREFIX(print_double_string) (const CHAR_T *where, const CHAR_T *string1, 1174 int size1, const CHAR_T *string2, int size2) 1175 { 1176 int this_char; 1177 1178 if (where == NULL) 1179 printf ("(null)"); 1180 else 1181 { 1182 int cnt; 1183 1184 if (FIRST_STRING_P (where)) 1185 { 1186 for (this_char = where - string1; this_char < size1; this_char++) 1187 PUT_CHAR (string1[this_char]); 1188 1189 where = string2; 1190 } 1191 1192 cnt = 0; 1193 for (this_char = where - string2; this_char < size2; this_char++) 1194 { 1195 PUT_CHAR (string2[this_char]); 1196 if (++cnt > 100) 1197 { 1198 fputs ("...", stdout); 1199 break; 1200 } 1201 } 1202 } 1203 } 1204 1205 # ifndef DEFINED_ONCE 1206 void 1207 printchar (int c) 1208 { 1209 putc (c, stderr); 1210 } 1211 # endif 1212 1213 # else /* not DEBUG */ 1214 1215 # ifndef DEFINED_ONCE 1216 # undef assert 1217 # define assert(e) 1218 1219 # define DEBUG_STATEMENT(e) 1220 # define DEBUG_PRINT1(x) 1221 # define DEBUG_PRINT2(x1, x2) 1222 # define DEBUG_PRINT3(x1, x2, x3) 1223 # define DEBUG_PRINT4(x1, x2, x3, x4) 1224 # endif /* not DEFINED_ONCE */ 1225 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) 1226 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) 1227 1228 # endif /* not DEBUG */ 1229 1230 1231 1232 # ifdef WCHAR 1233 /* This convert a multibyte string to a wide character string. 1234 And write their correspondances to offset_buffer(see below) 1235 and write whether each wchar_t is binary data to is_binary. 1236 This assume invalid multibyte sequences as binary data. 1237 We assume offset_buffer and is_binary is already allocated 1238 enough space. */ 1239 1240 static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src, 1241 size_t len, int *offset_buffer, 1242 char *is_binary); 1243 static size_t 1244 convert_mbs_to_wcs (CHAR_T *dest, const unsigned char*src, size_t len, 1245 int *offset_buffer, char *is_binary) 1246 /* It hold correspondances between src(char string) and 1247 dest(wchar_t string) for optimization. 1248 e.g. src = "xxxyzz" 1249 dest = {'X', 'Y', 'Z'} 1250 (each "xxx", "y" and "zz" represent one multibyte character 1251 corresponding to 'X', 'Y' and 'Z'.) 1252 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")} 1253 = {0, 3, 4, 6} 1254 */ 1255 { 1256 wchar_t *pdest = dest; 1257 const unsigned char *psrc = src; 1258 size_t wc_count = 0; 1259 1260 mbstate_t mbs; 1261 int i, consumed; 1262 size_t mb_remain = len; 1263 size_t mb_count = 0; 1264 1265 /* Initialize the conversion state. */ 1266 memset (&mbs, 0, sizeof (mbstate_t)); 1267 1268 offset_buffer[0] = 0; 1269 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed, 1270 psrc += consumed) 1271 { 1272 #ifdef _LIBC 1273 consumed = __mbrtowc (pdest, psrc, mb_remain, &mbs); 1274 #else 1275 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs); 1276 #endif 1277 1278 if (consumed <= 0) 1279 /* failed to convert. maybe src contains binary data. 1280 So we consume 1 byte manualy. */ 1281 { 1282 *pdest = *psrc; 1283 consumed = 1; 1284 is_binary[wc_count] = TRUE; 1285 } 1286 else 1287 is_binary[wc_count] = FALSE; 1288 /* In sjis encoding, we use yen sign as escape character in 1289 place of reverse solidus. So we convert 0x5c(yen sign in 1290 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse 1291 solidus in UCS2). */ 1292 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5) 1293 *pdest = (wchar_t) *psrc; 1294 1295 offset_buffer[wc_count + 1] = mb_count += consumed; 1296 } 1297 1298 /* Fill remain of the buffer with sentinel. */ 1299 for (i = wc_count + 1 ; i <= len ; i++) 1300 offset_buffer[i] = mb_count + 1; 1301 1302 return wc_count; 1303 } 1304 1305 # endif /* WCHAR */ 1306 1307 #else /* not INSIDE_RECURSION */ 1308 1309 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can 1310 also be assigned to arbitrarily: each pattern buffer stores its own 1311 syntax, so it can be changed between regex compilations. */ 1312 /* This has no initializer because initialized variables in Emacs 1313 become read-only after dumping. */ 1314 reg_syntax_t re_syntax_options; 1315 1316 1317 /* Specify the precise syntax of regexps for compilation. This provides 1318 for compatibility for various utilities which historically have 1319 different, incompatible syntaxes. 1320 1321 The argument SYNTAX is a bit mask comprised of the various bits 1322 defined in regex.h. We return the old syntax. */ 1323 1324 reg_syntax_t 1325 re_set_syntax (reg_syntax_t syntax) 1326 { 1327 reg_syntax_t ret = re_syntax_options; 1328 1329 re_syntax_options = syntax; 1330 # ifdef DEBUG 1331 if (syntax & RE_DEBUG) 1332 debug = 1; 1333 else if (debug) /* was on but now is not */ 1334 debug = 0; 1335 # endif /* DEBUG */ 1336 return ret; 1337 } 1338 # ifdef _LIBC 1339 weak_alias (__re_set_syntax, re_set_syntax) 1340 # endif 1341 1342 /* This table gives an error message for each of the error codes listed 1343 in regex.h. Obviously the order here has to be same as there. 1344 POSIX doesn't require that we do anything for REG_NOERROR, 1345 but why not be nice? */ 1346 1347 static const char *re_error_msgid[] = 1348 { 1349 gettext_noop ("Success"), /* REG_NOERROR */ 1350 gettext_noop ("No match"), /* REG_NOMATCH */ 1351 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ 1352 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ 1353 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ 1354 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ 1355 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ 1356 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ 1357 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ 1358 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ 1359 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ 1360 gettext_noop ("Invalid range end"), /* REG_ERANGE */ 1361 gettext_noop ("Memory exhausted"), /* REG_ESPACE */ 1362 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ 1363 gettext_noop ("Premature end of regular expression"), /* REG_EEND */ 1364 gettext_noop ("Regular expression too big"), /* REG_ESIZE */ 1365 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */ 1366 }; 1367 1368 #endif /* INSIDE_RECURSION */ 1369 1370 #ifndef DEFINED_ONCE 1371 /* Avoiding alloca during matching, to placate r_alloc. */ 1372 1373 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the 1374 searching and matching functions should not call alloca. On some 1375 systems, alloca is implemented in terms of malloc, and if we're 1376 using the relocating allocator routines, then malloc could cause a 1377 relocation, which might (if the strings being searched are in the 1378 ralloc heap) shift the data out from underneath the regexp 1379 routines. 1380 1381 Here's another reason to avoid allocation: Emacs 1382 processes input from X in a signal handler; processing X input may 1383 call malloc; if input arrives while a matching routine is calling 1384 malloc, then we're scrod. But Emacs can't just block input while 1385 calling matching routines; then we don't notice interrupts when 1386 they come in. So, Emacs blocks input around all regexp calls 1387 except the matching calls, which it leaves unprotected, in the 1388 faith that they will not malloc. */ 1389 1390 /* Normally, this is fine. */ 1391 # define MATCH_MAY_ALLOCATE 1392 1393 /* When using GNU C, we are not REALLY using the C alloca, no matter 1394 what config.h may say. So don't take precautions for it. */ 1395 # ifdef __GNUC__ 1396 # undef C_ALLOCA 1397 # endif 1398 1399 /* The match routines may not allocate if (1) they would do it with malloc 1400 and (2) it's not safe for them to use malloc. 1401 Note that if REL_ALLOC is defined, matching would not use malloc for the 1402 failure stack, but we would still use it for the register vectors; 1403 so REL_ALLOC should not affect this. */ 1404 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs 1405 # undef MATCH_MAY_ALLOCATE 1406 # endif 1407 #endif /* not DEFINED_ONCE */ 1408 1409 #ifdef INSIDE_RECURSION 1410 /* Failure stack declarations and macros; both re_compile_fastmap and 1411 re_match_2 use a failure stack. These have to be macros because of 1412 REGEX_ALLOCATE_STACK. */ 1413 1414 1415 /* Number of failure points for which to initially allocate space 1416 when matching. If this number is exceeded, we allocate more 1417 space, so it is not a hard limit. */ 1418 # ifndef INIT_FAILURE_ALLOC 1419 # define INIT_FAILURE_ALLOC 5 1420 # endif 1421 1422 /* Roughly the maximum number of failure points on the stack. Would be 1423 exactly that if always used MAX_FAILURE_ITEMS items each time we failed. 1424 This is a variable only so users of regex can assign to it; we never 1425 change it ourselves. */ 1426 1427 # ifdef INT_IS_16BIT 1428 1429 # ifndef DEFINED_ONCE 1430 # if defined MATCH_MAY_ALLOCATE 1431 /* 4400 was enough to cause a crash on Alpha OSF/1, 1432 whose default stack limit is 2mb. */ 1433 long int re_max_failures = 4000; 1434 # else 1435 long int re_max_failures = 2000; 1436 # endif 1437 # endif 1438 1439 union PREFIX(fail_stack_elt) 1440 { 1441 UCHAR_T *pointer; 1442 long int integer; 1443 }; 1444 1445 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); 1446 1447 typedef struct 1448 { 1449 PREFIX(fail_stack_elt_t) *stack; 1450 unsigned long int size; 1451 unsigned long int avail; /* Offset of next open position. */ 1452 } PREFIX(fail_stack_type); 1453 1454 # else /* not INT_IS_16BIT */ 1455 1456 # ifndef DEFINED_ONCE 1457 # if defined MATCH_MAY_ALLOCATE 1458 /* 4400 was enough to cause a crash on Alpha OSF/1, 1459 whose default stack limit is 2mb. */ 1460 int re_max_failures = 4000; 1461 # else 1462 int re_max_failures = 2000; 1463 # endif 1464 # endif 1465 1466 union PREFIX(fail_stack_elt) 1467 { 1468 UCHAR_T *pointer; 1469 int integer; 1470 }; 1471 1472 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); 1473 1474 typedef struct 1475 { 1476 PREFIX(fail_stack_elt_t) *stack; 1477 unsigned size; 1478 unsigned avail; /* Offset of next open position. */ 1479 } PREFIX(fail_stack_type); 1480 1481 # endif /* INT_IS_16BIT */ 1482 1483 # ifndef DEFINED_ONCE 1484 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0) 1485 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) 1486 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) 1487 # endif 1488 1489 1490 /* Define macros to initialize and free the failure stack. 1491 Do `return -2' if the alloc fails. */ 1492 1493 # ifdef MATCH_MAY_ALLOCATE 1494 # define INIT_FAIL_STACK() \ 1495 do { \ 1496 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \ 1497 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \ 1498 \ 1499 if (fail_stack.stack == NULL) \ 1500 return -2; \ 1501 \ 1502 fail_stack.size = INIT_FAILURE_ALLOC; \ 1503 fail_stack.avail = 0; \ 1504 } while (0) 1505 1506 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) 1507 # else 1508 # define INIT_FAIL_STACK() \ 1509 do { \ 1510 fail_stack.avail = 0; \ 1511 } while (0) 1512 1513 # define RESET_FAIL_STACK() 1514 # endif 1515 1516 1517 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. 1518 1519 Return 1 if succeeds, and 0 if either ran out of memory 1520 allocating space for it or it was already too large. 1521 1522 REGEX_REALLOCATE_STACK requires `destination' be declared. */ 1523 1524 # define DOUBLE_FAIL_STACK(fail_stack) \ 1525 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \ 1526 ? 0 \ 1527 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \ 1528 REGEX_REALLOCATE_STACK ((fail_stack).stack, \ 1529 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \ 1530 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\ 1531 \ 1532 (fail_stack).stack == NULL \ 1533 ? 0 \ 1534 : ((fail_stack).size <<= 1, \ 1535 1))) 1536 1537 1538 /* Push pointer POINTER on FAIL_STACK. 1539 Return 1 if was able to do so and 0 if ran out of memory allocating 1540 space to do so. */ 1541 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ 1542 ((FAIL_STACK_FULL () \ 1543 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \ 1544 ? 0 \ 1545 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ 1546 1)) 1547 1548 /* Push a pointer value onto the failure stack. 1549 Assumes the variable `fail_stack'. Probably should only 1550 be called from within `PUSH_FAILURE_POINT'. */ 1551 # define PUSH_FAILURE_POINTER(item) \ 1552 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item) 1553 1554 /* This pushes an integer-valued item onto the failure stack. 1555 Assumes the variable `fail_stack'. Probably should only 1556 be called from within `PUSH_FAILURE_POINT'. */ 1557 # define PUSH_FAILURE_INT(item) \ 1558 fail_stack.stack[fail_stack.avail++].integer = (item) 1559 1560 /* Push a fail_stack_elt_t value onto the failure stack. 1561 Assumes the variable `fail_stack'. Probably should only 1562 be called from within `PUSH_FAILURE_POINT'. */ 1563 # define PUSH_FAILURE_ELT(item) \ 1564 fail_stack.stack[fail_stack.avail++] = (item) 1565 1566 /* These three POP... operations complement the three PUSH... operations. 1567 All assume that `fail_stack' is nonempty. */ 1568 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer 1569 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer 1570 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] 1571 1572 /* Used to omit pushing failure point id's when we're not debugging. */ 1573 # ifdef DEBUG 1574 # define DEBUG_PUSH PUSH_FAILURE_INT 1575 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () 1576 # else 1577 # define DEBUG_PUSH(item) 1578 # define DEBUG_POP(item_addr) 1579 # endif 1580 1581 1582 /* Push the information about the state we will need 1583 if we ever fail back to it. 1584 1585 Requires variables fail_stack, regstart, regend, reg_info, and 1586 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination' 1587 be declared. 1588 1589 Does `return FAILURE_CODE' if runs out of memory. */ 1590 1591 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ 1592 do { \ 1593 char *destination; \ 1594 /* Must be int, so when we don't save any registers, the arithmetic \ 1595 of 0 + -1 isn't done as unsigned. */ \ 1596 /* Can't be int, since there is not a shred of a guarantee that int \ 1597 is wide enough to hold a value of something to which pointer can \ 1598 be assigned */ \ 1599 active_reg_t this_reg; \ 1600 \ 1601 DEBUG_STATEMENT (failure_id++); \ 1602 DEBUG_STATEMENT (nfailure_points_pushed++); \ 1603 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ 1604 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ 1605 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ 1606 \ 1607 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \ 1608 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ 1609 \ 1610 /* Ensure we have enough space allocated for what we will push. */ \ 1611 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ 1612 { \ 1613 if (!DOUBLE_FAIL_STACK (fail_stack)) \ 1614 return failure_code; \ 1615 \ 1616 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ 1617 (fail_stack).size); \ 1618 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ 1619 } \ 1620 \ 1621 /* Push the info, starting with the registers. */ \ 1622 DEBUG_PRINT1 ("\n"); \ 1623 \ 1624 if (1) \ 1625 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ 1626 this_reg++) \ 1627 { \ 1628 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \ 1629 DEBUG_STATEMENT (num_regs_pushed++); \ 1630 \ 1631 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ 1632 PUSH_FAILURE_POINTER (regstart[this_reg]); \ 1633 \ 1634 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ 1635 PUSH_FAILURE_POINTER (regend[this_reg]); \ 1636 \ 1637 DEBUG_PRINT2 (" info: %p\n ", \ 1638 reg_info[this_reg].word.pointer); \ 1639 DEBUG_PRINT2 (" match_null=%d", \ 1640 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ 1641 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ 1642 DEBUG_PRINT2 (" matched_something=%d", \ 1643 MATCHED_SOMETHING (reg_info[this_reg])); \ 1644 DEBUG_PRINT2 (" ever_matched=%d", \ 1645 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ 1646 DEBUG_PRINT1 ("\n"); \ 1647 PUSH_FAILURE_ELT (reg_info[this_reg].word); \ 1648 } \ 1649 \ 1650 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\ 1651 PUSH_FAILURE_INT (lowest_active_reg); \ 1652 \ 1653 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\ 1654 PUSH_FAILURE_INT (highest_active_reg); \ 1655 \ 1656 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \ 1657 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ 1658 PUSH_FAILURE_POINTER (pattern_place); \ 1659 \ 1660 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \ 1661 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ 1662 size2); \ 1663 DEBUG_PRINT1 ("'\n"); \ 1664 PUSH_FAILURE_POINTER (string_place); \ 1665 \ 1666 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ 1667 DEBUG_PUSH (failure_id); \ 1668 } while (0) 1669 1670 # ifndef DEFINED_ONCE 1671 /* This is the number of items that are pushed and popped on the stack 1672 for each register. */ 1673 # define NUM_REG_ITEMS 3 1674 1675 /* Individual items aside from the registers. */ 1676 # ifdef DEBUG 1677 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ 1678 # else 1679 # define NUM_NONREG_ITEMS 4 1680 # endif 1681 1682 /* We push at most this many items on the stack. */ 1683 /* We used to use (num_regs - 1), which is the number of registers 1684 this regexp will save; but that was changed to 5 1685 to avoid stack overflow for a regexp with lots of parens. */ 1686 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS) 1687 1688 /* We actually push this many items. */ 1689 # define NUM_FAILURE_ITEMS \ 1690 (((0 \ 1691 ? 0 : highest_active_reg - lowest_active_reg + 1) \ 1692 * NUM_REG_ITEMS) \ 1693 + NUM_NONREG_ITEMS) 1694 1695 /* How many items can still be added to the stack without overflowing it. */ 1696 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) 1697 # endif /* not DEFINED_ONCE */ 1698 1699 1700 /* Pops what PUSH_FAIL_STACK pushes. 1701 1702 We restore into the parameters, all of which should be lvalues: 1703 STR -- the saved data position. 1704 PAT -- the saved pattern position. 1705 LOW_REG, HIGH_REG -- the highest and lowest active registers. 1706 REGSTART, REGEND -- arrays of string positions. 1707 REG_INFO -- array of information about each subexpression. 1708 1709 Also assumes the variables `fail_stack' and (if debugging), `bufp', 1710 `pend', `string1', `size1', `string2', and `size2'. */ 1711 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ 1712 { \ 1713 DEBUG_STATEMENT (unsigned failure_id;) \ 1714 active_reg_t this_reg; \ 1715 const UCHAR_T *string_temp; \ 1716 \ 1717 assert (!FAIL_STACK_EMPTY ()); \ 1718 \ 1719 /* Remove failure points and point to how many regs pushed. */ \ 1720 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ 1721 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ 1722 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ 1723 \ 1724 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ 1725 \ 1726 DEBUG_POP (&failure_id); \ 1727 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ 1728 \ 1729 /* If the saved string location is NULL, it came from an \ 1730 on_failure_keep_string_jump opcode, and we want to throw away the \ 1731 saved NULL, thus retaining our current position in the string. */ \ 1732 string_temp = POP_FAILURE_POINTER (); \ 1733 if (string_temp != NULL) \ 1734 str = (const CHAR_T *) string_temp; \ 1735 \ 1736 DEBUG_PRINT2 (" Popping string %p: `", str); \ 1737 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ 1738 DEBUG_PRINT1 ("'\n"); \ 1739 \ 1740 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \ 1741 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \ 1742 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ 1743 \ 1744 /* Restore register info. */ \ 1745 high_reg = (active_reg_t) POP_FAILURE_INT (); \ 1746 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \ 1747 \ 1748 low_reg = (active_reg_t) POP_FAILURE_INT (); \ 1749 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \ 1750 \ 1751 if (1) \ 1752 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ 1753 { \ 1754 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \ 1755 \ 1756 reg_info[this_reg].word = POP_FAILURE_ELT (); \ 1757 DEBUG_PRINT2 (" info: %p\n", \ 1758 reg_info[this_reg].word.pointer); \ 1759 \ 1760 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ 1761 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ 1762 \ 1763 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ 1764 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ 1765 } \ 1766 else \ 1767 { \ 1768 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \ 1769 { \ 1770 reg_info[this_reg].word.integer = 0; \ 1771 regend[this_reg] = 0; \ 1772 regstart[this_reg] = 0; \ 1773 } \ 1774 highest_active_reg = high_reg; \ 1775 } \ 1776 \ 1777 set_regs_matched_done = 0; \ 1778 DEBUG_STATEMENT (nfailure_points_popped++); \ 1779 } /* POP_FAILURE_POINT */ 1780 1781 /* Structure for per-register (a.k.a. per-group) information. 1782 Other register information, such as the 1783 starting and ending positions (which are addresses), and the list of 1784 inner groups (which is a bits list) are maintained in separate 1785 variables. 1786 1787 We are making a (strictly speaking) nonportable assumption here: that 1788 the compiler will pack our bit fields into something that fits into 1789 the type of `word', i.e., is something that fits into one item on the 1790 failure stack. */ 1791 1792 1793 /* Declarations and macros for re_match_2. */ 1794 1795 typedef union 1796 { 1797 PREFIX(fail_stack_elt_t) word; 1798 struct 1799 { 1800 /* This field is one if this group can match the empty string, 1801 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ 1802 # define MATCH_NULL_UNSET_VALUE 3 1803 unsigned match_null_string_p : 2; 1804 unsigned is_active : 1; 1805 unsigned matched_something : 1; 1806 unsigned ever_matched_something : 1; 1807 } bits; 1808 } PREFIX(register_info_type); 1809 1810 # ifndef DEFINED_ONCE 1811 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) 1812 # define IS_ACTIVE(R) ((R).bits.is_active) 1813 # define MATCHED_SOMETHING(R) ((R).bits.matched_something) 1814 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) 1815 1816 1817 /* Call this when have matched a real character; it sets `matched' flags 1818 for the subexpressions which we are currently inside. Also records 1819 that those subexprs have matched. */ 1820 # define SET_REGS_MATCHED() \ 1821 do \ 1822 { \ 1823 if (!set_regs_matched_done) \ 1824 { \ 1825 active_reg_t r; \ 1826 set_regs_matched_done = 1; \ 1827 for (r = lowest_active_reg; r <= highest_active_reg; r++) \ 1828 { \ 1829 MATCHED_SOMETHING (reg_info[r]) \ 1830 = EVER_MATCHED_SOMETHING (reg_info[r]) \ 1831 = 1; \ 1832 } \ 1833 } \ 1834 } \ 1835 while (0) 1836 # endif /* not DEFINED_ONCE */ 1837 1838 /* Registers are set to a sentinel when they haven't yet matched. */ 1839 static CHAR_T PREFIX(reg_unset_dummy); 1840 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy)) 1841 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE) 1842 1843 /* Subroutine declarations and macros for regex_compile. */ 1844 static void PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg); 1845 static void PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc, 1846 int arg1, int arg2); 1847 static void PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc, 1848 int arg, UCHAR_T *end); 1849 static void PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc, 1850 int arg1, int arg2, UCHAR_T *end); 1851 static boolean PREFIX(at_begline_loc_p) (const CHAR_T *pattern, 1852 const CHAR_T *p, 1853 reg_syntax_t syntax); 1854 static boolean PREFIX(at_endline_loc_p) (const CHAR_T *p, 1855 const CHAR_T *pend, 1856 reg_syntax_t syntax); 1857 # ifdef WCHAR 1858 static reg_errcode_t wcs_compile_range (CHAR_T range_start, 1859 const CHAR_T **p_ptr, 1860 const CHAR_T *pend, 1861 char *translate, 1862 reg_syntax_t syntax, 1863 UCHAR_T *b, 1864 CHAR_T *char_set); 1865 static void insert_space (int num, CHAR_T *loc, CHAR_T *end); 1866 # else /* BYTE */ 1867 static reg_errcode_t byte_compile_range (unsigned int range_start, 1868 const char **p_ptr, 1869 const char *pend, 1870 char *translate, 1871 reg_syntax_t syntax, 1872 unsigned char *b); 1873 # endif /* WCHAR */ 1874 1875 /* Fetch the next character in the uncompiled pattern---translating it 1876 if necessary. Also cast from a signed character in the constant 1877 string passed to us by the user to an unsigned char that we can use 1878 as an array index (in, e.g., `translate'). */ 1879 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, 1880 because it is impossible to allocate 4GB array for some encodings 1881 which have 4 byte character_set like UCS4. */ 1882 # ifndef PATFETCH 1883 # ifdef WCHAR 1884 # define PATFETCH(c) \ 1885 do {if (p == pend) return REG_EEND; \ 1886 c = (UCHAR_T) *p++; \ 1887 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \ 1888 } while (0) 1889 # else /* BYTE */ 1890 # define PATFETCH(c) \ 1891 do {if (p == pend) return REG_EEND; \ 1892 c = (unsigned char) *p++; \ 1893 if (translate) c = (unsigned char) translate[c]; \ 1894 } while (0) 1895 # endif /* WCHAR */ 1896 # endif 1897 1898 /* Fetch the next character in the uncompiled pattern, with no 1899 translation. */ 1900 # define PATFETCH_RAW(c) \ 1901 do {if (p == pend) return REG_EEND; \ 1902 c = (UCHAR_T) *p++; \ 1903 } while (0) 1904 1905 /* Go backwards one character in the pattern. */ 1906 # define PATUNFETCH p-- 1907 1908 1909 /* If `translate' is non-null, return translate[D], else just D. We 1910 cast the subscript to translate because some data is declared as 1911 `char *', to avoid warnings when a string constant is passed. But 1912 when we use a character as a subscript we must make it unsigned. */ 1913 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, 1914 because it is impossible to allocate 4GB array for some encodings 1915 which have 4 byte character_set like UCS4. */ 1916 1917 # ifndef TRANSLATE 1918 # ifdef WCHAR 1919 # define TRANSLATE(d) \ 1920 ((translate && ((UCHAR_T) (d)) <= 0xff) \ 1921 ? (char) translate[(unsigned char) (d)] : (d)) 1922 # else /* BYTE */ 1923 # define TRANSLATE(d) \ 1924 (translate ? (char) translate[(unsigned char) (d)] : (char) (d)) 1925 # endif /* WCHAR */ 1926 # endif 1927 1928 1929 /* Macros for outputting the compiled pattern into `buffer'. */ 1930 1931 /* If the buffer isn't allocated when it comes in, use this. */ 1932 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T)) 1933 1934 /* Make sure we have at least N more bytes of space in buffer. */ 1935 # ifdef WCHAR 1936 # define GET_BUFFER_SPACE(n) \ 1937 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \ 1938 + (n)*sizeof(CHAR_T)) > bufp->allocated) \ 1939 EXTEND_BUFFER () 1940 # else /* BYTE */ 1941 # define GET_BUFFER_SPACE(n) \ 1942 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \ 1943 EXTEND_BUFFER () 1944 # endif /* WCHAR */ 1945 1946 /* Make sure we have one more byte of buffer space and then add C to it. */ 1947 # define BUF_PUSH(c) \ 1948 do { \ 1949 GET_BUFFER_SPACE (1); \ 1950 *b++ = (UCHAR_T) (c); \ 1951 } while (0) 1952 1953 1954 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ 1955 # define BUF_PUSH_2(c1, c2) \ 1956 do { \ 1957 GET_BUFFER_SPACE (2); \ 1958 *b++ = (UCHAR_T) (c1); \ 1959 *b++ = (UCHAR_T) (c2); \ 1960 } while (0) 1961 1962 1963 /* As with BUF_PUSH_2, except for three bytes. */ 1964 # define BUF_PUSH_3(c1, c2, c3) \ 1965 do { \ 1966 GET_BUFFER_SPACE (3); \ 1967 *b++ = (UCHAR_T) (c1); \ 1968 *b++ = (UCHAR_T) (c2); \ 1969 *b++ = (UCHAR_T) (c3); \ 1970 } while (0) 1971 1972 /* Store a jump with opcode OP at LOC to location TO. We store a 1973 relative address offset by the three bytes the jump itself occupies. */ 1974 # define STORE_JUMP(op, loc, to) \ 1975 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE))) 1976 1977 /* Likewise, for a two-argument jump. */ 1978 # define STORE_JUMP2(op, loc, to, arg) \ 1979 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg) 1980 1981 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ 1982 # define INSERT_JUMP(op, loc, to) \ 1983 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b) 1984 1985 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ 1986 # define INSERT_JUMP2(op, loc, to, arg) \ 1987 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\ 1988 arg, b) 1989 1990 /* This is not an arbitrary limit: the arguments which represent offsets 1991 into the pattern are two bytes long. So if 2^16 bytes turns out to 1992 be too small, many things would have to change. */ 1993 /* Any other compiler which, like MSC, has allocation limit below 2^16 1994 bytes will have to use approach similar to what was done below for 1995 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up 1996 reallocating to 0 bytes. Such thing is not going to work too well. 1997 You have been warned!! */ 1998 # ifndef DEFINED_ONCE 1999 # if defined _MSC_VER && !defined WIN32 2000 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. 2001 The REALLOC define eliminates a flurry of conversion warnings, 2002 but is not required. */ 2003 # define MAX_BUF_SIZE 65500L 2004 # define REALLOC(p,s) realloc ((p), (size_t) (s)) 2005 # else 2006 # define MAX_BUF_SIZE (1L << 16) 2007 # define REALLOC(p,s) realloc ((p), (s)) 2008 # endif 2009 2010 /* Extend the buffer by twice its current size via realloc and 2011 reset the pointers that pointed into the old block to point to the 2012 correct places in the new one. If extending the buffer results in it 2013 being larger than MAX_BUF_SIZE, then flag memory exhausted. */ 2014 # if __BOUNDED_POINTERS__ 2015 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) 2016 # define MOVE_BUFFER_POINTER(P) \ 2017 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) 2018 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ 2019 else \ 2020 { \ 2021 SET_HIGH_BOUND (b); \ 2022 SET_HIGH_BOUND (begalt); \ 2023 if (fixup_alt_jump) \ 2024 SET_HIGH_BOUND (fixup_alt_jump); \ 2025 if (laststart) \ 2026 SET_HIGH_BOUND (laststart); \ 2027 if (pending_exact) \ 2028 SET_HIGH_BOUND (pending_exact); \ 2029 } 2030 # else 2031 # define MOVE_BUFFER_POINTER(P) (P) += incr 2032 # define ELSE_EXTEND_BUFFER_HIGH_BOUND 2033 # endif 2034 # endif /* not DEFINED_ONCE */ 2035 2036 # ifdef WCHAR 2037 # define EXTEND_BUFFER() \ 2038 do { \ 2039 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ 2040 int wchar_count; \ 2041 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \ 2042 return REG_ESIZE; \ 2043 bufp->allocated <<= 1; \ 2044 if (bufp->allocated > MAX_BUF_SIZE) \ 2045 bufp->allocated = MAX_BUF_SIZE; \ 2046 /* How many characters the new buffer can have? */ \ 2047 wchar_count = bufp->allocated / sizeof(UCHAR_T); \ 2048 if (wchar_count == 0) wchar_count = 1; \ 2049 /* Truncate the buffer to CHAR_T align. */ \ 2050 bufp->allocated = wchar_count * sizeof(UCHAR_T); \ 2051 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \ 2052 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \ 2053 if (COMPILED_BUFFER_VAR == NULL) \ 2054 return REG_ESPACE; \ 2055 /* If the buffer moved, move all the pointers into it. */ \ 2056 if (old_buffer != COMPILED_BUFFER_VAR) \ 2057 { \ 2058 PTR_INT_TYPE incr = COMPILED_BUFFER_VAR - old_buffer; \ 2059 MOVE_BUFFER_POINTER (b); \ 2060 MOVE_BUFFER_POINTER (begalt); \ 2061 if (fixup_alt_jump) \ 2062 MOVE_BUFFER_POINTER (fixup_alt_jump); \ 2063 if (laststart) \ 2064 MOVE_BUFFER_POINTER (laststart); \ 2065 if (pending_exact) \ 2066 MOVE_BUFFER_POINTER (pending_exact); \ 2067 } \ 2068 ELSE_EXTEND_BUFFER_HIGH_BOUND \ 2069 } while (0) 2070 # else /* BYTE */ 2071 # define EXTEND_BUFFER() \ 2072 do { \ 2073 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ 2074 if (bufp->allocated == MAX_BUF_SIZE) \ 2075 return REG_ESIZE; \ 2076 bufp->allocated <<= 1; \ 2077 if (bufp->allocated > MAX_BUF_SIZE) \ 2078 bufp->allocated = MAX_BUF_SIZE; \ 2079 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \ 2080 bufp->allocated); \ 2081 if (COMPILED_BUFFER_VAR == NULL) \ 2082 return REG_ESPACE; \ 2083 /* If the buffer moved, move all the pointers into it. */ \ 2084 if (old_buffer != COMPILED_BUFFER_VAR) \ 2085 { \ 2086 PTR_INT_TYPE incr = COMPILED_BUFFER_VAR - old_buffer; \ 2087 MOVE_BUFFER_POINTER (b); \ 2088 MOVE_BUFFER_POINTER (begalt); \ 2089 if (fixup_alt_jump) \ 2090 MOVE_BUFFER_POINTER (fixup_alt_jump); \ 2091 if (laststart) \ 2092 MOVE_BUFFER_POINTER (laststart); \ 2093 if (pending_exact) \ 2094 MOVE_BUFFER_POINTER (pending_exact); \ 2095 } \ 2096 ELSE_EXTEND_BUFFER_HIGH_BOUND \ 2097 } while (0) 2098 # endif /* WCHAR */ 2099 2100 # ifndef DEFINED_ONCE 2101 /* Since we have one byte reserved for the register number argument to 2102 {start,stop}_memory, the maximum number of groups we can report 2103 things about is what fits in that byte. */ 2104 # define MAX_REGNUM 255 2105 2106 /* But patterns can have more than `MAX_REGNUM' registers. We just 2107 ignore the excess. */ 2108 typedef unsigned regnum_t; 2109 2110 2111 /* Macros for the compile stack. */ 2112 2113 /* Since offsets can go either forwards or backwards, this type needs to 2114 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ 2115 /* int may be not enough when sizeof(int) == 2. */ 2116 typedef long pattern_offset_t; 2117 2118 typedef struct 2119 { 2120 pattern_offset_t begalt_offset; 2121 pattern_offset_t fixup_alt_jump; 2122 pattern_offset_t inner_group_offset; 2123 pattern_offset_t laststart_offset; 2124 regnum_t regnum; 2125 } compile_stack_elt_t; 2126 2127 2128 typedef struct 2129 { 2130 compile_stack_elt_t *stack; 2131 unsigned size; 2132 unsigned avail; /* Offset of next open position. */ 2133 } compile_stack_type; 2134 2135 2136 # define INIT_COMPILE_STACK_SIZE 32 2137 2138 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0) 2139 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) 2140 2141 /* The next available element. */ 2142 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) 2143 2144 # endif /* not DEFINED_ONCE */ 2145 2146 /* Set the bit for character C in a list. */ 2147 # ifndef DEFINED_ONCE 2148 # define SET_LIST_BIT(c) \ 2149 (b[((unsigned char) (c)) / BYTEWIDTH] \ 2150 |= 1 << (((unsigned char) c) % BYTEWIDTH)) 2151 # endif /* DEFINED_ONCE */ 2152 2153 /* Get the next unsigned number in the uncompiled pattern. */ 2154 # define GET_UNSIGNED_NUMBER(num) \ 2155 { \ 2156 while (p != pend) \ 2157 { \ 2158 PATFETCH (c); \ 2159 if (c < '0' || c > '9') \ 2160 break; \ 2161 if (num <= RE_DUP_MAX) \ 2162 { \ 2163 if (num < 0) \ 2164 num = 0; \ 2165 num = num * 10 + c - '0'; \ 2166 } \ 2167 } \ 2168 } 2169 2170 # ifndef DEFINED_ONCE 2171 # if defined _LIBC || WIDE_CHAR_SUPPORT 2172 /* The GNU C library provides support for user-defined character classes 2173 and the functions from ISO C amendement 1. */ 2174 # ifdef CHARCLASS_NAME_MAX 2175 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX 2176 # else 2177 /* This shouldn't happen but some implementation might still have this 2178 problem. Use a reasonable default value. */ 2179 # define CHAR_CLASS_MAX_LENGTH 256 2180 # endif 2181 2182 # ifdef _LIBC 2183 # define IS_CHAR_CLASS(string) __wctype (string) 2184 # else 2185 # define IS_CHAR_CLASS(string) wctype (string) 2186 # endif 2187 # else 2188 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ 2189 2190 # define IS_CHAR_CLASS(string) \ 2191 (STREQ (string, "alpha") || STREQ (string, "upper") \ 2192 || STREQ (string, "lower") || STREQ (string, "digit") \ 2193 || STREQ (string, "alnum") || STREQ (string, "xdigit") \ 2194 || STREQ (string, "space") || STREQ (string, "print") \ 2195 || STREQ (string, "punct") || STREQ (string, "graph") \ 2196 || STREQ (string, "cntrl") || STREQ (string, "blank")) 2197 # endif 2198 # endif /* DEFINED_ONCE */ 2199 2200 # ifndef MATCH_MAY_ALLOCATE 2201 2202 /* If we cannot allocate large objects within re_match_2_internal, 2203 we make the fail stack and register vectors global. 2204 The fail stack, we grow to the maximum size when a regexp 2205 is compiled. 2206 The register vectors, we adjust in size each time we 2207 compile a regexp, according to the number of registers it needs. */ 2208 2209 static PREFIX(fail_stack_type) fail_stack; 2210 2211 /* Size with which the following vectors are currently allocated. 2212 That is so we can make them bigger as needed, 2213 but never make them smaller. */ 2214 # ifdef DEFINED_ONCE 2215 static int regs_allocated_size; 2216 2217 static const char ** regstart, ** regend; 2218 static const char ** old_regstart, ** old_regend; 2219 static const char **best_regstart, **best_regend; 2220 static const char **reg_dummy; 2221 # endif /* DEFINED_ONCE */ 2222 2223 static PREFIX(register_info_type) *PREFIX(reg_info); 2224 static PREFIX(register_info_type) *PREFIX(reg_info_dummy); 2225 2226 /* Make the register vectors big enough for NUM_REGS registers, 2227 but don't make them smaller. */ 2228 2229 static void 2230 PREFIX(regex_grow_registers) (int num_regs) 2231 { 2232 if (num_regs > regs_allocated_size) 2233 { 2234 RETALLOC_IF (regstart, num_regs, const char *); 2235 RETALLOC_IF (regend, num_regs, const char *); 2236 RETALLOC_IF (old_regstart, num_regs, const char *); 2237 RETALLOC_IF (old_regend, num_regs, const char *); 2238 RETALLOC_IF (best_regstart, num_regs, const char *); 2239 RETALLOC_IF (best_regend, num_regs, const char *); 2240 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type)); 2241 RETALLOC_IF (reg_dummy, num_regs, const char *); 2242 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type)); 2243 2244 regs_allocated_size = num_regs; 2245 } 2246 } 2247 2248 # endif /* not MATCH_MAY_ALLOCATE */ 2249 2250 # ifndef DEFINED_ONCE 2251 static boolean group_in_compile_stack (compile_stack_type compile_stack, 2252 regnum_t regnum); 2253 # endif /* not DEFINED_ONCE */ 2254 2255 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. 2256 Returns one of error codes defined in `regex.h', or zero for success. 2257 2258 Assumes the `allocated' (and perhaps `buffer') and `translate' 2259 fields are set in BUFP on entry. 2260 2261 If it succeeds, results are put in BUFP (if it returns an error, the 2262 contents of BUFP are undefined): 2263 `buffer' is the compiled pattern; 2264 `syntax' is set to SYNTAX; 2265 `used' is set to the length of the compiled pattern; 2266 `fastmap_accurate' is zero; 2267 `re_nsub' is the number of subexpressions in PATTERN; 2268 `not_bol' and `not_eol' are zero; 2269 2270 The `fastmap' and `newline_anchor' fields are neither 2271 examined nor set. */ 2272 2273 /* Return, freeing storage we allocated. */ 2274 # ifdef WCHAR 2275 # define FREE_STACK_RETURN(value) \ 2276 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value) 2277 # else 2278 # define FREE_STACK_RETURN(value) \ 2279 return (free (compile_stack.stack), value) 2280 # endif /* WCHAR */ 2281 2282 static reg_errcode_t 2283 PREFIX(regex_compile) (const char *ARG_PREFIX(pattern), 2284 size_t ARG_PREFIX(size), reg_syntax_t syntax, 2285 struct re_pattern_buffer *bufp) 2286 { 2287 /* We fetch characters from PATTERN here. Even though PATTERN is 2288 `char *' (i.e., signed), we declare these variables as unsigned, so 2289 they can be reliably used as array indices. */ 2290 register UCHAR_T c, c1; 2291 2292 #ifdef WCHAR 2293 /* A temporary space to keep wchar_t pattern and compiled pattern. */ 2294 CHAR_T *pattern, *COMPILED_BUFFER_VAR; 2295 size_t size; 2296 /* offset buffer for optimization. See convert_mbs_to_wc. */ 2297 int *mbs_offset = NULL; 2298 /* It hold whether each wchar_t is binary data or not. */ 2299 char *is_binary = NULL; 2300 /* A flag whether exactn is handling binary data or not. */ 2301 char is_exactn_bin = FALSE; 2302 #endif /* WCHAR */ 2303 2304 /* A random temporary spot in PATTERN. */ 2305 const CHAR_T *p1; 2306 2307 /* Points to the end of the buffer, where we should append. */ 2308 register UCHAR_T *b; 2309 2310 /* Keeps track of unclosed groups. */ 2311 compile_stack_type compile_stack; 2312 2313 /* Points to the current (ending) position in the pattern. */ 2314 #ifdef WCHAR 2315 const CHAR_T *p; 2316 const CHAR_T *pend; 2317 #else /* BYTE */ 2318 const CHAR_T *p = pattern; 2319 const CHAR_T *pend = pattern + size; 2320 #endif /* WCHAR */ 2321 2322 /* How to translate the characters in the pattern. */ 2323 RE_TRANSLATE_TYPE translate = bufp->translate; 2324 2325 /* Address of the count-byte of the most recently inserted `exactn' 2326 command. This makes it possible to tell if a new exact-match 2327 character can be added to that command or if the character requires 2328 a new `exactn' command. */ 2329 UCHAR_T *pending_exact = 0; 2330 2331 /* Address of start of the most recently finished expression. 2332 This tells, e.g., postfix * where to find the start of its 2333 operand. Reset at the beginning of groups and alternatives. */ 2334 UCHAR_T *laststart = 0; 2335 2336 /* Address of beginning of regexp, or inside of last group. */ 2337 UCHAR_T *begalt; 2338 2339 /* Address of the place where a forward jump should go to the end of 2340 the containing expression. Each alternative of an `or' -- except the 2341 last -- ends with a forward jump of this sort. */ 2342 UCHAR_T *fixup_alt_jump = 0; 2343 2344 /* Counts open-groups as they are encountered. Remembered for the 2345 matching close-group on the compile stack, so the same register 2346 number is put in the stop_memory as the start_memory. */ 2347 regnum_t regnum = 0; 2348 2349 #ifdef WCHAR 2350 /* Initialize the wchar_t PATTERN and offset_buffer. */ 2351 p = pend = pattern = TALLOC(csize + 1, CHAR_T); 2352 mbs_offset = TALLOC(csize + 1, int); 2353 is_binary = TALLOC(csize + 1, char); 2354 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL) 2355 { 2356 free(pattern); 2357 free(mbs_offset); 2358 free(is_binary); 2359 return REG_ESPACE; 2360 } 2361 pattern[csize] = L'\0'; /* sentinel */ 2362 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary); 2363 pend = p + size; 2364 if (size < 0) 2365 { 2366 free(pattern); 2367 free(mbs_offset); 2368 free(is_binary); 2369 return REG_BADPAT; 2370 } 2371 #endif 2372 2373 #ifdef DEBUG 2374 DEBUG_PRINT1 ("\nCompiling pattern: "); 2375 if (debug) 2376 { 2377 unsigned debug_count; 2378 2379 for (debug_count = 0; debug_count < size; debug_count++) 2380 PUT_CHAR (pattern[debug_count]); 2381 putchar ('\n'); 2382 } 2383 #endif /* DEBUG */ 2384 2385 /* Initialize the compile stack. */ 2386 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); 2387 if (compile_stack.stack == NULL) 2388 { 2389 #ifdef WCHAR 2390 free(pattern); 2391 free(mbs_offset); 2392 free(is_binary); 2393 #endif 2394 return REG_ESPACE; 2395 } 2396 2397 compile_stack.size = INIT_COMPILE_STACK_SIZE; 2398 compile_stack.avail = 0; 2399 2400 /* Initialize the pattern buffer. */ 2401 bufp->syntax = syntax; 2402 bufp->fastmap_accurate = 0; 2403 bufp->not_bol = bufp->not_eol = 0; 2404 2405 /* Set `used' to zero, so that if we return an error, the pattern 2406 printer (for debugging) will think there's no pattern. We reset it 2407 at the end. */ 2408 bufp->used = 0; 2409 2410 /* Always count groups, whether or not bufp->no_sub is set. */ 2411 bufp->re_nsub = 0; 2412 2413 #if !defined emacs && !defined SYNTAX_TABLE 2414 /* Initialize the syntax table. */ 2415 init_syntax_once (); 2416 #endif 2417 2418 if (bufp->allocated == 0) 2419 { 2420 if (bufp->buffer) 2421 { /* If zero allocated, but buffer is non-null, try to realloc 2422 enough space. This loses if buffer's address is bogus, but 2423 that is the user's responsibility. */ 2424 #ifdef WCHAR 2425 /* Free bufp->buffer and allocate an array for wchar_t pattern 2426 buffer. */ 2427 free(bufp->buffer); 2428 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T), 2429 UCHAR_T); 2430 #else 2431 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T); 2432 #endif /* WCHAR */ 2433 } 2434 else 2435 { /* Caller did not allocate a buffer. Do it for them. */ 2436 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T), 2437 UCHAR_T); 2438 } 2439 2440 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE); 2441 #ifdef WCHAR 2442 bufp->buffer = (char*)COMPILED_BUFFER_VAR; 2443 #endif /* WCHAR */ 2444 bufp->allocated = INIT_BUF_SIZE; 2445 } 2446 #ifdef WCHAR 2447 else 2448 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer; 2449 #endif 2450 2451 begalt = b = COMPILED_BUFFER_VAR; 2452 2453 /* Loop through the uncompiled pattern until we're at the end. */ 2454 while (p != pend) 2455 { 2456 PATFETCH (c); 2457 2458 switch (c) 2459 { 2460 case '^': 2461 { 2462 if ( /* If at start of pattern, it's an operator. */ 2463 p == pattern + 1 2464 /* If context independent, it's an operator. */ 2465 || syntax & RE_CONTEXT_INDEP_ANCHORS 2466 /* Otherwise, depends on what's come before. */ 2467 || PREFIX(at_begline_loc_p) (pattern, p, syntax)) 2468 BUF_PUSH (begline); 2469 else 2470 goto normal_char; 2471 } 2472 break; 2473 2474 2475 case '$': 2476 { 2477 if ( /* If at end of pattern, it's an operator. */ 2478 p == pend 2479 /* If context independent, it's an operator. */ 2480 || syntax & RE_CONTEXT_INDEP_ANCHORS 2481 /* Otherwise, depends on what's next. */ 2482 || PREFIX(at_endline_loc_p) (p, pend, syntax)) 2483 BUF_PUSH (endline); 2484 else 2485 goto normal_char; 2486 } 2487 break; 2488 2489 2490 case '+': 2491 case '?': 2492 if ((syntax & RE_BK_PLUS_QM) 2493 || (syntax & RE_LIMITED_OPS)) 2494 goto normal_char; 2495 /* Fall through. */ 2496 handle_plus: 2497 case '*': 2498 /* If there is no previous pattern... */ 2499 if (!laststart) 2500 { 2501 if (syntax & RE_CONTEXT_INVALID_OPS) 2502 FREE_STACK_RETURN (REG_BADRPT); 2503 else if (!(syntax & RE_CONTEXT_INDEP_OPS)) 2504 goto normal_char; 2505 } 2506 2507 { 2508 /* Are we optimizing this jump? */ 2509 boolean keep_string_p = false; 2510 2511 /* 1 means zero (many) matches is allowed. */ 2512 char zero_times_ok = 0, many_times_ok = 0; 2513 2514 /* If there is a sequence of repetition chars, collapse it 2515 down to just one (the right one). We can't combine 2516 interval operators with these because of, e.g., `a{2}*', 2517 which should only match an even number of `a's. */ 2518 2519 for (;;) 2520 { 2521 zero_times_ok |= c != '+'; 2522 many_times_ok |= c != '?'; 2523 2524 if (p == pend) 2525 break; 2526 2527 PATFETCH (c); 2528 2529 if (c == '*' 2530 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) 2531 ; 2532 2533 else if (syntax & RE_BK_PLUS_QM && c == '\\') 2534 { 2535 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); 2536 2537 PATFETCH (c1); 2538 if (!(c1 == '+' || c1 == '?')) 2539 { 2540 PATUNFETCH; 2541 PATUNFETCH; 2542 break; 2543 } 2544 2545 c = c1; 2546 } 2547 else 2548 { 2549 PATUNFETCH; 2550 break; 2551 } 2552 2553 /* If we get here, we found another repeat character. */ 2554 } 2555 2556 /* Star, etc. applied to an empty pattern is equivalent 2557 to an empty pattern. */ 2558 if (!laststart) 2559 break; 2560 2561 /* Now we know whether or not zero matches is allowed 2562 and also whether or not two or more matches is allowed. */ 2563 if (many_times_ok) 2564 { /* More than one repetition is allowed, so put in at the 2565 end a backward relative jump from `b' to before the next 2566 jump we're going to put in below (which jumps from 2567 laststart to after this jump). 2568 2569 But if we are at the `*' in the exact sequence `.*\n', 2570 insert an unconditional jump backwards to the ., 2571 instead of the beginning of the loop. This way we only 2572 push a failure point once, instead of every time 2573 through the loop. */ 2574 assert (p - 1 > pattern); 2575 2576 /* Allocate the space for the jump. */ 2577 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 2578 2579 /* We know we are not at the first character of the pattern, 2580 because laststart was nonzero. And we've already 2581 incremented `p', by the way, to be the character after 2582 the `*'. Do we have to do something analogous here 2583 for null bytes, because of RE_DOT_NOT_NULL? */ 2584 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') 2585 && zero_times_ok 2586 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') 2587 && !(syntax & RE_DOT_NEWLINE)) 2588 { /* We have .*\n. */ 2589 STORE_JUMP (jump, b, laststart); 2590 keep_string_p = true; 2591 } 2592 else 2593 /* Anything else. */ 2594 STORE_JUMP (maybe_pop_jump, b, laststart - 2595 (1 + OFFSET_ADDRESS_SIZE)); 2596 2597 /* We've added more stuff to the buffer. */ 2598 b += 1 + OFFSET_ADDRESS_SIZE; 2599 } 2600 2601 /* On failure, jump from laststart to b + 3, which will be the 2602 end of the buffer after this jump is inserted. */ 2603 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of 2604 'b + 3'. */ 2605 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 2606 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump 2607 : on_failure_jump, 2608 laststart, b + 1 + OFFSET_ADDRESS_SIZE); 2609 pending_exact = 0; 2610 b += 1 + OFFSET_ADDRESS_SIZE; 2611 2612 if (!zero_times_ok) 2613 { 2614 /* At least one repetition is required, so insert a 2615 `dummy_failure_jump' before the initial 2616 `on_failure_jump' instruction of the loop. This 2617 effects a skip over that instruction the first time 2618 we hit that loop. */ 2619 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 2620 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 2621 2 + 2 * OFFSET_ADDRESS_SIZE); 2622 b += 1 + OFFSET_ADDRESS_SIZE; 2623 } 2624 } 2625 break; 2626 2627 2628 case '.': 2629 laststart = b; 2630 BUF_PUSH (anychar); 2631 break; 2632 2633 2634 case '[': 2635 { 2636 boolean had_char_class = false; 2637 #ifdef WCHAR 2638 CHAR_T range_start = 0xffffffff; 2639 #else 2640 unsigned int range_start = 0xffffffff; 2641 #endif 2642 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2643 2644 #ifdef WCHAR 2645 /* We assume a charset(_not) structure as a wchar_t array. 2646 charset[0] = (re_opcode_t) charset(_not) 2647 charset[1] = l (= length of char_classes) 2648 charset[2] = m (= length of collating_symbols) 2649 charset[3] = n (= length of equivalence_classes) 2650 charset[4] = o (= length of char_ranges) 2651 charset[5] = p (= length of chars) 2652 2653 charset[6] = char_class (wctype_t) 2654 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t) 2655 ... 2656 charset[l+5] = char_class (wctype_t) 2657 2658 charset[l+6] = collating_symbol (wchar_t) 2659 ... 2660 charset[l+m+5] = collating_symbol (wchar_t) 2661 ifdef _LIBC we use the index if 2662 _NL_COLLATE_SYMB_EXTRAMB instead of 2663 wchar_t string. 2664 2665 charset[l+m+6] = equivalence_classes (wchar_t) 2666 ... 2667 charset[l+m+n+5] = equivalence_classes (wchar_t) 2668 ifdef _LIBC we use the index in 2669 _NL_COLLATE_WEIGHT instead of 2670 wchar_t string. 2671 2672 charset[l+m+n+6] = range_start 2673 charset[l+m+n+7] = range_end 2674 ... 2675 charset[l+m+n+2o+4] = range_start 2676 charset[l+m+n+2o+5] = range_end 2677 ifdef _LIBC we use the value looked up 2678 in _NL_COLLATE_COLLSEQ instead of 2679 wchar_t character. 2680 2681 charset[l+m+n+2o+6] = char 2682 ... 2683 charset[l+m+n+2o+p+5] = char 2684 2685 */ 2686 2687 /* We need at least 6 spaces: the opcode, the length of 2688 char_classes, the length of collating_symbols, the length of 2689 equivalence_classes, the length of char_ranges, the length of 2690 chars. */ 2691 GET_BUFFER_SPACE (6); 2692 2693 /* Save b as laststart. And We use laststart as the pointer 2694 to the first element of the charset here. 2695 In other words, laststart[i] indicates charset[i]. */ 2696 laststart = b; 2697 2698 /* We test `*p == '^' twice, instead of using an if 2699 statement, so we only need one BUF_PUSH. */ 2700 BUF_PUSH (*p == '^' ? charset_not : charset); 2701 if (*p == '^') 2702 p++; 2703 2704 /* Push the length of char_classes, the length of 2705 collating_symbols, the length of equivalence_classes, the 2706 length of char_ranges and the length of chars. */ 2707 BUF_PUSH_3 (0, 0, 0); 2708 BUF_PUSH_2 (0, 0); 2709 2710 /* Remember the first position in the bracket expression. */ 2711 p1 = p; 2712 2713 /* charset_not matches newline according to a syntax bit. */ 2714 if ((re_opcode_t) b[-6] == charset_not 2715 && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) 2716 { 2717 BUF_PUSH('\n'); 2718 laststart[5]++; /* Update the length of characters */ 2719 } 2720 2721 /* Read in characters and ranges, setting map bits. */ 2722 for (;;) 2723 { 2724 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2725 2726 PATFETCH (c); 2727 2728 /* \ might escape characters inside [...] and [^...]. */ 2729 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') 2730 { 2731 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); 2732 2733 PATFETCH (c1); 2734 BUF_PUSH(c1); 2735 laststart[5]++; /* Update the length of chars */ 2736 range_start = c1; 2737 continue; 2738 } 2739 2740 /* Could be the end of the bracket expression. If it's 2741 not (i.e., when the bracket expression is `[]' so 2742 far), the ']' character bit gets set way below. */ 2743 if (c == ']' && p != p1 + 1) 2744 break; 2745 2746 /* Look ahead to see if it's a range when the last thing 2747 was a character class. */ 2748 if (had_char_class && c == '-' && *p != ']') 2749 FREE_STACK_RETURN (REG_ERANGE); 2750 2751 /* Look ahead to see if it's a range when the last thing 2752 was a character: if this is a hyphen not at the 2753 beginning or the end of a list, then it's the range 2754 operator. */ 2755 if (c == '-' 2756 && !(p - 2 >= pattern && p[-2] == '[') 2757 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') 2758 && *p != ']') 2759 { 2760 reg_errcode_t ret; 2761 /* Allocate the space for range_start and range_end. */ 2762 GET_BUFFER_SPACE (2); 2763 /* Update the pointer to indicate end of buffer. */ 2764 b += 2; 2765 ret = wcs_compile_range (range_start, &p, pend, translate, 2766 syntax, b, laststart); 2767 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); 2768 range_start = 0xffffffff; 2769 } 2770 else if (p[0] == '-' && p[1] != ']') 2771 { /* This handles ranges made up of characters only. */ 2772 reg_errcode_t ret; 2773 2774 /* Move past the `-'. */ 2775 PATFETCH (c1); 2776 /* Allocate the space for range_start and range_end. */ 2777 GET_BUFFER_SPACE (2); 2778 /* Update the pointer to indicate end of buffer. */ 2779 b += 2; 2780 ret = wcs_compile_range (c, &p, pend, translate, syntax, b, 2781 laststart); 2782 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); 2783 range_start = 0xffffffff; 2784 } 2785 2786 /* See if we're at the beginning of a possible character 2787 class. */ 2788 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') 2789 { /* Leave room for the null. */ 2790 char str[CHAR_CLASS_MAX_LENGTH + 1]; 2791 2792 PATFETCH (c); 2793 c1 = 0; 2794 2795 /* If pattern is `[[:'. */ 2796 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2797 2798 for (;;) 2799 { 2800 PATFETCH (c); 2801 if ((c == ':' && *p == ']') || p == pend) 2802 break; 2803 if (c1 < CHAR_CLASS_MAX_LENGTH) 2804 str[c1++] = c; 2805 else 2806 /* This is in any case an invalid class name. */ 2807 str[0] = '\0'; 2808 } 2809 str[c1] = '\0'; 2810 2811 /* If isn't a word bracketed by `[:' and `:]': 2812 undo the ending character, the letters, and leave 2813 the leading `:' and `[' (but store them as character). */ 2814 if (c == ':' && *p == ']') 2815 { 2816 wctype_t wt; 2817 uintptr_t alignedp; 2818 2819 /* Query the character class as wctype_t. */ 2820 wt = IS_CHAR_CLASS (str); 2821 if (wt == 0) 2822 FREE_STACK_RETURN (REG_ECTYPE); 2823 2824 /* Throw away the ] at the end of the character 2825 class. */ 2826 PATFETCH (c); 2827 2828 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2829 2830 /* Allocate the space for character class. */ 2831 GET_BUFFER_SPACE(CHAR_CLASS_SIZE); 2832 /* Update the pointer to indicate end of buffer. */ 2833 b += CHAR_CLASS_SIZE; 2834 /* Move data which follow character classes 2835 not to violate the data. */ 2836 insert_space(CHAR_CLASS_SIZE, 2837 laststart + 6 + laststart[1], 2838 b - 1); 2839 alignedp = ((uintptr_t)(laststart + 6 + laststart[1]) 2840 + __alignof__(wctype_t) - 1) 2841 & ~(uintptr_t)(__alignof__(wctype_t) - 1); 2842 /* Store the character class. */ 2843 *((wctype_t*)alignedp) = wt; 2844 /* Update length of char_classes */ 2845 laststart[1] += CHAR_CLASS_SIZE; 2846 2847 had_char_class = true; 2848 } 2849 else 2850 { 2851 c1++; 2852 while (c1--) 2853 PATUNFETCH; 2854 BUF_PUSH ('['); 2855 BUF_PUSH (':'); 2856 laststart[5] += 2; /* Update the length of characters */ 2857 range_start = ':'; 2858 had_char_class = false; 2859 } 2860 } 2861 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '=' 2862 || *p == '.')) 2863 { 2864 CHAR_T str[128]; /* Should be large enough. */ 2865 CHAR_T delim = *p; /* '=' or '.' */ 2866 # ifdef _LIBC 2867 uint32_t nrules = 2868 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 2869 # endif 2870 PATFETCH (c); 2871 c1 = 0; 2872 2873 /* If pattern is `[[=' or '[[.'. */ 2874 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 2875 2876 for (;;) 2877 { 2878 PATFETCH (c); 2879 if ((c == delim && *p == ']') || p == pend) 2880 break; 2881 if (c1 < sizeof (str) - 1) 2882 str[c1++] = c; 2883 else 2884 /* This is in any case an invalid class name. */ 2885 str[0] = '\0'; 2886 } 2887 str[c1] = '\0'; 2888 2889 if (c == delim && *p == ']' && str[0] != '\0') 2890 { 2891 unsigned int i, offset; 2892 /* If we have no collation data we use the default 2893 collation in which each character is in a class 2894 by itself. It also means that ASCII is the 2895 character set and therefore we cannot have character 2896 with more than one byte in the multibyte 2897 representation. */ 2898 2899 /* If not defined _LIBC, we push the name and 2900 `\0' for the sake of matching performance. */ 2901 int datasize = c1 + 1; 2902 2903 # ifdef _LIBC 2904 int32_t idx = 0; 2905 if (nrules == 0) 2906 # endif 2907 { 2908 if (c1 != 1) 2909 FREE_STACK_RETURN (REG_ECOLLATE); 2910 } 2911 # ifdef _LIBC 2912 else 2913 { 2914 const int32_t *table; 2915 const int32_t *weights; 2916 const int32_t *extra; 2917 const int32_t *indirect; 2918 wint_t *cp; 2919 2920 /* This #include defines a local function! */ 2921 # include <locale/weightwc.h> 2922 2923 if(delim == '=') 2924 { 2925 /* We push the index for equivalence class. */ 2926 cp = (wint_t*)str; 2927 2928 table = (const int32_t *) 2929 _NL_CURRENT (LC_COLLATE, 2930 _NL_COLLATE_TABLEWC); 2931 weights = (const int32_t *) 2932 _NL_CURRENT (LC_COLLATE, 2933 _NL_COLLATE_WEIGHTWC); 2934 extra = (const int32_t *) 2935 _NL_CURRENT (LC_COLLATE, 2936 _NL_COLLATE_EXTRAWC); 2937 indirect = (const int32_t *) 2938 _NL_CURRENT (LC_COLLATE, 2939 _NL_COLLATE_INDIRECTWC); 2940 2941 idx = findidx ((const wint_t**)&cp); 2942 if (idx == 0 || cp < (wint_t*) str + c1) 2943 /* This is no valid character. */ 2944 FREE_STACK_RETURN (REG_ECOLLATE); 2945 2946 str[0] = (wchar_t)idx; 2947 } 2948 else /* delim == '.' */ 2949 { 2950 /* We push collation sequence value 2951 for collating symbol. */ 2952 int32_t table_size; 2953 const int32_t *symb_table; 2954 const unsigned char *extra; 2955 int32_t idx; 2956 int32_t elem; 2957 int32_t second; 2958 int32_t hash; 2959 char char_str[c1]; 2960 2961 /* We have to convert the name to a single-byte 2962 string. This is possible since the names 2963 consist of ASCII characters and the internal 2964 representation is UCS4. */ 2965 for (i = 0; i < c1; ++i) 2966 char_str[i] = str[i]; 2967 2968 table_size = 2969 _NL_CURRENT_WORD (LC_COLLATE, 2970 _NL_COLLATE_SYMB_HASH_SIZEMB); 2971 symb_table = (const int32_t *) 2972 _NL_CURRENT (LC_COLLATE, 2973 _NL_COLLATE_SYMB_TABLEMB); 2974 extra = (const unsigned char *) 2975 _NL_CURRENT (LC_COLLATE, 2976 _NL_COLLATE_SYMB_EXTRAMB); 2977 2978 /* Locate the character in the hashing table. */ 2979 hash = elem_hash (char_str, c1); 2980 2981 idx = 0; 2982 elem = hash % table_size; 2983 second = hash % (table_size - 2); 2984 while (symb_table[2 * elem] != 0) 2985 { 2986 /* First compare the hashing value. */ 2987 if (symb_table[2 * elem] == hash 2988 && c1 == extra[symb_table[2 * elem + 1]] 2989 && memcmp (char_str, 2990 &extra[symb_table[2 * elem + 1] 2991 + 1], c1) == 0) 2992 { 2993 /* Yep, this is the entry. */ 2994 idx = symb_table[2 * elem + 1]; 2995 idx += 1 + extra[idx]; 2996 break; 2997 } 2998 2999 /* Next entry. */ 3000 elem += second; 3001 } 3002 3003 if (symb_table[2 * elem] != 0) 3004 { 3005 /* Compute the index of the byte sequence 3006 in the table. */ 3007 idx += 1 + extra[idx]; 3008 /* Adjust for the alignment. */ 3009 idx = (idx + 3) & ~3; 3010 3011 str[0] = (wchar_t) idx + 4; 3012 } 3013 else if (symb_table[2 * elem] == 0 && c1 == 1) 3014 { 3015 /* No valid character. Match it as a 3016 single byte character. */ 3017 had_char_class = false; 3018 BUF_PUSH(str[0]); 3019 /* Update the length of characters */ 3020 laststart[5]++; 3021 range_start = str[0]; 3022 3023 /* Throw away the ] at the end of the 3024 collating symbol. */ 3025 PATFETCH (c); 3026 /* exit from the switch block. */ 3027 continue; 3028 } 3029 else 3030 FREE_STACK_RETURN (REG_ECOLLATE); 3031 } 3032 datasize = 1; 3033 } 3034 # endif 3035 /* Throw away the ] at the end of the equivalence 3036 class (or collating symbol). */ 3037 PATFETCH (c); 3038 3039 /* Allocate the space for the equivalence class 3040 (or collating symbol) (and '\0' if needed). */ 3041 GET_BUFFER_SPACE(datasize); 3042 /* Update the pointer to indicate end of buffer. */ 3043 b += datasize; 3044 3045 if (delim == '=') 3046 { /* equivalence class */ 3047 /* Calculate the offset of char_ranges, 3048 which is next to equivalence_classes. */ 3049 offset = laststart[1] + laststart[2] 3050 + laststart[3] +6; 3051 /* Insert space. */ 3052 insert_space(datasize, laststart + offset, b - 1); 3053 3054 /* Write the equivalence_class and \0. */ 3055 for (i = 0 ; i < datasize ; i++) 3056 laststart[offset + i] = str[i]; 3057 3058 /* Update the length of equivalence_classes. */ 3059 laststart[3] += datasize; 3060 had_char_class = true; 3061 } 3062 else /* delim == '.' */ 3063 { /* collating symbol */ 3064 /* Calculate the offset of the equivalence_classes, 3065 which is next to collating_symbols. */ 3066 offset = laststart[1] + laststart[2] + 6; 3067 /* Insert space and write the collationg_symbol 3068 and \0. */ 3069 insert_space(datasize, laststart + offset, b-1); 3070 for (i = 0 ; i < datasize ; i++) 3071 laststart[offset + i] = str[i]; 3072 3073 /* In re_match_2_internal if range_start < -1, we 3074 assume -range_start is the offset of the 3075 collating symbol which is specified as 3076 the character of the range start. So we assign 3077 -(laststart[1] + laststart[2] + 6) to 3078 range_start. */ 3079 range_start = -(laststart[1] + laststart[2] + 6); 3080 /* Update the length of collating_symbol. */ 3081 laststart[2] += datasize; 3082 had_char_class = false; 3083 } 3084 } 3085 else 3086 { 3087 c1++; 3088 while (c1--) 3089 PATUNFETCH; 3090 BUF_PUSH ('['); 3091 BUF_PUSH (delim); 3092 laststart[5] += 2; /* Update the length of characters */ 3093 range_start = delim; 3094 had_char_class = false; 3095 } 3096 } 3097 else 3098 { 3099 had_char_class = false; 3100 BUF_PUSH(c); 3101 laststart[5]++; /* Update the length of characters */ 3102 range_start = c; 3103 } 3104 } 3105 3106 #else /* BYTE */ 3107 /* Ensure that we have enough space to push a charset: the 3108 opcode, the length count, and the bitset; 34 bytes in all. */ 3109 GET_BUFFER_SPACE (34); 3110 3111 laststart = b; 3112 3113 /* We test `*p == '^' twice, instead of using an if 3114 statement, so we only need one BUF_PUSH. */ 3115 BUF_PUSH (*p == '^' ? charset_not : charset); 3116 if (*p == '^') 3117 p++; 3118 3119 /* Remember the first position in the bracket expression. */ 3120 p1 = p; 3121 3122 /* Push the number of bytes in the bitmap. */ 3123 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); 3124 3125 /* Clear the whole map. */ 3126 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); 3127 3128 /* charset_not matches newline according to a syntax bit. */ 3129 if ((re_opcode_t) b[-2] == charset_not 3130 && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) 3131 SET_LIST_BIT ('\n'); 3132 3133 /* Read in characters and ranges, setting map bits. */ 3134 for (;;) 3135 { 3136 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3137 3138 PATFETCH (c); 3139 3140 /* \ might escape characters inside [...] and [^...]. */ 3141 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') 3142 { 3143 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); 3144 3145 PATFETCH (c1); 3146 SET_LIST_BIT (c1); 3147 range_start = c1; 3148 continue; 3149 } 3150 3151 /* Could be the end of the bracket expression. If it's 3152 not (i.e., when the bracket expression is `[]' so 3153 far), the ']' character bit gets set way below. */ 3154 if (c == ']' && p != p1 + 1) 3155 break; 3156 3157 /* Look ahead to see if it's a range when the last thing 3158 was a character class. */ 3159 if (had_char_class && c == '-' && *p != ']') 3160 FREE_STACK_RETURN (REG_ERANGE); 3161 3162 /* Look ahead to see if it's a range when the last thing 3163 was a character: if this is a hyphen not at the 3164 beginning or the end of a list, then it's the range 3165 operator. */ 3166 if (c == '-' 3167 && !(p - 2 >= pattern && p[-2] == '[') 3168 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') 3169 && *p != ']') 3170 { 3171 reg_errcode_t ret 3172 = byte_compile_range (range_start, &p, pend, translate, 3173 syntax, b); 3174 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); 3175 range_start = 0xffffffff; 3176 } 3177 3178 else if (p[0] == '-' && p[1] != ']') 3179 { /* This handles ranges made up of characters only. */ 3180 reg_errcode_t ret; 3181 3182 /* Move past the `-'. */ 3183 PATFETCH (c1); 3184 3185 ret = byte_compile_range (c, &p, pend, translate, syntax, b); 3186 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); 3187 range_start = 0xffffffff; 3188 } 3189 3190 /* See if we're at the beginning of a possible character 3191 class. */ 3192 3193 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') 3194 { /* Leave room for the null. */ 3195 char str[CHAR_CLASS_MAX_LENGTH + 1]; 3196 3197 PATFETCH (c); 3198 c1 = 0; 3199 3200 /* If pattern is `[[:'. */ 3201 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3202 3203 for (;;) 3204 { 3205 PATFETCH (c); 3206 if ((c == ':' && *p == ']') || p == pend) 3207 break; 3208 if (c1 < CHAR_CLASS_MAX_LENGTH) 3209 str[c1++] = c; 3210 else 3211 /* This is in any case an invalid class name. */ 3212 str[0] = '\0'; 3213 } 3214 str[c1] = '\0'; 3215 3216 /* If isn't a word bracketed by `[:' and `:]': 3217 undo the ending character, the letters, and leave 3218 the leading `:' and `[' (but set bits for them). */ 3219 if (c == ':' && *p == ']') 3220 { 3221 # if defined _LIBC || WIDE_CHAR_SUPPORT 3222 boolean is_lower = STREQ (str, "lower"); 3223 boolean is_upper = STREQ (str, "upper"); 3224 wctype_t wt; 3225 int ch; 3226 3227 wt = IS_CHAR_CLASS (str); 3228 if (wt == 0) 3229 FREE_STACK_RETURN (REG_ECTYPE); 3230 3231 /* Throw away the ] at the end of the character 3232 class. */ 3233 PATFETCH (c); 3234 3235 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3236 3237 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) 3238 { 3239 # ifdef _LIBC 3240 if (__iswctype (__btowc (ch), wt)) 3241 SET_LIST_BIT (ch); 3242 # else 3243 if (iswctype (btowc (ch), wt)) 3244 SET_LIST_BIT (ch); 3245 # endif 3246 3247 if (translate && (is_upper || is_lower) 3248 && (ISUPPER (ch) || ISLOWER (ch))) 3249 SET_LIST_BIT (ch); 3250 } 3251 3252 had_char_class = true; 3253 # else 3254 int ch; 3255 boolean is_alnum = STREQ (str, "alnum"); 3256 boolean is_alpha = STREQ (str, "alpha"); 3257 boolean is_blank = STREQ (str, "blank"); 3258 boolean is_cntrl = STREQ (str, "cntrl"); 3259 boolean is_digit = STREQ (str, "digit"); 3260 boolean is_graph = STREQ (str, "graph"); 3261 boolean is_lower = STREQ (str, "lower"); 3262 boolean is_print = STREQ (str, "print"); 3263 boolean is_punct = STREQ (str, "punct"); 3264 boolean is_space = STREQ (str, "space"); 3265 boolean is_upper = STREQ (str, "upper"); 3266 boolean is_xdigit = STREQ (str, "xdigit"); 3267 3268 if (!IS_CHAR_CLASS (str)) 3269 FREE_STACK_RETURN (REG_ECTYPE); 3270 3271 /* Throw away the ] at the end of the character 3272 class. */ 3273 PATFETCH (c); 3274 3275 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3276 3277 for (ch = 0; ch < 1 << BYTEWIDTH; ch++) 3278 { 3279 /* This was split into 3 if's to 3280 avoid an arbitrary limit in some compiler. */ 3281 if ( (is_alnum && ISALNUM (ch)) 3282 || (is_alpha && ISALPHA (ch)) 3283 || (is_blank && ISBLANK (ch)) 3284 || (is_cntrl && ISCNTRL (ch))) 3285 SET_LIST_BIT (ch); 3286 if ( (is_digit && ISDIGIT (ch)) 3287 || (is_graph && ISGRAPH (ch)) 3288 || (is_lower && ISLOWER (ch)) 3289 || (is_print && ISPRINT (ch))) 3290 SET_LIST_BIT (ch); 3291 if ( (is_punct && ISPUNCT (ch)) 3292 || (is_space && ISSPACE (ch)) 3293 || (is_upper && ISUPPER (ch)) 3294 || (is_xdigit && ISXDIGIT (ch))) 3295 SET_LIST_BIT (ch); 3296 if ( translate && (is_upper || is_lower) 3297 && (ISUPPER (ch) || ISLOWER (ch))) 3298 SET_LIST_BIT (ch); 3299 } 3300 had_char_class = true; 3301 # endif /* libc || wctype.h */ 3302 } 3303 else 3304 { 3305 c1++; 3306 while (c1--) 3307 PATUNFETCH; 3308 SET_LIST_BIT ('['); 3309 SET_LIST_BIT (':'); 3310 range_start = ':'; 3311 had_char_class = false; 3312 } 3313 } 3314 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=') 3315 { 3316 unsigned char str[MB_LEN_MAX + 1]; 3317 # ifdef _LIBC 3318 uint32_t nrules = 3319 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 3320 # endif 3321 3322 PATFETCH (c); 3323 c1 = 0; 3324 3325 /* If pattern is `[[='. */ 3326 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3327 3328 for (;;) 3329 { 3330 PATFETCH (c); 3331 if ((c == '=' && *p == ']') || p == pend) 3332 break; 3333 if (c1 < MB_LEN_MAX) 3334 str[c1++] = c; 3335 else 3336 /* This is in any case an invalid class name. */ 3337 str[0] = '\0'; 3338 } 3339 str[c1] = '\0'; 3340 3341 if (c == '=' && *p == ']' && str[0] != '\0') 3342 { 3343 /* If we have no collation data we use the default 3344 collation in which each character is in a class 3345 by itself. It also means that ASCII is the 3346 character set and therefore we cannot have character 3347 with more than one byte in the multibyte 3348 representation. */ 3349 # ifdef _LIBC 3350 if (nrules == 0) 3351 # endif 3352 { 3353 if (c1 != 1) 3354 FREE_STACK_RETURN (REG_ECOLLATE); 3355 3356 /* Throw away the ] at the end of the equivalence 3357 class. */ 3358 PATFETCH (c); 3359 3360 /* Set the bit for the character. */ 3361 SET_LIST_BIT (str[0]); 3362 } 3363 # ifdef _LIBC 3364 else 3365 { 3366 /* Try to match the byte sequence in `str' against 3367 those known to the collate implementation. 3368 First find out whether the bytes in `str' are 3369 actually from exactly one character. */ 3370 const int32_t *table; 3371 const unsigned char *weights; 3372 const unsigned char *extra; 3373 const int32_t *indirect; 3374 int32_t idx; 3375 const unsigned char *cp = str; 3376 int ch; 3377 3378 /* This #include defines a local function! */ 3379 # include <locale/weight.h> 3380 3381 table = (const int32_t *) 3382 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB); 3383 weights = (const unsigned char *) 3384 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB); 3385 extra = (const unsigned char *) 3386 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB); 3387 indirect = (const int32_t *) 3388 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB); 3389 3390 idx = findidx (&cp); 3391 if (idx == 0 || cp < str + c1) 3392 /* This is no valid character. */ 3393 FREE_STACK_RETURN (REG_ECOLLATE); 3394 3395 /* Throw away the ] at the end of the equivalence 3396 class. */ 3397 PATFETCH (c); 3398 3399 /* Now we have to go through the whole table 3400 and find all characters which have the same 3401 first level weight. 3402 3403 XXX Note that this is not entirely correct. 3404 we would have to match multibyte sequences 3405 but this is not possible with the current 3406 implementation. */ 3407 for (ch = 1; ch < 256; ++ch) 3408 /* XXX This test would have to be changed if we 3409 would allow matching multibyte sequences. */ 3410 if (table[ch] > 0) 3411 { 3412 int32_t idx2 = table[ch]; 3413 size_t len = weights[idx2]; 3414 3415 /* Test whether the lenghts match. */ 3416 if (weights[idx] == len) 3417 { 3418 /* They do. New compare the bytes of 3419 the weight. */ 3420 size_t cnt = 0; 3421 3422 while (cnt < len 3423 && (weights[idx + 1 + cnt] 3424 == weights[idx2 + 1 + cnt])) 3425 ++cnt; 3426 3427 if (cnt == len) 3428 /* They match. Mark the character as 3429 acceptable. */ 3430 SET_LIST_BIT (ch); 3431 } 3432 } 3433 } 3434 # endif 3435 had_char_class = true; 3436 } 3437 else 3438 { 3439 c1++; 3440 while (c1--) 3441 PATUNFETCH; 3442 SET_LIST_BIT ('['); 3443 SET_LIST_BIT ('='); 3444 range_start = '='; 3445 had_char_class = false; 3446 } 3447 } 3448 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.') 3449 { 3450 unsigned char str[128]; /* Should be large enough. */ 3451 # ifdef _LIBC 3452 uint32_t nrules = 3453 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 3454 # endif 3455 3456 PATFETCH (c); 3457 c1 = 0; 3458 3459 /* If pattern is `[[.'. */ 3460 if (p == pend) FREE_STACK_RETURN (REG_EBRACK); 3461 3462 for (;;) 3463 { 3464 PATFETCH (c); 3465 if ((c == '.' && *p == ']') || p == pend) 3466 break; 3467 if (c1 < sizeof (str)) 3468 str[c1++] = c; 3469 else 3470 /* This is in any case an invalid class name. */ 3471 str[0] = '\0'; 3472 } 3473 str[c1] = '\0'; 3474 3475 if (c == '.' && *p == ']' && str[0] != '\0') 3476 { 3477 /* If we have no collation data we use the default 3478 collation in which each character is the name 3479 for its own class which contains only the one 3480 character. It also means that ASCII is the 3481 character set and therefore we cannot have character 3482 with more than one byte in the multibyte 3483 representation. */ 3484 # ifdef _LIBC 3485 if (nrules == 0) 3486 # endif 3487 { 3488 if (c1 != 1) 3489 FREE_STACK_RETURN (REG_ECOLLATE); 3490 3491 /* Throw away the ] at the end of the equivalence 3492 class. */ 3493 PATFETCH (c); 3494 3495 /* Set the bit for the character. */ 3496 SET_LIST_BIT (str[0]); 3497 range_start = ((const unsigned char *) str)[0]; 3498 } 3499 # ifdef _LIBC 3500 else 3501 { 3502 /* Try to match the byte sequence in `str' against 3503 those known to the collate implementation. 3504 First find out whether the bytes in `str' are 3505 actually from exactly one character. */ 3506 int32_t table_size; 3507 const int32_t *symb_table; 3508 const unsigned char *extra; 3509 int32_t idx; 3510 int32_t elem; 3511 int32_t second; 3512 int32_t hash; 3513 3514 table_size = 3515 _NL_CURRENT_WORD (LC_COLLATE, 3516 _NL_COLLATE_SYMB_HASH_SIZEMB); 3517 symb_table = (const int32_t *) 3518 _NL_CURRENT (LC_COLLATE, 3519 _NL_COLLATE_SYMB_TABLEMB); 3520 extra = (const unsigned char *) 3521 _NL_CURRENT (LC_COLLATE, 3522 _NL_COLLATE_SYMB_EXTRAMB); 3523 3524 /* Locate the character in the hashing table. */ 3525 hash = elem_hash (str, c1); 3526 3527 idx = 0; 3528 elem = hash % table_size; 3529 second = hash % (table_size - 2); 3530 while (symb_table[2 * elem] != 0) 3531 { 3532 /* First compare the hashing value. */ 3533 if (symb_table[2 * elem] == hash 3534 && c1 == extra[symb_table[2 * elem + 1]] 3535 && memcmp (str, 3536 &extra[symb_table[2 * elem + 1] 3537 + 1], 3538 c1) == 0) 3539 { 3540 /* Yep, this is the entry. */ 3541 idx = symb_table[2 * elem + 1]; 3542 idx += 1 + extra[idx]; 3543 break; 3544 } 3545 3546 /* Next entry. */ 3547 elem += second; 3548 } 3549 3550 if (symb_table[2 * elem] == 0) 3551 /* This is no valid character. */ 3552 FREE_STACK_RETURN (REG_ECOLLATE); 3553 3554 /* Throw away the ] at the end of the equivalence 3555 class. */ 3556 PATFETCH (c); 3557 3558 /* Now add the multibyte character(s) we found 3559 to the accept list. 3560 3561 XXX Note that this is not entirely correct. 3562 we would have to match multibyte sequences 3563 but this is not possible with the current 3564 implementation. Also, we have to match 3565 collating symbols, which expand to more than 3566 one file, as a whole and not allow the 3567 individual bytes. */ 3568 c1 = extra[idx++]; 3569 if (c1 == 1) 3570 range_start = extra[idx]; 3571 while (c1-- > 0) 3572 { 3573 SET_LIST_BIT (extra[idx]); 3574 ++idx; 3575 } 3576 } 3577 # endif 3578 had_char_class = false; 3579 } 3580 else 3581 { 3582 c1++; 3583 while (c1--) 3584 PATUNFETCH; 3585 SET_LIST_BIT ('['); 3586 SET_LIST_BIT ('.'); 3587 range_start = '.'; 3588 had_char_class = false; 3589 } 3590 } 3591 else 3592 { 3593 had_char_class = false; 3594 SET_LIST_BIT (c); 3595 range_start = c; 3596 } 3597 } 3598 3599 /* Discard any (non)matching list bytes that are all 0 at the 3600 end of the map. Decrease the map-length byte too. */ 3601 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) 3602 b[-1]--; 3603 b += b[-1]; 3604 #endif /* WCHAR */ 3605 } 3606 break; 3607 3608 3609 case '(': 3610 if (syntax & RE_NO_BK_PARENS) 3611 goto handle_open; 3612 else 3613 goto normal_char; 3614 3615 3616 case ')': 3617 if (syntax & RE_NO_BK_PARENS) 3618 goto handle_close; 3619 else 3620 goto normal_char; 3621 3622 3623 case '\n': 3624 if (syntax & RE_NEWLINE_ALT) 3625 goto handle_alt; 3626 else 3627 goto normal_char; 3628 3629 3630 case '|': 3631 if (syntax & RE_NO_BK_VBAR) 3632 goto handle_alt; 3633 else 3634 goto normal_char; 3635 3636 3637 case '{': 3638 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) 3639 goto handle_interval; 3640 else 3641 goto normal_char; 3642 3643 3644 case '\\': 3645 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); 3646 3647 /* Do not translate the character after the \, so that we can 3648 distinguish, e.g., \B from \b, even if we normally would 3649 translate, e.g., B to b. */ 3650 PATFETCH_RAW (c); 3651 3652 switch (c) 3653 { 3654 case '(': 3655 if (syntax & RE_NO_BK_PARENS) 3656 goto normal_backslash; 3657 3658 handle_open: 3659 bufp->re_nsub++; 3660 regnum++; 3661 3662 if (COMPILE_STACK_FULL) 3663 { 3664 RETALLOC (compile_stack.stack, compile_stack.size << 1, 3665 compile_stack_elt_t); 3666 if (compile_stack.stack == NULL) return REG_ESPACE; 3667 3668 compile_stack.size <<= 1; 3669 } 3670 3671 /* These are the values to restore when we hit end of this 3672 group. They are all relative offsets, so that if the 3673 whole pattern moves because of realloc, they will still 3674 be valid. */ 3675 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR; 3676 COMPILE_STACK_TOP.fixup_alt_jump 3677 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0; 3678 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR; 3679 COMPILE_STACK_TOP.regnum = regnum; 3680 3681 /* We will eventually replace the 0 with the number of 3682 groups inner to this one. But do not push a 3683 start_memory for groups beyond the last one we can 3684 represent in the compiled pattern. */ 3685 if (regnum <= MAX_REGNUM) 3686 { 3687 COMPILE_STACK_TOP.inner_group_offset = b 3688 - COMPILED_BUFFER_VAR + 2; 3689 BUF_PUSH_3 (start_memory, regnum, 0); 3690 } 3691 3692 compile_stack.avail++; 3693 3694 fixup_alt_jump = 0; 3695 laststart = 0; 3696 begalt = b; 3697 /* If we've reached MAX_REGNUM groups, then this open 3698 won't actually generate any code, so we'll have to 3699 clear pending_exact explicitly. */ 3700 pending_exact = 0; 3701 break; 3702 3703 3704 case ')': 3705 if (syntax & RE_NO_BK_PARENS) goto normal_backslash; 3706 3707 if (COMPILE_STACK_EMPTY) 3708 { 3709 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) 3710 goto normal_backslash; 3711 else 3712 FREE_STACK_RETURN (REG_ERPAREN); 3713 } 3714 3715 handle_close: 3716 if (fixup_alt_jump) 3717 { /* Push a dummy failure point at the end of the 3718 alternative for a possible future 3719 `pop_failure_jump' to pop. See comments at 3720 `push_dummy_failure' in `re_match_2'. */ 3721 BUF_PUSH (push_dummy_failure); 3722 3723 /* We allocated space for this jump when we assigned 3724 to `fixup_alt_jump', in the `handle_alt' case below. */ 3725 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); 3726 } 3727 3728 /* See similar code for backslashed left paren above. */ 3729 if (COMPILE_STACK_EMPTY) 3730 { 3731 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) 3732 goto normal_char; 3733 else 3734 FREE_STACK_RETURN (REG_ERPAREN); 3735 } 3736 3737 /* Since we just checked for an empty stack above, this 3738 ``can't happen''. */ 3739 assert (compile_stack.avail != 0); 3740 { 3741 /* We don't just want to restore into `regnum', because 3742 later groups should continue to be numbered higher, 3743 as in `(ab)c(de)' -- the second group is #2. */ 3744 regnum_t this_group_regnum; 3745 3746 compile_stack.avail--; 3747 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset; 3748 fixup_alt_jump 3749 = COMPILE_STACK_TOP.fixup_alt_jump 3750 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1 3751 : 0; 3752 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset; 3753 this_group_regnum = COMPILE_STACK_TOP.regnum; 3754 /* If we've reached MAX_REGNUM groups, then this open 3755 won't actually generate any code, so we'll have to 3756 clear pending_exact explicitly. */ 3757 pending_exact = 0; 3758 3759 /* We're at the end of the group, so now we know how many 3760 groups were inside this one. */ 3761 if (this_group_regnum <= MAX_REGNUM) 3762 { 3763 UCHAR_T *inner_group_loc 3764 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset; 3765 3766 *inner_group_loc = regnum - this_group_regnum; 3767 BUF_PUSH_3 (stop_memory, this_group_regnum, 3768 regnum - this_group_regnum); 3769 } 3770 } 3771 break; 3772 3773 3774 case '|': /* `\|'. */ 3775 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) 3776 goto normal_backslash; 3777 handle_alt: 3778 if (syntax & RE_LIMITED_OPS) 3779 goto normal_char; 3780 3781 /* Insert before the previous alternative a jump which 3782 jumps to this alternative if the former fails. */ 3783 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 3784 INSERT_JUMP (on_failure_jump, begalt, 3785 b + 2 + 2 * OFFSET_ADDRESS_SIZE); 3786 pending_exact = 0; 3787 b += 1 + OFFSET_ADDRESS_SIZE; 3788 3789 /* The alternative before this one has a jump after it 3790 which gets executed if it gets matched. Adjust that 3791 jump so it will jump to this alternative's analogous 3792 jump (put in below, which in turn will jump to the next 3793 (if any) alternative's such jump, etc.). The last such 3794 jump jumps to the correct final destination. A picture: 3795 _____ _____ 3796 | | | | 3797 | v | v 3798 a | b | c 3799 3800 If we are at `b', then fixup_alt_jump right now points to a 3801 three-byte space after `a'. We'll put in the jump, set 3802 fixup_alt_jump to right after `b', and leave behind three 3803 bytes which we'll fill in when we get to after `c'. */ 3804 3805 if (fixup_alt_jump) 3806 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); 3807 3808 /* Mark and leave space for a jump after this alternative, 3809 to be filled in later either by next alternative or 3810 when know we're at the end of a series of alternatives. */ 3811 fixup_alt_jump = b; 3812 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 3813 b += 1 + OFFSET_ADDRESS_SIZE; 3814 3815 laststart = 0; 3816 begalt = b; 3817 break; 3818 3819 3820 case '{': 3821 /* If \{ is a literal. */ 3822 if (!(syntax & RE_INTERVALS) 3823 /* If we're at `\{' and it's not the open-interval 3824 operator. */ 3825 || (syntax & RE_NO_BK_BRACES)) 3826 goto normal_backslash; 3827 3828 handle_interval: 3829 { 3830 /* If got here, then the syntax allows intervals. */ 3831 3832 /* At least (most) this many matches must be made. */ 3833 int lower_bound = -1, upper_bound = -1; 3834 3835 /* Place in the uncompiled pattern (i.e., just after 3836 the '{') to go back to if the interval is invalid. */ 3837 const CHAR_T *beg_interval = p; 3838 3839 if (p == pend) 3840 goto invalid_interval; 3841 3842 GET_UNSIGNED_NUMBER (lower_bound); 3843 3844 if (c == ',') 3845 { 3846 GET_UNSIGNED_NUMBER (upper_bound); 3847 if (upper_bound < 0) 3848 upper_bound = RE_DUP_MAX; 3849 } 3850 else 3851 /* Interval such as `{1}' => match exactly once. */ 3852 upper_bound = lower_bound; 3853 3854 if (! (0 <= lower_bound && lower_bound <= upper_bound)) 3855 goto invalid_interval; 3856 3857 if (!(syntax & RE_NO_BK_BRACES)) 3858 { 3859 if (c != '\\' || p == pend) 3860 goto invalid_interval; 3861 PATFETCH (c); 3862 } 3863 3864 if (c != '}') 3865 goto invalid_interval; 3866 3867 /* If it's invalid to have no preceding re. */ 3868 if (!laststart) 3869 { 3870 if (syntax & RE_CONTEXT_INVALID_OPS 3871 && !(syntax & RE_INVALID_INTERVAL_ORD)) 3872 FREE_STACK_RETURN (REG_BADRPT); 3873 else if (syntax & RE_CONTEXT_INDEP_OPS) 3874 laststart = b; 3875 else 3876 goto unfetch_interval; 3877 } 3878 3879 /* We just parsed a valid interval. */ 3880 3881 if (RE_DUP_MAX < upper_bound) 3882 FREE_STACK_RETURN (REG_BADBR); 3883 3884 /* If the upper bound is zero, don't want to succeed at 3885 all; jump from `laststart' to `b + 3', which will be 3886 the end of the buffer after we insert the jump. */ 3887 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' 3888 instead of 'b + 3'. */ 3889 if (upper_bound == 0) 3890 { 3891 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); 3892 INSERT_JUMP (jump, laststart, b + 1 3893 + OFFSET_ADDRESS_SIZE); 3894 b += 1 + OFFSET_ADDRESS_SIZE; 3895 } 3896 3897 /* Otherwise, we have a nontrivial interval. When 3898 we're all done, the pattern will look like: 3899 set_number_at <jump count> <upper bound> 3900 set_number_at <succeed_n count> <lower bound> 3901 succeed_n <after jump addr> <succeed_n count> 3902 <body of loop> 3903 jump_n <succeed_n addr> <jump count> 3904 (The upper bound and `jump_n' are omitted if 3905 `upper_bound' is 1, though.) */ 3906 else 3907 { /* If the upper bound is > 1, we need to insert 3908 more at the end of the loop. */ 3909 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE + 3910 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE); 3911 3912 GET_BUFFER_SPACE (nbytes); 3913 3914 /* Initialize lower bound of the `succeed_n', even 3915 though it will be set during matching by its 3916 attendant `set_number_at' (inserted next), 3917 because `re_compile_fastmap' needs to know. 3918 Jump to the `jump_n' we might insert below. */ 3919 INSERT_JUMP2 (succeed_n, laststart, 3920 b + 1 + 2 * OFFSET_ADDRESS_SIZE 3921 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE) 3922 , lower_bound); 3923 b += 1 + 2 * OFFSET_ADDRESS_SIZE; 3924 3925 /* Code to initialize the lower bound. Insert 3926 before the `succeed_n'. The `5' is the last two 3927 bytes of this `set_number_at', plus 3 bytes of 3928 the following `succeed_n'. */ 3929 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE' 3930 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE' 3931 of the following `succeed_n'. */ 3932 PREFIX(insert_op2) (set_number_at, laststart, 1 3933 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b); 3934 b += 1 + 2 * OFFSET_ADDRESS_SIZE; 3935 3936 if (upper_bound > 1) 3937 { /* More than one repetition is allowed, so 3938 append a backward jump to the `succeed_n' 3939 that starts this interval. 3940 3941 When we've reached this during matching, 3942 we'll have matched the interval once, so 3943 jump back only `upper_bound - 1' times. */ 3944 STORE_JUMP2 (jump_n, b, laststart 3945 + 2 * OFFSET_ADDRESS_SIZE + 1, 3946 upper_bound - 1); 3947 b += 1 + 2 * OFFSET_ADDRESS_SIZE; 3948 3949 /* The location we want to set is the second 3950 parameter of the `jump_n'; that is `b-2' as 3951 an absolute address. `laststart' will be 3952 the `set_number_at' we're about to insert; 3953 `laststart+3' the number to set, the source 3954 for the relative address. But we are 3955 inserting into the middle of the pattern -- 3956 so everything is getting moved up by 5. 3957 Conclusion: (b - 2) - (laststart + 3) + 5, 3958 i.e., b - laststart. 3959 3960 We insert this at the beginning of the loop 3961 so that if we fail during matching, we'll 3962 reinitialize the bounds. */ 3963 PREFIX(insert_op2) (set_number_at, laststart, 3964 b - laststart, 3965 upper_bound - 1, b); 3966 b += 1 + 2 * OFFSET_ADDRESS_SIZE; 3967 } 3968 } 3969 pending_exact = 0; 3970 break; 3971 3972 invalid_interval: 3973 if (!(syntax & RE_INVALID_INTERVAL_ORD)) 3974 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR); 3975 unfetch_interval: 3976 /* Match the characters as literals. */ 3977 p = beg_interval; 3978 c = '{'; 3979 if (syntax & RE_NO_BK_BRACES) 3980 goto normal_char; 3981 else 3982 goto normal_backslash; 3983 } 3984 3985 #ifdef emacs 3986 /* There is no way to specify the before_dot and after_dot 3987 operators. rms says this is ok. --karl */ 3988 case '=': 3989 BUF_PUSH (at_dot); 3990 break; 3991 3992 case 's': 3993 laststart = b; 3994 PATFETCH (c); 3995 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); 3996 break; 3997 3998 case 'S': 3999 laststart = b; 4000 PATFETCH (c); 4001 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); 4002 break; 4003 #endif /* emacs */ 4004 4005 4006 case 'w': 4007 if (syntax & RE_NO_GNU_OPS) 4008 goto normal_char; 4009 laststart = b; 4010 BUF_PUSH (wordchar); 4011 break; 4012 4013 4014 case 'W': 4015 if (syntax & RE_NO_GNU_OPS) 4016 goto normal_char; 4017 laststart = b; 4018 BUF_PUSH (notwordchar); 4019 break; 4020 4021 4022 case '<': 4023 if (syntax & RE_NO_GNU_OPS) 4024 goto normal_char; 4025 BUF_PUSH (wordbeg); 4026 break; 4027 4028 case '>': 4029 if (syntax & RE_NO_GNU_OPS) 4030 goto normal_char; 4031 BUF_PUSH (wordend); 4032 break; 4033 4034 case 'b': 4035 if (syntax & RE_NO_GNU_OPS) 4036 goto normal_char; 4037 BUF_PUSH (wordbound); 4038 break; 4039 4040 case 'B': 4041 if (syntax & RE_NO_GNU_OPS) 4042 goto normal_char; 4043 BUF_PUSH (notwordbound); 4044 break; 4045 4046 case '`': 4047 if (syntax & RE_NO_GNU_OPS) 4048 goto normal_char; 4049 BUF_PUSH (begbuf); 4050 break; 4051 4052 case '\'': 4053 if (syntax & RE_NO_GNU_OPS) 4054 goto normal_char; 4055 BUF_PUSH (endbuf); 4056 break; 4057 4058 case '1': case '2': case '3': case '4': case '5': 4059 case '6': case '7': case '8': case '9': 4060 if (syntax & RE_NO_BK_REFS) 4061 goto normal_char; 4062 4063 c1 = c - '0'; 4064 4065 if (c1 > regnum) 4066 FREE_STACK_RETURN (REG_ESUBREG); 4067 4068 /* Can't back reference to a subexpression if inside of it. */ 4069 if (group_in_compile_stack (compile_stack, (regnum_t) c1)) 4070 goto normal_char; 4071 4072 laststart = b; 4073 BUF_PUSH_2 (duplicate, c1); 4074 break; 4075 4076 4077 case '+': 4078 case '?': 4079 if (syntax & RE_BK_PLUS_QM) 4080 goto handle_plus; 4081 else 4082 goto normal_backslash; 4083 4084 default: 4085 normal_backslash: 4086 /* You might think it would be useful for \ to mean 4087 not to translate; but if we don't translate it 4088 it will never match anything. */ 4089 c = TRANSLATE (c); 4090 goto normal_char; 4091 } 4092 break; 4093 4094 4095 default: 4096 /* Expects the character in `c'. */ 4097 normal_char: 4098 /* If no exactn currently being built. */ 4099 if (!pending_exact 4100 #ifdef WCHAR 4101 /* If last exactn handle binary(or character) and 4102 new exactn handle character(or binary). */ 4103 || is_exactn_bin != is_binary[p - 1 - pattern] 4104 #endif /* WCHAR */ 4105 4106 /* If last exactn not at current position. */ 4107 || pending_exact + *pending_exact + 1 != b 4108 4109 /* We have only one byte following the exactn for the count. */ 4110 || *pending_exact == (1 << BYTEWIDTH) - 1 4111 4112 /* If followed by a repetition operator. */ 4113 || *p == '*' || *p == '^' 4114 || ((syntax & RE_BK_PLUS_QM) 4115 ? *p == '\\' && (p[1] == '+' || p[1] == '?') 4116 : (*p == '+' || *p == '?')) 4117 || ((syntax & RE_INTERVALS) 4118 && ((syntax & RE_NO_BK_BRACES) 4119 ? *p == '{' 4120 : (p[0] == '\\' && p[1] == '{')))) 4121 { 4122 /* Start building a new exactn. */ 4123 4124 laststart = b; 4125 4126 #ifdef WCHAR 4127 /* Is this exactn binary data or character? */ 4128 is_exactn_bin = is_binary[p - 1 - pattern]; 4129 if (is_exactn_bin) 4130 BUF_PUSH_2 (exactn_bin, 0); 4131 else 4132 BUF_PUSH_2 (exactn, 0); 4133 #else 4134 BUF_PUSH_2 (exactn, 0); 4135 #endif /* WCHAR */ 4136 pending_exact = b - 1; 4137 } 4138 4139 BUF_PUSH (c); 4140 (*pending_exact)++; 4141 break; 4142 } /* switch (c) */ 4143 } /* while p != pend */ 4144 4145 4146 /* Through the pattern now. */ 4147 4148 if (fixup_alt_jump) 4149 STORE_JUMP (jump_past_alt, fixup_alt_jump, b); 4150 4151 if (!COMPILE_STACK_EMPTY) 4152 FREE_STACK_RETURN (REG_EPAREN); 4153 4154 /* If we don't want backtracking, force success 4155 the first time we reach the end of the compiled pattern. */ 4156 if (syntax & RE_NO_POSIX_BACKTRACKING) 4157 BUF_PUSH (succeed); 4158 4159 #ifdef WCHAR 4160 free (pattern); 4161 free (mbs_offset); 4162 free (is_binary); 4163 #endif 4164 free (compile_stack.stack); 4165 4166 /* We have succeeded; set the length of the buffer. */ 4167 #ifdef WCHAR 4168 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR; 4169 #else 4170 bufp->used = b - bufp->buffer; 4171 #endif 4172 4173 #ifdef DEBUG 4174 if (debug) 4175 { 4176 DEBUG_PRINT1 ("\nCompiled pattern: \n"); 4177 PREFIX(print_compiled_pattern) (bufp); 4178 } 4179 #endif /* DEBUG */ 4180 4181 #ifndef MATCH_MAY_ALLOCATE 4182 /* Initialize the failure stack to the largest possible stack. This 4183 isn't necessary unless we're trying to avoid calling alloca in 4184 the search and match routines. */ 4185 { 4186 int num_regs = bufp->re_nsub + 1; 4187 4188 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size 4189 is strictly greater than re_max_failures, the largest possible stack 4190 is 2 * re_max_failures failure points. */ 4191 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) 4192 { 4193 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); 4194 4195 # ifdef emacs 4196 if (! fail_stack.stack) 4197 fail_stack.stack 4198 = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size 4199 * sizeof (PREFIX(fail_stack_elt_t))); 4200 else 4201 fail_stack.stack 4202 = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack, 4203 (fail_stack.size 4204 * sizeof (PREFIX(fail_stack_elt_t)))); 4205 # else /* not emacs */ 4206 if (! fail_stack.stack) 4207 fail_stack.stack 4208 = (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size 4209 * sizeof (PREFIX(fail_stack_elt_t))); 4210 else 4211 fail_stack.stack 4212 = (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack, 4213 (fail_stack.size 4214 * sizeof (PREFIX(fail_stack_elt_t)))); 4215 # endif /* not emacs */ 4216 } 4217 4218 PREFIX(regex_grow_registers) (num_regs); 4219 } 4220 #endif /* not MATCH_MAY_ALLOCATE */ 4221 4222 return REG_NOERROR; 4223 } /* regex_compile */ 4224 4225 /* Subroutines for `regex_compile'. */ 4226 4227 /* Store OP at LOC followed by two-byte integer parameter ARG. */ 4228 /* ifdef WCHAR, integer parameter is 1 wchar_t. */ 4229 4230 static void 4231 PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg) 4232 { 4233 *loc = (UCHAR_T) op; 4234 STORE_NUMBER (loc + 1, arg); 4235 } 4236 4237 4238 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ 4239 /* ifdef WCHAR, integer parameter is 1 wchar_t. */ 4240 4241 static void 4242 PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, int arg2) 4243 { 4244 *loc = (UCHAR_T) op; 4245 STORE_NUMBER (loc + 1, arg1); 4246 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2); 4247 } 4248 4249 4250 /* Copy the bytes from LOC to END to open up three bytes of space at LOC 4251 for OP followed by two-byte integer parameter ARG. */ 4252 /* ifdef WCHAR, integer parameter is 1 wchar_t. */ 4253 4254 static void 4255 PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc, int arg, UCHAR_T *end) 4256 { 4257 register UCHAR_T *pfrom = end; 4258 register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE; 4259 4260 while (pfrom != loc) 4261 *--pto = *--pfrom; 4262 4263 PREFIX(store_op1) (op, loc, arg); 4264 } 4265 4266 4267 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ 4268 /* ifdef WCHAR, integer parameter is 1 wchar_t. */ 4269 4270 static void 4271 PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, 4272 int arg2, UCHAR_T *end) 4273 { 4274 register UCHAR_T *pfrom = end; 4275 register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE; 4276 4277 while (pfrom != loc) 4278 *--pto = *--pfrom; 4279 4280 PREFIX(store_op2) (op, loc, arg1, arg2); 4281 } 4282 4283 4284 /* P points to just after a ^ in PATTERN. Return true if that ^ comes 4285 after an alternative or a begin-subexpression. We assume there is at 4286 least one character before the ^. */ 4287 4288 static boolean 4289 PREFIX(at_begline_loc_p) (const CHAR_T *pattern, const CHAR_T *p, 4290 reg_syntax_t syntax) 4291 { 4292 const CHAR_T *prev = p - 2; 4293 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; 4294 4295 return 4296 /* After a subexpression? */ 4297 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) 4298 /* After an alternative? */ 4299 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); 4300 } 4301 4302 4303 /* The dual of at_begline_loc_p. This one is for $. We assume there is 4304 at least one character after the $, i.e., `P < PEND'. */ 4305 4306 static boolean 4307 PREFIX(at_endline_loc_p) (const CHAR_T *p, const CHAR_T *pend, 4308 reg_syntax_t syntax) 4309 { 4310 const CHAR_T *next = p; 4311 boolean next_backslash = *next == '\\'; 4312 const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0; 4313 4314 return 4315 /* Before a subexpression? */ 4316 (syntax & RE_NO_BK_PARENS ? *next == ')' 4317 : next_backslash && next_next && *next_next == ')') 4318 /* Before an alternative? */ 4319 || (syntax & RE_NO_BK_VBAR ? *next == '|' 4320 : next_backslash && next_next && *next_next == '|'); 4321 } 4322 4323 #else /* not INSIDE_RECURSION */ 4324 4325 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and 4326 false if it's not. */ 4327 4328 static boolean 4329 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum) 4330 { 4331 int this_element; 4332 4333 for (this_element = compile_stack.avail - 1; 4334 this_element >= 0; 4335 this_element--) 4336 if (compile_stack.stack[this_element].regnum == regnum) 4337 return true; 4338 4339 return false; 4340 } 4341 #endif /* not INSIDE_RECURSION */ 4342 4343 #ifdef INSIDE_RECURSION 4344 4345 #ifdef WCHAR 4346 /* This insert space, which size is "num", into the pattern at "loc". 4347 "end" must point the end of the allocated buffer. */ 4348 static void 4349 insert_space (int num, CHAR_T *loc, CHAR_T *end) 4350 { 4351 register CHAR_T *pto = end; 4352 register CHAR_T *pfrom = end - num; 4353 4354 while (pfrom >= loc) 4355 *pto-- = *pfrom--; 4356 } 4357 #endif /* WCHAR */ 4358 4359 #ifdef WCHAR 4360 static reg_errcode_t 4361 wcs_compile_range (CHAR_T range_start_char, const CHAR_T **p_ptr, 4362 const CHAR_T *pend, RE_TRANSLATE_TYPE translate, 4363 reg_syntax_t syntax, CHAR_T *b, CHAR_T *char_set) 4364 { 4365 const CHAR_T *p = *p_ptr; 4366 CHAR_T range_start, range_end; 4367 reg_errcode_t ret; 4368 # ifdef _LIBC 4369 uint32_t nrules; 4370 uint32_t start_val, end_val; 4371 # endif 4372 if (p == pend) 4373 return REG_ERANGE; 4374 4375 # ifdef _LIBC 4376 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 4377 if (nrules != 0) 4378 { 4379 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE, 4380 _NL_COLLATE_COLLSEQWC); 4381 const unsigned char *extra = (const unsigned char *) 4382 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); 4383 4384 if (range_start_char < -1) 4385 { 4386 /* range_start is a collating symbol. */ 4387 int32_t *wextra; 4388 /* Retreive the index and get collation sequence value. */ 4389 wextra = (int32_t*)(extra + char_set[-range_start_char]); 4390 start_val = wextra[1 + *wextra]; 4391 } 4392 else 4393 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char)); 4394 4395 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0])); 4396 4397 /* Report an error if the range is empty and the syntax prohibits 4398 this. */ 4399 ret = ((syntax & RE_NO_EMPTY_RANGES) 4400 && (start_val > end_val))? REG_ERANGE : REG_NOERROR; 4401 4402 /* Insert space to the end of the char_ranges. */ 4403 insert_space(2, b - char_set[5] - 2, b - 1); 4404 *(b - char_set[5] - 2) = (wchar_t)start_val; 4405 *(b - char_set[5] - 1) = (wchar_t)end_val; 4406 char_set[4]++; /* ranges_index */ 4407 } 4408 else 4409 # endif 4410 { 4411 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char): 4412 range_start_char; 4413 range_end = TRANSLATE (p[0]); 4414 /* Report an error if the range is empty and the syntax prohibits 4415 this. */ 4416 ret = ((syntax & RE_NO_EMPTY_RANGES) 4417 && (range_start > range_end))? REG_ERANGE : REG_NOERROR; 4418 4419 /* Insert space to the end of the char_ranges. */ 4420 insert_space(2, b - char_set[5] - 2, b - 1); 4421 *(b - char_set[5] - 2) = range_start; 4422 *(b - char_set[5] - 1) = range_end; 4423 char_set[4]++; /* ranges_index */ 4424 } 4425 /* Have to increment the pointer into the pattern string, so the 4426 caller isn't still at the ending character. */ 4427 (*p_ptr)++; 4428 4429 return ret; 4430 } 4431 #else /* BYTE */ 4432 /* Read the ending character of a range (in a bracket expression) from the 4433 uncompiled pattern *P_PTR (which ends at PEND). We assume the 4434 starting character is in `P[-2]'. (`P[-1]' is the character `-'.) 4435 Then we set the translation of all bits between the starting and 4436 ending characters (inclusive) in the compiled pattern B. 4437 4438 Return an error code. 4439 4440 We use these short variable names so we can use the same macros as 4441 `regex_compile' itself. */ 4442 4443 static reg_errcode_t 4444 byte_compile_range (unsigned int range_start_char, const char **p_ptr, 4445 const char *pend, RE_TRANSLATE_TYPE translate, 4446 reg_syntax_t syntax, unsigned char *b) 4447 { 4448 unsigned this_char; 4449 const char *p = *p_ptr; 4450 reg_errcode_t ret; 4451 # if _LIBC 4452 const unsigned char *collseq; 4453 unsigned int start_colseq; 4454 unsigned int end_colseq; 4455 # else 4456 unsigned end_char; 4457 # endif 4458 4459 if (p == pend) 4460 return REG_ERANGE; 4461 4462 /* Have to increment the pointer into the pattern string, so the 4463 caller isn't still at the ending character. */ 4464 (*p_ptr)++; 4465 4466 /* Report an error if the range is empty and the syntax prohibits this. */ 4467 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; 4468 4469 # if _LIBC 4470 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE, 4471 _NL_COLLATE_COLLSEQMB); 4472 4473 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)]; 4474 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])]; 4475 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char) 4476 { 4477 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)]; 4478 4479 if (start_colseq <= this_colseq && this_colseq <= end_colseq) 4480 { 4481 SET_LIST_BIT (TRANSLATE (this_char)); 4482 ret = REG_NOERROR; 4483 } 4484 } 4485 # else 4486 /* Here we see why `this_char' has to be larger than an `unsigned 4487 char' -- we would otherwise go into an infinite loop, since all 4488 characters <= 0xff. */ 4489 range_start_char = TRANSLATE (range_start_char); 4490 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE, 4491 and some compilers cast it to int implicitly, so following for_loop 4492 may fall to (almost) infinite loop. 4493 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff. 4494 To avoid this, we cast p[0] to unsigned int and truncate it. */ 4495 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1)); 4496 4497 for (this_char = range_start_char; this_char <= end_char; ++this_char) 4498 { 4499 SET_LIST_BIT (TRANSLATE (this_char)); 4500 ret = REG_NOERROR; 4501 } 4502 # endif 4503 4504 return ret; 4505 } 4506 #endif /* WCHAR */ 4507 4508 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in 4509 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible 4510 characters can start a string that matches the pattern. This fastmap 4511 is used by re_search to skip quickly over impossible starting points. 4512 4513 The caller must supply the address of a (1 << BYTEWIDTH)-byte data 4514 area as BUFP->fastmap. 4515 4516 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in 4517 the pattern buffer. 4518 4519 Returns 0 if we succeed, -2 if an internal error. */ 4520 4521 #ifdef WCHAR 4522 /* local function for re_compile_fastmap. 4523 truncate wchar_t character to char. */ 4524 static unsigned char truncate_wchar (CHAR_T c); 4525 4526 static unsigned char 4527 truncate_wchar (CHAR_T c) 4528 { 4529 unsigned char buf[MB_CUR_MAX]; 4530 mbstate_t state; 4531 int retval; 4532 memset (&state, '\0', sizeof (state)); 4533 # ifdef _LIBC 4534 retval = __wcrtomb (buf, c, &state); 4535 # else 4536 retval = wcrtomb (buf, c, &state); 4537 # endif 4538 return retval > 0 ? buf[0] : (unsigned char) c; 4539 } 4540 #endif /* WCHAR */ 4541 4542 static int 4543 PREFIX(re_compile_fastmap) (struct re_pattern_buffer *bufp) 4544 { 4545 int j, k; 4546 #ifdef MATCH_MAY_ALLOCATE 4547 PREFIX(fail_stack_type) fail_stack; 4548 #endif 4549 #ifndef REGEX_MALLOC 4550 char *destination; 4551 #endif 4552 4553 register char *fastmap = bufp->fastmap; 4554 4555 #ifdef WCHAR 4556 /* We need to cast pattern to (wchar_t*), because we casted this compiled 4557 pattern to (char*) in regex_compile. */ 4558 UCHAR_T *pattern = (UCHAR_T*)bufp->buffer; 4559 register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used); 4560 #else /* BYTE */ 4561 UCHAR_T *pattern = bufp->buffer; 4562 register UCHAR_T *pend = pattern + bufp->used; 4563 #endif /* WCHAR */ 4564 UCHAR_T *p = pattern; 4565 4566 #ifdef REL_ALLOC 4567 /* This holds the pointer to the failure stack, when 4568 it is allocated relocatably. */ 4569 fail_stack_elt_t *failure_stack_ptr; 4570 #endif 4571 4572 /* Assume that each path through the pattern can be null until 4573 proven otherwise. We set this false at the bottom of switch 4574 statement, to which we get only if a particular path doesn't 4575 match the empty string. */ 4576 boolean path_can_be_null = true; 4577 4578 /* We aren't doing a `succeed_n' to begin with. */ 4579 boolean succeed_n_p = false; 4580 4581 assert (fastmap != NULL && p != NULL); 4582 4583 INIT_FAIL_STACK (); 4584 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ 4585 bufp->fastmap_accurate = 1; /* It will be when we're done. */ 4586 bufp->can_be_null = 0; 4587 4588 while (1) 4589 { 4590 if (p == pend || *p == (UCHAR_T) succeed) 4591 { 4592 /* We have reached the (effective) end of pattern. */ 4593 if (!FAIL_STACK_EMPTY ()) 4594 { 4595 bufp->can_be_null |= path_can_be_null; 4596 4597 /* Reset for next path. */ 4598 path_can_be_null = true; 4599 4600 p = fail_stack.stack[--fail_stack.avail].pointer; 4601 4602 continue; 4603 } 4604 else 4605 break; 4606 } 4607 4608 /* We should never be about to go beyond the end of the pattern. */ 4609 assert (p < pend); 4610 4611 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) 4612 { 4613 4614 /* I guess the idea here is to simply not bother with a fastmap 4615 if a backreference is used, since it's too hard to figure out 4616 the fastmap for the corresponding group. Setting 4617 `can_be_null' stops `re_search_2' from using the fastmap, so 4618 that is all we do. */ 4619 case duplicate: 4620 bufp->can_be_null = 1; 4621 goto done; 4622 4623 4624 /* Following are the cases which match a character. These end 4625 with `break'. */ 4626 4627 #ifdef WCHAR 4628 case exactn: 4629 fastmap[truncate_wchar(p[1])] = 1; 4630 break; 4631 #else /* BYTE */ 4632 case exactn: 4633 fastmap[p[1]] = 1; 4634 break; 4635 #endif /* WCHAR */ 4636 #ifdef MBS_SUPPORT 4637 case exactn_bin: 4638 fastmap[p[1]] = 1; 4639 break; 4640 #endif 4641 4642 #ifdef WCHAR 4643 /* It is hard to distinguish fastmap from (multi byte) characters 4644 which depends on current locale. */ 4645 case charset: 4646 case charset_not: 4647 case wordchar: 4648 case notwordchar: 4649 bufp->can_be_null = 1; 4650 goto done; 4651 #else /* BYTE */ 4652 case charset: 4653 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) 4654 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) 4655 fastmap[j] = 1; 4656 break; 4657 4658 4659 case charset_not: 4660 /* Chars beyond end of map must be allowed. */ 4661 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) 4662 fastmap[j] = 1; 4663 4664 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) 4665 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) 4666 fastmap[j] = 1; 4667 break; 4668 4669 4670 case wordchar: 4671 for (j = 0; j < (1 << BYTEWIDTH); j++) 4672 if (SYNTAX (j) == Sword) 4673 fastmap[j] = 1; 4674 break; 4675 4676 4677 case notwordchar: 4678 for (j = 0; j < (1 << BYTEWIDTH); j++) 4679 if (SYNTAX (j) != Sword) 4680 fastmap[j] = 1; 4681 break; 4682 #endif /* WCHAR */ 4683 4684 case anychar: 4685 { 4686 int fastmap_newline = fastmap['\n']; 4687 4688 /* `.' matches anything ... */ 4689 for (j = 0; j < (1 << BYTEWIDTH); j++) 4690 fastmap[j] = 1; 4691 4692 /* ... except perhaps newline. */ 4693 if (!(bufp->syntax & RE_DOT_NEWLINE)) 4694 fastmap['\n'] = fastmap_newline; 4695 4696 /* Return if we have already set `can_be_null'; if we have, 4697 then the fastmap is irrelevant. Something's wrong here. */ 4698 else if (bufp->can_be_null) 4699 goto done; 4700 4701 /* Otherwise, have to check alternative paths. */ 4702 break; 4703 } 4704 4705 #ifdef emacs 4706 case syntaxspec: 4707 k = *p++; 4708 for (j = 0; j < (1 << BYTEWIDTH); j++) 4709 if (SYNTAX (j) == (enum syntaxcode) k) 4710 fastmap[j] = 1; 4711 break; 4712 4713 4714 case notsyntaxspec: 4715 k = *p++; 4716 for (j = 0; j < (1 << BYTEWIDTH); j++) 4717 if (SYNTAX (j) != (enum syntaxcode) k) 4718 fastmap[j] = 1; 4719 break; 4720 4721 4722 /* All cases after this match the empty string. These end with 4723 `continue'. */ 4724 4725 4726 case before_dot: 4727 case at_dot: 4728 case after_dot: 4729 continue; 4730 #endif /* emacs */ 4731 4732 4733 case no_op: 4734 case begline: 4735 case endline: 4736 case begbuf: 4737 case endbuf: 4738 case wordbound: 4739 case notwordbound: 4740 case wordbeg: 4741 case wordend: 4742 case push_dummy_failure: 4743 continue; 4744 4745 4746 case jump_n: 4747 case pop_failure_jump: 4748 case maybe_pop_jump: 4749 case jump: 4750 case jump_past_alt: 4751 case dummy_failure_jump: 4752 EXTRACT_NUMBER_AND_INCR (j, p); 4753 p += j; 4754 if (j > 0) 4755 continue; 4756 4757 /* Jump backward implies we just went through the body of a 4758 loop and matched nothing. Opcode jumped to should be 4759 `on_failure_jump' or `succeed_n'. Just treat it like an 4760 ordinary jump. For a * loop, it has pushed its failure 4761 point already; if so, discard that as redundant. */ 4762 if ((re_opcode_t) *p != on_failure_jump 4763 && (re_opcode_t) *p != succeed_n) 4764 continue; 4765 4766 p++; 4767 EXTRACT_NUMBER_AND_INCR (j, p); 4768 p += j; 4769 4770 /* If what's on the stack is where we are now, pop it. */ 4771 if (!FAIL_STACK_EMPTY () 4772 && fail_stack.stack[fail_stack.avail - 1].pointer == p) 4773 fail_stack.avail--; 4774 4775 continue; 4776 4777 4778 case on_failure_jump: 4779 case on_failure_keep_string_jump: 4780 handle_on_failure_jump: 4781 EXTRACT_NUMBER_AND_INCR (j, p); 4782 4783 /* For some patterns, e.g., `(a?)?', `p+j' here points to the 4784 end of the pattern. We don't want to push such a point, 4785 since when we restore it above, entering the switch will 4786 increment `p' past the end of the pattern. We don't need 4787 to push such a point since we obviously won't find any more 4788 fastmap entries beyond `pend'. Such a pattern can match 4789 the null string, though. */ 4790 if (p + j < pend) 4791 { 4792 if (!PUSH_PATTERN_OP (p + j, fail_stack)) 4793 { 4794 RESET_FAIL_STACK (); 4795 return -2; 4796 } 4797 } 4798 else 4799 bufp->can_be_null = 1; 4800 4801 if (succeed_n_p) 4802 { 4803 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ 4804 succeed_n_p = false; 4805 } 4806 4807 continue; 4808 4809 4810 case succeed_n: 4811 /* Get to the number of times to succeed. */ 4812 p += OFFSET_ADDRESS_SIZE; 4813 4814 /* Increment p past the n for when k != 0. */ 4815 EXTRACT_NUMBER_AND_INCR (k, p); 4816 if (k == 0) 4817 { 4818 p -= 2 * OFFSET_ADDRESS_SIZE; 4819 succeed_n_p = true; /* Spaghetti code alert. */ 4820 goto handle_on_failure_jump; 4821 } 4822 continue; 4823 4824 4825 case set_number_at: 4826 p += 2 * OFFSET_ADDRESS_SIZE; 4827 continue; 4828 4829 4830 case start_memory: 4831 case stop_memory: 4832 p += 2; 4833 continue; 4834 4835 4836 default: 4837 abort (); /* We have listed all the cases. */ 4838 } /* switch *p++ */ 4839 4840 /* Getting here means we have found the possible starting 4841 characters for one path of the pattern -- and that the empty 4842 string does not match. We need not follow this path further. 4843 Instead, look at the next alternative (remembered on the 4844 stack), or quit if no more. The test at the top of the loop 4845 does these things. */ 4846 path_can_be_null = false; 4847 p = pend; 4848 } /* while p */ 4849 4850 /* Set `can_be_null' for the last path (also the first path, if the 4851 pattern is empty). */ 4852 bufp->can_be_null |= path_can_be_null; 4853 4854 done: 4855 RESET_FAIL_STACK (); 4856 return 0; 4857 } 4858 4859 #else /* not INSIDE_RECURSION */ 4860 4861 int 4862 re_compile_fastmap (struct re_pattern_buffer *bufp) 4863 { 4864 # ifdef MBS_SUPPORT 4865 if (MB_CUR_MAX != 1) 4866 return wcs_re_compile_fastmap(bufp); 4867 else 4868 # endif 4869 return byte_re_compile_fastmap(bufp); 4870 } /* re_compile_fastmap */ 4871 #ifdef _LIBC 4872 weak_alias (__re_compile_fastmap, re_compile_fastmap) 4873 #endif 4874 4875 4876 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and 4877 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use 4878 this memory for recording register information. STARTS and ENDS 4879 must be allocated using the malloc library routine, and must each 4880 be at least NUM_REGS * sizeof (regoff_t) bytes long. 4881 4882 If NUM_REGS == 0, then subsequent matches should allocate their own 4883 register data. 4884 4885 Unless this function is called, the first search or match using 4886 PATTERN_BUFFER will allocate its own register data, without 4887 freeing the old data. */ 4888 4889 void 4890 re_set_registers (struct re_pattern_buffer *bufp, 4891 struct re_registers *regs, unsigned num_regs, 4892 regoff_t *starts, regoff_t *ends) 4893 { 4894 if (num_regs) 4895 { 4896 bufp->regs_allocated = REGS_REALLOCATE; 4897 regs->num_regs = num_regs; 4898 regs->start = starts; 4899 regs->end = ends; 4900 } 4901 else 4902 { 4903 bufp->regs_allocated = REGS_UNALLOCATED; 4904 regs->num_regs = 0; 4905 regs->start = regs->end = (regoff_t *) 0; 4906 } 4907 } 4908 #ifdef _LIBC 4909 weak_alias (__re_set_registers, re_set_registers) 4910 #endif 4911 4912 /* Searching routines. */ 4913 4914 /* Like re_search_2, below, but only one string is specified, and 4915 doesn't let you say where to stop matching. */ 4916 4917 int 4918 re_search (struct re_pattern_buffer *bufp, const char *string, int size, 4919 int startpos, int range, struct re_registers *regs) 4920 { 4921 return re_search_2 (bufp, NULL, 0, string, size, startpos, range, 4922 regs, size); 4923 } 4924 #ifdef _LIBC 4925 weak_alias (__re_search, re_search) 4926 #endif 4927 4928 4929 /* Using the compiled pattern in BUFP->buffer, first tries to match the 4930 virtual concatenation of STRING1 and STRING2, starting first at index 4931 STARTPOS, then at STARTPOS + 1, and so on. 4932 4933 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. 4934 4935 RANGE is how far to scan while trying to match. RANGE = 0 means try 4936 only at STARTPOS; in general, the last start tried is STARTPOS + 4937 RANGE. 4938 4939 In REGS, return the indices of the virtual concatenation of STRING1 4940 and STRING2 that matched the entire BUFP->buffer and its contained 4941 subexpressions. 4942 4943 Do not consider matching one past the index STOP in the virtual 4944 concatenation of STRING1 and STRING2. 4945 4946 We return either the position in the strings at which the match was 4947 found, -1 if no match, or -2 if error (such as failure 4948 stack overflow). */ 4949 4950 int 4951 re_search_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, 4952 const char *string2, int size2, int startpos, int range, 4953 struct re_registers *regs, int stop) 4954 { 4955 # ifdef MBS_SUPPORT 4956 if (MB_CUR_MAX != 1) 4957 return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos, 4958 range, regs, stop); 4959 else 4960 # endif 4961 return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos, 4962 range, regs, stop); 4963 } /* re_search_2 */ 4964 #ifdef _LIBC 4965 weak_alias (__re_search_2, re_search_2) 4966 #endif 4967 4968 #endif /* not INSIDE_RECURSION */ 4969 4970 #ifdef INSIDE_RECURSION 4971 4972 #ifdef MATCH_MAY_ALLOCATE 4973 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL 4974 #else 4975 # define FREE_VAR(var) free (var); var = NULL 4976 #endif 4977 4978 #ifdef WCHAR 4979 # define MAX_ALLOCA_SIZE 2000 4980 4981 # define FREE_WCS_BUFFERS() \ 4982 do { \ 4983 if (size1 > MAX_ALLOCA_SIZE) \ 4984 { \ 4985 free (wcs_string1); \ 4986 free (mbs_offset1); \ 4987 } \ 4988 else \ 4989 { \ 4990 FREE_VAR (wcs_string1); \ 4991 FREE_VAR (mbs_offset1); \ 4992 } \ 4993 if (size2 > MAX_ALLOCA_SIZE) \ 4994 { \ 4995 free (wcs_string2); \ 4996 free (mbs_offset2); \ 4997 } \ 4998 else \ 4999 { \ 5000 FREE_VAR (wcs_string2); \ 5001 FREE_VAR (mbs_offset2); \ 5002 } \ 5003 } while (0) 5004 5005 #endif 5006 5007 5008 static int 5009 PREFIX(re_search_2) (struct re_pattern_buffer *bufp, const char *string1, 5010 int size1, const char *string2, int size2, 5011 int startpos, int range, 5012 struct re_registers *regs, int stop) 5013 { 5014 int val; 5015 register char *fastmap = bufp->fastmap; 5016 register RE_TRANSLATE_TYPE translate = bufp->translate; 5017 int total_size = size1 + size2; 5018 int endpos = startpos + range; 5019 #ifdef WCHAR 5020 /* We need wchar_t* buffers correspond to cstring1, cstring2. */ 5021 wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL; 5022 /* We need the size of wchar_t buffers correspond to csize1, csize2. */ 5023 int wcs_size1 = 0, wcs_size2 = 0; 5024 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ 5025 int *mbs_offset1 = NULL, *mbs_offset2 = NULL; 5026 /* They hold whether each wchar_t is binary data or not. */ 5027 char *is_binary = NULL; 5028 #endif /* WCHAR */ 5029 5030 /* Check for out-of-range STARTPOS. */ 5031 if (startpos < 0 || startpos > total_size) 5032 return -1; 5033 5034 /* Fix up RANGE if it might eventually take us outside 5035 the virtual concatenation of STRING1 and STRING2. 5036 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ 5037 if (endpos < 0) 5038 range = 0 - startpos; 5039 else if (endpos > total_size) 5040 range = total_size - startpos; 5041 5042 /* If the search isn't to be a backwards one, don't waste time in a 5043 search for a pattern that must be anchored. */ 5044 if (bufp->used > 0 && range > 0 5045 && ((re_opcode_t) bufp->buffer[0] == begbuf 5046 /* `begline' is like `begbuf' if it cannot match at newlines. */ 5047 || ((re_opcode_t) bufp->buffer[0] == begline 5048 && !bufp->newline_anchor))) 5049 { 5050 if (startpos > 0) 5051 return -1; 5052 else 5053 range = 1; 5054 } 5055 5056 #ifdef emacs 5057 /* In a forward search for something that starts with \=. 5058 don't keep searching past point. */ 5059 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) 5060 { 5061 range = PT - startpos; 5062 if (range <= 0) 5063 return -1; 5064 } 5065 #endif /* emacs */ 5066 5067 /* Update the fastmap now if not correct already. */ 5068 if (fastmap && !bufp->fastmap_accurate) 5069 if (re_compile_fastmap (bufp) == -2) 5070 return -2; 5071 5072 #ifdef WCHAR 5073 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and 5074 fill them with converted string. */ 5075 if (size1 != 0) 5076 { 5077 if (size1 > MAX_ALLOCA_SIZE) 5078 { 5079 wcs_string1 = TALLOC (size1 + 1, CHAR_T); 5080 mbs_offset1 = TALLOC (size1 + 1, int); 5081 is_binary = TALLOC (size1 + 1, char); 5082 } 5083 else 5084 { 5085 wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T); 5086 mbs_offset1 = REGEX_TALLOC (size1 + 1, int); 5087 is_binary = REGEX_TALLOC (size1 + 1, char); 5088 } 5089 if (!wcs_string1 || !mbs_offset1 || !is_binary) 5090 { 5091 if (size1 > MAX_ALLOCA_SIZE) 5092 { 5093 free (wcs_string1); 5094 free (mbs_offset1); 5095 free (is_binary); 5096 } 5097 else 5098 { 5099 FREE_VAR (wcs_string1); 5100 FREE_VAR (mbs_offset1); 5101 FREE_VAR (is_binary); 5102 } 5103 return -2; 5104 } 5105 wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1, 5106 mbs_offset1, is_binary); 5107 wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */ 5108 if (size1 > MAX_ALLOCA_SIZE) 5109 free (is_binary); 5110 else 5111 FREE_VAR (is_binary); 5112 } 5113 if (size2 != 0) 5114 { 5115 if (size2 > MAX_ALLOCA_SIZE) 5116 { 5117 wcs_string2 = TALLOC (size2 + 1, CHAR_T); 5118 mbs_offset2 = TALLOC (size2 + 1, int); 5119 is_binary = TALLOC (size2 + 1, char); 5120 } 5121 else 5122 { 5123 wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T); 5124 mbs_offset2 = REGEX_TALLOC (size2 + 1, int); 5125 is_binary = REGEX_TALLOC (size2 + 1, char); 5126 } 5127 if (!wcs_string2 || !mbs_offset2 || !is_binary) 5128 { 5129 FREE_WCS_BUFFERS (); 5130 if (size2 > MAX_ALLOCA_SIZE) 5131 free (is_binary); 5132 else 5133 FREE_VAR (is_binary); 5134 return -2; 5135 } 5136 wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2, 5137 mbs_offset2, is_binary); 5138 wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */ 5139 if (size2 > MAX_ALLOCA_SIZE) 5140 free (is_binary); 5141 else 5142 FREE_VAR (is_binary); 5143 } 5144 #endif /* WCHAR */ 5145 5146 5147 /* Loop through the string, looking for a place to start matching. */ 5148 for (;;) 5149 { 5150 /* If a fastmap is supplied, skip quickly over characters that 5151 cannot be the start of a match. If the pattern can match the 5152 null string, however, we don't need to skip characters; we want 5153 the first null string. */ 5154 if (fastmap && startpos < total_size && !bufp->can_be_null) 5155 { 5156 if (range > 0) /* Searching forwards. */ 5157 { 5158 register const char *d; 5159 register int lim = 0; 5160 int irange = range; 5161 5162 if (startpos < size1 && startpos + range >= size1) 5163 lim = range - (size1 - startpos); 5164 5165 d = (startpos >= size1 ? string2 - size1 : string1) + startpos; 5166 5167 /* Written out as an if-else to avoid testing `translate' 5168 inside the loop. */ 5169 if (translate) 5170 while (range > lim 5171 && !fastmap[(unsigned char) 5172 translate[(unsigned char) *d++]]) 5173 range--; 5174 else 5175 while (range > lim && !fastmap[(unsigned char) *d++]) 5176 range--; 5177 5178 startpos += irange - range; 5179 } 5180 else /* Searching backwards. */ 5181 { 5182 register CHAR_T c = (size1 == 0 || startpos >= size1 5183 ? string2[startpos - size1] 5184 : string1[startpos]); 5185 5186 if (!fastmap[(unsigned char) TRANSLATE (c)]) 5187 goto advance; 5188 } 5189 } 5190 5191 /* If can't match the null string, and that's all we have left, fail. */ 5192 if (range >= 0 && startpos == total_size && fastmap 5193 && !bufp->can_be_null) 5194 { 5195 #ifdef WCHAR 5196 FREE_WCS_BUFFERS (); 5197 #endif 5198 return -1; 5199 } 5200 5201 #ifdef WCHAR 5202 val = wcs_re_match_2_internal (bufp, string1, size1, string2, 5203 size2, startpos, regs, stop, 5204 wcs_string1, wcs_size1, 5205 wcs_string2, wcs_size2, 5206 mbs_offset1, mbs_offset2); 5207 #else /* BYTE */ 5208 val = byte_re_match_2_internal (bufp, string1, size1, string2, 5209 size2, startpos, regs, stop); 5210 #endif /* BYTE */ 5211 5212 #ifndef REGEX_MALLOC 5213 # ifdef C_ALLOCA 5214 alloca (0); 5215 # endif 5216 #endif 5217 5218 if (val >= 0) 5219 { 5220 #ifdef WCHAR 5221 FREE_WCS_BUFFERS (); 5222 #endif 5223 return startpos; 5224 } 5225 5226 if (val == -2) 5227 { 5228 #ifdef WCHAR 5229 FREE_WCS_BUFFERS (); 5230 #endif 5231 return -2; 5232 } 5233 5234 advance: 5235 if (!range) 5236 break; 5237 else if (range > 0) 5238 { 5239 range--; 5240 startpos++; 5241 } 5242 else 5243 { 5244 range++; 5245 startpos--; 5246 } 5247 } 5248 #ifdef WCHAR 5249 FREE_WCS_BUFFERS (); 5250 #endif 5251 return -1; 5252 } 5253 5254 #ifdef WCHAR 5255 /* This converts PTR, a pointer into one of the search wchar_t strings 5256 `string1' and `string2' into an multibyte string offset from the 5257 beginning of that string. We use mbs_offset to optimize. 5258 See convert_mbs_to_wcs. */ 5259 # define POINTER_TO_OFFSET(ptr) \ 5260 (FIRST_STRING_P (ptr) \ 5261 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \ 5262 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \ 5263 + csize1))) 5264 #else /* BYTE */ 5265 /* This converts PTR, a pointer into one of the search strings `string1' 5266 and `string2' into an offset from the beginning of that string. */ 5267 # define POINTER_TO_OFFSET(ptr) \ 5268 (FIRST_STRING_P (ptr) \ 5269 ? ((regoff_t) ((ptr) - string1)) \ 5270 : ((regoff_t) ((ptr) - string2 + size1))) 5271 #endif /* WCHAR */ 5272 5273 /* Macros for dealing with the split strings in re_match_2. */ 5274 5275 #define MATCHING_IN_FIRST_STRING (dend == end_match_1) 5276 5277 /* Call before fetching a character with *d. This switches over to 5278 string2 if necessary. */ 5279 #define PREFETCH() \ 5280 while (d == dend) \ 5281 { \ 5282 /* End of string2 => fail. */ \ 5283 if (dend == end_match_2) \ 5284 goto fail; \ 5285 /* End of string1 => advance to string2. */ \ 5286 d = string2; \ 5287 dend = end_match_2; \ 5288 } 5289 5290 /* Test if at very beginning or at very end of the virtual concatenation 5291 of `string1' and `string2'. If only one string, it's `string2'. */ 5292 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) 5293 #define AT_STRINGS_END(d) ((d) == end2) 5294 5295 5296 /* Test if D points to a character which is word-constituent. We have 5297 two special cases to check for: if past the end of string1, look at 5298 the first character in string2; and if before the beginning of 5299 string2, look at the last character in string1. */ 5300 #ifdef WCHAR 5301 /* Use internationalized API instead of SYNTAX. */ 5302 # define WORDCHAR_P(d) \ 5303 (iswalnum ((wint_t)((d) == end1 ? *string2 \ 5304 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \ 5305 || ((d) == end1 ? *string2 \ 5306 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_') 5307 #else /* BYTE */ 5308 # define WORDCHAR_P(d) \ 5309 (SYNTAX ((d) == end1 ? *string2 \ 5310 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ 5311 == Sword) 5312 #endif /* WCHAR */ 5313 5314 /* Disabled due to a compiler bug -- see comment at case wordbound */ 5315 #if 0 5316 /* Test if the character before D and the one at D differ with respect 5317 to being word-constituent. */ 5318 #define AT_WORD_BOUNDARY(d) \ 5319 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ 5320 || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) 5321 #endif 5322 5323 /* Free everything we malloc. */ 5324 #ifdef MATCH_MAY_ALLOCATE 5325 # ifdef WCHAR 5326 # define FREE_VARIABLES() \ 5327 do { \ 5328 REGEX_FREE_STACK (fail_stack.stack); \ 5329 FREE_VAR (regstart); \ 5330 FREE_VAR (regend); \ 5331 FREE_VAR (old_regstart); \ 5332 FREE_VAR (old_regend); \ 5333 FREE_VAR (best_regstart); \ 5334 FREE_VAR (best_regend); \ 5335 FREE_VAR (reg_info); \ 5336 FREE_VAR (reg_dummy); \ 5337 FREE_VAR (reg_info_dummy); \ 5338 if (!cant_free_wcs_buf) \ 5339 { \ 5340 FREE_VAR (string1); \ 5341 FREE_VAR (string2); \ 5342 FREE_VAR (mbs_offset1); \ 5343 FREE_VAR (mbs_offset2); \ 5344 } \ 5345 } while (0) 5346 # else /* BYTE */ 5347 # define FREE_VARIABLES() \ 5348 do { \ 5349 REGEX_FREE_STACK (fail_stack.stack); \ 5350 FREE_VAR (regstart); \ 5351 FREE_VAR (regend); \ 5352 FREE_VAR (old_regstart); \ 5353 FREE_VAR (old_regend); \ 5354 FREE_VAR (best_regstart); \ 5355 FREE_VAR (best_regend); \ 5356 FREE_VAR (reg_info); \ 5357 FREE_VAR (reg_dummy); \ 5358 FREE_VAR (reg_info_dummy); \ 5359 } while (0) 5360 # endif /* WCHAR */ 5361 #else 5362 # ifdef WCHAR 5363 # define FREE_VARIABLES() \ 5364 do { \ 5365 if (!cant_free_wcs_buf) \ 5366 { \ 5367 FREE_VAR (string1); \ 5368 FREE_VAR (string2); \ 5369 FREE_VAR (mbs_offset1); \ 5370 FREE_VAR (mbs_offset2); \ 5371 } \ 5372 } while (0) 5373 # else /* BYTE */ 5374 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ 5375 # endif /* WCHAR */ 5376 #endif /* not MATCH_MAY_ALLOCATE */ 5377 5378 /* These values must meet several constraints. They must not be valid 5379 register values; since we have a limit of 255 registers (because 5380 we use only one byte in the pattern for the register number), we can 5381 use numbers larger than 255. They must differ by 1, because of 5382 NUM_FAILURE_ITEMS above. And the value for the lowest register must 5383 be larger than the value for the highest register, so we do not try 5384 to actually save any registers when none are active. */ 5385 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) 5386 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) 5387 5388 #else /* not INSIDE_RECURSION */ 5389 /* Matching routines. */ 5390 5391 #ifndef emacs /* Emacs never uses this. */ 5392 /* re_match is like re_match_2 except it takes only a single string. */ 5393 5394 int 5395 re_match (struct re_pattern_buffer *bufp, const char *string, 5396 int size, int pos, struct re_registers *regs) 5397 { 5398 int result; 5399 # ifdef MBS_SUPPORT 5400 if (MB_CUR_MAX != 1) 5401 result = wcs_re_match_2_internal (bufp, NULL, 0, string, size, 5402 pos, regs, size, 5403 NULL, 0, NULL, 0, NULL, NULL); 5404 else 5405 # endif 5406 result = byte_re_match_2_internal (bufp, NULL, 0, string, size, 5407 pos, regs, size); 5408 # ifndef REGEX_MALLOC 5409 # ifdef C_ALLOCA 5410 alloca (0); 5411 # endif 5412 # endif 5413 return result; 5414 } 5415 # ifdef _LIBC 5416 weak_alias (__re_match, re_match) 5417 # endif 5418 #endif /* not emacs */ 5419 5420 #endif /* not INSIDE_RECURSION */ 5421 5422 #ifdef INSIDE_RECURSION 5423 static boolean PREFIX(group_match_null_string_p) (UCHAR_T **p, 5424 UCHAR_T *end, 5425 PREFIX(register_info_type) *reg_info); 5426 static boolean PREFIX(alt_match_null_string_p) (UCHAR_T *p, 5427 UCHAR_T *end, 5428 PREFIX(register_info_type) *reg_info); 5429 static boolean PREFIX(common_op_match_null_string_p) (UCHAR_T **p, 5430 UCHAR_T *end, 5431 PREFIX(register_info_type) *reg_info); 5432 static int PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2, 5433 int len, char *translate); 5434 #else /* not INSIDE_RECURSION */ 5435 5436 /* re_match_2 matches the compiled pattern in BUFP against the 5437 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 5438 and SIZE2, respectively). We start matching at POS, and stop 5439 matching at STOP. 5440 5441 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we 5442 store offsets for the substring each group matched in REGS. See the 5443 documentation for exactly how many groups we fill. 5444 5445 We return -1 if no match, -2 if an internal error (such as the 5446 failure stack overflowing). Otherwise, we return the length of the 5447 matched substring. */ 5448 5449 int 5450 re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, 5451 const char *string2, int size2, int pos, 5452 struct re_registers *regs, int stop) 5453 { 5454 int result; 5455 # ifdef MBS_SUPPORT 5456 if (MB_CUR_MAX != 1) 5457 result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2, 5458 pos, regs, stop, 5459 NULL, 0, NULL, 0, NULL, NULL); 5460 else 5461 # endif 5462 result = byte_re_match_2_internal (bufp, string1, size1, string2, size2, 5463 pos, regs, stop); 5464 5465 #ifndef REGEX_MALLOC 5466 # ifdef C_ALLOCA 5467 alloca (0); 5468 # endif 5469 #endif 5470 return result; 5471 } 5472 #ifdef _LIBC 5473 weak_alias (__re_match_2, re_match_2) 5474 #endif 5475 5476 #endif /* not INSIDE_RECURSION */ 5477 5478 #ifdef INSIDE_RECURSION 5479 5480 #ifdef WCHAR 5481 static int count_mbs_length (int *, int); 5482 5483 /* This check the substring (from 0, to length) of the multibyte string, 5484 to which offset_buffer correspond. And count how many wchar_t_characters 5485 the substring occupy. We use offset_buffer to optimization. 5486 See convert_mbs_to_wcs. */ 5487 5488 static int 5489 count_mbs_length(int *offset_buffer, int length) 5490 { 5491 int upper, lower; 5492 5493 /* Check whether the size is valid. */ 5494 if (length < 0) 5495 return -1; 5496 5497 if (offset_buffer == NULL) 5498 return 0; 5499 5500 /* If there are no multibyte character, offset_buffer[i] == i. 5501 Optmize for this case. */ 5502 if (offset_buffer[length] == length) 5503 return length; 5504 5505 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */ 5506 upper = length; 5507 lower = 0; 5508 5509 while (true) 5510 { 5511 int middle = (lower + upper) / 2; 5512 if (middle == lower || middle == upper) 5513 break; 5514 if (offset_buffer[middle] > length) 5515 upper = middle; 5516 else if (offset_buffer[middle] < length) 5517 lower = middle; 5518 else 5519 return middle; 5520 } 5521 5522 return -1; 5523 } 5524 #endif /* WCHAR */ 5525 5526 /* This is a separate function so that we can force an alloca cleanup 5527 afterwards. */ 5528 #ifdef WCHAR 5529 static int 5530 wcs_re_match_2_internal (struct re_pattern_buffer *bufp, 5531 const char *cstring1, int csize1, 5532 const char *cstring2, int csize2, 5533 int pos, 5534 struct re_registers *regs, 5535 int stop, 5536 /* string1 == string2 == NULL means string1/2, size1/2 and 5537 mbs_offset1/2 need seting up in this function. */ 5538 /* We need wchar_t* buffers correspond to cstring1, cstring2. */ 5539 wchar_t *string1, int size1, 5540 wchar_t *string2, int size2, 5541 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ 5542 int *mbs_offset1, int *mbs_offset2) 5543 #else /* BYTE */ 5544 static int 5545 byte_re_match_2_internal (struct re_pattern_buffer *bufp, 5546 const char *string1, int size1, 5547 const char *string2, int size2, 5548 int pos, 5549 struct re_registers *regs, int stop) 5550 #endif /* BYTE */ 5551 { 5552 /* General temporaries. */ 5553 int mcnt; 5554 UCHAR_T *p1; 5555 #ifdef WCHAR 5556 /* They hold whether each wchar_t is binary data or not. */ 5557 char *is_binary = NULL; 5558 /* If true, we can't free string1/2, mbs_offset1/2. */ 5559 int cant_free_wcs_buf = 1; 5560 #endif /* WCHAR */ 5561 5562 /* Just past the end of the corresponding string. */ 5563 const CHAR_T *end1, *end2; 5564 5565 /* Pointers into string1 and string2, just past the last characters in 5566 each to consider matching. */ 5567 const CHAR_T *end_match_1, *end_match_2; 5568 5569 /* Where we are in the data, and the end of the current string. */ 5570 const CHAR_T *d, *dend; 5571 5572 /* Where we are in the pattern, and the end of the pattern. */ 5573 #ifdef WCHAR 5574 UCHAR_T *pattern, *p; 5575 register UCHAR_T *pend; 5576 #else /* BYTE */ 5577 UCHAR_T *p = bufp->buffer; 5578 register UCHAR_T *pend = p + bufp->used; 5579 #endif /* WCHAR */ 5580 5581 /* Mark the opcode just after a start_memory, so we can test for an 5582 empty subpattern when we get to the stop_memory. */ 5583 UCHAR_T *just_past_start_mem = 0; 5584 5585 /* We use this to map every character in the string. */ 5586 RE_TRANSLATE_TYPE translate = bufp->translate; 5587 5588 /* Failure point stack. Each place that can handle a failure further 5589 down the line pushes a failure point on this stack. It consists of 5590 restart, regend, and reg_info for all registers corresponding to 5591 the subexpressions we're currently inside, plus the number of such 5592 registers, and, finally, two char *'s. The first char * is where 5593 to resume scanning the pattern; the second one is where to resume 5594 scanning the strings. If the latter is zero, the failure point is 5595 a ``dummy''; if a failure happens and the failure point is a dummy, 5596 it gets discarded and the next next one is tried. */ 5597 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ 5598 PREFIX(fail_stack_type) fail_stack; 5599 #endif 5600 #ifdef DEBUG 5601 static unsigned failure_id; 5602 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; 5603 #endif 5604 5605 #ifdef REL_ALLOC 5606 /* This holds the pointer to the failure stack, when 5607 it is allocated relocatably. */ 5608 fail_stack_elt_t *failure_stack_ptr; 5609 #endif 5610 5611 /* We fill all the registers internally, independent of what we 5612 return, for use in backreferences. The number here includes 5613 an element for register zero. */ 5614 size_t num_regs = bufp->re_nsub + 1; 5615 5616 /* The currently active registers. */ 5617 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG; 5618 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG; 5619 5620 /* Information on the contents of registers. These are pointers into 5621 the input strings; they record just what was matched (on this 5622 attempt) by a subexpression part of the pattern, that is, the 5623 regnum-th regstart pointer points to where in the pattern we began 5624 matching and the regnum-th regend points to right after where we 5625 stopped matching the regnum-th subexpression. (The zeroth register 5626 keeps track of what the whole pattern matches.) */ 5627 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ 5628 const CHAR_T **regstart, **regend; 5629 #endif 5630 5631 /* If a group that's operated upon by a repetition operator fails to 5632 match anything, then the register for its start will need to be 5633 restored because it will have been set to wherever in the string we 5634 are when we last see its open-group operator. Similarly for a 5635 register's end. */ 5636 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ 5637 const CHAR_T **old_regstart, **old_regend; 5638 #endif 5639 5640 /* The is_active field of reg_info helps us keep track of which (possibly 5641 nested) subexpressions we are currently in. The matched_something 5642 field of reg_info[reg_num] helps us tell whether or not we have 5643 matched any of the pattern so far this time through the reg_num-th 5644 subexpression. These two fields get reset each time through any 5645 loop their register is in. */ 5646 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ 5647 PREFIX(register_info_type) *reg_info; 5648 #endif 5649 5650 /* The following record the register info as found in the above 5651 variables when we find a match better than any we've seen before. 5652 This happens as we backtrack through the failure points, which in 5653 turn happens only if we have not yet matched the entire string. */ 5654 unsigned best_regs_set = false; 5655 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ 5656 const CHAR_T **best_regstart, **best_regend; 5657 #endif 5658 5659 /* Logically, this is `best_regend[0]'. But we don't want to have to 5660 allocate space for that if we're not allocating space for anything 5661 else (see below). Also, we never need info about register 0 for 5662 any of the other register vectors, and it seems rather a kludge to 5663 treat `best_regend' differently than the rest. So we keep track of 5664 the end of the best match so far in a separate variable. We 5665 initialize this to NULL so that when we backtrack the first time 5666 and need to test it, it's not garbage. */ 5667 const CHAR_T *match_end = NULL; 5668 5669 /* This helps SET_REGS_MATCHED avoid doing redundant work. */ 5670 int set_regs_matched_done = 0; 5671 5672 /* Used when we pop values we don't care about. */ 5673 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ 5674 const CHAR_T **reg_dummy; 5675 PREFIX(register_info_type) *reg_info_dummy; 5676 #endif 5677 5678 #ifdef DEBUG 5679 /* Counts the total number of registers pushed. */ 5680 unsigned num_regs_pushed = 0; 5681 #endif 5682 5683 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n"); 5684 5685 INIT_FAIL_STACK (); 5686 5687 #ifdef MATCH_MAY_ALLOCATE 5688 /* Do not bother to initialize all the register variables if there are 5689 no groups in the pattern, as it takes a fair amount of time. If 5690 there are groups, we include space for register 0 (the whole 5691 pattern), even though we never use it, since it simplifies the 5692 array indexing. We should fix this. */ 5693 if (bufp->re_nsub) 5694 { 5695 regstart = REGEX_TALLOC (num_regs, const CHAR_T *); 5696 regend = REGEX_TALLOC (num_regs, const CHAR_T *); 5697 old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *); 5698 old_regend = REGEX_TALLOC (num_regs, const CHAR_T *); 5699 best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *); 5700 best_regend = REGEX_TALLOC (num_regs, const CHAR_T *); 5701 reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type)); 5702 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *); 5703 reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type)); 5704 5705 if (!(regstart && regend && old_regstart && old_regend && reg_info 5706 && best_regstart && best_regend && reg_dummy && reg_info_dummy)) 5707 { 5708 FREE_VARIABLES (); 5709 return -2; 5710 } 5711 } 5712 else 5713 { 5714 /* We must initialize all our variables to NULL, so that 5715 `FREE_VARIABLES' doesn't try to free them. */ 5716 regstart = regend = old_regstart = old_regend = best_regstart 5717 = best_regend = reg_dummy = NULL; 5718 reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL; 5719 } 5720 #endif /* MATCH_MAY_ALLOCATE */ 5721 5722 /* The starting position is bogus. */ 5723 #ifdef WCHAR 5724 if (pos < 0 || pos > csize1 + csize2) 5725 #else /* BYTE */ 5726 if (pos < 0 || pos > size1 + size2) 5727 #endif 5728 { 5729 FREE_VARIABLES (); 5730 return -1; 5731 } 5732 5733 #ifdef WCHAR 5734 /* Allocate wchar_t array for string1 and string2 and 5735 fill them with converted string. */ 5736 if (string1 == NULL && string2 == NULL) 5737 { 5738 /* We need seting up buffers here. */ 5739 5740 /* We must free wcs buffers in this function. */ 5741 cant_free_wcs_buf = 0; 5742 5743 if (csize1 != 0) 5744 { 5745 string1 = REGEX_TALLOC (csize1 + 1, CHAR_T); 5746 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int); 5747 is_binary = REGEX_TALLOC (csize1 + 1, char); 5748 if (!string1 || !mbs_offset1 || !is_binary) 5749 { 5750 FREE_VAR (string1); 5751 FREE_VAR (mbs_offset1); 5752 FREE_VAR (is_binary); 5753 return -2; 5754 } 5755 } 5756 if (csize2 != 0) 5757 { 5758 string2 = REGEX_TALLOC (csize2 + 1, CHAR_T); 5759 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int); 5760 is_binary = REGEX_TALLOC (csize2 + 1, char); 5761 if (!string2 || !mbs_offset2 || !is_binary) 5762 { 5763 FREE_VAR (string1); 5764 FREE_VAR (mbs_offset1); 5765 FREE_VAR (string2); 5766 FREE_VAR (mbs_offset2); 5767 FREE_VAR (is_binary); 5768 return -2; 5769 } 5770 size2 = convert_mbs_to_wcs(string2, cstring2, csize2, 5771 mbs_offset2, is_binary); 5772 string2[size2] = L'\0'; /* for a sentinel */ 5773 FREE_VAR (is_binary); 5774 } 5775 } 5776 5777 /* We need to cast pattern to (wchar_t*), because we casted this compiled 5778 pattern to (char*) in regex_compile. */ 5779 p = pattern = (CHAR_T*)bufp->buffer; 5780 pend = (CHAR_T*)(bufp->buffer + bufp->used); 5781 5782 #endif /* WCHAR */ 5783 5784 /* Initialize subexpression text positions to -1 to mark ones that no 5785 start_memory/stop_memory has been seen for. Also initialize the 5786 register information struct. */ 5787 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) 5788 { 5789 regstart[mcnt] = regend[mcnt] 5790 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; 5791 5792 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; 5793 IS_ACTIVE (reg_info[mcnt]) = 0; 5794 MATCHED_SOMETHING (reg_info[mcnt]) = 0; 5795 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; 5796 } 5797 5798 /* We move `string1' into `string2' if the latter's empty -- but not if 5799 `string1' is null. */ 5800 if (size2 == 0 && string1 != NULL) 5801 { 5802 string2 = string1; 5803 size2 = size1; 5804 string1 = 0; 5805 size1 = 0; 5806 #ifdef WCHAR 5807 mbs_offset2 = mbs_offset1; 5808 csize2 = csize1; 5809 mbs_offset1 = NULL; 5810 csize1 = 0; 5811 #endif 5812 } 5813 end1 = string1 + size1; 5814 end2 = string2 + size2; 5815 5816 /* Compute where to stop matching, within the two strings. */ 5817 #ifdef WCHAR 5818 if (stop <= csize1) 5819 { 5820 mcnt = count_mbs_length(mbs_offset1, stop); 5821 end_match_1 = string1 + mcnt; 5822 end_match_2 = string2; 5823 } 5824 else 5825 { 5826 if (stop > csize1 + csize2) 5827 stop = csize1 + csize2; 5828 end_match_1 = end1; 5829 mcnt = count_mbs_length(mbs_offset2, stop-csize1); 5830 end_match_2 = string2 + mcnt; 5831 } 5832 if (mcnt < 0) 5833 { /* count_mbs_length return error. */ 5834 FREE_VARIABLES (); 5835 return -1; 5836 } 5837 #else 5838 if (stop <= size1) 5839 { 5840 end_match_1 = string1 + stop; 5841 end_match_2 = string2; 5842 } 5843 else 5844 { 5845 end_match_1 = end1; 5846 end_match_2 = string2 + stop - size1; 5847 } 5848 #endif /* WCHAR */ 5849 5850 /* `p' scans through the pattern as `d' scans through the data. 5851 `dend' is the end of the input string that `d' points within. `d' 5852 is advanced into the following input string whenever necessary, but 5853 this happens before fetching; therefore, at the beginning of the 5854 loop, `d' can be pointing at the end of a string, but it cannot 5855 equal `string2'. */ 5856 #ifdef WCHAR 5857 if (size1 > 0 && pos <= csize1) 5858 { 5859 mcnt = count_mbs_length(mbs_offset1, pos); 5860 d = string1 + mcnt; 5861 dend = end_match_1; 5862 } 5863 else 5864 { 5865 mcnt = count_mbs_length(mbs_offset2, pos-csize1); 5866 d = string2 + mcnt; 5867 dend = end_match_2; 5868 } 5869 5870 if (mcnt < 0) 5871 { /* count_mbs_length return error. */ 5872 FREE_VARIABLES (); 5873 return -1; 5874 } 5875 #else 5876 if (size1 > 0 && pos <= size1) 5877 { 5878 d = string1 + pos; 5879 dend = end_match_1; 5880 } 5881 else 5882 { 5883 d = string2 + pos - size1; 5884 dend = end_match_2; 5885 } 5886 #endif /* WCHAR */ 5887 5888 DEBUG_PRINT1 ("The compiled pattern is:\n"); 5889 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); 5890 DEBUG_PRINT1 ("The string to match is: `"); 5891 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); 5892 DEBUG_PRINT1 ("'\n"); 5893 5894 /* This loops over pattern commands. It exits by returning from the 5895 function if the match is complete, or it drops through if the match 5896 fails at this starting point in the input data. */ 5897 for (;;) 5898 { 5899 #ifdef _LIBC 5900 DEBUG_PRINT2 ("\n%p: ", p); 5901 #else 5902 DEBUG_PRINT2 ("\n0x%x: ", p); 5903 #endif 5904 5905 if (p == pend) 5906 { /* End of pattern means we might have succeeded. */ 5907 DEBUG_PRINT1 ("end of pattern ... "); 5908 5909 /* If we haven't matched the entire string, and we want the 5910 longest match, try backtracking. */ 5911 if (d != end_match_2) 5912 { 5913 /* 1 if this match ends in the same string (string1 or string2) 5914 as the best previous match. */ 5915 boolean same_str_p; 5916 5917 /* 1 if this match is the best seen so far. */ 5918 boolean best_match_p; 5919 5920 same_str_p = (FIRST_STRING_P (match_end) 5921 == MATCHING_IN_FIRST_STRING); 5922 5923 /* AIX compiler got confused when this was combined 5924 with the previous declaration. */ 5925 if (same_str_p) 5926 best_match_p = d > match_end; 5927 else 5928 best_match_p = !MATCHING_IN_FIRST_STRING; 5929 5930 DEBUG_PRINT1 ("backtracking.\n"); 5931 5932 if (!FAIL_STACK_EMPTY ()) 5933 { /* More failure points to try. */ 5934 5935 /* If exceeds best match so far, save it. */ 5936 if (!best_regs_set || best_match_p) 5937 { 5938 best_regs_set = true; 5939 match_end = d; 5940 5941 DEBUG_PRINT1 ("\nSAVING match as best so far.\n"); 5942 5943 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) 5944 { 5945 best_regstart[mcnt] = regstart[mcnt]; 5946 best_regend[mcnt] = regend[mcnt]; 5947 } 5948 } 5949 goto fail; 5950 } 5951 5952 /* If no failure points, don't restore garbage. And if 5953 last match is real best match, don't restore second 5954 best one. */ 5955 else if (best_regs_set && !best_match_p) 5956 { 5957 restore_best_regs: 5958 /* Restore best match. It may happen that `dend == 5959 end_match_1' while the restored d is in string2. 5960 For example, the pattern `x.*y.*z' against the 5961 strings `x-' and `y-z-', if the two strings are 5962 not consecutive in memory. */ 5963 DEBUG_PRINT1 ("Restoring best registers.\n"); 5964 5965 d = match_end; 5966 dend = ((d >= string1 && d <= end1) 5967 ? end_match_1 : end_match_2); 5968 5969 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) 5970 { 5971 regstart[mcnt] = best_regstart[mcnt]; 5972 regend[mcnt] = best_regend[mcnt]; 5973 } 5974 } 5975 } /* d != end_match_2 */ 5976 5977 succeed_label: 5978 DEBUG_PRINT1 ("Accepting match.\n"); 5979 /* If caller wants register contents data back, do it. */ 5980 if (regs && !bufp->no_sub) 5981 { 5982 /* Have the register data arrays been allocated? */ 5983 if (bufp->regs_allocated == REGS_UNALLOCATED) 5984 { /* No. So allocate them with malloc. We need one 5985 extra element beyond `num_regs' for the `-1' marker 5986 GNU code uses. */ 5987 regs->num_regs = MAX (RE_NREGS, num_regs + 1); 5988 regs->start = TALLOC (regs->num_regs, regoff_t); 5989 regs->end = TALLOC (regs->num_regs, regoff_t); 5990 if (regs->start == NULL || regs->end == NULL) 5991 { 5992 FREE_VARIABLES (); 5993 return -2; 5994 } 5995 bufp->regs_allocated = REGS_REALLOCATE; 5996 } 5997 else if (bufp->regs_allocated == REGS_REALLOCATE) 5998 { /* Yes. If we need more elements than were already 5999 allocated, reallocate them. If we need fewer, just 6000 leave it alone. */ 6001 if (regs->num_regs < num_regs + 1) 6002 { 6003 regs->num_regs = num_regs + 1; 6004 RETALLOC (regs->start, regs->num_regs, regoff_t); 6005 RETALLOC (regs->end, regs->num_regs, regoff_t); 6006 if (regs->start == NULL || regs->end == NULL) 6007 { 6008 FREE_VARIABLES (); 6009 return -2; 6010 } 6011 } 6012 } 6013 else 6014 { 6015 /* These braces fend off a "empty body in an else-statement" 6016 warning under GCC when assert expands to nothing. */ 6017 assert (bufp->regs_allocated == REGS_FIXED); 6018 } 6019 6020 /* Convert the pointer data in `regstart' and `regend' to 6021 indices. Register zero has to be set differently, 6022 since we haven't kept track of any info for it. */ 6023 if (regs->num_regs > 0) 6024 { 6025 regs->start[0] = pos; 6026 #ifdef WCHAR 6027 if (MATCHING_IN_FIRST_STRING) 6028 regs->end[0] = mbs_offset1 != NULL ? 6029 mbs_offset1[d-string1] : 0; 6030 else 6031 regs->end[0] = csize1 + (mbs_offset2 != NULL ? 6032 mbs_offset2[d-string2] : 0); 6033 #else 6034 regs->end[0] = (MATCHING_IN_FIRST_STRING 6035 ? ((regoff_t) (d - string1)) 6036 : ((regoff_t) (d - string2 + size1))); 6037 #endif /* WCHAR */ 6038 } 6039 6040 /* Go through the first `min (num_regs, regs->num_regs)' 6041 registers, since that is all we initialized. */ 6042 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs); 6043 mcnt++) 6044 { 6045 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) 6046 regs->start[mcnt] = regs->end[mcnt] = -1; 6047 else 6048 { 6049 regs->start[mcnt] 6050 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); 6051 regs->end[mcnt] 6052 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); 6053 } 6054 } 6055 6056 /* If the regs structure we return has more elements than 6057 were in the pattern, set the extra elements to -1. If 6058 we (re)allocated the registers, this is the case, 6059 because we always allocate enough to have at least one 6060 -1 at the end. */ 6061 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++) 6062 regs->start[mcnt] = regs->end[mcnt] = -1; 6063 } /* regs && !bufp->no_sub */ 6064 6065 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n", 6066 nfailure_points_pushed, nfailure_points_popped, 6067 nfailure_points_pushed - nfailure_points_popped); 6068 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed); 6069 6070 #ifdef WCHAR 6071 if (MATCHING_IN_FIRST_STRING) 6072 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0; 6073 else 6074 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) + 6075 csize1; 6076 mcnt -= pos; 6077 #else 6078 mcnt = d - pos - (MATCHING_IN_FIRST_STRING 6079 ? string1 6080 : string2 - size1); 6081 #endif /* WCHAR */ 6082 6083 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt); 6084 6085 FREE_VARIABLES (); 6086 return mcnt; 6087 } 6088 6089 /* Otherwise match next pattern command. */ 6090 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) 6091 { 6092 /* Ignore these. Used to ignore the n of succeed_n's which 6093 currently have n == 0. */ 6094 case no_op: 6095 DEBUG_PRINT1 ("EXECUTING no_op.\n"); 6096 break; 6097 6098 case succeed: 6099 DEBUG_PRINT1 ("EXECUTING succeed.\n"); 6100 goto succeed_label; 6101 6102 /* Match the next n pattern characters exactly. The following 6103 byte in the pattern defines n, and the n bytes after that 6104 are the characters to match. */ 6105 case exactn: 6106 #ifdef MBS_SUPPORT 6107 case exactn_bin: 6108 #endif 6109 mcnt = *p++; 6110 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt); 6111 6112 /* This is written out as an if-else so we don't waste time 6113 testing `translate' inside the loop. */ 6114 if (translate) 6115 { 6116 do 6117 { 6118 PREFETCH (); 6119 #ifdef WCHAR 6120 if (*d <= 0xff) 6121 { 6122 if ((UCHAR_T) translate[(unsigned char) *d++] 6123 != (UCHAR_T) *p++) 6124 goto fail; 6125 } 6126 else 6127 { 6128 if (*d++ != (CHAR_T) *p++) 6129 goto fail; 6130 } 6131 #else 6132 if ((UCHAR_T) translate[(unsigned char) *d++] 6133 != (UCHAR_T) *p++) 6134 goto fail; 6135 #endif /* WCHAR */ 6136 } 6137 while (--mcnt); 6138 } 6139 else 6140 { 6141 do 6142 { 6143 PREFETCH (); 6144 if (*d++ != (CHAR_T) *p++) goto fail; 6145 } 6146 while (--mcnt); 6147 } 6148 SET_REGS_MATCHED (); 6149 break; 6150 6151 6152 /* Match any character except possibly a newline or a null. */ 6153 case anychar: 6154 DEBUG_PRINT1 ("EXECUTING anychar.\n"); 6155 6156 PREFETCH (); 6157 6158 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') 6159 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) 6160 goto fail; 6161 6162 SET_REGS_MATCHED (); 6163 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d); 6164 d++; 6165 break; 6166 6167 6168 case charset: 6169 case charset_not: 6170 { 6171 register UCHAR_T c; 6172 #ifdef WCHAR 6173 unsigned int i, char_class_length, coll_symbol_length, 6174 equiv_class_length, ranges_length, chars_length, length; 6175 CHAR_T *workp, *workp2, *charset_top; 6176 #define WORK_BUFFER_SIZE 128 6177 CHAR_T str_buf[WORK_BUFFER_SIZE]; 6178 # ifdef _LIBC 6179 uint32_t nrules; 6180 # endif /* _LIBC */ 6181 #endif /* WCHAR */ 6182 boolean negate = (re_opcode_t) *(p - 1) == charset_not; 6183 6184 DEBUG_PRINT2 ("EXECUTING charset%s.\n", negate ? "_not" : ""); 6185 PREFETCH (); 6186 c = TRANSLATE (*d); /* The character to match. */ 6187 #ifdef WCHAR 6188 # ifdef _LIBC 6189 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); 6190 # endif /* _LIBC */ 6191 charset_top = p - 1; 6192 char_class_length = *p++; 6193 coll_symbol_length = *p++; 6194 equiv_class_length = *p++; 6195 ranges_length = *p++; 6196 chars_length = *p++; 6197 /* p points charset[6], so the address of the next instruction 6198 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'], 6199 where l=length of char_classes, m=length of collating_symbol, 6200 n=equivalence_class, o=length of char_range, 6201 p'=length of character. */ 6202 workp = p; 6203 /* Update p to indicate the next instruction. */ 6204 p += char_class_length + coll_symbol_length+ equiv_class_length + 6205 2*ranges_length + chars_length; 6206 6207 /* match with char_class? */ 6208 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE) 6209 { 6210 wctype_t wctype; 6211 uintptr_t alignedp = ((uintptr_t)workp 6212 + __alignof__(wctype_t) - 1) 6213 & ~(uintptr_t)(__alignof__(wctype_t) - 1); 6214 wctype = *((wctype_t*)alignedp); 6215 workp += CHAR_CLASS_SIZE; 6216 # ifdef _LIBC 6217 if (__iswctype((wint_t)c, wctype)) 6218 goto char_set_matched; 6219 # else 6220 if (iswctype((wint_t)c, wctype)) 6221 goto char_set_matched; 6222 # endif 6223 } 6224 6225 /* match with collating_symbol? */ 6226 # ifdef _LIBC 6227 if (nrules != 0) 6228 { 6229 const unsigned char *extra = (const unsigned char *) 6230 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); 6231 6232 for (workp2 = workp + coll_symbol_length ; workp < workp2 ; 6233 workp++) 6234 { 6235 int32_t *wextra; 6236 wextra = (int32_t*)(extra + *workp++); 6237 for (i = 0; i < *wextra; ++i) 6238 if (TRANSLATE(d[i]) != wextra[1 + i]) 6239 break; 6240 6241 if (i == *wextra) 6242 { 6243 /* Update d, however d will be incremented at 6244 char_set_matched:, we decrement d here. */ 6245 d += i - 1; 6246 goto char_set_matched; 6247 } 6248 } 6249 } 6250 else /* (nrules == 0) */ 6251 # endif 6252 /* If we can't look up collation data, we use wcscoll 6253 instead. */ 6254 { 6255 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;) 6256 { 6257 const CHAR_T *backup_d = d, *backup_dend = dend; 6258 # ifdef _LIBC 6259 length = __wcslen (workp); 6260 # else 6261 length = wcslen (workp); 6262 # endif 6263 6264 /* If wcscoll(the collating symbol, whole string) > 0, 6265 any substring of the string never match with the 6266 collating symbol. */ 6267 # ifdef _LIBC 6268 if (__wcscoll (workp, d) > 0) 6269 # else 6270 if (wcscoll (workp, d) > 0) 6271 # endif 6272 { 6273 workp += length + 1; 6274 continue; 6275 } 6276 6277 /* First, we compare the collating symbol with 6278 the first character of the string. 6279 If it don't match, we add the next character to 6280 the compare buffer in turn. */ 6281 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++) 6282 { 6283 int match; 6284 if (d == dend) 6285 { 6286 if (dend == end_match_2) 6287 break; 6288 d = string2; 6289 dend = end_match_2; 6290 } 6291 6292 /* add next character to the compare buffer. */ 6293 str_buf[i] = TRANSLATE(*d); 6294 str_buf[i+1] = '\0'; 6295 6296 # ifdef _LIBC 6297 match = __wcscoll (workp, str_buf); 6298 # else 6299 match = wcscoll (workp, str_buf); 6300 # endif 6301 if (match == 0) 6302 goto char_set_matched; 6303 6304 if (match < 0) 6305 /* (str_buf > workp) indicate (str_buf + X > workp), 6306 because for all X (str_buf + X > str_buf). 6307 So we don't need continue this loop. */ 6308 break; 6309 6310 /* Otherwise(str_buf < workp), 6311 (str_buf+next_character) may equals (workp). 6312 So we continue this loop. */ 6313 } 6314 /* not matched */ 6315 d = backup_d; 6316 dend = backup_dend; 6317 workp += length + 1; 6318 } 6319 } 6320 /* match with equivalence_class? */ 6321 # ifdef _LIBC 6322 if (nrules != 0) 6323 { 6324 const CHAR_T *backup_d = d, *backup_dend = dend; 6325 /* Try to match the equivalence class against 6326 those known to the collate implementation. */ 6327 const int32_t *table; 6328 const int32_t *weights; 6329 const int32_t *extra; 6330 const int32_t *indirect; 6331 int32_t idx, idx2; 6332 wint_t *cp; 6333 size_t len; 6334 6335 /* This #include defines a local function! */ 6336 # include <locale/weightwc.h> 6337 6338 table = (const int32_t *) 6339 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC); 6340 weights = (const wint_t *) 6341 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC); 6342 extra = (const wint_t *) 6343 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC); 6344 indirect = (const int32_t *) 6345 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC); 6346 6347 /* Write 1 collating element to str_buf, and 6348 get its index. */ 6349 idx2 = 0; 6350 6351 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++) 6352 { 6353 cp = (wint_t*)str_buf; 6354 if (d == dend) 6355 { 6356 if (dend == end_match_2) 6357 break; 6358 d = string2; 6359 dend = end_match_2; 6360 } 6361 str_buf[i] = TRANSLATE(*(d+i)); 6362 str_buf[i+1] = '\0'; /* sentinel */ 6363 idx2 = findidx ((const wint_t**)&cp); 6364 } 6365 6366 /* Update d, however d will be incremented at 6367 char_set_matched:, we decrement d here. */ 6368 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1); 6369 if (d >= dend) 6370 { 6371 if (dend == end_match_2) 6372 d = dend; 6373 else 6374 { 6375 d = string2; 6376 dend = end_match_2; 6377 } 6378 } 6379 6380 len = weights[idx2]; 6381 6382 for (workp2 = workp + equiv_class_length ; workp < workp2 ; 6383 workp++) 6384 { 6385 idx = (int32_t)*workp; 6386 /* We already checked idx != 0 in regex_compile. */ 6387 6388 if (idx2 != 0 && len == weights[idx]) 6389 { 6390 int cnt = 0; 6391 while (cnt < len && (weights[idx + 1 + cnt] 6392 == weights[idx2 + 1 + cnt])) 6393 ++cnt; 6394 6395 if (cnt == len) 6396 goto char_set_matched; 6397 } 6398 } 6399 /* not matched */ 6400 d = backup_d; 6401 dend = backup_dend; 6402 } 6403 else /* (nrules == 0) */ 6404 # endif 6405 /* If we can't look up collation data, we use wcscoll 6406 instead. */ 6407 { 6408 for (workp2 = workp + equiv_class_length ; workp < workp2 ;) 6409 { 6410 const CHAR_T *backup_d = d, *backup_dend = dend; 6411 # ifdef _LIBC 6412 length = __wcslen (workp); 6413 # else 6414 length = wcslen (workp); 6415 # endif 6416 6417 /* If wcscoll(the collating symbol, whole string) > 0, 6418 any substring of the string never match with the 6419 collating symbol. */ 6420 # ifdef _LIBC 6421 if (__wcscoll (workp, d) > 0) 6422 # else 6423 if (wcscoll (workp, d) > 0) 6424 # endif 6425 { 6426 workp += length + 1; 6427 break; 6428 } 6429 6430 /* First, we compare the equivalence class with 6431 the first character of the string. 6432 If it don't match, we add the next character to 6433 the compare buffer in turn. */ 6434 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++) 6435 { 6436 int match; 6437 if (d == dend) 6438 { 6439 if (dend == end_match_2) 6440 break; 6441 d = string2; 6442 dend = end_match_2; 6443 } 6444 6445 /* add next character to the compare buffer. */ 6446 str_buf[i] = TRANSLATE(*d); 6447 str_buf[i+1] = '\0'; 6448 6449 # ifdef _LIBC 6450 match = __wcscoll (workp, str_buf); 6451 # else 6452 match = wcscoll (workp, str_buf); 6453 # endif 6454 6455 if (match == 0) 6456 goto char_set_matched; 6457 6458 if (match < 0) 6459 /* (str_buf > workp) indicate (str_buf + X > workp), 6460 because for all X (str_buf + X > str_buf). 6461 So we don't need continue this loop. */ 6462 break; 6463 6464 /* Otherwise(str_buf < workp), 6465 (str_buf+next_character) may equals (workp). 6466 So we continue this loop. */ 6467 } 6468 /* not matched */ 6469 d = backup_d; 6470 dend = backup_dend; 6471 workp += length + 1; 6472 } 6473 } 6474 6475 /* match with char_range? */ 6476 # ifdef _LIBC 6477 if (nrules != 0) 6478 { 6479 uint32_t collseqval; 6480 const char *collseq = (const char *) 6481 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC); 6482 6483 collseqval = collseq_table_lookup (collseq, c); 6484 6485 for (; workp < p - chars_length ;) 6486 { 6487 uint32_t start_val, end_val; 6488 6489 /* We already compute the collation sequence value 6490 of the characters (or collating symbols). */ 6491 start_val = (uint32_t) *workp++; /* range_start */ 6492 end_val = (uint32_t) *workp++; /* range_end */ 6493 6494 if (start_val <= collseqval && collseqval <= end_val) 6495 goto char_set_matched; 6496 } 6497 } 6498 else 6499 # endif 6500 { 6501 /* We set range_start_char at str_buf[0], range_end_char 6502 at str_buf[4], and compared char at str_buf[2]. */ 6503 str_buf[1] = 0; 6504 str_buf[2] = c; 6505 str_buf[3] = 0; 6506 str_buf[5] = 0; 6507 for (; workp < p - chars_length ;) 6508 { 6509 wchar_t *range_start_char, *range_end_char; 6510 6511 /* match if (range_start_char <= c <= range_end_char). */ 6512 6513 /* If range_start(or end) < 0, we assume -range_start(end) 6514 is the offset of the collating symbol which is specified 6515 as the character of the range start(end). */ 6516 6517 /* range_start */ 6518 if (*workp < 0) 6519 range_start_char = charset_top - (*workp++); 6520 else 6521 { 6522 str_buf[0] = *workp++; 6523 range_start_char = str_buf; 6524 } 6525 6526 /* range_end */ 6527 if (*workp < 0) 6528 range_end_char = charset_top - (*workp++); 6529 else 6530 { 6531 str_buf[4] = *workp++; 6532 range_end_char = str_buf + 4; 6533 } 6534 6535 # ifdef _LIBC 6536 if (__wcscoll (range_start_char, str_buf+2) <= 0 6537 && __wcscoll (str_buf+2, range_end_char) <= 0) 6538 # else 6539 if (wcscoll (range_start_char, str_buf+2) <= 0 6540 && wcscoll (str_buf+2, range_end_char) <= 0) 6541 # endif 6542 goto char_set_matched; 6543 } 6544 } 6545 6546 /* match with char? */ 6547 for (; workp < p ; workp++) 6548 if (c == *workp) 6549 goto char_set_matched; 6550 6551 negate = !negate; 6552 6553 char_set_matched: 6554 if (negate) goto fail; 6555 #else 6556 /* Cast to `unsigned' instead of `unsigned char' in case the 6557 bit list is a full 32 bytes long. */ 6558 if (c < (unsigned) (*p * BYTEWIDTH) 6559 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) 6560 negate = !negate; 6561 6562 p += 1 + *p; 6563 6564 if (!negate) goto fail; 6565 #undef WORK_BUFFER_SIZE 6566 #endif /* WCHAR */ 6567 SET_REGS_MATCHED (); 6568 d++; 6569 break; 6570 } 6571 6572 6573 /* The beginning of a group is represented by start_memory. 6574 The arguments are the register number in the next byte, and the 6575 number of groups inner to this one in the next. The text 6576 matched within the group is recorded (in the internal 6577 registers data structure) under the register number. */ 6578 case start_memory: 6579 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n", 6580 (long int) *p, (long int) p[1]); 6581 6582 /* Find out if this group can match the empty string. */ 6583 p1 = p; /* To send to group_match_null_string_p. */ 6584 6585 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) 6586 REG_MATCH_NULL_STRING_P (reg_info[*p]) 6587 = PREFIX(group_match_null_string_p) (&p1, pend, reg_info); 6588 6589 /* Save the position in the string where we were the last time 6590 we were at this open-group operator in case the group is 6591 operated upon by a repetition operator, e.g., with `(a*)*b' 6592 against `ab'; then we want to ignore where we are now in 6593 the string in case this attempt to match fails. */ 6594 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) 6595 ? REG_UNSET (regstart[*p]) ? d : regstart[*p] 6596 : regstart[*p]; 6597 DEBUG_PRINT2 (" old_regstart: %d\n", 6598 POINTER_TO_OFFSET (old_regstart[*p])); 6599 6600 regstart[*p] = d; 6601 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p])); 6602 6603 IS_ACTIVE (reg_info[*p]) = 1; 6604 MATCHED_SOMETHING (reg_info[*p]) = 0; 6605 6606 /* Clear this whenever we change the register activity status. */ 6607 set_regs_matched_done = 0; 6608 6609 /* This is the new highest active register. */ 6610 highest_active_reg = *p; 6611 6612 /* If nothing was active before, this is the new lowest active 6613 register. */ 6614 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) 6615 lowest_active_reg = *p; 6616 6617 /* Move past the register number and inner group count. */ 6618 p += 2; 6619 just_past_start_mem = p; 6620 6621 break; 6622 6623 6624 /* The stop_memory opcode represents the end of a group. Its 6625 arguments are the same as start_memory's: the register 6626 number, and the number of inner groups. */ 6627 case stop_memory: 6628 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n", 6629 (long int) *p, (long int) p[1]); 6630 6631 /* We need to save the string position the last time we were at 6632 this close-group operator in case the group is operated 6633 upon by a repetition operator, e.g., with `((a*)*(b*)*)*' 6634 against `aba'; then we want to ignore where we are now in 6635 the string in case this attempt to match fails. */ 6636 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) 6637 ? REG_UNSET (regend[*p]) ? d : regend[*p] 6638 : regend[*p]; 6639 DEBUG_PRINT2 (" old_regend: %d\n", 6640 POINTER_TO_OFFSET (old_regend[*p])); 6641 6642 regend[*p] = d; 6643 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p])); 6644 6645 /* This register isn't active anymore. */ 6646 IS_ACTIVE (reg_info[*p]) = 0; 6647 6648 /* Clear this whenever we change the register activity status. */ 6649 set_regs_matched_done = 0; 6650 6651 /* If this was the only register active, nothing is active 6652 anymore. */ 6653 if (lowest_active_reg == highest_active_reg) 6654 { 6655 lowest_active_reg = NO_LOWEST_ACTIVE_REG; 6656 highest_active_reg = NO_HIGHEST_ACTIVE_REG; 6657 } 6658 else 6659 { /* We must scan for the new highest active register, since 6660 it isn't necessarily one less than now: consider 6661 (a(b)c(d(e)f)g). When group 3 ends, after the f), the 6662 new highest active register is 1. */ 6663 UCHAR_T r = *p - 1; 6664 while (r > 0 && !IS_ACTIVE (reg_info[r])) 6665 r--; 6666 6667 /* If we end up at register zero, that means that we saved 6668 the registers as the result of an `on_failure_jump', not 6669 a `start_memory', and we jumped to past the innermost 6670 `stop_memory'. For example, in ((.)*) we save 6671 registers 1 and 2 as a result of the *, but when we pop 6672 back to the second ), we are at the stop_memory 1. 6673 Thus, nothing is active. */ 6674 if (r == 0) 6675 { 6676 lowest_active_reg = NO_LOWEST_ACTIVE_REG; 6677 highest_active_reg = NO_HIGHEST_ACTIVE_REG; 6678 } 6679 else 6680 highest_active_reg = r; 6681 } 6682 6683 /* If just failed to match something this time around with a 6684 group that's operated on by a repetition operator, try to 6685 force exit from the ``loop'', and restore the register 6686 information for this group that we had before trying this 6687 last match. */ 6688 if ((!MATCHED_SOMETHING (reg_info[*p]) 6689 || just_past_start_mem == p - 1) 6690 && (p + 2) < pend) 6691 { 6692 boolean is_a_jump_n = false; 6693 6694 p1 = p + 2; 6695 mcnt = 0; 6696 switch ((re_opcode_t) *p1++) 6697 { 6698 case jump_n: 6699 is_a_jump_n = true; 6700 /* Fall through. */ 6701 case pop_failure_jump: 6702 case maybe_pop_jump: 6703 case jump: 6704 case dummy_failure_jump: 6705 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 6706 if (is_a_jump_n) 6707 p1 += OFFSET_ADDRESS_SIZE; 6708 break; 6709 6710 default: 6711 /* do nothing */ ; 6712 } 6713 p1 += mcnt; 6714 6715 /* If the next operation is a jump backwards in the pattern 6716 to an on_failure_jump right before the start_memory 6717 corresponding to this stop_memory, exit from the loop 6718 by forcing a failure after pushing on the stack the 6719 on_failure_jump's jump in the pattern, and d. */ 6720 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump 6721 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory 6722 && p1[2+OFFSET_ADDRESS_SIZE] == *p) 6723 { 6724 /* If this group ever matched anything, then restore 6725 what its registers were before trying this last 6726 failed match, e.g., with `(a*)*b' against `ab' for 6727 regstart[1], and, e.g., with `((a*)*(b*)*)*' 6728 against `aba' for regend[3]. 6729 6730 Also restore the registers for inner groups for, 6731 e.g., `((a*)(b*))*' against `aba' (register 3 would 6732 otherwise get trashed). */ 6733 6734 if (EVER_MATCHED_SOMETHING (reg_info[*p])) 6735 { 6736 unsigned r; 6737 6738 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; 6739 6740 /* Restore this and inner groups' (if any) registers. */ 6741 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1); 6742 r++) 6743 { 6744 regstart[r] = old_regstart[r]; 6745 6746 /* xx why this test? */ 6747 if (old_regend[r] >= regstart[r]) 6748 regend[r] = old_regend[r]; 6749 } 6750 } 6751 p1++; 6752 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 6753 PUSH_FAILURE_POINT (p1 + mcnt, d, -2); 6754 6755 goto fail; 6756 } 6757 } 6758 6759 /* Move past the register number and the inner group count. */ 6760 p += 2; 6761 break; 6762 6763 6764 /* \<digit> has been turned into a `duplicate' command which is 6765 followed by the numeric value of <digit> as the register number. */ 6766 case duplicate: 6767 { 6768 register const CHAR_T *d2, *dend2; 6769 int regno = *p++; /* Get which register to match against. */ 6770 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); 6771 6772 /* Can't back reference a group which we've never matched. */ 6773 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) 6774 goto fail; 6775 6776 /* Where in input to try to start matching. */ 6777 d2 = regstart[regno]; 6778 6779 /* Where to stop matching; if both the place to start and 6780 the place to stop matching are in the same string, then 6781 set to the place to stop, otherwise, for now have to use 6782 the end of the first string. */ 6783 6784 dend2 = ((FIRST_STRING_P (regstart[regno]) 6785 == FIRST_STRING_P (regend[regno])) 6786 ? regend[regno] : end_match_1); 6787 for (;;) 6788 { 6789 /* If necessary, advance to next segment in register 6790 contents. */ 6791 while (d2 == dend2) 6792 { 6793 if (dend2 == end_match_2) break; 6794 if (dend2 == regend[regno]) break; 6795 6796 /* End of string1 => advance to string2. */ 6797 d2 = string2; 6798 dend2 = regend[regno]; 6799 } 6800 /* At end of register contents => success */ 6801 if (d2 == dend2) break; 6802 6803 /* If necessary, advance to next segment in data. */ 6804 PREFETCH (); 6805 6806 /* How many characters left in this segment to match. */ 6807 mcnt = dend - d; 6808 6809 /* Want how many consecutive characters we can match in 6810 one shot, so, if necessary, adjust the count. */ 6811 if (mcnt > dend2 - d2) 6812 mcnt = dend2 - d2; 6813 6814 /* Compare that many; failure if mismatch, else move 6815 past them. */ 6816 if (translate 6817 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate) 6818 : memcmp (d, d2, mcnt*sizeof(UCHAR_T))) 6819 goto fail; 6820 d += mcnt, d2 += mcnt; 6821 6822 /* Do this because we've match some characters. */ 6823 SET_REGS_MATCHED (); 6824 } 6825 } 6826 break; 6827 6828 6829 /* begline matches the empty string at the beginning of the string 6830 (unless `not_bol' is set in `bufp'), and, if 6831 `newline_anchor' is set, after newlines. */ 6832 case begline: 6833 DEBUG_PRINT1 ("EXECUTING begline.\n"); 6834 6835 if (AT_STRINGS_BEG (d)) 6836 { 6837 if (!bufp->not_bol) break; 6838 } 6839 else if (d[-1] == '\n' && bufp->newline_anchor) 6840 { 6841 break; 6842 } 6843 /* In all other cases, we fail. */ 6844 goto fail; 6845 6846 6847 /* endline is the dual of begline. */ 6848 case endline: 6849 DEBUG_PRINT1 ("EXECUTING endline.\n"); 6850 6851 if (AT_STRINGS_END (d)) 6852 { 6853 if (!bufp->not_eol) break; 6854 } 6855 6856 /* We have to ``prefetch'' the next character. */ 6857 else if ((d == end1 ? *string2 : *d) == '\n' 6858 && bufp->newline_anchor) 6859 { 6860 break; 6861 } 6862 goto fail; 6863 6864 6865 /* Match at the very beginning of the data. */ 6866 case begbuf: 6867 DEBUG_PRINT1 ("EXECUTING begbuf.\n"); 6868 if (AT_STRINGS_BEG (d)) 6869 break; 6870 goto fail; 6871 6872 6873 /* Match at the very end of the data. */ 6874 case endbuf: 6875 DEBUG_PRINT1 ("EXECUTING endbuf.\n"); 6876 if (AT_STRINGS_END (d)) 6877 break; 6878 goto fail; 6879 6880 6881 /* on_failure_keep_string_jump is used to optimize `.*\n'. It 6882 pushes NULL as the value for the string on the stack. Then 6883 `pop_failure_point' will keep the current value for the 6884 string, instead of restoring it. To see why, consider 6885 matching `foo\nbar' against `.*\n'. The .* matches the foo; 6886 then the . fails against the \n. But the next thing we want 6887 to do is match the \n against the \n; if we restored the 6888 string value, we would be back at the foo. 6889 6890 Because this is used only in specific cases, we don't need to 6891 check all the things that `on_failure_jump' does, to make 6892 sure the right things get saved on the stack. Hence we don't 6893 share its code. The only reason to push anything on the 6894 stack at all is that otherwise we would have to change 6895 `anychar's code to do something besides goto fail in this 6896 case; that seems worse than this. */ 6897 case on_failure_keep_string_jump: 6898 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); 6899 6900 EXTRACT_NUMBER_AND_INCR (mcnt, p); 6901 #ifdef _LIBC 6902 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt); 6903 #else 6904 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); 6905 #endif 6906 6907 PUSH_FAILURE_POINT (p + mcnt, NULL, -2); 6908 break; 6909 6910 6911 /* Uses of on_failure_jump: 6912 6913 Each alternative starts with an on_failure_jump that points 6914 to the beginning of the next alternative. Each alternative 6915 except the last ends with a jump that in effect jumps past 6916 the rest of the alternatives. (They really jump to the 6917 ending jump of the following alternative, because tensioning 6918 these jumps is a hassle.) 6919 6920 Repeats start with an on_failure_jump that points past both 6921 the repetition text and either the following jump or 6922 pop_failure_jump back to this on_failure_jump. */ 6923 case on_failure_jump: 6924 on_failure: 6925 DEBUG_PRINT1 ("EXECUTING on_failure_jump"); 6926 6927 EXTRACT_NUMBER_AND_INCR (mcnt, p); 6928 #ifdef _LIBC 6929 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt); 6930 #else 6931 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); 6932 #endif 6933 6934 /* If this on_failure_jump comes right before a group (i.e., 6935 the original * applied to a group), save the information 6936 for that group and all inner ones, so that if we fail back 6937 to this point, the group's information will be correct. 6938 For example, in \(a*\)*\1, we need the preceding group, 6939 and in \(zz\(a*\)b*\)\2, we need the inner group. */ 6940 6941 /* We can't use `p' to check ahead because we push 6942 a failure point to `p + mcnt' after we do this. */ 6943 p1 = p; 6944 6945 /* We need to skip no_op's before we look for the 6946 start_memory in case this on_failure_jump is happening as 6947 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 6948 against aba. */ 6949 while (p1 < pend && (re_opcode_t) *p1 == no_op) 6950 p1++; 6951 6952 if (p1 < pend && (re_opcode_t) *p1 == start_memory) 6953 { 6954 /* We have a new highest active register now. This will 6955 get reset at the start_memory we are about to get to, 6956 but we will have saved all the registers relevant to 6957 this repetition op, as described above. */ 6958 highest_active_reg = *(p1 + 1) + *(p1 + 2); 6959 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) 6960 lowest_active_reg = *(p1 + 1); 6961 } 6962 6963 DEBUG_PRINT1 (":\n"); 6964 PUSH_FAILURE_POINT (p + mcnt, d, -2); 6965 break; 6966 6967 6968 /* A smart repeat ends with `maybe_pop_jump'. 6969 We change it to either `pop_failure_jump' or `jump'. */ 6970 case maybe_pop_jump: 6971 EXTRACT_NUMBER_AND_INCR (mcnt, p); 6972 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); 6973 { 6974 register UCHAR_T *p2 = p; 6975 6976 /* Compare the beginning of the repeat with what in the 6977 pattern follows its end. If we can establish that there 6978 is nothing that they would both match, i.e., that we 6979 would have to backtrack because of (as in, e.g., `a*a') 6980 then we can change to pop_failure_jump, because we'll 6981 never have to backtrack. 6982 6983 This is not true in the case of alternatives: in 6984 `(a|ab)*' we do need to backtrack to the `ab' alternative 6985 (e.g., if the string was `ab'). But instead of trying to 6986 detect that here, the alternative has put on a dummy 6987 failure point which is what we will end up popping. */ 6988 6989 /* Skip over open/close-group commands. 6990 If what follows this loop is a ...+ construct, 6991 look at what begins its body, since we will have to 6992 match at least one of that. */ 6993 while (1) 6994 { 6995 if (p2 + 2 < pend 6996 && ((re_opcode_t) *p2 == stop_memory 6997 || (re_opcode_t) *p2 == start_memory)) 6998 p2 += 3; 6999 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend 7000 && (re_opcode_t) *p2 == dummy_failure_jump) 7001 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE; 7002 else 7003 break; 7004 } 7005 7006 p1 = p + mcnt; 7007 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding 7008 to the `maybe_finalize_jump' of this case. Examine what 7009 follows. */ 7010 7011 /* If we're at the end of the pattern, we can change. */ 7012 if (p2 == pend) 7013 { 7014 /* Consider what happens when matching ":\(.*\)" 7015 against ":/". I don't really understand this code 7016 yet. */ 7017 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) 7018 pop_failure_jump; 7019 DEBUG_PRINT1 7020 (" End of pattern: change to `pop_failure_jump'.\n"); 7021 } 7022 7023 else if ((re_opcode_t) *p2 == exactn 7024 #ifdef MBS_SUPPORT 7025 || (re_opcode_t) *p2 == exactn_bin 7026 #endif 7027 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) 7028 { 7029 register UCHAR_T c 7030 = *p2 == (UCHAR_T) endline ? '\n' : p2[2]; 7031 7032 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn 7033 #ifdef MBS_SUPPORT 7034 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin 7035 #endif 7036 ) && p1[3+OFFSET_ADDRESS_SIZE] != c) 7037 { 7038 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) 7039 pop_failure_jump; 7040 #ifdef WCHAR 7041 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n", 7042 (wint_t) c, 7043 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]); 7044 #else 7045 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", 7046 (char) c, 7047 (char) p1[3+OFFSET_ADDRESS_SIZE]); 7048 #endif 7049 } 7050 7051 #ifndef WCHAR 7052 else if ((re_opcode_t) p1[3] == charset 7053 || (re_opcode_t) p1[3] == charset_not) 7054 { 7055 int negate = (re_opcode_t) p1[3] == charset_not; 7056 7057 if (c < (unsigned) (p1[4] * BYTEWIDTH) 7058 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) 7059 negate = !negate; 7060 7061 /* `negate' is equal to 1 if c would match, which means 7062 that we can't change to pop_failure_jump. */ 7063 if (!negate) 7064 { 7065 p[-3] = (unsigned char) pop_failure_jump; 7066 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7067 } 7068 } 7069 #endif /* not WCHAR */ 7070 } 7071 #ifndef WCHAR 7072 else if ((re_opcode_t) *p2 == charset) 7073 { 7074 /* We win if the first character of the loop is not part 7075 of the charset. */ 7076 if ((re_opcode_t) p1[3] == exactn 7077 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5] 7078 && (p2[2 + p1[5] / BYTEWIDTH] 7079 & (1 << (p1[5] % BYTEWIDTH))))) 7080 { 7081 p[-3] = (unsigned char) pop_failure_jump; 7082 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7083 } 7084 7085 else if ((re_opcode_t) p1[3] == charset_not) 7086 { 7087 int idx; 7088 /* We win if the charset_not inside the loop 7089 lists every character listed in the charset after. */ 7090 for (idx = 0; idx < (int) p2[1]; idx++) 7091 if (! (p2[2 + idx] == 0 7092 || (idx < (int) p1[4] 7093 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) 7094 break; 7095 7096 if (idx == p2[1]) 7097 { 7098 p[-3] = (unsigned char) pop_failure_jump; 7099 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7100 } 7101 } 7102 else if ((re_opcode_t) p1[3] == charset) 7103 { 7104 int idx; 7105 /* We win if the charset inside the loop 7106 has no overlap with the one after the loop. */ 7107 for (idx = 0; 7108 idx < (int) p2[1] && idx < (int) p1[4]; 7109 idx++) 7110 if ((p2[2 + idx] & p1[5 + idx]) != 0) 7111 break; 7112 7113 if (idx == p2[1] || idx == p1[4]) 7114 { 7115 p[-3] = (unsigned char) pop_failure_jump; 7116 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7117 } 7118 } 7119 } 7120 #endif /* not WCHAR */ 7121 } 7122 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */ 7123 if ((re_opcode_t) p[-1] != pop_failure_jump) 7124 { 7125 p[-1] = (UCHAR_T) jump; 7126 DEBUG_PRINT1 (" Match => jump.\n"); 7127 goto unconditional_jump; 7128 } 7129 /* Fall through. */ 7130 7131 7132 /* The end of a simple repeat has a pop_failure_jump back to 7133 its matching on_failure_jump, where the latter will push a 7134 failure point. The pop_failure_jump takes off failure 7135 points put on by this pop_failure_jump's matching 7136 on_failure_jump; we got through the pattern to here from the 7137 matching on_failure_jump, so didn't fail. */ 7138 case pop_failure_jump: 7139 { 7140 /* We need to pass separate storage for the lowest and 7141 highest registers, even though we don't care about the 7142 actual values. Otherwise, we will restore only one 7143 register from the stack, since lowest will == highest in 7144 `pop_failure_point'. */ 7145 active_reg_t dummy_low_reg, dummy_high_reg; 7146 UCHAR_T *pdummy ATTRIBUTE_UNUSED = NULL; 7147 const CHAR_T *sdummy ATTRIBUTE_UNUSED = NULL; 7148 7149 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); 7150 POP_FAILURE_POINT (sdummy, pdummy, 7151 dummy_low_reg, dummy_high_reg, 7152 reg_dummy, reg_dummy, reg_info_dummy); 7153 } 7154 /* Fall through. */ 7155 7156 unconditional_jump: 7157 #ifdef _LIBC 7158 DEBUG_PRINT2 ("\n%p: ", p); 7159 #else 7160 DEBUG_PRINT2 ("\n0x%x: ", p); 7161 #endif 7162 /* Note fall through. */ 7163 7164 /* Unconditionally jump (without popping any failure points). */ 7165 case jump: 7166 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ 7167 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); 7168 p += mcnt; /* Do the jump. */ 7169 #ifdef _LIBC 7170 DEBUG_PRINT2 ("(to %p).\n", p); 7171 #else 7172 DEBUG_PRINT2 ("(to 0x%x).\n", p); 7173 #endif 7174 break; 7175 7176 7177 /* We need this opcode so we can detect where alternatives end 7178 in `group_match_null_string_p' et al. */ 7179 case jump_past_alt: 7180 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); 7181 goto unconditional_jump; 7182 7183 7184 /* Normally, the on_failure_jump pushes a failure point, which 7185 then gets popped at pop_failure_jump. We will end up at 7186 pop_failure_jump, also, and with a pattern of, say, `a+', we 7187 are skipping over the on_failure_jump, so we have to push 7188 something meaningless for pop_failure_jump to pop. */ 7189 case dummy_failure_jump: 7190 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); 7191 /* It doesn't matter what we push for the string here. What 7192 the code at `fail' tests is the value for the pattern. */ 7193 PUSH_FAILURE_POINT (NULL, NULL, -2); 7194 goto unconditional_jump; 7195 7196 7197 /* At the end of an alternative, we need to push a dummy failure 7198 point in case we are followed by a `pop_failure_jump', because 7199 we don't want the failure point for the alternative to be 7200 popped. For example, matching `(a|ab)*' against `aab' 7201 requires that we match the `ab' alternative. */ 7202 case push_dummy_failure: 7203 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); 7204 /* See comments just above at `dummy_failure_jump' about the 7205 two zeroes. */ 7206 PUSH_FAILURE_POINT (NULL, NULL, -2); 7207 break; 7208 7209 /* Have to succeed matching what follows at least n times. 7210 After that, handle like `on_failure_jump'. */ 7211 case succeed_n: 7212 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); 7213 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); 7214 7215 assert (mcnt >= 0); 7216 /* Originally, this is how many times we HAVE to succeed. */ 7217 if (mcnt > 0) 7218 { 7219 mcnt--; 7220 p += OFFSET_ADDRESS_SIZE; 7221 STORE_NUMBER_AND_INCR (p, mcnt); 7222 #ifdef _LIBC 7223 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE 7224 , mcnt); 7225 #else 7226 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE 7227 , mcnt); 7228 #endif 7229 } 7230 else if (mcnt == 0) 7231 { 7232 #ifdef _LIBC 7233 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", 7234 p + OFFSET_ADDRESS_SIZE); 7235 #else 7236 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", 7237 p + OFFSET_ADDRESS_SIZE); 7238 #endif /* _LIBC */ 7239 7240 #ifdef WCHAR 7241 p[1] = (UCHAR_T) no_op; 7242 #else 7243 p[2] = (UCHAR_T) no_op; 7244 p[3] = (UCHAR_T) no_op; 7245 #endif /* WCHAR */ 7246 goto on_failure; 7247 } 7248 break; 7249 7250 case jump_n: 7251 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); 7252 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); 7253 7254 /* Originally, this is how many times we CAN jump. */ 7255 if (mcnt) 7256 { 7257 mcnt--; 7258 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt); 7259 7260 #ifdef _LIBC 7261 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE, 7262 mcnt); 7263 #else 7264 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE, 7265 mcnt); 7266 #endif /* _LIBC */ 7267 goto unconditional_jump; 7268 } 7269 /* If don't have to jump any more, skip over the rest of command. */ 7270 else 7271 p += 2 * OFFSET_ADDRESS_SIZE; 7272 break; 7273 7274 case set_number_at: 7275 { 7276 DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); 7277 7278 EXTRACT_NUMBER_AND_INCR (mcnt, p); 7279 p1 = p + mcnt; 7280 EXTRACT_NUMBER_AND_INCR (mcnt, p); 7281 #ifdef _LIBC 7282 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt); 7283 #else 7284 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); 7285 #endif 7286 STORE_NUMBER (p1, mcnt); 7287 break; 7288 } 7289 7290 #if 0 7291 /* The DEC Alpha C compiler 3.x generates incorrect code for the 7292 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of 7293 AT_WORD_BOUNDARY, so this code is disabled. Expanding the 7294 macro and introducing temporary variables works around the bug. */ 7295 7296 case wordbound: 7297 DEBUG_PRINT1 ("EXECUTING wordbound.\n"); 7298 if (AT_WORD_BOUNDARY (d)) 7299 break; 7300 goto fail; 7301 7302 case notwordbound: 7303 DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); 7304 if (AT_WORD_BOUNDARY (d)) 7305 goto fail; 7306 break; 7307 #else 7308 case wordbound: 7309 { 7310 boolean prevchar, thischar; 7311 7312 DEBUG_PRINT1 ("EXECUTING wordbound.\n"); 7313 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) 7314 break; 7315 7316 prevchar = WORDCHAR_P (d - 1); 7317 thischar = WORDCHAR_P (d); 7318 if (prevchar != thischar) 7319 break; 7320 goto fail; 7321 } 7322 7323 case notwordbound: 7324 { 7325 boolean prevchar, thischar; 7326 7327 DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); 7328 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) 7329 goto fail; 7330 7331 prevchar = WORDCHAR_P (d - 1); 7332 thischar = WORDCHAR_P (d); 7333 if (prevchar != thischar) 7334 goto fail; 7335 break; 7336 } 7337 #endif 7338 7339 case wordbeg: 7340 DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); 7341 if (!AT_STRINGS_END (d) && WORDCHAR_P (d) 7342 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) 7343 break; 7344 goto fail; 7345 7346 case wordend: 7347 DEBUG_PRINT1 ("EXECUTING wordend.\n"); 7348 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) 7349 && (AT_STRINGS_END (d) || !WORDCHAR_P (d))) 7350 break; 7351 goto fail; 7352 7353 #ifdef emacs 7354 case before_dot: 7355 DEBUG_PRINT1 ("EXECUTING before_dot.\n"); 7356 if (PTR_CHAR_POS ((unsigned char *) d) >= point) 7357 goto fail; 7358 break; 7359 7360 case at_dot: 7361 DEBUG_PRINT1 ("EXECUTING at_dot.\n"); 7362 if (PTR_CHAR_POS ((unsigned char *) d) != point) 7363 goto fail; 7364 break; 7365 7366 case after_dot: 7367 DEBUG_PRINT1 ("EXECUTING after_dot.\n"); 7368 if (PTR_CHAR_POS ((unsigned char *) d) <= point) 7369 goto fail; 7370 break; 7371 7372 case syntaxspec: 7373 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); 7374 mcnt = *p++; 7375 goto matchsyntax; 7376 7377 case wordchar: 7378 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); 7379 mcnt = (int) Sword; 7380 matchsyntax: 7381 PREFETCH (); 7382 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ 7383 d++; 7384 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt) 7385 goto fail; 7386 SET_REGS_MATCHED (); 7387 break; 7388 7389 case notsyntaxspec: 7390 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); 7391 mcnt = *p++; 7392 goto matchnotsyntax; 7393 7394 case notwordchar: 7395 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); 7396 mcnt = (int) Sword; 7397 matchnotsyntax: 7398 PREFETCH (); 7399 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ 7400 d++; 7401 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt) 7402 goto fail; 7403 SET_REGS_MATCHED (); 7404 break; 7405 7406 #else /* not emacs */ 7407 case wordchar: 7408 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); 7409 PREFETCH (); 7410 if (!WORDCHAR_P (d)) 7411 goto fail; 7412 SET_REGS_MATCHED (); 7413 d++; 7414 break; 7415 7416 case notwordchar: 7417 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); 7418 PREFETCH (); 7419 if (WORDCHAR_P (d)) 7420 goto fail; 7421 SET_REGS_MATCHED (); 7422 d++; 7423 break; 7424 #endif /* not emacs */ 7425 7426 default: 7427 abort (); 7428 } 7429 continue; /* Successfully executed one pattern command; keep going. */ 7430 7431 7432 /* We goto here if a matching operation fails. */ 7433 fail: 7434 if (!FAIL_STACK_EMPTY ()) 7435 { /* A restart point is known. Restore to that state. */ 7436 DEBUG_PRINT1 ("\nFAIL:\n"); 7437 POP_FAILURE_POINT (d, p, 7438 lowest_active_reg, highest_active_reg, 7439 regstart, regend, reg_info); 7440 7441 /* If this failure point is a dummy, try the next one. */ 7442 if (!p) 7443 goto fail; 7444 7445 /* If we failed to the end of the pattern, don't examine *p. */ 7446 assert (p <= pend); 7447 if (p < pend) 7448 { 7449 boolean is_a_jump_n = false; 7450 7451 /* If failed to a backwards jump that's part of a repetition 7452 loop, need to pop this failure point and use the next one. */ 7453 switch ((re_opcode_t) *p) 7454 { 7455 case jump_n: 7456 is_a_jump_n = true; 7457 /* Fall through. */ 7458 case maybe_pop_jump: 7459 case pop_failure_jump: 7460 case jump: 7461 p1 = p + 1; 7462 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7463 p1 += mcnt; 7464 7465 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) 7466 || (!is_a_jump_n 7467 && (re_opcode_t) *p1 == on_failure_jump)) 7468 goto fail; 7469 break; 7470 default: 7471 /* do nothing */ ; 7472 } 7473 } 7474 7475 if (d >= string1 && d <= end1) 7476 dend = end_match_1; 7477 } 7478 else 7479 break; /* Matching at this starting point really fails. */ 7480 } /* for (;;) */ 7481 7482 if (best_regs_set) 7483 goto restore_best_regs; 7484 7485 FREE_VARIABLES (); 7486 7487 return -1; /* Failure to match. */ 7488 } /* re_match_2 */ 7489 7490 /* Subroutine definitions for re_match_2. */ 7491 7492 7493 /* We are passed P pointing to a register number after a start_memory. 7494 7495 Return true if the pattern up to the corresponding stop_memory can 7496 match the empty string, and false otherwise. 7497 7498 If we find the matching stop_memory, sets P to point to one past its number. 7499 Otherwise, sets P to an undefined byte less than or equal to END. 7500 7501 We don't handle duplicates properly (yet). */ 7502 7503 static boolean 7504 PREFIX(group_match_null_string_p) (UCHAR_T **p, UCHAR_T *end, 7505 PREFIX(register_info_type) *reg_info) 7506 { 7507 int mcnt; 7508 /* Point to after the args to the start_memory. */ 7509 UCHAR_T *p1 = *p + 2; 7510 7511 while (p1 < end) 7512 { 7513 /* Skip over opcodes that can match nothing, and return true or 7514 false, as appropriate, when we get to one that can't, or to the 7515 matching stop_memory. */ 7516 7517 switch ((re_opcode_t) *p1) 7518 { 7519 /* Could be either a loop or a series of alternatives. */ 7520 case on_failure_jump: 7521 p1++; 7522 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7523 7524 /* If the next operation is not a jump backwards in the 7525 pattern. */ 7526 7527 if (mcnt >= 0) 7528 { 7529 /* Go through the on_failure_jumps of the alternatives, 7530 seeing if any of the alternatives cannot match nothing. 7531 The last alternative starts with only a jump, 7532 whereas the rest start with on_failure_jump and end 7533 with a jump, e.g., here is the pattern for `a|b|c': 7534 7535 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 7536 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 7537 /exactn/1/c 7538 7539 So, we have to first go through the first (n-1) 7540 alternatives and then deal with the last one separately. */ 7541 7542 7543 /* Deal with the first (n-1) alternatives, which start 7544 with an on_failure_jump (see above) that jumps to right 7545 past a jump_past_alt. */ 7546 7547 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] == 7548 jump_past_alt) 7549 { 7550 /* `mcnt' holds how many bytes long the alternative 7551 is, including the ending `jump_past_alt' and 7552 its number. */ 7553 7554 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt - 7555 (1 + OFFSET_ADDRESS_SIZE), 7556 reg_info)) 7557 return false; 7558 7559 /* Move to right after this alternative, including the 7560 jump_past_alt. */ 7561 p1 += mcnt; 7562 7563 /* Break if it's the beginning of an n-th alternative 7564 that doesn't begin with an on_failure_jump. */ 7565 if ((re_opcode_t) *p1 != on_failure_jump) 7566 break; 7567 7568 /* Still have to check that it's not an n-th 7569 alternative that starts with an on_failure_jump. */ 7570 p1++; 7571 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7572 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] != 7573 jump_past_alt) 7574 { 7575 /* Get to the beginning of the n-th alternative. */ 7576 p1 -= 1 + OFFSET_ADDRESS_SIZE; 7577 break; 7578 } 7579 } 7580 7581 /* Deal with the last alternative: go back and get number 7582 of the `jump_past_alt' just before it. `mcnt' contains 7583 the length of the alternative. */ 7584 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE); 7585 7586 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info)) 7587 return false; 7588 7589 p1 += mcnt; /* Get past the n-th alternative. */ 7590 } /* if mcnt > 0 */ 7591 break; 7592 7593 7594 case stop_memory: 7595 assert (p1[1] == **p); 7596 *p = p1 + 2; 7597 return true; 7598 7599 7600 default: 7601 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) 7602 return false; 7603 } 7604 } /* while p1 < end */ 7605 7606 return false; 7607 } /* group_match_null_string_p */ 7608 7609 7610 /* Similar to group_match_null_string_p, but doesn't deal with alternatives: 7611 It expects P to be the first byte of a single alternative and END one 7612 byte past the last. The alternative can contain groups. */ 7613 7614 static boolean 7615 PREFIX(alt_match_null_string_p) (UCHAR_T *p, UCHAR_T *end, 7616 PREFIX(register_info_type) *reg_info) 7617 { 7618 int mcnt; 7619 UCHAR_T *p1 = p; 7620 7621 while (p1 < end) 7622 { 7623 /* Skip over opcodes that can match nothing, and break when we get 7624 to one that can't. */ 7625 7626 switch ((re_opcode_t) *p1) 7627 { 7628 /* It's a loop. */ 7629 case on_failure_jump: 7630 p1++; 7631 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7632 p1 += mcnt; 7633 break; 7634 7635 default: 7636 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) 7637 return false; 7638 } 7639 } /* while p1 < end */ 7640 7641 return true; 7642 } /* alt_match_null_string_p */ 7643 7644 7645 /* Deals with the ops common to group_match_null_string_p and 7646 alt_match_null_string_p. 7647 7648 Sets P to one after the op and its arguments, if any. */ 7649 7650 static boolean 7651 PREFIX(common_op_match_null_string_p) (UCHAR_T **p, UCHAR_T *end, 7652 PREFIX(register_info_type) *reg_info) 7653 { 7654 int mcnt; 7655 boolean ret; 7656 int reg_no; 7657 UCHAR_T *p1 = *p; 7658 7659 switch ((re_opcode_t) *p1++) 7660 { 7661 case no_op: 7662 case begline: 7663 case endline: 7664 case begbuf: 7665 case endbuf: 7666 case wordbeg: 7667 case wordend: 7668 case wordbound: 7669 case notwordbound: 7670 #ifdef emacs 7671 case before_dot: 7672 case at_dot: 7673 case after_dot: 7674 #endif 7675 break; 7676 7677 case start_memory: 7678 reg_no = *p1; 7679 assert (reg_no > 0 && reg_no <= MAX_REGNUM); 7680 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info); 7681 7682 /* Have to set this here in case we're checking a group which 7683 contains a group and a back reference to it. */ 7684 7685 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) 7686 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; 7687 7688 if (!ret) 7689 return false; 7690 break; 7691 7692 /* If this is an optimized succeed_n for zero times, make the jump. */ 7693 case jump: 7694 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7695 if (mcnt >= 0) 7696 p1 += mcnt; 7697 else 7698 return false; 7699 break; 7700 7701 case succeed_n: 7702 /* Get to the number of times to succeed. */ 7703 p1 += OFFSET_ADDRESS_SIZE; 7704 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7705 7706 if (mcnt == 0) 7707 { 7708 p1 -= 2 * OFFSET_ADDRESS_SIZE; 7709 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7710 p1 += mcnt; 7711 } 7712 else 7713 return false; 7714 break; 7715 7716 case duplicate: 7717 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) 7718 return false; 7719 break; 7720 7721 case set_number_at: 7722 p1 += 2 * OFFSET_ADDRESS_SIZE; 7723 return false; 7724 7725 default: 7726 /* All other opcodes mean we cannot match the empty string. */ 7727 return false; 7728 } 7729 7730 *p = p1; 7731 return true; 7732 } /* common_op_match_null_string_p */ 7733 7734 7735 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN 7736 bytes; nonzero otherwise. */ 7737 7738 static int 7739 PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2, register int len, 7740 RE_TRANSLATE_TYPE translate) 7741 { 7742 register const UCHAR_T *p1 = (const UCHAR_T *) s1; 7743 register const UCHAR_T *p2 = (const UCHAR_T *) s2; 7744 while (len) 7745 { 7746 #ifdef WCHAR 7747 if (((*p1<=0xff)?translate[*p1++]:*p1++) 7748 != ((*p2<=0xff)?translate[*p2++]:*p2++)) 7749 return 1; 7750 #else /* BYTE */ 7751 if (translate[*p1++] != translate[*p2++]) return 1; 7752 #endif /* WCHAR */ 7753 len--; 7754 } 7755 return 0; 7756 } 7757 7758 7759 #else /* not INSIDE_RECURSION */ 7760 7761 /* Entry points for GNU code. */ 7762 7763 /* re_compile_pattern is the GNU regular expression compiler: it 7764 compiles PATTERN (of length SIZE) and puts the result in BUFP. 7765 Returns 0 if the pattern was valid, otherwise an error string. 7766 7767 Assumes the `allocated' (and perhaps `buffer') and `translate' fields 7768 are set in BUFP on entry. 7769 7770 We call regex_compile to do the actual compilation. */ 7771 7772 const char * 7773 re_compile_pattern (const char *pattern, size_t length, 7774 struct re_pattern_buffer *bufp) 7775 { 7776 reg_errcode_t ret; 7777 7778 /* GNU code is written to assume at least RE_NREGS registers will be set 7779 (and at least one extra will be -1). */ 7780 bufp->regs_allocated = REGS_UNALLOCATED; 7781 7782 /* And GNU code determines whether or not to get register information 7783 by passing null for the REGS argument to re_match, etc., not by 7784 setting no_sub. */ 7785 bufp->no_sub = 0; 7786 7787 /* Match anchors at newline. */ 7788 bufp->newline_anchor = 1; 7789 7790 # ifdef MBS_SUPPORT 7791 if (MB_CUR_MAX != 1) 7792 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp); 7793 else 7794 # endif 7795 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp); 7796 7797 if (!ret) 7798 return NULL; 7799 return gettext (re_error_msgid[(int) ret]); 7800 } 7801 #ifdef _LIBC 7802 weak_alias (__re_compile_pattern, re_compile_pattern) 7803 #endif 7804 7805 /* Entry points compatible with 4.2 BSD regex library. We don't define 7806 them unless specifically requested. */ 7807 7808 #if defined _REGEX_RE_COMP || defined _LIBC 7809 7810 /* BSD has one and only one pattern buffer. */ 7811 static struct re_pattern_buffer re_comp_buf; 7812 7813 char * 7814 #ifdef _LIBC 7815 /* Make these definitions weak in libc, so POSIX programs can redefine 7816 these names if they don't use our functions, and still use 7817 regcomp/regexec below without link errors. */ 7818 weak_function 7819 #endif 7820 re_comp (const char *s) 7821 { 7822 reg_errcode_t ret; 7823 7824 if (!s) 7825 { 7826 if (!re_comp_buf.buffer) 7827 return (char *) gettext ("No previous regular expression"); 7828 return 0; 7829 } 7830 7831 if (!re_comp_buf.buffer) 7832 { 7833 re_comp_buf.buffer = (unsigned char *) malloc (200); 7834 if (re_comp_buf.buffer == NULL) 7835 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); 7836 re_comp_buf.allocated = 200; 7837 7838 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); 7839 if (re_comp_buf.fastmap == NULL) 7840 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); 7841 } 7842 7843 /* Since `re_exec' always passes NULL for the `regs' argument, we 7844 don't need to initialize the pattern buffer fields which affect it. */ 7845 7846 /* Match anchors at newlines. */ 7847 re_comp_buf.newline_anchor = 1; 7848 7849 # ifdef MBS_SUPPORT 7850 if (MB_CUR_MAX != 1) 7851 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); 7852 else 7853 # endif 7854 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); 7855 7856 if (!ret) 7857 return NULL; 7858 7859 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ 7860 return (char *) gettext (re_error_msgid[(int) ret]); 7861 } 7862 7863 7864 int 7865 #ifdef _LIBC 7866 weak_function 7867 #endif 7868 re_exec (const char *s) 7869 { 7870 const int len = strlen (s); 7871 return 7872 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); 7873 } 7874 7875 #endif /* _REGEX_RE_COMP */ 7876 7877 /* POSIX.2 functions. Don't define these for Emacs. */ 7878 7879 #ifndef emacs 7880 7881 /* regcomp takes a regular expression as a string and compiles it. 7882 7883 PREG is a regex_t *. We do not expect any fields to be initialized, 7884 since POSIX says we shouldn't. Thus, we set 7885 7886 `buffer' to the compiled pattern; 7887 `used' to the length of the compiled pattern; 7888 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the 7889 REG_EXTENDED bit in CFLAGS is set; otherwise, to 7890 RE_SYNTAX_POSIX_BASIC; 7891 `newline_anchor' to REG_NEWLINE being set in CFLAGS; 7892 `fastmap' to an allocated space for the fastmap; 7893 `fastmap_accurate' to zero; 7894 `re_nsub' to the number of subexpressions in PATTERN. 7895 7896 PATTERN is the address of the pattern string. 7897 7898 CFLAGS is a series of bits which affect compilation. 7899 7900 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we 7901 use POSIX basic syntax. 7902 7903 If REG_NEWLINE is set, then . and [^...] don't match newline. 7904 Also, regexec will try a match beginning after every newline. 7905 7906 If REG_ICASE is set, then we considers upper- and lowercase 7907 versions of letters to be equivalent when matching. 7908 7909 If REG_NOSUB is set, then when PREG is passed to regexec, that 7910 routine will report only success or failure, and nothing about the 7911 registers. 7912 7913 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for 7914 the return codes and their meanings.) */ 7915 7916 int 7917 regcomp (regex_t *preg, const char *pattern, int cflags) 7918 { 7919 reg_errcode_t ret; 7920 reg_syntax_t syntax 7921 = (cflags & REG_EXTENDED) ? 7922 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; 7923 7924 /* regex_compile will allocate the space for the compiled pattern. */ 7925 preg->buffer = 0; 7926 preg->allocated = 0; 7927 preg->used = 0; 7928 7929 /* Try to allocate space for the fastmap. */ 7930 preg->fastmap = (char *) malloc (1 << BYTEWIDTH); 7931 7932 if (cflags & REG_ICASE) 7933 { 7934 int i; 7935 7936 preg->translate 7937 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE 7938 * sizeof (*(RE_TRANSLATE_TYPE)0)); 7939 if (preg->translate == NULL) 7940 return (int) REG_ESPACE; 7941 7942 /* Map uppercase characters to corresponding lowercase ones. */ 7943 for (i = 0; i < CHAR_SET_SIZE; i++) 7944 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; 7945 } 7946 else 7947 preg->translate = NULL; 7948 7949 /* If REG_NEWLINE is set, newlines are treated differently. */ 7950 if (cflags & REG_NEWLINE) 7951 { /* REG_NEWLINE implies neither . nor [^...] match newline. */ 7952 syntax &= ~RE_DOT_NEWLINE; 7953 syntax |= RE_HAT_LISTS_NOT_NEWLINE; 7954 /* It also changes the matching behavior. */ 7955 preg->newline_anchor = 1; 7956 } 7957 else 7958 preg->newline_anchor = 0; 7959 7960 preg->no_sub = !!(cflags & REG_NOSUB); 7961 7962 /* POSIX says a null character in the pattern terminates it, so we 7963 can use strlen here in compiling the pattern. */ 7964 # ifdef MBS_SUPPORT 7965 if (MB_CUR_MAX != 1) 7966 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg); 7967 else 7968 # endif 7969 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg); 7970 7971 /* POSIX doesn't distinguish between an unmatched open-group and an 7972 unmatched close-group: both are REG_EPAREN. */ 7973 if (ret == REG_ERPAREN) ret = REG_EPAREN; 7974 7975 if (ret == REG_NOERROR && preg->fastmap) 7976 { 7977 /* Compute the fastmap now, since regexec cannot modify the pattern 7978 buffer. */ 7979 if (re_compile_fastmap (preg) == -2) 7980 { 7981 /* Some error occurred while computing the fastmap, just forget 7982 about it. */ 7983 free (preg->fastmap); 7984 preg->fastmap = NULL; 7985 } 7986 } 7987 7988 return (int) ret; 7989 } 7990 #ifdef _LIBC 7991 weak_alias (__regcomp, regcomp) 7992 #endif 7993 7994 7995 /* regexec searches for a given pattern, specified by PREG, in the 7996 string STRING. 7997 7998 If NMATCH is zero or REG_NOSUB was set in the cflags argument to 7999 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at 8000 least NMATCH elements, and we set them to the offsets of the 8001 corresponding matched substrings. 8002 8003 EFLAGS specifies `execution flags' which affect matching: if 8004 REG_NOTBOL is set, then ^ does not match at the beginning of the 8005 string; if REG_NOTEOL is set, then $ does not match at the end. 8006 8007 We return 0 if we find a match and REG_NOMATCH if not. */ 8008 8009 int 8010 regexec (const regex_t *preg, const char *string, size_t nmatch, 8011 regmatch_t pmatch[], int eflags) 8012 { 8013 int ret; 8014 struct re_registers regs; 8015 regex_t private_preg; 8016 int len = strlen (string); 8017 boolean want_reg_info = !preg->no_sub && nmatch > 0; 8018 8019 private_preg = *preg; 8020 8021 private_preg.not_bol = !!(eflags & REG_NOTBOL); 8022 private_preg.not_eol = !!(eflags & REG_NOTEOL); 8023 8024 /* The user has told us exactly how many registers to return 8025 information about, via `nmatch'. We have to pass that on to the 8026 matching routines. */ 8027 private_preg.regs_allocated = REGS_FIXED; 8028 8029 if (want_reg_info) 8030 { 8031 regs.num_regs = nmatch; 8032 regs.start = TALLOC (nmatch * 2, regoff_t); 8033 if (regs.start == NULL) 8034 return (int) REG_NOMATCH; 8035 regs.end = regs.start + nmatch; 8036 } 8037 8038 /* Perform the searching operation. */ 8039 ret = re_search (&private_preg, string, len, 8040 /* start: */ 0, /* range: */ len, 8041 want_reg_info ? ®s : (struct re_registers *) 0); 8042 8043 /* Copy the register information to the POSIX structure. */ 8044 if (want_reg_info) 8045 { 8046 if (ret >= 0) 8047 { 8048 unsigned r; 8049 8050 for (r = 0; r < nmatch; r++) 8051 { 8052 pmatch[r].rm_so = regs.start[r]; 8053 pmatch[r].rm_eo = regs.end[r]; 8054 } 8055 } 8056 8057 /* If we needed the temporary register info, free the space now. */ 8058 free (regs.start); 8059 } 8060 8061 /* We want zero return to mean success, unlike `re_search'. */ 8062 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; 8063 } 8064 #ifdef _LIBC 8065 weak_alias (__regexec, regexec) 8066 #endif 8067 8068 8069 /* Returns a message corresponding to an error code, ERRCODE, returned 8070 from either regcomp or regexec. We don't use PREG here. */ 8071 8072 size_t 8073 regerror (int errcode, const regex_t *preg ATTRIBUTE_UNUSED, 8074 char *errbuf, size_t errbuf_size) 8075 { 8076 const char *msg; 8077 size_t msg_size; 8078 8079 if (errcode < 0 8080 || errcode >= (int) (sizeof (re_error_msgid) 8081 / sizeof (re_error_msgid[0]))) 8082 /* Only error codes returned by the rest of the code should be passed 8083 to this routine. If we are given anything else, or if other regex 8084 code generates an invalid error code, then the program has a bug. 8085 Dump core so we can fix it. */ 8086 abort (); 8087 8088 msg = gettext (re_error_msgid[errcode]); 8089 8090 msg_size = strlen (msg) + 1; /* Includes the null. */ 8091 8092 if (errbuf_size != 0) 8093 { 8094 if (msg_size > errbuf_size) 8095 { 8096 #if defined HAVE_MEMPCPY || defined _LIBC 8097 *((char *) mempcpy (errbuf, msg, errbuf_size - 1)) = '\0'; 8098 #else 8099 (void) memcpy (errbuf, msg, errbuf_size - 1); 8100 errbuf[errbuf_size - 1] = 0; 8101 #endif 8102 } 8103 else 8104 (void) memcpy (errbuf, msg, msg_size); 8105 } 8106 8107 return msg_size; 8108 } 8109 #ifdef _LIBC 8110 weak_alias (__regerror, regerror) 8111 #endif 8112 8113 8114 /* Free dynamically allocated space used by PREG. */ 8115 8116 void 8117 regfree (regex_t *preg) 8118 { 8119 free (preg->buffer); 8120 preg->buffer = NULL; 8121 8122 preg->allocated = 0; 8123 preg->used = 0; 8124 8125 free (preg->fastmap); 8126 preg->fastmap = NULL; 8127 preg->fastmap_accurate = 0; 8128 8129 free (preg->translate); 8130 preg->translate = NULL; 8131 } 8132 #ifdef _LIBC 8133 weak_alias (__regfree, regfree) 8134 #endif 8135 8136 #endif /* not emacs */ 8137 8138 #endif /* not INSIDE_RECURSION */ 8139 8140 8141 #undef STORE_NUMBER 8142 #undef STORE_NUMBER_AND_INCR 8143 #undef EXTRACT_NUMBER 8144 #undef EXTRACT_NUMBER_AND_INCR 8145 8146 #undef DEBUG_PRINT_COMPILED_PATTERN 8147 #undef DEBUG_PRINT_DOUBLE_STRING 8148 8149 #undef INIT_FAIL_STACK 8150 #undef RESET_FAIL_STACK 8151 #undef DOUBLE_FAIL_STACK 8152 #undef PUSH_PATTERN_OP 8153 #undef PUSH_FAILURE_POINTER 8154 #undef PUSH_FAILURE_INT 8155 #undef PUSH_FAILURE_ELT 8156 #undef POP_FAILURE_POINTER 8157 #undef POP_FAILURE_INT 8158 #undef POP_FAILURE_ELT 8159 #undef DEBUG_PUSH 8160 #undef DEBUG_POP 8161 #undef PUSH_FAILURE_POINT 8162 #undef POP_FAILURE_POINT 8163 8164 #undef REG_UNSET_VALUE 8165 #undef REG_UNSET 8166 8167 #undef PATFETCH 8168 #undef PATFETCH_RAW 8169 #undef PATUNFETCH 8170 #undef TRANSLATE 8171 8172 #undef INIT_BUF_SIZE 8173 #undef GET_BUFFER_SPACE 8174 #undef BUF_PUSH 8175 #undef BUF_PUSH_2 8176 #undef BUF_PUSH_3 8177 #undef STORE_JUMP 8178 #undef STORE_JUMP2 8179 #undef INSERT_JUMP 8180 #undef INSERT_JUMP2 8181 #undef EXTEND_BUFFER 8182 #undef GET_UNSIGNED_NUMBER 8183 #undef FREE_STACK_RETURN 8184 8185 # undef POINTER_TO_OFFSET 8186 # undef MATCHING_IN_FRST_STRING 8187 # undef PREFETCH 8188 # undef AT_STRINGS_BEG 8189 # undef AT_STRINGS_END 8190 # undef WORDCHAR_P 8191 # undef FREE_VAR 8192 # undef FREE_VARIABLES 8193 # undef NO_HIGHEST_ACTIVE_REG 8194 # undef NO_LOWEST_ACTIVE_REG 8195 8196 # undef CHAR_T 8197 # undef UCHAR_T 8198 # undef COMPILED_BUFFER_VAR 8199 # undef OFFSET_ADDRESS_SIZE 8200 # undef CHAR_CLASS_SIZE 8201 # undef PREFIX 8202 # undef ARG_PREFIX 8203 # undef PUT_CHAR 8204 # undef BYTE 8205 # undef WCHAR 8206 8207 # define DEFINED_ONCE 8208