1 /* Extended regular expression matching and search library, 2 version 0.12. 3 (Implements POSIX draft P1003.2/D11.2, except for some of the 4 internationalization features.) 5 6 Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 7 2002, 2005, 2010, 2013 Free Software Foundation, Inc. 8 This file is part of the GNU C Library. 9 10 The GNU C Library is free software; you can redistribute it and/or 11 modify it under the terms of the GNU Lesser General Public 12 License as published by the Free Software Foundation; either 13 version 2.1 of the License, or (at your option) any later version. 14 15 The GNU C Library is distributed in the hope that it will be useful, 16 but WITHOUT ANY WARRANTY; without even the implied warranty of 17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 18 Lesser General Public License for more details. 19 20 You should have received a copy of the GNU Lesser General Public 21 License along with the GNU C Library; if not, write to the Free 22 Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 23 02110-1301 USA. */ 24 25 /* This file has been modified for usage in libiberty. It includes "xregex.h" 26 instead of <regex.h>. The "xregex.h" header file renames all external 27 routines with an "x" prefix so they do not collide with the native regex 28 routines or with other components regex routines. */ 29 /* AIX requires this to be the first thing in the file. */ 30 #if defined _AIX && !defined __GNUC__ && !defined REGEX_MALLOC 31 #pragma alloca 32 #endif 33 34 #undef _GNU_SOURCE 35 #define _GNU_SOURCE 36 37 #ifndef INSIDE_RECURSION 38 # ifdef HAVE_CONFIG_H 39 # include <config.h> 40 # endif 41 #endif 42 43 #include <ansidecl.h> 44 45 #ifndef INSIDE_RECURSION 46 47 # if defined STDC_HEADERS && !defined emacs 48 # include <stddef.h> 49 # define PTR_INT_TYPE ptrdiff_t 50 # else 51 /* We need this for `regex.h', and perhaps for the Emacs include files. */ 52 # include <sys/types.h> 53 # define PTR_INT_TYPE long 54 # endif 55 56 # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC) 57 58 /* For platform which support the ISO C amendement 1 functionality we 59 support user defined character classes. */ 60 # if defined _LIBC || WIDE_CHAR_SUPPORT 61 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ 62 # include <wchar.h> 63 # include <wctype.h> 64 # endif 65 66 # ifdef _LIBC 67 /* We have to keep the namespace clean. */ 68 # define regfree(preg) __regfree (preg) 69 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) 70 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) 71 # define regerror(errcode, preg, errbuf, errbuf_size) \ 72 __regerror(errcode, preg, errbuf, errbuf_size) 73 # define re_set_registers(bu, re, nu, st, en) \ 74 __re_set_registers (bu, re, nu, st, en) 75 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ 76 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) 77 # define re_match(bufp, string, size, pos, regs) \ 78 __re_match (bufp, string, size, pos, regs) 79 # define re_search(bufp, string, size, startpos, range, regs) \ 80 __re_search (bufp, string, size, startpos, range, regs) 81 # define re_compile_pattern(pattern, length, bufp) \ 82 __re_compile_pattern (pattern, length, bufp) 83 # define re_set_syntax(syntax) __re_set_syntax (syntax) 84 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ 85 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) 86 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) 87 88 # define btowc __btowc 89 90 /* We are also using some library internals. */ 91 # include <locale/localeinfo.h> 92 # include <locale/elem-hash.h> 93 # include <langinfo.h> 94 # include <locale/coll-lookup.h> 95 # endif 96 97 /* This is for other GNU distributions with internationalized messages. */ 98 # if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC 99 # include <libintl.h> 100 # ifdef _LIBC 101 # undef gettext 102 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES) 103 # endif 104 # else 105 # define gettext(msgid) (msgid) 106 # endif 107 108 # ifndef gettext_noop 109 /* This define is so xgettext can find the internationalizable 110 strings. */ 111 # define gettext_noop(String) String 112 # endif 113 114 /* The `emacs' switch turns on certain matching commands 115 that make sense only in Emacs. */ 116 # ifdef emacs 117 118 # include "lisp.h" 119 # include "buffer.h" 120 # include "syntax.h" 121 122 # else /* not emacs */ 123 124 /* If we are not linking with Emacs proper, 125 we can't use the relocating allocator 126 even if config.h says that we can. */ 127 # undef REL_ALLOC 128 129 # if defined STDC_HEADERS || defined _LIBC 130 # include <stdlib.h> 131 # else 132 char *malloc (); 133 char *realloc (); 134 # endif 135 136 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. 137 If nothing else has been done, use the method below. */ 138 # ifdef INHIBIT_STRING_HEADER 139 # if !(defined HAVE_BZERO && defined HAVE_BCOPY) 140 # if !defined bzero && !defined bcopy 141 # undef INHIBIT_STRING_HEADER 142 # endif 143 # endif 144 # endif 145 146 /* This is the normal way of making sure we have a bcopy and a bzero. 147 This is used in most programs--a few other programs avoid this 148 by defining INHIBIT_STRING_HEADER. */ 149 # ifndef INHIBIT_STRING_HEADER 150 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC 151 # include <string.h> 152 # ifndef bzero 153 # ifndef _LIBC 154 # define bzero(s, n) (memset (s, '\0', n), (s)) 155 # else 156 # define bzero(s, n) __bzero (s, n) 157 # endif 158 # endif 159 # else 160 # include <strings.h> 161 # ifndef memcmp 162 # define memcmp(s1, s2, n) bcmp (s1, s2, n) 163 # endif 164 # ifndef memcpy 165 # define memcpy(d, s, n) (bcopy (s, d, n), (d)) 166 # endif 167 # endif 168 # endif 169 170 /* Define the syntax stuff for \<, \>, etc. */ 171 172 /* This must be nonzero for the wordchar and notwordchar pattern 173 commands in re_match_2. */ 174 # ifndef Sword 175 # define Sword 1 176 # endif 177 178 # ifdef SWITCH_ENUM_BUG 179 # define SWITCH_ENUM_CAST(x) ((int)(x)) 180 # else 181 # define SWITCH_ENUM_CAST(x) (x) 182 # endif 183 184 # endif /* not emacs */ 185 186 # if defined _LIBC || HAVE_LIMITS_H 187 # include <limits.h> 188 # endif 189 190 # ifndef MB_LEN_MAX 191 # define MB_LEN_MAX 1 192 # endif 193 194 /* Get the interface, including the syntax bits. */ 195 # include "xregex.h" /* change for libiberty */ 196 197 /* isalpha etc. are used for the character classes. */ 198 # include <ctype.h> 199 200 /* Jim Meyering writes: 201 202 "... Some ctype macros are valid only for character codes that 203 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when 204 using /bin/cc or gcc but without giving an ansi option). So, all 205 ctype uses should be through macros like ISPRINT... If 206 STDC_HEADERS is defined, then autoconf has verified that the ctype 207 macros don't need to be guarded with references to isascii. ... 208 Defining isascii to 1 should let any compiler worth its salt 209 eliminate the && through constant folding." 210 Solaris defines some of these symbols so we must undefine them first. */ 211 212 # undef ISASCII 213 # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) 214 # define ISASCII(c) 1 215 # else 216 # define ISASCII(c) isascii(c) 217 # endif 218 219 # ifdef isblank 220 # define ISBLANK(c) (ISASCII (c) && isblank (c)) 221 # else 222 # define ISBLANK(c) ((c) == ' ' || (c) == '\t') 223 # endif 224 # ifdef isgraph 225 # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) 226 # else 227 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) 228 # endif 229 230 # undef ISPRINT 231 # define ISPRINT(c) (ISASCII (c) && isprint (c)) 232 # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) 233 # define ISALNUM(c) (ISASCII (c) && isalnum (c)) 234 # define ISALPHA(c) (ISASCII (c) && isalpha (c)) 235 # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) 236 # define ISLOWER(c) (ISASCII (c) && islower (c)) 237 # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) 238 # define ISSPACE(c) (ISASCII (c) && isspace (c)) 239 # define ISUPPER(c) (ISASCII (c) && isupper (c)) 240 # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) 241 242 # ifdef _tolower 243 # define TOLOWER(c) _tolower(c) 244 # else 245 # define TOLOWER(c) tolower(c) 246 # endif 247 248 # ifndef NULL 249 # define NULL (void *)0 250 # endif 251 252 /* We remove any previous definition of `SIGN_EXTEND_CHAR', 253 since ours (we hope) works properly with all combinations of 254 machines, compilers, `char' and `unsigned char' argument types. 255 (Per Bothner suggested the basic approach.) */ 256 # undef SIGN_EXTEND_CHAR 257 # if __STDC__ 258 # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) 259 # else /* not __STDC__ */ 260 /* As in Harbison and Steele. */ 261 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) 262 # endif 263 264 # ifndef emacs 265 /* How many characters in the character set. */ 266 # define CHAR_SET_SIZE 256 267 268 # ifdef SYNTAX_TABLE 269 270 extern char *re_syntax_table; 271 272 # else /* not SYNTAX_TABLE */ 273 274 static char re_syntax_table[CHAR_SET_SIZE]; 275 276 static void init_syntax_once (void); 277 278 static void 279 init_syntax_once (void) 280 { 281 register int c; 282 static int done = 0; 283 284 if (done) 285 return; 286 bzero (re_syntax_table, sizeof re_syntax_table); 287 288 for (c = 0; c < CHAR_SET_SIZE; ++c) 289 if (ISALNUM (c)) 290 re_syntax_table[c] = Sword; 291 292 re_syntax_table['_'] = Sword; 293 294 done = 1; 295 } 296 297 # endif /* not SYNTAX_TABLE */ 298 299 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)] 300 301 # endif /* emacs */ 302 303 /* Integer type for pointers. */ 304 # if !defined _LIBC && !defined HAVE_UINTPTR_T 305 typedef unsigned long int uintptr_t; 306 # endif 307 308 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we 309 use `alloca' instead of `malloc'. This is because using malloc in 310 re_search* or re_match* could cause memory leaks when C-g is used in 311 Emacs; also, malloc is slower and causes storage fragmentation. On 312 the other hand, malloc is more portable, and easier to debug. 313 314 Because we sometimes use alloca, some routines have to be macros, 315 not functions -- `alloca'-allocated space disappears at the end of the 316 function it is called in. */ 317 318 # ifdef REGEX_MALLOC 319 320 # define REGEX_ALLOCATE malloc 321 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) 322 # define REGEX_FREE free 323 324 # else /* not REGEX_MALLOC */ 325 326 /* Emacs already defines alloca, sometimes. */ 327 # ifndef alloca 328 329 /* Make alloca work the best possible way. */ 330 # ifdef __GNUC__ 331 # define alloca __builtin_alloca 332 # else /* not __GNUC__ */ 333 # if HAVE_ALLOCA_H 334 # include <alloca.h> 335 # endif /* HAVE_ALLOCA_H */ 336 # endif /* not __GNUC__ */ 337 338 # endif /* not alloca */ 339 340 # define REGEX_ALLOCATE alloca 341 342 /* Assumes a `char *destination' variable. */ 343 # define REGEX_REALLOCATE(source, osize, nsize) \ 344 (destination = (char *) alloca (nsize), \ 345 memcpy (destination, source, osize)) 346 347 /* No need to do anything to free, after alloca. */ 348 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ 349 350 # endif /* not REGEX_MALLOC */ 351 352 /* Define how to allocate the failure stack. */ 353 354 # if defined REL_ALLOC && defined REGEX_MALLOC 355 356 # define REGEX_ALLOCATE_STACK(size) \ 357 r_alloc (&failure_stack_ptr, (size)) 358 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ 359 r_re_alloc (&failure_stack_ptr, (nsize)) 360 # define REGEX_FREE_STACK(ptr) \ 361 r_alloc_free (&failure_stack_ptr) 362 363 # else /* not using relocating allocator */ 364 365 # ifdef REGEX_MALLOC 366 367 # define REGEX_ALLOCATE_STACK malloc 368 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) 369 # define REGEX_FREE_STACK free 370 371 # else /* not REGEX_MALLOC */ 372 373 # define REGEX_ALLOCATE_STACK alloca 374 375 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ 376 REGEX_REALLOCATE (source, osize, nsize) 377 /* No need to explicitly free anything. */ 378 # define REGEX_FREE_STACK(arg) 379 380 # endif /* not REGEX_MALLOC */ 381 # endif /* not using relocating allocator */ 382 383 384 /* True if `size1' is non-NULL and PTR is pointing anywhere inside 385 `string1' or just past its end. This works if PTR is NULL, which is 386 a good thing. */ 387 # define FIRST_STRING_P(ptr) \ 388 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) 389 390 /* (Re)Allocate N items of type T using malloc, or fail. */ 391 # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) 392 # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) 393 # define RETALLOC_IF(addr, n, t) \ 394 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) 395 # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) 396 397 # define BYTEWIDTH 8 /* In bits. */ 398 399 # define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) 400 401 # undef MAX 402 # undef MIN 403 # define MAX(a, b) ((a) > (b) ? (a) : (b)) 404 # define MIN(a, b) ((a) < (b) ? (a) : (b)) 405 406 typedef char boolean; 407 # define false 0 408 # define true 1 409 410 static reg_errcode_t byte_regex_compile (const char *pattern, size_t size, 411 reg_syntax_t syntax, 412 struct re_pattern_buffer *bufp); 413 414 static int byte_re_match_2_internal (struct re_pattern_buffer *bufp, 415 const char *string1, int size1, 416 const char *string2, int size2, 417 int pos, 418 struct re_registers *regs, 419 int stop); 420 static int byte_re_search_2 (struct re_pattern_buffer *bufp, 421 const char *string1, int size1, 422 const char *string2, int size2, 423 int startpos, int range, 424 struct re_registers *regs, int stop); 425 static int byte_re_compile_fastmap (struct re_pattern_buffer *bufp); 426 427 #ifdef MBS_SUPPORT 428 static reg_errcode_t wcs_regex_compile (const char *pattern, size_t size, 429 reg_syntax_t syntax, 430 struct re_pattern_buffer *bufp); 431 432 433 static int wcs_re_match_2_internal (struct re_pattern_buffer *bufp, 434 const char *cstring1, int csize1, 435 const char *cstring2, int csize2, 436 int pos, 437 struct re_registers *regs, 438 int stop, 439 wchar_t *string1, int size1, 440 wchar_t *string2, int size2, 441 int *mbs_offset1, int *mbs_offset2); 442 static int wcs_re_search_2 (struct re_pattern_buffer *bufp, 443 const char *string1, int size1, 444 const char *string2, int size2, 445 int startpos, int range, 446 struct re_registers *regs, int stop); 447 static int wcs_re_compile_fastmap (struct re_pattern_buffer *bufp); 448 #endif 449 450 /* These are the command codes that appear in compiled regular 451 expressions. Some opcodes are followed by argument bytes. A 452 command code can specify any interpretation whatsoever for its 453 arguments. Zero bytes may appear in the compiled regular expression. */ 454 455 typedef enum 456 { 457 no_op = 0, 458 459 /* Succeed right away--no more backtracking. */ 460 succeed, 461 462 /* Followed by one byte giving n, then by n literal bytes. */ 463 exactn, 464 465 # ifdef MBS_SUPPORT 466 /* Same as exactn, but contains binary data. */ 467 exactn_bin, 468 # endif 469 470 /* Matches any (more or less) character. */ 471 anychar, 472 473 /* Matches any one char belonging to specified set. First 474 following byte is number of bitmap bytes. Then come bytes 475 for a bitmap saying which chars are in. Bits in each byte 476 are ordered low-bit-first. A character is in the set if its 477 bit is 1. A character too large to have a bit in the map is 478 automatically not in the set. */ 479 /* ifdef MBS_SUPPORT, following element is length of character 480 classes, length of collating symbols, length of equivalence 481 classes, length of character ranges, and length of characters. 482 Next, character class element, collating symbols elements, 483 equivalence class elements, range elements, and character 484 elements follow. 485 See regex_compile function. */ 486 charset, 487 488 /* Same parameters as charset, but match any character that is 489 not one of those specified. */ 490 charset_not, 491 492 /* Start remembering the text that is matched, for storing in a 493 register. Followed by one byte with the register number, in 494 the range 0 to one less than the pattern buffer's re_nsub 495 field. Then followed by one byte with the number of groups 496 inner to this one. (This last has to be part of the 497 start_memory only because we need it in the on_failure_jump 498 of re_match_2.) */ 499 start_memory, 500 501 /* Stop remembering the text that is matched and store it in a 502 memory register. Followed by one byte with the register 503 number, in the range 0 to one less than `re_nsub' in the 504 pattern buffer, and one byte with the number of inner groups, 505 just like `start_memory'. (We need the number of inner 506 groups here because we don't have any easy way of finding the 507 corresponding start_memory when we're at a stop_memory.) */ 508 stop_memory, 509 510 /* Match a duplicate of something remembered. Followed by one 511 byte containing the register number. */ 512 duplicate, 513 514 /* Fail unless at beginning of line. */ 515 begline, 516 517 /* Fail unless at end of line. */ 518 endline, 519 520 /* Succeeds if at beginning of buffer (if emacs) or at beginning 521 of string to be matched (if not). */ 522 begbuf, 523 524 /* Analogously, for end of buffer/string. */ 525 endbuf, 526 527 /* Followed by two byte relative address to which to jump. */ 528 jump, 529 530 /* Same as jump, but marks the end of an alternative. */ 531 jump_past_alt, 532 533 /* Followed by two-byte relative address of place to resume at 534 in case of failure. */ 535 /* ifdef MBS_SUPPORT, the size of address is 1. */ 536 on_failure_jump, 537 538 /* Like on_failure_jump, but pushes a placeholder instead of the 539 current string position when executed. */ 540 on_failure_keep_string_jump, 541 542 /* Throw away latest failure point and then jump to following 543 two-byte relative address. */ 544 /* ifdef MBS_SUPPORT, the size of address is 1. */ 545 pop_failure_jump, 546 547 /* Change to pop_failure_jump if know won't have to backtrack to 548 match; otherwise change to jump. This is used to jump 549 back to the beginning of a repeat. If what follows this jump 550 clearly won't match what the repeat does, such that we can be 551 sure that there is no use backtracking out of repetitions 552 already matched, then we change it to a pop_failure_jump. 553 Followed by two-byte address. */ 554 /* ifdef MBS_SUPPORT, the size of address is 1. */ 555 maybe_pop_jump, 556 557 /* Jump to following two-byte address, and push a dummy failure 558 point. This failure point will be thrown away if an attempt 559 is made to use it for a failure. A `+' construct makes this 560 before the first repeat. Also used as an intermediary kind 561 of jump when compiling an alternative. */ 562 /* ifdef MBS_SUPPORT, the size of address is 1. */ 563 dummy_failure_jump, 564 565 /* Push a dummy failure point and continue. Used at the end of 566 alternatives. */ 567 push_dummy_failure, 568 569 /* Followed by two-byte relative address and two-byte number n. 570 After matching N times, jump to the address upon failure. */ 571 /* ifdef MBS_SUPPORT, the size of address is 1. */ 572 succeed_n, 573 574 /* Followed by two-byte relative address, and two-byte number n. 575 Jump to the address N times, then fail. */ 576 /* ifdef MBS_SUPPORT, the size of address is 1. */ 577 jump_n, 578 579 /* Set the following two-byte relative address to the 580 subsequent two-byte number. The address *includes* the two 581 bytes of number. */ 582 /* ifdef MBS_SUPPORT, the size of address is 1. */ 583 set_number_at, 584 585 wordchar, /* Matches any word-constituent character. */ 586 notwordchar, /* Matches any char that is not a word-constituent. */ 587 588 wordbeg, /* Succeeds if at word beginning. */ 589 wordend, /* Succeeds if at word end. */ 590 591 wordbound, /* Succeeds if at a word boundary. */ 592 notwordbound /* Succeeds if not at a word boundary. */ 593 594 # ifdef emacs 595 ,before_dot, /* Succeeds if before point. */ 596 at_dot, /* Succeeds if at point. */ 597 after_dot, /* Succeeds if after point. */ 598 599 /* Matches any character whose syntax is specified. Followed by 600 a byte which contains a syntax code, e.g., Sword. */ 601 syntaxspec, 602 603 /* Matches any character whose syntax is not that specified. */ 604 notsyntaxspec 605 # endif /* emacs */ 606 } re_opcode_t; 607 #endif /* not INSIDE_RECURSION */ 608 609 610 #ifdef BYTE 611 # define CHAR_T char 612 # define UCHAR_T unsigned char 613 # define COMPILED_BUFFER_VAR bufp->buffer 614 # define OFFSET_ADDRESS_SIZE 2 615 # define PREFIX(name) byte_##name 616 # define ARG_PREFIX(name) name 617 # define PUT_CHAR(c) putchar (c) 618 #else 619 # ifdef WCHAR 620 # define CHAR_T wchar_t 621 # define UCHAR_T wchar_t 622 # define COMPILED_BUFFER_VAR wc_buffer 623 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */ 624 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1) 625 # define PREFIX(name) wcs_##name 626 # define ARG_PREFIX(name) c##name 627 /* Should we use wide stream?? */ 628 # define PUT_CHAR(c) printf ("%C", c); 629 # define TRUE 1 630 # define FALSE 0 631 # else 632 # ifdef MBS_SUPPORT 633 # define WCHAR 634 # define INSIDE_RECURSION 635 # include "regex.c" 636 # undef INSIDE_RECURSION 637 # endif 638 # define BYTE 639 # define INSIDE_RECURSION 640 # include "regex.c" 641 # undef INSIDE_RECURSION 642 # endif 643 #endif 644 645 #ifdef INSIDE_RECURSION 646 /* Common operations on the compiled pattern. */ 647 648 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ 649 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ 650 651 # ifdef WCHAR 652 # define STORE_NUMBER(destination, number) \ 653 do { \ 654 *(destination) = (UCHAR_T)(number); \ 655 } while (0) 656 # else /* BYTE */ 657 # define STORE_NUMBER(destination, number) \ 658 do { \ 659 (destination)[0] = (number) & 0377; \ 660 (destination)[1] = (number) >> 8; \ 661 } while (0) 662 # endif /* WCHAR */ 663 664 /* Same as STORE_NUMBER, except increment DESTINATION to 665 the byte after where the number is stored. Therefore, DESTINATION 666 must be an lvalue. */ 667 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ 668 669 # define STORE_NUMBER_AND_INCR(destination, number) \ 670 do { \ 671 STORE_NUMBER (destination, number); \ 672 (destination) += OFFSET_ADDRESS_SIZE; \ 673 } while (0) 674 675 /* Put into DESTINATION a number stored in two contiguous bytes starting 676 at SOURCE. */ 677 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ 678 679 # ifdef WCHAR 680 # define EXTRACT_NUMBER(destination, source) \ 681 do { \ 682 (destination) = *(source); \ 683 } while (0) 684 # else /* BYTE */ 685 # define EXTRACT_NUMBER(destination, source) \ 686 do { \ 687 (destination) = *(source) & 0377; \ 688 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \ 689 } while (0) 690 # endif 691 692 # ifdef DEBUG 693 static void PREFIX(extract_number) (int *dest, UCHAR_T *source); 694 static void 695 PREFIX(extract_number) (int *dest, UCHAR_T *source) 696 { 697 # ifdef WCHAR 698 *dest = *source; 699 # else /* BYTE */ 700 int temp = SIGN_EXTEND_CHAR (*(source + 1)); 701 *dest = *source & 0377; 702 *dest += temp << 8; 703 # endif 704 } 705 706 # ifndef EXTRACT_MACROS /* To debug the macros. */ 707 # undef EXTRACT_NUMBER 708 # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src) 709 # endif /* not EXTRACT_MACROS */ 710 711 # endif /* DEBUG */ 712 713 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. 714 SOURCE must be an lvalue. */ 715 716 # define EXTRACT_NUMBER_AND_INCR(destination, source) \ 717 do { \ 718 EXTRACT_NUMBER (destination, source); \ 719 (source) += OFFSET_ADDRESS_SIZE; \ 720 } while (0) 721 722 # ifdef DEBUG 723 static void PREFIX(extract_number_and_incr) (int *destination, 724 UCHAR_T **source); 725 static void 726 PREFIX(extract_number_and_incr) (int *destination, UCHAR_T **source) 727 { 728 PREFIX(extract_number) (destination, *source); 729 *source += OFFSET_ADDRESS_SIZE; 730 } 731 732 # ifndef EXTRACT_MACROS 733 # undef EXTRACT_NUMBER_AND_INCR 734 # define EXTRACT_NUMBER_AND_INCR(dest, src) \ 735 PREFIX(extract_number_and_incr) (&dest, &src) 736 # endif /* not EXTRACT_MACROS */ 737 738 # endif /* DEBUG */ 739 740 741 742 /* If DEBUG is defined, Regex prints many voluminous messages about what 743 it is doing (if the variable `debug' is nonzero). If linked with the 744 main program in `iregex.c', you can enter patterns and strings 745 interactively. And if linked with the main program in `main.c' and 746 the other test files, you can run the already-written tests. */ 747 748 # ifdef DEBUG 749 750 # ifndef DEFINED_ONCE 751 752 /* We use standard I/O for debugging. */ 753 # include <stdio.h> 754 755 /* It is useful to test things that ``must'' be true when debugging. */ 756 # include <assert.h> 757 758 static int debug; 759 760 # define DEBUG_STATEMENT(e) e 761 # define DEBUG_PRINT1(x) if (debug) printf (x) 762 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) 763 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) 764 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) 765 # endif /* not DEFINED_ONCE */ 766 767 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ 768 if (debug) PREFIX(print_partial_compiled_pattern) (s, e) 769 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ 770 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2) 771 772 773 /* Print the fastmap in human-readable form. */ 774 775 # ifndef DEFINED_ONCE 776 void 777 print_fastmap (char *fastmap) 778 { 779 unsigned was_a_range = 0; 780 unsigned i = 0; 781 782 while (i < (1 << BYTEWIDTH)) 783 { 784 if (fastmap[i++]) 785 { 786 was_a_range = 0; 787 putchar (i - 1); 788 while (i < (1 << BYTEWIDTH) && fastmap[i]) 789 { 790 was_a_range = 1; 791 i++; 792 } 793 if (was_a_range) 794 { 795 printf ("-"); 796 putchar (i - 1); 797 } 798 } 799 } 800 putchar ('\n'); 801 } 802 # endif /* not DEFINED_ONCE */ 803 804 805 /* Print a compiled pattern string in human-readable form, starting at 806 the START pointer into it and ending just before the pointer END. */ 807 808 void 809 PREFIX(print_partial_compiled_pattern) (UCHAR_T *start, UCHAR_T *end) 810 { 811 int mcnt, mcnt2; 812 UCHAR_T *p1; 813 UCHAR_T *p = start; 814 UCHAR_T *pend = end; 815 816 if (start == NULL) 817 { 818 printf ("(null)\n"); 819 return; 820 } 821 822 /* Loop over pattern commands. */ 823 while (p < pend) 824 { 825 # ifdef _LIBC 826 printf ("%td:\t", p - start); 827 # else 828 printf ("%ld:\t", (long int) (p - start)); 829 # endif 830 831 switch ((re_opcode_t) *p++) 832 { 833 case no_op: 834 printf ("/no_op"); 835 break; 836 837 case exactn: 838 mcnt = *p++; 839 printf ("/exactn/%d", mcnt); 840 do 841 { 842 putchar ('/'); 843 PUT_CHAR (*p++); 844 } 845 while (--mcnt); 846 break; 847 848 # ifdef MBS_SUPPORT 849 case exactn_bin: 850 mcnt = *p++; 851 printf ("/exactn_bin/%d", mcnt); 852 do 853 { 854 printf("/%lx", (long int) *p++); 855 } 856 while (--mcnt); 857 break; 858 # endif /* MBS_SUPPORT */ 859 860 case start_memory: 861 mcnt = *p++; 862 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++); 863 break; 864 865 case stop_memory: 866 mcnt = *p++; 867 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++); 868 break; 869 870 case duplicate: 871 printf ("/duplicate/%ld", (long int) *p++); 872 break; 873 874 case anychar: 875 printf ("/anychar"); 876 break; 877 878 case charset: 879 case charset_not: 880 { 881 # ifdef WCHAR 882 int i, length; 883 wchar_t *workp = p; 884 printf ("/charset [%s", 885 (re_opcode_t) *(workp - 1) == charset_not ? "^" : ""); 886 p += 5; 887 length = *workp++; /* the length of char_classes */ 888 for (i=0 ; i<length ; i++) 889 printf("[:%lx:]", (long int) *p++); 890 length = *workp++; /* the length of collating_symbol */ 891 for (i=0 ; i<length ;) 892 { 893 printf("[."); 894 while(*p != 0) 895 PUT_CHAR((i++,*p++)); 896 i++,p++; 897 printf(".]"); 898 } 899 length = *workp++; /* the length of equivalence_class */ 900 for (i=0 ; i<length ;) 901 { 902 printf("[="); 903 while(*p != 0) 904 PUT_CHAR((i++,*p++)); 905 i++,p++; 906 printf("=]"); 907 } 908 length = *workp++; /* the length of char_range */ 909 for (i=0 ; i<length ; i++) 910 { 911 wchar_t range_start = *p++; 912 wchar_t range_end = *p++; 913 printf("%C-%C", range_start, range_end); 914 } 915 length = *workp++; /* the length of char */ 916 for (i=0 ; i<length ; i++) 917 printf("%C", *p++); 918 putchar (']'); 919 # else 920 register int c, last = -100; 921 register int in_range = 0; 922 923 printf ("/charset [%s", 924 (re_opcode_t) *(p - 1) == charset_not ? "^" : ""); 925 926 assert (p + *p < pend); 927 928 for (c = 0; c < 256; c++) 929 if (c / 8 < *p 930 && (p[1 + (c/8)] & (1 << (c % 8)))) 931 { 932 /* Are we starting a range? */ 933 if (last + 1 == c && ! in_range) 934 { 935 putchar ('-'); 936 in_range = 1; 937 } 938 /* Have we broken a range? */ 939 else if (last + 1 != c && in_range) 940 { 941 putchar (last); 942 in_range = 0; 943 } 944 945 if (! in_range) 946 putchar (c); 947 948 last = c; 949 } 950 951 if (in_range) 952 putchar (last); 953 954 putchar (']'); 955 956 p += 1 + *p; 957 # endif /* WCHAR */ 958 } 959 break; 960 961 case begline: 962 printf ("/begline"); 963 break; 964 965 case endline: 966 printf ("/endline"); 967 break; 968 969 case on_failure_jump: 970 PREFIX(extract_number_and_incr) (&mcnt, &p); 971 # ifdef _LIBC 972 printf ("/on_failure_jump to %td", p + mcnt - start); 973 # else 974 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start)); 975 # endif 976 break; 977 978 case on_failure_keep_string_jump: 979 PREFIX(extract_number_and_incr) (&mcnt, &p); 980 # ifdef _LIBC 981 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start); 982 # else 983 printf ("/on_failure_keep_string_jump to %ld", 984 (long int) (p + mcnt - start)); 985 # endif 986 break; 987 988 case dummy_failure_jump: 989 PREFIX(extract_number_and_incr) (&mcnt, &p); 990 # ifdef _LIBC 991 printf ("/dummy_failure_jump to %td", p + mcnt - start); 992 # else 993 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start)); 994 # endif 995 break; 996 997 case push_dummy_failure: 998 printf ("/push_dummy_failure"); 999 break; 1000 1001 case maybe_pop_jump: 1002 PREFIX(extract_number_and_incr) (&mcnt, &p); 1003 # ifdef _LIBC 1004 printf ("/maybe_pop_jump to %td", p + mcnt - start); 1005 # else 1006 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start)); 1007 # endif 1008 break; 1009 1010 case pop_failure_jump: 1011 PREFIX(extract_number_and_incr) (&mcnt, &p); 1012 # ifdef _LIBC 1013 printf ("/pop_failure_jump to %td", p + mcnt - start); 1014 # else 1015 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start)); 1016 # endif 1017 break; 1018 1019 case jump_past_alt: 1020 PREFIX(extract_number_and_incr) (&mcnt, &p); 1021 # ifdef _LIBC 1022 printf ("/jump_past_alt to %td", p + mcnt - start); 1023 # else 1024 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start)); 1025 # endif 1026 break; 1027 1028 case jump: 1029 PREFIX(extract_number_and_incr) (&mcnt, &p); 1030 # ifdef _LIBC 1031 printf ("/jump to %td", p + mcnt - start); 1032 # else 1033 printf ("/jump to %ld", (long int) (p + mcnt - start)); 1034 # endif 1035 break; 1036 1037 case succeed_n: 1038 PREFIX(extract_number_and_incr) (&mcnt, &p); 1039 p1 = p + mcnt; 1040 PREFIX(extract_number_and_incr) (&mcnt2, &p); 1041 # ifdef _LIBC 1042 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2); 1043 # else 1044 printf ("/succeed_n to %ld, %d times", 1045 (long int) (p1 - start), mcnt2); 1046 # endif 1047 break; 1048 1049 case jump_n: 1050 PREFIX(extract_number_and_incr) (&mcnt, &p); 1051 p1 = p + mcnt; 1052 PREFIX(extract_number_and_incr) (&mcnt2, &p); 1053 printf ("/jump_n to %d, %d times", p1 - start, mcnt2); 1054 break; 1055 1056 case set_number_at: 1057 PREFIX(extract_number_and_incr) (&mcnt, &p); 1058 p1 = p + mcnt; 1059 PREFIX(extract_number_and_incr) (&mcnt2, &p); 1060 # ifdef _LIBC 1061 printf ("/set_number_at location %td to %d", p1 - start, mcnt2); 1062 # else 1063 printf ("/set_number_at location %ld to %d", 1064 (long int) (p1 - start), mcnt2); 1065 # endif 1066 break; 1067 1068 case wordbound: 1069 printf ("/wordbound"); 1070 break; 1071 1072 case notwordbound: 1073 printf ("/notwordbound"); 1074 break; 1075 1076 case wordbeg: 1077 printf ("/wordbeg"); 1078 break; 1079 1080 case wordend: 1081 printf ("/wordend"); 1082 break; 1083 1084 # ifdef emacs 1085 case before_dot: 1086 printf ("/before_dot"); 1087 break; 1088 1089 case at_dot: 1090 printf ("/at_dot"); 1091 break; 1092 1093 case after_dot: 1094 printf ("/after_dot"); 1095 break; 1096 1097 case syntaxspec: 1098 printf ("/syntaxspec"); 1099 mcnt = *p++; 1100 printf ("/%d", mcnt); 1101 break; 1102 1103 case notsyntaxspec: 1104 printf ("/notsyntaxspec"); 1105 mcnt = *p++; 1106 printf ("/%d", mcnt); 1107 break; 1108 # endif /* emacs */ 1109 1110 case wordchar: 1111 printf ("/wordchar"); 1112 break; 1113 1114 case notwordchar: 1115 printf ("/notwordchar"); 1116 break; 1117 1118 case begbuf: 1119 printf ("/begbuf"); 1120 break; 1121 1122 case endbuf: 1123 printf ("/endbuf"); 1124 break; 1125 1126 default: 1127 printf ("?%ld", (long int) *(p-1)); 1128 } 1129 1130 putchar ('\n'); 1131 } 1132 1133 # ifdef _LIBC 1134 printf ("%td:\tend of pattern.\n", p - start); 1135 # else 1136 printf ("%ld:\tend of pattern.\n", (long int) (p - start)); 1137 # endif 1138 } 1139 1140 1141 void 1142 PREFIX(print_compiled_pattern) (struct re_pattern_buffer *bufp) 1143 { 1144 UCHAR_T *buffer = (UCHAR_T*) bufp->buffer; 1145 1146 PREFIX(print_partial_compiled_pattern) (buffer, buffer 1147 + bufp->used / sizeof(UCHAR_T)); 1148 printf ("%ld bytes used/%ld bytes allocated.\n", 1149 bufp->used, bufp->allocated); 1150 1151 if (bufp->fastmap_accurate && bufp->fastmap) 1152 { 1153 printf ("fastmap: "); 1154 print_fastmap (bufp->fastmap); 1155 } 1156 1157 # ifdef _LIBC 1158 printf ("re_nsub: %Zd\t", bufp->re_nsub); 1159 # else 1160 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub); 1161 # endif 1162 printf ("regs_alloc: %d\t", bufp->regs_allocated); 1163 printf ("can_be_null: %d\t", bufp->can_be_null); 1164 printf ("newline_anchor: %d\n", bufp->newline_anchor); 1165 printf ("no_sub: %d\t", bufp->no_sub); 1166 printf ("not_bol: %d\t", bufp->not_bol); 1167 printf ("not_eol: %d\t", bufp->not_eol); 1168 printf ("syntax: %lx\n", bufp->syntax); 1169 /* Perhaps we should print the translate table? */ 1170 } 1171 1172 1173 void 1174 PREFIX(print_double_string) (const CHAR_T *where, const CHAR_T *string1, 1175 int size1, const CHAR_T *string2, int size2) 1176 { 1177 int this_char; 1178 1179 if (where == NULL) 1180 printf ("(null)"); 1181 else 1182 { 1183 int cnt; 1184 1185 if (FIRST_STRING_P (where)) 1186 { 1187 for (this_char = where - string1; this_char < size1; this_char++) 1188 PUT_CHAR (string1[this_char]); 1189 1190 where = string2; 1191 } 1192 1193 cnt = 0; 1194 for (this_char = where - string2; this_char < size2; this_char++) 1195 { 1196 PUT_CHAR (string2[this_char]); 1197 if (++cnt > 100) 1198 { 1199 fputs ("...", stdout); 1200 break; 1201 } 1202 } 1203 } 1204 } 1205 1206 # ifndef DEFINED_ONCE 1207 void 1208 printchar (int c) 1209 { 1210 putc (c, stderr); 1211 } 1212 # endif 1213 1214 # else /* not DEBUG */ 1215 1216 # ifndef DEFINED_ONCE 1217 # undef assert 1218 # define assert(e) 1219 1220 # define DEBUG_STATEMENT(e) 1221 # define DEBUG_PRINT1(x) 1222 # define DEBUG_PRINT2(x1, x2) 1223 # define DEBUG_PRINT3(x1, x2, x3) 1224 # define DEBUG_PRINT4(x1, x2, x3, x4) 1225 # endif /* not DEFINED_ONCE */ 1226 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) 1227 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) 1228 1229 # endif /* not DEBUG */ 1230 1231 1232 1233 # ifdef WCHAR 1234 /* This convert a multibyte string to a wide character string. 1235 And write their correspondances to offset_buffer(see below) 1236 and write whether each wchar_t is binary data to is_binary. 1237 This assume invalid multibyte sequences as binary data. 1238 We assume offset_buffer and is_binary is already allocated 1239 enough space. */ 1240 1241 static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src, 1242 size_t len, int *offset_buffer, 1243 char *is_binary); 1244 static size_t 1245 convert_mbs_to_wcs (CHAR_T *dest, const unsigned char*src, size_t len, 1246 int *offset_buffer, char *is_binary) 1247 /* It hold correspondances between src(char string) and 1248 dest(wchar_t string) for optimization. 1249 e.g. src = "xxxyzz" 1250 dest = {'X', 'Y', 'Z'} 1251 (each "xxx", "y" and "zz" represent one multibyte character 1252 corresponding to 'X', 'Y' and 'Z'.) 1253 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")} 1254 = {0, 3, 4, 6} 1255 */ 1256 { 1257 wchar_t *pdest = dest; 1258 const unsigned char *psrc = src; 1259 size_t wc_count = 0; 1260 1261 mbstate_t mbs; 1262 int i, consumed; 1263 size_t mb_remain = len; 1264 size_t mb_count = 0; 1265 1266 /* Initialize the conversion state. */ 1267 memset (&mbs, 0, sizeof (mbstate_t)); 1268 1269 offset_buffer[0] = 0; 1270 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed, 1271 psrc += consumed) 1272 { 1273 #ifdef _LIBC 1274 consumed = __mbrtowc (pdest, psrc, mb_remain, &mbs); 1275 #else 1276 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs); 1277 #endif 1278 1279 if (consumed <= 0) 1280 /* failed to convert. maybe src contains binary data. 1281 So we consume 1 byte manualy. */ 1282 { 1283 *pdest = *psrc; 1284 consumed = 1; 1285 is_binary[wc_count] = TRUE; 1286 } 1287 else 1288 is_binary[wc_count] = FALSE; 1289 /* In sjis encoding, we use yen sign as escape character in 1290 place of reverse solidus. So we convert 0x5c(yen sign in 1291 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse 1292 solidus in UCS2). */ 1293 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5) 1294 *pdest = (wchar_t) *psrc; 1295 1296 offset_buffer[wc_count + 1] = mb_count += consumed; 1297 } 1298 1299 /* Fill remain of the buffer with sentinel. */ 1300 for (i = wc_count + 1 ; i <= len ; i++) 1301 offset_buffer[i] = mb_count + 1; 1302 1303 return wc_count; 1304 } 1305 1306 # endif /* WCHAR */ 1307 1308 #else /* not INSIDE_RECURSION */ 1309 1310 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can 1311 also be assigned to arbitrarily: each pattern buffer stores its own 1312 syntax, so it can be changed between regex compilations. */ 1313 /* This has no initializer because initialized variables in Emacs 1314 become read-only after dumping. */ 1315 reg_syntax_t re_syntax_options; 1316 1317 1318 /* Specify the precise syntax of regexps for compilation. This provides 1319 for compatibility for various utilities which historically have 1320 different, incompatible syntaxes. 1321 1322 The argument SYNTAX is a bit mask comprised of the various bits 1323 defined in regex.h. We return the old syntax. */ 1324 1325 reg_syntax_t 1326 re_set_syntax (reg_syntax_t syntax) 1327 { 1328 reg_syntax_t ret = re_syntax_options; 1329 1330 re_syntax_options = syntax; 1331 # ifdef DEBUG 1332 if (syntax & RE_DEBUG) 1333 debug = 1; 1334 else if (debug) /* was on but now is not */ 1335 debug = 0; 1336 # endif /* DEBUG */ 1337 return ret; 1338 } 1339 # ifdef _LIBC 1340 weak_alias (__re_set_syntax, re_set_syntax) 1341 # endif 1342 1343 /* This table gives an error message for each of the error codes listed 1344 in regex.h. Obviously the order here has to be same as there. 1345 POSIX doesn't require that we do anything for REG_NOERROR, 1346 but why not be nice? */ 1347 1348 static const char *re_error_msgid[] = 1349 { 1350 gettext_noop ("Success"), /* REG_NOERROR */ 1351 gettext_noop ("No match"), /* REG_NOMATCH */ 1352 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */ 1353 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */ 1354 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */ 1355 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */ 1356 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */ 1357 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */ 1358 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */ 1359 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */ 1360 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */ 1361 gettext_noop ("Invalid range end"), /* REG_ERANGE */ 1362 gettext_noop ("Memory exhausted"), /* REG_ESPACE */ 1363 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */ 1364 gettext_noop ("Premature end of regular expression"), /* REG_EEND */ 1365 gettext_noop ("Regular expression too big"), /* REG_ESIZE */ 1366 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */ 1367 }; 1368 1369 #endif /* INSIDE_RECURSION */ 1370 1371 #ifndef DEFINED_ONCE 1372 /* Avoiding alloca during matching, to placate r_alloc. */ 1373 1374 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the 1375 searching and matching functions should not call alloca. On some 1376 systems, alloca is implemented in terms of malloc, and if we're 1377 using the relocating allocator routines, then malloc could cause a 1378 relocation, which might (if the strings being searched are in the 1379 ralloc heap) shift the data out from underneath the regexp 1380 routines. 1381 1382 Here's another reason to avoid allocation: Emacs 1383 processes input from X in a signal handler; processing X input may 1384 call malloc; if input arrives while a matching routine is calling 1385 malloc, then we're scrod. But Emacs can't just block input while 1386 calling matching routines; then we don't notice interrupts when 1387 they come in. So, Emacs blocks input around all regexp calls 1388 except the matching calls, which it leaves unprotected, in the 1389 faith that they will not malloc. */ 1390 1391 /* Normally, this is fine. */ 1392 # define MATCH_MAY_ALLOCATE 1393 1394 /* When using GNU C, we are not REALLY using the C alloca, no matter 1395 what config.h may say. So don't take precautions for it. */ 1396 # ifdef __GNUC__ 1397 # undef C_ALLOCA 1398 # endif 1399 1400 /* The match routines may not allocate if (1) they would do it with malloc 1401 and (2) it's not safe for them to use malloc. 1402 Note that if REL_ALLOC is defined, matching would not use malloc for the 1403 failure stack, but we would still use it for the register vectors; 1404 so REL_ALLOC should not affect this. */ 1405 # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs 1406 # undef MATCH_MAY_ALLOCATE 1407 # endif 1408 #endif /* not DEFINED_ONCE */ 1409 1410 #ifdef INSIDE_RECURSION 1411 /* Failure stack declarations and macros; both re_compile_fastmap and 1412 re_match_2 use a failure stack. These have to be macros because of 1413 REGEX_ALLOCATE_STACK. */ 1414 1415 1416 /* Number of failure points for which to initially allocate space 1417 when matching. If this number is exceeded, we allocate more 1418 space, so it is not a hard limit. */ 1419 # ifndef INIT_FAILURE_ALLOC 1420 # define INIT_FAILURE_ALLOC 5 1421 # endif 1422 1423 /* Roughly the maximum number of failure points on the stack. Would be 1424 exactly that if always used MAX_FAILURE_ITEMS items each time we failed. 1425 This is a variable only so users of regex can assign to it; we never 1426 change it ourselves. */ 1427 1428 # ifdef INT_IS_16BIT 1429 1430 # ifndef DEFINED_ONCE 1431 # if defined MATCH_MAY_ALLOCATE 1432 /* 4400 was enough to cause a crash on Alpha OSF/1, 1433 whose default stack limit is 2mb. */ 1434 long int re_max_failures = 4000; 1435 # else 1436 long int re_max_failures = 2000; 1437 # endif 1438 # endif 1439 1440 union PREFIX(fail_stack_elt) 1441 { 1442 UCHAR_T *pointer; 1443 long int integer; 1444 }; 1445 1446 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); 1447 1448 typedef struct 1449 { 1450 PREFIX(fail_stack_elt_t) *stack; 1451 unsigned long int size; 1452 unsigned long int avail; /* Offset of next open position. */ 1453 } PREFIX(fail_stack_type); 1454 1455 # else /* not INT_IS_16BIT */ 1456 1457 # ifndef DEFINED_ONCE 1458 # if defined MATCH_MAY_ALLOCATE 1459 /* 4400 was enough to cause a crash on Alpha OSF/1, 1460 whose default stack limit is 2mb. */ 1461 int re_max_failures = 4000; 1462 # else 1463 int re_max_failures = 2000; 1464 # endif 1465 # endif 1466 1467 union PREFIX(fail_stack_elt) 1468 { 1469 UCHAR_T *pointer; 1470 int integer; 1471 }; 1472 1473 typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); 1474 1475 typedef struct 1476 { 1477 PREFIX(fail_stack_elt_t) *stack; 1478 unsigned size; 1479 unsigned avail; /* Offset of next open position. */ 1480 } PREFIX(fail_stack_type); 1481 1482 # endif /* INT_IS_16BIT */ 1483 1484 # ifndef DEFINED_ONCE 1485 # define FAIL_STACK_EMPTY() (fail_stack.avail == 0) 1486 # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) 1487 # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) 1488 # endif 1489 1490 1491 /* Define macros to initialize and free the failure stack. 1492 Do `return -2' if the alloc fails. */ 1493 1494 # ifdef MATCH_MAY_ALLOCATE 1495 # define INIT_FAIL_STACK() \ 1496 do { \ 1497 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \ 1498 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \ 1499 \ 1500 if (fail_stack.stack == NULL) \ 1501 return -2; \ 1502 \ 1503 fail_stack.size = INIT_FAILURE_ALLOC; \ 1504 fail_stack.avail = 0; \ 1505 } while (0) 1506 1507 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) 1508 # else 1509 # define INIT_FAIL_STACK() \ 1510 do { \ 1511 fail_stack.avail = 0; \ 1512 } while (0) 1513 1514 # define RESET_FAIL_STACK() 1515 # endif 1516 1517 1518 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. 1519 1520 Return 1 if succeeds, and 0 if either ran out of memory 1521 allocating space for it or it was already too large. 1522 1523 REGEX_REALLOCATE_STACK requires `destination' be declared. */ 1524 1525 # define DOUBLE_FAIL_STACK(fail_stack) \ 1526 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \ 1527 ? 0 \ 1528 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \ 1529 REGEX_REALLOCATE_STACK ((fail_stack).stack, \ 1530 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \ 1531 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\ 1532 \ 1533 (fail_stack).stack == NULL \ 1534 ? 0 \ 1535 : ((fail_stack).size <<= 1, \ 1536 1))) 1537 1538 1539 /* Push pointer POINTER on FAIL_STACK. 1540 Return 1 if was able to do so and 0 if ran out of memory allocating 1541 space to do so. */ 1542 # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ 1543 ((FAIL_STACK_FULL () \ 1544 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \ 1545 ? 0 \ 1546 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ 1547 1)) 1548 1549 /* Push a pointer value onto the failure stack. 1550 Assumes the variable `fail_stack'. Probably should only 1551 be called from within `PUSH_FAILURE_POINT'. */ 1552 # define PUSH_FAILURE_POINTER(item) \ 1553 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item) 1554 1555 /* This pushes an integer-valued item onto the failure stack. 1556 Assumes the variable `fail_stack'. Probably should only 1557 be called from within `PUSH_FAILURE_POINT'. */ 1558 # define PUSH_FAILURE_INT(item) \ 1559 fail_stack.stack[fail_stack.avail++].integer = (item) 1560 1561 /* Push a fail_stack_elt_t value onto the failure stack. 1562 Assumes the variable `fail_stack'. Probably should only 1563 be called from within `PUSH_FAILURE_POINT'. */ 1564 # define PUSH_FAILURE_ELT(item) \ 1565 fail_stack.stack[fail_stack.avail++] = (item) 1566 1567 /* These three POP... operations complement the three PUSH... operations. 1568 All assume that `fail_stack' is nonempty. */ 1569 # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer 1570 # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer 1571 # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] 1572 1573 /* Used to omit pushing failure point id's when we're not debugging. */ 1574 # ifdef DEBUG 1575 # define DEBUG_PUSH PUSH_FAILURE_INT 1576 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () 1577 # else 1578 # define DEBUG_PUSH(item) 1579 # define DEBUG_POP(item_addr) 1580 # endif 1581 1582 1583 /* Push the information about the state we will need 1584 if we ever fail back to it. 1585 1586 Requires variables fail_stack, regstart, regend, reg_info, and 1587 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination' 1588 be declared. 1589 1590 Does `return FAILURE_CODE' if runs out of memory. */ 1591 1592 # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ 1593 do { \ 1594 char *destination; \ 1595 /* Must be int, so when we don't save any registers, the arithmetic \ 1596 of 0 + -1 isn't done as unsigned. */ \ 1597 /* Can't be int, since there is not a shred of a guarantee that int \ 1598 is wide enough to hold a value of something to which pointer can \ 1599 be assigned */ \ 1600 active_reg_t this_reg; \ 1601 \ 1602 DEBUG_STATEMENT (failure_id++); \ 1603 DEBUG_STATEMENT (nfailure_points_pushed++); \ 1604 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ 1605 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ 1606 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ 1607 \ 1608 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \ 1609 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ 1610 \ 1611 /* Ensure we have enough space allocated for what we will push. */ \ 1612 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ 1613 { \ 1614 if (!DOUBLE_FAIL_STACK (fail_stack)) \ 1615 return failure_code; \ 1616 \ 1617 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ 1618 (fail_stack).size); \ 1619 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ 1620 } \ 1621 \ 1622 /* Push the info, starting with the registers. */ \ 1623 DEBUG_PRINT1 ("\n"); \ 1624 \ 1625 if (1) \ 1626 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ 1627 this_reg++) \ 1628 { \ 1629 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \ 1630 DEBUG_STATEMENT (num_regs_pushed++); \ 1631 \ 1632 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ 1633 PUSH_FAILURE_POINTER (regstart[this_reg]); \ 1634 \ 1635 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ 1636 PUSH_FAILURE_POINTER (regend[this_reg]); \ 1637 \ 1638 DEBUG_PRINT2 (" info: %p\n ", \ 1639 reg_info[this_reg].word.pointer); \ 1640 DEBUG_PRINT2 (" match_null=%d", \ 1641 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ 1642 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ 1643 DEBUG_PRINT2 (" matched_something=%d", \ 1644 MATCHED_SOMETHING (reg_info[this_reg])); \ 1645 DEBUG_PRINT2 (" ever_matched=%d", \ 1646 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ 1647 DEBUG_PRINT1 ("\n"); \ 1648 PUSH_FAILURE_ELT (reg_info[this_reg].word); \ 1649 } \ 1650 \ 1651 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\ 1652 PUSH_FAILURE_INT (lowest_active_reg); \ 1653 \ 1654 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\ 1655 PUSH_FAILURE_INT (highest_active_reg); \ 1656 \ 1657 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \ 1658 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ 1659 PUSH_FAILURE_POINTER (pattern_place); \ 1660 \ 1661 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \ 1662 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ 1663 size2); \ 1664 DEBUG_PRINT1 ("'\n"); \ 1665 PUSH_FAILURE_POINTER (string_place); \ 1666 \ 1667 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ 1668 DEBUG_PUSH (failure_id); \ 1669 } while (0) 1670 1671 # ifndef DEFINED_ONCE 1672 /* This is the number of items that are pushed and popped on the stack 1673 for each register. */ 1674 # define NUM_REG_ITEMS 3 1675 1676 /* Individual items aside from the registers. */ 1677 # ifdef DEBUG 1678 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ 1679 # else 1680 # define NUM_NONREG_ITEMS 4 1681 # endif 1682 1683 /* We push at most this many items on the stack. */ 1684 /* We used to use (num_regs - 1), which is the number of registers 1685 this regexp will save; but that was changed to 5 1686 to avoid stack overflow for a regexp with lots of parens. */ 1687 # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS) 1688 1689 /* We actually push this many items. */ 1690 # define NUM_FAILURE_ITEMS \ 1691 (((0 \ 1692 ? 0 : highest_active_reg - lowest_active_reg + 1) \ 1693 * NUM_REG_ITEMS) \ 1694 + NUM_NONREG_ITEMS) 1695 1696 /* How many items can still be added to the stack without overflowing it. */ 1697 # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) 1698 # endif /* not DEFINED_ONCE */ 1699 1700 1701 /* Pops what PUSH_FAIL_STACK pushes. 1702 1703 We restore into the parameters, all of which should be lvalues: 1704 STR -- the saved data position. 1705 PAT -- the saved pattern position. 1706 LOW_REG, HIGH_REG -- the highest and lowest active registers. 1707 REGSTART, REGEND -- arrays of string positions. 1708 REG_INFO -- array of information about each subexpression. 1709 1710 Also assumes the variables `fail_stack' and (if debugging), `bufp', 1711 `pend', `string1', `size1', `string2', and `size2'. */ 1712 # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ 1713 { \ 1714 DEBUG_STATEMENT (unsigned failure_id;) \ 1715 active_reg_t this_reg; \ 1716 const UCHAR_T *string_temp; \ 1717 \ 1718 assert (!FAIL_STACK_EMPTY ()); \ 1719 \ 1720 /* Remove failure points and point to how many regs pushed. */ \ 1721 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ 1722 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ 1723 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ 1724 \ 1725 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ 1726 \ 1727 DEBUG_POP (&failure_id); \ 1728 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ 1729 \ 1730 /* If the saved string location is NULL, it came from an \ 1731 on_failure_keep_string_jump opcode, and we want to throw away the \ 1732 saved NULL, thus retaining our current position in the string. */ \ 1733 string_temp = POP_FAILURE_POINTER (); \ 1734 if (string_temp != NULL) \ 1735 str = (const CHAR_T *) string_temp; \ 1736 \ 1737 DEBUG_PRINT2 (" Popping string %p: `", str); \ 1738 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ 1739 DEBUG_PRINT1 ("'\n"); \ 1740 \ 1741 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \ 1742 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \ 1743 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ 1744 \ 1745 /* Restore register info. */ \ 1746 high_reg = (active_reg_t) POP_FAILURE_INT (); \ 1747 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \ 1748 \ 1749 low_reg = (active_reg_t) POP_FAILURE_INT (); \ 1750 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \ 1751 \ 1752 if (1) \ 1753 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ 1754 { \ 1755 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \ 1756 \ 1757 reg_info[this_reg].word = POP_FAILURE_ELT (); \ 1758 DEBUG_PRINT2 (" info: %p\n", \ 1759 reg_info[this_reg].word.pointer); \ 1760 \ 1761 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ 1762 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ 1763 \ 1764 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ 1765 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ 1766 } \ 1767 else \ 1768 { \ 1769 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \ 1770 { \ 1771 reg_info[this_reg].word.integer = 0; \ 1772 regend[this_reg] = 0; \ 1773 regstart[this_reg] = 0; \ 1774 } \ 1775 highest_active_reg = high_reg; \ 1776 } \ 1777 \ 1778 set_regs_matched_done = 0; \ 1779 DEBUG_STATEMENT (nfailure_points_popped++); \ 1780 } /* POP_FAILURE_POINT */ 1781 1782 /* Structure for per-register (a.k.a. per-group) information. 1783 Other register information, such as the 1784 starting and ending positions (which are addresses), and the list of 1785 inner groups (which is a bits list) are maintained in separate 1786 variables. 1787 1788 We are making a (strictly speaking) nonportable assumption here: that 1789 the compiler will pack our bit fields into something that fits into 1790 the type of `word', i.e., is something that fits into one item on the 1791 failure stack. */ 1792 1793 1794 /* Declarations and macros for re_match_2. */ 1795 1796 typedef union 1797 { 1798 PREFIX(fail_stack_elt_t) word; 1799 struct 1800 { 1801 /* This field is one if this group can match the empty string, 1802 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ 1803 # define MATCH_NULL_UNSET_VALUE 3 1804 unsigned match_null_string_p : 2; 1805 unsigned is_active : 1; 1806 unsigned matched_something : 1; 1807 unsigned ever_matched_something : 1; 1808 } bits; 1809 } PREFIX(register_info_type); 1810 1811 # ifndef DEFINED_ONCE 1812 # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) 1813 # define IS_ACTIVE(R) ((R).bits.is_active) 1814 # define MATCHED_SOMETHING(R) ((R).bits.matched_something) 1815 # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) 1816 1817 1818 /* Call this when have matched a real character; it sets `matched' flags 1819 for the subexpressions which we are currently inside. Also records 1820 that those subexprs have matched. */ 1821 # define SET_REGS_MATCHED() \ 1822 do \ 1823 { \ 1824 if (!set_regs_matched_done) \ 1825 { \ 1826 active_reg_t r; \ 1827 set_regs_matched_done = 1; \ 1828 for (r = lowest_active_reg; r <= highest_active_reg; r++) \ 1829 { \ 1830 MATCHED_SOMETHING (reg_info[r]) \ 1831 = EVER_MATCHED_SOMETHING (reg_info[r]) \ 1832 = 1; \ 1833 } \ 1834 } \ 1835 } \ 1836 while (0) 1837 # endif /* not DEFINED_ONCE */ 1838 1839 /* Registers are set to a sentinel when they haven't yet matched. */ 1840 static CHAR_T PREFIX(reg_unset_dummy); 1841 # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy)) 1842 # define REG_UNSET(e) ((e) == REG_UNSET_VALUE) 1843 1844 /* Subroutine declarations and macros for regex_compile. */ 1845 static void PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg); 1846 static void PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc, 1847 int arg1, int arg2); 1848 static void PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc, 1849 int arg, UCHAR_T *end); 1850 static void PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc, 1851 int arg1, int arg2, UCHAR_T *end); 1852 static boolean PREFIX(at_begline_loc_p) (const CHAR_T *pattern, 1853 const CHAR_T *p, 1854 reg_syntax_t syntax); 1855 static boolean PREFIX(at_endline_loc_p) (const CHAR_T *p, 1856 const CHAR_T *pend, 1857 reg_syntax_t syntax); 1858 # ifdef WCHAR 1859 static reg_errcode_t wcs_compile_range (CHAR_T range_start, 1860 const CHAR_T **p_ptr, 1861 const CHAR_T *pend, 1862 char *translate, 1863 reg_syntax_t syntax, 1864 UCHAR_T *b, 1865 CHAR_T *char_set); 1866 static void insert_space (int num, CHAR_T *loc, CHAR_T *end); 1867 # else /* BYTE */ 1868 static reg_errcode_t byte_compile_range (unsigned int range_start, 1869 const char **p_ptr, 1870 const char *pend, 1871 char *translate, 1872 reg_syntax_t syntax, 1873 unsigned char *b); 1874 # endif /* WCHAR */ 1875 1876 /* Fetch the next character in the uncompiled pattern---translating it 1877 if necessary. Also cast from a signed character in the constant 1878 string passed to us by the user to an unsigned char that we can use 1879 as an array index (in, e.g., `translate'). */ 1880 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, 1881 because it is impossible to allocate 4GB array for some encodings 1882 which have 4 byte character_set like UCS4. */ 1883 # ifndef PATFETCH 1884 # ifdef WCHAR 1885 # define PATFETCH(c) \ 1886 do {if (p == pend) return REG_EEND; \ 1887 c = (UCHAR_T) *p++; \ 1888 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \ 1889 } while (0) 1890 # else /* BYTE */ 1891 # define PATFETCH(c) \ 1892 do {if (p == pend) return REG_EEND; \ 1893 c = (unsigned char) *p++; \ 1894 if (translate) c = (unsigned char) translate[c]; \ 1895 } while (0) 1896 # endif /* WCHAR */ 1897 # endif 1898 1899 /* Fetch the next character in the uncompiled pattern, with no 1900 translation. */ 1901 # define PATFETCH_RAW(c) \ 1902 do {if (p == pend) return REG_EEND; \ 1903 c = (UCHAR_T) *p++; \ 1904 } while (0) 1905 1906 /* Go backwards one character in the pattern. */ 1907 # define PATUNFETCH p-- 1908 1909 1910 /* If `translate' is non-null, return translate[D], else just D. We 1911 cast the subscript to translate because some data is declared as 1912 `char *', to avoid warnings when a string constant is passed. But 1913 when we use a character as a subscript we must make it unsigned. */ 1914 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, 1915 because it is impossible to allocate 4GB array for some encodings 1916 which have 4 byte character_set like UCS4. */ 1917 1918 # ifndef TRANSLATE 1919 # ifdef WCHAR 1920 # define TRANSLATE(d) \ 1921 ((translate && ((UCHAR_T) (d)) <= 0xff) \ 1922 ? (char) translate[(unsigned char) (d)] : (d)) 1923 # else /* BYTE */ 1924 # define TRANSLATE(d) \ 1925 (translate ? (char) translate[(unsigned char) (d)] : (char) (d)) 1926 # endif /* WCHAR */ 1927 # endif 1928 1929 1930 /* Macros for outputting the compiled pattern into `buffer'. */ 1931 1932 /* If the buffer isn't allocated when it comes in, use this. */ 1933 # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T)) 1934 1935 /* Make sure we have at least N more bytes of space in buffer. */ 1936 # ifdef WCHAR 1937 # define GET_BUFFER_SPACE(n) \ 1938 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \ 1939 + (n)*sizeof(CHAR_T)) > bufp->allocated) \ 1940 EXTEND_BUFFER () 1941 # else /* BYTE */ 1942 # define GET_BUFFER_SPACE(n) \ 1943 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \ 1944 EXTEND_BUFFER () 1945 # endif /* WCHAR */ 1946 1947 /* Make sure we have one more byte of buffer space and then add C to it. */ 1948 # define BUF_PUSH(c) \ 1949 do { \ 1950 GET_BUFFER_SPACE (1); \ 1951 *b++ = (UCHAR_T) (c); \ 1952 } while (0) 1953 1954 1955 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ 1956 # define BUF_PUSH_2(c1, c2) \ 1957 do { \ 1958 GET_BUFFER_SPACE (2); \ 1959 *b++ = (UCHAR_T) (c1); \ 1960 *b++ = (UCHAR_T) (c2); \ 1961 } while (0) 1962 1963 1964 /* As with BUF_PUSH_2, except for three bytes. */ 1965 # define BUF_PUSH_3(c1, c2, c3) \ 1966 do { \ 1967 GET_BUFFER_SPACE (3); \ 1968 *b++ = (UCHAR_T) (c1); \ 1969 *b++ = (UCHAR_T) (c2); \ 1970 *b++ = (UCHAR_T) (c3); \ 1971 } while (0) 1972 1973 /* Store a jump with opcode OP at LOC to location TO. We store a 1974 relative address offset by the three bytes the jump itself occupies. */ 1975 # define STORE_JUMP(op, loc, to) \ 1976 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE))) 1977 1978 /* Likewise, for a two-argument jump. */ 1979 # define STORE_JUMP2(op, loc, to, arg) \ 1980 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg) 1981 1982 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ 1983 # define INSERT_JUMP(op, loc, to) \ 1984 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b) 1985 1986 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ 1987 # define INSERT_JUMP2(op, loc, to, arg) \ 1988 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\ 1989 arg, b) 1990 1991 /* This is not an arbitrary limit: the arguments which represent offsets 1992 into the pattern are two bytes long. So if 2^16 bytes turns out to 1993 be too small, many things would have to change. */ 1994 /* Any other compiler which, like MSC, has allocation limit below 2^16 1995 bytes will have to use approach similar to what was done below for 1996 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up 1997 reallocating to 0 bytes. Such thing is not going to work too well. 1998 You have been warned!! */ 1999 # ifndef DEFINED_ONCE 2000 # if defined _MSC_VER && !defined WIN32 2001 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. 2002 The REALLOC define eliminates a flurry of conversion warnings, 2003 but is not required. */ 2004 # define MAX_BUF_SIZE 65500L 2005 # define REALLOC(p,s) realloc ((p), (size_t) (s)) 2006 # else 2007 # define MAX_BUF_SIZE (1L << 16) 2008 # define REALLOC(p,s) realloc ((p), (s)) 2009 # endif 2010 2011 /* Extend the buffer by twice its current size via realloc and 2012 reset the pointers that pointed into the old block to point to the 2013 correct places in the new one. If extending the buffer results in it 2014 being larger than MAX_BUF_SIZE, then flag memory exhausted. */ 2015 # if __BOUNDED_POINTERS__ 2016 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) 2017 # define MOVE_BUFFER_POINTER(P) \ 2018 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) 2019 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ 2020 else \ 2021 { \ 2022 SET_HIGH_BOUND (b); \ 2023 SET_HIGH_BOUND (begalt); \ 2024 if (fixup_alt_jump) \ 2025 SET_HIGH_BOUND (fixup_alt_jump); \ 2026 if (laststart) \ 2027 SET_HIGH_BOUND (laststart); \ 2028 if (pending_exact) \ 2029 SET_HIGH_BOUND (pending_exact); \ 2030 } 2031 # else 2032 # define MOVE_BUFFER_POINTER(P) (P) += incr 2033 # define ELSE_EXTEND_BUFFER_HIGH_BOUND 2034 # endif 2035 # endif /* not DEFINED_ONCE */ 2036 2037 # ifdef WCHAR 2038 # define EXTEND_BUFFER() \ 2039 do { \ 2040 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ 2041 int wchar_count; \ 2042 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \ 2043 return REG_ESIZE; \ 2044 bufp->allocated <<= 1; \ 2045 if (bufp->allocated > MAX_BUF_SIZE) \ 2046 bufp->allocated = MAX_BUF_SIZE; \ 2047 /* How many characters the new buffer can have? */ \ 2048 wchar_count = bufp->allocated / sizeof(UCHAR_T); \ 2049 if (wchar_count == 0) wchar_count = 1; \ 2050 /* Truncate the buffer to CHAR_T align. */ \ 2051 bufp->allocated = wchar_count * sizeof(UCHAR_T); \ 2052 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \ 2053 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \ 2054 if (COMPILED_BUFFER_VAR == NULL) \ 2055 return REG_ESPACE; \ 2056 /* If the buffer moved, move all the pointers into it. */ \ 2057 if (old_buffer != COMPILED_BUFFER_VAR) \ 2058 { \ 2059 PTR_INT_TYPE incr = COMPILED_BUFFER_VAR - old_buffer; \ 2060 MOVE_BUFFER_POINTER (b); \ 2061 MOVE_BUFFER_POINTER (begalt); \ 2062 if (fixup_alt_jump) \ 2063 MOVE_BUFFER_POINTER (fixup_alt_jump); \ 2064 if (laststart) \ 2065 MOVE_BUFFER_POINTER (laststart); \ 2066 if (pending_exact) \ 2067 MOVE_BUFFER_POINTER (pending_exact); \ 2068 } \ 2069 ELSE_EXTEND_BUFFER_HIGH_BOUND \ 2070 } while (0) 2071 # else /* BYTE */ 2072 # define EXTEND_BUFFER() \ 2073 do { \ 2074 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ 2075 if (bufp->allocated == MAX_BUF_SIZE) \ 2076 return REG_ESIZE; \ 2077 bufp->allocated <<= 1; \ 2078 if (bufp->allocated > MAX_BUF_SIZE) \ 2079 bufp->allocated = MAX_BUF_SIZE; \ 2080 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \ 2081 bufp->allocated); \ 2082 if (COMPILED_BUFFER_VAR == NULL) \ 2083 return REG_ESPACE; \ 2084 /* If the buffer moved, move all the pointers into it. */ \ 2085 if (old_buffer != COMPILED_BUFFER_VAR) \ 2086 { \ 2087 PTR_INT_TYPE incr = COMPILED_BUFFER_VAR - old_buffer; \ 2088 MOVE_BUFFER_POINTER (b); \ 2089 MOVE_BUFFER_POINTER (begalt); \ 2090 if (fixup_alt_jump) \ 2091 MOVE_BUFFER_POINTER (fixup_alt_jump); \ 2092 if (laststart) \ 2093 MOVE_BUFFER_POINTER (laststart); \ 2094 if (pending_exact) \ 2095 MOVE_BUFFER_POINTER (pending_exact); \ 2096 } \ 2097 ELSE_EXTEND_BUFFER_HIGH_BOUND \ 2098 } while (0) 2099 # endif /* WCHAR */ 2100 2101 # ifndef DEFINED_ONCE 2102 /* Since we have one byte reserved for the register number argument to 2103 {start,stop}_memory, the maximum number of groups we can report 2104 things about is what fits in that byte. */ 2105 # define MAX_REGNUM 255 2106 2107 /* But patterns can have more than `MAX_REGNUM' registers. We just 2108 ignore the excess. */ 2109 typedef unsigned regnum_t; 2110 2111 2112 /* Macros for the compile stack. */ 2113 2114 /* Since offsets can go either forwards or backwards, this type needs to 2115 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ 2116 /* int may be not enough when sizeof(int) == 2. */ 2117 typedef long pattern_offset_t; 2118 2119 typedef struct 2120 { 2121 pattern_offset_t begalt_offset; 2122 pattern_offset_t fixup_alt_jump; 2123 pattern_offset_t inner_group_offset; 2124 pattern_offset_t laststart_offset; 2125 regnum_t regnum; 2126 } compile_stack_elt_t; 2127 2128 2129 typedef struct 2130 { 2131 compile_stack_elt_t *stack; 2132 unsigned size; 2133 unsigned avail; /* Offset of next open position. */ 2134 } compile_stack_type; 2135 2136 2137 # define INIT_COMPILE_STACK_SIZE 32 2138 2139 # define COMPILE_STACK_EMPTY (compile_stack.avail == 0) 2140 # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) 2141 2142 /* The next available element. */ 2143 # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) 2144 2145 # endif /* not DEFINED_ONCE */ 2146 2147 /* Set the bit for character C in a list. */ 2148 # ifndef DEFINED_ONCE 2149 # define SET_LIST_BIT(c) \ 2150 (b[((unsigned char) (c)) / BYTEWIDTH] \ 2151 |= 1 << (((unsigned char) c) % BYTEWIDTH)) 2152 # endif /* DEFINED_ONCE */ 2153 2154 /* Get the next unsigned number in the uncompiled pattern. */ 2155 # define GET_UNSIGNED_NUMBER(num) \ 2156 { \ 2157 while (p != pend) \ 2158 { \ 2159 PATFETCH (c); \ 2160 if (c < '0' || c > '9') \ 2161 break; \ 2162 if (num <= RE_DUP_MAX) \ 2163 { \ 2164 if (num < 0) \ 2165 num = 0; \ 2166 num = num * 10 + c - '0'; \ 2167 } \ 2168 } \ 2169 } 2170 2171 # ifndef DEFINED_ONCE 2172 # if defined _LIBC || WIDE_CHAR_SUPPORT 2173 /* The GNU C library provides support for user-defined character classes 2174 and the functions from ISO C amendement 1. */ 2175 # ifdef CHARCLASS_NAME_MAX 2176 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX 2177 # else 2178 /* This shouldn't happen but some implementation might still have this 2179 problem. Use a reasonable default value. */ 2180 # define CHAR_CLASS_MAX_LENGTH 256 2181 # endif 2182 2183 # ifdef _LIBC 2184 # define IS_CHAR_CLASS(string) __wctype (string) 2185 # else 2186 # define IS_CHAR_CLASS(string) wctype (string) 2187 # endif 2188 # else 2189 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ 2190 2191 # define IS_CHAR_CLASS(string) \ 2192 (STREQ (string, "alpha") || STREQ (string, "upper") \ 2193 || STREQ (string, "lower") || STREQ (string, "digit") \ 2194 || STREQ (string, "alnum") || STREQ (string, "xdigit") \ 2195 || STREQ (string, "space") || STREQ (string, "print") \ 2196 || STREQ (string, "punct") || STREQ (string, "graph") \ 2197 || STREQ (string, "cntrl") || STREQ (string, "blank")) 2198 # endif 2199 # endif /* DEFINED_ONCE */ 2200 2201 # ifndef MATCH_MAY_ALLOCATE 2202 2203 /* If we cannot allocate large objects within re_match_2_internal, 2204 we make the fail stack and register vectors global. 2205 The fail stack, we grow to the maximum size when a regexp 2206 is compiled. 2207 The register vectors, we adjust in size each time we 2208 compile a regexp, according to the number of registers it needs. */ 2209 2210 static PREFIX(fail_stack_type) fail_stack; 2211 2212 /* Size with which the following vectors are currently allocated. 2213 That is so we can make them bigger as needed, 2214 but never make them smaller. */ 2215 # ifdef DEFINED_ONCE 2216 static int regs_allocated_size; 2217 2218 static const char ** regstart, ** regend; 2219 static const char ** old_regstart, ** old_regend; 2220 static const char **best_regstart, **best_regend; 2221 static const char **reg_dummy; 2222 # endif /* DEFINED_ONCE */ 2223 2224 static PREFIX(register_info_type) *PREFIX(reg_info); 2225 static PREFIX(register_info_type) *PREFIX(reg_info_dummy); 2226 2227 /* Make the register vectors big enough for NUM_REGS registers, 2228 but don't make them smaller. */ 2229 2230 static void 2231 PREFIX(regex_grow_registers) (int num_regs) 2232 { 2233 if (num_regs > regs_allocated_size) 2234 { 2235 RETALLOC_IF (regstart, num_regs, const char *); 2236 RETALLOC_IF (regend, num_regs, const char *); 2237 RETALLOC_IF (old_regstart, num_regs, const char *); 2238 RETALLOC_IF (old_regend, num_regs, const char *); 2239 RETALLOC_IF (best_regstart, num_regs, const char *); 2240 RETALLOC_IF (best_regend, num_regs, const char *); 2241 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type)); 2242 RETALLOC_IF (reg_dummy, num_regs, const char *); 2243 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type)); 2244 2245 regs_allocated_size = num_regs; 2246 } 2247 } 2248 2249 # endif /* not MATCH_MAY_ALLOCATE */ 2250 2251 # ifndef DEFINED_ONCE 2252 static boolean group_in_compile_stack (compile_stack_type compile_stack, 2253 regnum_t regnum); 2254 # endif /* not DEFINED_ONCE */ 2255 2256 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. 2257 Returns one of error codes defined in `regex.h', or zero for success. 2258 2259 Assumes the `allocated' (and perhaps `buffer') and `translate' 2260 fields are set in BUFP on entry. 2261 2262 If it succeeds, results are put in BUFP (if it returns an error, the 2263 contents of BUFP are undefined): 2264 `buffer' is the compiled pattern; 2265 `syntax' is set to SYNTAX; 2266 `used' is set to the length of the compiled pattern; 2267 `fastmap_accurate' is zero; 2268 `re_nsub' is the number of subexpressions in PATTERN; 2269 `not_bol' and `not_eol' are zero; 2270 2271 The `fastmap' and `newline_anchor' fields are neither 2272 examined nor set. */ 2273 2274 /* Return, freeing storage we allocated. */ 2275 # ifdef WCHAR 2276 # define FREE_STACK_RETURN(value) \ 2277 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value) 2278 # else 2279 # define FREE_STACK_RETURN(value) \ 2280 return (free (compile_stack.stack), value) 2281 # endif /* WCHAR */ 2282 2283 static reg_errcode_t 2284 PREFIX(regex_compile) (const char *ARG_PREFIX(pattern), 2285 size_t ARG_PREFIX(size), reg_syntax_t syntax, 2286 struct re_pattern_buffer *bufp) 2287 { 2288 /* We fetch characters from PATTERN here. Even though PATTERN is 2289 `char *' (i.e., signed), we declare these variables as unsigned, so 2290 they can be reliably used as array indices. */ 2291 register UCHAR_T c, c1; 2292 2293 #ifdef WCHAR 2294 /* A temporary space to keep wchar_t pattern and compiled pattern. */ 2295 CHAR_T *pattern, *COMPILED_BUFFER_VAR; 2296 size_t size; 2297 /* offset buffer for optimization. See convert_mbs_to_wc. */ 2298 int *mbs_offset = NULL; 2299 /* It hold whether each wchar_t is binary data or not. */ 2300 char *is_binary = NULL; 2301 /* A flag whether exactn is handling binary data or not. */ 2302 char is_exactn_bin = FALSE; 2303 #endif /* WCHAR */ 2304 2305 /* A random temporary spot in PATTERN. */ 2306 const CHAR_T *p1; 2307 2308 /* Points to the end of the buffer, where we should append. */ 2309 register UCHAR_T *b; 2310 2311 /* Keeps track of unclosed groups. */ 2312 compile_stack_type compile_stack; 2313 2314 /* Points to the current (ending) position in the pattern. */ 2315 #ifdef WCHAR 2316 const CHAR_T *p; 2317 const CHAR_T *pend; 2318 #else /* BYTE */ 2319 const CHAR_T *p = pattern; 2320 const CHAR_T *pend = pattern + size; 2321 #endif /* WCHAR */ 2322 2323 /* How to translate the characters in the pattern. */ 2324 RE_TRANSLATE_TYPE translate = bufp->translate; 2325 2326 /* Address of the count-byte of the most recently inserted `exactn' 2327 command. This makes it possible to tell if a new exact-match 2328 character can be added to that command or if the character requires 2329 a new `exactn' command. */ 2330 UCHAR_T *pending_exact = 0; 2331 2332 /* Address of start of the most recently finished expression. 2333 This tells, e.g., postfix * where to find the start of its 2334 operand. Reset at the beginning of groups and alternatives. */ 2335 UCHAR_T *laststart = 0; 2336 2337 /* Address of beginning of regexp, or inside of last group. */ 2338 UCHAR_T *begalt; 2339 2340 /* Address of the place where a forward jump should go to the end of 2341 the containing expression. Each alternative of an `or' -- except the 2342 last -- ends with a forward jump of this sort. */ 2343 UCHAR_T *fixup_alt_jump = 0; 2344 2345 /* Counts open-groups as they are encountered. Remembered for the 2346 matching close-group on the compile stack, so the same register 2347 number is put in the stop_memory as the start_memory. */ 2348 regnum_t regnum = 0; 2349 2350 #ifdef WCHAR 2351 /* Initialize the wchar_t PATTERN and offset_buffer. */ 2352 p = pend = pattern = TALLOC(csize + 1, CHAR_T); 2353 mbs_offset = TALLOC(csize + 1, int); 2354 is_binary = TALLOC(csize + 1, char); 2355 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL) 2356 { 2357 free(pattern); 2358 free(mbs_offset); 2359 free(is_binary); 2360 return REG_ESPACE; 2361 } 2362 pattern[csize] = L'\0'; /* sentinel */ 2363 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary); 2364 pend = p + size; 2365 if (size < 0) 2366 { 2367 free(pattern); 2368 free(mbs_offset); 2369 free(is_binary); 2370 return REG_BADPAT; 2371 } 2372 #endif 2373 2374 #ifdef DEBUG 2375 DEBUG_PRINT1 ("\nCompiling pattern: "); 2376 if (debug) 2377 { 2378 unsigned debug_count; 2379 2380 for (debug_count = 0; debug_count < size; debug_count++) 2381 PUT_CHAR (pattern[debug_count]); 2382 putchar ('\n'); 2383 } 2384 #endif /* DEBUG */ 2385 2386 /* Initialize the compile stack. */ 2387 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); 2388 if (compile_stack.stack == NULL) 2389 { 2390 #ifdef WCHAR 2391 free(pattern); 2392 free(mbs_offset); 2393 free(is_binary); 2394 #endif 2395 return REG_ESPACE; 2396 } 2397 2398 compile_stack.size = INIT_COMPILE_STACK_SIZE; 2399 compile_stack.avail = 0; 2400 2401 /* Initialize the pattern buffer. */ 2402 bufp->syntax = syntax; 2403 bufp->fastmap_accurate = 0; 2404 bufp->not_bol = bufp->not_eol = 0; 2405 2406 /* Set `used' to zero, so that if we return an error, the pattern 2407 printer (for debugging) will think there's no pattern. We reset it 2408 at the end. */ 2409 bufp->used = 0; 2410 2411 /* Always count groups, whether or not bufp->no_sub is set. */ 2412 bufp->re_nsub = 0; 2413 2414 #if !defined emacs && !defined SYNTAX_TABLE 2415 /* Initialize the syntax table. */ 2416 init_syntax_once (); 2417 #endif 2418 2419 if (bufp->allocated == 0) 2420 { 2421 if (bufp->buffer) 2422 { /* If zero allocated, but buffer is non-null, try to realloc 2423 enough space. This loses if buffer's address is bogus, but 2424 that is the user's responsibility. */ 2425 #ifdef WCHAR 2426 /* Free bufp->buffer and allocate an array for wchar_t pattern 2427 buffer. */ 2428 free(bufp->buffer); 2429 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T), 2430 UCHAR_T); 2431 #else 2432 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T); 2433 #endif /* WCHAR */ 2434 } 2435 else 2436 { /* Caller did not allocate a buffer. Do it for them. */ 2437 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T), 2438 UCHAR_T); 2439 } 2440 2441 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE); 2442 #ifdef WCHAR 2443 bufp->buffer = (char*)COMPILED_BUFFER_VAR; 2444 #endif /* WCHAR */ 2445 bufp->allocated = INIT_BUF_SIZE; 2446 } 2447 #ifdef WCHAR 2448 else 2449 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer; 2450 #endif 2451 2452 begalt = b = COMPILED_BUFFER_VAR; 2453 2454 /* Loop through the uncompiled pattern until we're at the end. */ 2455 while (p != pend) 2456 { 2457 PATFETCH (c); 2458 2459 switch (c) 2460 { 2461 case '^': 2462 { 2463 if ( /* If at start of pattern, it's an operator. */ 2464 p == pattern + 1 2465 /* If context independent, it's an operator. */ 2466 || syntax & RE_CONTEXT_INDEP_ANCHORS 2467 /* Otherwise, depends on what's come before. */ 2468 || PREFIX(at_begline_loc_p) (pattern, p, syntax)) 2469 BUF_PUSH (begline); 2470 else 2471 goto normal_char; 2472 } 2473 break; 2474 2475 2476 case '$': 2477 { 2478 if ( /* If at end of pattern, it's an operator. */ 2479 p == pend 2480 /* If context independent, it's an operator. */ 2481 || syntax & RE_CONTEXT_INDEP_ANCHORS 2482 /* Otherwise, depends on what's next. */ 2483 || PREFIX(at_endline_loc_p) (p, pend, syntax)) 2484 BUF_PUSH (endline); 2485 else 2486 goto normal_char; 2487 } 2488 break; 2489 2490 2491 case '+': 2492 case '?': 2493 if ((syntax & RE_BK_PLUS_QM) 2494 || (syntax & RE_LIMITED_OPS)) 2495 goto normal_char; 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 throught 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 case pop_failure_jump: 6701 case maybe_pop_jump: 6702 case jump: 6703 case dummy_failure_jump: 6704 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 6705 if (is_a_jump_n) 6706 p1 += OFFSET_ADDRESS_SIZE; 6707 break; 6708 6709 default: 6710 /* do nothing */ ; 6711 } 6712 p1 += mcnt; 6713 6714 /* If the next operation is a jump backwards in the pattern 6715 to an on_failure_jump right before the start_memory 6716 corresponding to this stop_memory, exit from the loop 6717 by forcing a failure after pushing on the stack the 6718 on_failure_jump's jump in the pattern, and d. */ 6719 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump 6720 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory 6721 && p1[2+OFFSET_ADDRESS_SIZE] == *p) 6722 { 6723 /* If this group ever matched anything, then restore 6724 what its registers were before trying this last 6725 failed match, e.g., with `(a*)*b' against `ab' for 6726 regstart[1], and, e.g., with `((a*)*(b*)*)*' 6727 against `aba' for regend[3]. 6728 6729 Also restore the registers for inner groups for, 6730 e.g., `((a*)(b*))*' against `aba' (register 3 would 6731 otherwise get trashed). */ 6732 6733 if (EVER_MATCHED_SOMETHING (reg_info[*p])) 6734 { 6735 unsigned r; 6736 6737 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; 6738 6739 /* Restore this and inner groups' (if any) registers. */ 6740 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1); 6741 r++) 6742 { 6743 regstart[r] = old_regstart[r]; 6744 6745 /* xx why this test? */ 6746 if (old_regend[r] >= regstart[r]) 6747 regend[r] = old_regend[r]; 6748 } 6749 } 6750 p1++; 6751 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 6752 PUSH_FAILURE_POINT (p1 + mcnt, d, -2); 6753 6754 goto fail; 6755 } 6756 } 6757 6758 /* Move past the register number and the inner group count. */ 6759 p += 2; 6760 break; 6761 6762 6763 /* \<digit> has been turned into a `duplicate' command which is 6764 followed by the numeric value of <digit> as the register number. */ 6765 case duplicate: 6766 { 6767 register const CHAR_T *d2, *dend2; 6768 int regno = *p++; /* Get which register to match against. */ 6769 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno); 6770 6771 /* Can't back reference a group which we've never matched. */ 6772 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) 6773 goto fail; 6774 6775 /* Where in input to try to start matching. */ 6776 d2 = regstart[regno]; 6777 6778 /* Where to stop matching; if both the place to start and 6779 the place to stop matching are in the same string, then 6780 set to the place to stop, otherwise, for now have to use 6781 the end of the first string. */ 6782 6783 dend2 = ((FIRST_STRING_P (regstart[regno]) 6784 == FIRST_STRING_P (regend[regno])) 6785 ? regend[regno] : end_match_1); 6786 for (;;) 6787 { 6788 /* If necessary, advance to next segment in register 6789 contents. */ 6790 while (d2 == dend2) 6791 { 6792 if (dend2 == end_match_2) break; 6793 if (dend2 == regend[regno]) break; 6794 6795 /* End of string1 => advance to string2. */ 6796 d2 = string2; 6797 dend2 = regend[regno]; 6798 } 6799 /* At end of register contents => success */ 6800 if (d2 == dend2) break; 6801 6802 /* If necessary, advance to next segment in data. */ 6803 PREFETCH (); 6804 6805 /* How many characters left in this segment to match. */ 6806 mcnt = dend - d; 6807 6808 /* Want how many consecutive characters we can match in 6809 one shot, so, if necessary, adjust the count. */ 6810 if (mcnt > dend2 - d2) 6811 mcnt = dend2 - d2; 6812 6813 /* Compare that many; failure if mismatch, else move 6814 past them. */ 6815 if (translate 6816 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate) 6817 : memcmp (d, d2, mcnt*sizeof(UCHAR_T))) 6818 goto fail; 6819 d += mcnt, d2 += mcnt; 6820 6821 /* Do this because we've match some characters. */ 6822 SET_REGS_MATCHED (); 6823 } 6824 } 6825 break; 6826 6827 6828 /* begline matches the empty string at the beginning of the string 6829 (unless `not_bol' is set in `bufp'), and, if 6830 `newline_anchor' is set, after newlines. */ 6831 case begline: 6832 DEBUG_PRINT1 ("EXECUTING begline.\n"); 6833 6834 if (AT_STRINGS_BEG (d)) 6835 { 6836 if (!bufp->not_bol) break; 6837 } 6838 else if (d[-1] == '\n' && bufp->newline_anchor) 6839 { 6840 break; 6841 } 6842 /* In all other cases, we fail. */ 6843 goto fail; 6844 6845 6846 /* endline is the dual of begline. */ 6847 case endline: 6848 DEBUG_PRINT1 ("EXECUTING endline.\n"); 6849 6850 if (AT_STRINGS_END (d)) 6851 { 6852 if (!bufp->not_eol) break; 6853 } 6854 6855 /* We have to ``prefetch'' the next character. */ 6856 else if ((d == end1 ? *string2 : *d) == '\n' 6857 && bufp->newline_anchor) 6858 { 6859 break; 6860 } 6861 goto fail; 6862 6863 6864 /* Match at the very beginning of the data. */ 6865 case begbuf: 6866 DEBUG_PRINT1 ("EXECUTING begbuf.\n"); 6867 if (AT_STRINGS_BEG (d)) 6868 break; 6869 goto fail; 6870 6871 6872 /* Match at the very end of the data. */ 6873 case endbuf: 6874 DEBUG_PRINT1 ("EXECUTING endbuf.\n"); 6875 if (AT_STRINGS_END (d)) 6876 break; 6877 goto fail; 6878 6879 6880 /* on_failure_keep_string_jump is used to optimize `.*\n'. It 6881 pushes NULL as the value for the string on the stack. Then 6882 `pop_failure_point' will keep the current value for the 6883 string, instead of restoring it. To see why, consider 6884 matching `foo\nbar' against `.*\n'. The .* matches the foo; 6885 then the . fails against the \n. But the next thing we want 6886 to do is match the \n against the \n; if we restored the 6887 string value, we would be back at the foo. 6888 6889 Because this is used only in specific cases, we don't need to 6890 check all the things that `on_failure_jump' does, to make 6891 sure the right things get saved on the stack. Hence we don't 6892 share its code. The only reason to push anything on the 6893 stack at all is that otherwise we would have to change 6894 `anychar's code to do something besides goto fail in this 6895 case; that seems worse than this. */ 6896 case on_failure_keep_string_jump: 6897 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump"); 6898 6899 EXTRACT_NUMBER_AND_INCR (mcnt, p); 6900 #ifdef _LIBC 6901 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt); 6902 #else 6903 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt); 6904 #endif 6905 6906 PUSH_FAILURE_POINT (p + mcnt, NULL, -2); 6907 break; 6908 6909 6910 /* Uses of on_failure_jump: 6911 6912 Each alternative starts with an on_failure_jump that points 6913 to the beginning of the next alternative. Each alternative 6914 except the last ends with a jump that in effect jumps past 6915 the rest of the alternatives. (They really jump to the 6916 ending jump of the following alternative, because tensioning 6917 these jumps is a hassle.) 6918 6919 Repeats start with an on_failure_jump that points past both 6920 the repetition text and either the following jump or 6921 pop_failure_jump back to this on_failure_jump. */ 6922 case on_failure_jump: 6923 on_failure: 6924 DEBUG_PRINT1 ("EXECUTING on_failure_jump"); 6925 6926 EXTRACT_NUMBER_AND_INCR (mcnt, p); 6927 #ifdef _LIBC 6928 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt); 6929 #else 6930 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt); 6931 #endif 6932 6933 /* If this on_failure_jump comes right before a group (i.e., 6934 the original * applied to a group), save the information 6935 for that group and all inner ones, so that if we fail back 6936 to this point, the group's information will be correct. 6937 For example, in \(a*\)*\1, we need the preceding group, 6938 and in \(zz\(a*\)b*\)\2, we need the inner group. */ 6939 6940 /* We can't use `p' to check ahead because we push 6941 a failure point to `p + mcnt' after we do this. */ 6942 p1 = p; 6943 6944 /* We need to skip no_op's before we look for the 6945 start_memory in case this on_failure_jump is happening as 6946 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 6947 against aba. */ 6948 while (p1 < pend && (re_opcode_t) *p1 == no_op) 6949 p1++; 6950 6951 if (p1 < pend && (re_opcode_t) *p1 == start_memory) 6952 { 6953 /* We have a new highest active register now. This will 6954 get reset at the start_memory we are about to get to, 6955 but we will have saved all the registers relevant to 6956 this repetition op, as described above. */ 6957 highest_active_reg = *(p1 + 1) + *(p1 + 2); 6958 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) 6959 lowest_active_reg = *(p1 + 1); 6960 } 6961 6962 DEBUG_PRINT1 (":\n"); 6963 PUSH_FAILURE_POINT (p + mcnt, d, -2); 6964 break; 6965 6966 6967 /* A smart repeat ends with `maybe_pop_jump'. 6968 We change it to either `pop_failure_jump' or `jump'. */ 6969 case maybe_pop_jump: 6970 EXTRACT_NUMBER_AND_INCR (mcnt, p); 6971 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt); 6972 { 6973 register UCHAR_T *p2 = p; 6974 6975 /* Compare the beginning of the repeat with what in the 6976 pattern follows its end. If we can establish that there 6977 is nothing that they would both match, i.e., that we 6978 would have to backtrack because of (as in, e.g., `a*a') 6979 then we can change to pop_failure_jump, because we'll 6980 never have to backtrack. 6981 6982 This is not true in the case of alternatives: in 6983 `(a|ab)*' we do need to backtrack to the `ab' alternative 6984 (e.g., if the string was `ab'). But instead of trying to 6985 detect that here, the alternative has put on a dummy 6986 failure point which is what we will end up popping. */ 6987 6988 /* Skip over open/close-group commands. 6989 If what follows this loop is a ...+ construct, 6990 look at what begins its body, since we will have to 6991 match at least one of that. */ 6992 while (1) 6993 { 6994 if (p2 + 2 < pend 6995 && ((re_opcode_t) *p2 == stop_memory 6996 || (re_opcode_t) *p2 == start_memory)) 6997 p2 += 3; 6998 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend 6999 && (re_opcode_t) *p2 == dummy_failure_jump) 7000 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE; 7001 else 7002 break; 7003 } 7004 7005 p1 = p + mcnt; 7006 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding 7007 to the `maybe_finalize_jump' of this case. Examine what 7008 follows. */ 7009 7010 /* If we're at the end of the pattern, we can change. */ 7011 if (p2 == pend) 7012 { 7013 /* Consider what happens when matching ":\(.*\)" 7014 against ":/". I don't really understand this code 7015 yet. */ 7016 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) 7017 pop_failure_jump; 7018 DEBUG_PRINT1 7019 (" End of pattern: change to `pop_failure_jump'.\n"); 7020 } 7021 7022 else if ((re_opcode_t) *p2 == exactn 7023 #ifdef MBS_SUPPORT 7024 || (re_opcode_t) *p2 == exactn_bin 7025 #endif 7026 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) 7027 { 7028 register UCHAR_T c 7029 = *p2 == (UCHAR_T) endline ? '\n' : p2[2]; 7030 7031 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn 7032 #ifdef MBS_SUPPORT 7033 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin 7034 #endif 7035 ) && p1[3+OFFSET_ADDRESS_SIZE] != c) 7036 { 7037 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) 7038 pop_failure_jump; 7039 #ifdef WCHAR 7040 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n", 7041 (wint_t) c, 7042 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]); 7043 #else 7044 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n", 7045 (char) c, 7046 (char) p1[3+OFFSET_ADDRESS_SIZE]); 7047 #endif 7048 } 7049 7050 #ifndef WCHAR 7051 else if ((re_opcode_t) p1[3] == charset 7052 || (re_opcode_t) p1[3] == charset_not) 7053 { 7054 int negate = (re_opcode_t) p1[3] == charset_not; 7055 7056 if (c < (unsigned) (p1[4] * BYTEWIDTH) 7057 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) 7058 negate = !negate; 7059 7060 /* `negate' is equal to 1 if c would match, which means 7061 that we can't change to pop_failure_jump. */ 7062 if (!negate) 7063 { 7064 p[-3] = (unsigned char) pop_failure_jump; 7065 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7066 } 7067 } 7068 #endif /* not WCHAR */ 7069 } 7070 #ifndef WCHAR 7071 else if ((re_opcode_t) *p2 == charset) 7072 { 7073 /* We win if the first character of the loop is not part 7074 of the charset. */ 7075 if ((re_opcode_t) p1[3] == exactn 7076 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5] 7077 && (p2[2 + p1[5] / BYTEWIDTH] 7078 & (1 << (p1[5] % BYTEWIDTH))))) 7079 { 7080 p[-3] = (unsigned char) pop_failure_jump; 7081 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7082 } 7083 7084 else if ((re_opcode_t) p1[3] == charset_not) 7085 { 7086 int idx; 7087 /* We win if the charset_not inside the loop 7088 lists every character listed in the charset after. */ 7089 for (idx = 0; idx < (int) p2[1]; idx++) 7090 if (! (p2[2 + idx] == 0 7091 || (idx < (int) p1[4] 7092 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) 7093 break; 7094 7095 if (idx == p2[1]) 7096 { 7097 p[-3] = (unsigned char) pop_failure_jump; 7098 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7099 } 7100 } 7101 else if ((re_opcode_t) p1[3] == charset) 7102 { 7103 int idx; 7104 /* We win if the charset inside the loop 7105 has no overlap with the one after the loop. */ 7106 for (idx = 0; 7107 idx < (int) p2[1] && idx < (int) p1[4]; 7108 idx++) 7109 if ((p2[2 + idx] & p1[5 + idx]) != 0) 7110 break; 7111 7112 if (idx == p2[1] || idx == p1[4]) 7113 { 7114 p[-3] = (unsigned char) pop_failure_jump; 7115 DEBUG_PRINT1 (" No match => pop_failure_jump.\n"); 7116 } 7117 } 7118 } 7119 #endif /* not WCHAR */ 7120 } 7121 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */ 7122 if ((re_opcode_t) p[-1] != pop_failure_jump) 7123 { 7124 p[-1] = (UCHAR_T) jump; 7125 DEBUG_PRINT1 (" Match => jump.\n"); 7126 goto unconditional_jump; 7127 } 7128 /* Note fall through. */ 7129 7130 7131 /* The end of a simple repeat has a pop_failure_jump back to 7132 its matching on_failure_jump, where the latter will push a 7133 failure point. The pop_failure_jump takes off failure 7134 points put on by this pop_failure_jump's matching 7135 on_failure_jump; we got through the pattern to here from the 7136 matching on_failure_jump, so didn't fail. */ 7137 case pop_failure_jump: 7138 { 7139 /* We need to pass separate storage for the lowest and 7140 highest registers, even though we don't care about the 7141 actual values. Otherwise, we will restore only one 7142 register from the stack, since lowest will == highest in 7143 `pop_failure_point'. */ 7144 active_reg_t dummy_low_reg, dummy_high_reg; 7145 UCHAR_T *pdummy ATTRIBUTE_UNUSED = NULL; 7146 const CHAR_T *sdummy ATTRIBUTE_UNUSED = NULL; 7147 7148 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n"); 7149 POP_FAILURE_POINT (sdummy, pdummy, 7150 dummy_low_reg, dummy_high_reg, 7151 reg_dummy, reg_dummy, reg_info_dummy); 7152 } 7153 /* Note fall through. */ 7154 7155 unconditional_jump: 7156 #ifdef _LIBC 7157 DEBUG_PRINT2 ("\n%p: ", p); 7158 #else 7159 DEBUG_PRINT2 ("\n0x%x: ", p); 7160 #endif 7161 /* Note fall through. */ 7162 7163 /* Unconditionally jump (without popping any failure points). */ 7164 case jump: 7165 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ 7166 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt); 7167 p += mcnt; /* Do the jump. */ 7168 #ifdef _LIBC 7169 DEBUG_PRINT2 ("(to %p).\n", p); 7170 #else 7171 DEBUG_PRINT2 ("(to 0x%x).\n", p); 7172 #endif 7173 break; 7174 7175 7176 /* We need this opcode so we can detect where alternatives end 7177 in `group_match_null_string_p' et al. */ 7178 case jump_past_alt: 7179 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n"); 7180 goto unconditional_jump; 7181 7182 7183 /* Normally, the on_failure_jump pushes a failure point, which 7184 then gets popped at pop_failure_jump. We will end up at 7185 pop_failure_jump, also, and with a pattern of, say, `a+', we 7186 are skipping over the on_failure_jump, so we have to push 7187 something meaningless for pop_failure_jump to pop. */ 7188 case dummy_failure_jump: 7189 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n"); 7190 /* It doesn't matter what we push for the string here. What 7191 the code at `fail' tests is the value for the pattern. */ 7192 PUSH_FAILURE_POINT (NULL, NULL, -2); 7193 goto unconditional_jump; 7194 7195 7196 /* At the end of an alternative, we need to push a dummy failure 7197 point in case we are followed by a `pop_failure_jump', because 7198 we don't want the failure point for the alternative to be 7199 popped. For example, matching `(a|ab)*' against `aab' 7200 requires that we match the `ab' alternative. */ 7201 case push_dummy_failure: 7202 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n"); 7203 /* See comments just above at `dummy_failure_jump' about the 7204 two zeroes. */ 7205 PUSH_FAILURE_POINT (NULL, NULL, -2); 7206 break; 7207 7208 /* Have to succeed matching what follows at least n times. 7209 After that, handle like `on_failure_jump'. */ 7210 case succeed_n: 7211 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); 7212 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt); 7213 7214 assert (mcnt >= 0); 7215 /* Originally, this is how many times we HAVE to succeed. */ 7216 if (mcnt > 0) 7217 { 7218 mcnt--; 7219 p += OFFSET_ADDRESS_SIZE; 7220 STORE_NUMBER_AND_INCR (p, mcnt); 7221 #ifdef _LIBC 7222 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE 7223 , mcnt); 7224 #else 7225 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE 7226 , mcnt); 7227 #endif 7228 } 7229 else if (mcnt == 0) 7230 { 7231 #ifdef _LIBC 7232 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n", 7233 p + OFFSET_ADDRESS_SIZE); 7234 #else 7235 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", 7236 p + OFFSET_ADDRESS_SIZE); 7237 #endif /* _LIBC */ 7238 7239 #ifdef WCHAR 7240 p[1] = (UCHAR_T) no_op; 7241 #else 7242 p[2] = (UCHAR_T) no_op; 7243 p[3] = (UCHAR_T) no_op; 7244 #endif /* WCHAR */ 7245 goto on_failure; 7246 } 7247 break; 7248 7249 case jump_n: 7250 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); 7251 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt); 7252 7253 /* Originally, this is how many times we CAN jump. */ 7254 if (mcnt) 7255 { 7256 mcnt--; 7257 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt); 7258 7259 #ifdef _LIBC 7260 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE, 7261 mcnt); 7262 #else 7263 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE, 7264 mcnt); 7265 #endif /* _LIBC */ 7266 goto unconditional_jump; 7267 } 7268 /* If don't have to jump any more, skip over the rest of command. */ 7269 else 7270 p += 2 * OFFSET_ADDRESS_SIZE; 7271 break; 7272 7273 case set_number_at: 7274 { 7275 DEBUG_PRINT1 ("EXECUTING set_number_at.\n"); 7276 7277 EXTRACT_NUMBER_AND_INCR (mcnt, p); 7278 p1 = p + mcnt; 7279 EXTRACT_NUMBER_AND_INCR (mcnt, p); 7280 #ifdef _LIBC 7281 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt); 7282 #else 7283 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt); 7284 #endif 7285 STORE_NUMBER (p1, mcnt); 7286 break; 7287 } 7288 7289 #if 0 7290 /* The DEC Alpha C compiler 3.x generates incorrect code for the 7291 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of 7292 AT_WORD_BOUNDARY, so this code is disabled. Expanding the 7293 macro and introducing temporary variables works around the bug. */ 7294 7295 case wordbound: 7296 DEBUG_PRINT1 ("EXECUTING wordbound.\n"); 7297 if (AT_WORD_BOUNDARY (d)) 7298 break; 7299 goto fail; 7300 7301 case notwordbound: 7302 DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); 7303 if (AT_WORD_BOUNDARY (d)) 7304 goto fail; 7305 break; 7306 #else 7307 case wordbound: 7308 { 7309 boolean prevchar, thischar; 7310 7311 DEBUG_PRINT1 ("EXECUTING wordbound.\n"); 7312 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) 7313 break; 7314 7315 prevchar = WORDCHAR_P (d - 1); 7316 thischar = WORDCHAR_P (d); 7317 if (prevchar != thischar) 7318 break; 7319 goto fail; 7320 } 7321 7322 case notwordbound: 7323 { 7324 boolean prevchar, thischar; 7325 7326 DEBUG_PRINT1 ("EXECUTING notwordbound.\n"); 7327 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) 7328 goto fail; 7329 7330 prevchar = WORDCHAR_P (d - 1); 7331 thischar = WORDCHAR_P (d); 7332 if (prevchar != thischar) 7333 goto fail; 7334 break; 7335 } 7336 #endif 7337 7338 case wordbeg: 7339 DEBUG_PRINT1 ("EXECUTING wordbeg.\n"); 7340 if (!AT_STRINGS_END (d) && WORDCHAR_P (d) 7341 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) 7342 break; 7343 goto fail; 7344 7345 case wordend: 7346 DEBUG_PRINT1 ("EXECUTING wordend.\n"); 7347 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) 7348 && (AT_STRINGS_END (d) || !WORDCHAR_P (d))) 7349 break; 7350 goto fail; 7351 7352 #ifdef emacs 7353 case before_dot: 7354 DEBUG_PRINT1 ("EXECUTING before_dot.\n"); 7355 if (PTR_CHAR_POS ((unsigned char *) d) >= point) 7356 goto fail; 7357 break; 7358 7359 case at_dot: 7360 DEBUG_PRINT1 ("EXECUTING at_dot.\n"); 7361 if (PTR_CHAR_POS ((unsigned char *) d) != point) 7362 goto fail; 7363 break; 7364 7365 case after_dot: 7366 DEBUG_PRINT1 ("EXECUTING after_dot.\n"); 7367 if (PTR_CHAR_POS ((unsigned char *) d) <= point) 7368 goto fail; 7369 break; 7370 7371 case syntaxspec: 7372 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt); 7373 mcnt = *p++; 7374 goto matchsyntax; 7375 7376 case wordchar: 7377 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n"); 7378 mcnt = (int) Sword; 7379 matchsyntax: 7380 PREFETCH (); 7381 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ 7382 d++; 7383 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt) 7384 goto fail; 7385 SET_REGS_MATCHED (); 7386 break; 7387 7388 case notsyntaxspec: 7389 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt); 7390 mcnt = *p++; 7391 goto matchnotsyntax; 7392 7393 case notwordchar: 7394 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n"); 7395 mcnt = (int) Sword; 7396 matchnotsyntax: 7397 PREFETCH (); 7398 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ 7399 d++; 7400 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt) 7401 goto fail; 7402 SET_REGS_MATCHED (); 7403 break; 7404 7405 #else /* not emacs */ 7406 case wordchar: 7407 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n"); 7408 PREFETCH (); 7409 if (!WORDCHAR_P (d)) 7410 goto fail; 7411 SET_REGS_MATCHED (); 7412 d++; 7413 break; 7414 7415 case notwordchar: 7416 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n"); 7417 PREFETCH (); 7418 if (WORDCHAR_P (d)) 7419 goto fail; 7420 SET_REGS_MATCHED (); 7421 d++; 7422 break; 7423 #endif /* not emacs */ 7424 7425 default: 7426 abort (); 7427 } 7428 continue; /* Successfully executed one pattern command; keep going. */ 7429 7430 7431 /* We goto here if a matching operation fails. */ 7432 fail: 7433 if (!FAIL_STACK_EMPTY ()) 7434 { /* A restart point is known. Restore to that state. */ 7435 DEBUG_PRINT1 ("\nFAIL:\n"); 7436 POP_FAILURE_POINT (d, p, 7437 lowest_active_reg, highest_active_reg, 7438 regstart, regend, reg_info); 7439 7440 /* If this failure point is a dummy, try the next one. */ 7441 if (!p) 7442 goto fail; 7443 7444 /* If we failed to the end of the pattern, don't examine *p. */ 7445 assert (p <= pend); 7446 if (p < pend) 7447 { 7448 boolean is_a_jump_n = false; 7449 7450 /* If failed to a backwards jump that's part of a repetition 7451 loop, need to pop this failure point and use the next one. */ 7452 switch ((re_opcode_t) *p) 7453 { 7454 case jump_n: 7455 is_a_jump_n = true; 7456 case maybe_pop_jump: 7457 case pop_failure_jump: 7458 case jump: 7459 p1 = p + 1; 7460 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7461 p1 += mcnt; 7462 7463 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) 7464 || (!is_a_jump_n 7465 && (re_opcode_t) *p1 == on_failure_jump)) 7466 goto fail; 7467 break; 7468 default: 7469 /* do nothing */ ; 7470 } 7471 } 7472 7473 if (d >= string1 && d <= end1) 7474 dend = end_match_1; 7475 } 7476 else 7477 break; /* Matching at this starting point really fails. */ 7478 } /* for (;;) */ 7479 7480 if (best_regs_set) 7481 goto restore_best_regs; 7482 7483 FREE_VARIABLES (); 7484 7485 return -1; /* Failure to match. */ 7486 } /* re_match_2 */ 7487 7488 /* Subroutine definitions for re_match_2. */ 7489 7490 7491 /* We are passed P pointing to a register number after a start_memory. 7492 7493 Return true if the pattern up to the corresponding stop_memory can 7494 match the empty string, and false otherwise. 7495 7496 If we find the matching stop_memory, sets P to point to one past its number. 7497 Otherwise, sets P to an undefined byte less than or equal to END. 7498 7499 We don't handle duplicates properly (yet). */ 7500 7501 static boolean 7502 PREFIX(group_match_null_string_p) (UCHAR_T **p, UCHAR_T *end, 7503 PREFIX(register_info_type) *reg_info) 7504 { 7505 int mcnt; 7506 /* Point to after the args to the start_memory. */ 7507 UCHAR_T *p1 = *p + 2; 7508 7509 while (p1 < end) 7510 { 7511 /* Skip over opcodes that can match nothing, and return true or 7512 false, as appropriate, when we get to one that can't, or to the 7513 matching stop_memory. */ 7514 7515 switch ((re_opcode_t) *p1) 7516 { 7517 /* Could be either a loop or a series of alternatives. */ 7518 case on_failure_jump: 7519 p1++; 7520 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7521 7522 /* If the next operation is not a jump backwards in the 7523 pattern. */ 7524 7525 if (mcnt >= 0) 7526 { 7527 /* Go through the on_failure_jumps of the alternatives, 7528 seeing if any of the alternatives cannot match nothing. 7529 The last alternative starts with only a jump, 7530 whereas the rest start with on_failure_jump and end 7531 with a jump, e.g., here is the pattern for `a|b|c': 7532 7533 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 7534 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 7535 /exactn/1/c 7536 7537 So, we have to first go through the first (n-1) 7538 alternatives and then deal with the last one separately. */ 7539 7540 7541 /* Deal with the first (n-1) alternatives, which start 7542 with an on_failure_jump (see above) that jumps to right 7543 past a jump_past_alt. */ 7544 7545 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] == 7546 jump_past_alt) 7547 { 7548 /* `mcnt' holds how many bytes long the alternative 7549 is, including the ending `jump_past_alt' and 7550 its number. */ 7551 7552 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt - 7553 (1 + OFFSET_ADDRESS_SIZE), 7554 reg_info)) 7555 return false; 7556 7557 /* Move to right after this alternative, including the 7558 jump_past_alt. */ 7559 p1 += mcnt; 7560 7561 /* Break if it's the beginning of an n-th alternative 7562 that doesn't begin with an on_failure_jump. */ 7563 if ((re_opcode_t) *p1 != on_failure_jump) 7564 break; 7565 7566 /* Still have to check that it's not an n-th 7567 alternative that starts with an on_failure_jump. */ 7568 p1++; 7569 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7570 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] != 7571 jump_past_alt) 7572 { 7573 /* Get to the beginning of the n-th alternative. */ 7574 p1 -= 1 + OFFSET_ADDRESS_SIZE; 7575 break; 7576 } 7577 } 7578 7579 /* Deal with the last alternative: go back and get number 7580 of the `jump_past_alt' just before it. `mcnt' contains 7581 the length of the alternative. */ 7582 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE); 7583 7584 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info)) 7585 return false; 7586 7587 p1 += mcnt; /* Get past the n-th alternative. */ 7588 } /* if mcnt > 0 */ 7589 break; 7590 7591 7592 case stop_memory: 7593 assert (p1[1] == **p); 7594 *p = p1 + 2; 7595 return true; 7596 7597 7598 default: 7599 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) 7600 return false; 7601 } 7602 } /* while p1 < end */ 7603 7604 return false; 7605 } /* group_match_null_string_p */ 7606 7607 7608 /* Similar to group_match_null_string_p, but doesn't deal with alternatives: 7609 It expects P to be the first byte of a single alternative and END one 7610 byte past the last. The alternative can contain groups. */ 7611 7612 static boolean 7613 PREFIX(alt_match_null_string_p) (UCHAR_T *p, UCHAR_T *end, 7614 PREFIX(register_info_type) *reg_info) 7615 { 7616 int mcnt; 7617 UCHAR_T *p1 = p; 7618 7619 while (p1 < end) 7620 { 7621 /* Skip over opcodes that can match nothing, and break when we get 7622 to one that can't. */ 7623 7624 switch ((re_opcode_t) *p1) 7625 { 7626 /* It's a loop. */ 7627 case on_failure_jump: 7628 p1++; 7629 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7630 p1 += mcnt; 7631 break; 7632 7633 default: 7634 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) 7635 return false; 7636 } 7637 } /* while p1 < end */ 7638 7639 return true; 7640 } /* alt_match_null_string_p */ 7641 7642 7643 /* Deals with the ops common to group_match_null_string_p and 7644 alt_match_null_string_p. 7645 7646 Sets P to one after the op and its arguments, if any. */ 7647 7648 static boolean 7649 PREFIX(common_op_match_null_string_p) (UCHAR_T **p, UCHAR_T *end, 7650 PREFIX(register_info_type) *reg_info) 7651 { 7652 int mcnt; 7653 boolean ret; 7654 int reg_no; 7655 UCHAR_T *p1 = *p; 7656 7657 switch ((re_opcode_t) *p1++) 7658 { 7659 case no_op: 7660 case begline: 7661 case endline: 7662 case begbuf: 7663 case endbuf: 7664 case wordbeg: 7665 case wordend: 7666 case wordbound: 7667 case notwordbound: 7668 #ifdef emacs 7669 case before_dot: 7670 case at_dot: 7671 case after_dot: 7672 #endif 7673 break; 7674 7675 case start_memory: 7676 reg_no = *p1; 7677 assert (reg_no > 0 && reg_no <= MAX_REGNUM); 7678 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info); 7679 7680 /* Have to set this here in case we're checking a group which 7681 contains a group and a back reference to it. */ 7682 7683 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) 7684 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; 7685 7686 if (!ret) 7687 return false; 7688 break; 7689 7690 /* If this is an optimized succeed_n for zero times, make the jump. */ 7691 case jump: 7692 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7693 if (mcnt >= 0) 7694 p1 += mcnt; 7695 else 7696 return false; 7697 break; 7698 7699 case succeed_n: 7700 /* Get to the number of times to succeed. */ 7701 p1 += OFFSET_ADDRESS_SIZE; 7702 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7703 7704 if (mcnt == 0) 7705 { 7706 p1 -= 2 * OFFSET_ADDRESS_SIZE; 7707 EXTRACT_NUMBER_AND_INCR (mcnt, p1); 7708 p1 += mcnt; 7709 } 7710 else 7711 return false; 7712 break; 7713 7714 case duplicate: 7715 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) 7716 return false; 7717 break; 7718 7719 case set_number_at: 7720 p1 += 2 * OFFSET_ADDRESS_SIZE; 7721 7722 default: 7723 /* All other opcodes mean we cannot match the empty string. */ 7724 return false; 7725 } 7726 7727 *p = p1; 7728 return true; 7729 } /* common_op_match_null_string_p */ 7730 7731 7732 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN 7733 bytes; nonzero otherwise. */ 7734 7735 static int 7736 PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2, register int len, 7737 RE_TRANSLATE_TYPE translate) 7738 { 7739 register const UCHAR_T *p1 = (const UCHAR_T *) s1; 7740 register const UCHAR_T *p2 = (const UCHAR_T *) s2; 7741 while (len) 7742 { 7743 #ifdef WCHAR 7744 if (((*p1<=0xff)?translate[*p1++]:*p1++) 7745 != ((*p2<=0xff)?translate[*p2++]:*p2++)) 7746 return 1; 7747 #else /* BYTE */ 7748 if (translate[*p1++] != translate[*p2++]) return 1; 7749 #endif /* WCHAR */ 7750 len--; 7751 } 7752 return 0; 7753 } 7754 7755 7756 #else /* not INSIDE_RECURSION */ 7757 7758 /* Entry points for GNU code. */ 7759 7760 /* re_compile_pattern is the GNU regular expression compiler: it 7761 compiles PATTERN (of length SIZE) and puts the result in BUFP. 7762 Returns 0 if the pattern was valid, otherwise an error string. 7763 7764 Assumes the `allocated' (and perhaps `buffer') and `translate' fields 7765 are set in BUFP on entry. 7766 7767 We call regex_compile to do the actual compilation. */ 7768 7769 const char * 7770 re_compile_pattern (const char *pattern, size_t length, 7771 struct re_pattern_buffer *bufp) 7772 { 7773 reg_errcode_t ret; 7774 7775 /* GNU code is written to assume at least RE_NREGS registers will be set 7776 (and at least one extra will be -1). */ 7777 bufp->regs_allocated = REGS_UNALLOCATED; 7778 7779 /* And GNU code determines whether or not to get register information 7780 by passing null for the REGS argument to re_match, etc., not by 7781 setting no_sub. */ 7782 bufp->no_sub = 0; 7783 7784 /* Match anchors at newline. */ 7785 bufp->newline_anchor = 1; 7786 7787 # ifdef MBS_SUPPORT 7788 if (MB_CUR_MAX != 1) 7789 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp); 7790 else 7791 # endif 7792 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp); 7793 7794 if (!ret) 7795 return NULL; 7796 return gettext (re_error_msgid[(int) ret]); 7797 } 7798 #ifdef _LIBC 7799 weak_alias (__re_compile_pattern, re_compile_pattern) 7800 #endif 7801 7802 /* Entry points compatible with 4.2 BSD regex library. We don't define 7803 them unless specifically requested. */ 7804 7805 #if defined _REGEX_RE_COMP || defined _LIBC 7806 7807 /* BSD has one and only one pattern buffer. */ 7808 static struct re_pattern_buffer re_comp_buf; 7809 7810 char * 7811 #ifdef _LIBC 7812 /* Make these definitions weak in libc, so POSIX programs can redefine 7813 these names if they don't use our functions, and still use 7814 regcomp/regexec below without link errors. */ 7815 weak_function 7816 #endif 7817 re_comp (const char *s) 7818 { 7819 reg_errcode_t ret; 7820 7821 if (!s) 7822 { 7823 if (!re_comp_buf.buffer) 7824 return (char *) gettext ("No previous regular expression"); 7825 return 0; 7826 } 7827 7828 if (!re_comp_buf.buffer) 7829 { 7830 re_comp_buf.buffer = (unsigned char *) malloc (200); 7831 if (re_comp_buf.buffer == NULL) 7832 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); 7833 re_comp_buf.allocated = 200; 7834 7835 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH); 7836 if (re_comp_buf.fastmap == NULL) 7837 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); 7838 } 7839 7840 /* Since `re_exec' always passes NULL for the `regs' argument, we 7841 don't need to initialize the pattern buffer fields which affect it. */ 7842 7843 /* Match anchors at newlines. */ 7844 re_comp_buf.newline_anchor = 1; 7845 7846 # ifdef MBS_SUPPORT 7847 if (MB_CUR_MAX != 1) 7848 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); 7849 else 7850 # endif 7851 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); 7852 7853 if (!ret) 7854 return NULL; 7855 7856 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ 7857 return (char *) gettext (re_error_msgid[(int) ret]); 7858 } 7859 7860 7861 int 7862 #ifdef _LIBC 7863 weak_function 7864 #endif 7865 re_exec (const char *s) 7866 { 7867 const int len = strlen (s); 7868 return 7869 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0); 7870 } 7871 7872 #endif /* _REGEX_RE_COMP */ 7873 7874 /* POSIX.2 functions. Don't define these for Emacs. */ 7875 7876 #ifndef emacs 7877 7878 /* regcomp takes a regular expression as a string and compiles it. 7879 7880 PREG is a regex_t *. We do not expect any fields to be initialized, 7881 since POSIX says we shouldn't. Thus, we set 7882 7883 `buffer' to the compiled pattern; 7884 `used' to the length of the compiled pattern; 7885 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the 7886 REG_EXTENDED bit in CFLAGS is set; otherwise, to 7887 RE_SYNTAX_POSIX_BASIC; 7888 `newline_anchor' to REG_NEWLINE being set in CFLAGS; 7889 `fastmap' to an allocated space for the fastmap; 7890 `fastmap_accurate' to zero; 7891 `re_nsub' to the number of subexpressions in PATTERN. 7892 7893 PATTERN is the address of the pattern string. 7894 7895 CFLAGS is a series of bits which affect compilation. 7896 7897 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we 7898 use POSIX basic syntax. 7899 7900 If REG_NEWLINE is set, then . and [^...] don't match newline. 7901 Also, regexec will try a match beginning after every newline. 7902 7903 If REG_ICASE is set, then we considers upper- and lowercase 7904 versions of letters to be equivalent when matching. 7905 7906 If REG_NOSUB is set, then when PREG is passed to regexec, that 7907 routine will report only success or failure, and nothing about the 7908 registers. 7909 7910 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for 7911 the return codes and their meanings.) */ 7912 7913 int 7914 regcomp (regex_t *preg, const char *pattern, int cflags) 7915 { 7916 reg_errcode_t ret; 7917 reg_syntax_t syntax 7918 = (cflags & REG_EXTENDED) ? 7919 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; 7920 7921 /* regex_compile will allocate the space for the compiled pattern. */ 7922 preg->buffer = 0; 7923 preg->allocated = 0; 7924 preg->used = 0; 7925 7926 /* Try to allocate space for the fastmap. */ 7927 preg->fastmap = (char *) malloc (1 << BYTEWIDTH); 7928 7929 if (cflags & REG_ICASE) 7930 { 7931 int i; 7932 7933 preg->translate 7934 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE 7935 * sizeof (*(RE_TRANSLATE_TYPE)0)); 7936 if (preg->translate == NULL) 7937 return (int) REG_ESPACE; 7938 7939 /* Map uppercase characters to corresponding lowercase ones. */ 7940 for (i = 0; i < CHAR_SET_SIZE; i++) 7941 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; 7942 } 7943 else 7944 preg->translate = NULL; 7945 7946 /* If REG_NEWLINE is set, newlines are treated differently. */ 7947 if (cflags & REG_NEWLINE) 7948 { /* REG_NEWLINE implies neither . nor [^...] match newline. */ 7949 syntax &= ~RE_DOT_NEWLINE; 7950 syntax |= RE_HAT_LISTS_NOT_NEWLINE; 7951 /* It also changes the matching behavior. */ 7952 preg->newline_anchor = 1; 7953 } 7954 else 7955 preg->newline_anchor = 0; 7956 7957 preg->no_sub = !!(cflags & REG_NOSUB); 7958 7959 /* POSIX says a null character in the pattern terminates it, so we 7960 can use strlen here in compiling the pattern. */ 7961 # ifdef MBS_SUPPORT 7962 if (MB_CUR_MAX != 1) 7963 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg); 7964 else 7965 # endif 7966 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg); 7967 7968 /* POSIX doesn't distinguish between an unmatched open-group and an 7969 unmatched close-group: both are REG_EPAREN. */ 7970 if (ret == REG_ERPAREN) ret = REG_EPAREN; 7971 7972 if (ret == REG_NOERROR && preg->fastmap) 7973 { 7974 /* Compute the fastmap now, since regexec cannot modify the pattern 7975 buffer. */ 7976 if (re_compile_fastmap (preg) == -2) 7977 { 7978 /* Some error occurred while computing the fastmap, just forget 7979 about it. */ 7980 free (preg->fastmap); 7981 preg->fastmap = NULL; 7982 } 7983 } 7984 7985 return (int) ret; 7986 } 7987 #ifdef _LIBC 7988 weak_alias (__regcomp, regcomp) 7989 #endif 7990 7991 7992 /* regexec searches for a given pattern, specified by PREG, in the 7993 string STRING. 7994 7995 If NMATCH is zero or REG_NOSUB was set in the cflags argument to 7996 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at 7997 least NMATCH elements, and we set them to the offsets of the 7998 corresponding matched substrings. 7999 8000 EFLAGS specifies `execution flags' which affect matching: if 8001 REG_NOTBOL is set, then ^ does not match at the beginning of the 8002 string; if REG_NOTEOL is set, then $ does not match at the end. 8003 8004 We return 0 if we find a match and REG_NOMATCH if not. */ 8005 8006 int 8007 regexec (const regex_t *preg, const char *string, size_t nmatch, 8008 regmatch_t pmatch[], int eflags) 8009 { 8010 int ret; 8011 struct re_registers regs; 8012 regex_t private_preg; 8013 int len = strlen (string); 8014 boolean want_reg_info = !preg->no_sub && nmatch > 0; 8015 8016 private_preg = *preg; 8017 8018 private_preg.not_bol = !!(eflags & REG_NOTBOL); 8019 private_preg.not_eol = !!(eflags & REG_NOTEOL); 8020 8021 /* The user has told us exactly how many registers to return 8022 information about, via `nmatch'. We have to pass that on to the 8023 matching routines. */ 8024 private_preg.regs_allocated = REGS_FIXED; 8025 8026 if (want_reg_info) 8027 { 8028 regs.num_regs = nmatch; 8029 regs.start = TALLOC (nmatch * 2, regoff_t); 8030 if (regs.start == NULL) 8031 return (int) REG_NOMATCH; 8032 regs.end = regs.start + nmatch; 8033 } 8034 8035 /* Perform the searching operation. */ 8036 ret = re_search (&private_preg, string, len, 8037 /* start: */ 0, /* range: */ len, 8038 want_reg_info ? ®s : (struct re_registers *) 0); 8039 8040 /* Copy the register information to the POSIX structure. */ 8041 if (want_reg_info) 8042 { 8043 if (ret >= 0) 8044 { 8045 unsigned r; 8046 8047 for (r = 0; r < nmatch; r++) 8048 { 8049 pmatch[r].rm_so = regs.start[r]; 8050 pmatch[r].rm_eo = regs.end[r]; 8051 } 8052 } 8053 8054 /* If we needed the temporary register info, free the space now. */ 8055 free (regs.start); 8056 } 8057 8058 /* We want zero return to mean success, unlike `re_search'. */ 8059 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; 8060 } 8061 #ifdef _LIBC 8062 weak_alias (__regexec, regexec) 8063 #endif 8064 8065 8066 /* Returns a message corresponding to an error code, ERRCODE, returned 8067 from either regcomp or regexec. We don't use PREG here. */ 8068 8069 size_t 8070 regerror (int errcode, const regex_t *preg ATTRIBUTE_UNUSED, 8071 char *errbuf, size_t errbuf_size) 8072 { 8073 const char *msg; 8074 size_t msg_size; 8075 8076 if (errcode < 0 8077 || errcode >= (int) (sizeof (re_error_msgid) 8078 / sizeof (re_error_msgid[0]))) 8079 /* Only error codes returned by the rest of the code should be passed 8080 to this routine. If we are given anything else, or if other regex 8081 code generates an invalid error code, then the program has a bug. 8082 Dump core so we can fix it. */ 8083 abort (); 8084 8085 msg = gettext (re_error_msgid[errcode]); 8086 8087 msg_size = strlen (msg) + 1; /* Includes the null. */ 8088 8089 if (errbuf_size != 0) 8090 { 8091 if (msg_size > errbuf_size) 8092 { 8093 #if defined HAVE_MEMPCPY || defined _LIBC 8094 *((char *) mempcpy (errbuf, msg, errbuf_size - 1)) = '\0'; 8095 #else 8096 memcpy (errbuf, msg, errbuf_size - 1); 8097 errbuf[errbuf_size - 1] = 0; 8098 #endif 8099 } 8100 else 8101 memcpy (errbuf, msg, msg_size); 8102 } 8103 8104 return msg_size; 8105 } 8106 #ifdef _LIBC 8107 weak_alias (__regerror, regerror) 8108 #endif 8109 8110 8111 /* Free dynamically allocated space used by PREG. */ 8112 8113 void 8114 regfree (regex_t *preg) 8115 { 8116 free (preg->buffer); 8117 preg->buffer = NULL; 8118 8119 preg->allocated = 0; 8120 preg->used = 0; 8121 8122 free (preg->fastmap); 8123 preg->fastmap = NULL; 8124 preg->fastmap_accurate = 0; 8125 8126 free (preg->translate); 8127 preg->translate = NULL; 8128 } 8129 #ifdef _LIBC 8130 weak_alias (__regfree, regfree) 8131 #endif 8132 8133 #endif /* not emacs */ 8134 8135 #endif /* not INSIDE_RECURSION */ 8136 8137 8138 #undef STORE_NUMBER 8139 #undef STORE_NUMBER_AND_INCR 8140 #undef EXTRACT_NUMBER 8141 #undef EXTRACT_NUMBER_AND_INCR 8142 8143 #undef DEBUG_PRINT_COMPILED_PATTERN 8144 #undef DEBUG_PRINT_DOUBLE_STRING 8145 8146 #undef INIT_FAIL_STACK 8147 #undef RESET_FAIL_STACK 8148 #undef DOUBLE_FAIL_STACK 8149 #undef PUSH_PATTERN_OP 8150 #undef PUSH_FAILURE_POINTER 8151 #undef PUSH_FAILURE_INT 8152 #undef PUSH_FAILURE_ELT 8153 #undef POP_FAILURE_POINTER 8154 #undef POP_FAILURE_INT 8155 #undef POP_FAILURE_ELT 8156 #undef DEBUG_PUSH 8157 #undef DEBUG_POP 8158 #undef PUSH_FAILURE_POINT 8159 #undef POP_FAILURE_POINT 8160 8161 #undef REG_UNSET_VALUE 8162 #undef REG_UNSET 8163 8164 #undef PATFETCH 8165 #undef PATFETCH_RAW 8166 #undef PATUNFETCH 8167 #undef TRANSLATE 8168 8169 #undef INIT_BUF_SIZE 8170 #undef GET_BUFFER_SPACE 8171 #undef BUF_PUSH 8172 #undef BUF_PUSH_2 8173 #undef BUF_PUSH_3 8174 #undef STORE_JUMP 8175 #undef STORE_JUMP2 8176 #undef INSERT_JUMP 8177 #undef INSERT_JUMP2 8178 #undef EXTEND_BUFFER 8179 #undef GET_UNSIGNED_NUMBER 8180 #undef FREE_STACK_RETURN 8181 8182 # undef POINTER_TO_OFFSET 8183 # undef MATCHING_IN_FRST_STRING 8184 # undef PREFETCH 8185 # undef AT_STRINGS_BEG 8186 # undef AT_STRINGS_END 8187 # undef WORDCHAR_P 8188 # undef FREE_VAR 8189 # undef FREE_VARIABLES 8190 # undef NO_HIGHEST_ACTIVE_REG 8191 # undef NO_LOWEST_ACTIVE_REG 8192 8193 # undef CHAR_T 8194 # undef UCHAR_T 8195 # undef COMPILED_BUFFER_VAR 8196 # undef OFFSET_ADDRESS_SIZE 8197 # undef CHAR_CLASS_SIZE 8198 # undef PREFIX 8199 # undef ARG_PREFIX 8200 # undef PUT_CHAR 8201 # undef BYTE 8202 # undef WCHAR 8203 8204 # define DEFINED_ONCE 8205