projects / deliverable / binutils-gdb.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
daily update
[deliverable/binutils-gdb.git] / libiberty / regex.c
CommitLineData
2a6ef469
DD		 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.)
a2832523 5 Copyright (C) 1993-1999, 2000, 2001, 2002 Free Software Foundation, Inc.
6ad8a379 6 This file is part of the GNU C Library.
2a6ef469
DD		 7
8 The GNU C Library is free software; you can redistribute it and/or
6ad8a379
DD		 9 modify it under the terms of the GNU Lesser General Public
10 License as published by the Free Software Foundation; either
11 version 2.1 of the License, or (at your option) any later version.
2a6ef469
DD		 12
13 The GNU C Library is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
6ad8a379 16 Lesser General Public License for more details.
2a6ef469 17
6ad8a379
DD		 18 You should have received a copy of the GNU Lesser General Public
19 License along with the GNU C Library; if not, write to the Free
20 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
21 02111-1307 USA. */
2a6ef469
DD		 22
23/* This file has been modified for usage in libiberty. It includes "xregex.h"
24 instead of <regex.h>. The "xregex.h" header file renames all external
25 routines with an "x" prefix so they do not collide with the native regex
26 routines or with other components regex routines. */
6ad8a379 27/* AIX requires this to be the first thing in the file. */
2a6ef469
DD		 28#if defined _AIX && !defined REGEX_MALLOC
29 #pragma alloca
30#endif
31
32#undef _GNU_SOURCE
33#define _GNU_SOURCE
34
35#ifdef HAVE_CONFIG_H
36# include <config.h>
37#endif
38
39#ifndef PARAMS
40# if defined __GNUC__ || (defined __STDC__ && __STDC__)
41# define PARAMS(args) args
42# else
43# define PARAMS(args) ()
44# endif /* GCC. */
45#endif /* Not PARAMS. */
46
47#ifndef INSIDE_RECURSION
48
49# if defined STDC_HEADERS && !defined emacs
50# include <stddef.h>
51# else
52/* We need this for `regex.h', and perhaps for the Emacs include files. */
53# include <sys/types.h>
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. */
1a78a35a 98# if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC
2a6ef469
DD		 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
132char *malloc ();
133char *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\f
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\f
264# ifndef emacs
265/* How many characters in the character set. */
266# define CHAR_SET_SIZE 256
267
268# ifdef SYNTAX_TABLE
269
270extern char *re_syntax_table;
271
272# else /* not SYNTAX_TABLE */
273
274static char re_syntax_table[CHAR_SET_SIZE];
275
276static void init_syntax_once PARAMS ((void));
277
278static void
279init_syntax_once ()
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\f
303/* Integer type for pointers. */
dc579051 304# if !defined _LIBC && !defined HAVE_UINTPTR_T
2a6ef469
DD		 305typedef 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
406typedef char boolean;
407# define false 0
408# define true 1
409
410static reg_errcode_t byte_regex_compile _RE_ARGS ((const char *pattern, size_t size,
411 reg_syntax_t syntax,
412 struct re_pattern_buffer *bufp));
2a6ef469
DD		 413
414static int byte_re_match_2_internal PARAMS ((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));
86710ce2
DD		 420static int byte_re_search_2 PARAMS ((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));
425static int byte_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
426
427#ifdef MBS_SUPPORT
428static reg_errcode_t wcs_regex_compile _RE_ARGS ((const char *pattern, size_t size,
429 reg_syntax_t syntax,
430 struct re_pattern_buffer *bufp));
431
432
2a6ef469
DD		 433static int wcs_re_match_2_internal PARAMS ((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));
2a6ef469
DD		 442static int wcs_re_search_2 PARAMS ((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));
2a6ef469 447static int wcs_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
86710ce2 448#endif
2a6ef469
DD		 449\f
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
455typedef 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\f
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
941d74a0 615# if defined (__STDC__) || defined (ALMOST_STDC) || defined (HAVE_STRINGIZE)
dc579051
DD		 616# define PREFIX(name) byte_##name
617# else
618# define PREFIX(name) byte_/**/name
619# endif
2a6ef469
DD		 620# define ARG_PREFIX(name) name
621# define PUT_CHAR(c) putchar (c)
2a6ef469 622#else
86710ce2
DD		 623# ifdef WCHAR
624# define CHAR_T wchar_t
625# define UCHAR_T wchar_t
626# define COMPILED_BUFFER_VAR wc_buffer
627# define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
628# define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
941d74a0 629# if defined (__STDC__) || defined (ALMOST_STDC) || defined (HAVE_STRINGIZE)
dc579051
DD		 630# define PREFIX(name) wcs_##name
631# define ARG_PREFIX(name) c##name
632# else
633# define PREFIX(name) wcs_/**/name
634# define ARG_PREFIX(name) c/**/name
635# endif
86710ce2
DD		 636/* Should we use wide stream?? */
637# define PUT_CHAR(c) printf ("%C", c);
638# define TRUE 1
639# define FALSE 0
640# else
641# ifdef MBS_SUPPORT
642# define WCHAR
643# define INSIDE_RECURSION
644# include "regex.c"
645# undef INSIDE_RECURSION
646# endif
647# define BYTE
2a6ef469
DD		 648# define INSIDE_RECURSION
649# include "regex.c"
650# undef INSIDE_RECURSION
651# endif
2a6ef469
DD		 652#endif
653
654#ifdef INSIDE_RECURSION
655/* Common operations on the compiled pattern. */
656
657/* Store NUMBER in two contiguous bytes starting at DESTINATION. */
658/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
659
660# ifdef WCHAR
661# define STORE_NUMBER(destination, number) \
662 do { \
663 *(destination) = (UCHAR_T)(number); \
664 } while (0)
665# else /* BYTE */
666# define STORE_NUMBER(destination, number) \
667 do { \
668 (destination)[0] = (number) & 0377; \
669 (destination)[1] = (number) >> 8; \
670 } while (0)
671# endif /* WCHAR */
672
673/* Same as STORE_NUMBER, except increment DESTINATION to
674 the byte after where the number is stored. Therefore, DESTINATION
675 must be an lvalue. */
676/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
677
678# define STORE_NUMBER_AND_INCR(destination, number) \
679 do { \
680 STORE_NUMBER (destination, number); \
681 (destination) += OFFSET_ADDRESS_SIZE; \
682 } while (0)
683
684/* Put into DESTINATION a number stored in two contiguous bytes starting
685 at SOURCE. */
686/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
687
688# ifdef WCHAR
689# define EXTRACT_NUMBER(destination, source) \
690 do { \
691 (destination) = *(source); \
692 } while (0)
693# else /* BYTE */
694# define EXTRACT_NUMBER(destination, source) \
695 do { \
696 (destination) = *(source) & 0377; \
697 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
698 } while (0)
699# endif
700
701# ifdef DEBUG
702static void PREFIX(extract_number) _RE_ARGS ((int *dest, UCHAR_T *source));
703static void
704PREFIX(extract_number) (dest, source)
705 int *dest;
706 UCHAR_T *source;
707{
708# ifdef WCHAR
709 *dest = *source;
710# else /* BYTE */
711 int temp = SIGN_EXTEND_CHAR (*(source + 1));
712 *dest = *source & 0377;
713 *dest += temp << 8;
714# endif
715}
716
717# ifndef EXTRACT_MACROS /* To debug the macros. */
718# undef EXTRACT_NUMBER
719# define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
720# endif /* not EXTRACT_MACROS */
721
722# endif /* DEBUG */
723
724/* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
725 SOURCE must be an lvalue. */
726
727# define EXTRACT_NUMBER_AND_INCR(destination, source) \
728 do { \
729 EXTRACT_NUMBER (destination, source); \
730 (source) += OFFSET_ADDRESS_SIZE; \
731 } while (0)
732
733# ifdef DEBUG
734static void PREFIX(extract_number_and_incr) _RE_ARGS ((int *destination,
735 UCHAR_T **source));
736static void
737PREFIX(extract_number_and_incr) (destination, source)
738 int *destination;
739 UCHAR_T **source;
740{
741 PREFIX(extract_number) (destination, *source);
742 *source += OFFSET_ADDRESS_SIZE;
743}
744
745# ifndef EXTRACT_MACROS
746# undef EXTRACT_NUMBER_AND_INCR
747# define EXTRACT_NUMBER_AND_INCR(dest, src) \
748 PREFIX(extract_number_and_incr) (&dest, &src)
749# endif /* not EXTRACT_MACROS */
750
751# endif /* DEBUG */
752
753\f
754
755/* If DEBUG is defined, Regex prints many voluminous messages about what
756 it is doing (if the variable `debug' is nonzero). If linked with the
757 main program in `iregex.c', you can enter patterns and strings
758 interactively. And if linked with the main program in `main.c' and
759 the other test files, you can run the already-written tests. */
760
761# ifdef DEBUG
762
763# ifndef DEFINED_ONCE
764
765/* We use standard I/O for debugging. */
766# include <stdio.h>
767
768/* It is useful to test things that ``must'' be true when debugging. */
769# include <assert.h>
770
771static int debug;
772
773# define DEBUG_STATEMENT(e) e
774# define DEBUG_PRINT1(x) if (debug) printf (x)
775# define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
776# define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
777# define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
778# endif /* not DEFINED_ONCE */
779
780# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
781 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
782# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
783 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
784
785
786/* Print the fastmap in human-readable form. */
787
788# ifndef DEFINED_ONCE
789void
790print_fastmap (fastmap)
791 char *fastmap;
792{
793 unsigned was_a_range = 0;
794 unsigned i = 0;
795
796 while (i < (1 << BYTEWIDTH))
797 {
798 if (fastmap[i++])
799 {
800 was_a_range = 0;
801 putchar (i - 1);
802 while (i < (1 << BYTEWIDTH) && fastmap[i])
803 {
804 was_a_range = 1;
805 i++;
806 }
807 if (was_a_range)
808 {
809 printf ("-");
810 putchar (i - 1);
811 }
812 }
813 }
814 putchar ('\n');
815}
816# endif /* not DEFINED_ONCE */
817
818
819/* Print a compiled pattern string in human-readable form, starting at
820 the START pointer into it and ending just before the pointer END. */
821
822void
823PREFIX(print_partial_compiled_pattern) (start, end)
824 UCHAR_T *start;
825 UCHAR_T *end;
826{
827 int mcnt, mcnt2;
828 UCHAR_T *p1;
829 UCHAR_T *p = start;
830 UCHAR_T *pend = end;
831
832 if (start == NULL)
833 {
834 printf ("(null)\n");
835 return;
836 }
837
838 /* Loop over pattern commands. */
839 while (p < pend)
840 {
841# ifdef _LIBC
842 printf ("%td:\t", p - start);
843# else
844 printf ("%ld:\t", (long int) (p - start));
845# endif
846
847 switch ((re_opcode_t) *p++)
848 {
849 case no_op:
850 printf ("/no_op");
851 break;
852
853 case exactn:
854 mcnt = *p++;
855 printf ("/exactn/%d", mcnt);
856 do
857 {
858 putchar ('/');
859 PUT_CHAR (*p++);
860 }
861 while (--mcnt);
862 break;
863
864# ifdef MBS_SUPPORT
865 case exactn_bin:
866 mcnt = *p++;
867 printf ("/exactn_bin/%d", mcnt);
868 do
869 {
870 printf("/%lx", (long int) *p++);
871 }
872 while (--mcnt);
873 break;
874# endif /* MBS_SUPPORT */
875
876 case start_memory:
877 mcnt = *p++;
878 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
879 break;
880
881 case stop_memory:
882 mcnt = *p++;
883 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
884 break;
885
886 case duplicate:
887 printf ("/duplicate/%ld", (long int) *p++);
888 break;
889
890 case anychar:
891 printf ("/anychar");
892 break;
893
894 case charset:
895 case charset_not:
896 {
897# ifdef WCHAR
898 int i, length;
899 wchar_t *workp = p;
900 printf ("/charset [%s",
901 (re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
902 p += 5;
903 length = *workp++; /* the length of char_classes */
904 for (i=0 ; i<length ; i++)
905 printf("[:%lx:]", (long int) *p++);
906 length = *workp++; /* the length of collating_symbol */
907 for (i=0 ; i<length ;)
908 {
909 printf("[.");
910 while(*p != 0)
911 PUT_CHAR((i++,*p++));
912 i++,p++;
913 printf(".]");
914 }
915 length = *workp++; /* the length of equivalence_class */
916 for (i=0 ; i<length ;)
917 {
918 printf("[=");
919 while(*p != 0)
920 PUT_CHAR((i++,*p++));
921 i++,p++;
922 printf("=]");
923 }
924 length = *workp++; /* the length of char_range */
925 for (i=0 ; i<length ; i++)
926 {
927 wchar_t range_start = *p++;
928 wchar_t range_end = *p++;
929 printf("%C-%C", range_start, range_end);
930 }
931 length = *workp++; /* the length of char */
932 for (i=0 ; i<length ; i++)
933 printf("%C", *p++);
934 putchar (']');
935# else
936 register int c, last = -100;
937 register int in_range = 0;
938
939 printf ("/charset [%s",
940 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
941
942 assert (p + *p < pend);
943
944 for (c = 0; c < 256; c++)
945 if (c / 8 < *p
946 && (p[1 + (c/8)] & (1 << (c % 8))))
947 {
948 /* Are we starting a range? */
949 if (last + 1 == c && ! in_range)
950 {
951 putchar ('-');
952 in_range = 1;
953 }
954 /* Have we broken a range? */
955 else if (last + 1 != c && in_range)
956 {
957 putchar (last);
958 in_range = 0;
959 }
960
961 if (! in_range)
962 putchar (c);
963
964 last = c;
965 }
966
967 if (in_range)
968 putchar (last);
969
970 putchar (']');
971
972 p += 1 + *p;
973# endif /* WCHAR */
974 }
975 break;
976
977 case begline:
978 printf ("/begline");
979 break;
980
981 case endline:
982 printf ("/endline");
983 break;
984
985 case on_failure_jump:
986 PREFIX(extract_number_and_incr) (&mcnt, &p);
987# ifdef _LIBC
988 printf ("/on_failure_jump to %td", p + mcnt - start);
989# else
990 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
991# endif
992 break;
993
994 case on_failure_keep_string_jump:
995 PREFIX(extract_number_and_incr) (&mcnt, &p);
996# ifdef _LIBC
997 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
998# else
999 printf ("/on_failure_keep_string_jump to %ld",
1000 (long int) (p + mcnt - start));
1001# endif
1002 break;
1003
1004 case dummy_failure_jump:
1005 PREFIX(extract_number_and_incr) (&mcnt, &p);
1006# ifdef _LIBC
1007 printf ("/dummy_failure_jump to %td", p + mcnt - start);
1008# else
1009 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
1010# endif
1011 break;
1012
1013 case push_dummy_failure:
1014 printf ("/push_dummy_failure");
1015 break;
1016
1017 case maybe_pop_jump:
1018 PREFIX(extract_number_and_incr) (&mcnt, &p);
1019# ifdef _LIBC
1020 printf ("/maybe_pop_jump to %td", p + mcnt - start);
1021# else
1022 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
1023# endif
1024 break;
1025
1026 case pop_failure_jump:
1027 PREFIX(extract_number_and_incr) (&mcnt, &p);
1028# ifdef _LIBC
1029 printf ("/pop_failure_jump to %td", p + mcnt - start);
1030# else
1031 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
1032# endif
1033 break;
1034
1035 case jump_past_alt:
1036 PREFIX(extract_number_and_incr) (&mcnt, &p);
1037# ifdef _LIBC
1038 printf ("/jump_past_alt to %td", p + mcnt - start);
1039# else
1040 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
1041# endif
1042 break;
1043
1044 case jump:
1045 PREFIX(extract_number_and_incr) (&mcnt, &p);
1046# ifdef _LIBC
1047 printf ("/jump to %td", p + mcnt - start);
1048# else
1049 printf ("/jump to %ld", (long int) (p + mcnt - start));
1050# endif
1051 break;
1052
1053 case succeed_n:
1054 PREFIX(extract_number_and_incr) (&mcnt, &p);
1055 p1 = p + mcnt;
1056 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1057# ifdef _LIBC
1058 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
1059# else
1060 printf ("/succeed_n to %ld, %d times",
1061 (long int) (p1 - start), mcnt2);
1062# endif
1063 break;
1064
1065 case jump_n:
1066 PREFIX(extract_number_and_incr) (&mcnt, &p);
1067 p1 = p + mcnt;
1068 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1069 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
1070 break;
1071
1072 case set_number_at:
1073 PREFIX(extract_number_and_incr) (&mcnt, &p);
1074 p1 = p + mcnt;
1075 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1076# ifdef _LIBC
1077 printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1078# else
1079 printf ("/set_number_at location %ld to %d",
1080 (long int) (p1 - start), mcnt2);
1081# endif
1082 break;
1083
1084 case wordbound:
1085 printf ("/wordbound");
1086 break;
1087
1088 case notwordbound:
1089 printf ("/notwordbound");
1090 break;
1091
1092 case wordbeg:
1093 printf ("/wordbeg");
1094 break;
1095
1096 case wordend:
1097 printf ("/wordend");
1098 break;
1099
1100# ifdef emacs
1101 case before_dot:
1102 printf ("/before_dot");
1103 break;
1104
1105 case at_dot:
1106 printf ("/at_dot");
1107 break;
1108
1109 case after_dot:
1110 printf ("/after_dot");
1111 break;
1112
1113 case syntaxspec:
1114 printf ("/syntaxspec");
1115 mcnt = *p++;
1116 printf ("/%d", mcnt);
1117 break;
1118
1119 case notsyntaxspec:
1120 printf ("/notsyntaxspec");
1121 mcnt = *p++;
1122 printf ("/%d", mcnt);
1123 break;
1124# endif /* emacs */
1125
1126 case wordchar:
1127 printf ("/wordchar");
1128 break;
1129
1130 case notwordchar:
1131 printf ("/notwordchar");
1132 break;
1133
1134 case begbuf:
1135 printf ("/begbuf");
1136 break;
1137
1138 case endbuf:
1139 printf ("/endbuf");
1140 break;
1141
1142 default:
1143 printf ("?%ld", (long int) *(p-1));
1144 }
1145
1146 putchar ('\n');
1147 }
1148
1149# ifdef _LIBC
1150 printf ("%td:\tend of pattern.\n", p - start);
1151# else
1152 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1153# endif
1154}
1155
1156
1157void
1158PREFIX(print_compiled_pattern) (bufp)
1159 struct re_pattern_buffer *bufp;
1160{
1161 UCHAR_T *buffer = (UCHAR_T*) bufp->buffer;
1162
1163 PREFIX(print_partial_compiled_pattern) (buffer, buffer
1164 + bufp->used / sizeof(UCHAR_T));
1165 printf ("%ld bytes used/%ld bytes allocated.\n",
1166 bufp->used, bufp->allocated);
1167
1168 if (bufp->fastmap_accurate && bufp->fastmap)
1169 {
1170 printf ("fastmap: ");
1171 print_fastmap (bufp->fastmap);
1172 }
1173
1174# ifdef _LIBC
1175 printf ("re_nsub: %Zd\t", bufp->re_nsub);
1176# else
1177 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
1178# endif
1179 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1180 printf ("can_be_null: %d\t", bufp->can_be_null);
1181 printf ("newline_anchor: %d\n", bufp->newline_anchor);
1182 printf ("no_sub: %d\t", bufp->no_sub);
1183 printf ("not_bol: %d\t", bufp->not_bol);
1184 printf ("not_eol: %d\t", bufp->not_eol);
1185 printf ("syntax: %lx\n", bufp->syntax);
1186 /* Perhaps we should print the translate table? */
1187}
1188
1189
1190void
1191PREFIX(print_double_string) (where, string1, size1, string2, size2)
1192 const CHAR_T *where;
1193 const CHAR_T *string1;
1194 const CHAR_T *string2;
1195 int size1;
1196 int size2;
1197{
1198 int this_char;
1199
1200 if (where == NULL)
1201 printf ("(null)");
1202 else
1203 {
6ad8a379
DD		 1204 int cnt;
1205
2a6ef469
DD		 1206 if (FIRST_STRING_P (where))
1207 {
1208 for (this_char = where - string1; this_char < size1; this_char++)
1209 PUT_CHAR (string1[this_char]);
1210
1211 where = string2;
1212 }
1213
6ad8a379 1214 cnt = 0;
2a6ef469 1215 for (this_char = where - string2; this_char < size2; this_char++)
6ad8a379
DD		 1216 {
1217 PUT_CHAR (string2[this_char]);
1218 if (++cnt > 100)
1219 {
1220 fputs ("...", stdout);
1221 break;
1222 }
1223 }
2a6ef469
DD		 1224 }
1225}
1226
1227# ifndef DEFINED_ONCE
1228void
1229printchar (c)
1230 int c;
1231{
1232 putc (c, stderr);
1233}
1234# endif
1235
1236# else /* not DEBUG */
1237
1238# ifndef DEFINED_ONCE
1239# undef assert
1240# define assert(e)
1241
1242# define DEBUG_STATEMENT(e)
1243# define DEBUG_PRINT1(x)
1244# define DEBUG_PRINT2(x1, x2)
1245# define DEBUG_PRINT3(x1, x2, x3)
1246# define DEBUG_PRINT4(x1, x2, x3, x4)
1247# endif /* not DEFINED_ONCE */
1248# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1249# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1250
1251# endif /* not DEBUG */
1252
1253\f
1254
1255# ifdef WCHAR
1256/* This convert a multibyte string to a wide character string.
1257 And write their correspondances to offset_buffer(see below)
1258 and write whether each wchar_t is binary data to is_binary.
1259 This assume invalid multibyte sequences as binary data.
1260 We assume offset_buffer and is_binary is already allocated
1261 enough space. */
1262
1263static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src,
1264 size_t len, int *offset_buffer,
1265 char *is_binary);
1266static size_t
1267convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary)
1268 CHAR_T *dest;
1269 const unsigned char* src;
1270 size_t len; /* the length of multibyte string. */
1271
1272 /* It hold correspondances between src(char string) and
1273 dest(wchar_t string) for optimization.
1274 e.g. src = "xxxyzz"
1275 dest = {'X', 'Y', 'Z'}
1276 (each "xxx", "y" and "zz" represent one multibyte character
1277 corresponding to 'X', 'Y' and 'Z'.)
1278 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1279 = {0, 3, 4, 6}
1280 */
1281 int *offset_buffer;
1282 char *is_binary;
1283{
1284 wchar_t *pdest = dest;
1285 const unsigned char *psrc = src;
1286 size_t wc_count = 0;
1287
1288 mbstate_t mbs;
1289 int i, consumed;
1290 size_t mb_remain = len;
1291 size_t mb_count = 0;
1292
1293 /* Initialize the conversion state. */
1294 memset (&mbs, 0, sizeof (mbstate_t));
1295
1296 offset_buffer[0] = 0;
1297 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1298 psrc += consumed)
1299 {
dc676635
DD		 1300#ifdef _LIBC
1301 consumed = __mbrtowc (pdest, psrc, mb_remain, &mbs);
1302#else
2a6ef469 1303 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
dc676635 1304#endif
2a6ef469
DD		 1305
1306 if (consumed <= 0)
1307 /* failed to convert. maybe src contains binary data.
1308 So we consume 1 byte manualy. */
1309 {
1310 *pdest = *psrc;
1311 consumed = 1;
1312 is_binary[wc_count] = TRUE;
1313 }
1314 else
1315 is_binary[wc_count] = FALSE;
1316 /* In sjis encoding, we use yen sign as escape character in
1317 place of reverse solidus. So we convert 0x5c(yen sign in
1318 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1319 solidus in UCS2). */
1320 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1321 *pdest = (wchar_t) *psrc;
1322
1323 offset_buffer[wc_count + 1] = mb_count += consumed;
1324 }
1325
1326 /* Fill remain of the buffer with sentinel. */
1327 for (i = wc_count + 1 ; i <= len ; i++)
1328 offset_buffer[i] = mb_count + 1;
1329
1330 return wc_count;
1331}
1332
1333# endif /* WCHAR */
1334
1335#else /* not INSIDE_RECURSION */
1336
1337/* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1338 also be assigned to arbitrarily: each pattern buffer stores its own
1339 syntax, so it can be changed between regex compilations. */
1340/* This has no initializer because initialized variables in Emacs
1341 become read-only after dumping. */
1342reg_syntax_t re_syntax_options;
1343
1344
1345/* Specify the precise syntax of regexps for compilation. This provides
1346 for compatibility for various utilities which historically have
1347 different, incompatible syntaxes.
1348
1349 The argument SYNTAX is a bit mask comprised of the various bits
1350 defined in regex.h. We return the old syntax. */
1351
1352reg_syntax_t
1353re_set_syntax (syntax)
1354 reg_syntax_t syntax;
1355{
1356 reg_syntax_t ret = re_syntax_options;
1357
1358 re_syntax_options = syntax;
1359# ifdef DEBUG
1360 if (syntax & RE_DEBUG)
1361 debug = 1;
1362 else if (debug) /* was on but now is not */
1363 debug = 0;
1364# endif /* DEBUG */
1365 return ret;
1366}
1367# ifdef _LIBC
1368weak_alias (__re_set_syntax, re_set_syntax)
1369# endif
1370\f
1371/* This table gives an error message for each of the error codes listed
1372 in regex.h. Obviously the order here has to be same as there.
1373 POSIX doesn't require that we do anything for REG_NOERROR,
1374 but why not be nice? */
1375
a2832523 1376static const char *re_error_msgid[] =
2a6ef469 1377 {
a2832523
DD		 1378 gettext_noop ("Success"), /* REG_NOERROR */
1379 gettext_noop ("No match"), /* REG_NOMATCH */
1380 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1381 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1382 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1383 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1384 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1385 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1386 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1387 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1388 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1389 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1390 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1391 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1392 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1393 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
2a6ef469
DD		 1394 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1395 };
2a6ef469
DD		 1396\f
1397#endif /* INSIDE_RECURSION */
1398
1399#ifndef DEFINED_ONCE
1400/* Avoiding alloca during matching, to placate r_alloc. */
1401
1402/* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1403 searching and matching functions should not call alloca. On some
1404 systems, alloca is implemented in terms of malloc, and if we're
1405 using the relocating allocator routines, then malloc could cause a
1406 relocation, which might (if the strings being searched are in the
1407 ralloc heap) shift the data out from underneath the regexp
1408 routines.
1409
1410 Here's another reason to avoid allocation: Emacs
1411 processes input from X in a signal handler; processing X input may
1412 call malloc; if input arrives while a matching routine is calling
1413 malloc, then we're scrod. But Emacs can't just block input while
1414 calling matching routines; then we don't notice interrupts when
1415 they come in. So, Emacs blocks input around all regexp calls
1416 except the matching calls, which it leaves unprotected, in the
1417 faith that they will not malloc. */
1418
1419/* Normally, this is fine. */
1420# define MATCH_MAY_ALLOCATE
1421
1422/* When using GNU C, we are not REALLY using the C alloca, no matter
1423 what config.h may say. So don't take precautions for it. */
1424# ifdef __GNUC__
1425# undef C_ALLOCA
1426# endif
1427
1428/* The match routines may not allocate if (1) they would do it with malloc
1429 and (2) it's not safe for them to use malloc.
1430 Note that if REL_ALLOC is defined, matching would not use malloc for the
1431 failure stack, but we would still use it for the register vectors;
1432 so REL_ALLOC should not affect this. */
1433# if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1434# undef MATCH_MAY_ALLOCATE
1435# endif
1436#endif /* not DEFINED_ONCE */
1437\f
1438#ifdef INSIDE_RECURSION
1439/* Failure stack declarations and macros; both re_compile_fastmap and
1440 re_match_2 use a failure stack. These have to be macros because of
1441 REGEX_ALLOCATE_STACK. */
1442
1443
1444/* Number of failure points for which to initially allocate space
1445 when matching. If this number is exceeded, we allocate more
1446 space, so it is not a hard limit. */
1447# ifndef INIT_FAILURE_ALLOC
1448# define INIT_FAILURE_ALLOC 5
1449# endif
1450
1451/* Roughly the maximum number of failure points on the stack. Would be
1452 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1453 This is a variable only so users of regex can assign to it; we never
1454 change it ourselves. */
1455
1456# ifdef INT_IS_16BIT
1457
1458# ifndef DEFINED_ONCE
1459# if defined MATCH_MAY_ALLOCATE
1460/* 4400 was enough to cause a crash on Alpha OSF/1,
1461 whose default stack limit is 2mb. */
1462long int re_max_failures = 4000;
1463# else
1464long int re_max_failures = 2000;
1465# endif
1466# endif
1467
1468union PREFIX(fail_stack_elt)
1469{
1470 UCHAR_T *pointer;
1471 long int integer;
1472};
1473
1474typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1475
1476typedef struct
1477{
1478 PREFIX(fail_stack_elt_t) *stack;
1479 unsigned long int size;
1480 unsigned long int avail; /* Offset of next open position. */
1481} PREFIX(fail_stack_type);
1482
1483# else /* not INT_IS_16BIT */
1484
1485# ifndef DEFINED_ONCE
1486# if defined MATCH_MAY_ALLOCATE
1487/* 4400 was enough to cause a crash on Alpha OSF/1,
1488 whose default stack limit is 2mb. */
1489int re_max_failures = 4000;
1490# else
1491int re_max_failures = 2000;
1492# endif
1493# endif
1494
1495union PREFIX(fail_stack_elt)
1496{
1497 UCHAR_T *pointer;
1498 int integer;
1499};
1500
1501typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1502
1503typedef struct
1504{
1505 PREFIX(fail_stack_elt_t) *stack;
1506 unsigned size;
1507 unsigned avail; /* Offset of next open position. */
1508} PREFIX(fail_stack_type);
1509
1510# endif /* INT_IS_16BIT */
1511
1512# ifndef DEFINED_ONCE
1513# define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1514# define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1515# define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1516# endif
1517
1518
1519/* Define macros to initialize and free the failure stack.
1520 Do `return -2' if the alloc fails. */
1521
1522# ifdef MATCH_MAY_ALLOCATE
1523# define INIT_FAIL_STACK() \
1524 do { \
1525 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1526 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1527 \
1528 if (fail_stack.stack == NULL) \
1529 return -2; \
1530 \
1531 fail_stack.size = INIT_FAILURE_ALLOC; \
1532 fail_stack.avail = 0; \
1533 } while (0)
1534
1535# define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1536# else
1537# define INIT_FAIL_STACK() \
1538 do { \
1539 fail_stack.avail = 0; \
1540 } while (0)
1541
1542# define RESET_FAIL_STACK()
1543# endif
1544
1545
1546/* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1547
1548 Return 1 if succeeds, and 0 if either ran out of memory
1549 allocating space for it or it was already too large.
1550
1551 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1552
1553# define DOUBLE_FAIL_STACK(fail_stack) \
1554 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1555 ? 0 \
1556 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1557 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1558 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1559 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1560 \
1561 (fail_stack).stack == NULL \
1562 ? 0 \
1563 : ((fail_stack).size <<= 1, \
1564 1)))
1565
1566
1567/* Push pointer POINTER on FAIL_STACK.
1568 Return 1 if was able to do so and 0 if ran out of memory allocating
1569 space to do so. */
1570# define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1571 ((FAIL_STACK_FULL () \
1572 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1573 ? 0 \
1574 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1575 1))
1576
1577/* Push a pointer value onto the failure stack.
1578 Assumes the variable `fail_stack'. Probably should only
1579 be called from within `PUSH_FAILURE_POINT'. */
1580# define PUSH_FAILURE_POINTER(item) \
1581 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1582
1583/* This pushes an integer-valued item onto the failure stack.
1584 Assumes the variable `fail_stack'. Probably should only
1585 be called from within `PUSH_FAILURE_POINT'. */
1586# define PUSH_FAILURE_INT(item) \
1587 fail_stack.stack[fail_stack.avail++].integer = (item)
1588
1589/* Push a fail_stack_elt_t value onto the failure stack.
1590 Assumes the variable `fail_stack'. Probably should only
1591 be called from within `PUSH_FAILURE_POINT'. */
1592# define PUSH_FAILURE_ELT(item) \
1593 fail_stack.stack[fail_stack.avail++] = (item)
1594
1595/* These three POP... operations complement the three PUSH... operations.
1596 All assume that `fail_stack' is nonempty. */
1597# define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1598# define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1599# define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1600
1601/* Used to omit pushing failure point id's when we're not debugging. */
1602# ifdef DEBUG
1603# define DEBUG_PUSH PUSH_FAILURE_INT
1604# define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1605# else
1606# define DEBUG_PUSH(item)
1607# define DEBUG_POP(item_addr)
1608# endif
1609
1610
1611/* Push the information about the state we will need
1612 if we ever fail back to it.
1613
1614 Requires variables fail_stack, regstart, regend, reg_info, and
1615 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1616 be declared.
1617
1618 Does `return FAILURE_CODE' if runs out of memory. */
1619
1620# define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1621 do { \
1622 char *destination; \
1623 /* Must be int, so when we don't save any registers, the arithmetic \
1624 of 0 + -1 isn't done as unsigned. */ \
1625 /* Can't be int, since there is not a shred of a guarantee that int \
1626 is wide enough to hold a value of something to which pointer can \
1627 be assigned */ \
1628 active_reg_t this_reg; \
1629 \
1630 DEBUG_STATEMENT (failure_id++); \
1631 DEBUG_STATEMENT (nfailure_points_pushed++); \
1632 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1633 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1634 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1635 \
1636 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1637 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1638 \
1639 /* Ensure we have enough space allocated for what we will push. */ \
1640 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1641 { \
1642 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1643 return failure_code; \
1644 \
1645 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1646 (fail_stack).size); \
1647 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1648 } \
1649 \
1650 /* Push the info, starting with the registers. */ \
1651 DEBUG_PRINT1 ("\n"); \
1652 \
1653 if (1) \
1654 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1655 this_reg++) \
1656 { \
1657 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1658 DEBUG_STATEMENT (num_regs_pushed++); \
1659 \
1660 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1661 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1662 \
1663 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1664 PUSH_FAILURE_POINTER (regend[this_reg]); \
1665 \
1666 DEBUG_PRINT2 (" info: %p\n ", \
1667 reg_info[this_reg].word.pointer); \
1668 DEBUG_PRINT2 (" match_null=%d", \
1669 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1670 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1671 DEBUG_PRINT2 (" matched_something=%d", \
1672 MATCHED_SOMETHING (reg_info[this_reg])); \
1673 DEBUG_PRINT2 (" ever_matched=%d", \
1674 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1675 DEBUG_PRINT1 ("\n"); \
1676 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1677 } \
1678 \
1679 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1680 PUSH_FAILURE_INT (lowest_active_reg); \
1681 \
1682 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1683 PUSH_FAILURE_INT (highest_active_reg); \
1684 \
1685 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1686 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1687 PUSH_FAILURE_POINTER (pattern_place); \
1688 \
1689 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1690 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1691 size2); \
1692 DEBUG_PRINT1 ("'\n"); \
1693 PUSH_FAILURE_POINTER (string_place); \
1694 \
1695 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1696 DEBUG_PUSH (failure_id); \
1697 } while (0)
1698
1699# ifndef DEFINED_ONCE
1700/* This is the number of items that are pushed and popped on the stack
1701 for each register. */
1702# define NUM_REG_ITEMS 3
1703
1704/* Individual items aside from the registers. */
1705# ifdef DEBUG
1706# define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1707# else
1708# define NUM_NONREG_ITEMS 4
1709# endif
1710
1711/* We push at most this many items on the stack. */
1712/* We used to use (num_regs - 1), which is the number of registers
1713 this regexp will save; but that was changed to 5
1714 to avoid stack overflow for a regexp with lots of parens. */
1715# define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1716
1717/* We actually push this many items. */
1718# define NUM_FAILURE_ITEMS \
1719 (((0 \
1720 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1721 * NUM_REG_ITEMS) \
1722 + NUM_NONREG_ITEMS)
1723
1724/* How many items can still be added to the stack without overflowing it. */
1725# define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1726# endif /* not DEFINED_ONCE */
1727
1728
1729/* Pops what PUSH_FAIL_STACK pushes.
1730
1731 We restore into the parameters, all of which should be lvalues:
1732 STR -- the saved data position.
1733 PAT -- the saved pattern position.
1734 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1735 REGSTART, REGEND -- arrays of string positions.
1736 REG_INFO -- array of information about each subexpression.
1737
1738 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1739 `pend', `string1', `size1', `string2', and `size2'. */
1740# define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1741{ \
1742 DEBUG_STATEMENT (unsigned failure_id;) \
1743 active_reg_t this_reg; \
1744 const UCHAR_T *string_temp; \
1745 \
1746 assert (!FAIL_STACK_EMPTY ()); \
1747 \
1748 /* Remove failure points and point to how many regs pushed. */ \
1749 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1750 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1751 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1752 \
1753 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1754 \
1755 DEBUG_POP (&failure_id); \
1756 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1757 \
1758 /* If the saved string location is NULL, it came from an \
1759 on_failure_keep_string_jump opcode, and we want to throw away the \
1760 saved NULL, thus retaining our current position in the string. */ \
1761 string_temp = POP_FAILURE_POINTER (); \
1762 if (string_temp != NULL) \
1763 str = (const CHAR_T *) string_temp; \
1764 \
1765 DEBUG_PRINT2 (" Popping string %p: `", str); \
1766 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1767 DEBUG_PRINT1 ("'\n"); \
1768 \
1769 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1770 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1771 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1772 \
1773 /* Restore register info. */ \
1774 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1775 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1776 \
1777 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1778 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1779 \
1780 if (1) \
1781 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1782 { \
1783 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1784 \
1785 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1786 DEBUG_PRINT2 (" info: %p\n", \
1787 reg_info[this_reg].word.pointer); \
1788 \
1789 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1790 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1791 \
1792 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1793 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1794 } \
1795 else \
1796 { \
1797 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1798 { \
1799 reg_info[this_reg].word.integer = 0; \
1800 regend[this_reg] = 0; \
1801 regstart[this_reg] = 0; \
1802 } \
1803 highest_active_reg = high_reg; \
1804 } \
1805 \
1806 set_regs_matched_done = 0; \
1807 DEBUG_STATEMENT (nfailure_points_popped++); \
1808} /* POP_FAILURE_POINT */
1809\f
1810/* Structure for per-register (a.k.a. per-group) information.
1811 Other register information, such as the
1812 starting and ending positions (which are addresses), and the list of
1813 inner groups (which is a bits list) are maintained in separate
1814 variables.
1815
1816 We are making a (strictly speaking) nonportable assumption here: that
1817 the compiler will pack our bit fields into something that fits into
1818 the type of `word', i.e., is something that fits into one item on the
1819 failure stack. */
1820
1821
1822/* Declarations and macros for re_match_2. */
1823
1824typedef union
1825{
1826 PREFIX(fail_stack_elt_t) word;
1827 struct
1828 {
1829 /* This field is one if this group can match the empty string,
1830 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1831# define MATCH_NULL_UNSET_VALUE 3
1832 unsigned match_null_string_p : 2;
1833 unsigned is_active : 1;
1834 unsigned matched_something : 1;
1835 unsigned ever_matched_something : 1;
1836 } bits;
1837} PREFIX(register_info_type);
1838
1839# ifndef DEFINED_ONCE
1840# define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1841# define IS_ACTIVE(R) ((R).bits.is_active)
1842# define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1843# define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1844
1845
1846/* Call this when have matched a real character; it sets `matched' flags
1847 for the subexpressions which we are currently inside. Also records
1848 that those subexprs have matched. */
1849# define SET_REGS_MATCHED() \
1850 do \
1851 { \
1852 if (!set_regs_matched_done) \
1853 { \
1854 active_reg_t r; \
1855 set_regs_matched_done = 1; \
1856 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1857 { \
1858 MATCHED_SOMETHING (reg_info[r]) \
1859 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1860 = 1; \
1861 } \
1862 } \
1863 } \
1864 while (0)
1865# endif /* not DEFINED_ONCE */
1866
1867/* Registers are set to a sentinel when they haven't yet matched. */
1868static CHAR_T PREFIX(reg_unset_dummy);
1869# define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1870# define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1871
1872/* Subroutine declarations and macros for regex_compile. */
1873static void PREFIX(store_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, int arg));
1874static void PREFIX(store_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1875 int arg1, int arg2));
1876static void PREFIX(insert_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1877 int arg, UCHAR_T *end));
1878static void PREFIX(insert_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1879 int arg1, int arg2, UCHAR_T *end));
1880static boolean PREFIX(at_begline_loc_p) _RE_ARGS ((const CHAR_T *pattern,
1881 const CHAR_T *p,
1882 reg_syntax_t syntax));
1883static boolean PREFIX(at_endline_loc_p) _RE_ARGS ((const CHAR_T *p,
1884 const CHAR_T *pend,
1885 reg_syntax_t syntax));
1886# ifdef WCHAR
1887static reg_errcode_t wcs_compile_range _RE_ARGS ((CHAR_T range_start,
1888 const CHAR_T **p_ptr,
1889 const CHAR_T *pend,
1890 char *translate,
1891 reg_syntax_t syntax,
1892 UCHAR_T *b,
1893 CHAR_T *char_set));
1894static void insert_space _RE_ARGS ((int num, CHAR_T *loc, CHAR_T *end));
1895# else /* BYTE */
1896static reg_errcode_t byte_compile_range _RE_ARGS ((unsigned int range_start,
1897 const char **p_ptr,
1898 const char *pend,
1899 char *translate,
1900 reg_syntax_t syntax,
1901 unsigned char *b));
1902# endif /* WCHAR */
1903
1904/* Fetch the next character in the uncompiled pattern---translating it
1905 if necessary. Also cast from a signed character in the constant
1906 string passed to us by the user to an unsigned char that we can use
1907 as an array index (in, e.g., `translate'). */
1908/* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1909 because it is impossible to allocate 4GB array for some encodings
1910 which have 4 byte character_set like UCS4. */
1911# ifndef PATFETCH
1912# ifdef WCHAR
1913# define PATFETCH(c) \
1914 do {if (p == pend) return REG_EEND; \
1915 c = (UCHAR_T) *p++; \
1916 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1917 } while (0)
1918# else /* BYTE */
1919# define PATFETCH(c) \
1920 do {if (p == pend) return REG_EEND; \
1921 c = (unsigned char) *p++; \
1922 if (translate) c = (unsigned char) translate[c]; \
1923 } while (0)
1924# endif /* WCHAR */
1925# endif
1926
1927/* Fetch the next character in the uncompiled pattern, with no
1928 translation. */
1929# define PATFETCH_RAW(c) \
1930 do {if (p == pend) return REG_EEND; \
1931 c = (UCHAR_T) *p++; \
1932 } while (0)
1933
1934/* Go backwards one character in the pattern. */
1935# define PATUNFETCH p--
1936
1937
1938/* If `translate' is non-null, return translate[D], else just D. We
1939 cast the subscript to translate because some data is declared as
1940 `char *', to avoid warnings when a string constant is passed. But
1941 when we use a character as a subscript we must make it unsigned. */
1942/* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1943 because it is impossible to allocate 4GB array for some encodings
1944 which have 4 byte character_set like UCS4. */
1945
1946# ifndef TRANSLATE
1947# ifdef WCHAR
1948# define TRANSLATE(d) \
1949 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1950 ? (char) translate[(unsigned char) (d)] : (d))
1951# else /* BYTE */
1952# define TRANSLATE(d) \
1953 (translate ? (char) translate[(unsigned char) (d)] : (d))
1954# endif /* WCHAR */
1955# endif
1956
1957
1958/* Macros for outputting the compiled pattern into `buffer'. */
1959
1960/* If the buffer isn't allocated when it comes in, use this. */
1961# define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
1962
1963/* Make sure we have at least N more bytes of space in buffer. */
1964# ifdef WCHAR
1965# define GET_BUFFER_SPACE(n) \
1966 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
1967 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
1968 EXTEND_BUFFER ()
1969# else /* BYTE */
1970# define GET_BUFFER_SPACE(n) \
1971 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1972 EXTEND_BUFFER ()
1973# endif /* WCHAR */
1974
1975/* Make sure we have one more byte of buffer space and then add C to it. */
1976# define BUF_PUSH(c) \
1977 do { \
1978 GET_BUFFER_SPACE (1); \
1979 *b++ = (UCHAR_T) (c); \
1980 } while (0)
1981
1982
1983/* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1984# define BUF_PUSH_2(c1, c2) \
1985 do { \
1986 GET_BUFFER_SPACE (2); \
1987 *b++ = (UCHAR_T) (c1); \
1988 *b++ = (UCHAR_T) (c2); \
1989 } while (0)
1990
1991
1992/* As with BUF_PUSH_2, except for three bytes. */
1993# define BUF_PUSH_3(c1, c2, c3) \
1994 do { \
1995 GET_BUFFER_SPACE (3); \
1996 *b++ = (UCHAR_T) (c1); \
1997 *b++ = (UCHAR_T) (c2); \
1998 *b++ = (UCHAR_T) (c3); \
1999 } while (0)
2000
2001/* Store a jump with opcode OP at LOC to location TO. We store a
2002 relative address offset by the three bytes the jump itself occupies. */
2003# define STORE_JUMP(op, loc, to) \
2004 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2005
2006/* Likewise, for a two-argument jump. */
2007# define STORE_JUMP2(op, loc, to, arg) \
2008 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2009
2010/* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2011# define INSERT_JUMP(op, loc, to) \
2012 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2013
2014/* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2015# define INSERT_JUMP2(op, loc, to, arg) \
2016 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2017 arg, b)
2018
2019/* This is not an arbitrary limit: the arguments which represent offsets
2020 into the pattern are two bytes long. So if 2^16 bytes turns out to
2021 be too small, many things would have to change. */
2022/* Any other compiler which, like MSC, has allocation limit below 2^16
2023 bytes will have to use approach similar to what was done below for
2024 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2025 reallocating to 0 bytes. Such thing is not going to work too well.
2026 You have been warned!! */
2027# ifndef DEFINED_ONCE
2028# if defined _MSC_VER && !defined WIN32
2029/* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2030 The REALLOC define eliminates a flurry of conversion warnings,
2031 but is not required. */
2032# define MAX_BUF_SIZE 65500L
2033# define REALLOC(p,s) realloc ((p), (size_t) (s))
2034# else
2035# define MAX_BUF_SIZE (1L << 16)
2036# define REALLOC(p,s) realloc ((p), (s))
2037# endif
2038
2039/* Extend the buffer by twice its current size via realloc and
2040 reset the pointers that pointed into the old block to point to the
2041 correct places in the new one. If extending the buffer results in it
2042 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2043# if __BOUNDED_POINTERS__
2044# define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2045# define MOVE_BUFFER_POINTER(P) \
2046 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2047# define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2048 else \
2049 { \
2050 SET_HIGH_BOUND (b); \
2051 SET_HIGH_BOUND (begalt); \
2052 if (fixup_alt_jump) \
2053 SET_HIGH_BOUND (fixup_alt_jump); \
2054 if (laststart) \
2055 SET_HIGH_BOUND (laststart); \
2056 if (pending_exact) \
2057 SET_HIGH_BOUND (pending_exact); \
2058 }
2059# else
2060# define MOVE_BUFFER_POINTER(P) (P) += incr
2061# define ELSE_EXTEND_BUFFER_HIGH_BOUND
2062# endif
2063# endif /* not DEFINED_ONCE */
2064
2065# ifdef WCHAR
2066# define EXTEND_BUFFER() \
2067 do { \
2068 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2069 int wchar_count; \
2070 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2071 return REG_ESIZE; \
2072 bufp->allocated <<= 1; \
2073 if (bufp->allocated > MAX_BUF_SIZE) \
2074 bufp->allocated = MAX_BUF_SIZE; \
2075 /* How many characters the new buffer can have? */ \
2076 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2077 if (wchar_count == 0) wchar_count = 1; \
2078 /* Truncate the buffer to CHAR_T align. */ \
2079 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2080 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2081 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
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 int 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# else /* BYTE */
2100# define EXTEND_BUFFER() \
2101 do { \
2102 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2103 if (bufp->allocated == MAX_BUF_SIZE) \
2104 return REG_ESIZE; \
2105 bufp->allocated <<= 1; \
2106 if (bufp->allocated > MAX_BUF_SIZE) \
2107 bufp->allocated = MAX_BUF_SIZE; \
2108 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2109 bufp->allocated); \
2110 if (COMPILED_BUFFER_VAR == NULL) \
2111 return REG_ESPACE; \
2112 /* If the buffer moved, move all the pointers into it. */ \
2113 if (old_buffer != COMPILED_BUFFER_VAR) \
2114 { \
2115 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2116 MOVE_BUFFER_POINTER (b); \
2117 MOVE_BUFFER_POINTER (begalt); \
2118 if (fixup_alt_jump) \
2119 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2120 if (laststart) \
2121 MOVE_BUFFER_POINTER (laststart); \
2122 if (pending_exact) \
2123 MOVE_BUFFER_POINTER (pending_exact); \
2124 } \
2125 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2126 } while (0)
2127# endif /* WCHAR */
2128
2129# ifndef DEFINED_ONCE
2130/* Since we have one byte reserved for the register number argument to
2131 {start,stop}_memory, the maximum number of groups we can report
2132 things about is what fits in that byte. */
2133# define MAX_REGNUM 255
2134
2135/* But patterns can have more than `MAX_REGNUM' registers. We just
2136 ignore the excess. */
2137typedef unsigned regnum_t;
2138
2139
2140/* Macros for the compile stack. */
2141
2142/* Since offsets can go either forwards or backwards, this type needs to
2143 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2144/* int may be not enough when sizeof(int) == 2. */
2145typedef long pattern_offset_t;
2146
2147typedef struct
2148{
2149 pattern_offset_t begalt_offset;
2150 pattern_offset_t fixup_alt_jump;
2151 pattern_offset_t inner_group_offset;
2152 pattern_offset_t laststart_offset;
2153 regnum_t regnum;
2154} compile_stack_elt_t;
2155
2156
2157typedef struct
2158{
2159 compile_stack_elt_t *stack;
2160 unsigned size;
2161 unsigned avail; /* Offset of next open position. */
2162} compile_stack_type;
2163
2164
2165# define INIT_COMPILE_STACK_SIZE 32
2166
2167# define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2168# define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2169
2170/* The next available element. */
2171# define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2172
2173# endif /* not DEFINED_ONCE */
2174
2175/* Set the bit for character C in a list. */
2176# ifndef DEFINED_ONCE
2177# define SET_LIST_BIT(c) \
2178 (b[((unsigned char) (c)) / BYTEWIDTH] \
2179 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2180# endif /* DEFINED_ONCE */
2181
2182/* Get the next unsigned number in the uncompiled pattern. */
2183# define GET_UNSIGNED_NUMBER(num) \
2184 { \
2185 while (p != pend) \
2186 { \
2187 PATFETCH (c); \
2188 if (c < '0' || c > '9') \
2189 break; \
2190 if (num <= RE_DUP_MAX) \
2191 { \
2192 if (num < 0) \
2193 num = 0; \
2194 num = num * 10 + c - '0'; \
2195 } \
2196 } \
2197 }
2198
2199# ifndef DEFINED_ONCE
2200# if defined _LIBC || WIDE_CHAR_SUPPORT
2201/* The GNU C library provides support for user-defined character classes
2202 and the functions from ISO C amendement 1. */
2203# ifdef CHARCLASS_NAME_MAX
2204# define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2205# else
2206/* This shouldn't happen but some implementation might still have this
2207 problem. Use a reasonable default value. */
2208# define CHAR_CLASS_MAX_LENGTH 256
2209# endif
2210
2211# ifdef _LIBC
2212# define IS_CHAR_CLASS(string) __wctype (string)
2213# else
2214# define IS_CHAR_CLASS(string) wctype (string)
2215# endif
2216# else
2217# define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2218
2219# define IS_CHAR_CLASS(string) \
2220 (STREQ (string, "alpha") || STREQ (string, "upper") \
2221 || STREQ (string, "lower") || STREQ (string, "digit") \
2222 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2223 || STREQ (string, "space") || STREQ (string, "print") \
2224 || STREQ (string, "punct") || STREQ (string, "graph") \
2225 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2226# endif
2227# endif /* DEFINED_ONCE */
2228\f
2229# ifndef MATCH_MAY_ALLOCATE
2230
2231/* If we cannot allocate large objects within re_match_2_internal,
2232 we make the fail stack and register vectors global.
2233 The fail stack, we grow to the maximum size when a regexp
2234 is compiled.
2235 The register vectors, we adjust in size each time we
2236 compile a regexp, according to the number of registers it needs. */
2237
2238static PREFIX(fail_stack_type) fail_stack;
2239
2240/* Size with which the following vectors are currently allocated.
2241 That is so we can make them bigger as needed,
2242 but never make them smaller. */
2243# ifdef DEFINED_ONCE
2244static int regs_allocated_size;
2245
2246static const char ** regstart, ** regend;
2247static const char ** old_regstart, ** old_regend;
2248static const char **best_regstart, **best_regend;
2249static const char **reg_dummy;
2250# endif /* DEFINED_ONCE */
2251
2252static PREFIX(register_info_type) *PREFIX(reg_info);
2253static PREFIX(register_info_type) *PREFIX(reg_info_dummy);
2254
2255/* Make the register vectors big enough for NUM_REGS registers,
2256 but don't make them smaller. */
2257
2258static void
2259PREFIX(regex_grow_registers) (num_regs)
2260 int num_regs;
2261{
2262 if (num_regs > regs_allocated_size)
2263 {
2264 RETALLOC_IF (regstart, num_regs, const char *);
2265 RETALLOC_IF (regend, num_regs, const char *);
2266 RETALLOC_IF (old_regstart, num_regs, const char *);
2267 RETALLOC_IF (old_regend, num_regs, const char *);
2268 RETALLOC_IF (best_regstart, num_regs, const char *);
2269 RETALLOC_IF (best_regend, num_regs, const char *);
2270 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type));
2271 RETALLOC_IF (reg_dummy, num_regs, const char *);
2272 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type));
2273
2274 regs_allocated_size = num_regs;
2275 }
2276}
2277
2278# endif /* not MATCH_MAY_ALLOCATE */
2279\f
2280# ifndef DEFINED_ONCE
2281static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2282 compile_stack,
2283 regnum_t regnum));
2284# endif /* not DEFINED_ONCE */
2285
2286/* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2287 Returns one of error codes defined in `regex.h', or zero for success.
2288
2289 Assumes the `allocated' (and perhaps `buffer') and `translate'
2290 fields are set in BUFP on entry.
2291
2292 If it succeeds, results are put in BUFP (if it returns an error, the
2293 contents of BUFP are undefined):
2294 `buffer' is the compiled pattern;
2295 `syntax' is set to SYNTAX;
2296 `used' is set to the length of the compiled pattern;
2297 `fastmap_accurate' is zero;
2298 `re_nsub' is the number of subexpressions in PATTERN;
2299 `not_bol' and `not_eol' are zero;
2300
2301 The `fastmap' and `newline_anchor' fields are neither
2302 examined nor set. */
2303
2304/* Return, freeing storage we allocated. */
2305# ifdef WCHAR
2306# define FREE_STACK_RETURN(value) \
2307 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2308# else
2309# define FREE_STACK_RETURN(value) \
2310 return (free (compile_stack.stack), value)
2311# endif /* WCHAR */
2312
2313static reg_errcode_t
2314PREFIX(regex_compile) (ARG_PREFIX(pattern), ARG_PREFIX(size), syntax, bufp)
2315 const char *ARG_PREFIX(pattern);
2316 size_t ARG_PREFIX(size);
2317 reg_syntax_t syntax;
2318 struct re_pattern_buffer *bufp;
2319{
2320 /* We fetch characters from PATTERN here. Even though PATTERN is
2321 `char *' (i.e., signed), we declare these variables as unsigned, so
2322 they can be reliably used as array indices. */
2323 register UCHAR_T c, c1;
2324
2325#ifdef WCHAR
2326 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2327 CHAR_T *pattern, *COMPILED_BUFFER_VAR;
2328 size_t size;
2329 /* offset buffer for optimization. See convert_mbs_to_wc. */
2330 int *mbs_offset = NULL;
2331 /* It hold whether each wchar_t is binary data or not. */
2332 char *is_binary = NULL;
2333 /* A flag whether exactn is handling binary data or not. */
2334 char is_exactn_bin = FALSE;
2335#endif /* WCHAR */
2336
2337 /* A random temporary spot in PATTERN. */
2338 const CHAR_T *p1;
2339
2340 /* Points to the end of the buffer, where we should append. */
2341 register UCHAR_T *b;
2342
2343 /* Keeps track of unclosed groups. */
2344 compile_stack_type compile_stack;
2345
2346 /* Points to the current (ending) position in the pattern. */
2347#ifdef WCHAR
2348 const CHAR_T *p;
2349 const CHAR_T *pend;
2350#else /* BYTE */
2351 const CHAR_T *p = pattern;
2352 const CHAR_T *pend = pattern + size;
2353#endif /* WCHAR */
2354
2355 /* How to translate the characters in the pattern. */
2356 RE_TRANSLATE_TYPE translate = bufp->translate;
2357
2358 /* Address of the count-byte of the most recently inserted `exactn'
2359 command. This makes it possible to tell if a new exact-match
2360 character can be added to that command or if the character requires
2361 a new `exactn' command. */
2362 UCHAR_T *pending_exact = 0;
2363
2364 /* Address of start of the most recently finished expression.
2365 This tells, e.g., postfix * where to find the start of its
2366 operand. Reset at the beginning of groups and alternatives. */
2367 UCHAR_T *laststart = 0;
2368
2369 /* Address of beginning of regexp, or inside of last group. */
2370 UCHAR_T *begalt;
2371
2372 /* Address of the place where a forward jump should go to the end of
2373 the containing expression. Each alternative of an `or' -- except the
2374 last -- ends with a forward jump of this sort. */
2375 UCHAR_T *fixup_alt_jump = 0;
2376
2377 /* Counts open-groups as they are encountered. Remembered for the
2378 matching close-group on the compile stack, so the same register
2379 number is put in the stop_memory as the start_memory. */
2380 regnum_t regnum = 0;
2381
2382#ifdef WCHAR
2383 /* Initialize the wchar_t PATTERN and offset_buffer. */
2384 p = pend = pattern = TALLOC(csize + 1, CHAR_T);
2385 mbs_offset = TALLOC(csize + 1, int);
2386 is_binary = TALLOC(csize + 1, char);
2387 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2388 {
2389 free(pattern);
2390 free(mbs_offset);
2391 free(is_binary);
2392 return REG_ESPACE;
2393 }
2394 pattern[csize] = L'\0'; /* sentinel */
2395 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2396 pend = p + size;
2397 if (size < 0)
2398 {
2399 free(pattern);
2400 free(mbs_offset);
2401 free(is_binary);
2402 return REG_BADPAT;
2403 }
2404#endif
2405
2406#ifdef DEBUG
2407 DEBUG_PRINT1 ("\nCompiling pattern: ");
2408 if (debug)
2409 {
2410 unsigned debug_count;
2411
2412 for (debug_count = 0; debug_count < size; debug_count++)
2413 PUT_CHAR (pattern[debug_count]);
2414 putchar ('\n');
2415 }
2416#endif /* DEBUG */
2417
2418 /* Initialize the compile stack. */
2419 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2420 if (compile_stack.stack == NULL)
2421 {
2422#ifdef WCHAR
2423 free(pattern);
2424 free(mbs_offset);
2425 free(is_binary);
2426#endif
2427 return REG_ESPACE;
2428 }
2429
2430 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2431 compile_stack.avail = 0;
2432
2433 /* Initialize the pattern buffer. */
2434 bufp->syntax = syntax;
2435 bufp->fastmap_accurate = 0;
2436 bufp->not_bol = bufp->not_eol = 0;
2437
2438 /* Set `used' to zero, so that if we return an error, the pattern
2439 printer (for debugging) will think there's no pattern. We reset it
2440 at the end. */
2441 bufp->used = 0;
2442
2443 /* Always count groups, whether or not bufp->no_sub is set. */
2444 bufp->re_nsub = 0;
2445
2446#if !defined emacs && !defined SYNTAX_TABLE
2447 /* Initialize the syntax table. */
2448 init_syntax_once ();
2449#endif
2450
2451 if (bufp->allocated == 0)
2452 {
2453 if (bufp->buffer)
2454 { /* If zero allocated, but buffer is non-null, try to realloc
2455 enough space. This loses if buffer's address is bogus, but
2456 that is the user's responsibility. */
2457#ifdef WCHAR
2458 /* Free bufp->buffer and allocate an array for wchar_t pattern
2459 buffer. */
2460 free(bufp->buffer);
2461 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T),
2462 UCHAR_T);
2463#else
2464 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T);
2465#endif /* WCHAR */
2466 }
2467 else
2468 { /* Caller did not allocate a buffer. Do it for them. */
2469 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T),
2470 UCHAR_T);
2471 }
2472
2473 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2474#ifdef WCHAR
2475 bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2476#endif /* WCHAR */
2477 bufp->allocated = INIT_BUF_SIZE;
2478 }
2479#ifdef WCHAR
2480 else
2481 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer;
2482#endif
2483
2484 begalt = b = COMPILED_BUFFER_VAR;
2485
2486 /* Loop through the uncompiled pattern until we're at the end. */
2487 while (p != pend)
2488 {
2489 PATFETCH (c);
2490
2491 switch (c)
2492 {
2493 case '^':
2494 {
2495 if ( /* If at start of pattern, it's an operator. */
2496 p == pattern + 1
2497 /* If context independent, it's an operator. */
2498 || syntax & RE_CONTEXT_INDEP_ANCHORS
2499 /* Otherwise, depends on what's come before. */
2500 || PREFIX(at_begline_loc_p) (pattern, p, syntax))
2501 BUF_PUSH (begline);
2502 else
2503 goto normal_char;
2504 }
2505 break;
2506
2507
2508 case '$':
2509 {
2510 if ( /* If at end of pattern, it's an operator. */
2511 p == pend
2512 /* If context independent, it's an operator. */
2513 || syntax & RE_CONTEXT_INDEP_ANCHORS
2514 /* Otherwise, depends on what's next. */
2515 || PREFIX(at_endline_loc_p) (p, pend, syntax))
2516 BUF_PUSH (endline);
2517 else
2518 goto normal_char;
2519 }
2520 break;
2521
2522
2523 case '+':
2524 case '?':
2525 if ((syntax & RE_BK_PLUS_QM)
2526 || (syntax & RE_LIMITED_OPS))
2527 goto normal_char;
2528 handle_plus:
2529 case '*':
2530 /* If there is no previous pattern... */
2531 if (!laststart)
2532 {
2533 if (syntax & RE_CONTEXT_INVALID_OPS)
2534 FREE_STACK_RETURN (REG_BADRPT);
2535 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2536 goto normal_char;
2537 }
2538
2539 {
2540 /* Are we optimizing this jump? */
2541 boolean keep_string_p = false;
2542
2543 /* 1 means zero (many) matches is allowed. */
2544 char zero_times_ok = 0, many_times_ok = 0;
2545
2546 /* If there is a sequence of repetition chars, collapse it
2547 down to just one (the right one). We can't combine
2548 interval operators with these because of, e.g., `a{2}*',
2549 which should only match an even number of `a's. */
2550
2551 for (;;)
2552 {
2553 zero_times_ok |= c != '+';
2554 many_times_ok |= c != '?';
2555
2556 if (p == pend)
2557 break;
2558
2559 PATFETCH (c);
2560
2561 if (c == '*'
2562 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2563 ;
2564
2565 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2566 {
2567 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2568
2569 PATFETCH (c1);
2570 if (!(c1 == '+' || c1 == '?'))
2571 {
2572 PATUNFETCH;
2573 PATUNFETCH;
2574 break;
2575 }
2576
2577 c = c1;
2578 }
2579 else
2580 {
2581 PATUNFETCH;
2582 break;
2583 }
2584
2585 /* If we get here, we found another repeat character. */
2586 }
2587
2588 /* Star, etc. applied to an empty pattern is equivalent
2589 to an empty pattern. */
2590 if (!laststart)
2591 break;
2592
2593 /* Now we know whether or not zero matches is allowed
2594 and also whether or not two or more matches is allowed. */
2595 if (many_times_ok)
2596 { /* More than one repetition is allowed, so put in at the
2597 end a backward relative jump from `b' to before the next
2598 jump we're going to put in below (which jumps from
2599 laststart to after this jump).
2600
2601 But if we are at the `*' in the exact sequence `.*\n',
2602 insert an unconditional jump backwards to the .,
2603 instead of the beginning of the loop. This way we only
2604 push a failure point once, instead of every time
2605 through the loop. */
2606 assert (p - 1 > pattern);
2607
2608 /* Allocate the space for the jump. */
2609 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2610
2611 /* We know we are not at the first character of the pattern,
2612 because laststart was nonzero. And we've already
2613 incremented `p', by the way, to be the character after
2614 the `*'. Do we have to do something analogous here
2615 for null bytes, because of RE_DOT_NOT_NULL? */
2616 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2617 && zero_times_ok
2618 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2619 && !(syntax & RE_DOT_NEWLINE))
2620 { /* We have .*\n. */
2621 STORE_JUMP (jump, b, laststart);
2622 keep_string_p = true;
2623 }
2624 else
2625 /* Anything else. */
2626 STORE_JUMP (maybe_pop_jump, b, laststart -
2627 (1 + OFFSET_ADDRESS_SIZE));
2628
2629 /* We've added more stuff to the buffer. */
2630 b += 1 + OFFSET_ADDRESS_SIZE;
2631 }
2632
2633 /* On failure, jump from laststart to b + 3, which will be the
2634 end of the buffer after this jump is inserted. */
2635 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2636 'b + 3'. */
2637 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2638 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2639 : on_failure_jump,
2640 laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2641 pending_exact = 0;
2642 b += 1 + OFFSET_ADDRESS_SIZE;
2643
2644 if (!zero_times_ok)
2645 {
2646 /* At least one repetition is required, so insert a
2647 `dummy_failure_jump' before the initial
2648 `on_failure_jump' instruction of the loop. This
2649 effects a skip over that instruction the first time
2650 we hit that loop. */
2651 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2652 INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2653 2 + 2 * OFFSET_ADDRESS_SIZE);
2654 b += 1 + OFFSET_ADDRESS_SIZE;
2655 }
2656 }
2657 break;
2658
2659
2660 case '.':
2661 laststart = b;
2662 BUF_PUSH (anychar);
2663 break;
2664
2665
2666 case '[':
2667 {
2668 boolean had_char_class = false;
2669#ifdef WCHAR
2670 CHAR_T range_start = 0xffffffff;
2671#else
2672 unsigned int range_start = 0xffffffff;
2673#endif
2674 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2675
2676#ifdef WCHAR
2677 /* We assume a charset(_not) structure as a wchar_t array.
2678 charset[0] = (re_opcode_t) charset(_not)
2679 charset[1] = l (= length of char_classes)
2680 charset[2] = m (= length of collating_symbols)
2681 charset[3] = n (= length of equivalence_classes)
2682 charset[4] = o (= length of char_ranges)
2683 charset[5] = p (= length of chars)
2684
2685 charset[6] = char_class (wctype_t)
2686 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2687 ...
2688 charset[l+5] = char_class (wctype_t)
2689
2690 charset[l+6] = collating_symbol (wchar_t)
2691 ...
2692 charset[l+m+5] = collating_symbol (wchar_t)
2693 ifdef _LIBC we use the index if
2694 _NL_COLLATE_SYMB_EXTRAMB instead of
2695 wchar_t string.
2696
2697 charset[l+m+6] = equivalence_classes (wchar_t)
2698 ...
2699 charset[l+m+n+5] = equivalence_classes (wchar_t)
2700 ifdef _LIBC we use the index in
2701 _NL_COLLATE_WEIGHT instead of
2702 wchar_t string.
2703
2704 charset[l+m+n+6] = range_start
2705 charset[l+m+n+7] = range_end
2706 ...
2707 charset[l+m+n+2o+4] = range_start
2708 charset[l+m+n+2o+5] = range_end
2709 ifdef _LIBC we use the value looked up
2710 in _NL_COLLATE_COLLSEQ instead of
2711 wchar_t character.
2712
2713 charset[l+m+n+2o+6] = char
2714 ...
2715 charset[l+m+n+2o+p+5] = char
2716
2717 */
2718
2719 /* We need at least 6 spaces: the opcode, the length of
2720 char_classes, the length of collating_symbols, the length of
2721 equivalence_classes, the length of char_ranges, the length of
2722 chars. */
2723 GET_BUFFER_SPACE (6);
2724
2725 /* Save b as laststart. And We use laststart as the pointer
2726 to the first element of the charset here.
2727 In other words, laststart[i] indicates charset[i]. */
2728 laststart = b;
2729
2730 /* We test `*p == '^' twice, instead of using an if
2731 statement, so we only need one BUF_PUSH. */
2732 BUF_PUSH (*p == '^' ? charset_not : charset);
2733 if (*p == '^')
2734 p++;
2735
2736 /* Push the length of char_classes, the length of
2737 collating_symbols, the length of equivalence_classes, the
2738 length of char_ranges and the length of chars. */
2739 BUF_PUSH_3 (0, 0, 0);
2740 BUF_PUSH_2 (0, 0);
2741
2742 /* Remember the first position in the bracket expression. */
2743 p1 = p;
2744
2745 /* charset_not matches newline according to a syntax bit. */
2746 if ((re_opcode_t) b[-6] == charset_not
2747 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2748 {
2749 BUF_PUSH('\n');
2750 laststart[5]++; /* Update the length of characters */
2751 }
2752
2753 /* Read in characters and ranges, setting map bits. */
2754 for (;;)
2755 {
2756 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2757
2758 PATFETCH (c);
2759
2760 /* \ might escape characters inside [...] and [^...]. */
2761 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2762 {
2763 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2764
2765 PATFETCH (c1);
2766 BUF_PUSH(c1);
2767 laststart[5]++; /* Update the length of chars */
2768 range_start = c1;
2769 continue;
2770 }
2771
2772 /* Could be the end of the bracket expression. If it's
2773 not (i.e., when the bracket expression is `[]' so
2774 far), the ']' character bit gets set way below. */
2775 if (c == ']' && p != p1 + 1)
2776 break;
2777
2778 /* Look ahead to see if it's a range when the last thing
2779 was a character class. */
2780 if (had_char_class && c == '-' && *p != ']')
2781 FREE_STACK_RETURN (REG_ERANGE);
2782
2783 /* Look ahead to see if it's a range when the last thing
2784 was a character: if this is a hyphen not at the
2785 beginning or the end of a list, then it's the range
2786 operator. */
2787 if (c == '-'
2788 && !(p - 2 >= pattern && p[-2] == '[')
2789 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2790 && *p != ']')
2791 {
2792 reg_errcode_t ret;
2793 /* Allocate the space for range_start and range_end. */
2794 GET_BUFFER_SPACE (2);
2795 /* Update the pointer to indicate end of buffer. */
2796 b += 2;
2797 ret = wcs_compile_range (range_start, &p, pend, translate,
2798 syntax, b, laststart);
2799 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2800 range_start = 0xffffffff;
2801 }
2802 else if (p[0] == '-' && p[1] != ']')
2803 { /* This handles ranges made up of characters only. */
2804 reg_errcode_t ret;
2805
2806 /* Move past the `-'. */
2807 PATFETCH (c1);
2808 /* Allocate the space for range_start and range_end. */
2809 GET_BUFFER_SPACE (2);
2810 /* Update the pointer to indicate end of buffer. */
2811 b += 2;
2812 ret = wcs_compile_range (c, &p, pend, translate, syntax, b,
2813 laststart);
2814 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2815 range_start = 0xffffffff;
2816 }
2817
2818 /* See if we're at the beginning of a possible character
2819 class. */
2820 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2821 { /* Leave room for the null. */
2822 char str[CHAR_CLASS_MAX_LENGTH + 1];
2823
2824 PATFETCH (c);
2825 c1 = 0;
2826
2827 /* If pattern is `[[:'. */
2828 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2829
2830 for (;;)
2831 {
2832 PATFETCH (c);
2833 if ((c == ':' && *p == ']') || p == pend)
2834 break;
2835 if (c1 < CHAR_CLASS_MAX_LENGTH)
2836 str[c1++] = c;
2837 else
2838 /* This is in any case an invalid class name. */
2839 str[0] = '\0';
2840 }
2841 str[c1] = '\0';
2842
2843 /* If isn't a word bracketed by `[:' and `:]':
2844 undo the ending character, the letters, and leave
2845 the leading `:' and `[' (but store them as character). */
2846 if (c == ':' && *p == ']')
2847 {
2848 wctype_t wt;
2849 uintptr_t alignedp;
2850
2851 /* Query the character class as wctype_t. */
2852 wt = IS_CHAR_CLASS (str);
2853 if (wt == 0)
2854 FREE_STACK_RETURN (REG_ECTYPE);
2855
2856 /* Throw away the ] at the end of the character
2857 class. */
2858 PATFETCH (c);
2859
2860 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2861
2862 /* Allocate the space for character class. */
2863 GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2864 /* Update the pointer to indicate end of buffer. */
2865 b += CHAR_CLASS_SIZE;
2866 /* Move data which follow character classes
2867 not to violate the data. */
2868 insert_space(CHAR_CLASS_SIZE,
2869 laststart + 6 + laststart[1],
2870 b - 1);
2871 alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2872 + __alignof__(wctype_t) - 1)
2873 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2874 /* Store the character class. */
2875 *((wctype_t*)alignedp) = wt;
2876 /* Update length of char_classes */
2877 laststart[1] += CHAR_CLASS_SIZE;
2878
2879 had_char_class = true;
2880 }
2881 else
2882 {
2883 c1++;
2884 while (c1--)
2885 PATUNFETCH;
2886 BUF_PUSH ('[');
2887 BUF_PUSH (':');
2888 laststart[5] += 2; /* Update the length of characters */
2889 range_start = ':';
2890 had_char_class = false;
2891 }
2892 }
2893 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2894 || *p == '.'))
2895 {
2896 CHAR_T str[128]; /* Should be large enough. */
2897 CHAR_T delim = *p; /* '=' or '.' */
2898# ifdef _LIBC
2899 uint32_t nrules =
2900 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2901# endif
2902 PATFETCH (c);
2903 c1 = 0;
2904
2905 /* If pattern is `[[=' or '[[.'. */
2906 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2907
2908 for (;;)
2909 {
2910 PATFETCH (c);
2911 if ((c == delim && *p == ']') || p == pend)
2912 break;
2913 if (c1 < sizeof (str) - 1)
2914 str[c1++] = c;
2915 else
2916 /* This is in any case an invalid class name. */
2917 str[0] = '\0';
2918 }
2919 str[c1] = '\0';
2920
2921 if (c == delim && *p == ']' && str[0] != '\0')
2922 {
2923 unsigned int i, offset;
2924 /* If we have no collation data we use the default
2925 collation in which each character is in a class
2926 by itself. It also means that ASCII is the
2927 character set and therefore we cannot have character
2928 with more than one byte in the multibyte
2929 representation. */
2930
2931 /* If not defined _LIBC, we push the name and
2932 `\0' for the sake of matching performance. */
2933 int datasize = c1 + 1;
2934
2935# ifdef _LIBC
2936 int32_t idx = 0;
2937 if (nrules == 0)
2938# endif
2939 {
2940 if (c1 != 1)
2941 FREE_STACK_RETURN (REG_ECOLLATE);
2942 }
2943# ifdef _LIBC
2944 else
2945 {
2946 const int32_t *table;
2947 const int32_t *weights;
2948 const int32_t *extra;
2949 const int32_t *indirect;
2950 wint_t *cp;
2951
2952 /* This #include defines a local function! */
2953# include <locale/weightwc.h>
2954
2955 if(delim == '=')
2956 {
2957 /* We push the index for equivalence class. */
2958 cp = (wint_t*)str;
2959
2960 table = (const int32_t *)
2961 _NL_CURRENT (LC_COLLATE,
2962 _NL_COLLATE_TABLEWC);
2963 weights = (const int32_t *)
2964 _NL_CURRENT (LC_COLLATE,
2965 _NL_COLLATE_WEIGHTWC);
2966 extra = (const int32_t *)
2967 _NL_CURRENT (LC_COLLATE,
2968 _NL_COLLATE_EXTRAWC);
2969 indirect = (const int32_t *)
2970 _NL_CURRENT (LC_COLLATE,
2971 _NL_COLLATE_INDIRECTWC);
2972
2973 idx = findidx ((const wint_t**)&cp);
2974 if (idx == 0 || cp < (wint_t*) str + c1)
2975 /* This is no valid character. */
2976 FREE_STACK_RETURN (REG_ECOLLATE);
2977
2978 str[0] = (wchar_t)idx;
2979 }
2980 else /* delim == '.' */
2981 {
2982 /* We push collation sequence value
2983 for collating symbol. */
2984 int32_t table_size;
2985 const int32_t *symb_table;
2986 const unsigned char *extra;
2987 int32_t idx;
2988 int32_t elem;
2989 int32_t second;
2990 int32_t hash;
2991 char char_str[c1];
2992
2993 /* We have to convert the name to a single-byte
2994 string. This is possible since the names
2995 consist of ASCII characters and the internal
2996 representation is UCS4. */
2997 for (i = 0; i < c1; ++i)
2998 char_str[i] = str[i];
2999
3000 table_size =
3001 _NL_CURRENT_WORD (LC_COLLATE,
3002 _NL_COLLATE_SYMB_HASH_SIZEMB);
3003 symb_table = (const int32_t *)
3004 _NL_CURRENT (LC_COLLATE,
3005 _NL_COLLATE_SYMB_TABLEMB);
3006 extra = (const unsigned char *)
3007 _NL_CURRENT (LC_COLLATE,
3008 _NL_COLLATE_SYMB_EXTRAMB);
3009
3010 /* Locate the character in the hashing table. */
3011 hash = elem_hash (char_str, c1);
3012
3013 idx = 0;
3014 elem = hash % table_size;
3015 second = hash % (table_size - 2);
3016 while (symb_table[2 * elem] != 0)
3017 {
3018 /* First compare the hashing value. */
3019 if (symb_table[2 * elem] == hash
3020 && c1 == extra[symb_table[2 * elem + 1]]
a9022147 3021 && memcmp (char_str,
2a6ef469
DD		 3022 &extra[symb_table[2 * elem + 1]
3023 + 1], c1) == 0)
3024 {
3025 /* Yep, this is the entry. */
3026 idx = symb_table[2 * elem + 1];
3027 idx += 1 + extra[idx];
3028 break;
3029 }
3030
3031 /* Next entry. */
3032 elem += second;
3033 }
3034
3035 if (symb_table[2 * elem] != 0)
3036 {
3037 /* Compute the index of the byte sequence
3038 in the table. */
3039 idx += 1 + extra[idx];
3040 /* Adjust for the alignment. */
a9022147 3041 idx = (idx + 3) & ~3;
2a6ef469
DD		 3042
3043 str[0] = (wchar_t) idx + 4;
3044 }
3045 else if (symb_table[2 * elem] == 0 && c1 == 1)
3046 {
3047 /* No valid character. Match it as a
3048 single byte character. */
3049 had_char_class = false;
3050 BUF_PUSH(str[0]);
3051 /* Update the length of characters */
3052 laststart[5]++;
3053 range_start = str[0];
3054
3055 /* Throw away the ] at the end of the
3056 collating symbol. */
3057 PATFETCH (c);
3058 /* exit from the switch block. */
3059 continue;
3060 }
3061 else
3062 FREE_STACK_RETURN (REG_ECOLLATE);
3063 }
3064 datasize = 1;
3065 }
3066# endif
3067 /* Throw away the ] at the end of the equivalence
3068 class (or collating symbol). */
3069 PATFETCH (c);
3070
3071 /* Allocate the space for the equivalence class
3072 (or collating symbol) (and '\0' if needed). */
3073 GET_BUFFER_SPACE(datasize);
3074 /* Update the pointer to indicate end of buffer. */
3075 b += datasize;
3076
3077 if (delim == '=')
3078 { /* equivalence class */
3079 /* Calculate the offset of char_ranges,
3080 which is next to equivalence_classes. */
3081 offset = laststart[1] + laststart[2]
3082 + laststart[3] +6;
3083 /* Insert space. */
3084 insert_space(datasize, laststart + offset, b - 1);
3085
3086 /* Write the equivalence_class and \0. */
3087 for (i = 0 ; i < datasize ; i++)
3088 laststart[offset + i] = str[i];
3089
3090 /* Update the length of equivalence_classes. */
3091 laststart[3] += datasize;
3092 had_char_class = true;
3093 }
3094 else /* delim == '.' */
3095 { /* collating symbol */
3096 /* Calculate the offset of the equivalence_classes,
3097 which is next to collating_symbols. */
3098 offset = laststart[1] + laststart[2] + 6;
3099 /* Insert space and write the collationg_symbol
3100 and \0. */
3101 insert_space(datasize, laststart + offset, b-1);
3102 for (i = 0 ; i < datasize ; i++)
3103 laststart[offset + i] = str[i];
3104
3105 /* In re_match_2_internal if range_start < -1, we
3106 assume -range_start is the offset of the
3107 collating symbol which is specified as
3108 the character of the range start. So we assign
3109 -(laststart[1] + laststart[2] + 6) to
3110 range_start. */
3111 range_start = -(laststart[1] + laststart[2] + 6);
3112 /* Update the length of collating_symbol. */
3113 laststart[2] += datasize;
3114 had_char_class = false;
3115 }
3116 }
3117 else
3118 {
3119 c1++;
3120 while (c1--)
3121 PATUNFETCH;
3122 BUF_PUSH ('[');
3123 BUF_PUSH (delim);
3124 laststart[5] += 2; /* Update the length of characters */
3125 range_start = delim;
3126 had_char_class = false;
3127 }
3128 }
3129 else
3130 {
3131 had_char_class = false;
3132 BUF_PUSH(c);
3133 laststart[5]++; /* Update the length of characters */
3134 range_start = c;
3135 }
3136 }
3137
3138#else /* BYTE */
3139 /* Ensure that we have enough space to push a charset: the
3140 opcode, the length count, and the bitset; 34 bytes in all. */
3141 GET_BUFFER_SPACE (34);
3142
3143 laststart = b;
3144
3145 /* We test `*p == '^' twice, instead of using an if
3146 statement, so we only need one BUF_PUSH. */
3147 BUF_PUSH (*p == '^' ? charset_not : charset);
3148 if (*p == '^')
3149 p++;
3150
3151 /* Remember the first position in the bracket expression. */
3152 p1 = p;
3153
3154 /* Push the number of bytes in the bitmap. */
3155 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3156
3157 /* Clear the whole map. */
3158 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3159
3160 /* charset_not matches newline according to a syntax bit. */
3161 if ((re_opcode_t) b[-2] == charset_not
3162 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3163 SET_LIST_BIT ('\n');
3164
3165 /* Read in characters and ranges, setting map bits. */
3166 for (;;)
3167 {
3168 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3169
3170 PATFETCH (c);
3171
3172 /* \ might escape characters inside [...] and [^...]. */
3173 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3174 {
3175 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3176
3177 PATFETCH (c1);
3178 SET_LIST_BIT (c1);
3179 range_start = c1;
3180 continue;
3181 }
3182
3183 /* Could be the end of the bracket expression. If it's
3184 not (i.e., when the bracket expression is `[]' so
3185 far), the ']' character bit gets set way below. */
3186 if (c == ']' && p != p1 + 1)
3187 break;
3188
3189 /* Look ahead to see if it's a range when the last thing
3190 was a character class. */
3191 if (had_char_class && c == '-' && *p != ']')
3192 FREE_STACK_RETURN (REG_ERANGE);
3193
3194 /* Look ahead to see if it's a range when the last thing
3195 was a character: if this is a hyphen not at the
3196 beginning or the end of a list, then it's the range
3197 operator. */
3198 if (c == '-'
3199 && !(p - 2 >= pattern && p[-2] == '[')
3200 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3201 && *p != ']')
3202 {
3203 reg_errcode_t ret
3204 = byte_compile_range (range_start, &p, pend, translate,
3205 syntax, b);
3206 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3207 range_start = 0xffffffff;
3208 }
3209
3210 else if (p[0] == '-' && p[1] != ']')
3211 { /* This handles ranges made up of characters only. */
3212 reg_errcode_t ret;
3213
3214 /* Move past the `-'. */
3215 PATFETCH (c1);
3216
3217 ret = byte_compile_range (c, &p, pend, translate, syntax, b);
3218 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3219 range_start = 0xffffffff;
3220 }
3221
3222 /* See if we're at the beginning of a possible character
3223 class. */
3224
3225 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3226 { /* Leave room for the null. */
3227 char str[CHAR_CLASS_MAX_LENGTH + 1];
3228
3229 PATFETCH (c);
3230 c1 = 0;
3231
3232 /* If pattern is `[[:'. */
3233 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3234
3235 for (;;)
3236 {
3237 PATFETCH (c);
3238 if ((c == ':' && *p == ']') || p == pend)
3239 break;
3240 if (c1 < CHAR_CLASS_MAX_LENGTH)
3241 str[c1++] = c;
3242 else
3243 /* This is in any case an invalid class name. */
3244 str[0] = '\0';
3245 }
3246 str[c1] = '\0';
3247
3248 /* If isn't a word bracketed by `[:' and `:]':
3249 undo the ending character, the letters, and leave
3250 the leading `:' and `[' (but set bits for them). */
3251 if (c == ':' && *p == ']')
3252 {
3253# if defined _LIBC || WIDE_CHAR_SUPPORT
3254 boolean is_lower = STREQ (str, "lower");
3255 boolean is_upper = STREQ (str, "upper");
3256 wctype_t wt;
3257 int ch;
3258
3259 wt = IS_CHAR_CLASS (str);
3260 if (wt == 0)
3261 FREE_STACK_RETURN (REG_ECTYPE);
3262
3263 /* Throw away the ] at the end of the character
3264 class. */
3265 PATFETCH (c);
3266
3267 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3268
3269 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3270 {
3271# ifdef _LIBC
3272 if (__iswctype (__btowc (ch), wt))
3273 SET_LIST_BIT (ch);
3274# else
3275 if (iswctype (btowc (ch), wt))
3276 SET_LIST_BIT (ch);
3277# endif
3278
3279 if (translate && (is_upper || is_lower)
3280 && (ISUPPER (ch) || ISLOWER (ch)))
3281 SET_LIST_BIT (ch);
3282 }
3283
3284 had_char_class = true;
3285# else
3286 int ch;
3287 boolean is_alnum = STREQ (str, "alnum");
3288 boolean is_alpha = STREQ (str, "alpha");
3289 boolean is_blank = STREQ (str, "blank");
3290 boolean is_cntrl = STREQ (str, "cntrl");
3291 boolean is_digit = STREQ (str, "digit");
3292 boolean is_graph = STREQ (str, "graph");
3293 boolean is_lower = STREQ (str, "lower");
3294 boolean is_print = STREQ (str, "print");
3295 boolean is_punct = STREQ (str, "punct");
3296 boolean is_space = STREQ (str, "space");
3297 boolean is_upper = STREQ (str, "upper");
3298 boolean is_xdigit = STREQ (str, "xdigit");
3299
3300 if (!IS_CHAR_CLASS (str))
3301 FREE_STACK_RETURN (REG_ECTYPE);
3302
3303 /* Throw away the ] at the end of the character
3304 class. */
3305 PATFETCH (c);
3306
3307 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3308
3309 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3310 {
3311 /* This was split into 3 if's to
3312 avoid an arbitrary limit in some compiler. */
3313 if ( (is_alnum && ISALNUM (ch))
3314 || (is_alpha && ISALPHA (ch))
3315 || (is_blank && ISBLANK (ch))
3316 || (is_cntrl && ISCNTRL (ch)))
3317 SET_LIST_BIT (ch);
3318 if ( (is_digit && ISDIGIT (ch))
3319 || (is_graph && ISGRAPH (ch))
3320 || (is_lower && ISLOWER (ch))
3321 || (is_print && ISPRINT (ch)))
3322 SET_LIST_BIT (ch);
3323 if ( (is_punct && ISPUNCT (ch))
3324 || (is_space && ISSPACE (ch))
3325 || (is_upper && ISUPPER (ch))
3326 || (is_xdigit && ISXDIGIT (ch)))
3327 SET_LIST_BIT (ch);
3328 if ( translate && (is_upper || is_lower)
3329 && (ISUPPER (ch) || ISLOWER (ch)))
3330 SET_LIST_BIT (ch);
3331 }
3332 had_char_class = true;
3333# endif /* libc || wctype.h */
3334 }
3335 else
3336 {
3337 c1++;
3338 while (c1--)
3339 PATUNFETCH;
3340 SET_LIST_BIT ('[');
3341 SET_LIST_BIT (':');
3342 range_start = ':';
3343 had_char_class = false;
3344 }
3345 }
3346 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3347 {
3348 unsigned char str[MB_LEN_MAX + 1];
3349# ifdef _LIBC
3350 uint32_t nrules =
3351 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3352# endif
3353
3354 PATFETCH (c);
3355 c1 = 0;
3356
3357 /* If pattern is `[[='. */
3358 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3359
3360 for (;;)
3361 {
3362 PATFETCH (c);
3363 if ((c == '=' && *p == ']') || p == pend)
3364 break;
3365 if (c1 < MB_LEN_MAX)
3366 str[c1++] = c;
3367 else
3368 /* This is in any case an invalid class name. */
3369 str[0] = '\0';
3370 }
3371 str[c1] = '\0';
3372
3373 if (c == '=' && *p == ']' && str[0] != '\0')
3374 {
3375 /* If we have no collation data we use the default
3376 collation in which each character is in a class
3377 by itself. It also means that ASCII is the
3378 character set and therefore we cannot have character
3379 with more than one byte in the multibyte
3380 representation. */
3381# ifdef _LIBC
3382 if (nrules == 0)
3383# endif
3384 {
3385 if (c1 != 1)
3386 FREE_STACK_RETURN (REG_ECOLLATE);
3387
3388 /* Throw away the ] at the end of the equivalence
3389 class. */
3390 PATFETCH (c);
3391
3392 /* Set the bit for the character. */
3393 SET_LIST_BIT (str[0]);
3394 }
3395# ifdef _LIBC
3396 else
3397 {
3398 /* Try to match the byte sequence in `str' against
3399 those known to the collate implementation.
3400 First find out whether the bytes in `str' are
3401 actually from exactly one character. */
3402 const int32_t *table;
3403 const unsigned char *weights;
3404 const unsigned char *extra;
3405 const int32_t *indirect;
3406 int32_t idx;
3407 const unsigned char *cp = str;
3408 int ch;
3409
3410 /* This #include defines a local function! */
3411# include <locale/weight.h>
3412
3413 table = (const int32_t *)
3414 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3415 weights = (const unsigned char *)
3416 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3417 extra = (const unsigned char *)
3418 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3419 indirect = (const int32_t *)
3420 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3421
3422 idx = findidx (&cp);
3423 if (idx == 0 || cp < str + c1)
3424 /* This is no valid character. */
3425 FREE_STACK_RETURN (REG_ECOLLATE);
3426
3427 /* Throw away the ] at the end of the equivalence
3428 class. */
3429 PATFETCH (c);
3430
3431 /* Now we have to go throught the whole table
3432 and find all characters which have the same
3433 first level weight.
3434
3435 XXX Note that this is not entirely correct.
3436 we would have to match multibyte sequences
3437 but this is not possible with the current
3438 implementation. */
3439 for (ch = 1; ch < 256; ++ch)
3440 /* XXX This test would have to be changed if we
3441 would allow matching multibyte sequences. */
3442 if (table[ch] > 0)
3443 {
3444 int32_t idx2 = table[ch];
3445 size_t len = weights[idx2];
3446
3447 /* Test whether the lenghts match. */
3448 if (weights[idx] == len)
3449 {
3450 /* They do. New compare the bytes of
3451 the weight. */
3452 size_t cnt = 0;
3453
3454 while (cnt < len
3455 && (weights[idx + 1 + cnt]
3456 == weights[idx2 + 1 + cnt]))
3457 ++cnt;
3458
3459 if (cnt == len)
3460 /* They match. Mark the character as
3461 acceptable. */
3462 SET_LIST_BIT (ch);
3463 }
3464 }
3465 }
3466# endif
3467 had_char_class = true;
3468 }
3469 else
3470 {
3471 c1++;
3472 while (c1--)
3473 PATUNFETCH;
3474 SET_LIST_BIT ('[');
3475 SET_LIST_BIT ('=');
3476 range_start = '=';
3477 had_char_class = false;
3478 }
3479 }
3480 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3481 {
3482 unsigned char str[128]; /* Should be large enough. */
3483# ifdef _LIBC
3484 uint32_t nrules =
3485 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3486# endif
3487
3488 PATFETCH (c);
3489 c1 = 0;
3490
3491 /* If pattern is `[[.'. */
3492 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3493
3494 for (;;)
3495 {
3496 PATFETCH (c);
3497 if ((c == '.' && *p == ']') || p == pend)
3498 break;
3499 if (c1 < sizeof (str))
3500 str[c1++] = c;
3501 else
3502 /* This is in any case an invalid class name. */
3503 str[0] = '\0';
3504 }
3505 str[c1] = '\0';
3506
3507 if (c == '.' && *p == ']' && str[0] != '\0')
3508 {
3509 /* If we have no collation data we use the default
3510 collation in which each character is the name
3511 for its own class which contains only the one
3512 character. It also means that ASCII is the
3513 character set and therefore we cannot have character
3514 with more than one byte in the multibyte
3515 representation. */
3516# ifdef _LIBC
3517 if (nrules == 0)
3518# endif
3519 {
3520 if (c1 != 1)
3521 FREE_STACK_RETURN (REG_ECOLLATE);
3522
3523 /* Throw away the ] at the end of the equivalence
3524 class. */
3525 PATFETCH (c);
3526
3527 /* Set the bit for the character. */
3528 SET_LIST_BIT (str[0]);
3529 range_start = ((const unsigned char *) str)[0];
3530 }
3531# ifdef _LIBC
3532 else
3533 {
3534 /* Try to match the byte sequence in `str' against
3535 those known to the collate implementation.
3536 First find out whether the bytes in `str' are
3537 actually from exactly one character. */
3538 int32_t table_size;
3539 const int32_t *symb_table;
3540 const unsigned char *extra;
3541 int32_t idx;
3542 int32_t elem;
3543 int32_t second;
3544 int32_t hash;
3545
3546 table_size =
3547 _NL_CURRENT_WORD (LC_COLLATE,
3548 _NL_COLLATE_SYMB_HASH_SIZEMB);
3549 symb_table = (const int32_t *)
3550 _NL_CURRENT (LC_COLLATE,
3551 _NL_COLLATE_SYMB_TABLEMB);
3552 extra = (const unsigned char *)
3553 _NL_CURRENT (LC_COLLATE,
3554 _NL_COLLATE_SYMB_EXTRAMB);
3555
3556 /* Locate the character in the hashing table. */
3557 hash = elem_hash (str, c1);
3558
3559 idx = 0;
3560 elem = hash % table_size;
3561 second = hash % (table_size - 2);
3562 while (symb_table[2 * elem] != 0)
3563 {
3564 /* First compare the hashing value. */
3565 if (symb_table[2 * elem] == hash
3566 && c1 == extra[symb_table[2 * elem + 1]]
3567 && memcmp (str,
3568 &extra[symb_table[2 * elem + 1]
3569 + 1],
3570 c1) == 0)
3571 {
3572 /* Yep, this is the entry. */
3573 idx = symb_table[2 * elem + 1];
3574 idx += 1 + extra[idx];
3575 break;
3576 }
3577
3578 /* Next entry. */
3579 elem += second;
3580 }
3581
3582 if (symb_table[2 * elem] == 0)
3583 /* This is no valid character. */
3584 FREE_STACK_RETURN (REG_ECOLLATE);
3585
3586 /* Throw away the ] at the end of the equivalence
3587 class. */
3588 PATFETCH (c);
3589
3590 /* Now add the multibyte character(s) we found
3591 to the accept list.
3592
3593 XXX Note that this is not entirely correct.
3594 we would have to match multibyte sequences
3595 but this is not possible with the current
3596 implementation. Also, we have to match
3597 collating symbols, which expand to more than
3598 one file, as a whole and not allow the
3599 individual bytes. */
3600 c1 = extra[idx++];
3601 if (c1 == 1)
3602 range_start = extra[idx];
3603 while (c1-- > 0)
3604 {
3605 SET_LIST_BIT (extra[idx]);
3606 ++idx;
3607 }
3608 }
3609# endif
3610 had_char_class = false;
3611 }
3612 else
3613 {
3614 c1++;
3615 while (c1--)
3616 PATUNFETCH;
3617 SET_LIST_BIT ('[');
3618 SET_LIST_BIT ('.');
3619 range_start = '.';
3620 had_char_class = false;
3621 }
3622 }
3623 else
3624 {
3625 had_char_class = false;
3626 SET_LIST_BIT (c);
3627 range_start = c;
3628 }
3629 }
3630
3631 /* Discard any (non)matching list bytes that are all 0 at the
3632 end of the map. Decrease the map-length byte too. */
3633 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3634 b[-1]--;
3635 b += b[-1];
3636#endif /* WCHAR */
3637 }
3638 break;
3639
3640
3641 case '(':
3642 if (syntax & RE_NO_BK_PARENS)
3643 goto handle_open;
3644 else
3645 goto normal_char;
3646
3647
3648 case ')':
3649 if (syntax & RE_NO_BK_PARENS)
3650 goto handle_close;
3651 else
3652 goto normal_char;
3653
3654
3655 case '\n':
3656 if (syntax & RE_NEWLINE_ALT)
3657 goto handle_alt;
3658 else
3659 goto normal_char;
3660
3661
3662 case '|':
3663 if (syntax & RE_NO_BK_VBAR)
3664 goto handle_alt;
3665 else
3666 goto normal_char;
3667
3668
3669 case '{':
3670 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3671 goto handle_interval;
3672 else
3673 goto normal_char;
3674
3675
3676 case '\\':
3677 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3678
3679 /* Do not translate the character after the \, so that we can
3680 distinguish, e.g., \B from \b, even if we normally would
3681 translate, e.g., B to b. */
3682 PATFETCH_RAW (c);
3683
3684 switch (c)
3685 {
3686 case '(':
3687 if (syntax & RE_NO_BK_PARENS)
3688 goto normal_backslash;
3689
3690 handle_open:
3691 bufp->re_nsub++;
3692 regnum++;
3693
3694 if (COMPILE_STACK_FULL)
3695 {
3696 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3697 compile_stack_elt_t);
3698 if (compile_stack.stack == NULL) return REG_ESPACE;
3699
3700 compile_stack.size <<= 1;
3701 }
3702
3703 /* These are the values to restore when we hit end of this
3704 group. They are all relative offsets, so that if the
3705 whole pattern moves because of realloc, they will still
3706 be valid. */
3707 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3708 COMPILE_STACK_TOP.fixup_alt_jump
3709 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3710 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3711 COMPILE_STACK_TOP.regnum = regnum;
3712
3713 /* We will eventually replace the 0 with the number of
3714 groups inner to this one. But do not push a
3715 start_memory for groups beyond the last one we can
3716 represent in the compiled pattern. */
3717 if (regnum <= MAX_REGNUM)
3718 {
3719 COMPILE_STACK_TOP.inner_group_offset = b
3720 - COMPILED_BUFFER_VAR + 2;
3721 BUF_PUSH_3 (start_memory, regnum, 0);
3722 }
3723
3724 compile_stack.avail++;
3725
3726 fixup_alt_jump = 0;
3727 laststart = 0;
3728 begalt = b;
3729 /* If we've reached MAX_REGNUM groups, then this open
3730 won't actually generate any code, so we'll have to
3731 clear pending_exact explicitly. */
3732 pending_exact = 0;
3733 break;
3734
3735
3736 case ')':
3737 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3738
3739 if (COMPILE_STACK_EMPTY)
3740 {
3741 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3742 goto normal_backslash;
3743 else
3744 FREE_STACK_RETURN (REG_ERPAREN);
3745 }
3746
3747 handle_close:
3748 if (fixup_alt_jump)
3749 { /* Push a dummy failure point at the end of the
3750 alternative for a possible future
3751 `pop_failure_jump' to pop. See comments at
3752 `push_dummy_failure' in `re_match_2'. */
3753 BUF_PUSH (push_dummy_failure);
3754
3755 /* We allocated space for this jump when we assigned
3756 to `fixup_alt_jump', in the `handle_alt' case below. */
3757 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3758 }
3759
3760 /* See similar code for backslashed left paren above. */
3761 if (COMPILE_STACK_EMPTY)
3762 {
3763 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3764 goto normal_char;
3765 else
3766 FREE_STACK_RETURN (REG_ERPAREN);
3767 }
3768
3769 /* Since we just checked for an empty stack above, this
3770 ``can't happen''. */
3771 assert (compile_stack.avail != 0);
3772 {
3773 /* We don't just want to restore into `regnum', because
3774 later groups should continue to be numbered higher,
3775 as in `(ab)c(de)' -- the second group is #2. */
3776 regnum_t this_group_regnum;
3777
3778 compile_stack.avail--;
3779 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3780 fixup_alt_jump
3781 = COMPILE_STACK_TOP.fixup_alt_jump
3782 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3783 : 0;
3784 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3785 this_group_regnum = COMPILE_STACK_TOP.regnum;
3786 /* If we've reached MAX_REGNUM groups, then this open
3787 won't actually generate any code, so we'll have to
3788 clear pending_exact explicitly. */
3789 pending_exact = 0;
3790
3791 /* We're at the end of the group, so now we know how many
3792 groups were inside this one. */
3793 if (this_group_regnum <= MAX_REGNUM)
3794 {
3795 UCHAR_T *inner_group_loc
3796 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3797
3798 *inner_group_loc = regnum - this_group_regnum;
3799 BUF_PUSH_3 (stop_memory, this_group_regnum,
3800 regnum - this_group_regnum);
3801 }
3802 }
3803 break;
3804
3805
3806 case '|': /* `\|'. */
3807 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3808 goto normal_backslash;
3809 handle_alt:
3810 if (syntax & RE_LIMITED_OPS)
3811 goto normal_char;
3812
3813 /* Insert before the previous alternative a jump which
3814 jumps to this alternative if the former fails. */
3815 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3816 INSERT_JUMP (on_failure_jump, begalt,
3817 b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3818 pending_exact = 0;
3819 b += 1 + OFFSET_ADDRESS_SIZE;
3820
3821 /* The alternative before this one has a jump after it
3822 which gets executed if it gets matched. Adjust that
3823 jump so it will jump to this alternative's analogous
3824 jump (put in below, which in turn will jump to the next
3825 (if any) alternative's such jump, etc.). The last such
3826 jump jumps to the correct final destination. A picture:
3827 _____ _____
3828 | | | |
3829 | v | v
3830 a | b | c
3831
3832 If we are at `b', then fixup_alt_jump right now points to a
3833 three-byte space after `a'. We'll put in the jump, set
3834 fixup_alt_jump to right after `b', and leave behind three
3835 bytes which we'll fill in when we get to after `c'. */
3836
3837 if (fixup_alt_jump)
3838 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3839
3840 /* Mark and leave space for a jump after this alternative,
3841 to be filled in later either by next alternative or
3842 when know we're at the end of a series of alternatives. */
3843 fixup_alt_jump = b;
3844 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3845 b += 1 + OFFSET_ADDRESS_SIZE;
3846
3847 laststart = 0;
3848 begalt = b;
3849 break;
3850
3851
3852 case '{':
3853 /* If \{ is a literal. */
3854 if (!(syntax & RE_INTERVALS)
3855 /* If we're at `\{' and it's not the open-interval
3856 operator. */
3857 || (syntax & RE_NO_BK_BRACES))
3858 goto normal_backslash;
3859
3860 handle_interval:
3861 {
3862 /* If got here, then the syntax allows intervals. */
3863
3864 /* At least (most) this many matches must be made. */
3865 int lower_bound = -1, upper_bound = -1;
3866
3867 /* Place in the uncompiled pattern (i.e., just after
3868 the '{') to go back to if the interval is invalid. */
3869 const CHAR_T *beg_interval = p;
3870
3871 if (p == pend)
3872 goto invalid_interval;
3873
3874 GET_UNSIGNED_NUMBER (lower_bound);
3875
3876 if (c == ',')
3877 {
3878 GET_UNSIGNED_NUMBER (upper_bound);
3879 if (upper_bound < 0)
3880 upper_bound = RE_DUP_MAX;
3881 }
3882 else
3883 /* Interval such as `{1}' => match exactly once. */
3884 upper_bound = lower_bound;
3885
3886 if (! (0 <= lower_bound && lower_bound <= upper_bound))
3887 goto invalid_interval;
3888
3889 if (!(syntax & RE_NO_BK_BRACES))
3890 {
3891 if (c != '\\' || p == pend)
3892 goto invalid_interval;
3893 PATFETCH (c);
3894 }
3895
3896 if (c != '}')
3897 goto invalid_interval;
3898
3899 /* If it's invalid to have no preceding re. */
3900 if (!laststart)
3901 {
3902 if (syntax & RE_CONTEXT_INVALID_OPS
3903 && !(syntax & RE_INVALID_INTERVAL_ORD))
3904 FREE_STACK_RETURN (REG_BADRPT);
3905 else if (syntax & RE_CONTEXT_INDEP_OPS)
3906 laststart = b;
3907 else
3908 goto unfetch_interval;
3909 }
3910
3911 /* We just parsed a valid interval. */
3912
3913 if (RE_DUP_MAX < upper_bound)
3914 FREE_STACK_RETURN (REG_BADBR);
3915
3916 /* If the upper bound is zero, don't want to succeed at
3917 all; jump from `laststart' to `b + 3', which will be
3918 the end of the buffer after we insert the jump. */
3919 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3920 instead of 'b + 3'. */
3921 if (upper_bound == 0)
3922 {
3923 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3924 INSERT_JUMP (jump, laststart, b + 1
3925 + OFFSET_ADDRESS_SIZE);
3926 b += 1 + OFFSET_ADDRESS_SIZE;
3927 }
3928
3929 /* Otherwise, we have a nontrivial interval. When
3930 we're all done, the pattern will look like:
3931 set_number_at <jump count> <upper bound>
3932 set_number_at <succeed_n count> <lower bound>
3933 succeed_n <after jump addr> <succeed_n count>
3934 <body of loop>
3935 jump_n <succeed_n addr> <jump count>
3936 (The upper bound and `jump_n' are omitted if
3937 `upper_bound' is 1, though.) */
3938 else
3939 { /* If the upper bound is > 1, we need to insert
3940 more at the end of the loop. */
3941 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3942 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3943
3944 GET_BUFFER_SPACE (nbytes);
3945
3946 /* Initialize lower bound of the `succeed_n', even
3947 though it will be set during matching by its
3948 attendant `set_number_at' (inserted next),
3949 because `re_compile_fastmap' needs to know.
3950 Jump to the `jump_n' we might insert below. */
3951 INSERT_JUMP2 (succeed_n, laststart,
3952 b + 1 + 2 * OFFSET_ADDRESS_SIZE
3953 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
3954 , lower_bound);
3955 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3956
3957 /* Code to initialize the lower bound. Insert
3958 before the `succeed_n'. The `5' is the last two
3959 bytes of this `set_number_at', plus 3 bytes of
3960 the following `succeed_n'. */
3961 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3962 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
3963 of the following `succeed_n'. */
3964 PREFIX(insert_op2) (set_number_at, laststart, 1
3965 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
3966 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3967
3968 if (upper_bound > 1)
3969 { /* More than one repetition is allowed, so
3970 append a backward jump to the `succeed_n'
3971 that starts this interval.
3972
3973 When we've reached this during matching,
3974 we'll have matched the interval once, so
3975 jump back only `upper_bound - 1' times. */
3976 STORE_JUMP2 (jump_n, b, laststart
3977 + 2 * OFFSET_ADDRESS_SIZE + 1,
3978 upper_bound - 1);
3979 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3980
3981 /* The location we want to set is the second
3982 parameter of the `jump_n'; that is `b-2' as
3983 an absolute address. `laststart' will be
3984 the `set_number_at' we're about to insert;
3985 `laststart+3' the number to set, the source
3986 for the relative address. But we are
3987 inserting into the middle of the pattern --
3988 so everything is getting moved up by 5.
3989 Conclusion: (b - 2) - (laststart + 3) + 5,
3990 i.e., b - laststart.
3991
3992 We insert this at the beginning of the loop
3993 so that if we fail during matching, we'll
3994 reinitialize the bounds. */
3995 PREFIX(insert_op2) (set_number_at, laststart,
3996 b - laststart,
3997 upper_bound - 1, b);
3998 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3999 }
4000 }
4001 pending_exact = 0;
4002 break;
4003
4004 invalid_interval:
4005 if (!(syntax & RE_INVALID_INTERVAL_ORD))
4006 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
4007 unfetch_interval:
4008 /* Match the characters as literals. */
4009 p = beg_interval;
4010 c = '{';
4011 if (syntax & RE_NO_BK_BRACES)
4012 goto normal_char;
4013 else
4014 goto normal_backslash;
4015 }
4016
4017#ifdef emacs
4018 /* There is no way to specify the before_dot and after_dot
4019 operators. rms says this is ok. --karl */
4020 case '=':
4021 BUF_PUSH (at_dot);
4022 break;
4023
4024 case 's':
4025 laststart = b;
4026 PATFETCH (c);
4027 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
4028 break;
4029
4030 case 'S':
4031 laststart = b;
4032 PATFETCH (c);
4033 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
4034 break;
4035#endif /* emacs */
4036
4037
4038 case 'w':
4039 if (syntax & RE_NO_GNU_OPS)
4040 goto normal_char;
4041 laststart = b;
4042 BUF_PUSH (wordchar);
4043 break;
4044
4045
4046 case 'W':
4047 if (syntax & RE_NO_GNU_OPS)
4048 goto normal_char;
4049 laststart = b;
4050 BUF_PUSH (notwordchar);
4051 break;
4052
4053
4054 case '<':
4055 if (syntax & RE_NO_GNU_OPS)
4056 goto normal_char;
4057 BUF_PUSH (wordbeg);
4058 break;
4059
4060 case '>':
4061 if (syntax & RE_NO_GNU_OPS)
4062 goto normal_char;
4063 BUF_PUSH (wordend);
4064 break;
4065
4066 case 'b':
4067 if (syntax & RE_NO_GNU_OPS)
4068 goto normal_char;
4069 BUF_PUSH (wordbound);
4070 break;
4071
4072 case 'B':
4073 if (syntax & RE_NO_GNU_OPS)
4074 goto normal_char;
4075 BUF_PUSH (notwordbound);
4076 break;
4077
4078 case '`':
4079 if (syntax & RE_NO_GNU_OPS)
4080 goto normal_char;
4081 BUF_PUSH (begbuf);
4082 break;
4083
4084 case '\'':
4085 if (syntax & RE_NO_GNU_OPS)
4086 goto normal_char;
4087 BUF_PUSH (endbuf);
4088 break;
4089
4090 case '1': case '2': case '3': case '4': case '5':
4091 case '6': case '7': case '8': case '9':
4092 if (syntax & RE_NO_BK_REFS)
4093 goto normal_char;
4094
4095 c1 = c - '0';
4096
4097 if (c1 > regnum)
4098 FREE_STACK_RETURN (REG_ESUBREG);
4099
4100 /* Can't back reference to a subexpression if inside of it. */
4101 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4102 goto normal_char;
4103
4104 laststart = b;
4105 BUF_PUSH_2 (duplicate, c1);
4106 break;
4107
4108
4109 case '+':
4110 case '?':
4111 if (syntax & RE_BK_PLUS_QM)
4112 goto handle_plus;
4113 else
4114 goto normal_backslash;
4115
4116 default:
4117 normal_backslash:
4118 /* You might think it would be useful for \ to mean
4119 not to translate; but if we don't translate it
4120 it will never match anything. */
4121 c = TRANSLATE (c);
4122 goto normal_char;
4123 }
4124 break;
4125
4126
4127 default:
4128 /* Expects the character in `c'. */
4129 normal_char:
4130 /* If no exactn currently being built. */
4131 if (!pending_exact
4132#ifdef WCHAR
4133 /* If last exactn handle binary(or character) and
4134 new exactn handle character(or binary). */
4135 || is_exactn_bin != is_binary[p - 1 - pattern]
4136#endif /* WCHAR */
4137
4138 /* If last exactn not at current position. */
4139 || pending_exact + *pending_exact + 1 != b
4140
4141 /* We have only one byte following the exactn for the count. */
4142 || *pending_exact == (1 << BYTEWIDTH) - 1
4143
4144 /* If followed by a repetition operator. */
4145 || *p == '*' || *p == '^'
4146 || ((syntax & RE_BK_PLUS_QM)
4147 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
4148 : (*p == '+' || *p == '?'))
4149 || ((syntax & RE_INTERVALS)
4150 && ((syntax & RE_NO_BK_BRACES)
4151 ? *p == '{'
4152 : (p[0] == '\\' && p[1] == '{'))))
4153 {
4154 /* Start building a new exactn. */
4155
4156 laststart = b;
4157
4158#ifdef WCHAR
4159 /* Is this exactn binary data or character? */
4160 is_exactn_bin = is_binary[p - 1 - pattern];
4161 if (is_exactn_bin)
4162 BUF_PUSH_2 (exactn_bin, 0);
4163 else
4164 BUF_PUSH_2 (exactn, 0);
4165#else
4166 BUF_PUSH_2 (exactn, 0);
4167#endif /* WCHAR */
4168 pending_exact = b - 1;
4169 }
4170
4171 BUF_PUSH (c);
4172 (*pending_exact)++;
4173 break;
4174 } /* switch (c) */
4175 } /* while p != pend */
4176
4177
4178 /* Through the pattern now. */
4179
4180 if (fixup_alt_jump)
4181 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4182
4183 if (!COMPILE_STACK_EMPTY)
4184 FREE_STACK_RETURN (REG_EPAREN);
4185
4186 /* If we don't want backtracking, force success
4187 the first time we reach the end of the compiled pattern. */
4188 if (syntax & RE_NO_POSIX_BACKTRACKING)
4189 BUF_PUSH (succeed);
4190
4191#ifdef WCHAR
4192 free (pattern);
4193 free (mbs_offset);
4194 free (is_binary);
4195#endif
4196 free (compile_stack.stack);
4197
4198 /* We have succeeded; set the length of the buffer. */
4199#ifdef WCHAR
4200 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4201#else
4202 bufp->used = b - bufp->buffer;
4203#endif
4204
4205#ifdef DEBUG
4206 if (debug)
4207 {
4208 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4209 PREFIX(print_compiled_pattern) (bufp);
4210 }
4211#endif /* DEBUG */
4212
4213#ifndef MATCH_MAY_ALLOCATE
4214 /* Initialize the failure stack to the largest possible stack. This
4215 isn't necessary unless we're trying to avoid calling alloca in
4216 the search and match routines. */
4217 {
4218 int num_regs = bufp->re_nsub + 1;
4219
4220 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4221 is strictly greater than re_max_failures, the largest possible stack
4222 is 2 * re_max_failures failure points. */
4223 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4224 {
4225 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4226
4227# ifdef emacs
4228 if (! fail_stack.stack)
4229 fail_stack.stack
4230 = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size
4231 * sizeof (PREFIX(fail_stack_elt_t)));
4232 else
4233 fail_stack.stack
4234 = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack,
4235 (fail_stack.size
4236 * sizeof (PREFIX(fail_stack_elt_t))));
4237# else /* not emacs */
4238 if (! fail_stack.stack)
4239 fail_stack.stack
4240 = (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size
4241 * sizeof (PREFIX(fail_stack_elt_t)));
4242 else
4243 fail_stack.stack
4244 = (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack,
4245 (fail_stack.size
4246 * sizeof (PREFIX(fail_stack_elt_t))));
4247# endif /* not emacs */
4248 }
4249
4250 PREFIX(regex_grow_registers) (num_regs);
4251 }
4252#endif /* not MATCH_MAY_ALLOCATE */
4253
4254 return REG_NOERROR;
4255} /* regex_compile */
4256
4257/* Subroutines for `regex_compile'. */
4258
4259/* Store OP at LOC followed by two-byte integer parameter ARG. */
4260/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4261
4262static void
4263PREFIX(store_op1) (op, loc, arg)
4264 re_opcode_t op;
4265 UCHAR_T *loc;
4266 int arg;
4267{
4268 *loc = (UCHAR_T) op;
4269 STORE_NUMBER (loc + 1, arg);
4270}
4271
4272
4273/* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4274/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4275
4276static void
4277PREFIX(store_op2) (op, loc, arg1, arg2)
4278 re_opcode_t op;
4279 UCHAR_T *loc;
4280 int arg1, arg2;
4281{
4282 *loc = (UCHAR_T) op;
4283 STORE_NUMBER (loc + 1, arg1);
4284 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4285}
4286
4287
4288/* Copy the bytes from LOC to END to open up three bytes of space at LOC
4289 for OP followed by two-byte integer parameter ARG. */
4290/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4291
4292static void
4293PREFIX(insert_op1) (op, loc, arg, end)
4294 re_opcode_t op;
4295 UCHAR_T *loc;
4296 int arg;
4297 UCHAR_T *end;
4298{
4299 register UCHAR_T *pfrom = end;
4300 register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4301
4302 while (pfrom != loc)
4303 *--pto = *--pfrom;
4304
4305 PREFIX(store_op1) (op, loc, arg);
4306}
4307
4308
4309/* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4310/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4311
4312static void
4313PREFIX(insert_op2) (op, loc, arg1, arg2, end)
4314 re_opcode_t op;
4315 UCHAR_T *loc;
4316 int arg1, arg2;
4317 UCHAR_T *end;
4318{
4319 register UCHAR_T *pfrom = end;
4320 register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4321
4322 while (pfrom != loc)
4323 *--pto = *--pfrom;
4324
4325 PREFIX(store_op2) (op, loc, arg1, arg2);
4326}
4327
4328
4329/* P points to just after a ^ in PATTERN. Return true if that ^ comes
4330 after an alternative or a begin-subexpression. We assume there is at
4331 least one character before the ^. */
4332
4333static boolean
4334PREFIX(at_begline_loc_p) (pattern, p, syntax)
4335 const CHAR_T *pattern, *p;
4336 reg_syntax_t syntax;
4337{
4338 const CHAR_T *prev = p - 2;
4339 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
4340
4341 return
4342 /* After a subexpression? */
4343 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
4344 /* After an alternative? */
4345 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
4346}
4347
4348
4349/* The dual of at_begline_loc_p. This one is for $. We assume there is
4350 at least one character after the $, i.e., `P < PEND'. */
4351
4352static boolean
4353PREFIX(at_endline_loc_p) (p, pend, syntax)
4354 const CHAR_T *p, *pend;
4355 reg_syntax_t syntax;
4356{
4357 const CHAR_T *next = p;
4358 boolean next_backslash = *next == '\\';
4359 const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0;
4360
4361 return
4362 /* Before a subexpression? */
4363 (syntax & RE_NO_BK_PARENS ? *next == ')'
4364 : next_backslash && next_next && *next_next == ')')
4365 /* Before an alternative? */
4366 || (syntax & RE_NO_BK_VBAR ? *next == '|'
4367 : next_backslash && next_next && *next_next == '|');
4368}
4369
4370#else /* not INSIDE_RECURSION */
4371
4372/* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4373 false if it's not. */
4374
4375static boolean
4376group_in_compile_stack (compile_stack, regnum)
4377 compile_stack_type compile_stack;
4378 regnum_t regnum;
4379{
4380 int this_element;
4381
4382 for (this_element = compile_stack.avail - 1;
4383 this_element >= 0;
4384 this_element--)
4385 if (compile_stack.stack[this_element].regnum == regnum)
4386 return true;
4387
4388 return false;
4389}
4390#endif /* not INSIDE_RECURSION */
4391
4392#ifdef INSIDE_RECURSION
4393
4394#ifdef WCHAR
4395/* This insert space, which size is "num", into the pattern at "loc".
4396 "end" must point the end of the allocated buffer. */
4397static void
4398insert_space (num, loc, end)
4399 int num;
4400 CHAR_T *loc;
4401 CHAR_T *end;
4402{
4403 register CHAR_T *pto = end;
4404 register CHAR_T *pfrom = end - num;
4405
4406 while (pfrom >= loc)
4407 *pto-- = *pfrom--;
4408}
4409#endif /* WCHAR */
4410
4411#ifdef WCHAR
4412static reg_errcode_t
4413wcs_compile_range (range_start_char, p_ptr, pend, translate, syntax, b,
4414 char_set)
4415 CHAR_T range_start_char;
4416 const CHAR_T **p_ptr, *pend;
4417 CHAR_T *char_set, *b;
4418 RE_TRANSLATE_TYPE translate;
4419 reg_syntax_t syntax;
4420{
4421 const CHAR_T *p = *p_ptr;
4422 CHAR_T range_start, range_end;
4423 reg_errcode_t ret;
4424# ifdef _LIBC
4425 uint32_t nrules;
4426 uint32_t start_val, end_val;
4427# endif
4428 if (p == pend)
4429 return REG_ERANGE;
4430
4431# ifdef _LIBC
4432 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4433 if (nrules != 0)
4434 {
4435 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4436 _NL_COLLATE_COLLSEQWC);
4437 const unsigned char *extra = (const unsigned char *)
4438 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4439
4440 if (range_start_char < -1)
4441 {
4442 /* range_start is a collating symbol. */
4443 int32_t *wextra;
4444 /* Retreive the index and get collation sequence value. */
4445 wextra = (int32_t*)(extra + char_set[-range_start_char]);
4446 start_val = wextra[1 + *wextra];
4447 }
4448 else
4449 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4450
4451 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4452
4453 /* Report an error if the range is empty and the syntax prohibits
4454 this. */
4455 ret = ((syntax & RE_NO_EMPTY_RANGES)
4456 && (start_val > end_val))? REG_ERANGE : REG_NOERROR;
4457
4458 /* Insert space to the end of the char_ranges. */
4459 insert_space(2, b - char_set[5] - 2, b - 1);
4460 *(b - char_set[5] - 2) = (wchar_t)start_val;
4461 *(b - char_set[5] - 1) = (wchar_t)end_val;
4462 char_set[4]++; /* ranges_index */
4463 }
4464 else
4465# endif
4466 {
4467 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4468 range_start_char;
4469 range_end = TRANSLATE (p[0]);
4470 /* Report an error if the range is empty and the syntax prohibits
4471 this. */
4472 ret = ((syntax & RE_NO_EMPTY_RANGES)
4473 && (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4474
4475 /* Insert space to the end of the char_ranges. */
4476 insert_space(2, b - char_set[5] - 2, b - 1);
4477 *(b - char_set[5] - 2) = range_start;
4478 *(b - char_set[5] - 1) = range_end;
4479 char_set[4]++; /* ranges_index */
4480 }
4481 /* Have to increment the pointer into the pattern string, so the
4482 caller isn't still at the ending character. */
4483 (*p_ptr)++;
4484
4485 return ret;
4486}
4487#else /* BYTE */
4488/* Read the ending character of a range (in a bracket expression) from the
4489 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4490 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4491 Then we set the translation of all bits between the starting and
4492 ending characters (inclusive) in the compiled pattern B.
4493
4494 Return an error code.
4495
4496 We use these short variable names so we can use the same macros as
4497 `regex_compile' itself. */
4498
4499static reg_errcode_t
4500byte_compile_range (range_start_char, p_ptr, pend, translate, syntax, b)
4501 unsigned int range_start_char;
4502 const char **p_ptr, *pend;
4503 RE_TRANSLATE_TYPE translate;
4504 reg_syntax_t syntax;
4505 unsigned char *b;
4506{
4507 unsigned this_char;
4508 const char *p = *p_ptr;
4509 reg_errcode_t ret;
4510# if _LIBC
4511 const unsigned char *collseq;
4512 unsigned int start_colseq;
4513 unsigned int end_colseq;
4514# else
4515 unsigned end_char;
4516# endif
4517
4518 if (p == pend)
4519 return REG_ERANGE;
4520
4521 /* Have to increment the pointer into the pattern string, so the
4522 caller isn't still at the ending character. */
4523 (*p_ptr)++;
4524
4525 /* Report an error if the range is empty and the syntax prohibits this. */
4526 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4527
4528# if _LIBC
4529 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4530 _NL_COLLATE_COLLSEQMB);
4531
4532 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4533 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4534 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4535 {
4536 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4537
4538 if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4539 {
4540 SET_LIST_BIT (TRANSLATE (this_char));
4541 ret = REG_NOERROR;
4542 }
4543 }
4544# else
4545 /* Here we see why `this_char' has to be larger than an `unsigned
4546 char' -- we would otherwise go into an infinite loop, since all
4547 characters <= 0xff. */
4548 range_start_char = TRANSLATE (range_start_char);
4549 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4550 and some compilers cast it to int implicitly, so following for_loop
4551 may fall to (almost) infinite loop.
4552 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4553 To avoid this, we cast p[0] to unsigned int and truncate it. */
4554 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4555
4556 for (this_char = range_start_char; this_char <= end_char; ++this_char)
4557 {
4558 SET_LIST_BIT (TRANSLATE (this_char));
4559 ret = REG_NOERROR;
4560 }
4561# endif
4562
4563 return ret;
4564}
4565#endif /* WCHAR */
4566\f
4567/* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4568 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4569 characters can start a string that matches the pattern. This fastmap
4570 is used by re_search to skip quickly over impossible starting points.
4571
4572 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4573 area as BUFP->fastmap.
4574
4575 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4576 the pattern buffer.
4577
4578 Returns 0 if we succeed, -2 if an internal error. */
4579
4580#ifdef WCHAR
4581/* local function for re_compile_fastmap.
4582 truncate wchar_t character to char. */
4583static unsigned char truncate_wchar (CHAR_T c);
4584
4585static unsigned char
4586truncate_wchar (c)
4587 CHAR_T c;
4588{
dc676635
DD		 4589 unsigned char buf[MB_CUR_MAX];
4590 mbstate_t state;
4591 int retval;
4592 memset (&state, '\0', sizeof (state));
4593# ifdef _LIBC
4594 retval = __wcrtomb (buf, c, &state);
4595# else
4596 retval = wcrtomb (buf, c, &state);
4597# endif
4598 return retval > 0 ? buf[0] : (unsigned char) c;
2a6ef469
DD		 4599}
4600#endif /* WCHAR */
4601
4602static int
4603PREFIX(re_compile_fastmap) (bufp)
4604 struct re_pattern_buffer *bufp;
4605{
4606 int j, k;
4607#ifdef MATCH_MAY_ALLOCATE
4608 PREFIX(fail_stack_type) fail_stack;
4609#endif
4610#ifndef REGEX_MALLOC
4611 char *destination;
4612#endif
4613
4614 register char *fastmap = bufp->fastmap;
4615
4616#ifdef WCHAR
4617 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4618 pattern to (char*) in regex_compile. */
4619 UCHAR_T *pattern = (UCHAR_T*)bufp->buffer;
4620 register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used);
4621#else /* BYTE */
4622 UCHAR_T *pattern = bufp->buffer;
4623 register UCHAR_T *pend = pattern + bufp->used;
4624#endif /* WCHAR */
4625 UCHAR_T *p = pattern;
4626
4627#ifdef REL_ALLOC
4628 /* This holds the pointer to the failure stack, when
4629 it is allocated relocatably. */
4630 fail_stack_elt_t *failure_stack_ptr;
4631#endif
4632
4633 /* Assume that each path through the pattern can be null until
4634 proven otherwise. We set this false at the bottom of switch
4635 statement, to which we get only if a particular path doesn't
4636 match the empty string. */
4637 boolean path_can_be_null = true;
4638
4639 /* We aren't doing a `succeed_n' to begin with. */
4640 boolean succeed_n_p = false;
4641
4642 assert (fastmap != NULL && p != NULL);
4643
4644 INIT_FAIL_STACK ();
4645 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4646 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4647 bufp->can_be_null = 0;
4648
4649 while (1)
4650 {
4651 if (p == pend || *p == succeed)
4652 {
4653 /* We have reached the (effective) end of pattern. */
4654 if (!FAIL_STACK_EMPTY ())
4655 {
4656 bufp->can_be_null |= path_can_be_null;
4657
4658 /* Reset for next path. */
4659 path_can_be_null = true;
4660
4661 p = fail_stack.stack[--fail_stack.avail].pointer;
4662
4663 continue;
4664 }
4665 else
4666 break;
4667 }
4668
4669 /* We should never be about to go beyond the end of the pattern. */
4670 assert (p < pend);
4671
4672 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4673 {
4674
4675 /* I guess the idea here is to simply not bother with a fastmap
4676 if a backreference is used, since it's too hard to figure out
4677 the fastmap for the corresponding group. Setting
4678 `can_be_null' stops `re_search_2' from using the fastmap, so
4679 that is all we do. */
4680 case duplicate:
4681 bufp->can_be_null = 1;
4682 goto done;
4683
4684
4685 /* Following are the cases which match a character. These end
4686 with `break'. */
4687
4688#ifdef WCHAR
4689 case exactn:
4690 fastmap[truncate_wchar(p[1])] = 1;
4691 break;
4692#else /* BYTE */
4693 case exactn:
4694 fastmap[p[1]] = 1;
4695 break;
4696#endif /* WCHAR */
4697#ifdef MBS_SUPPORT
4698 case exactn_bin:
4699 fastmap[p[1]] = 1;
4700 break;
4701#endif
4702
4703#ifdef WCHAR
4704 /* It is hard to distinguish fastmap from (multi byte) characters
4705 which depends on current locale. */
4706 case charset:
4707 case charset_not:
4708 case wordchar:
4709 case notwordchar:
4710 bufp->can_be_null = 1;
4711 goto done;
4712#else /* BYTE */
4713 case charset:
4714 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4715 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4716 fastmap[j] = 1;
4717 break;
4718
4719
4720 case charset_not:
4721 /* Chars beyond end of map must be allowed. */
4722 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4723 fastmap[j] = 1;
4724
4725 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4726 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4727 fastmap[j] = 1;
4728 break;
4729
4730
4731 case wordchar:
4732 for (j = 0; j < (1 << BYTEWIDTH); j++)
4733 if (SYNTAX (j) == Sword)
4734 fastmap[j] = 1;
4735 break;
4736
4737
4738 case notwordchar:
4739 for (j = 0; j < (1 << BYTEWIDTH); j++)
4740 if (SYNTAX (j) != Sword)
4741 fastmap[j] = 1;
4742 break;
4743#endif /* WCHAR */
4744
4745 case anychar:
4746 {
4747 int fastmap_newline = fastmap['\n'];
4748
4749 /* `.' matches anything ... */
4750 for (j = 0; j < (1 << BYTEWIDTH); j++)
4751 fastmap[j] = 1;
4752
4753 /* ... except perhaps newline. */
4754 if (!(bufp->syntax & RE_DOT_NEWLINE))
4755 fastmap['\n'] = fastmap_newline;
4756
4757 /* Return if we have already set `can_be_null'; if we have,
4758 then the fastmap is irrelevant. Something's wrong here. */
4759 else if (bufp->can_be_null)
4760 goto done;
4761
4762 /* Otherwise, have to check alternative paths. */
4763 break;
4764 }
4765
4766#ifdef emacs
4767 case syntaxspec:
4768 k = *p++;
4769 for (j = 0; j < (1 << BYTEWIDTH); j++)
4770 if (SYNTAX (j) == (enum syntaxcode) k)
4771 fastmap[j] = 1;
4772 break;
4773
4774
4775 case notsyntaxspec:
4776 k = *p++;
4777 for (j = 0; j < (1 << BYTEWIDTH); j++)
4778 if (SYNTAX (j) != (enum syntaxcode) k)
4779 fastmap[j] = 1;
4780 break;
4781
4782
4783 /* All cases after this match the empty string. These end with
4784 `continue'. */
4785
4786
4787 case before_dot:
4788 case at_dot:
4789 case after_dot:
4790 continue;
4791#endif /* emacs */
4792
4793
4794 case no_op:
4795 case begline:
4796 case endline:
4797 case begbuf:
4798 case endbuf:
4799 case wordbound:
4800 case notwordbound:
4801 case wordbeg:
4802 case wordend:
4803 case push_dummy_failure:
4804 continue;
4805
4806
4807 case jump_n:
4808 case pop_failure_jump:
4809 case maybe_pop_jump:
4810 case jump:
4811 case jump_past_alt:
4812 case dummy_failure_jump:
4813 EXTRACT_NUMBER_AND_INCR (j, p);
4814 p += j;
4815 if (j > 0)
4816 continue;
4817
4818 /* Jump backward implies we just went through the body of a
4819 loop and matched nothing. Opcode jumped to should be
4820 `on_failure_jump' or `succeed_n'. Just treat it like an
4821 ordinary jump. For a * loop, it has pushed its failure
4822 point already; if so, discard that as redundant. */
4823 if ((re_opcode_t) *p != on_failure_jump
4824 && (re_opcode_t) *p != succeed_n)
4825 continue;
4826
4827 p++;
4828 EXTRACT_NUMBER_AND_INCR (j, p);
4829 p += j;
4830
4831 /* If what's on the stack is where we are now, pop it. */
4832 if (!FAIL_STACK_EMPTY ()
4833 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
4834 fail_stack.avail--;
4835
4836 continue;
4837
4838
4839 case on_failure_jump:
4840 case on_failure_keep_string_jump:
4841 handle_on_failure_jump:
4842 EXTRACT_NUMBER_AND_INCR (j, p);
4843
4844 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4845 end of the pattern. We don't want to push such a point,
4846 since when we restore it above, entering the switch will
4847 increment `p' past the end of the pattern. We don't need
4848 to push such a point since we obviously won't find any more
4849 fastmap entries beyond `pend'. Such a pattern can match
4850 the null string, though. */
4851 if (p + j < pend)
4852 {
4853 if (!PUSH_PATTERN_OP (p + j, fail_stack))
4854 {
4855 RESET_FAIL_STACK ();
4856 return -2;
4857 }
4858 }
4859 else
4860 bufp->can_be_null = 1;
4861
4862 if (succeed_n_p)
4863 {
4864 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4865 succeed_n_p = false;
4866 }
4867
4868 continue;
4869
4870
4871 case succeed_n:
4872 /* Get to the number of times to succeed. */
4873 p += OFFSET_ADDRESS_SIZE;
4874
4875 /* Increment p past the n for when k != 0. */
4876 EXTRACT_NUMBER_AND_INCR (k, p);
4877 if (k == 0)
4878 {
4879 p -= 2 * OFFSET_ADDRESS_SIZE;
4880 succeed_n_p = true; /* Spaghetti code alert. */
4881 goto handle_on_failure_jump;
4882 }
4883 continue;
4884
4885
4886 case set_number_at:
4887 p += 2 * OFFSET_ADDRESS_SIZE;
4888 continue;
4889
4890
4891 case start_memory:
4892 case stop_memory:
4893 p += 2;
4894 continue;
4895
4896
4897 default:
4898 abort (); /* We have listed all the cases. */
4899 } /* switch *p++ */
4900
4901 /* Getting here means we have found the possible starting
4902 characters for one path of the pattern -- and that the empty
4903 string does not match. We need not follow this path further.
4904 Instead, look at the next alternative (remembered on the
4905 stack), or quit if no more. The test at the top of the loop
4906 does these things. */
4907 path_can_be_null = false;
4908 p = pend;
4909 } /* while p */
4910
4911 /* Set `can_be_null' for the last path (also the first path, if the
4912 pattern is empty). */
4913 bufp->can_be_null |= path_can_be_null;
4914
4915 done:
4916 RESET_FAIL_STACK ();
4917 return 0;
4918}
4919
4920#else /* not INSIDE_RECURSION */
4921
4922int
4923re_compile_fastmap (bufp)
4924 struct re_pattern_buffer *bufp;
4925{
4926# ifdef MBS_SUPPORT
4927 if (MB_CUR_MAX != 1)
4928 return wcs_re_compile_fastmap(bufp);
4929 else
4930# endif
4931 return byte_re_compile_fastmap(bufp);
4932} /* re_compile_fastmap */
4933#ifdef _LIBC
4934weak_alias (__re_compile_fastmap, re_compile_fastmap)
4935#endif
4936\f
4937
4938/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4939 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4940 this memory for recording register information. STARTS and ENDS
4941 must be allocated using the malloc library routine, and must each
4942 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4943
4944 If NUM_REGS == 0, then subsequent matches should allocate their own
4945 register data.
4946
4947 Unless this function is called, the first search or match using
4948 PATTERN_BUFFER will allocate its own register data, without
4949 freeing the old data. */
4950
4951void
4952re_set_registers (bufp, regs, num_regs, starts, ends)
4953 struct re_pattern_buffer *bufp;
4954 struct re_registers *regs;
4955 unsigned num_regs;
4956 regoff_t *starts, *ends;
4957{
4958 if (num_regs)
4959 {
4960 bufp->regs_allocated = REGS_REALLOCATE;
4961 regs->num_regs = num_regs;
4962 regs->start = starts;
4963 regs->end = ends;
4964 }
4965 else
4966 {
4967 bufp->regs_allocated = REGS_UNALLOCATED;
4968 regs->num_regs = 0;
4969 regs->start = regs->end = (regoff_t *) 0;
4970 }
4971}
4972#ifdef _LIBC
4973weak_alias (__re_set_registers, re_set_registers)
4974#endif
4975\f
4976/* Searching routines. */
4977
4978/* Like re_search_2, below, but only one string is specified, and
4979 doesn't let you say where to stop matching. */
4980
4981int
4982re_search (bufp, string, size, startpos, range, regs)
4983 struct re_pattern_buffer *bufp;
4984 const char *string;
4985 int size, startpos, range;
4986 struct re_registers *regs;
4987{
4988 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4989 regs, size);
4990}
4991#ifdef _LIBC
4992weak_alias (__re_search, re_search)
4993#endif
4994
4995
4996/* Using the compiled pattern in BUFP->buffer, first tries to match the
4997 virtual concatenation of STRING1 and STRING2, starting first at index
4998 STARTPOS, then at STARTPOS + 1, and so on.
4999
5000 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5001
5002 RANGE is how far to scan while trying to match. RANGE = 0 means try
5003 only at STARTPOS; in general, the last start tried is STARTPOS +
5004 RANGE.
5005
5006 In REGS, return the indices of the virtual concatenation of STRING1
5007 and STRING2 that matched the entire BUFP->buffer and its contained
5008 subexpressions.
5009
5010 Do not consider matching one past the index STOP in the virtual
5011 concatenation of STRING1 and STRING2.
5012
5013 We return either the position in the strings at which the match was
5014 found, -1 if no match, or -2 if error (such as failure
5015 stack overflow). */
5016
5017int
5018re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
5019 struct re_pattern_buffer *bufp;
5020 const char *string1, *string2;
5021 int size1, size2;
5022 int startpos;
5023 int range;
5024 struct re_registers *regs;
5025 int stop;
5026{
5027# ifdef MBS_SUPPORT
5028 if (MB_CUR_MAX != 1)
5029 return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5030 range, regs, stop);
5031 else
5032# endif
5033 return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5034 range, regs, stop);
5035} /* re_search_2 */
5036#ifdef _LIBC
5037weak_alias (__re_search_2, re_search_2)
5038#endif
5039
5040#endif /* not INSIDE_RECURSION */
5041
5042#ifdef INSIDE_RECURSION
5043
5044#ifdef MATCH_MAY_ALLOCATE
5045# define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5046#else
5047# define FREE_VAR(var) if (var) free (var); var = NULL
5048#endif
5049
5050#ifdef WCHAR
86710ce2
DD		 5051# define MAX_ALLOCA_SIZE 2000
5052
5053# define FREE_WCS_BUFFERS() \
5054 do { \
5055 if (size1 > MAX_ALLOCA_SIZE) \
5056 { \
5057 free (wcs_string1); \
5058 free (mbs_offset1); \
5059 } \
5060 else \
5061 { \
5062 FREE_VAR (wcs_string1); \
5063 FREE_VAR (mbs_offset1); \
5064 } \
5065 if (size2 > MAX_ALLOCA_SIZE) \
5066 { \
5067 free (wcs_string2); \
5068 free (mbs_offset2); \
5069 } \
5070 else \
5071 { \
5072 FREE_VAR (wcs_string2); \
5073 FREE_VAR (mbs_offset2); \
5074 } \
2a6ef469
DD		 5075 } while (0)
5076
5077#endif
5078
86710ce2 5079
2a6ef469
DD		 5080static int
5081PREFIX(re_search_2) (bufp, string1, size1, string2, size2, startpos, range,
5082 regs, stop)
5083 struct re_pattern_buffer *bufp;
5084 const char *string1, *string2;
5085 int size1, size2;
5086 int startpos;
5087 int range;
5088 struct re_registers *regs;
5089 int stop;
5090{
5091 int val;
5092 register char *fastmap = bufp->fastmap;
5093 register RE_TRANSLATE_TYPE translate = bufp->translate;
5094 int total_size = size1 + size2;
5095 int endpos = startpos + range;
5096#ifdef WCHAR
5097 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5098 wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL;
5099 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5100 int wcs_size1 = 0, wcs_size2 = 0;
5101 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5102 int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5103 /* They hold whether each wchar_t is binary data or not. */
5104 char *is_binary = NULL;
5105#endif /* WCHAR */
5106
5107 /* Check for out-of-range STARTPOS. */
5108 if (startpos < 0 || startpos > total_size)
5109 return -1;
5110
5111 /* Fix up RANGE if it might eventually take us outside
5112 the virtual concatenation of STRING1 and STRING2.
5113 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5114 if (endpos < 0)
5115 range = 0 - startpos;
5116 else if (endpos > total_size)
5117 range = total_size - startpos;
5118
5119 /* If the search isn't to be a backwards one, don't waste time in a
5120 search for a pattern that must be anchored. */
5121 if (bufp->used > 0 && range > 0
5122 && ((re_opcode_t) bufp->buffer[0] == begbuf
5123 /* `begline' is like `begbuf' if it cannot match at newlines. */
5124 || ((re_opcode_t) bufp->buffer[0] == begline
5125 && !bufp->newline_anchor)))
5126 {
5127 if (startpos > 0)
5128 return -1;
5129 else
5130 range = 1;
5131 }
5132
5133#ifdef emacs
5134 /* In a forward search for something that starts with \=.
5135 don't keep searching past point. */
5136 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
5137 {
5138 range = PT - startpos;
5139 if (range <= 0)
5140 return -1;
5141 }
5142#endif /* emacs */
5143
5144 /* Update the fastmap now if not correct already. */
5145 if (fastmap && !bufp->fastmap_accurate)
5146 if (re_compile_fastmap (bufp) == -2)
5147 return -2;
5148
5149#ifdef WCHAR
5150 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5151 fill them with converted string. */
5152 if (size1 != 0)
5153 {
86710ce2
DD		 5154 if (size1 > MAX_ALLOCA_SIZE)
5155 {
5156 wcs_string1 = TALLOC (size1 + 1, CHAR_T);
5157 mbs_offset1 = TALLOC (size1 + 1, int);
5158 is_binary = TALLOC (size1 + 1, char);
5159 }
5160 else
5161 {
5162 wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T);
5163 mbs_offset1 = REGEX_TALLOC (size1 + 1, int);
5164 is_binary = REGEX_TALLOC (size1 + 1, char);
5165 }
2a6ef469
DD		 5166 if (!wcs_string1 || !mbs_offset1 || !is_binary)
5167 {
86710ce2
DD		 5168 if (size1 > MAX_ALLOCA_SIZE)
5169 {
5170 free (wcs_string1);
5171 free (mbs_offset1);
5172 free (is_binary);
5173 }
5174 else
5175 {
5176 FREE_VAR (wcs_string1);
5177 FREE_VAR (mbs_offset1);
5178 FREE_VAR (is_binary);
5179 }
2a6ef469
DD		 5180 return -2;
5181 }
5182 wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1,
5183 mbs_offset1, is_binary);
5184 wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */
86710ce2
DD		 5185 if (size1 > MAX_ALLOCA_SIZE)
5186 free (is_binary);
5187 else
5188 FREE_VAR (is_binary);
2a6ef469
DD		 5189 }
5190 if (size2 != 0)
5191 {
86710ce2
DD		 5192 if (size2 > MAX_ALLOCA_SIZE)
5193 {
5194 wcs_string2 = TALLOC (size2 + 1, CHAR_T);
5195 mbs_offset2 = TALLOC (size2 + 1, int);
5196 is_binary = TALLOC (size2 + 1, char);
5197 }
5198 else
5199 {
5200 wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T);
5201 mbs_offset2 = REGEX_TALLOC (size2 + 1, int);
5202 is_binary = REGEX_TALLOC (size2 + 1, char);
5203 }
2a6ef469
DD		 5204 if (!wcs_string2 || !mbs_offset2 || !is_binary)
5205 {
5206 FREE_WCS_BUFFERS ();
86710ce2
DD		 5207 if (size2 > MAX_ALLOCA_SIZE)
5208 free (is_binary);
5209 else
5210 FREE_VAR (is_binary);
2a6ef469
DD		 5211 return -2;
5212 }
5213 wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2,
5214 mbs_offset2, is_binary);
5215 wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */
86710ce2
DD		 5216 if (size2 > MAX_ALLOCA_SIZE)
5217 free (is_binary);
5218 else
5219 FREE_VAR (is_binary);
2a6ef469
DD		 5220 }
5221#endif /* WCHAR */
5222
5223
5224 /* Loop through the string, looking for a place to start matching. */
5225 for (;;)
5226 {
5227 /* If a fastmap is supplied, skip quickly over characters that
5228 cannot be the start of a match. If the pattern can match the
5229 null string, however, we don't need to skip characters; we want
5230 the first null string. */
5231 if (fastmap && startpos < total_size && !bufp->can_be_null)
5232 {
5233 if (range > 0) /* Searching forwards. */
5234 {
5235 register const char *d;
5236 register int lim = 0;
5237 int irange = range;
5238
5239 if (startpos < size1 && startpos + range >= size1)
5240 lim = range - (size1 - startpos);
5241
5242 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5243
5244 /* Written out as an if-else to avoid testing `translate'
5245 inside the loop. */
5246 if (translate)
5247 while (range > lim
5248 && !fastmap[(unsigned char)
5249 translate[(unsigned char) *d++]])
5250 range--;
5251 else
5252 while (range > lim && !fastmap[(unsigned char) *d++])
5253 range--;
5254
5255 startpos += irange - range;
5256 }
5257 else /* Searching backwards. */
5258 {
5259 register CHAR_T c = (size1 == 0 || startpos >= size1
5260 ? string2[startpos - size1]
5261 : string1[startpos]);
5262
5263 if (!fastmap[(unsigned char) TRANSLATE (c)])
5264 goto advance;
5265 }
5266 }
5267
5268 /* If can't match the null string, and that's all we have left, fail. */
5269 if (range >= 0 && startpos == total_size && fastmap
5270 && !bufp->can_be_null)
5271 {
5272#ifdef WCHAR
5273 FREE_WCS_BUFFERS ();
5274#endif
5275 return -1;
5276 }
5277
5278#ifdef WCHAR
5279 val = wcs_re_match_2_internal (bufp, string1, size1, string2,
5280 size2, startpos, regs, stop,
5281 wcs_string1, wcs_size1,
5282 wcs_string2, wcs_size2,
5283 mbs_offset1, mbs_offset2);
5284#else /* BYTE */
5285 val = byte_re_match_2_internal (bufp, string1, size1, string2,
5286 size2, startpos, regs, stop);
5287#endif /* BYTE */
5288
5289#ifndef REGEX_MALLOC
5290# ifdef C_ALLOCA
5291 alloca (0);
5292# endif
5293#endif
5294
5295 if (val >= 0)
5296 {
5297#ifdef WCHAR
5298 FREE_WCS_BUFFERS ();
5299#endif
5300 return startpos;
5301 }
5302
5303 if (val == -2)
5304 {
5305#ifdef WCHAR
5306 FREE_WCS_BUFFERS ();
5307#endif
5308 return -2;
5309 }
5310
5311 advance:
5312 if (!range)
5313 break;
5314 else if (range > 0)
5315 {
5316 range--;
5317 startpos++;
5318 }
5319 else
5320 {
5321 range++;
5322 startpos--;
5323 }
5324 }
5325#ifdef WCHAR
5326 FREE_WCS_BUFFERS ();
5327#endif
5328 return -1;
5329}
5330
5331#ifdef WCHAR
5332/* This converts PTR, a pointer into one of the search wchar_t strings
5333 `string1' and `string2' into an multibyte string offset from the
5334 beginning of that string. We use mbs_offset to optimize.
5335 See convert_mbs_to_wcs. */
5336# define POINTER_TO_OFFSET(ptr) \
5337 (FIRST_STRING_P (ptr) \
5338 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5339 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5340 + csize1)))
5341#else /* BYTE */
5342/* This converts PTR, a pointer into one of the search strings `string1'
5343 and `string2' into an offset from the beginning of that string. */
5344# define POINTER_TO_OFFSET(ptr) \
5345 (FIRST_STRING_P (ptr) \
5346 ? ((regoff_t) ((ptr) - string1)) \
5347 : ((regoff_t) ((ptr) - string2 + size1)))
5348#endif /* WCHAR */
5349
5350/* Macros for dealing with the split strings in re_match_2. */
5351
5352#define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5353
5354/* Call before fetching a character with *d. This switches over to
5355 string2 if necessary. */
5356#define PREFETCH() \
5357 while (d == dend) \
5358 { \
5359 /* End of string2 => fail. */ \
5360 if (dend == end_match_2) \
5361 goto fail; \
5362 /* End of string1 => advance to string2. */ \
5363 d = string2; \
5364 dend = end_match_2; \
5365 }
5366
5367/* Test if at very beginning or at very end of the virtual concatenation
5368 of `string1' and `string2'. If only one string, it's `string2'. */
5369#define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5370#define AT_STRINGS_END(d) ((d) == end2)
5371
5372
5373/* Test if D points to a character which is word-constituent. We have
5374 two special cases to check for: if past the end of string1, look at
5375 the first character in string2; and if before the beginning of
5376 string2, look at the last character in string1. */
5377#ifdef WCHAR
5378/* Use internationalized API instead of SYNTAX. */
5379# define WORDCHAR_P(d) \
5380 (iswalnum ((wint_t)((d) == end1 ? *string2 \
6ad8a379
DD		 5381 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \
5382 || ((d) == end1 ? *string2 \
5383 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_')
2a6ef469
DD		 5384#else /* BYTE */
5385# define WORDCHAR_P(d) \
5386 (SYNTAX ((d) == end1 ? *string2 \
5387 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5388 == Sword)
5389#endif /* WCHAR */
5390
5391/* Disabled due to a compiler bug -- see comment at case wordbound */
5392#if 0
5393/* Test if the character before D and the one at D differ with respect
5394 to being word-constituent. */
5395#define AT_WORD_BOUNDARY(d) \
5396 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5397 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5398#endif
5399
5400/* Free everything we malloc. */
5401#ifdef MATCH_MAY_ALLOCATE
5402# ifdef WCHAR
5403# define FREE_VARIABLES() \
5404 do { \
5405 REGEX_FREE_STACK (fail_stack.stack); \
5406 FREE_VAR (regstart); \
5407 FREE_VAR (regend); \
5408 FREE_VAR (old_regstart); \
5409 FREE_VAR (old_regend); \
5410 FREE_VAR (best_regstart); \
5411 FREE_VAR (best_regend); \
5412 FREE_VAR (reg_info); \
5413 FREE_VAR (reg_dummy); \
5414 FREE_VAR (reg_info_dummy); \
5415 if (!cant_free_wcs_buf) \
5416 { \
5417 FREE_VAR (string1); \
5418 FREE_VAR (string2); \
5419 FREE_VAR (mbs_offset1); \
5420 FREE_VAR (mbs_offset2); \
5421 } \
5422 } while (0)
5423# else /* BYTE */
5424# define FREE_VARIABLES() \
5425 do { \
5426 REGEX_FREE_STACK (fail_stack.stack); \
5427 FREE_VAR (regstart); \
5428 FREE_VAR (regend); \
5429 FREE_VAR (old_regstart); \
5430 FREE_VAR (old_regend); \
5431 FREE_VAR (best_regstart); \
5432 FREE_VAR (best_regend); \
5433 FREE_VAR (reg_info); \
5434 FREE_VAR (reg_dummy); \
5435 FREE_VAR (reg_info_dummy); \
5436 } while (0)
5437# endif /* WCHAR */
5438#else
5439# ifdef WCHAR
5440# define FREE_VARIABLES() \
5441 do { \
5442 if (!cant_free_wcs_buf) \
5443 { \
5444 FREE_VAR (string1); \
5445 FREE_VAR (string2); \
5446 FREE_VAR (mbs_offset1); \
5447 FREE_VAR (mbs_offset2); \
5448 } \
5449 } while (0)
5450# else /* BYTE */
5451# define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5452# endif /* WCHAR */
5453#endif /* not MATCH_MAY_ALLOCATE */
5454
5455/* These values must meet several constraints. They must not be valid
5456 register values; since we have a limit of 255 registers (because
5457 we use only one byte in the pattern for the register number), we can
5458 use numbers larger than 255. They must differ by 1, because of
5459 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5460 be larger than the value for the highest register, so we do not try
5461 to actually save any registers when none are active. */
5462#define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5463#define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5464\f
5465#else /* not INSIDE_RECURSION */
5466/* Matching routines. */
5467
5468#ifndef emacs /* Emacs never uses this. */
5469/* re_match is like re_match_2 except it takes only a single string. */
5470
5471int
5472re_match (bufp, string, size, pos, regs)
5473 struct re_pattern_buffer *bufp;
5474 const char *string;
5475 int size, pos;
5476 struct re_registers *regs;
5477{
5478 int result;
5479# ifdef MBS_SUPPORT
5480 if (MB_CUR_MAX != 1)
5481 result = wcs_re_match_2_internal (bufp, NULL, 0, string, size,
5482 pos, regs, size,
5483 NULL, 0, NULL, 0, NULL, NULL);
5484 else
5485# endif
5486 result = byte_re_match_2_internal (bufp, NULL, 0, string, size,
5487 pos, regs, size);
5488# ifndef REGEX_MALLOC
5489# ifdef C_ALLOCA
5490 alloca (0);
5491# endif
5492# endif
5493 return result;
5494}
5495# ifdef _LIBC
5496weak_alias (__re_match, re_match)
5497# endif
5498#endif /* not emacs */
5499
5500#endif /* not INSIDE_RECURSION */
5501
5502#ifdef INSIDE_RECURSION
5503static boolean PREFIX(group_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5504 UCHAR_T *end,
5505 PREFIX(register_info_type) *reg_info));
5506static boolean PREFIX(alt_match_null_string_p) _RE_ARGS ((UCHAR_T *p,
5507 UCHAR_T *end,
5508 PREFIX(register_info_type) *reg_info));
5509static boolean PREFIX(common_op_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5510 UCHAR_T *end,
5511 PREFIX(register_info_type) *reg_info));
5512static int PREFIX(bcmp_translate) _RE_ARGS ((const CHAR_T *s1, const CHAR_T *s2,
5513 int len, char *translate));
5514#else /* not INSIDE_RECURSION */
5515
5516/* re_match_2 matches the compiled pattern in BUFP against the
5517 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5518 and SIZE2, respectively). We start matching at POS, and stop
5519 matching at STOP.
5520
5521 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5522 store offsets for the substring each group matched in REGS. See the
5523 documentation for exactly how many groups we fill.
5524
5525 We return -1 if no match, -2 if an internal error (such as the
5526 failure stack overflowing). Otherwise, we return the length of the
5527 matched substring. */
5528
5529int
5530re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5531 struct re_pattern_buffer *bufp;
5532 const char *string1, *string2;
5533 int size1, size2;
5534 int pos;
5535 struct re_registers *regs;
5536 int stop;
5537{
5538 int result;
5539# ifdef MBS_SUPPORT
5540 if (MB_CUR_MAX != 1)
5541 result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2,
5542 pos, regs, stop,
5543 NULL, 0, NULL, 0, NULL, NULL);
5544 else
5545# endif
5546 result = byte_re_match_2_internal (bufp, string1, size1, string2, size2,
5547 pos, regs, stop);
5548
5549#ifndef REGEX_MALLOC
5550# ifdef C_ALLOCA
5551 alloca (0);
5552# endif
5553#endif
5554 return result;
5555}
5556#ifdef _LIBC
5557weak_alias (__re_match_2, re_match_2)
5558#endif
5559
5560#endif /* not INSIDE_RECURSION */
5561
5562#ifdef INSIDE_RECURSION
5563
5564#ifdef WCHAR
5565static int count_mbs_length PARAMS ((int *, int));
5566
5567/* This check the substring (from 0, to length) of the multibyte string,
5568 to which offset_buffer correspond. And count how many wchar_t_characters
5569 the substring occupy. We use offset_buffer to optimization.
5570 See convert_mbs_to_wcs. */
5571
5572static int
5573count_mbs_length(offset_buffer, length)
5574 int *offset_buffer;
5575 int length;
5576{
5577 int upper, lower;
5578
5579 /* Check whether the size is valid. */
5580 if (length < 0)
5581 return -1;
5582
5583 if (offset_buffer == NULL)
5584 return 0;
5585
5586 /* If there are no multibyte character, offset_buffer[i] == i.
5587 Optmize for this case. */
5588 if (offset_buffer[length] == length)
5589 return length;
5590
5591 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5592 upper = length;
5593 lower = 0;
5594
5595 while (true)
5596 {
5597 int middle = (lower + upper) / 2;
5598 if (middle == lower || middle == upper)
5599 break;
5600 if (offset_buffer[middle] > length)
5601 upper = middle;
5602 else if (offset_buffer[middle] < length)
5603 lower = middle;
5604 else
5605 return middle;
5606 }
5607
5608 return -1;
5609}
5610#endif /* WCHAR */
5611
5612/* This is a separate function so that we can force an alloca cleanup
5613 afterwards. */
5614#ifdef WCHAR
5615static int
5616wcs_re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos,
5617 regs, stop, string1, size1, string2, size2,
5618 mbs_offset1, mbs_offset2)
5619 struct re_pattern_buffer *bufp;
5620 const char *cstring1, *cstring2;
5621 int csize1, csize2;
5622 int pos;
5623 struct re_registers *regs;
5624 int stop;
5625 /* string1 == string2 == NULL means string1/2, size1/2 and
5626 mbs_offset1/2 need seting up in this function. */
5627 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5628 wchar_t *string1, *string2;
5629 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5630 int size1, size2;
5631 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5632 int *mbs_offset1, *mbs_offset2;
5633#else /* BYTE */
5634static int
5635byte_re_match_2_internal (bufp, string1, size1,string2, size2, pos,
5636 regs, stop)
5637 struct re_pattern_buffer *bufp;
5638 const char *string1, *string2;
5639 int size1, size2;
5640 int pos;
5641 struct re_registers *regs;
5642 int stop;
5643#endif /* BYTE */
5644{
5645 /* General temporaries. */
5646 int mcnt;
5647 UCHAR_T *p1;
5648#ifdef WCHAR
5649 /* They hold whether each wchar_t is binary data or not. */
5650 char *is_binary = NULL;
5651 /* If true, we can't free string1/2, mbs_offset1/2. */
5652 int cant_free_wcs_buf = 1;
5653#endif /* WCHAR */
5654
5655 /* Just past the end of the corresponding string. */
5656 const CHAR_T *end1, *end2;
5657
5658 /* Pointers into string1 and string2, just past the last characters in
5659 each to consider matching. */
5660 const CHAR_T *end_match_1, *end_match_2;
5661
5662 /* Where we are in the data, and the end of the current string. */
5663 const CHAR_T *d, *dend;
5664
5665 /* Where we are in the pattern, and the end of the pattern. */
5666#ifdef WCHAR
5667 UCHAR_T *pattern, *p;
5668 register UCHAR_T *pend;
5669#else /* BYTE */
5670 UCHAR_T *p = bufp->buffer;
5671 register UCHAR_T *pend = p + bufp->used;
5672#endif /* WCHAR */
5673
5674 /* Mark the opcode just after a start_memory, so we can test for an
5675 empty subpattern when we get to the stop_memory. */
5676 UCHAR_T *just_past_start_mem = 0;
5677
5678 /* We use this to map every character in the string. */
5679 RE_TRANSLATE_TYPE translate = bufp->translate;
5680
5681 /* Failure point stack. Each place that can handle a failure further
5682 down the line pushes a failure point on this stack. It consists of
5683 restart, regend, and reg_info for all registers corresponding to
5684 the subexpressions we're currently inside, plus the number of such
5685 registers, and, finally, two char *'s. The first char * is where
5686 to resume scanning the pattern; the second one is where to resume
5687 scanning the strings. If the latter is zero, the failure point is
5688 a ``dummy''; if a failure happens and the failure point is a dummy,
5689 it gets discarded and the next next one is tried. */
5690#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5691 PREFIX(fail_stack_type) fail_stack;
5692#endif
5693#ifdef DEBUG
5694 static unsigned failure_id;
5695 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5696#endif
5697
5698#ifdef REL_ALLOC
5699 /* This holds the pointer to the failure stack, when
5700 it is allocated relocatably. */
5701 fail_stack_elt_t *failure_stack_ptr;
5702#endif
5703
5704 /* We fill all the registers internally, independent of what we
5705 return, for use in backreferences. The number here includes
5706 an element for register zero. */
5707 size_t num_regs = bufp->re_nsub + 1;
5708
5709 /* The currently active registers. */
5710 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5711 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5712
5713 /* Information on the contents of registers. These are pointers into
5714 the input strings; they record just what was matched (on this
5715 attempt) by a subexpression part of the pattern, that is, the
5716 regnum-th regstart pointer points to where in the pattern we began
5717 matching and the regnum-th regend points to right after where we
5718 stopped matching the regnum-th subexpression. (The zeroth register
5719 keeps track of what the whole pattern matches.) */
5720#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5721 const CHAR_T **regstart, **regend;
5722#endif
5723
5724 /* If a group that's operated upon by a repetition operator fails to
5725 match anything, then the register for its start will need to be
5726 restored because it will have been set to wherever in the string we
5727 are when we last see its open-group operator. Similarly for a
5728 register's end. */
5729#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5730 const CHAR_T **old_regstart, **old_regend;
5731#endif
5732
5733 /* The is_active field of reg_info helps us keep track of which (possibly
5734 nested) subexpressions we are currently in. The matched_something
5735 field of reg_info[reg_num] helps us tell whether or not we have
5736 matched any of the pattern so far this time through the reg_num-th
5737 subexpression. These two fields get reset each time through any
5738 loop their register is in. */
5739#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5740 PREFIX(register_info_type) *reg_info;
5741#endif
5742
5743 /* The following record the register info as found in the above
5744 variables when we find a match better than any we've seen before.
5745 This happens as we backtrack through the failure points, which in
5746 turn happens only if we have not yet matched the entire string. */
5747 unsigned best_regs_set = false;
5748#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5749 const CHAR_T **best_regstart, **best_regend;
5750#endif
5751
5752 /* Logically, this is `best_regend[0]'. But we don't want to have to
5753 allocate space for that if we're not allocating space for anything
5754 else (see below). Also, we never need info about register 0 for
5755 any of the other register vectors, and it seems rather a kludge to
5756 treat `best_regend' differently than the rest. So we keep track of
5757 the end of the best match so far in a separate variable. We
5758 initialize this to NULL so that when we backtrack the first time
5759 and need to test it, it's not garbage. */
5760 const CHAR_T *match_end = NULL;
5761
5762 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5763 int set_regs_matched_done = 0;
5764
5765 /* Used when we pop values we don't care about. */
5766#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5767 const CHAR_T **reg_dummy;
5768 PREFIX(register_info_type) *reg_info_dummy;
5769#endif
5770
5771#ifdef DEBUG
5772 /* Counts the total number of registers pushed. */
5773 unsigned num_regs_pushed = 0;
5774#endif
5775
5776 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5777
5778 INIT_FAIL_STACK ();
5779
5780#ifdef MATCH_MAY_ALLOCATE
5781 /* Do not bother to initialize all the register variables if there are
5782 no groups in the pattern, as it takes a fair amount of time. If
5783 there are groups, we include space for register 0 (the whole
5784 pattern), even though we never use it, since it simplifies the
5785 array indexing. We should fix this. */
5786 if (bufp->re_nsub)
5787 {
5788 regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5789 regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5790 old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5791 old_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5792 best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5793 best_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5794 reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5795 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *);
5796 reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5797
5798 if (!(regstart && regend && old_regstart && old_regend && reg_info
5799 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
5800 {
5801 FREE_VARIABLES ();
5802 return -2;
5803 }
5804 }
5805 else
5806 {
5807 /* We must initialize all our variables to NULL, so that
5808 `FREE_VARIABLES' doesn't try to free them. */
5809 regstart = regend = old_regstart = old_regend = best_regstart
5810 = best_regend = reg_dummy = NULL;
5811 reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL;
5812 }
5813#endif /* MATCH_MAY_ALLOCATE */
5814
5815 /* The starting position is bogus. */
5816#ifdef WCHAR
5817 if (pos < 0 || pos > csize1 + csize2)
5818#else /* BYTE */
5819 if (pos < 0 || pos > size1 + size2)
5820#endif
5821 {
5822 FREE_VARIABLES ();
5823 return -1;
5824 }
5825
5826#ifdef WCHAR
5827 /* Allocate wchar_t array for string1 and string2 and
5828 fill them with converted string. */
5829 if (string1 == NULL && string2 == NULL)
5830 {
5831 /* We need seting up buffers here. */
5832
5833 /* We must free wcs buffers in this function. */
5834 cant_free_wcs_buf = 0;
5835
5836 if (csize1 != 0)
5837 {
5838 string1 = REGEX_TALLOC (csize1 + 1, CHAR_T);
5839 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5840 is_binary = REGEX_TALLOC (csize1 + 1, char);
5841 if (!string1 || !mbs_offset1 || !is_binary)
5842 {
5843 FREE_VAR (string1);
5844 FREE_VAR (mbs_offset1);
5845 FREE_VAR (is_binary);
5846 return -2;
5847 }
5848 }
5849 if (csize2 != 0)
5850 {
5851 string2 = REGEX_TALLOC (csize2 + 1, CHAR_T);
5852 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5853 is_binary = REGEX_TALLOC (csize2 + 1, char);
5854 if (!string2 || !mbs_offset2 || !is_binary)
5855 {
5856 FREE_VAR (string1);
5857 FREE_VAR (mbs_offset1);
5858 FREE_VAR (string2);
5859 FREE_VAR (mbs_offset2);
5860 FREE_VAR (is_binary);
5861 return -2;
5862 }
5863 size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5864 mbs_offset2, is_binary);
5865 string2[size2] = L'\0'; /* for a sentinel */
5866 FREE_VAR (is_binary);
5867 }
5868 }
5869
5870 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5871 pattern to (char*) in regex_compile. */
5872 p = pattern = (CHAR_T*)bufp->buffer;
5873 pend = (CHAR_T*)(bufp->buffer + bufp->used);
5874
5875#endif /* WCHAR */
5876
5877 /* Initialize subexpression text positions to -1 to mark ones that no
5878 start_memory/stop_memory has been seen for. Also initialize the
5879 register information struct. */
5880 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5881 {
5882 regstart[mcnt] = regend[mcnt]
5883 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5884
5885 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
5886 IS_ACTIVE (reg_info[mcnt]) = 0;
5887 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5888 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5889 }
5890
5891 /* We move `string1' into `string2' if the latter's empty -- but not if
5892 `string1' is null. */
5893 if (size2 == 0 && string1 != NULL)
5894 {
5895 string2 = string1;
5896 size2 = size1;
5897 string1 = 0;
5898 size1 = 0;
5899#ifdef WCHAR
5900 mbs_offset2 = mbs_offset1;
5901 csize2 = csize1;
5902 mbs_offset1 = NULL;
5903 csize1 = 0;
5904#endif
5905 }
5906 end1 = string1 + size1;
5907 end2 = string2 + size2;
5908
5909 /* Compute where to stop matching, within the two strings. */
5910#ifdef WCHAR
5911 if (stop <= csize1)
5912 {
5913 mcnt = count_mbs_length(mbs_offset1, stop);
5914 end_match_1 = string1 + mcnt;
5915 end_match_2 = string2;
5916 }
5917 else
5918 {
5919 if (stop > csize1 + csize2)
5920 stop = csize1 + csize2;
5921 end_match_1 = end1;
5922 mcnt = count_mbs_length(mbs_offset2, stop-csize1);
5923 end_match_2 = string2 + mcnt;
5924 }
5925 if (mcnt < 0)
5926 { /* count_mbs_length return error. */
5927 FREE_VARIABLES ();
5928 return -1;
5929 }
5930#else
5931 if (stop <= size1)
5932 {
5933 end_match_1 = string1 + stop;
5934 end_match_2 = string2;
5935 }
5936 else
5937 {
5938 end_match_1 = end1;
5939 end_match_2 = string2 + stop - size1;
5940 }
5941#endif /* WCHAR */
5942
5943 /* `p' scans through the pattern as `d' scans through the data.
5944 `dend' is the end of the input string that `d' points within. `d'
5945 is advanced into the following input string whenever necessary, but
5946 this happens before fetching; therefore, at the beginning of the
5947 loop, `d' can be pointing at the end of a string, but it cannot
5948 equal `string2'. */
5949#ifdef WCHAR
5950 if (size1 > 0 && pos <= csize1)
5951 {
5952 mcnt = count_mbs_length(mbs_offset1, pos);
5953 d = string1 + mcnt;
5954 dend = end_match_1;
5955 }
5956 else
5957 {
5958 mcnt = count_mbs_length(mbs_offset2, pos-csize1);
5959 d = string2 + mcnt;
5960 dend = end_match_2;
5961 }
5962
5963 if (mcnt < 0)
5964 { /* count_mbs_length return error. */
5965 FREE_VARIABLES ();
5966 return -1;
5967 }
5968#else
5969 if (size1 > 0 && pos <= size1)
5970 {
5971 d = string1 + pos;
5972 dend = end_match_1;
5973 }
5974 else
5975 {
5976 d = string2 + pos - size1;
5977 dend = end_match_2;
5978 }
5979#endif /* WCHAR */
5980
5981 DEBUG_PRINT1 ("The compiled pattern is:\n");
5982 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5983 DEBUG_PRINT1 ("The string to match is: `");
5984 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5985 DEBUG_PRINT1 ("'\n");
5986
5987 /* This loops over pattern commands. It exits by returning from the
5988 function if the match is complete, or it drops through if the match
5989 fails at this starting point in the input data. */
5990 for (;;)
5991 {
5992#ifdef _LIBC
5993 DEBUG_PRINT2 ("\n%p: ", p);
5994#else
5995 DEBUG_PRINT2 ("\n0x%x: ", p);
5996#endif
5997
5998 if (p == pend)
5999 { /* End of pattern means we might have succeeded. */
6000 DEBUG_PRINT1 ("end of pattern ... ");
6001
6002 /* If we haven't matched the entire string, and we want the
6003 longest match, try backtracking. */
6004 if (d != end_match_2)
6005 {
6006 /* 1 if this match ends in the same string (string1 or string2)
6007 as the best previous match. */
6008 boolean same_str_p = (FIRST_STRING_P (match_end)
6009 == MATCHING_IN_FIRST_STRING);
6010 /* 1 if this match is the best seen so far. */
6011 boolean best_match_p;
6012
6013 /* AIX compiler got confused when this was combined
6014 with the previous declaration. */
6015 if (same_str_p)
6016 best_match_p = d > match_end;
6017 else
6018 best_match_p = !MATCHING_IN_FIRST_STRING;
6019
6020 DEBUG_PRINT1 ("backtracking.\n");
6021
6022 if (!FAIL_STACK_EMPTY ())
6023 { /* More failure points to try. */
6024
6025 /* If exceeds best match so far, save it. */
6026 if (!best_regs_set || best_match_p)
6027 {
6028 best_regs_set = true;
6029 match_end = d;
6030
6031 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
6032
6033 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6034 {
6035 best_regstart[mcnt] = regstart[mcnt];
6036 best_regend[mcnt] = regend[mcnt];
6037 }
6038 }
6039 goto fail;
6040 }
6041
6042 /* If no failure points, don't restore garbage. And if
6043 last match is real best match, don't restore second
6044 best one. */
6045 else if (best_regs_set && !best_match_p)
6046 {
6047 restore_best_regs:
6048 /* Restore best match. It may happen that `dend ==
6049 end_match_1' while the restored d is in string2.
6050 For example, the pattern `x.*y.*z' against the
6051 strings `x-' and `y-z-', if the two strings are
6052 not consecutive in memory. */
6053 DEBUG_PRINT1 ("Restoring best registers.\n");
6054
6055 d = match_end;
6056 dend = ((d >= string1 && d <= end1)
6057 ? end_match_1 : end_match_2);
6058
6059 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6060 {
6061 regstart[mcnt] = best_regstart[mcnt];
6062 regend[mcnt] = best_regend[mcnt];
6063 }
6064 }
6065 } /* d != end_match_2 */
6066
6067 succeed_label:
6068 DEBUG_PRINT1 ("Accepting match.\n");
6069 /* If caller wants register contents data back, do it. */
6070 if (regs && !bufp->no_sub)
6071 {
6072 /* Have the register data arrays been allocated? */
6073 if (bufp->regs_allocated == REGS_UNALLOCATED)
6074 { /* No. So allocate them with malloc. We need one
6075 extra element beyond `num_regs' for the `-1' marker
6076 GNU code uses. */
6077 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
6078 regs->start = TALLOC (regs->num_regs, regoff_t);
6079 regs->end = TALLOC (regs->num_regs, regoff_t);
6080 if (regs->start == NULL || regs->end == NULL)
6081 {
6082 FREE_VARIABLES ();
6083 return -2;
6084 }
6085 bufp->regs_allocated = REGS_REALLOCATE;
6086 }
6087 else if (bufp->regs_allocated == REGS_REALLOCATE)
6088 { /* Yes. If we need more elements than were already
6089 allocated, reallocate them. If we need fewer, just
6090 leave it alone. */
6091 if (regs->num_regs < num_regs + 1)
6092 {
6093 regs->num_regs = num_regs + 1;
6094 RETALLOC (regs->start, regs->num_regs, regoff_t);
6095 RETALLOC (regs->end, regs->num_regs, regoff_t);
6096 if (regs->start == NULL || regs->end == NULL)
6097 {
6098 FREE_VARIABLES ();
6099 return -2;
6100 }
6101 }
6102 }
6103 else
6104 {
6105 /* These braces fend off a "empty body in an else-statement"
6106 warning under GCC when assert expands to nothing. */
6107 assert (bufp->regs_allocated == REGS_FIXED);
6108 }
6109
6110 /* Convert the pointer data in `regstart' and `regend' to
6111 indices. Register zero has to be set differently,
6112 since we haven't kept track of any info for it. */
6113 if (regs->num_regs > 0)
6114 {
6115 regs->start[0] = pos;
6116#ifdef WCHAR
6117 if (MATCHING_IN_FIRST_STRING)
6118 regs->end[0] = mbs_offset1 != NULL ?
6119 mbs_offset1[d-string1] : 0;
6120 else
6121 regs->end[0] = csize1 + (mbs_offset2 != NULL ?
6122 mbs_offset2[d-string2] : 0);
6123#else
6124 regs->end[0] = (MATCHING_IN_FIRST_STRING
6125 ? ((regoff_t) (d - string1))
6126 : ((regoff_t) (d - string2 + size1)));
6127#endif /* WCHAR */
6128 }
6129
6130 /* Go through the first `min (num_regs, regs->num_regs)'
6131 registers, since that is all we initialized. */
6132 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
6133 mcnt++)
6134 {
6135 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
6136 regs->start[mcnt] = regs->end[mcnt] = -1;
6137 else
6138 {
6139 regs->start[mcnt]
6140 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
6141 regs->end[mcnt]
6142 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
6143 }
6144 }
6145
6146 /* If the regs structure we return has more elements than
6147 were in the pattern, set the extra elements to -1. If
6148 we (re)allocated the registers, this is the case,
6149 because we always allocate enough to have at least one
6150 -1 at the end. */
6151 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
6152 regs->start[mcnt] = regs->end[mcnt] = -1;
6153 } /* regs && !bufp->no_sub */
6154
6155 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6156 nfailure_points_pushed, nfailure_points_popped,
6157 nfailure_points_pushed - nfailure_points_popped);
6158 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
6159
6160#ifdef WCHAR
6161 if (MATCHING_IN_FIRST_STRING)
6162 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
6163 else
6164 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
6165 csize1;
6166 mcnt -= pos;
6167#else
6168 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
6169 ? string1
6170 : string2 - size1);
6171#endif /* WCHAR */
6172
6173 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
6174
6175 FREE_VARIABLES ();
6176 return mcnt;
6177 }
6178
6179 /* Otherwise match next pattern command. */
6180 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
6181 {
6182 /* Ignore these. Used to ignore the n of succeed_n's which
6183 currently have n == 0. */
6184 case no_op:
6185 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6186 break;
6187
6188 case succeed:
6189 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6190 goto succeed_label;
6191
6192 /* Match the next n pattern characters exactly. The following
6193 byte in the pattern defines n, and the n bytes after that
6194 are the characters to match. */
6195 case exactn:
6196#ifdef MBS_SUPPORT
6197 case exactn_bin:
6198#endif
6199 mcnt = *p++;
6200 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
6201
6202 /* This is written out as an if-else so we don't waste time
6203 testing `translate' inside the loop. */
6204 if (translate)
6205 {
6206 do
6207 {
6208 PREFETCH ();
6209#ifdef WCHAR
6210 if (*d <= 0xff)
6211 {
6212 if ((UCHAR_T) translate[(unsigned char) *d++]
6213 != (UCHAR_T) *p++)
6214 goto fail;
6215 }
6216 else
6217 {
6218 if (*d++ != (CHAR_T) *p++)
6219 goto fail;
6220 }
6221#else
6222 if ((UCHAR_T) translate[(unsigned char) *d++]
6223 != (UCHAR_T) *p++)
6224 goto fail;
6225#endif /* WCHAR */
6226 }
6227 while (--mcnt);
6228 }
6229 else
6230 {
6231 do
6232 {
6233 PREFETCH ();
6234 if (*d++ != (CHAR_T) *p++) goto fail;
6235 }
6236 while (--mcnt);
6237 }
6238 SET_REGS_MATCHED ();
6239 break;
6240
6241
6242 /* Match any character except possibly a newline or a null. */
6243 case anychar:
6244 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6245
6246 PREFETCH ();
6247
6248 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
6249 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
6250 goto fail;
6251
6252 SET_REGS_MATCHED ();
6253 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
6254 d++;
6255 break;
6256
6257
6258 case charset:
6259 case charset_not:
6260 {
6261 register UCHAR_T c;
6262#ifdef WCHAR
6263 unsigned int i, char_class_length, coll_symbol_length,
6264 equiv_class_length, ranges_length, chars_length, length;
6265 CHAR_T *workp, *workp2, *charset_top;
6266#define WORK_BUFFER_SIZE 128
6267 CHAR_T str_buf[WORK_BUFFER_SIZE];
6268# ifdef _LIBC
6269 uint32_t nrules;
6270# endif /* _LIBC */
6271#endif /* WCHAR */
6272 boolean not = (re_opcode_t) *(p - 1) == charset_not;
6273
6274 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6275 PREFETCH ();
6276 c = TRANSLATE (*d); /* The character to match. */
6277#ifdef WCHAR
6278# ifdef _LIBC
6279 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
6280# endif /* _LIBC */
6281 charset_top = p - 1;
6282 char_class_length = *p++;
6283 coll_symbol_length = *p++;
6284 equiv_class_length = *p++;
6285 ranges_length = *p++;
6286 chars_length = *p++;
6287 /* p points charset[6], so the address of the next instruction
6288 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6289 where l=length of char_classes, m=length of collating_symbol,
6290 n=equivalence_class, o=length of char_range,
6291 p'=length of character. */
6292 workp = p;
6293 /* Update p to indicate the next instruction. */
6294 p += char_class_length + coll_symbol_length+ equiv_class_length +
6295 2*ranges_length + chars_length;
6296
6297 /* match with char_class? */
6298 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
6299 {
6300 wctype_t wctype;
6301 uintptr_t alignedp = ((uintptr_t)workp
6302 + __alignof__(wctype_t) - 1)
6303 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6304 wctype = *((wctype_t*)alignedp);
6305 workp += CHAR_CLASS_SIZE;
dc676635
DD		 6306# ifdef _LIBC
6307 if (__iswctype((wint_t)c, wctype))
6308 goto char_set_matched;
6309# else
2a6ef469
DD		 6310 if (iswctype((wint_t)c, wctype))
6311 goto char_set_matched;
dc676635 6312# endif
2a6ef469
DD		 6313 }
6314
6315 /* match with collating_symbol? */
6316# ifdef _LIBC
6317 if (nrules != 0)
6318 {
6319 const unsigned char *extra = (const unsigned char *)
6320 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6321
6322 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6323 workp++)
6324 {
6325 int32_t *wextra;
6326 wextra = (int32_t*)(extra + *workp++);
6327 for (i = 0; i < *wextra; ++i)
6328 if (TRANSLATE(d[i]) != wextra[1 + i])
6329 break;
6330
6331 if (i == *wextra)
6332 {
6333 /* Update d, however d will be incremented at
6334 char_set_matched:, we decrement d here. */
6335 d += i - 1;
6336 goto char_set_matched;
6337 }
6338 }
6339 }
6340 else /* (nrules == 0) */
6341# endif
6342 /* If we can't look up collation data, we use wcscoll
6343 instead. */
6344 {
6345 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6346 {
6347 const CHAR_T *backup_d = d, *backup_dend = dend;
dc676635
DD		 6348# ifdef _LIBC
6349 length = __wcslen (workp);
6350# else
6351 length = wcslen (workp);
6352# endif
2a6ef469
DD		 6353
6354 /* If wcscoll(the collating symbol, whole string) > 0,
6355 any substring of the string never match with the
6356 collating symbol. */
dc676635
DD		 6357# ifdef _LIBC
6358 if (__wcscoll (workp, d) > 0)
6359# else
6360 if (wcscoll (workp, d) > 0)
6361# endif
2a6ef469
DD		 6362 {
6363 workp += length + 1;
6364 continue;
6365 }
6366
6367 /* First, we compare the collating symbol with
6368 the first character of the string.
6369 If it don't match, we add the next character to
6370 the compare buffer in turn. */
6371 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6372 {
6373 int match;
6374 if (d == dend)
6375 {
6376 if (dend == end_match_2)
6377 break;
6378 d = string2;
6379 dend = end_match_2;
6380 }
6381
6382 /* add next character to the compare buffer. */
6383 str_buf[i] = TRANSLATE(*d);
6384 str_buf[i+1] = '\0';
6385
dc676635
DD		 6386# ifdef _LIBC
6387 match = __wcscoll (workp, str_buf);
6388# else
6389 match = wcscoll (workp, str_buf);
6390# endif
2a6ef469
DD		 6391 if (match == 0)
6392 goto char_set_matched;
6393
6394 if (match < 0)
6395 /* (str_buf > workp) indicate (str_buf + X > workp),
6396 because for all X (str_buf + X > str_buf).
6397 So we don't need continue this loop. */
6398 break;
6399
6400 /* Otherwise(str_buf < workp),
6401 (str_buf+next_character) may equals (workp).
6402 So we continue this loop. */
6403 }
6404 /* not matched */
6405 d = backup_d;
6406 dend = backup_dend;
6407 workp += length + 1;
6408 }
6409 }
6410 /* match with equivalence_class? */
6411# ifdef _LIBC
6412 if (nrules != 0)
6413 {
6414 const CHAR_T *backup_d = d, *backup_dend = dend;
6415 /* Try to match the equivalence class against
6416 those known to the collate implementation. */
6417 const int32_t *table;
6418 const int32_t *weights;
6419 const int32_t *extra;
6420 const int32_t *indirect;
6421 int32_t idx, idx2;
6422 wint_t *cp;
6423 size_t len;
6424
6425 /* This #include defines a local function! */
6426# include <locale/weightwc.h>
6427
6428 table = (const int32_t *)
6429 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6430 weights = (const wint_t *)
6431 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6432 extra = (const wint_t *)
6433 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6434 indirect = (const int32_t *)
6435 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6436
6437 /* Write 1 collating element to str_buf, and
6438 get its index. */
6439 idx2 = 0;
6440
6441 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6442 {
6443 cp = (wint_t*)str_buf;
6444 if (d == dend)
6445 {
6446 if (dend == end_match_2)
6447 break;
6448 d = string2;
6449 dend = end_match_2;
6450 }
6451 str_buf[i] = TRANSLATE(*(d+i));
6452 str_buf[i+1] = '\0'; /* sentinel */
6453 idx2 = findidx ((const wint_t**)&cp);
6454 }
6455
6456 /* Update d, however d will be incremented at
6457 char_set_matched:, we decrement d here. */
6458 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6459 if (d >= dend)
6460 {
6461 if (dend == end_match_2)
6462 d = dend;
6463 else
6464 {
6465 d = string2;
6466 dend = end_match_2;
6467 }
6468 }
6469
6470 len = weights[idx2];
6471
6472 for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6473 workp++)
6474 {
6475 idx = (int32_t)*workp;
6476 /* We already checked idx != 0 in regex_compile. */
6477
6478 if (idx2 != 0 && len == weights[idx])
6479 {
6480 int cnt = 0;
6481 while (cnt < len && (weights[idx + 1 + cnt]
6482 == weights[idx2 + 1 + cnt]))
6483 ++cnt;
6484
6485 if (cnt == len)
6486 goto char_set_matched;
6487 }
6488 }
6489 /* not matched */
6490 d = backup_d;
6491 dend = backup_dend;
6492 }
6493 else /* (nrules == 0) */
6494# endif
6495 /* If we can't look up collation data, we use wcscoll
6496 instead. */
6497 {
6498 for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6499 {
6500 const CHAR_T *backup_d = d, *backup_dend = dend;
dc676635
DD		 6501# ifdef _LIBC
6502 length = __wcslen (workp);
6503# else
6504 length = wcslen (workp);
6505# endif
2a6ef469
DD		 6506
6507 /* If wcscoll(the collating symbol, whole string) > 0,
6508 any substring of the string never match with the
6509 collating symbol. */
dc676635
DD		 6510# ifdef _LIBC
6511 if (__wcscoll (workp, d) > 0)
6512# else
6513 if (wcscoll (workp, d) > 0)
6514# endif
2a6ef469
DD		 6515 {
6516 workp += length + 1;
6517 break;
6518 }
6519
6520 /* First, we compare the equivalence class with
6521 the first character of the string.
6522 If it don't match, we add the next character to
6523 the compare buffer in turn. */
6524 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6525 {
6526 int match;
6527 if (d == dend)
6528 {
6529 if (dend == end_match_2)
6530 break;
6531 d = string2;
6532 dend = end_match_2;
6533 }
6534
6535 /* add next character to the compare buffer. */
6536 str_buf[i] = TRANSLATE(*d);
6537 str_buf[i+1] = '\0';
6538
dc676635
DD		 6539# ifdef _LIBC
6540 match = __wcscoll (workp, str_buf);
6541# else
6542 match = wcscoll (workp, str_buf);
6543# endif
2a6ef469
DD		 6544
6545 if (match == 0)
6546 goto char_set_matched;
6547
6548 if (match < 0)
6549 /* (str_buf > workp) indicate (str_buf + X > workp),
6550 because for all X (str_buf + X > str_buf).
6551 So we don't need continue this loop. */
6552 break;
6553
6554 /* Otherwise(str_buf < workp),
6555 (str_buf+next_character) may equals (workp).
6556 So we continue this loop. */
6557 }
6558 /* not matched */
6559 d = backup_d;
6560 dend = backup_dend;
6561 workp += length + 1;
6562 }
6563 }
6564
6565 /* match with char_range? */
dc676635 6566# ifdef _LIBC
2a6ef469
DD		 6567 if (nrules != 0)
6568 {
6569 uint32_t collseqval;
6570 const char *collseq = (const char *)
6571 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6572
6573 collseqval = collseq_table_lookup (collseq, c);
6574
6575 for (; workp < p - chars_length ;)
6576 {
6577 uint32_t start_val, end_val;
6578
6579 /* We already compute the collation sequence value
6580 of the characters (or collating symbols). */
6581 start_val = (uint32_t) *workp++; /* range_start */
6582 end_val = (uint32_t) *workp++; /* range_end */
6583
6584 if (start_val <= collseqval && collseqval <= end_val)
6585 goto char_set_matched;
6586 }
6587 }
6588 else
dc676635 6589# endif
2a6ef469
DD		 6590 {
6591 /* We set range_start_char at str_buf[0], range_end_char
6592 at str_buf[4], and compared char at str_buf[2]. */
6593 str_buf[1] = 0;
6594 str_buf[2] = c;
6595 str_buf[3] = 0;
6596 str_buf[5] = 0;
6597 for (; workp < p - chars_length ;)
6598 {
6599 wchar_t *range_start_char, *range_end_char;
6600
6601 /* match if (range_start_char <= c <= range_end_char). */
6602
6603 /* If range_start(or end) < 0, we assume -range_start(end)
6604 is the offset of the collating symbol which is specified
6605 as the character of the range start(end). */
6606
6607 /* range_start */
6608 if (*workp < 0)
6609 range_start_char = charset_top - (*workp++);
6610 else
6611 {
6612 str_buf[0] = *workp++;
6613 range_start_char = str_buf;
6614 }
6615
6616 /* range_end */
6617 if (*workp < 0)
6618 range_end_char = charset_top - (*workp++);
6619 else
6620 {
6621 str_buf[4] = *workp++;
6622 range_end_char = str_buf + 4;
6623 }
6624
dc676635
DD		 6625# ifdef _LIBC
6626 if (__wcscoll (range_start_char, str_buf+2) <= 0
6627 && __wcscoll (str_buf+2, range_end_char) <= 0)
6628# else
6629 if (wcscoll (range_start_char, str_buf+2) <= 0
6630 && wcscoll (str_buf+2, range_end_char) <= 0)
6631# endif
2a6ef469
DD		 6632 goto char_set_matched;
6633 }
6634 }
6635
6636 /* match with char? */
6637 for (; workp < p ; workp++)
6638 if (c == *workp)
6639 goto char_set_matched;
6640
6641 not = !not;
6642
6643 char_set_matched:
6644 if (not) goto fail;
6645#else
6646 /* Cast to `unsigned' instead of `unsigned char' in case the
6647 bit list is a full 32 bytes long. */
6648 if (c < (unsigned) (*p * BYTEWIDTH)
6649 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6650 not = !not;
6651
6652 p += 1 + *p;
6653
6654 if (!not) goto fail;
6655#undef WORK_BUFFER_SIZE
6656#endif /* WCHAR */
6657 SET_REGS_MATCHED ();
6658 d++;
6659 break;
6660 }
6661
6662
6663 /* The beginning of a group is represented by start_memory.
6664 The arguments are the register number in the next byte, and the
6665 number of groups inner to this one in the next. The text
6666 matched within the group is recorded (in the internal
6667 registers data structure) under the register number. */
6668 case start_memory:
6669 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6670 (long int) *p, (long int) p[1]);
6671
6672 /* Find out if this group can match the empty string. */
6673 p1 = p; /* To send to group_match_null_string_p. */
6674
6675 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6676 REG_MATCH_NULL_STRING_P (reg_info[*p])
6677 = PREFIX(group_match_null_string_p) (&p1, pend, reg_info);
6678
6679 /* Save the position in the string where we were the last time
6680 we were at this open-group operator in case the group is
6681 operated upon by a repetition operator, e.g., with `(a*)*b'
6682 against `ab'; then we want to ignore where we are now in
6683 the string in case this attempt to match fails. */
6684 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6685 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6686 : regstart[*p];
6687 DEBUG_PRINT2 (" old_regstart: %d\n",
6688 POINTER_TO_OFFSET (old_regstart[*p]));
6689
6690 regstart[*p] = d;
6691 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6692
6693 IS_ACTIVE (reg_info[*p]) = 1;
6694 MATCHED_SOMETHING (reg_info[*p]) = 0;
6695
6696 /* Clear this whenever we change the register activity status. */
6697 set_regs_matched_done = 0;
6698
6699 /* This is the new highest active register. */
6700 highest_active_reg = *p;
6701
6702 /* If nothing was active before, this is the new lowest active
6703 register. */
6704 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6705 lowest_active_reg = *p;
6706
6707 /* Move past the register number and inner group count. */
6708 p += 2;
6709 just_past_start_mem = p;
6710
6711 break;
6712
6713
6714 /* The stop_memory opcode represents the end of a group. Its
6715 arguments are the same as start_memory's: the register
6716 number, and the number of inner groups. */
6717 case stop_memory:
6718 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6719 (long int) *p, (long int) p[1]);
6720
6721 /* We need to save the string position the last time we were at
6722 this close-group operator in case the group is operated
6723 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6724 against `aba'; then we want to ignore where we are now in
6725 the string in case this attempt to match fails. */
6726 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6727 ? REG_UNSET (regend[*p]) ? d : regend[*p]
6728 : regend[*p];
6729 DEBUG_PRINT2 (" old_regend: %d\n",
6730 POINTER_TO_OFFSET (old_regend[*p]));
6731
6732 regend[*p] = d;
6733 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6734
6735 /* This register isn't active anymore. */
6736 IS_ACTIVE (reg_info[*p]) = 0;
6737
6738 /* Clear this whenever we change the register activity status. */
6739 set_regs_matched_done = 0;
6740
6741 /* If this was the only register active, nothing is active
6742 anymore. */
6743 if (lowest_active_reg == highest_active_reg)
6744 {
6745 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6746 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6747 }
6748 else
6749 { /* We must scan for the new highest active register, since
6750 it isn't necessarily one less than now: consider
6751 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6752 new highest active register is 1. */
6753 UCHAR_T r = *p - 1;
6754 while (r > 0 && !IS_ACTIVE (reg_info[r]))
6755 r--;
6756
6757 /* If we end up at register zero, that means that we saved
6758 the registers as the result of an `on_failure_jump', not
6759 a `start_memory', and we jumped to past the innermost
6760 `stop_memory'. For example, in ((.)*) we save
6761 registers 1 and 2 as a result of the *, but when we pop
6762 back to the second ), we are at the stop_memory 1.
6763 Thus, nothing is active. */
6764 if (r == 0)
6765 {
6766 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6767 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6768 }
6769 else
6770 highest_active_reg = r;
6771 }
6772
6773 /* If just failed to match something this time around with a
6774 group that's operated on by a repetition operator, try to
6775 force exit from the ``loop'', and restore the register
6776 information for this group that we had before trying this
6777 last match. */
6778 if ((!MATCHED_SOMETHING (reg_info[*p])
6779 || just_past_start_mem == p - 1)
6780 && (p + 2) < pend)
6781 {
6782 boolean is_a_jump_n = false;
6783
6784 p1 = p + 2;
6785 mcnt = 0;
6786 switch ((re_opcode_t) *p1++)
6787 {
6788 case jump_n:
6789 is_a_jump_n = true;
6790 case pop_failure_jump:
6791 case maybe_pop_jump:
6792 case jump:
6793 case dummy_failure_jump:
6794 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6795 if (is_a_jump_n)
6796 p1 += OFFSET_ADDRESS_SIZE;
6797 break;
6798
6799 default:
6800 /* do nothing */ ;
6801 }
6802 p1 += mcnt;
6803
6804 /* If the next operation is a jump backwards in the pattern
6805 to an on_failure_jump right before the start_memory
6806 corresponding to this stop_memory, exit from the loop
6807 by forcing a failure after pushing on the stack the
6808 on_failure_jump's jump in the pattern, and d. */
6809 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6810 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6811 && p1[2+OFFSET_ADDRESS_SIZE] == *p)
6812 {
6813 /* If this group ever matched anything, then restore
6814 what its registers were before trying this last
6815 failed match, e.g., with `(a*)*b' against `ab' for
6816 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6817 against `aba' for regend[3].
6818
6819 Also restore the registers for inner groups for,
6820 e.g., `((a*)(b*))*' against `aba' (register 3 would
6821 otherwise get trashed). */
6822
6823 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6824 {
6825 unsigned r;
6826
6827 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6828
6829 /* Restore this and inner groups' (if any) registers. */
6830 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6831 r++)
6832 {
6833 regstart[r] = old_regstart[r];
6834
6835 /* xx why this test? */
6836 if (old_regend[r] >= regstart[r])
6837 regend[r] = old_regend[r];
6838 }
6839 }
6840 p1++;
6841 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6842 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6843
6844 goto fail;
6845 }
6846 }
6847
6848 /* Move past the register number and the inner group count. */
6849 p += 2;
6850 break;
6851
6852
6853 /* \<digit> has been turned into a `duplicate' command which is
6854 followed by the numeric value of <digit> as the register number. */
6855 case duplicate:
6856 {
6857 register const CHAR_T *d2, *dend2;
6858 int regno = *p++; /* Get which register to match against. */
6859 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6860
6861 /* Can't back reference a group which we've never matched. */
6862 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6863 goto fail;
6864
6865 /* Where in input to try to start matching. */
6866 d2 = regstart[regno];
6867
6868 /* Where to stop matching; if both the place to start and
6869 the place to stop matching are in the same string, then
6870 set to the place to stop, otherwise, for now have to use
6871 the end of the first string. */
6872
6873 dend2 = ((FIRST_STRING_P (regstart[regno])
6874 == FIRST_STRING_P (regend[regno]))
6875 ? regend[regno] : end_match_1);
6876 for (;;)
6877 {
6878 /* If necessary, advance to next segment in register
6879 contents. */
6880 while (d2 == dend2)
6881 {
6882 if (dend2 == end_match_2) break;
6883 if (dend2 == regend[regno]) break;
6884
6885 /* End of string1 => advance to string2. */
6886 d2 = string2;
6887 dend2 = regend[regno];
6888 }
6889 /* At end of register contents => success */
6890 if (d2 == dend2) break;
6891
6892 /* If necessary, advance to next segment in data. */
6893 PREFETCH ();
6894
6895 /* How many characters left in this segment to match. */
6896 mcnt = dend - d;
6897
6898 /* Want how many consecutive characters we can match in
6899 one shot, so, if necessary, adjust the count. */
6900 if (mcnt > dend2 - d2)
6901 mcnt = dend2 - d2;
6902
6903 /* Compare that many; failure if mismatch, else move
6904 past them. */
6905 if (translate
6906 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate)
6907 : memcmp (d, d2, mcnt*sizeof(UCHAR_T)))
6908 goto fail;
6909 d += mcnt, d2 += mcnt;
6910
6911 /* Do this because we've match some characters. */
6912 SET_REGS_MATCHED ();
6913 }
6914 }
6915 break;
6916
6917
6918 /* begline matches the empty string at the beginning of the string
6919 (unless `not_bol' is set in `bufp'), and, if
6920 `newline_anchor' is set, after newlines. */
6921 case begline:
6922 DEBUG_PRINT1 ("EXECUTING begline.\n");
6923
6924 if (AT_STRINGS_BEG (d))
6925 {
6926 if (!bufp->not_bol) break;
6927 }
6928 else if (d[-1] == '\n' && bufp->newline_anchor)
6929 {
6930 break;
6931 }
6932 /* In all other cases, we fail. */
6933 goto fail;
6934
6935
6936 /* endline is the dual of begline. */
6937 case endline:
6938 DEBUG_PRINT1 ("EXECUTING endline.\n");
6939
6940 if (AT_STRINGS_END (d))
6941 {
6942 if (!bufp->not_eol) break;
6943 }
6944
6945 /* We have to ``prefetch'' the next character. */
6946 else if ((d == end1 ? *string2 : *d) == '\n'
6947 && bufp->newline_anchor)
6948 {
6949 break;
6950 }
6951 goto fail;
6952
6953
6954 /* Match at the very beginning of the data. */
6955 case begbuf:
6956 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6957 if (AT_STRINGS_BEG (d))
6958 break;
6959 goto fail;
6960
6961
6962 /* Match at the very end of the data. */
6963 case endbuf:
6964 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
6965 if (AT_STRINGS_END (d))
6966 break;
6967 goto fail;
6968
6969
6970 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
6971 pushes NULL as the value for the string on the stack. Then
6972 `pop_failure_point' will keep the current value for the
6973 string, instead of restoring it. To see why, consider
6974 matching `foo\nbar' against `.*\n'. The .* matches the foo;
6975 then the . fails against the \n. But the next thing we want
6976 to do is match the \n against the \n; if we restored the
6977 string value, we would be back at the foo.
6978
6979 Because this is used only in specific cases, we don't need to
6980 check all the things that `on_failure_jump' does, to make
6981 sure the right things get saved on the stack. Hence we don't
6982 share its code. The only reason to push anything on the
6983 stack at all is that otherwise we would have to change
6984 `anychar's code to do something besides goto fail in this
6985 case; that seems worse than this. */
6986 case on_failure_keep_string_jump:
6987 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
6988
6989 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6990#ifdef _LIBC
6991 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
6992#else
6993 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
6994#endif
6995
6996 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
6997 break;
6998
6999
7000 /* Uses of on_failure_jump:
7001
7002 Each alternative starts with an on_failure_jump that points
7003 to the beginning of the next alternative. Each alternative
7004 except the last ends with a jump that in effect jumps past
7005 the rest of the alternatives. (They really jump to the
7006 ending jump of the following alternative, because tensioning
7007 these jumps is a hassle.)
7008
7009 Repeats start with an on_failure_jump that points past both
7010 the repetition text and either the following jump or
7011 pop_failure_jump back to this on_failure_jump. */
7012 case on_failure_jump:
7013 on_failure:
7014 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
7015
7016 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7017#ifdef _LIBC
7018 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
7019#else
7020 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
7021#endif
7022
7023 /* If this on_failure_jump comes right before a group (i.e.,
7024 the original * applied to a group), save the information
7025 for that group and all inner ones, so that if we fail back
7026 to this point, the group's information will be correct.
7027 For example, in \(a*\)*\1, we need the preceding group,
7028 and in \(zz\(a*\)b*\)\2, we need the inner group. */
7029
7030 /* We can't use `p' to check ahead because we push
7031 a failure point to `p + mcnt' after we do this. */
7032 p1 = p;
7033
7034 /* We need to skip no_op's before we look for the
7035 start_memory in case this on_failure_jump is happening as
7036 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
7037 against aba. */
7038 while (p1 < pend && (re_opcode_t) *p1 == no_op)
7039 p1++;
7040
7041 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
7042 {
7043 /* We have a new highest active register now. This will
7044 get reset at the start_memory we are about to get to,
7045 but we will have saved all the registers relevant to
7046 this repetition op, as described above. */
7047 highest_active_reg = *(p1 + 1) + *(p1 + 2);
7048 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
7049 lowest_active_reg = *(p1 + 1);
7050 }
7051
7052 DEBUG_PRINT1 (":\n");
7053 PUSH_FAILURE_POINT (p + mcnt, d, -2);
7054 break;
7055
7056
7057 /* A smart repeat ends with `maybe_pop_jump'.
7058 We change it to either `pop_failure_jump' or `jump'. */
7059 case maybe_pop_jump:
7060 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7061 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
7062 {
7063 register UCHAR_T *p2 = p;
7064
7065 /* Compare the beginning of the repeat with what in the
7066 pattern follows its end. If we can establish that there
7067 is nothing that they would both match, i.e., that we
7068 would have to backtrack because of (as in, e.g., `a*a')
7069 then we can change to pop_failure_jump, because we'll
7070 never have to backtrack.
7071
7072 This is not true in the case of alternatives: in
7073 `(a|ab)*' we do need to backtrack to the `ab' alternative
7074 (e.g., if the string was `ab'). But instead of trying to
7075 detect that here, the alternative has put on a dummy
7076 failure point which is what we will end up popping. */
7077
7078 /* Skip over open/close-group commands.
7079 If what follows this loop is a ...+ construct,
7080 look at what begins its body, since we will have to
7081 match at least one of that. */
7082 while (1)
7083 {
7084 if (p2 + 2 < pend
7085 && ((re_opcode_t) *p2 == stop_memory
7086 || (re_opcode_t) *p2 == start_memory))
7087 p2 += 3;
7088 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
7089 && (re_opcode_t) *p2 == dummy_failure_jump)
7090 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
7091 else
7092 break;
7093 }
7094
7095 p1 = p + mcnt;
7096 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7097 to the `maybe_finalize_jump' of this case. Examine what
7098 follows. */
7099
7100 /* If we're at the end of the pattern, we can change. */
7101 if (p2 == pend)
7102 {
7103 /* Consider what happens when matching ":\(.*\)"
7104 against ":/". I don't really understand this code
7105 yet. */
7106 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7107 pop_failure_jump;
7108 DEBUG_PRINT1
7109 (" End of pattern: change to `pop_failure_jump'.\n");
7110 }
7111
7112 else if ((re_opcode_t) *p2 == exactn
7113#ifdef MBS_SUPPORT
7114 || (re_opcode_t) *p2 == exactn_bin
7115#endif
7116 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
7117 {
7118 register UCHAR_T c
7119 = *p2 == (UCHAR_T) endline ? '\n' : p2[2];
7120
7121 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
7122#ifdef MBS_SUPPORT
7123 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
7124#endif
7125 ) && p1[3+OFFSET_ADDRESS_SIZE] != c)
7126 {
7127 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7128 pop_failure_jump;
7129#ifdef WCHAR
7130 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7131 (wint_t) c,
7132 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
7133#else
7134 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7135 (char) c,
7136 (char) p1[3+OFFSET_ADDRESS_SIZE]);
7137#endif
7138 }
7139
7140#ifndef WCHAR
7141 else if ((re_opcode_t) p1[3] == charset
7142 || (re_opcode_t) p1[3] == charset_not)
7143 {
7144 int not = (re_opcode_t) p1[3] == charset_not;
7145
7146 if (c < (unsigned) (p1[4] * BYTEWIDTH)
7147 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
7148 not = !not;
7149
7150 /* `not' is equal to 1 if c would match, which means
7151 that we can't change to pop_failure_jump. */
7152 if (!not)
7153 {
7154 p[-3] = (unsigned char) pop_failure_jump;
7155 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7156 }
7157 }
7158#endif /* not WCHAR */
7159 }
7160#ifndef WCHAR
7161 else if ((re_opcode_t) *p2 == charset)
7162 {
7163 /* We win if the first character of the loop is not part
7164 of the charset. */
7165 if ((re_opcode_t) p1[3] == exactn
7166 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
7167 && (p2[2 + p1[5] / BYTEWIDTH]
7168 & (1 << (p1[5] % BYTEWIDTH)))))
7169 {
7170 p[-3] = (unsigned char) pop_failure_jump;
7171 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7172 }
7173
7174 else if ((re_opcode_t) p1[3] == charset_not)
7175 {
7176 int idx;
7177 /* We win if the charset_not inside the loop
7178 lists every character listed in the charset after. */
7179 for (idx = 0; idx < (int) p2[1]; idx++)
7180 if (! (p2[2 + idx] == 0
7181 || (idx < (int) p1[4]
7182 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
7183 break;
7184
7185 if (idx == p2[1])
7186 {
7187 p[-3] = (unsigned char) pop_failure_jump;
7188 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7189 }
7190 }
7191 else if ((re_opcode_t) p1[3] == charset)
7192 {
7193 int idx;
7194 /* We win if the charset inside the loop
7195 has no overlap with the one after the loop. */
7196 for (idx = 0;
7197 idx < (int) p2[1] && idx < (int) p1[4];
7198 idx++)
7199 if ((p2[2 + idx] & p1[5 + idx]) != 0)
7200 break;
7201
7202 if (idx == p2[1] || idx == p1[4])
7203 {
7204 p[-3] = (unsigned char) pop_failure_jump;
7205 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7206 }
7207 }
7208 }
7209#endif /* not WCHAR */
7210 }
7211 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
7212 if ((re_opcode_t) p[-1] != pop_failure_jump)
7213 {
7214 p[-1] = (UCHAR_T) jump;
7215 DEBUG_PRINT1 (" Match => jump.\n");
7216 goto unconditional_jump;
7217 }
7218 /* Note fall through. */
7219
7220
7221 /* The end of a simple repeat has a pop_failure_jump back to
7222 its matching on_failure_jump, where the latter will push a
7223 failure point. The pop_failure_jump takes off failure
7224 points put on by this pop_failure_jump's matching
7225 on_failure_jump; we got through the pattern to here from the
7226 matching on_failure_jump, so didn't fail. */
7227 case pop_failure_jump:
7228 {
7229 /* We need to pass separate storage for the lowest and
7230 highest registers, even though we don't care about the
7231 actual values. Otherwise, we will restore only one
7232 register from the stack, since lowest will == highest in
7233 `pop_failure_point'. */
7234 active_reg_t dummy_low_reg, dummy_high_reg;
7235 UCHAR_T *pdummy = NULL;
7236 const CHAR_T *sdummy = NULL;
7237
7238 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7239 POP_FAILURE_POINT (sdummy, pdummy,
7240 dummy_low_reg, dummy_high_reg,
7241 reg_dummy, reg_dummy, reg_info_dummy);
7242 }
7243 /* Note fall through. */
7244
7245 unconditional_jump:
7246#ifdef _LIBC
7247 DEBUG_PRINT2 ("\n%p: ", p);
7248#else
7249 DEBUG_PRINT2 ("\n0x%x: ", p);
7250#endif
7251 /* Note fall through. */
7252
7253 /* Unconditionally jump (without popping any failure points). */
7254 case jump:
7255 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
7256 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
7257 p += mcnt; /* Do the jump. */
7258#ifdef _LIBC
7259 DEBUG_PRINT2 ("(to %p).\n", p);
7260#else
7261 DEBUG_PRINT2 ("(to 0x%x).\n", p);
7262#endif
7263 break;
7264
7265
7266 /* We need this opcode so we can detect where alternatives end
7267 in `group_match_null_string_p' et al. */
7268 case jump_past_alt:
7269 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7270 goto unconditional_jump;
7271
7272
7273 /* Normally, the on_failure_jump pushes a failure point, which
7274 then gets popped at pop_failure_jump. We will end up at
7275 pop_failure_jump, also, and with a pattern of, say, `a+', we
7276 are skipping over the on_failure_jump, so we have to push
7277 something meaningless for pop_failure_jump to pop. */
7278 case dummy_failure_jump:
7279 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7280 /* It doesn't matter what we push for the string here. What
7281 the code at `fail' tests is the value for the pattern. */
7282 PUSH_FAILURE_POINT (NULL, NULL, -2);
7283 goto unconditional_jump;
7284
7285
7286 /* At the end of an alternative, we need to push a dummy failure
7287 point in case we are followed by a `pop_failure_jump', because
7288 we don't want the failure point for the alternative to be
7289 popped. For example, matching `(a|ab)*' against `aab'
7290 requires that we match the `ab' alternative. */
7291 case push_dummy_failure:
7292 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7293 /* See comments just above at `dummy_failure_jump' about the
7294 two zeroes. */
7295 PUSH_FAILURE_POINT (NULL, NULL, -2);
7296 break;
7297
7298 /* Have to succeed matching what follows at least n times.
7299 After that, handle like `on_failure_jump'. */
7300 case succeed_n:
7301 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7302 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
7303
7304 assert (mcnt >= 0);
7305 /* Originally, this is how many times we HAVE to succeed. */
7306 if (mcnt > 0)
7307 {
7308 mcnt--;
7309 p += OFFSET_ADDRESS_SIZE;
7310 STORE_NUMBER_AND_INCR (p, mcnt);
7311#ifdef _LIBC
7312 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
7313 , mcnt);
7314#else
7315 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
7316 , mcnt);
7317#endif
7318 }
7319 else if (mcnt == 0)
7320 {
7321#ifdef _LIBC
7322 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7323 p + OFFSET_ADDRESS_SIZE);
7324#else
7325 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7326 p + OFFSET_ADDRESS_SIZE);
7327#endif /* _LIBC */
7328
7329#ifdef WCHAR
7330 p[1] = (UCHAR_T) no_op;
7331#else
7332 p[2] = (UCHAR_T) no_op;
7333 p[3] = (UCHAR_T) no_op;
7334#endif /* WCHAR */
7335 goto on_failure;
7336 }
7337 break;
7338
7339 case jump_n:
7340 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7341 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7342
7343 /* Originally, this is how many times we CAN jump. */
7344 if (mcnt)
7345 {
7346 mcnt--;
7347 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7348
7349#ifdef _LIBC
7350 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7351 mcnt);
7352#else
7353 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7354 mcnt);
7355#endif /* _LIBC */
7356 goto unconditional_jump;
7357 }
7358 /* If don't have to jump any more, skip over the rest of command. */
7359 else
7360 p += 2 * OFFSET_ADDRESS_SIZE;
7361 break;
7362
7363 case set_number_at:
7364 {
7365 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7366
7367 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7368 p1 = p + mcnt;
7369 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7370#ifdef _LIBC
7371 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7372#else
7373 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7374#endif
7375 STORE_NUMBER (p1, mcnt);
7376 break;
7377 }
7378
7379#if 0
7380 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7381 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7382 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7383 macro and introducing temporary variables works around the bug. */
7384
7385 case wordbound:
7386 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7387 if (AT_WORD_BOUNDARY (d))
7388 break;
7389 goto fail;
7390
7391 case notwordbound:
7392 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7393 if (AT_WORD_BOUNDARY (d))
7394 goto fail;
7395 break;
7396#else
7397 case wordbound:
7398 {
7399 boolean prevchar, thischar;
7400
7401 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7402 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7403 break;
7404
7405 prevchar = WORDCHAR_P (d - 1);
7406 thischar = WORDCHAR_P (d);
7407 if (prevchar != thischar)
7408 break;
7409 goto fail;
7410 }
7411
7412 case notwordbound:
7413 {
7414 boolean prevchar, thischar;
7415
7416 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7417 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7418 goto fail;
7419
7420 prevchar = WORDCHAR_P (d - 1);
7421 thischar = WORDCHAR_P (d);
7422 if (prevchar != thischar)
7423 goto fail;
7424 break;
7425 }
7426#endif
7427
7428 case wordbeg:
7429 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7430 if (!AT_STRINGS_END (d) && WORDCHAR_P (d)
7431 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7432 break;
7433 goto fail;
7434
7435 case wordend:
7436 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7437 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7438 && (AT_STRINGS_END (d) || !WORDCHAR_P (d)))
7439 break;
7440 goto fail;
7441
7442#ifdef emacs
7443 case before_dot:
7444 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7445 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7446 goto fail;
7447 break;
7448
7449 case at_dot:
7450 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7451 if (PTR_CHAR_POS ((unsigned char *) d) != point)
7452 goto fail;
7453 break;
7454
7455 case after_dot:
7456 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7457 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7458 goto fail;
7459 break;
7460
7461 case syntaxspec:
7462 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7463 mcnt = *p++;
7464 goto matchsyntax;
7465
7466 case wordchar:
7467 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7468 mcnt = (int) Sword;
7469 matchsyntax:
7470 PREFETCH ();
7471 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7472 d++;
7473 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7474 goto fail;
7475 SET_REGS_MATCHED ();
7476 break;
7477
7478 case notsyntaxspec:
7479 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7480 mcnt = *p++;
7481 goto matchnotsyntax;
7482
7483 case notwordchar:
7484 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7485 mcnt = (int) Sword;
7486 matchnotsyntax:
7487 PREFETCH ();
7488 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7489 d++;
7490 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7491 goto fail;
7492 SET_REGS_MATCHED ();
7493 break;
7494
7495#else /* not emacs */
7496 case wordchar:
7497 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7498 PREFETCH ();
7499 if (!WORDCHAR_P (d))
7500 goto fail;
7501 SET_REGS_MATCHED ();
7502 d++;
7503 break;
7504
7505 case notwordchar:
7506 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7507 PREFETCH ();
7508 if (WORDCHAR_P (d))
7509 goto fail;
7510 SET_REGS_MATCHED ();
7511 d++;
7512 break;
7513#endif /* not emacs */
7514
7515 default:
7516 abort ();
7517 }
7518 continue; /* Successfully executed one pattern command; keep going. */
7519
7520
7521 /* We goto here if a matching operation fails. */
7522 fail:
7523 if (!FAIL_STACK_EMPTY ())
7524 { /* A restart point is known. Restore to that state. */
7525 DEBUG_PRINT1 ("\nFAIL:\n");
7526 POP_FAILURE_POINT (d, p,
7527 lowest_active_reg, highest_active_reg,
7528 regstart, regend, reg_info);
7529
7530 /* If this failure point is a dummy, try the next one. */
7531 if (!p)
7532 goto fail;
7533
7534 /* If we failed to the end of the pattern, don't examine *p. */
7535 assert (p <= pend);
7536 if (p < pend)
7537 {
7538 boolean is_a_jump_n = false;
7539
7540 /* If failed to a backwards jump that's part of a repetition
7541 loop, need to pop this failure point and use the next one. */
7542 switch ((re_opcode_t) *p)
7543 {
7544 case jump_n:
7545 is_a_jump_n = true;
7546 case maybe_pop_jump:
7547 case pop_failure_jump:
7548 case jump:
7549 p1 = p + 1;
7550 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7551 p1 += mcnt;
7552
7553 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7554 || (!is_a_jump_n
7555 && (re_opcode_t) *p1 == on_failure_jump))
7556 goto fail;
7557 break;
7558 default:
7559 /* do nothing */ ;
7560 }
7561 }
7562
7563 if (d >= string1 && d <= end1)
7564 dend = end_match_1;
7565 }
7566 else
7567 break; /* Matching at this starting point really fails. */
7568 } /* for (;;) */
7569
7570 if (best_regs_set)
7571 goto restore_best_regs;
7572
7573 FREE_VARIABLES ();
7574
7575 return -1; /* Failure to match. */
7576} /* re_match_2 */
7577\f
7578/* Subroutine definitions for re_match_2. */
7579
7580
7581/* We are passed P pointing to a register number after a start_memory.
7582
7583 Return true if the pattern up to the corresponding stop_memory can
7584 match the empty string, and false otherwise.
7585
7586 If we find the matching stop_memory, sets P to point to one past its number.
7587 Otherwise, sets P to an undefined byte less than or equal to END.
7588
7589 We don't handle duplicates properly (yet). */
7590
7591static boolean
7592PREFIX(group_match_null_string_p) (p, end, reg_info)
7593 UCHAR_T **p, *end;
7594 PREFIX(register_info_type) *reg_info;
7595{
7596 int mcnt;
7597 /* Point to after the args to the start_memory. */
7598 UCHAR_T *p1 = *p + 2;
7599
7600 while (p1 < end)
7601 {
7602 /* Skip over opcodes that can match nothing, and return true or
7603 false, as appropriate, when we get to one that can't, or to the
7604 matching stop_memory. */
7605
7606 switch ((re_opcode_t) *p1)
7607 {
7608 /* Could be either a loop or a series of alternatives. */
7609 case on_failure_jump:
7610 p1++;
7611 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7612
7613 /* If the next operation is not a jump backwards in the
7614 pattern. */
7615
7616 if (mcnt >= 0)
7617 {
7618 /* Go through the on_failure_jumps of the alternatives,
7619 seeing if any of the alternatives cannot match nothing.
7620 The last alternative starts with only a jump,
7621 whereas the rest start with on_failure_jump and end
7622 with a jump, e.g., here is the pattern for `a|b|c':
7623
7624 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7625 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7626 /exactn/1/c
7627
7628 So, we have to first go through the first (n-1)
7629 alternatives and then deal with the last one separately. */
7630
7631
7632 /* Deal with the first (n-1) alternatives, which start
7633 with an on_failure_jump (see above) that jumps to right
7634 past a jump_past_alt. */
7635
7636 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7637 jump_past_alt)
7638 {
7639 /* `mcnt' holds how many bytes long the alternative
7640 is, including the ending `jump_past_alt' and
7641 its number. */
7642
7643 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt -
7644 (1 + OFFSET_ADDRESS_SIZE),
7645 reg_info))
7646 return false;
7647
7648 /* Move to right after this alternative, including the
7649 jump_past_alt. */
7650 p1 += mcnt;
7651
7652 /* Break if it's the beginning of an n-th alternative
7653 that doesn't begin with an on_failure_jump. */
7654 if ((re_opcode_t) *p1 != on_failure_jump)
7655 break;
7656
7657 /* Still have to check that it's not an n-th
7658 alternative that starts with an on_failure_jump. */
7659 p1++;
7660 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7661 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7662 jump_past_alt)
7663 {
7664 /* Get to the beginning of the n-th alternative. */
7665 p1 -= 1 + OFFSET_ADDRESS_SIZE;
7666 break;
7667 }
7668 }
7669
7670 /* Deal with the last alternative: go back and get number
7671 of the `jump_past_alt' just before it. `mcnt' contains
7672 the length of the alternative. */
7673 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7674
7675 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info))
7676 return false;
7677
7678 p1 += mcnt; /* Get past the n-th alternative. */
7679 } /* if mcnt > 0 */
7680 break;
7681
7682
7683 case stop_memory:
7684 assert (p1[1] == **p);
7685 *p = p1 + 2;
7686 return true;
7687
7688
7689 default:
7690 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7691 return false;
7692 }
7693 } /* while p1 < end */
7694
7695 return false;
7696} /* group_match_null_string_p */
7697
7698
7699/* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7700 It expects P to be the first byte of a single alternative and END one
7701 byte past the last. The alternative can contain groups. */
7702
7703static boolean
7704PREFIX(alt_match_null_string_p) (p, end, reg_info)
7705 UCHAR_T *p, *end;
7706 PREFIX(register_info_type) *reg_info;
7707{
7708 int mcnt;
7709 UCHAR_T *p1 = p;
7710
7711 while (p1 < end)
7712 {
7713 /* Skip over opcodes that can match nothing, and break when we get
7714 to one that can't. */
7715
7716 switch ((re_opcode_t) *p1)
7717 {
7718 /* It's a loop. */
7719 case on_failure_jump:
7720 p1++;
7721 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7722 p1 += mcnt;
7723 break;
7724
7725 default:
7726 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7727 return false;
7728 }
7729 } /* while p1 < end */
7730
7731 return true;
7732} /* alt_match_null_string_p */
7733
7734
7735/* Deals with the ops common to group_match_null_string_p and
7736 alt_match_null_string_p.
7737
7738 Sets P to one after the op and its arguments, if any. */
7739
7740static boolean
7741PREFIX(common_op_match_null_string_p) (p, end, reg_info)
7742 UCHAR_T **p, *end;
7743 PREFIX(register_info_type) *reg_info;
7744{
7745 int mcnt;
7746 boolean ret;
7747 int reg_no;
7748 UCHAR_T *p1 = *p;
7749
7750 switch ((re_opcode_t) *p1++)
7751 {
7752 case no_op:
7753 case begline:
7754 case endline:
7755 case begbuf:
7756 case endbuf:
7757 case wordbeg:
7758 case wordend:
7759 case wordbound:
7760 case notwordbound:
7761#ifdef emacs
7762 case before_dot:
7763 case at_dot:
7764 case after_dot:
7765#endif
7766 break;
7767
7768 case start_memory:
7769 reg_no = *p1;
7770 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7771 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info);
7772
7773 /* Have to set this here in case we're checking a group which
7774 contains a group and a back reference to it. */
7775
7776 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7777 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7778
7779 if (!ret)
7780 return false;
7781 break;
7782
7783 /* If this is an optimized succeed_n for zero times, make the jump. */
7784 case jump:
7785 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7786 if (mcnt >= 0)
7787 p1 += mcnt;
7788 else
7789 return false;
7790 break;
7791
7792 case succeed_n:
7793 /* Get to the number of times to succeed. */
7794 p1 += OFFSET_ADDRESS_SIZE;
7795 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7796
7797 if (mcnt == 0)
7798 {
7799 p1 -= 2 * OFFSET_ADDRESS_SIZE;
7800 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7801 p1 += mcnt;
7802 }
7803 else
7804 return false;
7805 break;
7806
7807 case duplicate:
7808 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7809 return false;
7810 break;
7811
7812 case set_number_at:
7813 p1 += 2 * OFFSET_ADDRESS_SIZE;
7814
7815 default:
7816 /* All other opcodes mean we cannot match the empty string. */
7817 return false;
7818 }
7819
7820 *p = p1;
7821 return true;
7822} /* common_op_match_null_string_p */
7823
7824
7825/* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7826 bytes; nonzero otherwise. */
7827
7828static int
7829PREFIX(bcmp_translate) (s1, s2, len, translate)
7830 const CHAR_T *s1, *s2;
7831 register int len;
7832 RE_TRANSLATE_TYPE translate;
7833{
7834 register const UCHAR_T *p1 = (const UCHAR_T *) s1;
7835 register const UCHAR_T *p2 = (const UCHAR_T *) s2;
7836 while (len)
7837 {
7838#ifdef WCHAR
7839 if (((*p1<=0xff)?translate[*p1++]:*p1++)
7840 != ((*p2<=0xff)?translate[*p2++]:*p2++))
7841 return 1;
7842#else /* BYTE */
7843 if (translate[*p1++] != translate[*p2++]) return 1;
7844#endif /* WCHAR */
7845 len--;
7846 }
7847 return 0;
7848}
7849\f
7850
7851#else /* not INSIDE_RECURSION */
7852
7853/* Entry points for GNU code. */
7854
7855/* re_compile_pattern is the GNU regular expression compiler: it
7856 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7857 Returns 0 if the pattern was valid, otherwise an error string.
7858
7859 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7860 are set in BUFP on entry.
7861
7862 We call regex_compile to do the actual compilation. */
7863
7864const char *
7865re_compile_pattern (pattern, length, bufp)
7866 const char *pattern;
7867 size_t length;
7868 struct re_pattern_buffer *bufp;
7869{
7870 reg_errcode_t ret;
7871
7872 /* GNU code is written to assume at least RE_NREGS registers will be set
7873 (and at least one extra will be -1). */
7874 bufp->regs_allocated = REGS_UNALLOCATED;
7875
7876 /* And GNU code determines whether or not to get register information
7877 by passing null for the REGS argument to re_match, etc., not by
7878 setting no_sub. */
7879 bufp->no_sub = 0;
7880
7881 /* Match anchors at newline. */
7882 bufp->newline_anchor = 1;
7883
7884# ifdef MBS_SUPPORT
7885 if (MB_CUR_MAX != 1)
7886 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp);
7887 else
7888# endif
7889 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp);
7890
7891 if (!ret)
7892 return NULL;
a2832523 7893 return gettext (re_error_msgid[(int) ret]);
2a6ef469
DD		 7894}
7895#ifdef _LIBC
7896weak_alias (__re_compile_pattern, re_compile_pattern)
7897#endif
7898\f
7899/* Entry points compatible with 4.2 BSD regex library. We don't define
7900 them unless specifically requested. */
7901
7902#if defined _REGEX_RE_COMP || defined _LIBC
7903
7904/* BSD has one and only one pattern buffer. */
7905static struct re_pattern_buffer re_comp_buf;
7906
7907char *
7908#ifdef _LIBC
7909/* Make these definitions weak in libc, so POSIX programs can redefine
7910 these names if they don't use our functions, and still use
7911 regcomp/regexec below without link errors. */
7912weak_function
7913#endif
7914re_comp (s)
7915 const char *s;
7916{
7917 reg_errcode_t ret;
7918
7919 if (!s)
7920 {
7921 if (!re_comp_buf.buffer)
7922 return gettext ("No previous regular expression");
7923 return 0;
7924 }
7925
7926 if (!re_comp_buf.buffer)
7927 {
7928 re_comp_buf.buffer = (unsigned char *) malloc (200);
7929 if (re_comp_buf.buffer == NULL)
a2832523 7930 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
2a6ef469
DD		 7931 re_comp_buf.allocated = 200;
7932
7933 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
7934 if (re_comp_buf.fastmap == NULL)
a2832523 7935 return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
2a6ef469
DD		 7936 }
7937
7938 /* Since `re_exec' always passes NULL for the `regs' argument, we
7939 don't need to initialize the pattern buffer fields which affect it. */
7940
7941 /* Match anchors at newlines. */
7942 re_comp_buf.newline_anchor = 1;
7943
7944# ifdef MBS_SUPPORT
7945 if (MB_CUR_MAX != 1)
7946 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7947 else
7948# endif
7949 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7950
7951 if (!ret)
7952 return NULL;
7953
7954 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
a2832523 7955 return (char *) gettext (re_error_msgid[(int) ret]);
2a6ef469
DD		 7956}
7957
7958
7959int
7960#ifdef _LIBC
7961weak_function
7962#endif
7963re_exec (s)
7964 const char *s;
7965{
7966 const int len = strlen (s);
7967 return
7968 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
7969}
7970
7971#endif /* _REGEX_RE_COMP */
7972\f
7973/* POSIX.2 functions. Don't define these for Emacs. */
7974
7975#ifndef emacs
7976
7977/* regcomp takes a regular expression as a string and compiles it.
7978
7979 PREG is a regex_t *. We do not expect any fields to be initialized,
7980 since POSIX says we shouldn't. Thus, we set
7981
7982 `buffer' to the compiled pattern;
7983 `used' to the length of the compiled pattern;
7984 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
7985 REG_EXTENDED bit in CFLAGS is set; otherwise, to
7986 RE_SYNTAX_POSIX_BASIC;
7987 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
7988 `fastmap' to an allocated space for the fastmap;
7989 `fastmap_accurate' to zero;
7990 `re_nsub' to the number of subexpressions in PATTERN.
7991
7992 PATTERN is the address of the pattern string.
7993
7994 CFLAGS is a series of bits which affect compilation.
7995
7996 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
7997 use POSIX basic syntax.
7998
7999 If REG_NEWLINE is set, then . and [^...] don't match newline.
8000 Also, regexec will try a match beginning after every newline.
8001
8002 If REG_ICASE is set, then we considers upper- and lowercase
8003 versions of letters to be equivalent when matching.
8004
8005 If REG_NOSUB is set, then when PREG is passed to regexec, that
8006 routine will report only success or failure, and nothing about the
8007 registers.
8008
8009 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
8010 the return codes and their meanings.) */
8011
8012int
8013regcomp (preg, pattern, cflags)
8014 regex_t *preg;
8015 const char *pattern;
8016 int cflags;
8017{
8018 reg_errcode_t ret;
8019 reg_syntax_t syntax
8020 = (cflags & REG_EXTENDED) ?
8021 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
8022
8023 /* regex_compile will allocate the space for the compiled pattern. */
8024 preg->buffer = 0;
8025 preg->allocated = 0;
8026 preg->used = 0;
8027
8028 /* Try to allocate space for the fastmap. */
8029 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
8030
8031 if (cflags & REG_ICASE)
8032 {
8033 unsigned i;
8034
8035 preg->translate
8036 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
8037 * sizeof (*(RE_TRANSLATE_TYPE)0));
8038 if (preg->translate == NULL)
8039 return (int) REG_ESPACE;
8040
8041 /* Map uppercase characters to corresponding lowercase ones. */
8042 for (i = 0; i < CHAR_SET_SIZE; i++)
8043 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
8044 }
8045 else
8046 preg->translate = NULL;
8047
8048 /* If REG_NEWLINE is set, newlines are treated differently. */
8049 if (cflags & REG_NEWLINE)
8050 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
8051 syntax &= ~RE_DOT_NEWLINE;
8052 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
8053 /* It also changes the matching behavior. */
8054 preg->newline_anchor = 1;
8055 }
8056 else
8057 preg->newline_anchor = 0;
8058
8059 preg->no_sub = !!(cflags & REG_NOSUB);
8060
8061 /* POSIX says a null character in the pattern terminates it, so we
8062 can use strlen here in compiling the pattern. */
8063# ifdef MBS_SUPPORT
8064 if (MB_CUR_MAX != 1)
8065 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg);
8066 else
8067# endif
8068 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg);
8069
8070 /* POSIX doesn't distinguish between an unmatched open-group and an
8071 unmatched close-group: both are REG_EPAREN. */
8072 if (ret == REG_ERPAREN) ret = REG_EPAREN;
8073
8074 if (ret == REG_NOERROR && preg->fastmap)
8075 {
8076 /* Compute the fastmap now, since regexec cannot modify the pattern
8077 buffer. */
8078 if (re_compile_fastmap (preg) == -2)
8079 {
8080 /* Some error occurred while computing the fastmap, just forget
8081 about it. */
8082 free (preg->fastmap);
8083 preg->fastmap = NULL;
8084 }
8085 }
8086
8087 return (int) ret;
8088}
8089#ifdef _LIBC
8090weak_alias (__regcomp, regcomp)
8091#endif
8092
8093
8094/* regexec searches for a given pattern, specified by PREG, in the
8095 string STRING.
8096
8097 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8098 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8099 least NMATCH elements, and we set them to the offsets of the
8100 corresponding matched substrings.
8101
8102 EFLAGS specifies `execution flags' which affect matching: if
8103 REG_NOTBOL is set, then ^ does not match at the beginning of the
8104 string; if REG_NOTEOL is set, then $ does not match at the end.
8105
8106 We return 0 if we find a match and REG_NOMATCH if not. */
8107
8108int
8109regexec (preg, string, nmatch, pmatch, eflags)
8110 const regex_t *preg;
8111 const char *string;
8112 size_t nmatch;
8113 regmatch_t pmatch[];
8114 int eflags;
8115{
8116 int ret;
8117 struct re_registers regs;
8118 regex_t private_preg;
8119 int len = strlen (string);
8120 boolean want_reg_info = !preg->no_sub && nmatch > 0;
8121
8122 private_preg = *preg;
8123
8124 private_preg.not_bol = !!(eflags & REG_NOTBOL);
8125 private_preg.not_eol = !!(eflags & REG_NOTEOL);
8126
8127 /* The user has told us exactly how many registers to return
8128 information about, via `nmatch'. We have to pass that on to the
8129 matching routines. */
8130 private_preg.regs_allocated = REGS_FIXED;
8131
8132 if (want_reg_info)
8133 {
8134 regs.num_regs = nmatch;
8135 regs.start = TALLOC (nmatch * 2, regoff_t);
8136 if (regs.start == NULL)
8137 return (int) REG_NOMATCH;
8138 regs.end = regs.start + nmatch;
8139 }
8140
8141 /* Perform the searching operation. */
8142 ret = re_search (&private_preg, string, len,
8143 /* start: */ 0, /* range: */ len,
8144 want_reg_info ? &regs : (struct re_registers *) 0);
8145
8146 /* Copy the register information to the POSIX structure. */
8147 if (want_reg_info)
8148 {
8149 if (ret >= 0)
8150 {
8151 unsigned r;
8152
8153 for (r = 0; r < nmatch; r++)
8154 {
8155 pmatch[r].rm_so = regs.start[r];
8156 pmatch[r].rm_eo = regs.end[r];
8157 }
8158 }
8159
8160 /* If we needed the temporary register info, free the space now. */
8161 free (regs.start);
8162 }
8163
8164 /* We want zero return to mean success, unlike `re_search'. */
8165 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
8166}
8167#ifdef _LIBC
8168weak_alias (__regexec, regexec)
8169#endif
8170
8171
8172/* Returns a message corresponding to an error code, ERRCODE, returned
8173 from either regcomp or regexec. We don't use PREG here. */
8174
8175size_t
8176regerror (errcode, preg, errbuf, errbuf_size)
8177 int errcode;
8178 const regex_t *preg;
8179 char *errbuf;
8180 size_t errbuf_size;
8181{
8182 const char *msg;
8183 size_t msg_size;
8184
8185 if (errcode < 0
a2832523
DD		 8186 || errcode >= (int) (sizeof (re_error_msgid)
8187 / sizeof (re_error_msgid[0])))
2a6ef469
DD		 8188 /* Only error codes returned by the rest of the code should be passed
8189 to this routine. If we are given anything else, or if other regex
8190 code generates an invalid error code, then the program has a bug.
8191 Dump core so we can fix it. */
8192 abort ();
8193
a2832523 8194 msg = gettext (re_error_msgid[errcode]);
2a6ef469
DD		 8195
8196 msg_size = strlen (msg) + 1; /* Includes the null. */
8197
8198 if (errbuf_size != 0)
8199 {
8200 if (msg_size > errbuf_size)
8201 {
8202#if defined HAVE_MEMPCPY || defined _LIBC
8203 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
8204#else
8205 memcpy (errbuf, msg, errbuf_size - 1);
8206 errbuf[errbuf_size - 1] = 0;
8207#endif
8208 }
8209 else
8210 memcpy (errbuf, msg, msg_size);
8211 }
8212
8213 return msg_size;
8214}
8215#ifdef _LIBC
8216weak_alias (__regerror, regerror)
8217#endif
8218
8219
8220/* Free dynamically allocated space used by PREG. */
8221
8222void
8223regfree (preg)
8224 regex_t *preg;
8225{
8226 if (preg->buffer != NULL)
8227 free (preg->buffer);
8228 preg->buffer = NULL;
8229
8230 preg->allocated = 0;
8231 preg->used = 0;
8232
8233 if (preg->fastmap != NULL)
8234 free (preg->fastmap);
8235 preg->fastmap = NULL;
8236 preg->fastmap_accurate = 0;
8237
8238 if (preg->translate != NULL)
8239 free (preg->translate);
8240 preg->translate = NULL;
8241}
8242#ifdef _LIBC
8243weak_alias (__regfree, regfree)
8244#endif
8245
8246#endif /* not emacs */
8247
8248#endif /* not INSIDE_RECURSION */
8249
8250\f
8251#undef STORE_NUMBER
8252#undef STORE_NUMBER_AND_INCR
8253#undef EXTRACT_NUMBER
8254#undef EXTRACT_NUMBER_AND_INCR
8255
8256#undef DEBUG_PRINT_COMPILED_PATTERN
8257#undef DEBUG_PRINT_DOUBLE_STRING
8258
8259#undef INIT_FAIL_STACK
8260#undef RESET_FAIL_STACK
8261#undef DOUBLE_FAIL_STACK
8262#undef PUSH_PATTERN_OP
8263#undef PUSH_FAILURE_POINTER
8264#undef PUSH_FAILURE_INT
8265#undef PUSH_FAILURE_ELT
8266#undef POP_FAILURE_POINTER
8267#undef POP_FAILURE_INT
8268#undef POP_FAILURE_ELT
8269#undef DEBUG_PUSH
8270#undef DEBUG_POP
8271#undef PUSH_FAILURE_POINT
8272#undef POP_FAILURE_POINT
8273
8274#undef REG_UNSET_VALUE
8275#undef REG_UNSET
8276
8277#undef PATFETCH
8278#undef PATFETCH_RAW
8279#undef PATUNFETCH
8280#undef TRANSLATE
8281
8282#undef INIT_BUF_SIZE
8283#undef GET_BUFFER_SPACE
8284#undef BUF_PUSH
8285#undef BUF_PUSH_2
8286#undef BUF_PUSH_3
8287#undef STORE_JUMP
8288#undef STORE_JUMP2
8289#undef INSERT_JUMP
8290#undef INSERT_JUMP2
8291#undef EXTEND_BUFFER
8292#undef GET_UNSIGNED_NUMBER
8293#undef FREE_STACK_RETURN
8294
8295# undef POINTER_TO_OFFSET
8296# undef MATCHING_IN_FRST_STRING
8297# undef PREFETCH
8298# undef AT_STRINGS_BEG
8299# undef AT_STRINGS_END
8300# undef WORDCHAR_P
8301# undef FREE_VAR
8302# undef FREE_VARIABLES
8303# undef NO_HIGHEST_ACTIVE_REG
8304# undef NO_LOWEST_ACTIVE_REG
8305
8306# undef CHAR_T
8307# undef UCHAR_T
8308# undef COMPILED_BUFFER_VAR
8309# undef OFFSET_ADDRESS_SIZE
8310# undef CHAR_CLASS_SIZE
8311# undef PREFIX
8312# undef ARG_PREFIX
8313# undef PUT_CHAR
8314# undef BYTE
8315# undef WCHAR
8316
8317# define DEFINED_ONCE
GDB Binutils - Deliverables
This page took 0.515073 seconds and 4 git commands to generate.