projects / deliverable / binutils-gdb.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
merge from gcc
[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.)
5 Copyright (C) 1993-1999, 2000, 2001 Free Software Foundation, Inc.
6
7 The GNU C Library is free software; you can redistribute it and/or
8 modify it under the terms of the GNU Library General Public License as
9 published by the Free Software Foundation; either version 2 of the
10 License, or (at your option) any later version.
11
12 The GNU C Library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Library General Public License for more details.
16
17 You should have received a copy of the GNU Library General Public
18 License along with the GNU C Library; see the file COPYING.LIB. If not,
19 write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
20 Boston, MA 02111-1307, USA. */
21
22/* This file has been modified for usage in libiberty. It includes "xregex.h"
23 instead of <regex.h>. The "xregex.h" header file renames all external
24 routines with an "x" prefix so they do not collide with the native regex
25 routines or with other components regex routines. */
26#if defined _AIX && !defined REGEX_MALLOC
27 #pragma alloca
28#endif
29
30#undef _GNU_SOURCE
31#define _GNU_SOURCE
32
33#ifdef HAVE_CONFIG_H
34# include <config.h>
35#endif
36
37#ifndef PARAMS
38# if defined __GNUC__ || (defined __STDC__ && __STDC__)
39# define PARAMS(args) args
40# else
41# define PARAMS(args) ()
42# endif /* GCC. */
43#endif /* Not PARAMS. */
44
45#ifndef INSIDE_RECURSION
46
47# if defined STDC_HEADERS && !defined emacs
48# include <stddef.h>
49# else
50/* We need this for `regex.h', and perhaps for the Emacs include files. */
51# include <sys/types.h>
52# endif
53
54# define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
55
56/* For platform which support the ISO C amendement 1 functionality we
57 support user defined character classes. */
58# if defined _LIBC || WIDE_CHAR_SUPPORT
59/* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
60# include <wchar.h>
61# include <wctype.h>
62# endif
63
64# ifdef _LIBC
65/* We have to keep the namespace clean. */
66# define regfree(preg) __regfree (preg)
67# define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
68# define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
69# define regerror(errcode, preg, errbuf, errbuf_size) \
70 __regerror(errcode, preg, errbuf, errbuf_size)
71# define re_set_registers(bu, re, nu, st, en) \
72 __re_set_registers (bu, re, nu, st, en)
73# define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
74 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
75# define re_match(bufp, string, size, pos, regs) \
76 __re_match (bufp, string, size, pos, regs)
77# define re_search(bufp, string, size, startpos, range, regs) \
78 __re_search (bufp, string, size, startpos, range, regs)
79# define re_compile_pattern(pattern, length, bufp) \
80 __re_compile_pattern (pattern, length, bufp)
81# define re_set_syntax(syntax) __re_set_syntax (syntax)
82# define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
83 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
84# define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
85
86# define btowc __btowc
87
88/* We are also using some library internals. */
89# include <locale/localeinfo.h>
90# include <locale/elem-hash.h>
91# include <langinfo.h>
92# include <locale/coll-lookup.h>
93# endif
94
95/* This is for other GNU distributions with internationalized messages. */
96# if HAVE_LIBINTL_H || defined _LIBC
97# include <libintl.h>
98# ifdef _LIBC
99# undef gettext
100# define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
101# endif
102# else
103# define gettext(msgid) (msgid)
104# endif
105
106# ifndef gettext_noop
107/* This define is so xgettext can find the internationalizable
108 strings. */
109# define gettext_noop(String) String
110# endif
111
112/* The `emacs' switch turns on certain matching commands
113 that make sense only in Emacs. */
114# ifdef emacs
115
116# include "lisp.h"
117# include "buffer.h"
118# include "syntax.h"
119
120# else /* not emacs */
121
122/* If we are not linking with Emacs proper,
123 we can't use the relocating allocator
124 even if config.h says that we can. */
125# undef REL_ALLOC
126
127# if defined STDC_HEADERS || defined _LIBC
128# include <stdlib.h>
129# else
130char *malloc ();
131char *realloc ();
132# endif
133
134/* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
135 If nothing else has been done, use the method below. */
136# ifdef INHIBIT_STRING_HEADER
137# if !(defined HAVE_BZERO && defined HAVE_BCOPY)
138# if !defined bzero && !defined bcopy
139# undef INHIBIT_STRING_HEADER
140# endif
141# endif
142# endif
143
144/* This is the normal way of making sure we have a bcopy and a bzero.
145 This is used in most programs--a few other programs avoid this
146 by defining INHIBIT_STRING_HEADER. */
147# ifndef INHIBIT_STRING_HEADER
148# if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
149# include <string.h>
150# ifndef bzero
151# ifndef _LIBC
152# define bzero(s, n) (memset (s, '\0', n), (s))
153# else
154# define bzero(s, n) __bzero (s, n)
155# endif
156# endif
157# else
158# include <strings.h>
159# ifndef memcmp
160# define memcmp(s1, s2, n) bcmp (s1, s2, n)
161# endif
162# ifndef memcpy
163# define memcpy(d, s, n) (bcopy (s, d, n), (d))
164# endif
165# endif
166# endif
167
168/* Define the syntax stuff for \<, \>, etc. */
169
170/* This must be nonzero for the wordchar and notwordchar pattern
171 commands in re_match_2. */
172# ifndef Sword
173# define Sword 1
174# endif
175
176# ifdef SWITCH_ENUM_BUG
177# define SWITCH_ENUM_CAST(x) ((int)(x))
178# else
179# define SWITCH_ENUM_CAST(x) (x)
180# endif
181
182# endif /* not emacs */
183
184# if defined _LIBC || HAVE_LIMITS_H
185# include <limits.h>
186# endif
187
188# ifndef MB_LEN_MAX
189# define MB_LEN_MAX 1
190# endif
191\f
192/* Get the interface, including the syntax bits. */
193# include "xregex.h" /* change for libiberty */
194
195/* isalpha etc. are used for the character classes. */
196# include <ctype.h>
197
198/* Jim Meyering writes:
199
200 "... Some ctype macros are valid only for character codes that
201 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
202 using /bin/cc or gcc but without giving an ansi option). So, all
203 ctype uses should be through macros like ISPRINT... If
204 STDC_HEADERS is defined, then autoconf has verified that the ctype
205 macros don't need to be guarded with references to isascii. ...
206 Defining isascii to 1 should let any compiler worth its salt
207 eliminate the && through constant folding."
208 Solaris defines some of these symbols so we must undefine them first. */
209
210# undef ISASCII
211# if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
212# define ISASCII(c) 1
213# else
214# define ISASCII(c) isascii(c)
215# endif
216
217# ifdef isblank
218# define ISBLANK(c) (ISASCII (c) && isblank (c))
219# else
220# define ISBLANK(c) ((c) == ' ' || (c) == '\t')
221# endif
222# ifdef isgraph
223# define ISGRAPH(c) (ISASCII (c) && isgraph (c))
224# else
225# define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
226# endif
227
228# undef ISPRINT
229# define ISPRINT(c) (ISASCII (c) && isprint (c))
230# define ISDIGIT(c) (ISASCII (c) && isdigit (c))
231# define ISALNUM(c) (ISASCII (c) && isalnum (c))
232# define ISALPHA(c) (ISASCII (c) && isalpha (c))
233# define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
234# define ISLOWER(c) (ISASCII (c) && islower (c))
235# define ISPUNCT(c) (ISASCII (c) && ispunct (c))
236# define ISSPACE(c) (ISASCII (c) && isspace (c))
237# define ISUPPER(c) (ISASCII (c) && isupper (c))
238# define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
239
240# ifdef _tolower
241# define TOLOWER(c) _tolower(c)
242# else
243# define TOLOWER(c) tolower(c)
244# endif
245
246# ifndef NULL
247# define NULL (void *)0
248# endif
249
250/* We remove any previous definition of `SIGN_EXTEND_CHAR',
251 since ours (we hope) works properly with all combinations of
252 machines, compilers, `char' and `unsigned char' argument types.
253 (Per Bothner suggested the basic approach.) */
254# undef SIGN_EXTEND_CHAR
255# if __STDC__
256# define SIGN_EXTEND_CHAR(c) ((signed char) (c))
257# else /* not __STDC__ */
258/* As in Harbison and Steele. */
259# define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
260# endif
261\f
262# ifndef emacs
263/* How many characters in the character set. */
264# define CHAR_SET_SIZE 256
265
266# ifdef SYNTAX_TABLE
267
268extern char *re_syntax_table;
269
270# else /* not SYNTAX_TABLE */
271
272static char re_syntax_table[CHAR_SET_SIZE];
273
274static void init_syntax_once PARAMS ((void));
275
276static void
277init_syntax_once ()
278{
279 register int c;
280 static int done = 0;
281
282 if (done)
283 return;
284 bzero (re_syntax_table, sizeof re_syntax_table);
285
286 for (c = 0; c < CHAR_SET_SIZE; ++c)
287 if (ISALNUM (c))
288 re_syntax_table[c] = Sword;
289
290 re_syntax_table['_'] = Sword;
291
292 done = 1;
293}
294
295# endif /* not SYNTAX_TABLE */
296
297# define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
298
299# endif /* emacs */
300\f
301/* Integer type for pointers. */
302# if !defined _LIBC
303typedef unsigned long int uintptr_t;
304# endif
305
306/* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
307 use `alloca' instead of `malloc'. This is because using malloc in
308 re_search* or re_match* could cause memory leaks when C-g is used in
309 Emacs; also, malloc is slower and causes storage fragmentation. On
310 the other hand, malloc is more portable, and easier to debug.
311
312 Because we sometimes use alloca, some routines have to be macros,
313 not functions -- `alloca'-allocated space disappears at the end of the
314 function it is called in. */
315
316# ifdef REGEX_MALLOC
317
318# define REGEX_ALLOCATE malloc
319# define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
320# define REGEX_FREE free
321
322# else /* not REGEX_MALLOC */
323
324/* Emacs already defines alloca, sometimes. */
325# ifndef alloca
326
327/* Make alloca work the best possible way. */
328# ifdef __GNUC__
329# define alloca __builtin_alloca
330# else /* not __GNUC__ */
331# if HAVE_ALLOCA_H
332# include <alloca.h>
333# endif /* HAVE_ALLOCA_H */
334# endif /* not __GNUC__ */
335
336# endif /* not alloca */
337
338# define REGEX_ALLOCATE alloca
339
340/* Assumes a `char *destination' variable. */
341# define REGEX_REALLOCATE(source, osize, nsize) \
342 (destination = (char *) alloca (nsize), \
343 memcpy (destination, source, osize))
344
345/* No need to do anything to free, after alloca. */
346# define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
347
348# endif /* not REGEX_MALLOC */
349
350/* Define how to allocate the failure stack. */
351
352# if defined REL_ALLOC && defined REGEX_MALLOC
353
354# define REGEX_ALLOCATE_STACK(size) \
355 r_alloc (&failure_stack_ptr, (size))
356# define REGEX_REALLOCATE_STACK(source, osize, nsize) \
357 r_re_alloc (&failure_stack_ptr, (nsize))
358# define REGEX_FREE_STACK(ptr) \
359 r_alloc_free (&failure_stack_ptr)
360
361# else /* not using relocating allocator */
362
363# ifdef REGEX_MALLOC
364
365# define REGEX_ALLOCATE_STACK malloc
366# define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
367# define REGEX_FREE_STACK free
368
369# else /* not REGEX_MALLOC */
370
371# define REGEX_ALLOCATE_STACK alloca
372
373# define REGEX_REALLOCATE_STACK(source, osize, nsize) \
374 REGEX_REALLOCATE (source, osize, nsize)
375/* No need to explicitly free anything. */
376# define REGEX_FREE_STACK(arg)
377
378# endif /* not REGEX_MALLOC */
379# endif /* not using relocating allocator */
380
381
382/* True if `size1' is non-NULL and PTR is pointing anywhere inside
383 `string1' or just past its end. This works if PTR is NULL, which is
384 a good thing. */
385# define FIRST_STRING_P(ptr) \
386 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
387
388/* (Re)Allocate N items of type T using malloc, or fail. */
389# define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
390# define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
391# define RETALLOC_IF(addr, n, t) \
392 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
393# define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
394
395# define BYTEWIDTH 8 /* In bits. */
396
397# define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
398
399# undef MAX
400# undef MIN
401# define MAX(a, b) ((a) > (b) ? (a) : (b))
402# define MIN(a, b) ((a) < (b) ? (a) : (b))
403
404typedef char boolean;
405# define false 0
406# define true 1
407
408static reg_errcode_t byte_regex_compile _RE_ARGS ((const char *pattern, size_t size,
409 reg_syntax_t syntax,
410 struct re_pattern_buffer *bufp));
411static reg_errcode_t wcs_regex_compile _RE_ARGS ((const char *pattern, size_t size,
412 reg_syntax_t syntax,
413 struct re_pattern_buffer *bufp));
414
415static int byte_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
416 const char *string1, int size1,
417 const char *string2, int size2,
418 int pos,
419 struct re_registers *regs,
420 int stop));
421static int wcs_re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
422 const char *cstring1, int csize1,
423 const char *cstring2, int csize2,
424 int pos,
425 struct re_registers *regs,
426 int stop,
427 wchar_t *string1, int size1,
428 wchar_t *string2, int size2,
429 int *mbs_offset1, int *mbs_offset2));
430static int byte_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
431 const char *string1, int size1,
432 const char *string2, int size2,
433 int startpos, int range,
434 struct re_registers *regs, int stop));
435static int wcs_re_search_2 PARAMS ((struct re_pattern_buffer *bufp,
436 const char *string1, int size1,
437 const char *string2, int size2,
438 int startpos, int range,
439 struct re_registers *regs, int stop));
440static int byte_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
441static int wcs_re_compile_fastmap PARAMS ((struct re_pattern_buffer *bufp));
442
443\f
444/* These are the command codes that appear in compiled regular
445 expressions. Some opcodes are followed by argument bytes. A
446 command code can specify any interpretation whatsoever for its
447 arguments. Zero bytes may appear in the compiled regular expression. */
448
449typedef enum
450{
451 no_op = 0,
452
453 /* Succeed right away--no more backtracking. */
454 succeed,
455
456 /* Followed by one byte giving n, then by n literal bytes. */
457 exactn,
458
459# ifdef MBS_SUPPORT
460 /* Same as exactn, but contains binary data. */
461 exactn_bin,
462# endif
463
464 /* Matches any (more or less) character. */
465 anychar,
466
467 /* Matches any one char belonging to specified set. First
468 following byte is number of bitmap bytes. Then come bytes
469 for a bitmap saying which chars are in. Bits in each byte
470 are ordered low-bit-first. A character is in the set if its
471 bit is 1. A character too large to have a bit in the map is
472 automatically not in the set. */
473 /* ifdef MBS_SUPPORT, following element is length of character
474 classes, length of collating symbols, length of equivalence
475 classes, length of character ranges, and length of characters.
476 Next, character class element, collating symbols elements,
477 equivalence class elements, range elements, and character
478 elements follow.
479 See regex_compile function. */
480 charset,
481
482 /* Same parameters as charset, but match any character that is
483 not one of those specified. */
484 charset_not,
485
486 /* Start remembering the text that is matched, for storing in a
487 register. Followed by one byte with the register number, in
488 the range 0 to one less than the pattern buffer's re_nsub
489 field. Then followed by one byte with the number of groups
490 inner to this one. (This last has to be part of the
491 start_memory only because we need it in the on_failure_jump
492 of re_match_2.) */
493 start_memory,
494
495 /* Stop remembering the text that is matched and store it in a
496 memory register. Followed by one byte with the register
497 number, in the range 0 to one less than `re_nsub' in the
498 pattern buffer, and one byte with the number of inner groups,
499 just like `start_memory'. (We need the number of inner
500 groups here because we don't have any easy way of finding the
501 corresponding start_memory when we're at a stop_memory.) */
502 stop_memory,
503
504 /* Match a duplicate of something remembered. Followed by one
505 byte containing the register number. */
506 duplicate,
507
508 /* Fail unless at beginning of line. */
509 begline,
510
511 /* Fail unless at end of line. */
512 endline,
513
514 /* Succeeds if at beginning of buffer (if emacs) or at beginning
515 of string to be matched (if not). */
516 begbuf,
517
518 /* Analogously, for end of buffer/string. */
519 endbuf,
520
521 /* Followed by two byte relative address to which to jump. */
522 jump,
523
524 /* Same as jump, but marks the end of an alternative. */
525 jump_past_alt,
526
527 /* Followed by two-byte relative address of place to resume at
528 in case of failure. */
529 /* ifdef MBS_SUPPORT, the size of address is 1. */
530 on_failure_jump,
531
532 /* Like on_failure_jump, but pushes a placeholder instead of the
533 current string position when executed. */
534 on_failure_keep_string_jump,
535
536 /* Throw away latest failure point and then jump to following
537 two-byte relative address. */
538 /* ifdef MBS_SUPPORT, the size of address is 1. */
539 pop_failure_jump,
540
541 /* Change to pop_failure_jump if know won't have to backtrack to
542 match; otherwise change to jump. This is used to jump
543 back to the beginning of a repeat. If what follows this jump
544 clearly won't match what the repeat does, such that we can be
545 sure that there is no use backtracking out of repetitions
546 already matched, then we change it to a pop_failure_jump.
547 Followed by two-byte address. */
548 /* ifdef MBS_SUPPORT, the size of address is 1. */
549 maybe_pop_jump,
550
551 /* Jump to following two-byte address, and push a dummy failure
552 point. This failure point will be thrown away if an attempt
553 is made to use it for a failure. A `+' construct makes this
554 before the first repeat. Also used as an intermediary kind
555 of jump when compiling an alternative. */
556 /* ifdef MBS_SUPPORT, the size of address is 1. */
557 dummy_failure_jump,
558
559 /* Push a dummy failure point and continue. Used at the end of
560 alternatives. */
561 push_dummy_failure,
562
563 /* Followed by two-byte relative address and two-byte number n.
564 After matching N times, jump to the address upon failure. */
565 /* ifdef MBS_SUPPORT, the size of address is 1. */
566 succeed_n,
567
568 /* Followed by two-byte relative address, and two-byte number n.
569 Jump to the address N times, then fail. */
570 /* ifdef MBS_SUPPORT, the size of address is 1. */
571 jump_n,
572
573 /* Set the following two-byte relative address to the
574 subsequent two-byte number. The address *includes* the two
575 bytes of number. */
576 /* ifdef MBS_SUPPORT, the size of address is 1. */
577 set_number_at,
578
579 wordchar, /* Matches any word-constituent character. */
580 notwordchar, /* Matches any char that is not a word-constituent. */
581
582 wordbeg, /* Succeeds if at word beginning. */
583 wordend, /* Succeeds if at word end. */
584
585 wordbound, /* Succeeds if at a word boundary. */
586 notwordbound /* Succeeds if not at a word boundary. */
587
588# ifdef emacs
589 ,before_dot, /* Succeeds if before point. */
590 at_dot, /* Succeeds if at point. */
591 after_dot, /* Succeeds if after point. */
592
593 /* Matches any character whose syntax is specified. Followed by
594 a byte which contains a syntax code, e.g., Sword. */
595 syntaxspec,
596
597 /* Matches any character whose syntax is not that specified. */
598 notsyntaxspec
599# endif /* emacs */
600} re_opcode_t;
601#endif /* not INSIDE_RECURSION */
602\f
603
604#ifdef BYTE
605# define CHAR_T char
606# define UCHAR_T unsigned char
607# define COMPILED_BUFFER_VAR bufp->buffer
608# define OFFSET_ADDRESS_SIZE 2
609# define PREFIX(name) byte_##name
610# define ARG_PREFIX(name) name
611# define PUT_CHAR(c) putchar (c)
612#elif defined WCHAR
613# define CHAR_T wchar_t
614# define UCHAR_T wchar_t
615# define COMPILED_BUFFER_VAR wc_buffer
616# define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
617# define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1)
618# define PREFIX(name) wcs_##name
619# define ARG_PREFIX(name) c##name
620/* Should we use wide stream?? */
621# define PUT_CHAR(c) printf ("%C", c);
622# define TRUE 1
623# define FALSE 0
624#else
625# ifdef MBS_SUPPORT
626# define WCHAR
627# define INSIDE_RECURSION
628# include "regex.c"
629# undef INSIDE_RECURSION
630# endif
631# define BYTE
632# define INSIDE_RECURSION
633# include "regex.c"
634# undef INSIDE_RECURSION
635#endif
636
637#ifdef INSIDE_RECURSION
638/* Common operations on the compiled pattern. */
639
640/* Store NUMBER in two contiguous bytes starting at DESTINATION. */
641/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
642
643# ifdef WCHAR
644# define STORE_NUMBER(destination, number) \
645 do { \
646 *(destination) = (UCHAR_T)(number); \
647 } while (0)
648# else /* BYTE */
649# define STORE_NUMBER(destination, number) \
650 do { \
651 (destination)[0] = (number) & 0377; \
652 (destination)[1] = (number) >> 8; \
653 } while (0)
654# endif /* WCHAR */
655
656/* Same as STORE_NUMBER, except increment DESTINATION to
657 the byte after where the number is stored. Therefore, DESTINATION
658 must be an lvalue. */
659/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
660
661# define STORE_NUMBER_AND_INCR(destination, number) \
662 do { \
663 STORE_NUMBER (destination, number); \
664 (destination) += OFFSET_ADDRESS_SIZE; \
665 } while (0)
666
667/* Put into DESTINATION a number stored in two contiguous bytes starting
668 at SOURCE. */
669/* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
670
671# ifdef WCHAR
672# define EXTRACT_NUMBER(destination, source) \
673 do { \
674 (destination) = *(source); \
675 } while (0)
676# else /* BYTE */
677# define EXTRACT_NUMBER(destination, source) \
678 do { \
679 (destination) = *(source) & 0377; \
680 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
681 } while (0)
682# endif
683
684# ifdef DEBUG
685static void PREFIX(extract_number) _RE_ARGS ((int *dest, UCHAR_T *source));
686static void
687PREFIX(extract_number) (dest, source)
688 int *dest;
689 UCHAR_T *source;
690{
691# ifdef WCHAR
692 *dest = *source;
693# else /* BYTE */
694 int temp = SIGN_EXTEND_CHAR (*(source + 1));
695 *dest = *source & 0377;
696 *dest += temp << 8;
697# endif
698}
699
700# ifndef EXTRACT_MACROS /* To debug the macros. */
701# undef EXTRACT_NUMBER
702# define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src)
703# endif /* not EXTRACT_MACROS */
704
705# endif /* DEBUG */
706
707/* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
708 SOURCE must be an lvalue. */
709
710# define EXTRACT_NUMBER_AND_INCR(destination, source) \
711 do { \
712 EXTRACT_NUMBER (destination, source); \
713 (source) += OFFSET_ADDRESS_SIZE; \
714 } while (0)
715
716# ifdef DEBUG
717static void PREFIX(extract_number_and_incr) _RE_ARGS ((int *destination,
718 UCHAR_T **source));
719static void
720PREFIX(extract_number_and_incr) (destination, source)
721 int *destination;
722 UCHAR_T **source;
723{
724 PREFIX(extract_number) (destination, *source);
725 *source += OFFSET_ADDRESS_SIZE;
726}
727
728# ifndef EXTRACT_MACROS
729# undef EXTRACT_NUMBER_AND_INCR
730# define EXTRACT_NUMBER_AND_INCR(dest, src) \
731 PREFIX(extract_number_and_incr) (&dest, &src)
732# endif /* not EXTRACT_MACROS */
733
734# endif /* DEBUG */
735
736\f
737
738/* If DEBUG is defined, Regex prints many voluminous messages about what
739 it is doing (if the variable `debug' is nonzero). If linked with the
740 main program in `iregex.c', you can enter patterns and strings
741 interactively. And if linked with the main program in `main.c' and
742 the other test files, you can run the already-written tests. */
743
744# ifdef DEBUG
745
746# ifndef DEFINED_ONCE
747
748/* We use standard I/O for debugging. */
749# include <stdio.h>
750
751/* It is useful to test things that ``must'' be true when debugging. */
752# include <assert.h>
753
754static int debug;
755
756# define DEBUG_STATEMENT(e) e
757# define DEBUG_PRINT1(x) if (debug) printf (x)
758# define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
759# define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
760# define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
761# endif /* not DEFINED_ONCE */
762
763# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
764 if (debug) PREFIX(print_partial_compiled_pattern) (s, e)
765# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
766 if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2)
767
768
769/* Print the fastmap in human-readable form. */
770
771# ifndef DEFINED_ONCE
772void
773print_fastmap (fastmap)
774 char *fastmap;
775{
776 unsigned was_a_range = 0;
777 unsigned i = 0;
778
779 while (i < (1 << BYTEWIDTH))
780 {
781 if (fastmap[i++])
782 {
783 was_a_range = 0;
784 putchar (i - 1);
785 while (i < (1 << BYTEWIDTH) && fastmap[i])
786 {
787 was_a_range = 1;
788 i++;
789 }
790 if (was_a_range)
791 {
792 printf ("-");
793 putchar (i - 1);
794 }
795 }
796 }
797 putchar ('\n');
798}
799# endif /* not DEFINED_ONCE */
800
801
802/* Print a compiled pattern string in human-readable form, starting at
803 the START pointer into it and ending just before the pointer END. */
804
805void
806PREFIX(print_partial_compiled_pattern) (start, end)
807 UCHAR_T *start;
808 UCHAR_T *end;
809{
810 int mcnt, mcnt2;
811 UCHAR_T *p1;
812 UCHAR_T *p = start;
813 UCHAR_T *pend = end;
814
815 if (start == NULL)
816 {
817 printf ("(null)\n");
818 return;
819 }
820
821 /* Loop over pattern commands. */
822 while (p < pend)
823 {
824# ifdef _LIBC
825 printf ("%td:\t", p - start);
826# else
827 printf ("%ld:\t", (long int) (p - start));
828# endif
829
830 switch ((re_opcode_t) *p++)
831 {
832 case no_op:
833 printf ("/no_op");
834 break;
835
836 case exactn:
837 mcnt = *p++;
838 printf ("/exactn/%d", mcnt);
839 do
840 {
841 putchar ('/');
842 PUT_CHAR (*p++);
843 }
844 while (--mcnt);
845 break;
846
847# ifdef MBS_SUPPORT
848 case exactn_bin:
849 mcnt = *p++;
850 printf ("/exactn_bin/%d", mcnt);
851 do
852 {
853 printf("/%lx", (long int) *p++);
854 }
855 while (--mcnt);
856 break;
857# endif /* MBS_SUPPORT */
858
859 case start_memory:
860 mcnt = *p++;
861 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
862 break;
863
864 case stop_memory:
865 mcnt = *p++;
866 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
867 break;
868
869 case duplicate:
870 printf ("/duplicate/%ld", (long int) *p++);
871 break;
872
873 case anychar:
874 printf ("/anychar");
875 break;
876
877 case charset:
878 case charset_not:
879 {
880# ifdef WCHAR
881 int i, length;
882 wchar_t *workp = p;
883 printf ("/charset [%s",
884 (re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
885 p += 5;
886 length = *workp++; /* the length of char_classes */
887 for (i=0 ; i<length ; i++)
888 printf("[:%lx:]", (long int) *p++);
889 length = *workp++; /* the length of collating_symbol */
890 for (i=0 ; i<length ;)
891 {
892 printf("[.");
893 while(*p != 0)
894 PUT_CHAR((i++,*p++));
895 i++,p++;
896 printf(".]");
897 }
898 length = *workp++; /* the length of equivalence_class */
899 for (i=0 ; i<length ;)
900 {
901 printf("[=");
902 while(*p != 0)
903 PUT_CHAR((i++,*p++));
904 i++,p++;
905 printf("=]");
906 }
907 length = *workp++; /* the length of char_range */
908 for (i=0 ; i<length ; i++)
909 {
910 wchar_t range_start = *p++;
911 wchar_t range_end = *p++;
912 printf("%C-%C", range_start, range_end);
913 }
914 length = *workp++; /* the length of char */
915 for (i=0 ; i<length ; i++)
916 printf("%C", *p++);
917 putchar (']');
918# else
919 register int c, last = -100;
920 register int in_range = 0;
921
922 printf ("/charset [%s",
923 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
924
925 assert (p + *p < pend);
926
927 for (c = 0; c < 256; c++)
928 if (c / 8 < *p
929 && (p[1 + (c/8)] & (1 << (c % 8))))
930 {
931 /* Are we starting a range? */
932 if (last + 1 == c && ! in_range)
933 {
934 putchar ('-');
935 in_range = 1;
936 }
937 /* Have we broken a range? */
938 else if (last + 1 != c && in_range)
939 {
940 putchar (last);
941 in_range = 0;
942 }
943
944 if (! in_range)
945 putchar (c);
946
947 last = c;
948 }
949
950 if (in_range)
951 putchar (last);
952
953 putchar (']');
954
955 p += 1 + *p;
956# endif /* WCHAR */
957 }
958 break;
959
960 case begline:
961 printf ("/begline");
962 break;
963
964 case endline:
965 printf ("/endline");
966 break;
967
968 case on_failure_jump:
969 PREFIX(extract_number_and_incr) (&mcnt, &p);
970# ifdef _LIBC
971 printf ("/on_failure_jump to %td", p + mcnt - start);
972# else
973 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
974# endif
975 break;
976
977 case on_failure_keep_string_jump:
978 PREFIX(extract_number_and_incr) (&mcnt, &p);
979# ifdef _LIBC
980 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
981# else
982 printf ("/on_failure_keep_string_jump to %ld",
983 (long int) (p + mcnt - start));
984# endif
985 break;
986
987 case dummy_failure_jump:
988 PREFIX(extract_number_and_incr) (&mcnt, &p);
989# ifdef _LIBC
990 printf ("/dummy_failure_jump to %td", p + mcnt - start);
991# else
992 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
993# endif
994 break;
995
996 case push_dummy_failure:
997 printf ("/push_dummy_failure");
998 break;
999
1000 case maybe_pop_jump:
1001 PREFIX(extract_number_and_incr) (&mcnt, &p);
1002# ifdef _LIBC
1003 printf ("/maybe_pop_jump to %td", p + mcnt - start);
1004# else
1005 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
1006# endif
1007 break;
1008
1009 case pop_failure_jump:
1010 PREFIX(extract_number_and_incr) (&mcnt, &p);
1011# ifdef _LIBC
1012 printf ("/pop_failure_jump to %td", p + mcnt - start);
1013# else
1014 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
1015# endif
1016 break;
1017
1018 case jump_past_alt:
1019 PREFIX(extract_number_and_incr) (&mcnt, &p);
1020# ifdef _LIBC
1021 printf ("/jump_past_alt to %td", p + mcnt - start);
1022# else
1023 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
1024# endif
1025 break;
1026
1027 case jump:
1028 PREFIX(extract_number_and_incr) (&mcnt, &p);
1029# ifdef _LIBC
1030 printf ("/jump to %td", p + mcnt - start);
1031# else
1032 printf ("/jump to %ld", (long int) (p + mcnt - start));
1033# endif
1034 break;
1035
1036 case succeed_n:
1037 PREFIX(extract_number_and_incr) (&mcnt, &p);
1038 p1 = p + mcnt;
1039 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1040# ifdef _LIBC
1041 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
1042# else
1043 printf ("/succeed_n to %ld, %d times",
1044 (long int) (p1 - start), mcnt2);
1045# endif
1046 break;
1047
1048 case jump_n:
1049 PREFIX(extract_number_and_incr) (&mcnt, &p);
1050 p1 = p + mcnt;
1051 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1052 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
1053 break;
1054
1055 case set_number_at:
1056 PREFIX(extract_number_and_incr) (&mcnt, &p);
1057 p1 = p + mcnt;
1058 PREFIX(extract_number_and_incr) (&mcnt2, &p);
1059# ifdef _LIBC
1060 printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1061# else
1062 printf ("/set_number_at location %ld to %d",
1063 (long int) (p1 - start), mcnt2);
1064# endif
1065 break;
1066
1067 case wordbound:
1068 printf ("/wordbound");
1069 break;
1070
1071 case notwordbound:
1072 printf ("/notwordbound");
1073 break;
1074
1075 case wordbeg:
1076 printf ("/wordbeg");
1077 break;
1078
1079 case wordend:
1080 printf ("/wordend");
1081 break;
1082
1083# ifdef emacs
1084 case before_dot:
1085 printf ("/before_dot");
1086 break;
1087
1088 case at_dot:
1089 printf ("/at_dot");
1090 break;
1091
1092 case after_dot:
1093 printf ("/after_dot");
1094 break;
1095
1096 case syntaxspec:
1097 printf ("/syntaxspec");
1098 mcnt = *p++;
1099 printf ("/%d", mcnt);
1100 break;
1101
1102 case notsyntaxspec:
1103 printf ("/notsyntaxspec");
1104 mcnt = *p++;
1105 printf ("/%d", mcnt);
1106 break;
1107# endif /* emacs */
1108
1109 case wordchar:
1110 printf ("/wordchar");
1111 break;
1112
1113 case notwordchar:
1114 printf ("/notwordchar");
1115 break;
1116
1117 case begbuf:
1118 printf ("/begbuf");
1119 break;
1120
1121 case endbuf:
1122 printf ("/endbuf");
1123 break;
1124
1125 default:
1126 printf ("?%ld", (long int) *(p-1));
1127 }
1128
1129 putchar ('\n');
1130 }
1131
1132# ifdef _LIBC
1133 printf ("%td:\tend of pattern.\n", p - start);
1134# else
1135 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1136# endif
1137}
1138
1139
1140void
1141PREFIX(print_compiled_pattern) (bufp)
1142 struct re_pattern_buffer *bufp;
1143{
1144 UCHAR_T *buffer = (UCHAR_T*) bufp->buffer;
1145
1146 PREFIX(print_partial_compiled_pattern) (buffer, buffer
1147 + bufp->used / sizeof(UCHAR_T));
1148 printf ("%ld bytes used/%ld bytes allocated.\n",
1149 bufp->used, bufp->allocated);
1150
1151 if (bufp->fastmap_accurate && bufp->fastmap)
1152 {
1153 printf ("fastmap: ");
1154 print_fastmap (bufp->fastmap);
1155 }
1156
1157# ifdef _LIBC
1158 printf ("re_nsub: %Zd\t", bufp->re_nsub);
1159# else
1160 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
1161# endif
1162 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1163 printf ("can_be_null: %d\t", bufp->can_be_null);
1164 printf ("newline_anchor: %d\n", bufp->newline_anchor);
1165 printf ("no_sub: %d\t", bufp->no_sub);
1166 printf ("not_bol: %d\t", bufp->not_bol);
1167 printf ("not_eol: %d\t", bufp->not_eol);
1168 printf ("syntax: %lx\n", bufp->syntax);
1169 /* Perhaps we should print the translate table? */
1170}
1171
1172
1173void
1174PREFIX(print_double_string) (where, string1, size1, string2, size2)
1175 const CHAR_T *where;
1176 const CHAR_T *string1;
1177 const CHAR_T *string2;
1178 int size1;
1179 int size2;
1180{
1181 int this_char;
1182
1183 if (where == NULL)
1184 printf ("(null)");
1185 else
1186 {
1187 if (FIRST_STRING_P (where))
1188 {
1189 for (this_char = where - string1; this_char < size1; this_char++)
1190 PUT_CHAR (string1[this_char]);
1191
1192 where = string2;
1193 }
1194
1195 for (this_char = where - string2; this_char < size2; this_char++)
1196 PUT_CHAR (string2[this_char]);
1197 }
1198}
1199
1200# ifndef DEFINED_ONCE
1201void
1202printchar (c)
1203 int c;
1204{
1205 putc (c, stderr);
1206}
1207# endif
1208
1209# else /* not DEBUG */
1210
1211# ifndef DEFINED_ONCE
1212# undef assert
1213# define assert(e)
1214
1215# define DEBUG_STATEMENT(e)
1216# define DEBUG_PRINT1(x)
1217# define DEBUG_PRINT2(x1, x2)
1218# define DEBUG_PRINT3(x1, x2, x3)
1219# define DEBUG_PRINT4(x1, x2, x3, x4)
1220# endif /* not DEFINED_ONCE */
1221# define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1222# define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1223
1224# endif /* not DEBUG */
1225
1226\f
1227
1228# ifdef WCHAR
1229/* This convert a multibyte string to a wide character string.
1230 And write their correspondances to offset_buffer(see below)
1231 and write whether each wchar_t is binary data to is_binary.
1232 This assume invalid multibyte sequences as binary data.
1233 We assume offset_buffer and is_binary is already allocated
1234 enough space. */
1235
1236static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src,
1237 size_t len, int *offset_buffer,
1238 char *is_binary);
1239static size_t
1240convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary)
1241 CHAR_T *dest;
1242 const unsigned char* src;
1243 size_t len; /* the length of multibyte string. */
1244
1245 /* It hold correspondances between src(char string) and
1246 dest(wchar_t string) for optimization.
1247 e.g. src = "xxxyzz"
1248 dest = {'X', 'Y', 'Z'}
1249 (each "xxx", "y" and "zz" represent one multibyte character
1250 corresponding to 'X', 'Y' and 'Z'.)
1251 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1252 = {0, 3, 4, 6}
1253 */
1254 int *offset_buffer;
1255 char *is_binary;
1256{
1257 wchar_t *pdest = dest;
1258 const unsigned char *psrc = src;
1259 size_t wc_count = 0;
1260
1261 mbstate_t mbs;
1262 int i, consumed;
1263 size_t mb_remain = len;
1264 size_t mb_count = 0;
1265
1266 /* Initialize the conversion state. */
1267 memset (&mbs, 0, sizeof (mbstate_t));
1268
1269 offset_buffer[0] = 0;
1270 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1271 psrc += consumed)
1272 {
1273 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
1274
1275 if (consumed <= 0)
1276 /* failed to convert. maybe src contains binary data.
1277 So we consume 1 byte manualy. */
1278 {
1279 *pdest = *psrc;
1280 consumed = 1;
1281 is_binary[wc_count] = TRUE;
1282 }
1283 else
1284 is_binary[wc_count] = FALSE;
1285 /* In sjis encoding, we use yen sign as escape character in
1286 place of reverse solidus. So we convert 0x5c(yen sign in
1287 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1288 solidus in UCS2). */
1289 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1290 *pdest = (wchar_t) *psrc;
1291
1292 offset_buffer[wc_count + 1] = mb_count += consumed;
1293 }
1294
1295 /* Fill remain of the buffer with sentinel. */
1296 for (i = wc_count + 1 ; i <= len ; i++)
1297 offset_buffer[i] = mb_count + 1;
1298
1299 return wc_count;
1300}
1301
1302# endif /* WCHAR */
1303
1304#else /* not INSIDE_RECURSION */
1305
1306/* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1307 also be assigned to arbitrarily: each pattern buffer stores its own
1308 syntax, so it can be changed between regex compilations. */
1309/* This has no initializer because initialized variables in Emacs
1310 become read-only after dumping. */
1311reg_syntax_t re_syntax_options;
1312
1313
1314/* Specify the precise syntax of regexps for compilation. This provides
1315 for compatibility for various utilities which historically have
1316 different, incompatible syntaxes.
1317
1318 The argument SYNTAX is a bit mask comprised of the various bits
1319 defined in regex.h. We return the old syntax. */
1320
1321reg_syntax_t
1322re_set_syntax (syntax)
1323 reg_syntax_t syntax;
1324{
1325 reg_syntax_t ret = re_syntax_options;
1326
1327 re_syntax_options = syntax;
1328# ifdef DEBUG
1329 if (syntax & RE_DEBUG)
1330 debug = 1;
1331 else if (debug) /* was on but now is not */
1332 debug = 0;
1333# endif /* DEBUG */
1334 return ret;
1335}
1336# ifdef _LIBC
1337weak_alias (__re_set_syntax, re_set_syntax)
1338# endif
1339\f
1340/* This table gives an error message for each of the error codes listed
1341 in regex.h. Obviously the order here has to be same as there.
1342 POSIX doesn't require that we do anything for REG_NOERROR,
1343 but why not be nice? */
1344
1345static const char re_error_msgid[] =
1346 {
1347# define REG_NOERROR_IDX 0
1348 gettext_noop ("Success") /* REG_NOERROR */
1349 "\0"
1350# define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1351 gettext_noop ("No match") /* REG_NOMATCH */
1352 "\0"
1353# define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1354 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1355 "\0"
1356# define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1357 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1358 "\0"
1359# define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1360 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1361 "\0"
1362# define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1363 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1364 "\0"
1365# define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1366 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1367 "\0"
1368# define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1369 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1370 "\0"
1371# define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1372 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1373 "\0"
1374# define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1375 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1376 "\0"
1377# define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1378 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1379 "\0"
1380# define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1381 gettext_noop ("Invalid range end") /* REG_ERANGE */
1382 "\0"
1383# define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1384 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1385 "\0"
1386# define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1387 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1388 "\0"
1389# define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1390 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1391 "\0"
1392# define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1393 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1394 "\0"
1395# define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1396 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1397 };
1398
1399static const size_t re_error_msgid_idx[] =
1400 {
1401 REG_NOERROR_IDX,
1402 REG_NOMATCH_IDX,
1403 REG_BADPAT_IDX,
1404 REG_ECOLLATE_IDX,
1405 REG_ECTYPE_IDX,
1406 REG_EESCAPE_IDX,
1407 REG_ESUBREG_IDX,
1408 REG_EBRACK_IDX,
1409 REG_EPAREN_IDX,
1410 REG_EBRACE_IDX,
1411 REG_BADBR_IDX,
1412 REG_ERANGE_IDX,
1413 REG_ESPACE_IDX,
1414 REG_BADRPT_IDX,
1415 REG_EEND_IDX,
1416 REG_ESIZE_IDX,
1417 REG_ERPAREN_IDX
1418 };
1419\f
1420#endif /* INSIDE_RECURSION */
1421
1422#ifndef DEFINED_ONCE
1423/* Avoiding alloca during matching, to placate r_alloc. */
1424
1425/* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1426 searching and matching functions should not call alloca. On some
1427 systems, alloca is implemented in terms of malloc, and if we're
1428 using the relocating allocator routines, then malloc could cause a
1429 relocation, which might (if the strings being searched are in the
1430 ralloc heap) shift the data out from underneath the regexp
1431 routines.
1432
1433 Here's another reason to avoid allocation: Emacs
1434 processes input from X in a signal handler; processing X input may
1435 call malloc; if input arrives while a matching routine is calling
1436 malloc, then we're scrod. But Emacs can't just block input while
1437 calling matching routines; then we don't notice interrupts when
1438 they come in. So, Emacs blocks input around all regexp calls
1439 except the matching calls, which it leaves unprotected, in the
1440 faith that they will not malloc. */
1441
1442/* Normally, this is fine. */
1443# define MATCH_MAY_ALLOCATE
1444
1445/* When using GNU C, we are not REALLY using the C alloca, no matter
1446 what config.h may say. So don't take precautions for it. */
1447# ifdef __GNUC__
1448# undef C_ALLOCA
1449# endif
1450
1451/* The match routines may not allocate if (1) they would do it with malloc
1452 and (2) it's not safe for them to use malloc.
1453 Note that if REL_ALLOC is defined, matching would not use malloc for the
1454 failure stack, but we would still use it for the register vectors;
1455 so REL_ALLOC should not affect this. */
1456# if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1457# undef MATCH_MAY_ALLOCATE
1458# endif
1459#endif /* not DEFINED_ONCE */
1460\f
1461#ifdef INSIDE_RECURSION
1462/* Failure stack declarations and macros; both re_compile_fastmap and
1463 re_match_2 use a failure stack. These have to be macros because of
1464 REGEX_ALLOCATE_STACK. */
1465
1466
1467/* Number of failure points for which to initially allocate space
1468 when matching. If this number is exceeded, we allocate more
1469 space, so it is not a hard limit. */
1470# ifndef INIT_FAILURE_ALLOC
1471# define INIT_FAILURE_ALLOC 5
1472# endif
1473
1474/* Roughly the maximum number of failure points on the stack. Would be
1475 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1476 This is a variable only so users of regex can assign to it; we never
1477 change it ourselves. */
1478
1479# ifdef INT_IS_16BIT
1480
1481# ifndef DEFINED_ONCE
1482# if defined MATCH_MAY_ALLOCATE
1483/* 4400 was enough to cause a crash on Alpha OSF/1,
1484 whose default stack limit is 2mb. */
1485long int re_max_failures = 4000;
1486# else
1487long int re_max_failures = 2000;
1488# endif
1489# endif
1490
1491union PREFIX(fail_stack_elt)
1492{
1493 UCHAR_T *pointer;
1494 long int integer;
1495};
1496
1497typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1498
1499typedef struct
1500{
1501 PREFIX(fail_stack_elt_t) *stack;
1502 unsigned long int size;
1503 unsigned long int avail; /* Offset of next open position. */
1504} PREFIX(fail_stack_type);
1505
1506# else /* not INT_IS_16BIT */
1507
1508# ifndef DEFINED_ONCE
1509# if defined MATCH_MAY_ALLOCATE
1510/* 4400 was enough to cause a crash on Alpha OSF/1,
1511 whose default stack limit is 2mb. */
1512int re_max_failures = 4000;
1513# else
1514int re_max_failures = 2000;
1515# endif
1516# endif
1517
1518union PREFIX(fail_stack_elt)
1519{
1520 UCHAR_T *pointer;
1521 int integer;
1522};
1523
1524typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t);
1525
1526typedef struct
1527{
1528 PREFIX(fail_stack_elt_t) *stack;
1529 unsigned size;
1530 unsigned avail; /* Offset of next open position. */
1531} PREFIX(fail_stack_type);
1532
1533# endif /* INT_IS_16BIT */
1534
1535# ifndef DEFINED_ONCE
1536# define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1537# define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1538# define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1539# endif
1540
1541
1542/* Define macros to initialize and free the failure stack.
1543 Do `return -2' if the alloc fails. */
1544
1545# ifdef MATCH_MAY_ALLOCATE
1546# define INIT_FAIL_STACK() \
1547 do { \
1548 fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \
1549 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \
1550 \
1551 if (fail_stack.stack == NULL) \
1552 return -2; \
1553 \
1554 fail_stack.size = INIT_FAILURE_ALLOC; \
1555 fail_stack.avail = 0; \
1556 } while (0)
1557
1558# define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1559# else
1560# define INIT_FAIL_STACK() \
1561 do { \
1562 fail_stack.avail = 0; \
1563 } while (0)
1564
1565# define RESET_FAIL_STACK()
1566# endif
1567
1568
1569/* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1570
1571 Return 1 if succeeds, and 0 if either ran out of memory
1572 allocating space for it or it was already too large.
1573
1574 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1575
1576# define DOUBLE_FAIL_STACK(fail_stack) \
1577 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1578 ? 0 \
1579 : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \
1580 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1581 (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \
1582 ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\
1583 \
1584 (fail_stack).stack == NULL \
1585 ? 0 \
1586 : ((fail_stack).size <<= 1, \
1587 1)))
1588
1589
1590/* Push pointer POINTER on FAIL_STACK.
1591 Return 1 if was able to do so and 0 if ran out of memory allocating
1592 space to do so. */
1593# define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1594 ((FAIL_STACK_FULL () \
1595 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1596 ? 0 \
1597 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1598 1))
1599
1600/* Push a pointer value onto the failure stack.
1601 Assumes the variable `fail_stack'. Probably should only
1602 be called from within `PUSH_FAILURE_POINT'. */
1603# define PUSH_FAILURE_POINTER(item) \
1604 fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item)
1605
1606/* This pushes an integer-valued item onto the failure stack.
1607 Assumes the variable `fail_stack'. Probably should only
1608 be called from within `PUSH_FAILURE_POINT'. */
1609# define PUSH_FAILURE_INT(item) \
1610 fail_stack.stack[fail_stack.avail++].integer = (item)
1611
1612/* Push a fail_stack_elt_t value onto the failure stack.
1613 Assumes the variable `fail_stack'. Probably should only
1614 be called from within `PUSH_FAILURE_POINT'. */
1615# define PUSH_FAILURE_ELT(item) \
1616 fail_stack.stack[fail_stack.avail++] = (item)
1617
1618/* These three POP... operations complement the three PUSH... operations.
1619 All assume that `fail_stack' is nonempty. */
1620# define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1621# define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1622# define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1623
1624/* Used to omit pushing failure point id's when we're not debugging. */
1625# ifdef DEBUG
1626# define DEBUG_PUSH PUSH_FAILURE_INT
1627# define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1628# else
1629# define DEBUG_PUSH(item)
1630# define DEBUG_POP(item_addr)
1631# endif
1632
1633
1634/* Push the information about the state we will need
1635 if we ever fail back to it.
1636
1637 Requires variables fail_stack, regstart, regend, reg_info, and
1638 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1639 be declared.
1640
1641 Does `return FAILURE_CODE' if runs out of memory. */
1642
1643# define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1644 do { \
1645 char *destination; \
1646 /* Must be int, so when we don't save any registers, the arithmetic \
1647 of 0 + -1 isn't done as unsigned. */ \
1648 /* Can't be int, since there is not a shred of a guarantee that int \
1649 is wide enough to hold a value of something to which pointer can \
1650 be assigned */ \
1651 active_reg_t this_reg; \
1652 \
1653 DEBUG_STATEMENT (failure_id++); \
1654 DEBUG_STATEMENT (nfailure_points_pushed++); \
1655 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1656 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1657 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1658 \
1659 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1660 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1661 \
1662 /* Ensure we have enough space allocated for what we will push. */ \
1663 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1664 { \
1665 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1666 return failure_code; \
1667 \
1668 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1669 (fail_stack).size); \
1670 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1671 } \
1672 \
1673 /* Push the info, starting with the registers. */ \
1674 DEBUG_PRINT1 ("\n"); \
1675 \
1676 if (1) \
1677 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1678 this_reg++) \
1679 { \
1680 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1681 DEBUG_STATEMENT (num_regs_pushed++); \
1682 \
1683 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1684 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1685 \
1686 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1687 PUSH_FAILURE_POINTER (regend[this_reg]); \
1688 \
1689 DEBUG_PRINT2 (" info: %p\n ", \
1690 reg_info[this_reg].word.pointer); \
1691 DEBUG_PRINT2 (" match_null=%d", \
1692 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1693 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1694 DEBUG_PRINT2 (" matched_something=%d", \
1695 MATCHED_SOMETHING (reg_info[this_reg])); \
1696 DEBUG_PRINT2 (" ever_matched=%d", \
1697 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1698 DEBUG_PRINT1 ("\n"); \
1699 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1700 } \
1701 \
1702 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1703 PUSH_FAILURE_INT (lowest_active_reg); \
1704 \
1705 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1706 PUSH_FAILURE_INT (highest_active_reg); \
1707 \
1708 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1709 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1710 PUSH_FAILURE_POINTER (pattern_place); \
1711 \
1712 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1713 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1714 size2); \
1715 DEBUG_PRINT1 ("'\n"); \
1716 PUSH_FAILURE_POINTER (string_place); \
1717 \
1718 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1719 DEBUG_PUSH (failure_id); \
1720 } while (0)
1721
1722# ifndef DEFINED_ONCE
1723/* This is the number of items that are pushed and popped on the stack
1724 for each register. */
1725# define NUM_REG_ITEMS 3
1726
1727/* Individual items aside from the registers. */
1728# ifdef DEBUG
1729# define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1730# else
1731# define NUM_NONREG_ITEMS 4
1732# endif
1733
1734/* We push at most this many items on the stack. */
1735/* We used to use (num_regs - 1), which is the number of registers
1736 this regexp will save; but that was changed to 5
1737 to avoid stack overflow for a regexp with lots of parens. */
1738# define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1739
1740/* We actually push this many items. */
1741# define NUM_FAILURE_ITEMS \
1742 (((0 \
1743 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1744 * NUM_REG_ITEMS) \
1745 + NUM_NONREG_ITEMS)
1746
1747/* How many items can still be added to the stack without overflowing it. */
1748# define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1749# endif /* not DEFINED_ONCE */
1750
1751
1752/* Pops what PUSH_FAIL_STACK pushes.
1753
1754 We restore into the parameters, all of which should be lvalues:
1755 STR -- the saved data position.
1756 PAT -- the saved pattern position.
1757 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1758 REGSTART, REGEND -- arrays of string positions.
1759 REG_INFO -- array of information about each subexpression.
1760
1761 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1762 `pend', `string1', `size1', `string2', and `size2'. */
1763# define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1764{ \
1765 DEBUG_STATEMENT (unsigned failure_id;) \
1766 active_reg_t this_reg; \
1767 const UCHAR_T *string_temp; \
1768 \
1769 assert (!FAIL_STACK_EMPTY ()); \
1770 \
1771 /* Remove failure points and point to how many regs pushed. */ \
1772 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1773 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1774 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1775 \
1776 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1777 \
1778 DEBUG_POP (&failure_id); \
1779 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1780 \
1781 /* If the saved string location is NULL, it came from an \
1782 on_failure_keep_string_jump opcode, and we want to throw away the \
1783 saved NULL, thus retaining our current position in the string. */ \
1784 string_temp = POP_FAILURE_POINTER (); \
1785 if (string_temp != NULL) \
1786 str = (const CHAR_T *) string_temp; \
1787 \
1788 DEBUG_PRINT2 (" Popping string %p: `", str); \
1789 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1790 DEBUG_PRINT1 ("'\n"); \
1791 \
1792 pat = (UCHAR_T *) POP_FAILURE_POINTER (); \
1793 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1794 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1795 \
1796 /* Restore register info. */ \
1797 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1798 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1799 \
1800 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1801 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1802 \
1803 if (1) \
1804 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1805 { \
1806 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1807 \
1808 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1809 DEBUG_PRINT2 (" info: %p\n", \
1810 reg_info[this_reg].word.pointer); \
1811 \
1812 regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1813 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1814 \
1815 regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \
1816 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1817 } \
1818 else \
1819 { \
1820 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1821 { \
1822 reg_info[this_reg].word.integer = 0; \
1823 regend[this_reg] = 0; \
1824 regstart[this_reg] = 0; \
1825 } \
1826 highest_active_reg = high_reg; \
1827 } \
1828 \
1829 set_regs_matched_done = 0; \
1830 DEBUG_STATEMENT (nfailure_points_popped++); \
1831} /* POP_FAILURE_POINT */
1832\f
1833/* Structure for per-register (a.k.a. per-group) information.
1834 Other register information, such as the
1835 starting and ending positions (which are addresses), and the list of
1836 inner groups (which is a bits list) are maintained in separate
1837 variables.
1838
1839 We are making a (strictly speaking) nonportable assumption here: that
1840 the compiler will pack our bit fields into something that fits into
1841 the type of `word', i.e., is something that fits into one item on the
1842 failure stack. */
1843
1844
1845/* Declarations and macros for re_match_2. */
1846
1847typedef union
1848{
1849 PREFIX(fail_stack_elt_t) word;
1850 struct
1851 {
1852 /* This field is one if this group can match the empty string,
1853 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1854# define MATCH_NULL_UNSET_VALUE 3
1855 unsigned match_null_string_p : 2;
1856 unsigned is_active : 1;
1857 unsigned matched_something : 1;
1858 unsigned ever_matched_something : 1;
1859 } bits;
1860} PREFIX(register_info_type);
1861
1862# ifndef DEFINED_ONCE
1863# define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1864# define IS_ACTIVE(R) ((R).bits.is_active)
1865# define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1866# define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1867
1868
1869/* Call this when have matched a real character; it sets `matched' flags
1870 for the subexpressions which we are currently inside. Also records
1871 that those subexprs have matched. */
1872# define SET_REGS_MATCHED() \
1873 do \
1874 { \
1875 if (!set_regs_matched_done) \
1876 { \
1877 active_reg_t r; \
1878 set_regs_matched_done = 1; \
1879 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1880 { \
1881 MATCHED_SOMETHING (reg_info[r]) \
1882 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1883 = 1; \
1884 } \
1885 } \
1886 } \
1887 while (0)
1888# endif /* not DEFINED_ONCE */
1889
1890/* Registers are set to a sentinel when they haven't yet matched. */
1891static CHAR_T PREFIX(reg_unset_dummy);
1892# define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy))
1893# define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1894
1895/* Subroutine declarations and macros for regex_compile. */
1896static void PREFIX(store_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc, int arg));
1897static void PREFIX(store_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1898 int arg1, int arg2));
1899static void PREFIX(insert_op1) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1900 int arg, UCHAR_T *end));
1901static void PREFIX(insert_op2) _RE_ARGS ((re_opcode_t op, UCHAR_T *loc,
1902 int arg1, int arg2, UCHAR_T *end));
1903static boolean PREFIX(at_begline_loc_p) _RE_ARGS ((const CHAR_T *pattern,
1904 const CHAR_T *p,
1905 reg_syntax_t syntax));
1906static boolean PREFIX(at_endline_loc_p) _RE_ARGS ((const CHAR_T *p,
1907 const CHAR_T *pend,
1908 reg_syntax_t syntax));
1909# ifdef WCHAR
1910static reg_errcode_t wcs_compile_range _RE_ARGS ((CHAR_T range_start,
1911 const CHAR_T **p_ptr,
1912 const CHAR_T *pend,
1913 char *translate,
1914 reg_syntax_t syntax,
1915 UCHAR_T *b,
1916 CHAR_T *char_set));
1917static void insert_space _RE_ARGS ((int num, CHAR_T *loc, CHAR_T *end));
1918# else /* BYTE */
1919static reg_errcode_t byte_compile_range _RE_ARGS ((unsigned int range_start,
1920 const char **p_ptr,
1921 const char *pend,
1922 char *translate,
1923 reg_syntax_t syntax,
1924 unsigned char *b));
1925# endif /* WCHAR */
1926
1927/* Fetch the next character in the uncompiled pattern---translating it
1928 if necessary. Also cast from a signed character in the constant
1929 string passed to us by the user to an unsigned char that we can use
1930 as an array index (in, e.g., `translate'). */
1931/* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1932 because it is impossible to allocate 4GB array for some encodings
1933 which have 4 byte character_set like UCS4. */
1934# ifndef PATFETCH
1935# ifdef WCHAR
1936# define PATFETCH(c) \
1937 do {if (p == pend) return REG_EEND; \
1938 c = (UCHAR_T) *p++; \
1939 if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \
1940 } while (0)
1941# else /* BYTE */
1942# define PATFETCH(c) \
1943 do {if (p == pend) return REG_EEND; \
1944 c = (unsigned char) *p++; \
1945 if (translate) c = (unsigned char) translate[c]; \
1946 } while (0)
1947# endif /* WCHAR */
1948# endif
1949
1950/* Fetch the next character in the uncompiled pattern, with no
1951 translation. */
1952# define PATFETCH_RAW(c) \
1953 do {if (p == pend) return REG_EEND; \
1954 c = (UCHAR_T) *p++; \
1955 } while (0)
1956
1957/* Go backwards one character in the pattern. */
1958# define PATUNFETCH p--
1959
1960
1961/* If `translate' is non-null, return translate[D], else just D. We
1962 cast the subscript to translate because some data is declared as
1963 `char *', to avoid warnings when a string constant is passed. But
1964 when we use a character as a subscript we must make it unsigned. */
1965/* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1966 because it is impossible to allocate 4GB array for some encodings
1967 which have 4 byte character_set like UCS4. */
1968
1969# ifndef TRANSLATE
1970# ifdef WCHAR
1971# define TRANSLATE(d) \
1972 ((translate && ((UCHAR_T) (d)) <= 0xff) \
1973 ? (char) translate[(unsigned char) (d)] : (d))
1974# else /* BYTE */
1975# define TRANSLATE(d) \
1976 (translate ? (char) translate[(unsigned char) (d)] : (d))
1977# endif /* WCHAR */
1978# endif
1979
1980
1981/* Macros for outputting the compiled pattern into `buffer'. */
1982
1983/* If the buffer isn't allocated when it comes in, use this. */
1984# define INIT_BUF_SIZE (32 * sizeof(UCHAR_T))
1985
1986/* Make sure we have at least N more bytes of space in buffer. */
1987# ifdef WCHAR
1988# define GET_BUFFER_SPACE(n) \
1989 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
1990 + (n)*sizeof(CHAR_T)) > bufp->allocated) \
1991 EXTEND_BUFFER ()
1992# else /* BYTE */
1993# define GET_BUFFER_SPACE(n) \
1994 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1995 EXTEND_BUFFER ()
1996# endif /* WCHAR */
1997
1998/* Make sure we have one more byte of buffer space and then add C to it. */
1999# define BUF_PUSH(c) \
2000 do { \
2001 GET_BUFFER_SPACE (1); \
2002 *b++ = (UCHAR_T) (c); \
2003 } while (0)
2004
2005
2006/* Ensure we have two more bytes of buffer space and then append C1 and C2. */
2007# define BUF_PUSH_2(c1, c2) \
2008 do { \
2009 GET_BUFFER_SPACE (2); \
2010 *b++ = (UCHAR_T) (c1); \
2011 *b++ = (UCHAR_T) (c2); \
2012 } while (0)
2013
2014
2015/* As with BUF_PUSH_2, except for three bytes. */
2016# define BUF_PUSH_3(c1, c2, c3) \
2017 do { \
2018 GET_BUFFER_SPACE (3); \
2019 *b++ = (UCHAR_T) (c1); \
2020 *b++ = (UCHAR_T) (c2); \
2021 *b++ = (UCHAR_T) (c3); \
2022 } while (0)
2023
2024/* Store a jump with opcode OP at LOC to location TO. We store a
2025 relative address offset by the three bytes the jump itself occupies. */
2026# define STORE_JUMP(op, loc, to) \
2027 PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
2028
2029/* Likewise, for a two-argument jump. */
2030# define STORE_JUMP2(op, loc, to, arg) \
2031 PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
2032
2033/* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
2034# define INSERT_JUMP(op, loc, to) \
2035 PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
2036
2037/* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
2038# define INSERT_JUMP2(op, loc, to, arg) \
2039 PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
2040 arg, b)
2041
2042/* This is not an arbitrary limit: the arguments which represent offsets
2043 into the pattern are two bytes long. So if 2^16 bytes turns out to
2044 be too small, many things would have to change. */
2045/* Any other compiler which, like MSC, has allocation limit below 2^16
2046 bytes will have to use approach similar to what was done below for
2047 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
2048 reallocating to 0 bytes. Such thing is not going to work too well.
2049 You have been warned!! */
2050# ifndef DEFINED_ONCE
2051# if defined _MSC_VER && !defined WIN32
2052/* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
2053 The REALLOC define eliminates a flurry of conversion warnings,
2054 but is not required. */
2055# define MAX_BUF_SIZE 65500L
2056# define REALLOC(p,s) realloc ((p), (size_t) (s))
2057# else
2058# define MAX_BUF_SIZE (1L << 16)
2059# define REALLOC(p,s) realloc ((p), (s))
2060# endif
2061
2062/* Extend the buffer by twice its current size via realloc and
2063 reset the pointers that pointed into the old block to point to the
2064 correct places in the new one. If extending the buffer results in it
2065 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2066# if __BOUNDED_POINTERS__
2067# define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2068# define MOVE_BUFFER_POINTER(P) \
2069 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2070# define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2071 else \
2072 { \
2073 SET_HIGH_BOUND (b); \
2074 SET_HIGH_BOUND (begalt); \
2075 if (fixup_alt_jump) \
2076 SET_HIGH_BOUND (fixup_alt_jump); \
2077 if (laststart) \
2078 SET_HIGH_BOUND (laststart); \
2079 if (pending_exact) \
2080 SET_HIGH_BOUND (pending_exact); \
2081 }
2082# else
2083# define MOVE_BUFFER_POINTER(P) (P) += incr
2084# define ELSE_EXTEND_BUFFER_HIGH_BOUND
2085# endif
2086# endif /* not DEFINED_ONCE */
2087
2088# ifdef WCHAR
2089# define EXTEND_BUFFER() \
2090 do { \
2091 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2092 int wchar_count; \
2093 if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \
2094 return REG_ESIZE; \
2095 bufp->allocated <<= 1; \
2096 if (bufp->allocated > MAX_BUF_SIZE) \
2097 bufp->allocated = MAX_BUF_SIZE; \
2098 /* How many characters the new buffer can have? */ \
2099 wchar_count = bufp->allocated / sizeof(UCHAR_T); \
2100 if (wchar_count == 0) wchar_count = 1; \
2101 /* Truncate the buffer to CHAR_T align. */ \
2102 bufp->allocated = wchar_count * sizeof(UCHAR_T); \
2103 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \
2104 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2105 if (COMPILED_BUFFER_VAR == NULL) \
2106 return REG_ESPACE; \
2107 /* If the buffer moved, move all the pointers into it. */ \
2108 if (old_buffer != COMPILED_BUFFER_VAR) \
2109 { \
2110 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2111 MOVE_BUFFER_POINTER (b); \
2112 MOVE_BUFFER_POINTER (begalt); \
2113 if (fixup_alt_jump) \
2114 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2115 if (laststart) \
2116 MOVE_BUFFER_POINTER (laststart); \
2117 if (pending_exact) \
2118 MOVE_BUFFER_POINTER (pending_exact); \
2119 } \
2120 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2121 } while (0)
2122# else /* BYTE */
2123# define EXTEND_BUFFER() \
2124 do { \
2125 UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \
2126 if (bufp->allocated == MAX_BUF_SIZE) \
2127 return REG_ESIZE; \
2128 bufp->allocated <<= 1; \
2129 if (bufp->allocated > MAX_BUF_SIZE) \
2130 bufp->allocated = MAX_BUF_SIZE; \
2131 bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \
2132 bufp->allocated); \
2133 if (COMPILED_BUFFER_VAR == NULL) \
2134 return REG_ESPACE; \
2135 /* If the buffer moved, move all the pointers into it. */ \
2136 if (old_buffer != COMPILED_BUFFER_VAR) \
2137 { \
2138 int incr = COMPILED_BUFFER_VAR - old_buffer; \
2139 MOVE_BUFFER_POINTER (b); \
2140 MOVE_BUFFER_POINTER (begalt); \
2141 if (fixup_alt_jump) \
2142 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2143 if (laststart) \
2144 MOVE_BUFFER_POINTER (laststart); \
2145 if (pending_exact) \
2146 MOVE_BUFFER_POINTER (pending_exact); \
2147 } \
2148 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2149 } while (0)
2150# endif /* WCHAR */
2151
2152# ifndef DEFINED_ONCE
2153/* Since we have one byte reserved for the register number argument to
2154 {start,stop}_memory, the maximum number of groups we can report
2155 things about is what fits in that byte. */
2156# define MAX_REGNUM 255
2157
2158/* But patterns can have more than `MAX_REGNUM' registers. We just
2159 ignore the excess. */
2160typedef unsigned regnum_t;
2161
2162
2163/* Macros for the compile stack. */
2164
2165/* Since offsets can go either forwards or backwards, this type needs to
2166 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2167/* int may be not enough when sizeof(int) == 2. */
2168typedef long pattern_offset_t;
2169
2170typedef struct
2171{
2172 pattern_offset_t begalt_offset;
2173 pattern_offset_t fixup_alt_jump;
2174 pattern_offset_t inner_group_offset;
2175 pattern_offset_t laststart_offset;
2176 regnum_t regnum;
2177} compile_stack_elt_t;
2178
2179
2180typedef struct
2181{
2182 compile_stack_elt_t *stack;
2183 unsigned size;
2184 unsigned avail; /* Offset of next open position. */
2185} compile_stack_type;
2186
2187
2188# define INIT_COMPILE_STACK_SIZE 32
2189
2190# define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2191# define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2192
2193/* The next available element. */
2194# define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2195
2196# endif /* not DEFINED_ONCE */
2197
2198/* Set the bit for character C in a list. */
2199# ifndef DEFINED_ONCE
2200# define SET_LIST_BIT(c) \
2201 (b[((unsigned char) (c)) / BYTEWIDTH] \
2202 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2203# endif /* DEFINED_ONCE */
2204
2205/* Get the next unsigned number in the uncompiled pattern. */
2206# define GET_UNSIGNED_NUMBER(num) \
2207 { \
2208 while (p != pend) \
2209 { \
2210 PATFETCH (c); \
2211 if (c < '0' || c > '9') \
2212 break; \
2213 if (num <= RE_DUP_MAX) \
2214 { \
2215 if (num < 0) \
2216 num = 0; \
2217 num = num * 10 + c - '0'; \
2218 } \
2219 } \
2220 }
2221
2222# ifndef DEFINED_ONCE
2223# if defined _LIBC || WIDE_CHAR_SUPPORT
2224/* The GNU C library provides support for user-defined character classes
2225 and the functions from ISO C amendement 1. */
2226# ifdef CHARCLASS_NAME_MAX
2227# define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2228# else
2229/* This shouldn't happen but some implementation might still have this
2230 problem. Use a reasonable default value. */
2231# define CHAR_CLASS_MAX_LENGTH 256
2232# endif
2233
2234# ifdef _LIBC
2235# define IS_CHAR_CLASS(string) __wctype (string)
2236# else
2237# define IS_CHAR_CLASS(string) wctype (string)
2238# endif
2239# else
2240# define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2241
2242# define IS_CHAR_CLASS(string) \
2243 (STREQ (string, "alpha") || STREQ (string, "upper") \
2244 || STREQ (string, "lower") || STREQ (string, "digit") \
2245 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2246 || STREQ (string, "space") || STREQ (string, "print") \
2247 || STREQ (string, "punct") || STREQ (string, "graph") \
2248 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2249# endif
2250# endif /* DEFINED_ONCE */
2251\f
2252# ifndef MATCH_MAY_ALLOCATE
2253
2254/* If we cannot allocate large objects within re_match_2_internal,
2255 we make the fail stack and register vectors global.
2256 The fail stack, we grow to the maximum size when a regexp
2257 is compiled.
2258 The register vectors, we adjust in size each time we
2259 compile a regexp, according to the number of registers it needs. */
2260
2261static PREFIX(fail_stack_type) fail_stack;
2262
2263/* Size with which the following vectors are currently allocated.
2264 That is so we can make them bigger as needed,
2265 but never make them smaller. */
2266# ifdef DEFINED_ONCE
2267static int regs_allocated_size;
2268
2269static const char ** regstart, ** regend;
2270static const char ** old_regstart, ** old_regend;
2271static const char **best_regstart, **best_regend;
2272static const char **reg_dummy;
2273# endif /* DEFINED_ONCE */
2274
2275static PREFIX(register_info_type) *PREFIX(reg_info);
2276static PREFIX(register_info_type) *PREFIX(reg_info_dummy);
2277
2278/* Make the register vectors big enough for NUM_REGS registers,
2279 but don't make them smaller. */
2280
2281static void
2282PREFIX(regex_grow_registers) (num_regs)
2283 int num_regs;
2284{
2285 if (num_regs > regs_allocated_size)
2286 {
2287 RETALLOC_IF (regstart, num_regs, const char *);
2288 RETALLOC_IF (regend, num_regs, const char *);
2289 RETALLOC_IF (old_regstart, num_regs, const char *);
2290 RETALLOC_IF (old_regend, num_regs, const char *);
2291 RETALLOC_IF (best_regstart, num_regs, const char *);
2292 RETALLOC_IF (best_regend, num_regs, const char *);
2293 RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type));
2294 RETALLOC_IF (reg_dummy, num_regs, const char *);
2295 RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type));
2296
2297 regs_allocated_size = num_regs;
2298 }
2299}
2300
2301# endif /* not MATCH_MAY_ALLOCATE */
2302\f
2303# ifndef DEFINED_ONCE
2304static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2305 compile_stack,
2306 regnum_t regnum));
2307# endif /* not DEFINED_ONCE */
2308
2309/* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2310 Returns one of error codes defined in `regex.h', or zero for success.
2311
2312 Assumes the `allocated' (and perhaps `buffer') and `translate'
2313 fields are set in BUFP on entry.
2314
2315 If it succeeds, results are put in BUFP (if it returns an error, the
2316 contents of BUFP are undefined):
2317 `buffer' is the compiled pattern;
2318 `syntax' is set to SYNTAX;
2319 `used' is set to the length of the compiled pattern;
2320 `fastmap_accurate' is zero;
2321 `re_nsub' is the number of subexpressions in PATTERN;
2322 `not_bol' and `not_eol' are zero;
2323
2324 The `fastmap' and `newline_anchor' fields are neither
2325 examined nor set. */
2326
2327/* Return, freeing storage we allocated. */
2328# ifdef WCHAR
2329# define FREE_STACK_RETURN(value) \
2330 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2331# else
2332# define FREE_STACK_RETURN(value) \
2333 return (free (compile_stack.stack), value)
2334# endif /* WCHAR */
2335
2336static reg_errcode_t
2337PREFIX(regex_compile) (ARG_PREFIX(pattern), ARG_PREFIX(size), syntax, bufp)
2338 const char *ARG_PREFIX(pattern);
2339 size_t ARG_PREFIX(size);
2340 reg_syntax_t syntax;
2341 struct re_pattern_buffer *bufp;
2342{
2343 /* We fetch characters from PATTERN here. Even though PATTERN is
2344 `char *' (i.e., signed), we declare these variables as unsigned, so
2345 they can be reliably used as array indices. */
2346 register UCHAR_T c, c1;
2347
2348#ifdef WCHAR
2349 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2350 CHAR_T *pattern, *COMPILED_BUFFER_VAR;
2351 size_t size;
2352 /* offset buffer for optimization. See convert_mbs_to_wc. */
2353 int *mbs_offset = NULL;
2354 /* It hold whether each wchar_t is binary data or not. */
2355 char *is_binary = NULL;
2356 /* A flag whether exactn is handling binary data or not. */
2357 char is_exactn_bin = FALSE;
2358#endif /* WCHAR */
2359
2360 /* A random temporary spot in PATTERN. */
2361 const CHAR_T *p1;
2362
2363 /* Points to the end of the buffer, where we should append. */
2364 register UCHAR_T *b;
2365
2366 /* Keeps track of unclosed groups. */
2367 compile_stack_type compile_stack;
2368
2369 /* Points to the current (ending) position in the pattern. */
2370#ifdef WCHAR
2371 const CHAR_T *p;
2372 const CHAR_T *pend;
2373#else /* BYTE */
2374 const CHAR_T *p = pattern;
2375 const CHAR_T *pend = pattern + size;
2376#endif /* WCHAR */
2377
2378 /* How to translate the characters in the pattern. */
2379 RE_TRANSLATE_TYPE translate = bufp->translate;
2380
2381 /* Address of the count-byte of the most recently inserted `exactn'
2382 command. This makes it possible to tell if a new exact-match
2383 character can be added to that command or if the character requires
2384 a new `exactn' command. */
2385 UCHAR_T *pending_exact = 0;
2386
2387 /* Address of start of the most recently finished expression.
2388 This tells, e.g., postfix * where to find the start of its
2389 operand. Reset at the beginning of groups and alternatives. */
2390 UCHAR_T *laststart = 0;
2391
2392 /* Address of beginning of regexp, or inside of last group. */
2393 UCHAR_T *begalt;
2394
2395 /* Address of the place where a forward jump should go to the end of
2396 the containing expression. Each alternative of an `or' -- except the
2397 last -- ends with a forward jump of this sort. */
2398 UCHAR_T *fixup_alt_jump = 0;
2399
2400 /* Counts open-groups as they are encountered. Remembered for the
2401 matching close-group on the compile stack, so the same register
2402 number is put in the stop_memory as the start_memory. */
2403 regnum_t regnum = 0;
2404
2405#ifdef WCHAR
2406 /* Initialize the wchar_t PATTERN and offset_buffer. */
2407 p = pend = pattern = TALLOC(csize + 1, CHAR_T);
2408 mbs_offset = TALLOC(csize + 1, int);
2409 is_binary = TALLOC(csize + 1, char);
2410 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2411 {
2412 free(pattern);
2413 free(mbs_offset);
2414 free(is_binary);
2415 return REG_ESPACE;
2416 }
2417 pattern[csize] = L'\0'; /* sentinel */
2418 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2419 pend = p + size;
2420 if (size < 0)
2421 {
2422 free(pattern);
2423 free(mbs_offset);
2424 free(is_binary);
2425 return REG_BADPAT;
2426 }
2427#endif
2428
2429#ifdef DEBUG
2430 DEBUG_PRINT1 ("\nCompiling pattern: ");
2431 if (debug)
2432 {
2433 unsigned debug_count;
2434
2435 for (debug_count = 0; debug_count < size; debug_count++)
2436 PUT_CHAR (pattern[debug_count]);
2437 putchar ('\n');
2438 }
2439#endif /* DEBUG */
2440
2441 /* Initialize the compile stack. */
2442 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2443 if (compile_stack.stack == NULL)
2444 {
2445#ifdef WCHAR
2446 free(pattern);
2447 free(mbs_offset);
2448 free(is_binary);
2449#endif
2450 return REG_ESPACE;
2451 }
2452
2453 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2454 compile_stack.avail = 0;
2455
2456 /* Initialize the pattern buffer. */
2457 bufp->syntax = syntax;
2458 bufp->fastmap_accurate = 0;
2459 bufp->not_bol = bufp->not_eol = 0;
2460
2461 /* Set `used' to zero, so that if we return an error, the pattern
2462 printer (for debugging) will think there's no pattern. We reset it
2463 at the end. */
2464 bufp->used = 0;
2465
2466 /* Always count groups, whether or not bufp->no_sub is set. */
2467 bufp->re_nsub = 0;
2468
2469#if !defined emacs && !defined SYNTAX_TABLE
2470 /* Initialize the syntax table. */
2471 init_syntax_once ();
2472#endif
2473
2474 if (bufp->allocated == 0)
2475 {
2476 if (bufp->buffer)
2477 { /* If zero allocated, but buffer is non-null, try to realloc
2478 enough space. This loses if buffer's address is bogus, but
2479 that is the user's responsibility. */
2480#ifdef WCHAR
2481 /* Free bufp->buffer and allocate an array for wchar_t pattern
2482 buffer. */
2483 free(bufp->buffer);
2484 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T),
2485 UCHAR_T);
2486#else
2487 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T);
2488#endif /* WCHAR */
2489 }
2490 else
2491 { /* Caller did not allocate a buffer. Do it for them. */
2492 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T),
2493 UCHAR_T);
2494 }
2495
2496 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2497#ifdef WCHAR
2498 bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2499#endif /* WCHAR */
2500 bufp->allocated = INIT_BUF_SIZE;
2501 }
2502#ifdef WCHAR
2503 else
2504 COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer;
2505#endif
2506
2507 begalt = b = COMPILED_BUFFER_VAR;
2508
2509 /* Loop through the uncompiled pattern until we're at the end. */
2510 while (p != pend)
2511 {
2512 PATFETCH (c);
2513
2514 switch (c)
2515 {
2516 case '^':
2517 {
2518 if ( /* If at start of pattern, it's an operator. */
2519 p == pattern + 1
2520 /* If context independent, it's an operator. */
2521 || syntax & RE_CONTEXT_INDEP_ANCHORS
2522 /* Otherwise, depends on what's come before. */
2523 || PREFIX(at_begline_loc_p) (pattern, p, syntax))
2524 BUF_PUSH (begline);
2525 else
2526 goto normal_char;
2527 }
2528 break;
2529
2530
2531 case '$':
2532 {
2533 if ( /* If at end of pattern, it's an operator. */
2534 p == pend
2535 /* If context independent, it's an operator. */
2536 || syntax & RE_CONTEXT_INDEP_ANCHORS
2537 /* Otherwise, depends on what's next. */
2538 || PREFIX(at_endline_loc_p) (p, pend, syntax))
2539 BUF_PUSH (endline);
2540 else
2541 goto normal_char;
2542 }
2543 break;
2544
2545
2546 case '+':
2547 case '?':
2548 if ((syntax & RE_BK_PLUS_QM)
2549 || (syntax & RE_LIMITED_OPS))
2550 goto normal_char;
2551 handle_plus:
2552 case '*':
2553 /* If there is no previous pattern... */
2554 if (!laststart)
2555 {
2556 if (syntax & RE_CONTEXT_INVALID_OPS)
2557 FREE_STACK_RETURN (REG_BADRPT);
2558 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2559 goto normal_char;
2560 }
2561
2562 {
2563 /* Are we optimizing this jump? */
2564 boolean keep_string_p = false;
2565
2566 /* 1 means zero (many) matches is allowed. */
2567 char zero_times_ok = 0, many_times_ok = 0;
2568
2569 /* If there is a sequence of repetition chars, collapse it
2570 down to just one (the right one). We can't combine
2571 interval operators with these because of, e.g., `a{2}*',
2572 which should only match an even number of `a's. */
2573
2574 for (;;)
2575 {
2576 zero_times_ok |= c != '+';
2577 many_times_ok |= c != '?';
2578
2579 if (p == pend)
2580 break;
2581
2582 PATFETCH (c);
2583
2584 if (c == '*'
2585 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2586 ;
2587
2588 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2589 {
2590 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2591
2592 PATFETCH (c1);
2593 if (!(c1 == '+' || c1 == '?'))
2594 {
2595 PATUNFETCH;
2596 PATUNFETCH;
2597 break;
2598 }
2599
2600 c = c1;
2601 }
2602 else
2603 {
2604 PATUNFETCH;
2605 break;
2606 }
2607
2608 /* If we get here, we found another repeat character. */
2609 }
2610
2611 /* Star, etc. applied to an empty pattern is equivalent
2612 to an empty pattern. */
2613 if (!laststart)
2614 break;
2615
2616 /* Now we know whether or not zero matches is allowed
2617 and also whether or not two or more matches is allowed. */
2618 if (many_times_ok)
2619 { /* More than one repetition is allowed, so put in at the
2620 end a backward relative jump from `b' to before the next
2621 jump we're going to put in below (which jumps from
2622 laststart to after this jump).
2623
2624 But if we are at the `*' in the exact sequence `.*\n',
2625 insert an unconditional jump backwards to the .,
2626 instead of the beginning of the loop. This way we only
2627 push a failure point once, instead of every time
2628 through the loop. */
2629 assert (p - 1 > pattern);
2630
2631 /* Allocate the space for the jump. */
2632 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2633
2634 /* We know we are not at the first character of the pattern,
2635 because laststart was nonzero. And we've already
2636 incremented `p', by the way, to be the character after
2637 the `*'. Do we have to do something analogous here
2638 for null bytes, because of RE_DOT_NOT_NULL? */
2639 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2640 && zero_times_ok
2641 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2642 && !(syntax & RE_DOT_NEWLINE))
2643 { /* We have .*\n. */
2644 STORE_JUMP (jump, b, laststart);
2645 keep_string_p = true;
2646 }
2647 else
2648 /* Anything else. */
2649 STORE_JUMP (maybe_pop_jump, b, laststart -
2650 (1 + OFFSET_ADDRESS_SIZE));
2651
2652 /* We've added more stuff to the buffer. */
2653 b += 1 + OFFSET_ADDRESS_SIZE;
2654 }
2655
2656 /* On failure, jump from laststart to b + 3, which will be the
2657 end of the buffer after this jump is inserted. */
2658 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2659 'b + 3'. */
2660 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2661 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2662 : on_failure_jump,
2663 laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2664 pending_exact = 0;
2665 b += 1 + OFFSET_ADDRESS_SIZE;
2666
2667 if (!zero_times_ok)
2668 {
2669 /* At least one repetition is required, so insert a
2670 `dummy_failure_jump' before the initial
2671 `on_failure_jump' instruction of the loop. This
2672 effects a skip over that instruction the first time
2673 we hit that loop. */
2674 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2675 INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2676 2 + 2 * OFFSET_ADDRESS_SIZE);
2677 b += 1 + OFFSET_ADDRESS_SIZE;
2678 }
2679 }
2680 break;
2681
2682
2683 case '.':
2684 laststart = b;
2685 BUF_PUSH (anychar);
2686 break;
2687
2688
2689 case '[':
2690 {
2691 boolean had_char_class = false;
2692#ifdef WCHAR
2693 CHAR_T range_start = 0xffffffff;
2694#else
2695 unsigned int range_start = 0xffffffff;
2696#endif
2697 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2698
2699#ifdef WCHAR
2700 /* We assume a charset(_not) structure as a wchar_t array.
2701 charset[0] = (re_opcode_t) charset(_not)
2702 charset[1] = l (= length of char_classes)
2703 charset[2] = m (= length of collating_symbols)
2704 charset[3] = n (= length of equivalence_classes)
2705 charset[4] = o (= length of char_ranges)
2706 charset[5] = p (= length of chars)
2707
2708 charset[6] = char_class (wctype_t)
2709 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2710 ...
2711 charset[l+5] = char_class (wctype_t)
2712
2713 charset[l+6] = collating_symbol (wchar_t)
2714 ...
2715 charset[l+m+5] = collating_symbol (wchar_t)
2716 ifdef _LIBC we use the index if
2717 _NL_COLLATE_SYMB_EXTRAMB instead of
2718 wchar_t string.
2719
2720 charset[l+m+6] = equivalence_classes (wchar_t)
2721 ...
2722 charset[l+m+n+5] = equivalence_classes (wchar_t)
2723 ifdef _LIBC we use the index in
2724 _NL_COLLATE_WEIGHT instead of
2725 wchar_t string.
2726
2727 charset[l+m+n+6] = range_start
2728 charset[l+m+n+7] = range_end
2729 ...
2730 charset[l+m+n+2o+4] = range_start
2731 charset[l+m+n+2o+5] = range_end
2732 ifdef _LIBC we use the value looked up
2733 in _NL_COLLATE_COLLSEQ instead of
2734 wchar_t character.
2735
2736 charset[l+m+n+2o+6] = char
2737 ...
2738 charset[l+m+n+2o+p+5] = char
2739
2740 */
2741
2742 /* We need at least 6 spaces: the opcode, the length of
2743 char_classes, the length of collating_symbols, the length of
2744 equivalence_classes, the length of char_ranges, the length of
2745 chars. */
2746 GET_BUFFER_SPACE (6);
2747
2748 /* Save b as laststart. And We use laststart as the pointer
2749 to the first element of the charset here.
2750 In other words, laststart[i] indicates charset[i]. */
2751 laststart = b;
2752
2753 /* We test `*p == '^' twice, instead of using an if
2754 statement, so we only need one BUF_PUSH. */
2755 BUF_PUSH (*p == '^' ? charset_not : charset);
2756 if (*p == '^')
2757 p++;
2758
2759 /* Push the length of char_classes, the length of
2760 collating_symbols, the length of equivalence_classes, the
2761 length of char_ranges and the length of chars. */
2762 BUF_PUSH_3 (0, 0, 0);
2763 BUF_PUSH_2 (0, 0);
2764
2765 /* Remember the first position in the bracket expression. */
2766 p1 = p;
2767
2768 /* charset_not matches newline according to a syntax bit. */
2769 if ((re_opcode_t) b[-6] == charset_not
2770 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2771 {
2772 BUF_PUSH('\n');
2773 laststart[5]++; /* Update the length of characters */
2774 }
2775
2776 /* Read in characters and ranges, setting map bits. */
2777 for (;;)
2778 {
2779 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2780
2781 PATFETCH (c);
2782
2783 /* \ might escape characters inside [...] and [^...]. */
2784 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2785 {
2786 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2787
2788 PATFETCH (c1);
2789 BUF_PUSH(c1);
2790 laststart[5]++; /* Update the length of chars */
2791 range_start = c1;
2792 continue;
2793 }
2794
2795 /* Could be the end of the bracket expression. If it's
2796 not (i.e., when the bracket expression is `[]' so
2797 far), the ']' character bit gets set way below. */
2798 if (c == ']' && p != p1 + 1)
2799 break;
2800
2801 /* Look ahead to see if it's a range when the last thing
2802 was a character class. */
2803 if (had_char_class && c == '-' && *p != ']')
2804 FREE_STACK_RETURN (REG_ERANGE);
2805
2806 /* Look ahead to see if it's a range when the last thing
2807 was a character: if this is a hyphen not at the
2808 beginning or the end of a list, then it's the range
2809 operator. */
2810 if (c == '-'
2811 && !(p - 2 >= pattern && p[-2] == '[')
2812 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2813 && *p != ']')
2814 {
2815 reg_errcode_t ret;
2816 /* Allocate the space for range_start and range_end. */
2817 GET_BUFFER_SPACE (2);
2818 /* Update the pointer to indicate end of buffer. */
2819 b += 2;
2820 ret = wcs_compile_range (range_start, &p, pend, translate,
2821 syntax, b, laststart);
2822 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2823 range_start = 0xffffffff;
2824 }
2825 else if (p[0] == '-' && p[1] != ']')
2826 { /* This handles ranges made up of characters only. */
2827 reg_errcode_t ret;
2828
2829 /* Move past the `-'. */
2830 PATFETCH (c1);
2831 /* Allocate the space for range_start and range_end. */
2832 GET_BUFFER_SPACE (2);
2833 /* Update the pointer to indicate end of buffer. */
2834 b += 2;
2835 ret = wcs_compile_range (c, &p, pend, translate, syntax, b,
2836 laststart);
2837 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2838 range_start = 0xffffffff;
2839 }
2840
2841 /* See if we're at the beginning of a possible character
2842 class. */
2843 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2844 { /* Leave room for the null. */
2845 char str[CHAR_CLASS_MAX_LENGTH + 1];
2846
2847 PATFETCH (c);
2848 c1 = 0;
2849
2850 /* If pattern is `[[:'. */
2851 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2852
2853 for (;;)
2854 {
2855 PATFETCH (c);
2856 if ((c == ':' && *p == ']') || p == pend)
2857 break;
2858 if (c1 < CHAR_CLASS_MAX_LENGTH)
2859 str[c1++] = c;
2860 else
2861 /* This is in any case an invalid class name. */
2862 str[0] = '\0';
2863 }
2864 str[c1] = '\0';
2865
2866 /* If isn't a word bracketed by `[:' and `:]':
2867 undo the ending character, the letters, and leave
2868 the leading `:' and `[' (but store them as character). */
2869 if (c == ':' && *p == ']')
2870 {
2871 wctype_t wt;
2872 uintptr_t alignedp;
2873
2874 /* Query the character class as wctype_t. */
2875 wt = IS_CHAR_CLASS (str);
2876 if (wt == 0)
2877 FREE_STACK_RETURN (REG_ECTYPE);
2878
2879 /* Throw away the ] at the end of the character
2880 class. */
2881 PATFETCH (c);
2882
2883 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2884
2885 /* Allocate the space for character class. */
2886 GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2887 /* Update the pointer to indicate end of buffer. */
2888 b += CHAR_CLASS_SIZE;
2889 /* Move data which follow character classes
2890 not to violate the data. */
2891 insert_space(CHAR_CLASS_SIZE,
2892 laststart + 6 + laststart[1],
2893 b - 1);
2894 alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2895 + __alignof__(wctype_t) - 1)
2896 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2897 /* Store the character class. */
2898 *((wctype_t*)alignedp) = wt;
2899 /* Update length of char_classes */
2900 laststart[1] += CHAR_CLASS_SIZE;
2901
2902 had_char_class = true;
2903 }
2904 else
2905 {
2906 c1++;
2907 while (c1--)
2908 PATUNFETCH;
2909 BUF_PUSH ('[');
2910 BUF_PUSH (':');
2911 laststart[5] += 2; /* Update the length of characters */
2912 range_start = ':';
2913 had_char_class = false;
2914 }
2915 }
2916 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2917 || *p == '.'))
2918 {
2919 CHAR_T str[128]; /* Should be large enough. */
2920 CHAR_T delim = *p; /* '=' or '.' */
2921# ifdef _LIBC
2922 uint32_t nrules =
2923 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2924# endif
2925 PATFETCH (c);
2926 c1 = 0;
2927
2928 /* If pattern is `[[=' or '[[.'. */
2929 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2930
2931 for (;;)
2932 {
2933 PATFETCH (c);
2934 if ((c == delim && *p == ']') || p == pend)
2935 break;
2936 if (c1 < sizeof (str) - 1)
2937 str[c1++] = c;
2938 else
2939 /* This is in any case an invalid class name. */
2940 str[0] = '\0';
2941 }
2942 str[c1] = '\0';
2943
2944 if (c == delim && *p == ']' && str[0] != '\0')
2945 {
2946 unsigned int i, offset;
2947 /* If we have no collation data we use the default
2948 collation in which each character is in a class
2949 by itself. It also means that ASCII is the
2950 character set and therefore we cannot have character
2951 with more than one byte in the multibyte
2952 representation. */
2953
2954 /* If not defined _LIBC, we push the name and
2955 `\0' for the sake of matching performance. */
2956 int datasize = c1 + 1;
2957
2958# ifdef _LIBC
2959 int32_t idx = 0;
2960 if (nrules == 0)
2961# endif
2962 {
2963 if (c1 != 1)
2964 FREE_STACK_RETURN (REG_ECOLLATE);
2965 }
2966# ifdef _LIBC
2967 else
2968 {
2969 const int32_t *table;
2970 const int32_t *weights;
2971 const int32_t *extra;
2972 const int32_t *indirect;
2973 wint_t *cp;
2974
2975 /* This #include defines a local function! */
2976# include <locale/weightwc.h>
2977
2978 if(delim == '=')
2979 {
2980 /* We push the index for equivalence class. */
2981 cp = (wint_t*)str;
2982
2983 table = (const int32_t *)
2984 _NL_CURRENT (LC_COLLATE,
2985 _NL_COLLATE_TABLEWC);
2986 weights = (const int32_t *)
2987 _NL_CURRENT (LC_COLLATE,
2988 _NL_COLLATE_WEIGHTWC);
2989 extra = (const int32_t *)
2990 _NL_CURRENT (LC_COLLATE,
2991 _NL_COLLATE_EXTRAWC);
2992 indirect = (const int32_t *)
2993 _NL_CURRENT (LC_COLLATE,
2994 _NL_COLLATE_INDIRECTWC);
2995
2996 idx = findidx ((const wint_t**)&cp);
2997 if (idx == 0 || cp < (wint_t*) str + c1)
2998 /* This is no valid character. */
2999 FREE_STACK_RETURN (REG_ECOLLATE);
3000
3001 str[0] = (wchar_t)idx;
3002 }
3003 else /* delim == '.' */
3004 {
3005 /* We push collation sequence value
3006 for collating symbol. */
3007 int32_t table_size;
3008 const int32_t *symb_table;
3009 const unsigned char *extra;
3010 int32_t idx;
3011 int32_t elem;
3012 int32_t second;
3013 int32_t hash;
3014 char char_str[c1];
3015
3016 /* We have to convert the name to a single-byte
3017 string. This is possible since the names
3018 consist of ASCII characters and the internal
3019 representation is UCS4. */
3020 for (i = 0; i < c1; ++i)
3021 char_str[i] = str[i];
3022
3023 table_size =
3024 _NL_CURRENT_WORD (LC_COLLATE,
3025 _NL_COLLATE_SYMB_HASH_SIZEMB);
3026 symb_table = (const int32_t *)
3027 _NL_CURRENT (LC_COLLATE,
3028 _NL_COLLATE_SYMB_TABLEMB);
3029 extra = (const unsigned char *)
3030 _NL_CURRENT (LC_COLLATE,
3031 _NL_COLLATE_SYMB_EXTRAMB);
3032
3033 /* Locate the character in the hashing table. */
3034 hash = elem_hash (char_str, c1);
3035
3036 idx = 0;
3037 elem = hash % table_size;
3038 second = hash % (table_size - 2);
3039 while (symb_table[2 * elem] != 0)
3040 {
3041 /* First compare the hashing value. */
3042 if (symb_table[2 * elem] == hash
3043 && c1 == extra[symb_table[2 * elem + 1]]
3044 && memcmp (str,
3045 &extra[symb_table[2 * elem + 1]
3046 + 1], c1) == 0)
3047 {
3048 /* Yep, this is the entry. */
3049 idx = symb_table[2 * elem + 1];
3050 idx += 1 + extra[idx];
3051 break;
3052 }
3053
3054 /* Next entry. */
3055 elem += second;
3056 }
3057
3058 if (symb_table[2 * elem] != 0)
3059 {
3060 /* Compute the index of the byte sequence
3061 in the table. */
3062 idx += 1 + extra[idx];
3063 /* Adjust for the alignment. */
3064 idx = (idx + 3) & ~4;
3065
3066 str[0] = (wchar_t) idx + 4;
3067 }
3068 else if (symb_table[2 * elem] == 0 && c1 == 1)
3069 {
3070 /* No valid character. Match it as a
3071 single byte character. */
3072 had_char_class = false;
3073 BUF_PUSH(str[0]);
3074 /* Update the length of characters */
3075 laststart[5]++;
3076 range_start = str[0];
3077
3078 /* Throw away the ] at the end of the
3079 collating symbol. */
3080 PATFETCH (c);
3081 /* exit from the switch block. */
3082 continue;
3083 }
3084 else
3085 FREE_STACK_RETURN (REG_ECOLLATE);
3086 }
3087 datasize = 1;
3088 }
3089# endif
3090 /* Throw away the ] at the end of the equivalence
3091 class (or collating symbol). */
3092 PATFETCH (c);
3093
3094 /* Allocate the space for the equivalence class
3095 (or collating symbol) (and '\0' if needed). */
3096 GET_BUFFER_SPACE(datasize);
3097 /* Update the pointer to indicate end of buffer. */
3098 b += datasize;
3099
3100 if (delim == '=')
3101 { /* equivalence class */
3102 /* Calculate the offset of char_ranges,
3103 which is next to equivalence_classes. */
3104 offset = laststart[1] + laststart[2]
3105 + laststart[3] +6;
3106 /* Insert space. */
3107 insert_space(datasize, laststart + offset, b - 1);
3108
3109 /* Write the equivalence_class and \0. */
3110 for (i = 0 ; i < datasize ; i++)
3111 laststart[offset + i] = str[i];
3112
3113 /* Update the length of equivalence_classes. */
3114 laststart[3] += datasize;
3115 had_char_class = true;
3116 }
3117 else /* delim == '.' */
3118 { /* collating symbol */
3119 /* Calculate the offset of the equivalence_classes,
3120 which is next to collating_symbols. */
3121 offset = laststart[1] + laststart[2] + 6;
3122 /* Insert space and write the collationg_symbol
3123 and \0. */
3124 insert_space(datasize, laststart + offset, b-1);
3125 for (i = 0 ; i < datasize ; i++)
3126 laststart[offset + i] = str[i];
3127
3128 /* In re_match_2_internal if range_start < -1, we
3129 assume -range_start is the offset of the
3130 collating symbol which is specified as
3131 the character of the range start. So we assign
3132 -(laststart[1] + laststart[2] + 6) to
3133 range_start. */
3134 range_start = -(laststart[1] + laststart[2] + 6);
3135 /* Update the length of collating_symbol. */
3136 laststart[2] += datasize;
3137 had_char_class = false;
3138 }
3139 }
3140 else
3141 {
3142 c1++;
3143 while (c1--)
3144 PATUNFETCH;
3145 BUF_PUSH ('[');
3146 BUF_PUSH (delim);
3147 laststart[5] += 2; /* Update the length of characters */
3148 range_start = delim;
3149 had_char_class = false;
3150 }
3151 }
3152 else
3153 {
3154 had_char_class = false;
3155 BUF_PUSH(c);
3156 laststart[5]++; /* Update the length of characters */
3157 range_start = c;
3158 }
3159 }
3160
3161#else /* BYTE */
3162 /* Ensure that we have enough space to push a charset: the
3163 opcode, the length count, and the bitset; 34 bytes in all. */
3164 GET_BUFFER_SPACE (34);
3165
3166 laststart = b;
3167
3168 /* We test `*p == '^' twice, instead of using an if
3169 statement, so we only need one BUF_PUSH. */
3170 BUF_PUSH (*p == '^' ? charset_not : charset);
3171 if (*p == '^')
3172 p++;
3173
3174 /* Remember the first position in the bracket expression. */
3175 p1 = p;
3176
3177 /* Push the number of bytes in the bitmap. */
3178 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3179
3180 /* Clear the whole map. */
3181 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3182
3183 /* charset_not matches newline according to a syntax bit. */
3184 if ((re_opcode_t) b[-2] == charset_not
3185 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3186 SET_LIST_BIT ('\n');
3187
3188 /* Read in characters and ranges, setting map bits. */
3189 for (;;)
3190 {
3191 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3192
3193 PATFETCH (c);
3194
3195 /* \ might escape characters inside [...] and [^...]. */
3196 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3197 {
3198 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3199
3200 PATFETCH (c1);
3201 SET_LIST_BIT (c1);
3202 range_start = c1;
3203 continue;
3204 }
3205
3206 /* Could be the end of the bracket expression. If it's
3207 not (i.e., when the bracket expression is `[]' so
3208 far), the ']' character bit gets set way below. */
3209 if (c == ']' && p != p1 + 1)
3210 break;
3211
3212 /* Look ahead to see if it's a range when the last thing
3213 was a character class. */
3214 if (had_char_class && c == '-' && *p != ']')
3215 FREE_STACK_RETURN (REG_ERANGE);
3216
3217 /* Look ahead to see if it's a range when the last thing
3218 was a character: if this is a hyphen not at the
3219 beginning or the end of a list, then it's the range
3220 operator. */
3221 if (c == '-'
3222 && !(p - 2 >= pattern && p[-2] == '[')
3223 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3224 && *p != ']')
3225 {
3226 reg_errcode_t ret
3227 = byte_compile_range (range_start, &p, pend, translate,
3228 syntax, b);
3229 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3230 range_start = 0xffffffff;
3231 }
3232
3233 else if (p[0] == '-' && p[1] != ']')
3234 { /* This handles ranges made up of characters only. */
3235 reg_errcode_t ret;
3236
3237 /* Move past the `-'. */
3238 PATFETCH (c1);
3239
3240 ret = byte_compile_range (c, &p, pend, translate, syntax, b);
3241 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3242 range_start = 0xffffffff;
3243 }
3244
3245 /* See if we're at the beginning of a possible character
3246 class. */
3247
3248 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3249 { /* Leave room for the null. */
3250 char str[CHAR_CLASS_MAX_LENGTH + 1];
3251
3252 PATFETCH (c);
3253 c1 = 0;
3254
3255 /* If pattern is `[[:'. */
3256 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3257
3258 for (;;)
3259 {
3260 PATFETCH (c);
3261 if ((c == ':' && *p == ']') || p == pend)
3262 break;
3263 if (c1 < CHAR_CLASS_MAX_LENGTH)
3264 str[c1++] = c;
3265 else
3266 /* This is in any case an invalid class name. */
3267 str[0] = '\0';
3268 }
3269 str[c1] = '\0';
3270
3271 /* If isn't a word bracketed by `[:' and `:]':
3272 undo the ending character, the letters, and leave
3273 the leading `:' and `[' (but set bits for them). */
3274 if (c == ':' && *p == ']')
3275 {
3276# if defined _LIBC || WIDE_CHAR_SUPPORT
3277 boolean is_lower = STREQ (str, "lower");
3278 boolean is_upper = STREQ (str, "upper");
3279 wctype_t wt;
3280 int ch;
3281
3282 wt = IS_CHAR_CLASS (str);
3283 if (wt == 0)
3284 FREE_STACK_RETURN (REG_ECTYPE);
3285
3286 /* Throw away the ] at the end of the character
3287 class. */
3288 PATFETCH (c);
3289
3290 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3291
3292 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3293 {
3294# ifdef _LIBC
3295 if (__iswctype (__btowc (ch), wt))
3296 SET_LIST_BIT (ch);
3297# else
3298 if (iswctype (btowc (ch), wt))
3299 SET_LIST_BIT (ch);
3300# endif
3301
3302 if (translate && (is_upper || is_lower)
3303 && (ISUPPER (ch) || ISLOWER (ch)))
3304 SET_LIST_BIT (ch);
3305 }
3306
3307 had_char_class = true;
3308# else
3309 int ch;
3310 boolean is_alnum = STREQ (str, "alnum");
3311 boolean is_alpha = STREQ (str, "alpha");
3312 boolean is_blank = STREQ (str, "blank");
3313 boolean is_cntrl = STREQ (str, "cntrl");
3314 boolean is_digit = STREQ (str, "digit");
3315 boolean is_graph = STREQ (str, "graph");
3316 boolean is_lower = STREQ (str, "lower");
3317 boolean is_print = STREQ (str, "print");
3318 boolean is_punct = STREQ (str, "punct");
3319 boolean is_space = STREQ (str, "space");
3320 boolean is_upper = STREQ (str, "upper");
3321 boolean is_xdigit = STREQ (str, "xdigit");
3322
3323 if (!IS_CHAR_CLASS (str))
3324 FREE_STACK_RETURN (REG_ECTYPE);
3325
3326 /* Throw away the ] at the end of the character
3327 class. */
3328 PATFETCH (c);
3329
3330 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3331
3332 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3333 {
3334 /* This was split into 3 if's to
3335 avoid an arbitrary limit in some compiler. */
3336 if ( (is_alnum && ISALNUM (ch))
3337 || (is_alpha && ISALPHA (ch))
3338 || (is_blank && ISBLANK (ch))
3339 || (is_cntrl && ISCNTRL (ch)))
3340 SET_LIST_BIT (ch);
3341 if ( (is_digit && ISDIGIT (ch))
3342 || (is_graph && ISGRAPH (ch))
3343 || (is_lower && ISLOWER (ch))
3344 || (is_print && ISPRINT (ch)))
3345 SET_LIST_BIT (ch);
3346 if ( (is_punct && ISPUNCT (ch))
3347 || (is_space && ISSPACE (ch))
3348 || (is_upper && ISUPPER (ch))
3349 || (is_xdigit && ISXDIGIT (ch)))
3350 SET_LIST_BIT (ch);
3351 if ( translate && (is_upper || is_lower)
3352 && (ISUPPER (ch) || ISLOWER (ch)))
3353 SET_LIST_BIT (ch);
3354 }
3355 had_char_class = true;
3356# endif /* libc || wctype.h */
3357 }
3358 else
3359 {
3360 c1++;
3361 while (c1--)
3362 PATUNFETCH;
3363 SET_LIST_BIT ('[');
3364 SET_LIST_BIT (':');
3365 range_start = ':';
3366 had_char_class = false;
3367 }
3368 }
3369 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3370 {
3371 unsigned char str[MB_LEN_MAX + 1];
3372# ifdef _LIBC
3373 uint32_t nrules =
3374 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3375# endif
3376
3377 PATFETCH (c);
3378 c1 = 0;
3379
3380 /* If pattern is `[[='. */
3381 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3382
3383 for (;;)
3384 {
3385 PATFETCH (c);
3386 if ((c == '=' && *p == ']') || p == pend)
3387 break;
3388 if (c1 < MB_LEN_MAX)
3389 str[c1++] = c;
3390 else
3391 /* This is in any case an invalid class name. */
3392 str[0] = '\0';
3393 }
3394 str[c1] = '\0';
3395
3396 if (c == '=' && *p == ']' && str[0] != '\0')
3397 {
3398 /* If we have no collation data we use the default
3399 collation in which each character is in a class
3400 by itself. It also means that ASCII is the
3401 character set and therefore we cannot have character
3402 with more than one byte in the multibyte
3403 representation. */
3404# ifdef _LIBC
3405 if (nrules == 0)
3406# endif
3407 {
3408 if (c1 != 1)
3409 FREE_STACK_RETURN (REG_ECOLLATE);
3410
3411 /* Throw away the ] at the end of the equivalence
3412 class. */
3413 PATFETCH (c);
3414
3415 /* Set the bit for the character. */
3416 SET_LIST_BIT (str[0]);
3417 }
3418# ifdef _LIBC
3419 else
3420 {
3421 /* Try to match the byte sequence in `str' against
3422 those known to the collate implementation.
3423 First find out whether the bytes in `str' are
3424 actually from exactly one character. */
3425 const int32_t *table;
3426 const unsigned char *weights;
3427 const unsigned char *extra;
3428 const int32_t *indirect;
3429 int32_t idx;
3430 const unsigned char *cp = str;
3431 int ch;
3432
3433 /* This #include defines a local function! */
3434# include <locale/weight.h>
3435
3436 table = (const int32_t *)
3437 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3438 weights = (const unsigned char *)
3439 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3440 extra = (const unsigned char *)
3441 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3442 indirect = (const int32_t *)
3443 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3444
3445 idx = findidx (&cp);
3446 if (idx == 0 || cp < str + c1)
3447 /* This is no valid character. */
3448 FREE_STACK_RETURN (REG_ECOLLATE);
3449
3450 /* Throw away the ] at the end of the equivalence
3451 class. */
3452 PATFETCH (c);
3453
3454 /* Now we have to go throught the whole table
3455 and find all characters which have the same
3456 first level weight.
3457
3458 XXX Note that this is not entirely correct.
3459 we would have to match multibyte sequences
3460 but this is not possible with the current
3461 implementation. */
3462 for (ch = 1; ch < 256; ++ch)
3463 /* XXX This test would have to be changed if we
3464 would allow matching multibyte sequences. */
3465 if (table[ch] > 0)
3466 {
3467 int32_t idx2 = table[ch];
3468 size_t len = weights[idx2];
3469
3470 /* Test whether the lenghts match. */
3471 if (weights[idx] == len)
3472 {
3473 /* They do. New compare the bytes of
3474 the weight. */
3475 size_t cnt = 0;
3476
3477 while (cnt < len
3478 && (weights[idx + 1 + cnt]
3479 == weights[idx2 + 1 + cnt]))
3480 ++cnt;
3481
3482 if (cnt == len)
3483 /* They match. Mark the character as
3484 acceptable. */
3485 SET_LIST_BIT (ch);
3486 }
3487 }
3488 }
3489# endif
3490 had_char_class = true;
3491 }
3492 else
3493 {
3494 c1++;
3495 while (c1--)
3496 PATUNFETCH;
3497 SET_LIST_BIT ('[');
3498 SET_LIST_BIT ('=');
3499 range_start = '=';
3500 had_char_class = false;
3501 }
3502 }
3503 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3504 {
3505 unsigned char str[128]; /* Should be large enough. */
3506# ifdef _LIBC
3507 uint32_t nrules =
3508 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3509# endif
3510
3511 PATFETCH (c);
3512 c1 = 0;
3513
3514 /* If pattern is `[[.'. */
3515 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3516
3517 for (;;)
3518 {
3519 PATFETCH (c);
3520 if ((c == '.' && *p == ']') || p == pend)
3521 break;
3522 if (c1 < sizeof (str))
3523 str[c1++] = c;
3524 else
3525 /* This is in any case an invalid class name. */
3526 str[0] = '\0';
3527 }
3528 str[c1] = '\0';
3529
3530 if (c == '.' && *p == ']' && str[0] != '\0')
3531 {
3532 /* If we have no collation data we use the default
3533 collation in which each character is the name
3534 for its own class which contains only the one
3535 character. It also means that ASCII is the
3536 character set and therefore we cannot have character
3537 with more than one byte in the multibyte
3538 representation. */
3539# ifdef _LIBC
3540 if (nrules == 0)
3541# endif
3542 {
3543 if (c1 != 1)
3544 FREE_STACK_RETURN (REG_ECOLLATE);
3545
3546 /* Throw away the ] at the end of the equivalence
3547 class. */
3548 PATFETCH (c);
3549
3550 /* Set the bit for the character. */
3551 SET_LIST_BIT (str[0]);
3552 range_start = ((const unsigned char *) str)[0];
3553 }
3554# ifdef _LIBC
3555 else
3556 {
3557 /* Try to match the byte sequence in `str' against
3558 those known to the collate implementation.
3559 First find out whether the bytes in `str' are
3560 actually from exactly one character. */
3561 int32_t table_size;
3562 const int32_t *symb_table;
3563 const unsigned char *extra;
3564 int32_t idx;
3565 int32_t elem;
3566 int32_t second;
3567 int32_t hash;
3568
3569 table_size =
3570 _NL_CURRENT_WORD (LC_COLLATE,
3571 _NL_COLLATE_SYMB_HASH_SIZEMB);
3572 symb_table = (const int32_t *)
3573 _NL_CURRENT (LC_COLLATE,
3574 _NL_COLLATE_SYMB_TABLEMB);
3575 extra = (const unsigned char *)
3576 _NL_CURRENT (LC_COLLATE,
3577 _NL_COLLATE_SYMB_EXTRAMB);
3578
3579 /* Locate the character in the hashing table. */
3580 hash = elem_hash (str, c1);
3581
3582 idx = 0;
3583 elem = hash % table_size;
3584 second = hash % (table_size - 2);
3585 while (symb_table[2 * elem] != 0)
3586 {
3587 /* First compare the hashing value. */
3588 if (symb_table[2 * elem] == hash
3589 && c1 == extra[symb_table[2 * elem + 1]]
3590 && memcmp (str,
3591 &extra[symb_table[2 * elem + 1]
3592 + 1],
3593 c1) == 0)
3594 {
3595 /* Yep, this is the entry. */
3596 idx = symb_table[2 * elem + 1];
3597 idx += 1 + extra[idx];
3598 break;
3599 }
3600
3601 /* Next entry. */
3602 elem += second;
3603 }
3604
3605 if (symb_table[2 * elem] == 0)
3606 /* This is no valid character. */
3607 FREE_STACK_RETURN (REG_ECOLLATE);
3608
3609 /* Throw away the ] at the end of the equivalence
3610 class. */
3611 PATFETCH (c);
3612
3613 /* Now add the multibyte character(s) we found
3614 to the accept list.
3615
3616 XXX Note that this is not entirely correct.
3617 we would have to match multibyte sequences
3618 but this is not possible with the current
3619 implementation. Also, we have to match
3620 collating symbols, which expand to more than
3621 one file, as a whole and not allow the
3622 individual bytes. */
3623 c1 = extra[idx++];
3624 if (c1 == 1)
3625 range_start = extra[idx];
3626 while (c1-- > 0)
3627 {
3628 SET_LIST_BIT (extra[idx]);
3629 ++idx;
3630 }
3631 }
3632# endif
3633 had_char_class = false;
3634 }
3635 else
3636 {
3637 c1++;
3638 while (c1--)
3639 PATUNFETCH;
3640 SET_LIST_BIT ('[');
3641 SET_LIST_BIT ('.');
3642 range_start = '.';
3643 had_char_class = false;
3644 }
3645 }
3646 else
3647 {
3648 had_char_class = false;
3649 SET_LIST_BIT (c);
3650 range_start = c;
3651 }
3652 }
3653
3654 /* Discard any (non)matching list bytes that are all 0 at the
3655 end of the map. Decrease the map-length byte too. */
3656 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3657 b[-1]--;
3658 b += b[-1];
3659#endif /* WCHAR */
3660 }
3661 break;
3662
3663
3664 case '(':
3665 if (syntax & RE_NO_BK_PARENS)
3666 goto handle_open;
3667 else
3668 goto normal_char;
3669
3670
3671 case ')':
3672 if (syntax & RE_NO_BK_PARENS)
3673 goto handle_close;
3674 else
3675 goto normal_char;
3676
3677
3678 case '\n':
3679 if (syntax & RE_NEWLINE_ALT)
3680 goto handle_alt;
3681 else
3682 goto normal_char;
3683
3684
3685 case '|':
3686 if (syntax & RE_NO_BK_VBAR)
3687 goto handle_alt;
3688 else
3689 goto normal_char;
3690
3691
3692 case '{':
3693 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3694 goto handle_interval;
3695 else
3696 goto normal_char;
3697
3698
3699 case '\\':
3700 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3701
3702 /* Do not translate the character after the \, so that we can
3703 distinguish, e.g., \B from \b, even if we normally would
3704 translate, e.g., B to b. */
3705 PATFETCH_RAW (c);
3706
3707 switch (c)
3708 {
3709 case '(':
3710 if (syntax & RE_NO_BK_PARENS)
3711 goto normal_backslash;
3712
3713 handle_open:
3714 bufp->re_nsub++;
3715 regnum++;
3716
3717 if (COMPILE_STACK_FULL)
3718 {
3719 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3720 compile_stack_elt_t);
3721 if (compile_stack.stack == NULL) return REG_ESPACE;
3722
3723 compile_stack.size <<= 1;
3724 }
3725
3726 /* These are the values to restore when we hit end of this
3727 group. They are all relative offsets, so that if the
3728 whole pattern moves because of realloc, they will still
3729 be valid. */
3730 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3731 COMPILE_STACK_TOP.fixup_alt_jump
3732 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3733 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3734 COMPILE_STACK_TOP.regnum = regnum;
3735
3736 /* We will eventually replace the 0 with the number of
3737 groups inner to this one. But do not push a
3738 start_memory for groups beyond the last one we can
3739 represent in the compiled pattern. */
3740 if (regnum <= MAX_REGNUM)
3741 {
3742 COMPILE_STACK_TOP.inner_group_offset = b
3743 - COMPILED_BUFFER_VAR + 2;
3744 BUF_PUSH_3 (start_memory, regnum, 0);
3745 }
3746
3747 compile_stack.avail++;
3748
3749 fixup_alt_jump = 0;
3750 laststart = 0;
3751 begalt = b;
3752 /* If we've reached MAX_REGNUM groups, then this open
3753 won't actually generate any code, so we'll have to
3754 clear pending_exact explicitly. */
3755 pending_exact = 0;
3756 break;
3757
3758
3759 case ')':
3760 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3761
3762 if (COMPILE_STACK_EMPTY)
3763 {
3764 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3765 goto normal_backslash;
3766 else
3767 FREE_STACK_RETURN (REG_ERPAREN);
3768 }
3769
3770 handle_close:
3771 if (fixup_alt_jump)
3772 { /* Push a dummy failure point at the end of the
3773 alternative for a possible future
3774 `pop_failure_jump' to pop. See comments at
3775 `push_dummy_failure' in `re_match_2'. */
3776 BUF_PUSH (push_dummy_failure);
3777
3778 /* We allocated space for this jump when we assigned
3779 to `fixup_alt_jump', in the `handle_alt' case below. */
3780 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3781 }
3782
3783 /* See similar code for backslashed left paren above. */
3784 if (COMPILE_STACK_EMPTY)
3785 {
3786 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3787 goto normal_char;
3788 else
3789 FREE_STACK_RETURN (REG_ERPAREN);
3790 }
3791
3792 /* Since we just checked for an empty stack above, this
3793 ``can't happen''. */
3794 assert (compile_stack.avail != 0);
3795 {
3796 /* We don't just want to restore into `regnum', because
3797 later groups should continue to be numbered higher,
3798 as in `(ab)c(de)' -- the second group is #2. */
3799 regnum_t this_group_regnum;
3800
3801 compile_stack.avail--;
3802 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3803 fixup_alt_jump
3804 = COMPILE_STACK_TOP.fixup_alt_jump
3805 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3806 : 0;
3807 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3808 this_group_regnum = COMPILE_STACK_TOP.regnum;
3809 /* If we've reached MAX_REGNUM groups, then this open
3810 won't actually generate any code, so we'll have to
3811 clear pending_exact explicitly. */
3812 pending_exact = 0;
3813
3814 /* We're at the end of the group, so now we know how many
3815 groups were inside this one. */
3816 if (this_group_regnum <= MAX_REGNUM)
3817 {
3818 UCHAR_T *inner_group_loc
3819 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3820
3821 *inner_group_loc = regnum - this_group_regnum;
3822 BUF_PUSH_3 (stop_memory, this_group_regnum,
3823 regnum - this_group_regnum);
3824 }
3825 }
3826 break;
3827
3828
3829 case '|': /* `\|'. */
3830 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3831 goto normal_backslash;
3832 handle_alt:
3833 if (syntax & RE_LIMITED_OPS)
3834 goto normal_char;
3835
3836 /* Insert before the previous alternative a jump which
3837 jumps to this alternative if the former fails. */
3838 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3839 INSERT_JUMP (on_failure_jump, begalt,
3840 b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3841 pending_exact = 0;
3842 b += 1 + OFFSET_ADDRESS_SIZE;
3843
3844 /* The alternative before this one has a jump after it
3845 which gets executed if it gets matched. Adjust that
3846 jump so it will jump to this alternative's analogous
3847 jump (put in below, which in turn will jump to the next
3848 (if any) alternative's such jump, etc.). The last such
3849 jump jumps to the correct final destination. A picture:
3850 _____ _____
3851 | | | |
3852 | v | v
3853 a | b | c
3854
3855 If we are at `b', then fixup_alt_jump right now points to a
3856 three-byte space after `a'. We'll put in the jump, set
3857 fixup_alt_jump to right after `b', and leave behind three
3858 bytes which we'll fill in when we get to after `c'. */
3859
3860 if (fixup_alt_jump)
3861 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3862
3863 /* Mark and leave space for a jump after this alternative,
3864 to be filled in later either by next alternative or
3865 when know we're at the end of a series of alternatives. */
3866 fixup_alt_jump = b;
3867 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3868 b += 1 + OFFSET_ADDRESS_SIZE;
3869
3870 laststart = 0;
3871 begalt = b;
3872 break;
3873
3874
3875 case '{':
3876 /* If \{ is a literal. */
3877 if (!(syntax & RE_INTERVALS)
3878 /* If we're at `\{' and it's not the open-interval
3879 operator. */
3880 || (syntax & RE_NO_BK_BRACES))
3881 goto normal_backslash;
3882
3883 handle_interval:
3884 {
3885 /* If got here, then the syntax allows intervals. */
3886
3887 /* At least (most) this many matches must be made. */
3888 int lower_bound = -1, upper_bound = -1;
3889
3890 /* Place in the uncompiled pattern (i.e., just after
3891 the '{') to go back to if the interval is invalid. */
3892 const CHAR_T *beg_interval = p;
3893
3894 if (p == pend)
3895 goto invalid_interval;
3896
3897 GET_UNSIGNED_NUMBER (lower_bound);
3898
3899 if (c == ',')
3900 {
3901 GET_UNSIGNED_NUMBER (upper_bound);
3902 if (upper_bound < 0)
3903 upper_bound = RE_DUP_MAX;
3904 }
3905 else
3906 /* Interval such as `{1}' => match exactly once. */
3907 upper_bound = lower_bound;
3908
3909 if (! (0 <= lower_bound && lower_bound <= upper_bound))
3910 goto invalid_interval;
3911
3912 if (!(syntax & RE_NO_BK_BRACES))
3913 {
3914 if (c != '\\' || p == pend)
3915 goto invalid_interval;
3916 PATFETCH (c);
3917 }
3918
3919 if (c != '}')
3920 goto invalid_interval;
3921
3922 /* If it's invalid to have no preceding re. */
3923 if (!laststart)
3924 {
3925 if (syntax & RE_CONTEXT_INVALID_OPS
3926 && !(syntax & RE_INVALID_INTERVAL_ORD))
3927 FREE_STACK_RETURN (REG_BADRPT);
3928 else if (syntax & RE_CONTEXT_INDEP_OPS)
3929 laststart = b;
3930 else
3931 goto unfetch_interval;
3932 }
3933
3934 /* We just parsed a valid interval. */
3935
3936 if (RE_DUP_MAX < upper_bound)
3937 FREE_STACK_RETURN (REG_BADBR);
3938
3939 /* If the upper bound is zero, don't want to succeed at
3940 all; jump from `laststart' to `b + 3', which will be
3941 the end of the buffer after we insert the jump. */
3942 /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE'
3943 instead of 'b + 3'. */
3944 if (upper_bound == 0)
3945 {
3946 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3947 INSERT_JUMP (jump, laststart, b + 1
3948 + OFFSET_ADDRESS_SIZE);
3949 b += 1 + OFFSET_ADDRESS_SIZE;
3950 }
3951
3952 /* Otherwise, we have a nontrivial interval. When
3953 we're all done, the pattern will look like:
3954 set_number_at <jump count> <upper bound>
3955 set_number_at <succeed_n count> <lower bound>
3956 succeed_n <after jump addr> <succeed_n count>
3957 <body of loop>
3958 jump_n <succeed_n addr> <jump count>
3959 (The upper bound and `jump_n' are omitted if
3960 `upper_bound' is 1, though.) */
3961 else
3962 { /* If the upper bound is > 1, we need to insert
3963 more at the end of the loop. */
3964 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3965 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3966
3967 GET_BUFFER_SPACE (nbytes);
3968
3969 /* Initialize lower bound of the `succeed_n', even
3970 though it will be set during matching by its
3971 attendant `set_number_at' (inserted next),
3972 because `re_compile_fastmap' needs to know.
3973 Jump to the `jump_n' we might insert below. */
3974 INSERT_JUMP2 (succeed_n, laststart,
3975 b + 1 + 2 * OFFSET_ADDRESS_SIZE
3976 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
3977 , lower_bound);
3978 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3979
3980 /* Code to initialize the lower bound. Insert
3981 before the `succeed_n'. The `5' is the last two
3982 bytes of this `set_number_at', plus 3 bytes of
3983 the following `succeed_n'. */
3984 /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE'
3985 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
3986 of the following `succeed_n'. */
3987 PREFIX(insert_op2) (set_number_at, laststart, 1
3988 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
3989 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3990
3991 if (upper_bound > 1)
3992 { /* More than one repetition is allowed, so
3993 append a backward jump to the `succeed_n'
3994 that starts this interval.
3995
3996 When we've reached this during matching,
3997 we'll have matched the interval once, so
3998 jump back only `upper_bound - 1' times. */
3999 STORE_JUMP2 (jump_n, b, laststart
4000 + 2 * OFFSET_ADDRESS_SIZE + 1,
4001 upper_bound - 1);
4002 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4003
4004 /* The location we want to set is the second
4005 parameter of the `jump_n'; that is `b-2' as
4006 an absolute address. `laststart' will be
4007 the `set_number_at' we're about to insert;
4008 `laststart+3' the number to set, the source
4009 for the relative address. But we are
4010 inserting into the middle of the pattern --
4011 so everything is getting moved up by 5.
4012 Conclusion: (b - 2) - (laststart + 3) + 5,
4013 i.e., b - laststart.
4014
4015 We insert this at the beginning of the loop
4016 so that if we fail during matching, we'll
4017 reinitialize the bounds. */
4018 PREFIX(insert_op2) (set_number_at, laststart,
4019 b - laststart,
4020 upper_bound - 1, b);
4021 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
4022 }
4023 }
4024 pending_exact = 0;
4025 break;
4026
4027 invalid_interval:
4028 if (!(syntax & RE_INVALID_INTERVAL_ORD))
4029 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
4030 unfetch_interval:
4031 /* Match the characters as literals. */
4032 p = beg_interval;
4033 c = '{';
4034 if (syntax & RE_NO_BK_BRACES)
4035 goto normal_char;
4036 else
4037 goto normal_backslash;
4038 }
4039
4040#ifdef emacs
4041 /* There is no way to specify the before_dot and after_dot
4042 operators. rms says this is ok. --karl */
4043 case '=':
4044 BUF_PUSH (at_dot);
4045 break;
4046
4047 case 's':
4048 laststart = b;
4049 PATFETCH (c);
4050 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
4051 break;
4052
4053 case 'S':
4054 laststart = b;
4055 PATFETCH (c);
4056 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
4057 break;
4058#endif /* emacs */
4059
4060
4061 case 'w':
4062 if (syntax & RE_NO_GNU_OPS)
4063 goto normal_char;
4064 laststart = b;
4065 BUF_PUSH (wordchar);
4066 break;
4067
4068
4069 case 'W':
4070 if (syntax & RE_NO_GNU_OPS)
4071 goto normal_char;
4072 laststart = b;
4073 BUF_PUSH (notwordchar);
4074 break;
4075
4076
4077 case '<':
4078 if (syntax & RE_NO_GNU_OPS)
4079 goto normal_char;
4080 BUF_PUSH (wordbeg);
4081 break;
4082
4083 case '>':
4084 if (syntax & RE_NO_GNU_OPS)
4085 goto normal_char;
4086 BUF_PUSH (wordend);
4087 break;
4088
4089 case 'b':
4090 if (syntax & RE_NO_GNU_OPS)
4091 goto normal_char;
4092 BUF_PUSH (wordbound);
4093 break;
4094
4095 case 'B':
4096 if (syntax & RE_NO_GNU_OPS)
4097 goto normal_char;
4098 BUF_PUSH (notwordbound);
4099 break;
4100
4101 case '`':
4102 if (syntax & RE_NO_GNU_OPS)
4103 goto normal_char;
4104 BUF_PUSH (begbuf);
4105 break;
4106
4107 case '\'':
4108 if (syntax & RE_NO_GNU_OPS)
4109 goto normal_char;
4110 BUF_PUSH (endbuf);
4111 break;
4112
4113 case '1': case '2': case '3': case '4': case '5':
4114 case '6': case '7': case '8': case '9':
4115 if (syntax & RE_NO_BK_REFS)
4116 goto normal_char;
4117
4118 c1 = c - '0';
4119
4120 if (c1 > regnum)
4121 FREE_STACK_RETURN (REG_ESUBREG);
4122
4123 /* Can't back reference to a subexpression if inside of it. */
4124 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4125 goto normal_char;
4126
4127 laststart = b;
4128 BUF_PUSH_2 (duplicate, c1);
4129 break;
4130
4131
4132 case '+':
4133 case '?':
4134 if (syntax & RE_BK_PLUS_QM)
4135 goto handle_plus;
4136 else
4137 goto normal_backslash;
4138
4139 default:
4140 normal_backslash:
4141 /* You might think it would be useful for \ to mean
4142 not to translate; but if we don't translate it
4143 it will never match anything. */
4144 c = TRANSLATE (c);
4145 goto normal_char;
4146 }
4147 break;
4148
4149
4150 default:
4151 /* Expects the character in `c'. */
4152 normal_char:
4153 /* If no exactn currently being built. */
4154 if (!pending_exact
4155#ifdef WCHAR
4156 /* If last exactn handle binary(or character) and
4157 new exactn handle character(or binary). */
4158 || is_exactn_bin != is_binary[p - 1 - pattern]
4159#endif /* WCHAR */
4160
4161 /* If last exactn not at current position. */
4162 || pending_exact + *pending_exact + 1 != b
4163
4164 /* We have only one byte following the exactn for the count. */
4165 || *pending_exact == (1 << BYTEWIDTH) - 1
4166
4167 /* If followed by a repetition operator. */
4168 || *p == '*' || *p == '^'
4169 || ((syntax & RE_BK_PLUS_QM)
4170 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
4171 : (*p == '+' || *p == '?'))
4172 || ((syntax & RE_INTERVALS)
4173 && ((syntax & RE_NO_BK_BRACES)
4174 ? *p == '{'
4175 : (p[0] == '\\' && p[1] == '{'))))
4176 {
4177 /* Start building a new exactn. */
4178
4179 laststart = b;
4180
4181#ifdef WCHAR
4182 /* Is this exactn binary data or character? */
4183 is_exactn_bin = is_binary[p - 1 - pattern];
4184 if (is_exactn_bin)
4185 BUF_PUSH_2 (exactn_bin, 0);
4186 else
4187 BUF_PUSH_2 (exactn, 0);
4188#else
4189 BUF_PUSH_2 (exactn, 0);
4190#endif /* WCHAR */
4191 pending_exact = b - 1;
4192 }
4193
4194 BUF_PUSH (c);
4195 (*pending_exact)++;
4196 break;
4197 } /* switch (c) */
4198 } /* while p != pend */
4199
4200
4201 /* Through the pattern now. */
4202
4203 if (fixup_alt_jump)
4204 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4205
4206 if (!COMPILE_STACK_EMPTY)
4207 FREE_STACK_RETURN (REG_EPAREN);
4208
4209 /* If we don't want backtracking, force success
4210 the first time we reach the end of the compiled pattern. */
4211 if (syntax & RE_NO_POSIX_BACKTRACKING)
4212 BUF_PUSH (succeed);
4213
4214#ifdef WCHAR
4215 free (pattern);
4216 free (mbs_offset);
4217 free (is_binary);
4218#endif
4219 free (compile_stack.stack);
4220
4221 /* We have succeeded; set the length of the buffer. */
4222#ifdef WCHAR
4223 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4224#else
4225 bufp->used = b - bufp->buffer;
4226#endif
4227
4228#ifdef DEBUG
4229 if (debug)
4230 {
4231 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4232 PREFIX(print_compiled_pattern) (bufp);
4233 }
4234#endif /* DEBUG */
4235
4236#ifndef MATCH_MAY_ALLOCATE
4237 /* Initialize the failure stack to the largest possible stack. This
4238 isn't necessary unless we're trying to avoid calling alloca in
4239 the search and match routines. */
4240 {
4241 int num_regs = bufp->re_nsub + 1;
4242
4243 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4244 is strictly greater than re_max_failures, the largest possible stack
4245 is 2 * re_max_failures failure points. */
4246 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4247 {
4248 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4249
4250# ifdef emacs
4251 if (! fail_stack.stack)
4252 fail_stack.stack
4253 = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size
4254 * sizeof (PREFIX(fail_stack_elt_t)));
4255 else
4256 fail_stack.stack
4257 = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack,
4258 (fail_stack.size
4259 * sizeof (PREFIX(fail_stack_elt_t))));
4260# else /* not emacs */
4261 if (! fail_stack.stack)
4262 fail_stack.stack
4263 = (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size
4264 * sizeof (PREFIX(fail_stack_elt_t)));
4265 else
4266 fail_stack.stack
4267 = (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack,
4268 (fail_stack.size
4269 * sizeof (PREFIX(fail_stack_elt_t))));
4270# endif /* not emacs */
4271 }
4272
4273 PREFIX(regex_grow_registers) (num_regs);
4274 }
4275#endif /* not MATCH_MAY_ALLOCATE */
4276
4277 return REG_NOERROR;
4278} /* regex_compile */
4279
4280/* Subroutines for `regex_compile'. */
4281
4282/* Store OP at LOC followed by two-byte integer parameter ARG. */
4283/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4284
4285static void
4286PREFIX(store_op1) (op, loc, arg)
4287 re_opcode_t op;
4288 UCHAR_T *loc;
4289 int arg;
4290{
4291 *loc = (UCHAR_T) op;
4292 STORE_NUMBER (loc + 1, arg);
4293}
4294
4295
4296/* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4297/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4298
4299static void
4300PREFIX(store_op2) (op, loc, arg1, arg2)
4301 re_opcode_t op;
4302 UCHAR_T *loc;
4303 int arg1, arg2;
4304{
4305 *loc = (UCHAR_T) op;
4306 STORE_NUMBER (loc + 1, arg1);
4307 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4308}
4309
4310
4311/* Copy the bytes from LOC to END to open up three bytes of space at LOC
4312 for OP followed by two-byte integer parameter ARG. */
4313/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4314
4315static void
4316PREFIX(insert_op1) (op, loc, arg, end)
4317 re_opcode_t op;
4318 UCHAR_T *loc;
4319 int arg;
4320 UCHAR_T *end;
4321{
4322 register UCHAR_T *pfrom = end;
4323 register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4324
4325 while (pfrom != loc)
4326 *--pto = *--pfrom;
4327
4328 PREFIX(store_op1) (op, loc, arg);
4329}
4330
4331
4332/* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4333/* ifdef WCHAR, integer parameter is 1 wchar_t. */
4334
4335static void
4336PREFIX(insert_op2) (op, loc, arg1, arg2, end)
4337 re_opcode_t op;
4338 UCHAR_T *loc;
4339 int arg1, arg2;
4340 UCHAR_T *end;
4341{
4342 register UCHAR_T *pfrom = end;
4343 register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4344
4345 while (pfrom != loc)
4346 *--pto = *--pfrom;
4347
4348 PREFIX(store_op2) (op, loc, arg1, arg2);
4349}
4350
4351
4352/* P points to just after a ^ in PATTERN. Return true if that ^ comes
4353 after an alternative or a begin-subexpression. We assume there is at
4354 least one character before the ^. */
4355
4356static boolean
4357PREFIX(at_begline_loc_p) (pattern, p, syntax)
4358 const CHAR_T *pattern, *p;
4359 reg_syntax_t syntax;
4360{
4361 const CHAR_T *prev = p - 2;
4362 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
4363
4364 return
4365 /* After a subexpression? */
4366 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
4367 /* After an alternative? */
4368 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
4369}
4370
4371
4372/* The dual of at_begline_loc_p. This one is for $. We assume there is
4373 at least one character after the $, i.e., `P < PEND'. */
4374
4375static boolean
4376PREFIX(at_endline_loc_p) (p, pend, syntax)
4377 const CHAR_T *p, *pend;
4378 reg_syntax_t syntax;
4379{
4380 const CHAR_T *next = p;
4381 boolean next_backslash = *next == '\\';
4382 const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0;
4383
4384 return
4385 /* Before a subexpression? */
4386 (syntax & RE_NO_BK_PARENS ? *next == ')'
4387 : next_backslash && next_next && *next_next == ')')
4388 /* Before an alternative? */
4389 || (syntax & RE_NO_BK_VBAR ? *next == '|'
4390 : next_backslash && next_next && *next_next == '|');
4391}
4392
4393#else /* not INSIDE_RECURSION */
4394
4395/* Returns true if REGNUM is in one of COMPILE_STACK's elements and
4396 false if it's not. */
4397
4398static boolean
4399group_in_compile_stack (compile_stack, regnum)
4400 compile_stack_type compile_stack;
4401 regnum_t regnum;
4402{
4403 int this_element;
4404
4405 for (this_element = compile_stack.avail - 1;
4406 this_element >= 0;
4407 this_element--)
4408 if (compile_stack.stack[this_element].regnum == regnum)
4409 return true;
4410
4411 return false;
4412}
4413#endif /* not INSIDE_RECURSION */
4414
4415#ifdef INSIDE_RECURSION
4416
4417#ifdef WCHAR
4418/* This insert space, which size is "num", into the pattern at "loc".
4419 "end" must point the end of the allocated buffer. */
4420static void
4421insert_space (num, loc, end)
4422 int num;
4423 CHAR_T *loc;
4424 CHAR_T *end;
4425{
4426 register CHAR_T *pto = end;
4427 register CHAR_T *pfrom = end - num;
4428
4429 while (pfrom >= loc)
4430 *pto-- = *pfrom--;
4431}
4432#endif /* WCHAR */
4433
4434#ifdef WCHAR
4435static reg_errcode_t
4436wcs_compile_range (range_start_char, p_ptr, pend, translate, syntax, b,
4437 char_set)
4438 CHAR_T range_start_char;
4439 const CHAR_T **p_ptr, *pend;
4440 CHAR_T *char_set, *b;
4441 RE_TRANSLATE_TYPE translate;
4442 reg_syntax_t syntax;
4443{
4444 const CHAR_T *p = *p_ptr;
4445 CHAR_T range_start, range_end;
4446 reg_errcode_t ret;
4447# ifdef _LIBC
4448 uint32_t nrules;
4449 uint32_t start_val, end_val;
4450# endif
4451 if (p == pend)
4452 return REG_ERANGE;
4453
4454# ifdef _LIBC
4455 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4456 if (nrules != 0)
4457 {
4458 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4459 _NL_COLLATE_COLLSEQWC);
4460 const unsigned char *extra = (const unsigned char *)
4461 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4462
4463 if (range_start_char < -1)
4464 {
4465 /* range_start is a collating symbol. */
4466 int32_t *wextra;
4467 /* Retreive the index and get collation sequence value. */
4468 wextra = (int32_t*)(extra + char_set[-range_start_char]);
4469 start_val = wextra[1 + *wextra];
4470 }
4471 else
4472 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4473
4474 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4475
4476 /* Report an error if the range is empty and the syntax prohibits
4477 this. */
4478 ret = ((syntax & RE_NO_EMPTY_RANGES)
4479 && (start_val > end_val))? REG_ERANGE : REG_NOERROR;
4480
4481 /* Insert space to the end of the char_ranges. */
4482 insert_space(2, b - char_set[5] - 2, b - 1);
4483 *(b - char_set[5] - 2) = (wchar_t)start_val;
4484 *(b - char_set[5] - 1) = (wchar_t)end_val;
4485 char_set[4]++; /* ranges_index */
4486 }
4487 else
4488# endif
4489 {
4490 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4491 range_start_char;
4492 range_end = TRANSLATE (p[0]);
4493 /* Report an error if the range is empty and the syntax prohibits
4494 this. */
4495 ret = ((syntax & RE_NO_EMPTY_RANGES)
4496 && (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4497
4498 /* Insert space to the end of the char_ranges. */
4499 insert_space(2, b - char_set[5] - 2, b - 1);
4500 *(b - char_set[5] - 2) = range_start;
4501 *(b - char_set[5] - 1) = range_end;
4502 char_set[4]++; /* ranges_index */
4503 }
4504 /* Have to increment the pointer into the pattern string, so the
4505 caller isn't still at the ending character. */
4506 (*p_ptr)++;
4507
4508 return ret;
4509}
4510#else /* BYTE */
4511/* Read the ending character of a range (in a bracket expression) from the
4512 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4513 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4514 Then we set the translation of all bits between the starting and
4515 ending characters (inclusive) in the compiled pattern B.
4516
4517 Return an error code.
4518
4519 We use these short variable names so we can use the same macros as
4520 `regex_compile' itself. */
4521
4522static reg_errcode_t
4523byte_compile_range (range_start_char, p_ptr, pend, translate, syntax, b)
4524 unsigned int range_start_char;
4525 const char **p_ptr, *pend;
4526 RE_TRANSLATE_TYPE translate;
4527 reg_syntax_t syntax;
4528 unsigned char *b;
4529{
4530 unsigned this_char;
4531 const char *p = *p_ptr;
4532 reg_errcode_t ret;
4533# if _LIBC
4534 const unsigned char *collseq;
4535 unsigned int start_colseq;
4536 unsigned int end_colseq;
4537# else
4538 unsigned end_char;
4539# endif
4540
4541 if (p == pend)
4542 return REG_ERANGE;
4543
4544 /* Have to increment the pointer into the pattern string, so the
4545 caller isn't still at the ending character. */
4546 (*p_ptr)++;
4547
4548 /* Report an error if the range is empty and the syntax prohibits this. */
4549 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4550
4551# if _LIBC
4552 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4553 _NL_COLLATE_COLLSEQMB);
4554
4555 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4556 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4557 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4558 {
4559 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4560
4561 if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4562 {
4563 SET_LIST_BIT (TRANSLATE (this_char));
4564 ret = REG_NOERROR;
4565 }
4566 }
4567# else
4568 /* Here we see why `this_char' has to be larger than an `unsigned
4569 char' -- we would otherwise go into an infinite loop, since all
4570 characters <= 0xff. */
4571 range_start_char = TRANSLATE (range_start_char);
4572 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4573 and some compilers cast it to int implicitly, so following for_loop
4574 may fall to (almost) infinite loop.
4575 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4576 To avoid this, we cast p[0] to unsigned int and truncate it. */
4577 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4578
4579 for (this_char = range_start_char; this_char <= end_char; ++this_char)
4580 {
4581 SET_LIST_BIT (TRANSLATE (this_char));
4582 ret = REG_NOERROR;
4583 }
4584# endif
4585
4586 return ret;
4587}
4588#endif /* WCHAR */
4589\f
4590/* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4591 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4592 characters can start a string that matches the pattern. This fastmap
4593 is used by re_search to skip quickly over impossible starting points.
4594
4595 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4596 area as BUFP->fastmap.
4597
4598 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4599 the pattern buffer.
4600
4601 Returns 0 if we succeed, -2 if an internal error. */
4602
4603#ifdef WCHAR
4604/* local function for re_compile_fastmap.
4605 truncate wchar_t character to char. */
4606static unsigned char truncate_wchar (CHAR_T c);
4607
4608static unsigned char
4609truncate_wchar (c)
4610 CHAR_T c;
4611{
4612 unsigned char buf[MB_LEN_MAX];
4613 int retval = wctomb(buf, c);
4614 return retval > 0 ? buf[0] : (unsigned char)c;
4615}
4616#endif /* WCHAR */
4617
4618static int
4619PREFIX(re_compile_fastmap) (bufp)
4620 struct re_pattern_buffer *bufp;
4621{
4622 int j, k;
4623#ifdef MATCH_MAY_ALLOCATE
4624 PREFIX(fail_stack_type) fail_stack;
4625#endif
4626#ifndef REGEX_MALLOC
4627 char *destination;
4628#endif
4629
4630 register char *fastmap = bufp->fastmap;
4631
4632#ifdef WCHAR
4633 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4634 pattern to (char*) in regex_compile. */
4635 UCHAR_T *pattern = (UCHAR_T*)bufp->buffer;
4636 register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used);
4637#else /* BYTE */
4638 UCHAR_T *pattern = bufp->buffer;
4639 register UCHAR_T *pend = pattern + bufp->used;
4640#endif /* WCHAR */
4641 UCHAR_T *p = pattern;
4642
4643#ifdef REL_ALLOC
4644 /* This holds the pointer to the failure stack, when
4645 it is allocated relocatably. */
4646 fail_stack_elt_t *failure_stack_ptr;
4647#endif
4648
4649 /* Assume that each path through the pattern can be null until
4650 proven otherwise. We set this false at the bottom of switch
4651 statement, to which we get only if a particular path doesn't
4652 match the empty string. */
4653 boolean path_can_be_null = true;
4654
4655 /* We aren't doing a `succeed_n' to begin with. */
4656 boolean succeed_n_p = false;
4657
4658 assert (fastmap != NULL && p != NULL);
4659
4660 INIT_FAIL_STACK ();
4661 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4662 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4663 bufp->can_be_null = 0;
4664
4665 while (1)
4666 {
4667 if (p == pend || *p == succeed)
4668 {
4669 /* We have reached the (effective) end of pattern. */
4670 if (!FAIL_STACK_EMPTY ())
4671 {
4672 bufp->can_be_null |= path_can_be_null;
4673
4674 /* Reset for next path. */
4675 path_can_be_null = true;
4676
4677 p = fail_stack.stack[--fail_stack.avail].pointer;
4678
4679 continue;
4680 }
4681 else
4682 break;
4683 }
4684
4685 /* We should never be about to go beyond the end of the pattern. */
4686 assert (p < pend);
4687
4688 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4689 {
4690
4691 /* I guess the idea here is to simply not bother with a fastmap
4692 if a backreference is used, since it's too hard to figure out
4693 the fastmap for the corresponding group. Setting
4694 `can_be_null' stops `re_search_2' from using the fastmap, so
4695 that is all we do. */
4696 case duplicate:
4697 bufp->can_be_null = 1;
4698 goto done;
4699
4700
4701 /* Following are the cases which match a character. These end
4702 with `break'. */
4703
4704#ifdef WCHAR
4705 case exactn:
4706 fastmap[truncate_wchar(p[1])] = 1;
4707 break;
4708#else /* BYTE */
4709 case exactn:
4710 fastmap[p[1]] = 1;
4711 break;
4712#endif /* WCHAR */
4713#ifdef MBS_SUPPORT
4714 case exactn_bin:
4715 fastmap[p[1]] = 1;
4716 break;
4717#endif
4718
4719#ifdef WCHAR
4720 /* It is hard to distinguish fastmap from (multi byte) characters
4721 which depends on current locale. */
4722 case charset:
4723 case charset_not:
4724 case wordchar:
4725 case notwordchar:
4726 bufp->can_be_null = 1;
4727 goto done;
4728#else /* BYTE */
4729 case charset:
4730 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4731 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4732 fastmap[j] = 1;
4733 break;
4734
4735
4736 case charset_not:
4737 /* Chars beyond end of map must be allowed. */
4738 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4739 fastmap[j] = 1;
4740
4741 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4742 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4743 fastmap[j] = 1;
4744 break;
4745
4746
4747 case wordchar:
4748 for (j = 0; j < (1 << BYTEWIDTH); j++)
4749 if (SYNTAX (j) == Sword)
4750 fastmap[j] = 1;
4751 break;
4752
4753
4754 case notwordchar:
4755 for (j = 0; j < (1 << BYTEWIDTH); j++)
4756 if (SYNTAX (j) != Sword)
4757 fastmap[j] = 1;
4758 break;
4759#endif /* WCHAR */
4760
4761 case anychar:
4762 {
4763 int fastmap_newline = fastmap['\n'];
4764
4765 /* `.' matches anything ... */
4766 for (j = 0; j < (1 << BYTEWIDTH); j++)
4767 fastmap[j] = 1;
4768
4769 /* ... except perhaps newline. */
4770 if (!(bufp->syntax & RE_DOT_NEWLINE))
4771 fastmap['\n'] = fastmap_newline;
4772
4773 /* Return if we have already set `can_be_null'; if we have,
4774 then the fastmap is irrelevant. Something's wrong here. */
4775 else if (bufp->can_be_null)
4776 goto done;
4777
4778 /* Otherwise, have to check alternative paths. */
4779 break;
4780 }
4781
4782#ifdef emacs
4783 case syntaxspec:
4784 k = *p++;
4785 for (j = 0; j < (1 << BYTEWIDTH); j++)
4786 if (SYNTAX (j) == (enum syntaxcode) k)
4787 fastmap[j] = 1;
4788 break;
4789
4790
4791 case notsyntaxspec:
4792 k = *p++;
4793 for (j = 0; j < (1 << BYTEWIDTH); j++)
4794 if (SYNTAX (j) != (enum syntaxcode) k)
4795 fastmap[j] = 1;
4796 break;
4797
4798
4799 /* All cases after this match the empty string. These end with
4800 `continue'. */
4801
4802
4803 case before_dot:
4804 case at_dot:
4805 case after_dot:
4806 continue;
4807#endif /* emacs */
4808
4809
4810 case no_op:
4811 case begline:
4812 case endline:
4813 case begbuf:
4814 case endbuf:
4815 case wordbound:
4816 case notwordbound:
4817 case wordbeg:
4818 case wordend:
4819 case push_dummy_failure:
4820 continue;
4821
4822
4823 case jump_n:
4824 case pop_failure_jump:
4825 case maybe_pop_jump:
4826 case jump:
4827 case jump_past_alt:
4828 case dummy_failure_jump:
4829 EXTRACT_NUMBER_AND_INCR (j, p);
4830 p += j;
4831 if (j > 0)
4832 continue;
4833
4834 /* Jump backward implies we just went through the body of a
4835 loop and matched nothing. Opcode jumped to should be
4836 `on_failure_jump' or `succeed_n'. Just treat it like an
4837 ordinary jump. For a * loop, it has pushed its failure
4838 point already; if so, discard that as redundant. */
4839 if ((re_opcode_t) *p != on_failure_jump
4840 && (re_opcode_t) *p != succeed_n)
4841 continue;
4842
4843 p++;
4844 EXTRACT_NUMBER_AND_INCR (j, p);
4845 p += j;
4846
4847 /* If what's on the stack is where we are now, pop it. */
4848 if (!FAIL_STACK_EMPTY ()
4849 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
4850 fail_stack.avail--;
4851
4852 continue;
4853
4854
4855 case on_failure_jump:
4856 case on_failure_keep_string_jump:
4857 handle_on_failure_jump:
4858 EXTRACT_NUMBER_AND_INCR (j, p);
4859
4860 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4861 end of the pattern. We don't want to push such a point,
4862 since when we restore it above, entering the switch will
4863 increment `p' past the end of the pattern. We don't need
4864 to push such a point since we obviously won't find any more
4865 fastmap entries beyond `pend'. Such a pattern can match
4866 the null string, though. */
4867 if (p + j < pend)
4868 {
4869 if (!PUSH_PATTERN_OP (p + j, fail_stack))
4870 {
4871 RESET_FAIL_STACK ();
4872 return -2;
4873 }
4874 }
4875 else
4876 bufp->can_be_null = 1;
4877
4878 if (succeed_n_p)
4879 {
4880 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4881 succeed_n_p = false;
4882 }
4883
4884 continue;
4885
4886
4887 case succeed_n:
4888 /* Get to the number of times to succeed. */
4889 p += OFFSET_ADDRESS_SIZE;
4890
4891 /* Increment p past the n for when k != 0. */
4892 EXTRACT_NUMBER_AND_INCR (k, p);
4893 if (k == 0)
4894 {
4895 p -= 2 * OFFSET_ADDRESS_SIZE;
4896 succeed_n_p = true; /* Spaghetti code alert. */
4897 goto handle_on_failure_jump;
4898 }
4899 continue;
4900
4901
4902 case set_number_at:
4903 p += 2 * OFFSET_ADDRESS_SIZE;
4904 continue;
4905
4906
4907 case start_memory:
4908 case stop_memory:
4909 p += 2;
4910 continue;
4911
4912
4913 default:
4914 abort (); /* We have listed all the cases. */
4915 } /* switch *p++ */
4916
4917 /* Getting here means we have found the possible starting
4918 characters for one path of the pattern -- and that the empty
4919 string does not match. We need not follow this path further.
4920 Instead, look at the next alternative (remembered on the
4921 stack), or quit if no more. The test at the top of the loop
4922 does these things. */
4923 path_can_be_null = false;
4924 p = pend;
4925 } /* while p */
4926
4927 /* Set `can_be_null' for the last path (also the first path, if the
4928 pattern is empty). */
4929 bufp->can_be_null |= path_can_be_null;
4930
4931 done:
4932 RESET_FAIL_STACK ();
4933 return 0;
4934}
4935
4936#else /* not INSIDE_RECURSION */
4937
4938int
4939re_compile_fastmap (bufp)
4940 struct re_pattern_buffer *bufp;
4941{
4942# ifdef MBS_SUPPORT
4943 if (MB_CUR_MAX != 1)
4944 return wcs_re_compile_fastmap(bufp);
4945 else
4946# endif
4947 return byte_re_compile_fastmap(bufp);
4948} /* re_compile_fastmap */
4949#ifdef _LIBC
4950weak_alias (__re_compile_fastmap, re_compile_fastmap)
4951#endif
4952\f
4953
4954/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4955 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4956 this memory for recording register information. STARTS and ENDS
4957 must be allocated using the malloc library routine, and must each
4958 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4959
4960 If NUM_REGS == 0, then subsequent matches should allocate their own
4961 register data.
4962
4963 Unless this function is called, the first search or match using
4964 PATTERN_BUFFER will allocate its own register data, without
4965 freeing the old data. */
4966
4967void
4968re_set_registers (bufp, regs, num_regs, starts, ends)
4969 struct re_pattern_buffer *bufp;
4970 struct re_registers *regs;
4971 unsigned num_regs;
4972 regoff_t *starts, *ends;
4973{
4974 if (num_regs)
4975 {
4976 bufp->regs_allocated = REGS_REALLOCATE;
4977 regs->num_regs = num_regs;
4978 regs->start = starts;
4979 regs->end = ends;
4980 }
4981 else
4982 {
4983 bufp->regs_allocated = REGS_UNALLOCATED;
4984 regs->num_regs = 0;
4985 regs->start = regs->end = (regoff_t *) 0;
4986 }
4987}
4988#ifdef _LIBC
4989weak_alias (__re_set_registers, re_set_registers)
4990#endif
4991\f
4992/* Searching routines. */
4993
4994/* Like re_search_2, below, but only one string is specified, and
4995 doesn't let you say where to stop matching. */
4996
4997int
4998re_search (bufp, string, size, startpos, range, regs)
4999 struct re_pattern_buffer *bufp;
5000 const char *string;
5001 int size, startpos, range;
5002 struct re_registers *regs;
5003{
5004 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
5005 regs, size);
5006}
5007#ifdef _LIBC
5008weak_alias (__re_search, re_search)
5009#endif
5010
5011
5012/* Using the compiled pattern in BUFP->buffer, first tries to match the
5013 virtual concatenation of STRING1 and STRING2, starting first at index
5014 STARTPOS, then at STARTPOS + 1, and so on.
5015
5016 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
5017
5018 RANGE is how far to scan while trying to match. RANGE = 0 means try
5019 only at STARTPOS; in general, the last start tried is STARTPOS +
5020 RANGE.
5021
5022 In REGS, return the indices of the virtual concatenation of STRING1
5023 and STRING2 that matched the entire BUFP->buffer and its contained
5024 subexpressions.
5025
5026 Do not consider matching one past the index STOP in the virtual
5027 concatenation of STRING1 and STRING2.
5028
5029 We return either the position in the strings at which the match was
5030 found, -1 if no match, or -2 if error (such as failure
5031 stack overflow). */
5032
5033int
5034re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
5035 struct re_pattern_buffer *bufp;
5036 const char *string1, *string2;
5037 int size1, size2;
5038 int startpos;
5039 int range;
5040 struct re_registers *regs;
5041 int stop;
5042{
5043# ifdef MBS_SUPPORT
5044 if (MB_CUR_MAX != 1)
5045 return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5046 range, regs, stop);
5047 else
5048# endif
5049 return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos,
5050 range, regs, stop);
5051} /* re_search_2 */
5052#ifdef _LIBC
5053weak_alias (__re_search_2, re_search_2)
5054#endif
5055
5056#endif /* not INSIDE_RECURSION */
5057
5058#ifdef INSIDE_RECURSION
5059
5060#ifdef MATCH_MAY_ALLOCATE
5061# define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5062#else
5063# define FREE_VAR(var) if (var) free (var); var = NULL
5064#endif
5065
5066#ifdef WCHAR
5067# define FREE_WCS_BUFFERS() \
5068 do { \
5069 FREE_VAR (string1); \
5070 FREE_VAR (string2); \
5071 FREE_VAR (mbs_offset1); \
5072 FREE_VAR (mbs_offset2); \
5073 } while (0)
5074
5075#endif
5076
5077static int
5078PREFIX(re_search_2) (bufp, string1, size1, string2, size2, startpos, range,
5079 regs, stop)
5080 struct re_pattern_buffer *bufp;
5081 const char *string1, *string2;
5082 int size1, size2;
5083 int startpos;
5084 int range;
5085 struct re_registers *regs;
5086 int stop;
5087{
5088 int val;
5089 register char *fastmap = bufp->fastmap;
5090 register RE_TRANSLATE_TYPE translate = bufp->translate;
5091 int total_size = size1 + size2;
5092 int endpos = startpos + range;
5093#ifdef WCHAR
5094 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5095 wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL;
5096 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5097 int wcs_size1 = 0, wcs_size2 = 0;
5098 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5099 int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5100 /* They hold whether each wchar_t is binary data or not. */
5101 char *is_binary = NULL;
5102#endif /* WCHAR */
5103
5104 /* Check for out-of-range STARTPOS. */
5105 if (startpos < 0 || startpos > total_size)
5106 return -1;
5107
5108 /* Fix up RANGE if it might eventually take us outside
5109 the virtual concatenation of STRING1 and STRING2.
5110 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
5111 if (endpos < 0)
5112 range = 0 - startpos;
5113 else if (endpos > total_size)
5114 range = total_size - startpos;
5115
5116 /* If the search isn't to be a backwards one, don't waste time in a
5117 search for a pattern that must be anchored. */
5118 if (bufp->used > 0 && range > 0
5119 && ((re_opcode_t) bufp->buffer[0] == begbuf
5120 /* `begline' is like `begbuf' if it cannot match at newlines. */
5121 || ((re_opcode_t) bufp->buffer[0] == begline
5122 && !bufp->newline_anchor)))
5123 {
5124 if (startpos > 0)
5125 return -1;
5126 else
5127 range = 1;
5128 }
5129
5130#ifdef emacs
5131 /* In a forward search for something that starts with \=.
5132 don't keep searching past point. */
5133 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
5134 {
5135 range = PT - startpos;
5136 if (range <= 0)
5137 return -1;
5138 }
5139#endif /* emacs */
5140
5141 /* Update the fastmap now if not correct already. */
5142 if (fastmap && !bufp->fastmap_accurate)
5143 if (re_compile_fastmap (bufp) == -2)
5144 return -2;
5145
5146#ifdef WCHAR
5147 /* Allocate wchar_t array for wcs_string1 and wcs_string2 and
5148 fill them with converted string. */
5149 if (size1 != 0)
5150 {
5151 wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T);
5152 mbs_offset1 = REGEX_TALLOC (size1 + 1, int);
5153 is_binary = REGEX_TALLOC (size1 + 1, char);
5154 if (!wcs_string1 || !mbs_offset1 || !is_binary)
5155 {
5156 FREE_VAR (wcs_string1);
5157 FREE_VAR (mbs_offset1);
5158 FREE_VAR (is_binary);
5159 return -2;
5160 }
5161 wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1,
5162 mbs_offset1, is_binary);
5163 wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */
5164 FREE_VAR (is_binary);
5165 }
5166 if (size2 != 0)
5167 {
5168 wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T);
5169 mbs_offset2 = REGEX_TALLOC (size2 + 1, int);
5170 is_binary = REGEX_TALLOC (size2 + 1, char);
5171 if (!wcs_string2 || !mbs_offset2 || !is_binary)
5172 {
5173 FREE_WCS_BUFFERS ();
5174 FREE_VAR (is_binary);
5175 return -2;
5176 }
5177 wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2,
5178 mbs_offset2, is_binary);
5179 wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */
5180 FREE_VAR (is_binary);
5181 }
5182#endif /* WCHAR */
5183
5184
5185 /* Loop through the string, looking for a place to start matching. */
5186 for (;;)
5187 {
5188 /* If a fastmap is supplied, skip quickly over characters that
5189 cannot be the start of a match. If the pattern can match the
5190 null string, however, we don't need to skip characters; we want
5191 the first null string. */
5192 if (fastmap && startpos < total_size && !bufp->can_be_null)
5193 {
5194 if (range > 0) /* Searching forwards. */
5195 {
5196 register const char *d;
5197 register int lim = 0;
5198 int irange = range;
5199
5200 if (startpos < size1 && startpos + range >= size1)
5201 lim = range - (size1 - startpos);
5202
5203 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5204
5205 /* Written out as an if-else to avoid testing `translate'
5206 inside the loop. */
5207 if (translate)
5208 while (range > lim
5209 && !fastmap[(unsigned char)
5210 translate[(unsigned char) *d++]])
5211 range--;
5212 else
5213 while (range > lim && !fastmap[(unsigned char) *d++])
5214 range--;
5215
5216 startpos += irange - range;
5217 }
5218 else /* Searching backwards. */
5219 {
5220 register CHAR_T c = (size1 == 0 || startpos >= size1
5221 ? string2[startpos - size1]
5222 : string1[startpos]);
5223
5224 if (!fastmap[(unsigned char) TRANSLATE (c)])
5225 goto advance;
5226 }
5227 }
5228
5229 /* If can't match the null string, and that's all we have left, fail. */
5230 if (range >= 0 && startpos == total_size && fastmap
5231 && !bufp->can_be_null)
5232 {
5233#ifdef WCHAR
5234 FREE_WCS_BUFFERS ();
5235#endif
5236 return -1;
5237 }
5238
5239#ifdef WCHAR
5240 val = wcs_re_match_2_internal (bufp, string1, size1, string2,
5241 size2, startpos, regs, stop,
5242 wcs_string1, wcs_size1,
5243 wcs_string2, wcs_size2,
5244 mbs_offset1, mbs_offset2);
5245#else /* BYTE */
5246 val = byte_re_match_2_internal (bufp, string1, size1, string2,
5247 size2, startpos, regs, stop);
5248#endif /* BYTE */
5249
5250#ifndef REGEX_MALLOC
5251# ifdef C_ALLOCA
5252 alloca (0);
5253# endif
5254#endif
5255
5256 if (val >= 0)
5257 {
5258#ifdef WCHAR
5259 FREE_WCS_BUFFERS ();
5260#endif
5261 return startpos;
5262 }
5263
5264 if (val == -2)
5265 {
5266#ifdef WCHAR
5267 FREE_WCS_BUFFERS ();
5268#endif
5269 return -2;
5270 }
5271
5272 advance:
5273 if (!range)
5274 break;
5275 else if (range > 0)
5276 {
5277 range--;
5278 startpos++;
5279 }
5280 else
5281 {
5282 range++;
5283 startpos--;
5284 }
5285 }
5286#ifdef WCHAR
5287 FREE_WCS_BUFFERS ();
5288#endif
5289 return -1;
5290}
5291
5292#ifdef WCHAR
5293/* This converts PTR, a pointer into one of the search wchar_t strings
5294 `string1' and `string2' into an multibyte string offset from the
5295 beginning of that string. We use mbs_offset to optimize.
5296 See convert_mbs_to_wcs. */
5297# define POINTER_TO_OFFSET(ptr) \
5298 (FIRST_STRING_P (ptr) \
5299 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5300 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5301 + csize1)))
5302#else /* BYTE */
5303/* This converts PTR, a pointer into one of the search strings `string1'
5304 and `string2' into an offset from the beginning of that string. */
5305# define POINTER_TO_OFFSET(ptr) \
5306 (FIRST_STRING_P (ptr) \
5307 ? ((regoff_t) ((ptr) - string1)) \
5308 : ((regoff_t) ((ptr) - string2 + size1)))
5309#endif /* WCHAR */
5310
5311/* Macros for dealing with the split strings in re_match_2. */
5312
5313#define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5314
5315/* Call before fetching a character with *d. This switches over to
5316 string2 if necessary. */
5317#define PREFETCH() \
5318 while (d == dend) \
5319 { \
5320 /* End of string2 => fail. */ \
5321 if (dend == end_match_2) \
5322 goto fail; \
5323 /* End of string1 => advance to string2. */ \
5324 d = string2; \
5325 dend = end_match_2; \
5326 }
5327
5328/* Test if at very beginning or at very end of the virtual concatenation
5329 of `string1' and `string2'. If only one string, it's `string2'. */
5330#define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5331#define AT_STRINGS_END(d) ((d) == end2)
5332
5333
5334/* Test if D points to a character which is word-constituent. We have
5335 two special cases to check for: if past the end of string1, look at
5336 the first character in string2; and if before the beginning of
5337 string2, look at the last character in string1. */
5338#ifdef WCHAR
5339/* Use internationalized API instead of SYNTAX. */
5340# define WORDCHAR_P(d) \
5341 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5342 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0)
5343#else /* BYTE */
5344# define WORDCHAR_P(d) \
5345 (SYNTAX ((d) == end1 ? *string2 \
5346 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5347 == Sword)
5348#endif /* WCHAR */
5349
5350/* Disabled due to a compiler bug -- see comment at case wordbound */
5351#if 0
5352/* Test if the character before D and the one at D differ with respect
5353 to being word-constituent. */
5354#define AT_WORD_BOUNDARY(d) \
5355 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5356 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5357#endif
5358
5359/* Free everything we malloc. */
5360#ifdef MATCH_MAY_ALLOCATE
5361# ifdef WCHAR
5362# define FREE_VARIABLES() \
5363 do { \
5364 REGEX_FREE_STACK (fail_stack.stack); \
5365 FREE_VAR (regstart); \
5366 FREE_VAR (regend); \
5367 FREE_VAR (old_regstart); \
5368 FREE_VAR (old_regend); \
5369 FREE_VAR (best_regstart); \
5370 FREE_VAR (best_regend); \
5371 FREE_VAR (reg_info); \
5372 FREE_VAR (reg_dummy); \
5373 FREE_VAR (reg_info_dummy); \
5374 if (!cant_free_wcs_buf) \
5375 { \
5376 FREE_VAR (string1); \
5377 FREE_VAR (string2); \
5378 FREE_VAR (mbs_offset1); \
5379 FREE_VAR (mbs_offset2); \
5380 } \
5381 } while (0)
5382# else /* BYTE */
5383# define FREE_VARIABLES() \
5384 do { \
5385 REGEX_FREE_STACK (fail_stack.stack); \
5386 FREE_VAR (regstart); \
5387 FREE_VAR (regend); \
5388 FREE_VAR (old_regstart); \
5389 FREE_VAR (old_regend); \
5390 FREE_VAR (best_regstart); \
5391 FREE_VAR (best_regend); \
5392 FREE_VAR (reg_info); \
5393 FREE_VAR (reg_dummy); \
5394 FREE_VAR (reg_info_dummy); \
5395 } while (0)
5396# endif /* WCHAR */
5397#else
5398# ifdef WCHAR
5399# define FREE_VARIABLES() \
5400 do { \
5401 if (!cant_free_wcs_buf) \
5402 { \
5403 FREE_VAR (string1); \
5404 FREE_VAR (string2); \
5405 FREE_VAR (mbs_offset1); \
5406 FREE_VAR (mbs_offset2); \
5407 } \
5408 } while (0)
5409# else /* BYTE */
5410# define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5411# endif /* WCHAR */
5412#endif /* not MATCH_MAY_ALLOCATE */
5413
5414/* These values must meet several constraints. They must not be valid
5415 register values; since we have a limit of 255 registers (because
5416 we use only one byte in the pattern for the register number), we can
5417 use numbers larger than 255. They must differ by 1, because of
5418 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5419 be larger than the value for the highest register, so we do not try
5420 to actually save any registers when none are active. */
5421#define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5422#define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5423\f
5424#else /* not INSIDE_RECURSION */
5425/* Matching routines. */
5426
5427#ifndef emacs /* Emacs never uses this. */
5428/* re_match is like re_match_2 except it takes only a single string. */
5429
5430int
5431re_match (bufp, string, size, pos, regs)
5432 struct re_pattern_buffer *bufp;
5433 const char *string;
5434 int size, pos;
5435 struct re_registers *regs;
5436{
5437 int result;
5438# ifdef MBS_SUPPORT
5439 if (MB_CUR_MAX != 1)
5440 result = wcs_re_match_2_internal (bufp, NULL, 0, string, size,
5441 pos, regs, size,
5442 NULL, 0, NULL, 0, NULL, NULL);
5443 else
5444# endif
5445 result = byte_re_match_2_internal (bufp, NULL, 0, string, size,
5446 pos, regs, size);
5447# ifndef REGEX_MALLOC
5448# ifdef C_ALLOCA
5449 alloca (0);
5450# endif
5451# endif
5452 return result;
5453}
5454# ifdef _LIBC
5455weak_alias (__re_match, re_match)
5456# endif
5457#endif /* not emacs */
5458
5459#endif /* not INSIDE_RECURSION */
5460
5461#ifdef INSIDE_RECURSION
5462static boolean PREFIX(group_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5463 UCHAR_T *end,
5464 PREFIX(register_info_type) *reg_info));
5465static boolean PREFIX(alt_match_null_string_p) _RE_ARGS ((UCHAR_T *p,
5466 UCHAR_T *end,
5467 PREFIX(register_info_type) *reg_info));
5468static boolean PREFIX(common_op_match_null_string_p) _RE_ARGS ((UCHAR_T **p,
5469 UCHAR_T *end,
5470 PREFIX(register_info_type) *reg_info));
5471static int PREFIX(bcmp_translate) _RE_ARGS ((const CHAR_T *s1, const CHAR_T *s2,
5472 int len, char *translate));
5473#else /* not INSIDE_RECURSION */
5474
5475/* re_match_2 matches the compiled pattern in BUFP against the
5476 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5477 and SIZE2, respectively). We start matching at POS, and stop
5478 matching at STOP.
5479
5480 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5481 store offsets for the substring each group matched in REGS. See the
5482 documentation for exactly how many groups we fill.
5483
5484 We return -1 if no match, -2 if an internal error (such as the
5485 failure stack overflowing). Otherwise, we return the length of the
5486 matched substring. */
5487
5488int
5489re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5490 struct re_pattern_buffer *bufp;
5491 const char *string1, *string2;
5492 int size1, size2;
5493 int pos;
5494 struct re_registers *regs;
5495 int stop;
5496{
5497 int result;
5498# ifdef MBS_SUPPORT
5499 if (MB_CUR_MAX != 1)
5500 result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2,
5501 pos, regs, stop,
5502 NULL, 0, NULL, 0, NULL, NULL);
5503 else
5504# endif
5505 result = byte_re_match_2_internal (bufp, string1, size1, string2, size2,
5506 pos, regs, stop);
5507
5508#ifndef REGEX_MALLOC
5509# ifdef C_ALLOCA
5510 alloca (0);
5511# endif
5512#endif
5513 return result;
5514}
5515#ifdef _LIBC
5516weak_alias (__re_match_2, re_match_2)
5517#endif
5518
5519#endif /* not INSIDE_RECURSION */
5520
5521#ifdef INSIDE_RECURSION
5522
5523#ifdef WCHAR
5524static int count_mbs_length PARAMS ((int *, int));
5525
5526/* This check the substring (from 0, to length) of the multibyte string,
5527 to which offset_buffer correspond. And count how many wchar_t_characters
5528 the substring occupy. We use offset_buffer to optimization.
5529 See convert_mbs_to_wcs. */
5530
5531static int
5532count_mbs_length(offset_buffer, length)
5533 int *offset_buffer;
5534 int length;
5535{
5536 int upper, lower;
5537
5538 /* Check whether the size is valid. */
5539 if (length < 0)
5540 return -1;
5541
5542 if (offset_buffer == NULL)
5543 return 0;
5544
5545 /* If there are no multibyte character, offset_buffer[i] == i.
5546 Optmize for this case. */
5547 if (offset_buffer[length] == length)
5548 return length;
5549
5550 /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */
5551 upper = length;
5552 lower = 0;
5553
5554 while (true)
5555 {
5556 int middle = (lower + upper) / 2;
5557 if (middle == lower || middle == upper)
5558 break;
5559 if (offset_buffer[middle] > length)
5560 upper = middle;
5561 else if (offset_buffer[middle] < length)
5562 lower = middle;
5563 else
5564 return middle;
5565 }
5566
5567 return -1;
5568}
5569#endif /* WCHAR */
5570
5571/* This is a separate function so that we can force an alloca cleanup
5572 afterwards. */
5573#ifdef WCHAR
5574static int
5575wcs_re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos,
5576 regs, stop, string1, size1, string2, size2,
5577 mbs_offset1, mbs_offset2)
5578 struct re_pattern_buffer *bufp;
5579 const char *cstring1, *cstring2;
5580 int csize1, csize2;
5581 int pos;
5582 struct re_registers *regs;
5583 int stop;
5584 /* string1 == string2 == NULL means string1/2, size1/2 and
5585 mbs_offset1/2 need seting up in this function. */
5586 /* We need wchar_t* buffers correspond to cstring1, cstring2. */
5587 wchar_t *string1, *string2;
5588 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5589 int size1, size2;
5590 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5591 int *mbs_offset1, *mbs_offset2;
5592#else /* BYTE */
5593static int
5594byte_re_match_2_internal (bufp, string1, size1,string2, size2, pos,
5595 regs, stop)
5596 struct re_pattern_buffer *bufp;
5597 const char *string1, *string2;
5598 int size1, size2;
5599 int pos;
5600 struct re_registers *regs;
5601 int stop;
5602#endif /* BYTE */
5603{
5604 /* General temporaries. */
5605 int mcnt;
5606 UCHAR_T *p1;
5607#ifdef WCHAR
5608 /* They hold whether each wchar_t is binary data or not. */
5609 char *is_binary = NULL;
5610 /* If true, we can't free string1/2, mbs_offset1/2. */
5611 int cant_free_wcs_buf = 1;
5612#endif /* WCHAR */
5613
5614 /* Just past the end of the corresponding string. */
5615 const CHAR_T *end1, *end2;
5616
5617 /* Pointers into string1 and string2, just past the last characters in
5618 each to consider matching. */
5619 const CHAR_T *end_match_1, *end_match_2;
5620
5621 /* Where we are in the data, and the end of the current string. */
5622 const CHAR_T *d, *dend;
5623
5624 /* Where we are in the pattern, and the end of the pattern. */
5625#ifdef WCHAR
5626 UCHAR_T *pattern, *p;
5627 register UCHAR_T *pend;
5628#else /* BYTE */
5629 UCHAR_T *p = bufp->buffer;
5630 register UCHAR_T *pend = p + bufp->used;
5631#endif /* WCHAR */
5632
5633 /* Mark the opcode just after a start_memory, so we can test for an
5634 empty subpattern when we get to the stop_memory. */
5635 UCHAR_T *just_past_start_mem = 0;
5636
5637 /* We use this to map every character in the string. */
5638 RE_TRANSLATE_TYPE translate = bufp->translate;
5639
5640 /* Failure point stack. Each place that can handle a failure further
5641 down the line pushes a failure point on this stack. It consists of
5642 restart, regend, and reg_info for all registers corresponding to
5643 the subexpressions we're currently inside, plus the number of such
5644 registers, and, finally, two char *'s. The first char * is where
5645 to resume scanning the pattern; the second one is where to resume
5646 scanning the strings. If the latter is zero, the failure point is
5647 a ``dummy''; if a failure happens and the failure point is a dummy,
5648 it gets discarded and the next next one is tried. */
5649#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5650 PREFIX(fail_stack_type) fail_stack;
5651#endif
5652#ifdef DEBUG
5653 static unsigned failure_id;
5654 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5655#endif
5656
5657#ifdef REL_ALLOC
5658 /* This holds the pointer to the failure stack, when
5659 it is allocated relocatably. */
5660 fail_stack_elt_t *failure_stack_ptr;
5661#endif
5662
5663 /* We fill all the registers internally, independent of what we
5664 return, for use in backreferences. The number here includes
5665 an element for register zero. */
5666 size_t num_regs = bufp->re_nsub + 1;
5667
5668 /* The currently active registers. */
5669 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5670 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5671
5672 /* Information on the contents of registers. These are pointers into
5673 the input strings; they record just what was matched (on this
5674 attempt) by a subexpression part of the pattern, that is, the
5675 regnum-th regstart pointer points to where in the pattern we began
5676 matching and the regnum-th regend points to right after where we
5677 stopped matching the regnum-th subexpression. (The zeroth register
5678 keeps track of what the whole pattern matches.) */
5679#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5680 const CHAR_T **regstart, **regend;
5681#endif
5682
5683 /* If a group that's operated upon by a repetition operator fails to
5684 match anything, then the register for its start will need to be
5685 restored because it will have been set to wherever in the string we
5686 are when we last see its open-group operator. Similarly for a
5687 register's end. */
5688#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5689 const CHAR_T **old_regstart, **old_regend;
5690#endif
5691
5692 /* The is_active field of reg_info helps us keep track of which (possibly
5693 nested) subexpressions we are currently in. The matched_something
5694 field of reg_info[reg_num] helps us tell whether or not we have
5695 matched any of the pattern so far this time through the reg_num-th
5696 subexpression. These two fields get reset each time through any
5697 loop their register is in. */
5698#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5699 PREFIX(register_info_type) *reg_info;
5700#endif
5701
5702 /* The following record the register info as found in the above
5703 variables when we find a match better than any we've seen before.
5704 This happens as we backtrack through the failure points, which in
5705 turn happens only if we have not yet matched the entire string. */
5706 unsigned best_regs_set = false;
5707#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5708 const CHAR_T **best_regstart, **best_regend;
5709#endif
5710
5711 /* Logically, this is `best_regend[0]'. But we don't want to have to
5712 allocate space for that if we're not allocating space for anything
5713 else (see below). Also, we never need info about register 0 for
5714 any of the other register vectors, and it seems rather a kludge to
5715 treat `best_regend' differently than the rest. So we keep track of
5716 the end of the best match so far in a separate variable. We
5717 initialize this to NULL so that when we backtrack the first time
5718 and need to test it, it's not garbage. */
5719 const CHAR_T *match_end = NULL;
5720
5721 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5722 int set_regs_matched_done = 0;
5723
5724 /* Used when we pop values we don't care about. */
5725#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5726 const CHAR_T **reg_dummy;
5727 PREFIX(register_info_type) *reg_info_dummy;
5728#endif
5729
5730#ifdef DEBUG
5731 /* Counts the total number of registers pushed. */
5732 unsigned num_regs_pushed = 0;
5733#endif
5734
5735 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5736
5737 INIT_FAIL_STACK ();
5738
5739#ifdef MATCH_MAY_ALLOCATE
5740 /* Do not bother to initialize all the register variables if there are
5741 no groups in the pattern, as it takes a fair amount of time. If
5742 there are groups, we include space for register 0 (the whole
5743 pattern), even though we never use it, since it simplifies the
5744 array indexing. We should fix this. */
5745 if (bufp->re_nsub)
5746 {
5747 regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5748 regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5749 old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5750 old_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5751 best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *);
5752 best_regend = REGEX_TALLOC (num_regs, const CHAR_T *);
5753 reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5754 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *);
5755 reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type));
5756
5757 if (!(regstart && regend && old_regstart && old_regend && reg_info
5758 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
5759 {
5760 FREE_VARIABLES ();
5761 return -2;
5762 }
5763 }
5764 else
5765 {
5766 /* We must initialize all our variables to NULL, so that
5767 `FREE_VARIABLES' doesn't try to free them. */
5768 regstart = regend = old_regstart = old_regend = best_regstart
5769 = best_regend = reg_dummy = NULL;
5770 reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL;
5771 }
5772#endif /* MATCH_MAY_ALLOCATE */
5773
5774 /* The starting position is bogus. */
5775#ifdef WCHAR
5776 if (pos < 0 || pos > csize1 + csize2)
5777#else /* BYTE */
5778 if (pos < 0 || pos > size1 + size2)
5779#endif
5780 {
5781 FREE_VARIABLES ();
5782 return -1;
5783 }
5784
5785#ifdef WCHAR
5786 /* Allocate wchar_t array for string1 and string2 and
5787 fill them with converted string. */
5788 if (string1 == NULL && string2 == NULL)
5789 {
5790 /* We need seting up buffers here. */
5791
5792 /* We must free wcs buffers in this function. */
5793 cant_free_wcs_buf = 0;
5794
5795 if (csize1 != 0)
5796 {
5797 string1 = REGEX_TALLOC (csize1 + 1, CHAR_T);
5798 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5799 is_binary = REGEX_TALLOC (csize1 + 1, char);
5800 if (!string1 || !mbs_offset1 || !is_binary)
5801 {
5802 FREE_VAR (string1);
5803 FREE_VAR (mbs_offset1);
5804 FREE_VAR (is_binary);
5805 return -2;
5806 }
5807 }
5808 if (csize2 != 0)
5809 {
5810 string2 = REGEX_TALLOC (csize2 + 1, CHAR_T);
5811 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5812 is_binary = REGEX_TALLOC (csize2 + 1, char);
5813 if (!string2 || !mbs_offset2 || !is_binary)
5814 {
5815 FREE_VAR (string1);
5816 FREE_VAR (mbs_offset1);
5817 FREE_VAR (string2);
5818 FREE_VAR (mbs_offset2);
5819 FREE_VAR (is_binary);
5820 return -2;
5821 }
5822 size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5823 mbs_offset2, is_binary);
5824 string2[size2] = L'\0'; /* for a sentinel */
5825 FREE_VAR (is_binary);
5826 }
5827 }
5828
5829 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5830 pattern to (char*) in regex_compile. */
5831 p = pattern = (CHAR_T*)bufp->buffer;
5832 pend = (CHAR_T*)(bufp->buffer + bufp->used);
5833
5834#endif /* WCHAR */
5835
5836 /* Initialize subexpression text positions to -1 to mark ones that no
5837 start_memory/stop_memory has been seen for. Also initialize the
5838 register information struct. */
5839 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5840 {
5841 regstart[mcnt] = regend[mcnt]
5842 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5843
5844 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
5845 IS_ACTIVE (reg_info[mcnt]) = 0;
5846 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5847 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5848 }
5849
5850 /* We move `string1' into `string2' if the latter's empty -- but not if
5851 `string1' is null. */
5852 if (size2 == 0 && string1 != NULL)
5853 {
5854 string2 = string1;
5855 size2 = size1;
5856 string1 = 0;
5857 size1 = 0;
5858#ifdef WCHAR
5859 mbs_offset2 = mbs_offset1;
5860 csize2 = csize1;
5861 mbs_offset1 = NULL;
5862 csize1 = 0;
5863#endif
5864 }
5865 end1 = string1 + size1;
5866 end2 = string2 + size2;
5867
5868 /* Compute where to stop matching, within the two strings. */
5869#ifdef WCHAR
5870 if (stop <= csize1)
5871 {
5872 mcnt = count_mbs_length(mbs_offset1, stop);
5873 end_match_1 = string1 + mcnt;
5874 end_match_2 = string2;
5875 }
5876 else
5877 {
5878 if (stop > csize1 + csize2)
5879 stop = csize1 + csize2;
5880 end_match_1 = end1;
5881 mcnt = count_mbs_length(mbs_offset2, stop-csize1);
5882 end_match_2 = string2 + mcnt;
5883 }
5884 if (mcnt < 0)
5885 { /* count_mbs_length return error. */
5886 FREE_VARIABLES ();
5887 return -1;
5888 }
5889#else
5890 if (stop <= size1)
5891 {
5892 end_match_1 = string1 + stop;
5893 end_match_2 = string2;
5894 }
5895 else
5896 {
5897 end_match_1 = end1;
5898 end_match_2 = string2 + stop - size1;
5899 }
5900#endif /* WCHAR */
5901
5902 /* `p' scans through the pattern as `d' scans through the data.
5903 `dend' is the end of the input string that `d' points within. `d'
5904 is advanced into the following input string whenever necessary, but
5905 this happens before fetching; therefore, at the beginning of the
5906 loop, `d' can be pointing at the end of a string, but it cannot
5907 equal `string2'. */
5908#ifdef WCHAR
5909 if (size1 > 0 && pos <= csize1)
5910 {
5911 mcnt = count_mbs_length(mbs_offset1, pos);
5912 d = string1 + mcnt;
5913 dend = end_match_1;
5914 }
5915 else
5916 {
5917 mcnt = count_mbs_length(mbs_offset2, pos-csize1);
5918 d = string2 + mcnt;
5919 dend = end_match_2;
5920 }
5921
5922 if (mcnt < 0)
5923 { /* count_mbs_length return error. */
5924 FREE_VARIABLES ();
5925 return -1;
5926 }
5927#else
5928 if (size1 > 0 && pos <= size1)
5929 {
5930 d = string1 + pos;
5931 dend = end_match_1;
5932 }
5933 else
5934 {
5935 d = string2 + pos - size1;
5936 dend = end_match_2;
5937 }
5938#endif /* WCHAR */
5939
5940 DEBUG_PRINT1 ("The compiled pattern is:\n");
5941 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5942 DEBUG_PRINT1 ("The string to match is: `");
5943 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5944 DEBUG_PRINT1 ("'\n");
5945
5946 /* This loops over pattern commands. It exits by returning from the
5947 function if the match is complete, or it drops through if the match
5948 fails at this starting point in the input data. */
5949 for (;;)
5950 {
5951#ifdef _LIBC
5952 DEBUG_PRINT2 ("\n%p: ", p);
5953#else
5954 DEBUG_PRINT2 ("\n0x%x: ", p);
5955#endif
5956
5957 if (p == pend)
5958 { /* End of pattern means we might have succeeded. */
5959 DEBUG_PRINT1 ("end of pattern ... ");
5960
5961 /* If we haven't matched the entire string, and we want the
5962 longest match, try backtracking. */
5963 if (d != end_match_2)
5964 {
5965 /* 1 if this match ends in the same string (string1 or string2)
5966 as the best previous match. */
5967 boolean same_str_p = (FIRST_STRING_P (match_end)
5968 == MATCHING_IN_FIRST_STRING);
5969 /* 1 if this match is the best seen so far. */
5970 boolean best_match_p;
5971
5972 /* AIX compiler got confused when this was combined
5973 with the previous declaration. */
5974 if (same_str_p)
5975 best_match_p = d > match_end;
5976 else
5977 best_match_p = !MATCHING_IN_FIRST_STRING;
5978
5979 DEBUG_PRINT1 ("backtracking.\n");
5980
5981 if (!FAIL_STACK_EMPTY ())
5982 { /* More failure points to try. */
5983
5984 /* If exceeds best match so far, save it. */
5985 if (!best_regs_set || best_match_p)
5986 {
5987 best_regs_set = true;
5988 match_end = d;
5989
5990 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5991
5992 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5993 {
5994 best_regstart[mcnt] = regstart[mcnt];
5995 best_regend[mcnt] = regend[mcnt];
5996 }
5997 }
5998 goto fail;
5999 }
6000
6001 /* If no failure points, don't restore garbage. And if
6002 last match is real best match, don't restore second
6003 best one. */
6004 else if (best_regs_set && !best_match_p)
6005 {
6006 restore_best_regs:
6007 /* Restore best match. It may happen that `dend ==
6008 end_match_1' while the restored d is in string2.
6009 For example, the pattern `x.*y.*z' against the
6010 strings `x-' and `y-z-', if the two strings are
6011 not consecutive in memory. */
6012 DEBUG_PRINT1 ("Restoring best registers.\n");
6013
6014 d = match_end;
6015 dend = ((d >= string1 && d <= end1)
6016 ? end_match_1 : end_match_2);
6017
6018 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
6019 {
6020 regstart[mcnt] = best_regstart[mcnt];
6021 regend[mcnt] = best_regend[mcnt];
6022 }
6023 }
6024 } /* d != end_match_2 */
6025
6026 succeed_label:
6027 DEBUG_PRINT1 ("Accepting match.\n");
6028 /* If caller wants register contents data back, do it. */
6029 if (regs && !bufp->no_sub)
6030 {
6031 /* Have the register data arrays been allocated? */
6032 if (bufp->regs_allocated == REGS_UNALLOCATED)
6033 { /* No. So allocate them with malloc. We need one
6034 extra element beyond `num_regs' for the `-1' marker
6035 GNU code uses. */
6036 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
6037 regs->start = TALLOC (regs->num_regs, regoff_t);
6038 regs->end = TALLOC (regs->num_regs, regoff_t);
6039 if (regs->start == NULL || regs->end == NULL)
6040 {
6041 FREE_VARIABLES ();
6042 return -2;
6043 }
6044 bufp->regs_allocated = REGS_REALLOCATE;
6045 }
6046 else if (bufp->regs_allocated == REGS_REALLOCATE)
6047 { /* Yes. If we need more elements than were already
6048 allocated, reallocate them. If we need fewer, just
6049 leave it alone. */
6050 if (regs->num_regs < num_regs + 1)
6051 {
6052 regs->num_regs = num_regs + 1;
6053 RETALLOC (regs->start, regs->num_regs, regoff_t);
6054 RETALLOC (regs->end, regs->num_regs, regoff_t);
6055 if (regs->start == NULL || regs->end == NULL)
6056 {
6057 FREE_VARIABLES ();
6058 return -2;
6059 }
6060 }
6061 }
6062 else
6063 {
6064 /* These braces fend off a "empty body in an else-statement"
6065 warning under GCC when assert expands to nothing. */
6066 assert (bufp->regs_allocated == REGS_FIXED);
6067 }
6068
6069 /* Convert the pointer data in `regstart' and `regend' to
6070 indices. Register zero has to be set differently,
6071 since we haven't kept track of any info for it. */
6072 if (regs->num_regs > 0)
6073 {
6074 regs->start[0] = pos;
6075#ifdef WCHAR
6076 if (MATCHING_IN_FIRST_STRING)
6077 regs->end[0] = mbs_offset1 != NULL ?
6078 mbs_offset1[d-string1] : 0;
6079 else
6080 regs->end[0] = csize1 + (mbs_offset2 != NULL ?
6081 mbs_offset2[d-string2] : 0);
6082#else
6083 regs->end[0] = (MATCHING_IN_FIRST_STRING
6084 ? ((regoff_t) (d - string1))
6085 : ((regoff_t) (d - string2 + size1)));
6086#endif /* WCHAR */
6087 }
6088
6089 /* Go through the first `min (num_regs, regs->num_regs)'
6090 registers, since that is all we initialized. */
6091 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
6092 mcnt++)
6093 {
6094 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
6095 regs->start[mcnt] = regs->end[mcnt] = -1;
6096 else
6097 {
6098 regs->start[mcnt]
6099 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
6100 regs->end[mcnt]
6101 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
6102 }
6103 }
6104
6105 /* If the regs structure we return has more elements than
6106 were in the pattern, set the extra elements to -1. If
6107 we (re)allocated the registers, this is the case,
6108 because we always allocate enough to have at least one
6109 -1 at the end. */
6110 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
6111 regs->start[mcnt] = regs->end[mcnt] = -1;
6112 } /* regs && !bufp->no_sub */
6113
6114 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
6115 nfailure_points_pushed, nfailure_points_popped,
6116 nfailure_points_pushed - nfailure_points_popped);
6117 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
6118
6119#ifdef WCHAR
6120 if (MATCHING_IN_FIRST_STRING)
6121 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
6122 else
6123 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
6124 csize1;
6125 mcnt -= pos;
6126#else
6127 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
6128 ? string1
6129 : string2 - size1);
6130#endif /* WCHAR */
6131
6132 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
6133
6134 FREE_VARIABLES ();
6135 return mcnt;
6136 }
6137
6138 /* Otherwise match next pattern command. */
6139 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
6140 {
6141 /* Ignore these. Used to ignore the n of succeed_n's which
6142 currently have n == 0. */
6143 case no_op:
6144 DEBUG_PRINT1 ("EXECUTING no_op.\n");
6145 break;
6146
6147 case succeed:
6148 DEBUG_PRINT1 ("EXECUTING succeed.\n");
6149 goto succeed_label;
6150
6151 /* Match the next n pattern characters exactly. The following
6152 byte in the pattern defines n, and the n bytes after that
6153 are the characters to match. */
6154 case exactn:
6155#ifdef MBS_SUPPORT
6156 case exactn_bin:
6157#endif
6158 mcnt = *p++;
6159 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
6160
6161 /* This is written out as an if-else so we don't waste time
6162 testing `translate' inside the loop. */
6163 if (translate)
6164 {
6165 do
6166 {
6167 PREFETCH ();
6168#ifdef WCHAR
6169 if (*d <= 0xff)
6170 {
6171 if ((UCHAR_T) translate[(unsigned char) *d++]
6172 != (UCHAR_T) *p++)
6173 goto fail;
6174 }
6175 else
6176 {
6177 if (*d++ != (CHAR_T) *p++)
6178 goto fail;
6179 }
6180#else
6181 if ((UCHAR_T) translate[(unsigned char) *d++]
6182 != (UCHAR_T) *p++)
6183 goto fail;
6184#endif /* WCHAR */
6185 }
6186 while (--mcnt);
6187 }
6188 else
6189 {
6190 do
6191 {
6192 PREFETCH ();
6193 if (*d++ != (CHAR_T) *p++) goto fail;
6194 }
6195 while (--mcnt);
6196 }
6197 SET_REGS_MATCHED ();
6198 break;
6199
6200
6201 /* Match any character except possibly a newline or a null. */
6202 case anychar:
6203 DEBUG_PRINT1 ("EXECUTING anychar.\n");
6204
6205 PREFETCH ();
6206
6207 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
6208 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
6209 goto fail;
6210
6211 SET_REGS_MATCHED ();
6212 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
6213 d++;
6214 break;
6215
6216
6217 case charset:
6218 case charset_not:
6219 {
6220 register UCHAR_T c;
6221#ifdef WCHAR
6222 unsigned int i, char_class_length, coll_symbol_length,
6223 equiv_class_length, ranges_length, chars_length, length;
6224 CHAR_T *workp, *workp2, *charset_top;
6225#define WORK_BUFFER_SIZE 128
6226 CHAR_T str_buf[WORK_BUFFER_SIZE];
6227# ifdef _LIBC
6228 uint32_t nrules;
6229# endif /* _LIBC */
6230#endif /* WCHAR */
6231 boolean not = (re_opcode_t) *(p - 1) == charset_not;
6232
6233 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
6234 PREFETCH ();
6235 c = TRANSLATE (*d); /* The character to match. */
6236#ifdef WCHAR
6237# ifdef _LIBC
6238 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
6239# endif /* _LIBC */
6240 charset_top = p - 1;
6241 char_class_length = *p++;
6242 coll_symbol_length = *p++;
6243 equiv_class_length = *p++;
6244 ranges_length = *p++;
6245 chars_length = *p++;
6246 /* p points charset[6], so the address of the next instruction
6247 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
6248 where l=length of char_classes, m=length of collating_symbol,
6249 n=equivalence_class, o=length of char_range,
6250 p'=length of character. */
6251 workp = p;
6252 /* Update p to indicate the next instruction. */
6253 p += char_class_length + coll_symbol_length+ equiv_class_length +
6254 2*ranges_length + chars_length;
6255
6256 /* match with char_class? */
6257 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
6258 {
6259 wctype_t wctype;
6260 uintptr_t alignedp = ((uintptr_t)workp
6261 + __alignof__(wctype_t) - 1)
6262 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
6263 wctype = *((wctype_t*)alignedp);
6264 workp += CHAR_CLASS_SIZE;
6265 if (iswctype((wint_t)c, wctype))
6266 goto char_set_matched;
6267 }
6268
6269 /* match with collating_symbol? */
6270# ifdef _LIBC
6271 if (nrules != 0)
6272 {
6273 const unsigned char *extra = (const unsigned char *)
6274 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6275
6276 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6277 workp++)
6278 {
6279 int32_t *wextra;
6280 wextra = (int32_t*)(extra + *workp++);
6281 for (i = 0; i < *wextra; ++i)
6282 if (TRANSLATE(d[i]) != wextra[1 + i])
6283 break;
6284
6285 if (i == *wextra)
6286 {
6287 /* Update d, however d will be incremented at
6288 char_set_matched:, we decrement d here. */
6289 d += i - 1;
6290 goto char_set_matched;
6291 }
6292 }
6293 }
6294 else /* (nrules == 0) */
6295# endif
6296 /* If we can't look up collation data, we use wcscoll
6297 instead. */
6298 {
6299 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6300 {
6301 const CHAR_T *backup_d = d, *backup_dend = dend;
6302 length = wcslen(workp);
6303
6304 /* If wcscoll(the collating symbol, whole string) > 0,
6305 any substring of the string never match with the
6306 collating symbol. */
6307 if (wcscoll(workp, d) > 0)
6308 {
6309 workp += length + 1;
6310 continue;
6311 }
6312
6313 /* First, we compare the collating symbol with
6314 the first character of the string.
6315 If it don't match, we add the next character to
6316 the compare buffer in turn. */
6317 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6318 {
6319 int match;
6320 if (d == dend)
6321 {
6322 if (dend == end_match_2)
6323 break;
6324 d = string2;
6325 dend = end_match_2;
6326 }
6327
6328 /* add next character to the compare buffer. */
6329 str_buf[i] = TRANSLATE(*d);
6330 str_buf[i+1] = '\0';
6331
6332 match = wcscoll(workp, str_buf);
6333 if (match == 0)
6334 goto char_set_matched;
6335
6336 if (match < 0)
6337 /* (str_buf > workp) indicate (str_buf + X > workp),
6338 because for all X (str_buf + X > str_buf).
6339 So we don't need continue this loop. */
6340 break;
6341
6342 /* Otherwise(str_buf < workp),
6343 (str_buf+next_character) may equals (workp).
6344 So we continue this loop. */
6345 }
6346 /* not matched */
6347 d = backup_d;
6348 dend = backup_dend;
6349 workp += length + 1;
6350 }
6351 }
6352 /* match with equivalence_class? */
6353# ifdef _LIBC
6354 if (nrules != 0)
6355 {
6356 const CHAR_T *backup_d = d, *backup_dend = dend;
6357 /* Try to match the equivalence class against
6358 those known to the collate implementation. */
6359 const int32_t *table;
6360 const int32_t *weights;
6361 const int32_t *extra;
6362 const int32_t *indirect;
6363 int32_t idx, idx2;
6364 wint_t *cp;
6365 size_t len;
6366
6367 /* This #include defines a local function! */
6368# include <locale/weightwc.h>
6369
6370 table = (const int32_t *)
6371 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6372 weights = (const wint_t *)
6373 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6374 extra = (const wint_t *)
6375 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6376 indirect = (const int32_t *)
6377 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6378
6379 /* Write 1 collating element to str_buf, and
6380 get its index. */
6381 idx2 = 0;
6382
6383 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6384 {
6385 cp = (wint_t*)str_buf;
6386 if (d == dend)
6387 {
6388 if (dend == end_match_2)
6389 break;
6390 d = string2;
6391 dend = end_match_2;
6392 }
6393 str_buf[i] = TRANSLATE(*(d+i));
6394 str_buf[i+1] = '\0'; /* sentinel */
6395 idx2 = findidx ((const wint_t**)&cp);
6396 }
6397
6398 /* Update d, however d will be incremented at
6399 char_set_matched:, we decrement d here. */
6400 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6401 if (d >= dend)
6402 {
6403 if (dend == end_match_2)
6404 d = dend;
6405 else
6406 {
6407 d = string2;
6408 dend = end_match_2;
6409 }
6410 }
6411
6412 len = weights[idx2];
6413
6414 for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6415 workp++)
6416 {
6417 idx = (int32_t)*workp;
6418 /* We already checked idx != 0 in regex_compile. */
6419
6420 if (idx2 != 0 && len == weights[idx])
6421 {
6422 int cnt = 0;
6423 while (cnt < len && (weights[idx + 1 + cnt]
6424 == weights[idx2 + 1 + cnt]))
6425 ++cnt;
6426
6427 if (cnt == len)
6428 goto char_set_matched;
6429 }
6430 }
6431 /* not matched */
6432 d = backup_d;
6433 dend = backup_dend;
6434 }
6435 else /* (nrules == 0) */
6436# endif
6437 /* If we can't look up collation data, we use wcscoll
6438 instead. */
6439 {
6440 for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6441 {
6442 const CHAR_T *backup_d = d, *backup_dend = dend;
6443 length = wcslen(workp);
6444
6445 /* If wcscoll(the collating symbol, whole string) > 0,
6446 any substring of the string never match with the
6447 collating symbol. */
6448 if (wcscoll(workp, d) > 0)
6449 {
6450 workp += length + 1;
6451 break;
6452 }
6453
6454 /* First, we compare the equivalence class with
6455 the first character of the string.
6456 If it don't match, we add the next character to
6457 the compare buffer in turn. */
6458 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6459 {
6460 int match;
6461 if (d == dend)
6462 {
6463 if (dend == end_match_2)
6464 break;
6465 d = string2;
6466 dend = end_match_2;
6467 }
6468
6469 /* add next character to the compare buffer. */
6470 str_buf[i] = TRANSLATE(*d);
6471 str_buf[i+1] = '\0';
6472
6473 match = wcscoll(workp, str_buf);
6474
6475 if (match == 0)
6476 goto char_set_matched;
6477
6478 if (match < 0)
6479 /* (str_buf > workp) indicate (str_buf + X > workp),
6480 because for all X (str_buf + X > str_buf).
6481 So we don't need continue this loop. */
6482 break;
6483
6484 /* Otherwise(str_buf < workp),
6485 (str_buf+next_character) may equals (workp).
6486 So we continue this loop. */
6487 }
6488 /* not matched */
6489 d = backup_d;
6490 dend = backup_dend;
6491 workp += length + 1;
6492 }
6493 }
6494
6495 /* match with char_range? */
6496#ifdef _LIBC
6497 if (nrules != 0)
6498 {
6499 uint32_t collseqval;
6500 const char *collseq = (const char *)
6501 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6502
6503 collseqval = collseq_table_lookup (collseq, c);
6504
6505 for (; workp < p - chars_length ;)
6506 {
6507 uint32_t start_val, end_val;
6508
6509 /* We already compute the collation sequence value
6510 of the characters (or collating symbols). */
6511 start_val = (uint32_t) *workp++; /* range_start */
6512 end_val = (uint32_t) *workp++; /* range_end */
6513
6514 if (start_val <= collseqval && collseqval <= end_val)
6515 goto char_set_matched;
6516 }
6517 }
6518 else
6519#endif
6520 {
6521 /* We set range_start_char at str_buf[0], range_end_char
6522 at str_buf[4], and compared char at str_buf[2]. */
6523 str_buf[1] = 0;
6524 str_buf[2] = c;
6525 str_buf[3] = 0;
6526 str_buf[5] = 0;
6527 for (; workp < p - chars_length ;)
6528 {
6529 wchar_t *range_start_char, *range_end_char;
6530
6531 /* match if (range_start_char <= c <= range_end_char). */
6532
6533 /* If range_start(or end) < 0, we assume -range_start(end)
6534 is the offset of the collating symbol which is specified
6535 as the character of the range start(end). */
6536
6537 /* range_start */
6538 if (*workp < 0)
6539 range_start_char = charset_top - (*workp++);
6540 else
6541 {
6542 str_buf[0] = *workp++;
6543 range_start_char = str_buf;
6544 }
6545
6546 /* range_end */
6547 if (*workp < 0)
6548 range_end_char = charset_top - (*workp++);
6549 else
6550 {
6551 str_buf[4] = *workp++;
6552 range_end_char = str_buf + 4;
6553 }
6554
6555 if (wcscoll(range_start_char, str_buf+2) <= 0 &&
6556 wcscoll(str_buf+2, range_end_char) <= 0)
6557
6558 goto char_set_matched;
6559 }
6560 }
6561
6562 /* match with char? */
6563 for (; workp < p ; workp++)
6564 if (c == *workp)
6565 goto char_set_matched;
6566
6567 not = !not;
6568
6569 char_set_matched:
6570 if (not) goto fail;
6571#else
6572 /* Cast to `unsigned' instead of `unsigned char' in case the
6573 bit list is a full 32 bytes long. */
6574 if (c < (unsigned) (*p * BYTEWIDTH)
6575 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6576 not = !not;
6577
6578 p += 1 + *p;
6579
6580 if (!not) goto fail;
6581#undef WORK_BUFFER_SIZE
6582#endif /* WCHAR */
6583 SET_REGS_MATCHED ();
6584 d++;
6585 break;
6586 }
6587
6588
6589 /* The beginning of a group is represented by start_memory.
6590 The arguments are the register number in the next byte, and the
6591 number of groups inner to this one in the next. The text
6592 matched within the group is recorded (in the internal
6593 registers data structure) under the register number. */
6594 case start_memory:
6595 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6596 (long int) *p, (long int) p[1]);
6597
6598 /* Find out if this group can match the empty string. */
6599 p1 = p; /* To send to group_match_null_string_p. */
6600
6601 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6602 REG_MATCH_NULL_STRING_P (reg_info[*p])
6603 = PREFIX(group_match_null_string_p) (&p1, pend, reg_info);
6604
6605 /* Save the position in the string where we were the last time
6606 we were at this open-group operator in case the group is
6607 operated upon by a repetition operator, e.g., with `(a*)*b'
6608 against `ab'; then we want to ignore where we are now in
6609 the string in case this attempt to match fails. */
6610 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6611 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6612 : regstart[*p];
6613 DEBUG_PRINT2 (" old_regstart: %d\n",
6614 POINTER_TO_OFFSET (old_regstart[*p]));
6615
6616 regstart[*p] = d;
6617 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6618
6619 IS_ACTIVE (reg_info[*p]) = 1;
6620 MATCHED_SOMETHING (reg_info[*p]) = 0;
6621
6622 /* Clear this whenever we change the register activity status. */
6623 set_regs_matched_done = 0;
6624
6625 /* This is the new highest active register. */
6626 highest_active_reg = *p;
6627
6628 /* If nothing was active before, this is the new lowest active
6629 register. */
6630 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6631 lowest_active_reg = *p;
6632
6633 /* Move past the register number and inner group count. */
6634 p += 2;
6635 just_past_start_mem = p;
6636
6637 break;
6638
6639
6640 /* The stop_memory opcode represents the end of a group. Its
6641 arguments are the same as start_memory's: the register
6642 number, and the number of inner groups. */
6643 case stop_memory:
6644 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6645 (long int) *p, (long int) p[1]);
6646
6647 /* We need to save the string position the last time we were at
6648 this close-group operator in case the group is operated
6649 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6650 against `aba'; then we want to ignore where we are now in
6651 the string in case this attempt to match fails. */
6652 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6653 ? REG_UNSET (regend[*p]) ? d : regend[*p]
6654 : regend[*p];
6655 DEBUG_PRINT2 (" old_regend: %d\n",
6656 POINTER_TO_OFFSET (old_regend[*p]));
6657
6658 regend[*p] = d;
6659 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6660
6661 /* This register isn't active anymore. */
6662 IS_ACTIVE (reg_info[*p]) = 0;
6663
6664 /* Clear this whenever we change the register activity status. */
6665 set_regs_matched_done = 0;
6666
6667 /* If this was the only register active, nothing is active
6668 anymore. */
6669 if (lowest_active_reg == highest_active_reg)
6670 {
6671 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6672 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6673 }
6674 else
6675 { /* We must scan for the new highest active register, since
6676 it isn't necessarily one less than now: consider
6677 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6678 new highest active register is 1. */
6679 UCHAR_T r = *p - 1;
6680 while (r > 0 && !IS_ACTIVE (reg_info[r]))
6681 r--;
6682
6683 /* If we end up at register zero, that means that we saved
6684 the registers as the result of an `on_failure_jump', not
6685 a `start_memory', and we jumped to past the innermost
6686 `stop_memory'. For example, in ((.)*) we save
6687 registers 1 and 2 as a result of the *, but when we pop
6688 back to the second ), we are at the stop_memory 1.
6689 Thus, nothing is active. */
6690 if (r == 0)
6691 {
6692 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6693 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6694 }
6695 else
6696 highest_active_reg = r;
6697 }
6698
6699 /* If just failed to match something this time around with a
6700 group that's operated on by a repetition operator, try to
6701 force exit from the ``loop'', and restore the register
6702 information for this group that we had before trying this
6703 last match. */
6704 if ((!MATCHED_SOMETHING (reg_info[*p])
6705 || just_past_start_mem == p - 1)
6706 && (p + 2) < pend)
6707 {
6708 boolean is_a_jump_n = false;
6709
6710 p1 = p + 2;
6711 mcnt = 0;
6712 switch ((re_opcode_t) *p1++)
6713 {
6714 case jump_n:
6715 is_a_jump_n = true;
6716 case pop_failure_jump:
6717 case maybe_pop_jump:
6718 case jump:
6719 case dummy_failure_jump:
6720 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6721 if (is_a_jump_n)
6722 p1 += OFFSET_ADDRESS_SIZE;
6723 break;
6724
6725 default:
6726 /* do nothing */ ;
6727 }
6728 p1 += mcnt;
6729
6730 /* If the next operation is a jump backwards in the pattern
6731 to an on_failure_jump right before the start_memory
6732 corresponding to this stop_memory, exit from the loop
6733 by forcing a failure after pushing on the stack the
6734 on_failure_jump's jump in the pattern, and d. */
6735 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6736 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6737 && p1[2+OFFSET_ADDRESS_SIZE] == *p)
6738 {
6739 /* If this group ever matched anything, then restore
6740 what its registers were before trying this last
6741 failed match, e.g., with `(a*)*b' against `ab' for
6742 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6743 against `aba' for regend[3].
6744
6745 Also restore the registers for inner groups for,
6746 e.g., `((a*)(b*))*' against `aba' (register 3 would
6747 otherwise get trashed). */
6748
6749 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6750 {
6751 unsigned r;
6752
6753 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6754
6755 /* Restore this and inner groups' (if any) registers. */
6756 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6757 r++)
6758 {
6759 regstart[r] = old_regstart[r];
6760
6761 /* xx why this test? */
6762 if (old_regend[r] >= regstart[r])
6763 regend[r] = old_regend[r];
6764 }
6765 }
6766 p1++;
6767 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6768 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6769
6770 goto fail;
6771 }
6772 }
6773
6774 /* Move past the register number and the inner group count. */
6775 p += 2;
6776 break;
6777
6778
6779 /* \<digit> has been turned into a `duplicate' command which is
6780 followed by the numeric value of <digit> as the register number. */
6781 case duplicate:
6782 {
6783 register const CHAR_T *d2, *dend2;
6784 int regno = *p++; /* Get which register to match against. */
6785 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6786
6787 /* Can't back reference a group which we've never matched. */
6788 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6789 goto fail;
6790
6791 /* Where in input to try to start matching. */
6792 d2 = regstart[regno];
6793
6794 /* Where to stop matching; if both the place to start and
6795 the place to stop matching are in the same string, then
6796 set to the place to stop, otherwise, for now have to use
6797 the end of the first string. */
6798
6799 dend2 = ((FIRST_STRING_P (regstart[regno])
6800 == FIRST_STRING_P (regend[regno]))
6801 ? regend[regno] : end_match_1);
6802 for (;;)
6803 {
6804 /* If necessary, advance to next segment in register
6805 contents. */
6806 while (d2 == dend2)
6807 {
6808 if (dend2 == end_match_2) break;
6809 if (dend2 == regend[regno]) break;
6810
6811 /* End of string1 => advance to string2. */
6812 d2 = string2;
6813 dend2 = regend[regno];
6814 }
6815 /* At end of register contents => success */
6816 if (d2 == dend2) break;
6817
6818 /* If necessary, advance to next segment in data. */
6819 PREFETCH ();
6820
6821 /* How many characters left in this segment to match. */
6822 mcnt = dend - d;
6823
6824 /* Want how many consecutive characters we can match in
6825 one shot, so, if necessary, adjust the count. */
6826 if (mcnt > dend2 - d2)
6827 mcnt = dend2 - d2;
6828
6829 /* Compare that many; failure if mismatch, else move
6830 past them. */
6831 if (translate
6832 ? PREFIX(bcmp_translate) (d, d2, mcnt, translate)
6833 : memcmp (d, d2, mcnt*sizeof(UCHAR_T)))
6834 goto fail;
6835 d += mcnt, d2 += mcnt;
6836
6837 /* Do this because we've match some characters. */
6838 SET_REGS_MATCHED ();
6839 }
6840 }
6841 break;
6842
6843
6844 /* begline matches the empty string at the beginning of the string
6845 (unless `not_bol' is set in `bufp'), and, if
6846 `newline_anchor' is set, after newlines. */
6847 case begline:
6848 DEBUG_PRINT1 ("EXECUTING begline.\n");
6849
6850 if (AT_STRINGS_BEG (d))
6851 {
6852 if (!bufp->not_bol) break;
6853 }
6854 else if (d[-1] == '\n' && bufp->newline_anchor)
6855 {
6856 break;
6857 }
6858 /* In all other cases, we fail. */
6859 goto fail;
6860
6861
6862 /* endline is the dual of begline. */
6863 case endline:
6864 DEBUG_PRINT1 ("EXECUTING endline.\n");
6865
6866 if (AT_STRINGS_END (d))
6867 {
6868 if (!bufp->not_eol) break;
6869 }
6870
6871 /* We have to ``prefetch'' the next character. */
6872 else if ((d == end1 ? *string2 : *d) == '\n'
6873 && bufp->newline_anchor)
6874 {
6875 break;
6876 }
6877 goto fail;
6878
6879
6880 /* Match at the very beginning of the data. */
6881 case begbuf:
6882 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6883 if (AT_STRINGS_BEG (d))
6884 break;
6885 goto fail;
6886
6887
6888 /* Match at the very end of the data. */
6889 case endbuf:
6890 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
6891 if (AT_STRINGS_END (d))
6892 break;
6893 goto fail;
6894
6895
6896 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
6897 pushes NULL as the value for the string on the stack. Then
6898 `pop_failure_point' will keep the current value for the
6899 string, instead of restoring it. To see why, consider
6900 matching `foo\nbar' against `.*\n'. The .* matches the foo;
6901 then the . fails against the \n. But the next thing we want
6902 to do is match the \n against the \n; if we restored the
6903 string value, we would be back at the foo.
6904
6905 Because this is used only in specific cases, we don't need to
6906 check all the things that `on_failure_jump' does, to make
6907 sure the right things get saved on the stack. Hence we don't
6908 share its code. The only reason to push anything on the
6909 stack at all is that otherwise we would have to change
6910 `anychar's code to do something besides goto fail in this
6911 case; that seems worse than this. */
6912 case on_failure_keep_string_jump:
6913 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
6914
6915 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6916#ifdef _LIBC
6917 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
6918#else
6919 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
6920#endif
6921
6922 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
6923 break;
6924
6925
6926 /* Uses of on_failure_jump:
6927
6928 Each alternative starts with an on_failure_jump that points
6929 to the beginning of the next alternative. Each alternative
6930 except the last ends with a jump that in effect jumps past
6931 the rest of the alternatives. (They really jump to the
6932 ending jump of the following alternative, because tensioning
6933 these jumps is a hassle.)
6934
6935 Repeats start with an on_failure_jump that points past both
6936 the repetition text and either the following jump or
6937 pop_failure_jump back to this on_failure_jump. */
6938 case on_failure_jump:
6939 on_failure:
6940 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
6941
6942 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6943#ifdef _LIBC
6944 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
6945#else
6946 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
6947#endif
6948
6949 /* If this on_failure_jump comes right before a group (i.e.,
6950 the original * applied to a group), save the information
6951 for that group and all inner ones, so that if we fail back
6952 to this point, the group's information will be correct.
6953 For example, in \(a*\)*\1, we need the preceding group,
6954 and in \(zz\(a*\)b*\)\2, we need the inner group. */
6955
6956 /* We can't use `p' to check ahead because we push
6957 a failure point to `p + mcnt' after we do this. */
6958 p1 = p;
6959
6960 /* We need to skip no_op's before we look for the
6961 start_memory in case this on_failure_jump is happening as
6962 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
6963 against aba. */
6964 while (p1 < pend && (re_opcode_t) *p1 == no_op)
6965 p1++;
6966
6967 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
6968 {
6969 /* We have a new highest active register now. This will
6970 get reset at the start_memory we are about to get to,
6971 but we will have saved all the registers relevant to
6972 this repetition op, as described above. */
6973 highest_active_reg = *(p1 + 1) + *(p1 + 2);
6974 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6975 lowest_active_reg = *(p1 + 1);
6976 }
6977
6978 DEBUG_PRINT1 (":\n");
6979 PUSH_FAILURE_POINT (p + mcnt, d, -2);
6980 break;
6981
6982
6983 /* A smart repeat ends with `maybe_pop_jump'.
6984 We change it to either `pop_failure_jump' or `jump'. */
6985 case maybe_pop_jump:
6986 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6987 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
6988 {
6989 register UCHAR_T *p2 = p;
6990
6991 /* Compare the beginning of the repeat with what in the
6992 pattern follows its end. If we can establish that there
6993 is nothing that they would both match, i.e., that we
6994 would have to backtrack because of (as in, e.g., `a*a')
6995 then we can change to pop_failure_jump, because we'll
6996 never have to backtrack.
6997
6998 This is not true in the case of alternatives: in
6999 `(a|ab)*' we do need to backtrack to the `ab' alternative
7000 (e.g., if the string was `ab'). But instead of trying to
7001 detect that here, the alternative has put on a dummy
7002 failure point which is what we will end up popping. */
7003
7004 /* Skip over open/close-group commands.
7005 If what follows this loop is a ...+ construct,
7006 look at what begins its body, since we will have to
7007 match at least one of that. */
7008 while (1)
7009 {
7010 if (p2 + 2 < pend
7011 && ((re_opcode_t) *p2 == stop_memory
7012 || (re_opcode_t) *p2 == start_memory))
7013 p2 += 3;
7014 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
7015 && (re_opcode_t) *p2 == dummy_failure_jump)
7016 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
7017 else
7018 break;
7019 }
7020
7021 p1 = p + mcnt;
7022 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
7023 to the `maybe_finalize_jump' of this case. Examine what
7024 follows. */
7025
7026 /* If we're at the end of the pattern, we can change. */
7027 if (p2 == pend)
7028 {
7029 /* Consider what happens when matching ":\(.*\)"
7030 against ":/". I don't really understand this code
7031 yet. */
7032 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7033 pop_failure_jump;
7034 DEBUG_PRINT1
7035 (" End of pattern: change to `pop_failure_jump'.\n");
7036 }
7037
7038 else if ((re_opcode_t) *p2 == exactn
7039#ifdef MBS_SUPPORT
7040 || (re_opcode_t) *p2 == exactn_bin
7041#endif
7042 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
7043 {
7044 register UCHAR_T c
7045 = *p2 == (UCHAR_T) endline ? '\n' : p2[2];
7046
7047 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
7048#ifdef MBS_SUPPORT
7049 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
7050#endif
7051 ) && p1[3+OFFSET_ADDRESS_SIZE] != c)
7052 {
7053 p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T)
7054 pop_failure_jump;
7055#ifdef WCHAR
7056 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
7057 (wint_t) c,
7058 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
7059#else
7060 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
7061 (char) c,
7062 (char) p1[3+OFFSET_ADDRESS_SIZE]);
7063#endif
7064 }
7065
7066#ifndef WCHAR
7067 else if ((re_opcode_t) p1[3] == charset
7068 || (re_opcode_t) p1[3] == charset_not)
7069 {
7070 int not = (re_opcode_t) p1[3] == charset_not;
7071
7072 if (c < (unsigned) (p1[4] * BYTEWIDTH)
7073 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
7074 not = !not;
7075
7076 /* `not' is equal to 1 if c would match, which means
7077 that we can't change to pop_failure_jump. */
7078 if (!not)
7079 {
7080 p[-3] = (unsigned char) pop_failure_jump;
7081 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7082 }
7083 }
7084#endif /* not WCHAR */
7085 }
7086#ifndef WCHAR
7087 else if ((re_opcode_t) *p2 == charset)
7088 {
7089 /* We win if the first character of the loop is not part
7090 of the charset. */
7091 if ((re_opcode_t) p1[3] == exactn
7092 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
7093 && (p2[2 + p1[5] / BYTEWIDTH]
7094 & (1 << (p1[5] % BYTEWIDTH)))))
7095 {
7096 p[-3] = (unsigned char) pop_failure_jump;
7097 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7098 }
7099
7100 else if ((re_opcode_t) p1[3] == charset_not)
7101 {
7102 int idx;
7103 /* We win if the charset_not inside the loop
7104 lists every character listed in the charset after. */
7105 for (idx = 0; idx < (int) p2[1]; idx++)
7106 if (! (p2[2 + idx] == 0
7107 || (idx < (int) p1[4]
7108 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
7109 break;
7110
7111 if (idx == p2[1])
7112 {
7113 p[-3] = (unsigned char) pop_failure_jump;
7114 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7115 }
7116 }
7117 else if ((re_opcode_t) p1[3] == charset)
7118 {
7119 int idx;
7120 /* We win if the charset inside the loop
7121 has no overlap with the one after the loop. */
7122 for (idx = 0;
7123 idx < (int) p2[1] && idx < (int) p1[4];
7124 idx++)
7125 if ((p2[2 + idx] & p1[5 + idx]) != 0)
7126 break;
7127
7128 if (idx == p2[1] || idx == p1[4])
7129 {
7130 p[-3] = (unsigned char) pop_failure_jump;
7131 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
7132 }
7133 }
7134 }
7135#endif /* not WCHAR */
7136 }
7137 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
7138 if ((re_opcode_t) p[-1] != pop_failure_jump)
7139 {
7140 p[-1] = (UCHAR_T) jump;
7141 DEBUG_PRINT1 (" Match => jump.\n");
7142 goto unconditional_jump;
7143 }
7144 /* Note fall through. */
7145
7146
7147 /* The end of a simple repeat has a pop_failure_jump back to
7148 its matching on_failure_jump, where the latter will push a
7149 failure point. The pop_failure_jump takes off failure
7150 points put on by this pop_failure_jump's matching
7151 on_failure_jump; we got through the pattern to here from the
7152 matching on_failure_jump, so didn't fail. */
7153 case pop_failure_jump:
7154 {
7155 /* We need to pass separate storage for the lowest and
7156 highest registers, even though we don't care about the
7157 actual values. Otherwise, we will restore only one
7158 register from the stack, since lowest will == highest in
7159 `pop_failure_point'. */
7160 active_reg_t dummy_low_reg, dummy_high_reg;
7161 UCHAR_T *pdummy = NULL;
7162 const CHAR_T *sdummy = NULL;
7163
7164 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
7165 POP_FAILURE_POINT (sdummy, pdummy,
7166 dummy_low_reg, dummy_high_reg,
7167 reg_dummy, reg_dummy, reg_info_dummy);
7168 }
7169 /* Note fall through. */
7170
7171 unconditional_jump:
7172#ifdef _LIBC
7173 DEBUG_PRINT2 ("\n%p: ", p);
7174#else
7175 DEBUG_PRINT2 ("\n0x%x: ", p);
7176#endif
7177 /* Note fall through. */
7178
7179 /* Unconditionally jump (without popping any failure points). */
7180 case jump:
7181 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
7182 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
7183 p += mcnt; /* Do the jump. */
7184#ifdef _LIBC
7185 DEBUG_PRINT2 ("(to %p).\n", p);
7186#else
7187 DEBUG_PRINT2 ("(to 0x%x).\n", p);
7188#endif
7189 break;
7190
7191
7192 /* We need this opcode so we can detect where alternatives end
7193 in `group_match_null_string_p' et al. */
7194 case jump_past_alt:
7195 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
7196 goto unconditional_jump;
7197
7198
7199 /* Normally, the on_failure_jump pushes a failure point, which
7200 then gets popped at pop_failure_jump. We will end up at
7201 pop_failure_jump, also, and with a pattern of, say, `a+', we
7202 are skipping over the on_failure_jump, so we have to push
7203 something meaningless for pop_failure_jump to pop. */
7204 case dummy_failure_jump:
7205 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
7206 /* It doesn't matter what we push for the string here. What
7207 the code at `fail' tests is the value for the pattern. */
7208 PUSH_FAILURE_POINT (NULL, NULL, -2);
7209 goto unconditional_jump;
7210
7211
7212 /* At the end of an alternative, we need to push a dummy failure
7213 point in case we are followed by a `pop_failure_jump', because
7214 we don't want the failure point for the alternative to be
7215 popped. For example, matching `(a|ab)*' against `aab'
7216 requires that we match the `ab' alternative. */
7217 case push_dummy_failure:
7218 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
7219 /* See comments just above at `dummy_failure_jump' about the
7220 two zeroes. */
7221 PUSH_FAILURE_POINT (NULL, NULL, -2);
7222 break;
7223
7224 /* Have to succeed matching what follows at least n times.
7225 After that, handle like `on_failure_jump'. */
7226 case succeed_n:
7227 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7228 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
7229
7230 assert (mcnt >= 0);
7231 /* Originally, this is how many times we HAVE to succeed. */
7232 if (mcnt > 0)
7233 {
7234 mcnt--;
7235 p += OFFSET_ADDRESS_SIZE;
7236 STORE_NUMBER_AND_INCR (p, mcnt);
7237#ifdef _LIBC
7238 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
7239 , mcnt);
7240#else
7241 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
7242 , mcnt);
7243#endif
7244 }
7245 else if (mcnt == 0)
7246 {
7247#ifdef _LIBC
7248 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
7249 p + OFFSET_ADDRESS_SIZE);
7250#else
7251 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
7252 p + OFFSET_ADDRESS_SIZE);
7253#endif /* _LIBC */
7254
7255#ifdef WCHAR
7256 p[1] = (UCHAR_T) no_op;
7257#else
7258 p[2] = (UCHAR_T) no_op;
7259 p[3] = (UCHAR_T) no_op;
7260#endif /* WCHAR */
7261 goto on_failure;
7262 }
7263 break;
7264
7265 case jump_n:
7266 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7267 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7268
7269 /* Originally, this is how many times we CAN jump. */
7270 if (mcnt)
7271 {
7272 mcnt--;
7273 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7274
7275#ifdef _LIBC
7276 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7277 mcnt);
7278#else
7279 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7280 mcnt);
7281#endif /* _LIBC */
7282 goto unconditional_jump;
7283 }
7284 /* If don't have to jump any more, skip over the rest of command. */
7285 else
7286 p += 2 * OFFSET_ADDRESS_SIZE;
7287 break;
7288
7289 case set_number_at:
7290 {
7291 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7292
7293 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7294 p1 = p + mcnt;
7295 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7296#ifdef _LIBC
7297 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7298#else
7299 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7300#endif
7301 STORE_NUMBER (p1, mcnt);
7302 break;
7303 }
7304
7305#if 0
7306 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7307 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7308 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7309 macro and introducing temporary variables works around the bug. */
7310
7311 case wordbound:
7312 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7313 if (AT_WORD_BOUNDARY (d))
7314 break;
7315 goto fail;
7316
7317 case notwordbound:
7318 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7319 if (AT_WORD_BOUNDARY (d))
7320 goto fail;
7321 break;
7322#else
7323 case wordbound:
7324 {
7325 boolean prevchar, thischar;
7326
7327 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7328 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7329 break;
7330
7331 prevchar = WORDCHAR_P (d - 1);
7332 thischar = WORDCHAR_P (d);
7333 if (prevchar != thischar)
7334 break;
7335 goto fail;
7336 }
7337
7338 case notwordbound:
7339 {
7340 boolean prevchar, thischar;
7341
7342 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7343 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7344 goto fail;
7345
7346 prevchar = WORDCHAR_P (d - 1);
7347 thischar = WORDCHAR_P (d);
7348 if (prevchar != thischar)
7349 goto fail;
7350 break;
7351 }
7352#endif
7353
7354 case wordbeg:
7355 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7356 if (!AT_STRINGS_END (d) && WORDCHAR_P (d)
7357 && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7358 break;
7359 goto fail;
7360
7361 case wordend:
7362 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7363 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7364 && (AT_STRINGS_END (d) || !WORDCHAR_P (d)))
7365 break;
7366 goto fail;
7367
7368#ifdef emacs
7369 case before_dot:
7370 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7371 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7372 goto fail;
7373 break;
7374
7375 case at_dot:
7376 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7377 if (PTR_CHAR_POS ((unsigned char *) d) != point)
7378 goto fail;
7379 break;
7380
7381 case after_dot:
7382 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7383 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7384 goto fail;
7385 break;
7386
7387 case syntaxspec:
7388 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7389 mcnt = *p++;
7390 goto matchsyntax;
7391
7392 case wordchar:
7393 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7394 mcnt = (int) Sword;
7395 matchsyntax:
7396 PREFETCH ();
7397 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7398 d++;
7399 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7400 goto fail;
7401 SET_REGS_MATCHED ();
7402 break;
7403
7404 case notsyntaxspec:
7405 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7406 mcnt = *p++;
7407 goto matchnotsyntax;
7408
7409 case notwordchar:
7410 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7411 mcnt = (int) Sword;
7412 matchnotsyntax:
7413 PREFETCH ();
7414 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7415 d++;
7416 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7417 goto fail;
7418 SET_REGS_MATCHED ();
7419 break;
7420
7421#else /* not emacs */
7422 case wordchar:
7423 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7424 PREFETCH ();
7425 if (!WORDCHAR_P (d))
7426 goto fail;
7427 SET_REGS_MATCHED ();
7428 d++;
7429 break;
7430
7431 case notwordchar:
7432 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7433 PREFETCH ();
7434 if (WORDCHAR_P (d))
7435 goto fail;
7436 SET_REGS_MATCHED ();
7437 d++;
7438 break;
7439#endif /* not emacs */
7440
7441 default:
7442 abort ();
7443 }
7444 continue; /* Successfully executed one pattern command; keep going. */
7445
7446
7447 /* We goto here if a matching operation fails. */
7448 fail:
7449 if (!FAIL_STACK_EMPTY ())
7450 { /* A restart point is known. Restore to that state. */
7451 DEBUG_PRINT1 ("\nFAIL:\n");
7452 POP_FAILURE_POINT (d, p,
7453 lowest_active_reg, highest_active_reg,
7454 regstart, regend, reg_info);
7455
7456 /* If this failure point is a dummy, try the next one. */
7457 if (!p)
7458 goto fail;
7459
7460 /* If we failed to the end of the pattern, don't examine *p. */
7461 assert (p <= pend);
7462 if (p < pend)
7463 {
7464 boolean is_a_jump_n = false;
7465
7466 /* If failed to a backwards jump that's part of a repetition
7467 loop, need to pop this failure point and use the next one. */
7468 switch ((re_opcode_t) *p)
7469 {
7470 case jump_n:
7471 is_a_jump_n = true;
7472 case maybe_pop_jump:
7473 case pop_failure_jump:
7474 case jump:
7475 p1 = p + 1;
7476 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7477 p1 += mcnt;
7478
7479 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7480 || (!is_a_jump_n
7481 && (re_opcode_t) *p1 == on_failure_jump))
7482 goto fail;
7483 break;
7484 default:
7485 /* do nothing */ ;
7486 }
7487 }
7488
7489 if (d >= string1 && d <= end1)
7490 dend = end_match_1;
7491 }
7492 else
7493 break; /* Matching at this starting point really fails. */
7494 } /* for (;;) */
7495
7496 if (best_regs_set)
7497 goto restore_best_regs;
7498
7499 FREE_VARIABLES ();
7500
7501 return -1; /* Failure to match. */
7502} /* re_match_2 */
7503\f
7504/* Subroutine definitions for re_match_2. */
7505
7506
7507/* We are passed P pointing to a register number after a start_memory.
7508
7509 Return true if the pattern up to the corresponding stop_memory can
7510 match the empty string, and false otherwise.
7511
7512 If we find the matching stop_memory, sets P to point to one past its number.
7513 Otherwise, sets P to an undefined byte less than or equal to END.
7514
7515 We don't handle duplicates properly (yet). */
7516
7517static boolean
7518PREFIX(group_match_null_string_p) (p, end, reg_info)
7519 UCHAR_T **p, *end;
7520 PREFIX(register_info_type) *reg_info;
7521{
7522 int mcnt;
7523 /* Point to after the args to the start_memory. */
7524 UCHAR_T *p1 = *p + 2;
7525
7526 while (p1 < end)
7527 {
7528 /* Skip over opcodes that can match nothing, and return true or
7529 false, as appropriate, when we get to one that can't, or to the
7530 matching stop_memory. */
7531
7532 switch ((re_opcode_t) *p1)
7533 {
7534 /* Could be either a loop or a series of alternatives. */
7535 case on_failure_jump:
7536 p1++;
7537 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7538
7539 /* If the next operation is not a jump backwards in the
7540 pattern. */
7541
7542 if (mcnt >= 0)
7543 {
7544 /* Go through the on_failure_jumps of the alternatives,
7545 seeing if any of the alternatives cannot match nothing.
7546 The last alternative starts with only a jump,
7547 whereas the rest start with on_failure_jump and end
7548 with a jump, e.g., here is the pattern for `a|b|c':
7549
7550 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7551 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7552 /exactn/1/c
7553
7554 So, we have to first go through the first (n-1)
7555 alternatives and then deal with the last one separately. */
7556
7557
7558 /* Deal with the first (n-1) alternatives, which start
7559 with an on_failure_jump (see above) that jumps to right
7560 past a jump_past_alt. */
7561
7562 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7563 jump_past_alt)
7564 {
7565 /* `mcnt' holds how many bytes long the alternative
7566 is, including the ending `jump_past_alt' and
7567 its number. */
7568
7569 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt -
7570 (1 + OFFSET_ADDRESS_SIZE),
7571 reg_info))
7572 return false;
7573
7574 /* Move to right after this alternative, including the
7575 jump_past_alt. */
7576 p1 += mcnt;
7577
7578 /* Break if it's the beginning of an n-th alternative
7579 that doesn't begin with an on_failure_jump. */
7580 if ((re_opcode_t) *p1 != on_failure_jump)
7581 break;
7582
7583 /* Still have to check that it's not an n-th
7584 alternative that starts with an on_failure_jump. */
7585 p1++;
7586 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7587 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7588 jump_past_alt)
7589 {
7590 /* Get to the beginning of the n-th alternative. */
7591 p1 -= 1 + OFFSET_ADDRESS_SIZE;
7592 break;
7593 }
7594 }
7595
7596 /* Deal with the last alternative: go back and get number
7597 of the `jump_past_alt' just before it. `mcnt' contains
7598 the length of the alternative. */
7599 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7600
7601 if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info))
7602 return false;
7603
7604 p1 += mcnt; /* Get past the n-th alternative. */
7605 } /* if mcnt > 0 */
7606 break;
7607
7608
7609 case stop_memory:
7610 assert (p1[1] == **p);
7611 *p = p1 + 2;
7612 return true;
7613
7614
7615 default:
7616 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7617 return false;
7618 }
7619 } /* while p1 < end */
7620
7621 return false;
7622} /* group_match_null_string_p */
7623
7624
7625/* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7626 It expects P to be the first byte of a single alternative and END one
7627 byte past the last. The alternative can contain groups. */
7628
7629static boolean
7630PREFIX(alt_match_null_string_p) (p, end, reg_info)
7631 UCHAR_T *p, *end;
7632 PREFIX(register_info_type) *reg_info;
7633{
7634 int mcnt;
7635 UCHAR_T *p1 = p;
7636
7637 while (p1 < end)
7638 {
7639 /* Skip over opcodes that can match nothing, and break when we get
7640 to one that can't. */
7641
7642 switch ((re_opcode_t) *p1)
7643 {
7644 /* It's a loop. */
7645 case on_failure_jump:
7646 p1++;
7647 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7648 p1 += mcnt;
7649 break;
7650
7651 default:
7652 if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info))
7653 return false;
7654 }
7655 } /* while p1 < end */
7656
7657 return true;
7658} /* alt_match_null_string_p */
7659
7660
7661/* Deals with the ops common to group_match_null_string_p and
7662 alt_match_null_string_p.
7663
7664 Sets P to one after the op and its arguments, if any. */
7665
7666static boolean
7667PREFIX(common_op_match_null_string_p) (p, end, reg_info)
7668 UCHAR_T **p, *end;
7669 PREFIX(register_info_type) *reg_info;
7670{
7671 int mcnt;
7672 boolean ret;
7673 int reg_no;
7674 UCHAR_T *p1 = *p;
7675
7676 switch ((re_opcode_t) *p1++)
7677 {
7678 case no_op:
7679 case begline:
7680 case endline:
7681 case begbuf:
7682 case endbuf:
7683 case wordbeg:
7684 case wordend:
7685 case wordbound:
7686 case notwordbound:
7687#ifdef emacs
7688 case before_dot:
7689 case at_dot:
7690 case after_dot:
7691#endif
7692 break;
7693
7694 case start_memory:
7695 reg_no = *p1;
7696 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7697 ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info);
7698
7699 /* Have to set this here in case we're checking a group which
7700 contains a group and a back reference to it. */
7701
7702 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7703 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7704
7705 if (!ret)
7706 return false;
7707 break;
7708
7709 /* If this is an optimized succeed_n for zero times, make the jump. */
7710 case jump:
7711 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7712 if (mcnt >= 0)
7713 p1 += mcnt;
7714 else
7715 return false;
7716 break;
7717
7718 case succeed_n:
7719 /* Get to the number of times to succeed. */
7720 p1 += OFFSET_ADDRESS_SIZE;
7721 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7722
7723 if (mcnt == 0)
7724 {
7725 p1 -= 2 * OFFSET_ADDRESS_SIZE;
7726 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7727 p1 += mcnt;
7728 }
7729 else
7730 return false;
7731 break;
7732
7733 case duplicate:
7734 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7735 return false;
7736 break;
7737
7738 case set_number_at:
7739 p1 += 2 * OFFSET_ADDRESS_SIZE;
7740
7741 default:
7742 /* All other opcodes mean we cannot match the empty string. */
7743 return false;
7744 }
7745
7746 *p = p1;
7747 return true;
7748} /* common_op_match_null_string_p */
7749
7750
7751/* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7752 bytes; nonzero otherwise. */
7753
7754static int
7755PREFIX(bcmp_translate) (s1, s2, len, translate)
7756 const CHAR_T *s1, *s2;
7757 register int len;
7758 RE_TRANSLATE_TYPE translate;
7759{
7760 register const UCHAR_T *p1 = (const UCHAR_T *) s1;
7761 register const UCHAR_T *p2 = (const UCHAR_T *) s2;
7762 while (len)
7763 {
7764#ifdef WCHAR
7765 if (((*p1<=0xff)?translate[*p1++]:*p1++)
7766 != ((*p2<=0xff)?translate[*p2++]:*p2++))
7767 return 1;
7768#else /* BYTE */
7769 if (translate[*p1++] != translate[*p2++]) return 1;
7770#endif /* WCHAR */
7771 len--;
7772 }
7773 return 0;
7774}
7775\f
7776
7777#else /* not INSIDE_RECURSION */
7778
7779/* Entry points for GNU code. */
7780
7781/* re_compile_pattern is the GNU regular expression compiler: it
7782 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7783 Returns 0 if the pattern was valid, otherwise an error string.
7784
7785 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7786 are set in BUFP on entry.
7787
7788 We call regex_compile to do the actual compilation. */
7789
7790const char *
7791re_compile_pattern (pattern, length, bufp)
7792 const char *pattern;
7793 size_t length;
7794 struct re_pattern_buffer *bufp;
7795{
7796 reg_errcode_t ret;
7797
7798 /* GNU code is written to assume at least RE_NREGS registers will be set
7799 (and at least one extra will be -1). */
7800 bufp->regs_allocated = REGS_UNALLOCATED;
7801
7802 /* And GNU code determines whether or not to get register information
7803 by passing null for the REGS argument to re_match, etc., not by
7804 setting no_sub. */
7805 bufp->no_sub = 0;
7806
7807 /* Match anchors at newline. */
7808 bufp->newline_anchor = 1;
7809
7810# ifdef MBS_SUPPORT
7811 if (MB_CUR_MAX != 1)
7812 ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp);
7813 else
7814# endif
7815 ret = byte_regex_compile (pattern, length, re_syntax_options, bufp);
7816
7817 if (!ret)
7818 return NULL;
7819 return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7820}
7821#ifdef _LIBC
7822weak_alias (__re_compile_pattern, re_compile_pattern)
7823#endif
7824\f
7825/* Entry points compatible with 4.2 BSD regex library. We don't define
7826 them unless specifically requested. */
7827
7828#if defined _REGEX_RE_COMP || defined _LIBC
7829
7830/* BSD has one and only one pattern buffer. */
7831static struct re_pattern_buffer re_comp_buf;
7832
7833char *
7834#ifdef _LIBC
7835/* Make these definitions weak in libc, so POSIX programs can redefine
7836 these names if they don't use our functions, and still use
7837 regcomp/regexec below without link errors. */
7838weak_function
7839#endif
7840re_comp (s)
7841 const char *s;
7842{
7843 reg_errcode_t ret;
7844
7845 if (!s)
7846 {
7847 if (!re_comp_buf.buffer)
7848 return gettext ("No previous regular expression");
7849 return 0;
7850 }
7851
7852 if (!re_comp_buf.buffer)
7853 {
7854 re_comp_buf.buffer = (unsigned char *) malloc (200);
7855 if (re_comp_buf.buffer == NULL)
7856 return (char *) gettext (re_error_msgid
7857 + re_error_msgid_idx[(int) REG_ESPACE]);
7858 re_comp_buf.allocated = 200;
7859
7860 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
7861 if (re_comp_buf.fastmap == NULL)
7862 return (char *) gettext (re_error_msgid
7863 + re_error_msgid_idx[(int) REG_ESPACE]);
7864 }
7865
7866 /* Since `re_exec' always passes NULL for the `regs' argument, we
7867 don't need to initialize the pattern buffer fields which affect it. */
7868
7869 /* Match anchors at newlines. */
7870 re_comp_buf.newline_anchor = 1;
7871
7872# ifdef MBS_SUPPORT
7873 if (MB_CUR_MAX != 1)
7874 ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7875 else
7876# endif
7877 ret = byte_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7878
7879 if (!ret)
7880 return NULL;
7881
7882 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
7883 return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7884}
7885
7886
7887int
7888#ifdef _LIBC
7889weak_function
7890#endif
7891re_exec (s)
7892 const char *s;
7893{
7894 const int len = strlen (s);
7895 return
7896 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
7897}
7898
7899#endif /* _REGEX_RE_COMP */
7900\f
7901/* POSIX.2 functions. Don't define these for Emacs. */
7902
7903#ifndef emacs
7904
7905/* regcomp takes a regular expression as a string and compiles it.
7906
7907 PREG is a regex_t *. We do not expect any fields to be initialized,
7908 since POSIX says we shouldn't. Thus, we set
7909
7910 `buffer' to the compiled pattern;
7911 `used' to the length of the compiled pattern;
7912 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
7913 REG_EXTENDED bit in CFLAGS is set; otherwise, to
7914 RE_SYNTAX_POSIX_BASIC;
7915 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
7916 `fastmap' to an allocated space for the fastmap;
7917 `fastmap_accurate' to zero;
7918 `re_nsub' to the number of subexpressions in PATTERN.
7919
7920 PATTERN is the address of the pattern string.
7921
7922 CFLAGS is a series of bits which affect compilation.
7923
7924 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
7925 use POSIX basic syntax.
7926
7927 If REG_NEWLINE is set, then . and [^...] don't match newline.
7928 Also, regexec will try a match beginning after every newline.
7929
7930 If REG_ICASE is set, then we considers upper- and lowercase
7931 versions of letters to be equivalent when matching.
7932
7933 If REG_NOSUB is set, then when PREG is passed to regexec, that
7934 routine will report only success or failure, and nothing about the
7935 registers.
7936
7937 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
7938 the return codes and their meanings.) */
7939
7940int
7941regcomp (preg, pattern, cflags)
7942 regex_t *preg;
7943 const char *pattern;
7944 int cflags;
7945{
7946 reg_errcode_t ret;
7947 reg_syntax_t syntax
7948 = (cflags & REG_EXTENDED) ?
7949 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
7950
7951 /* regex_compile will allocate the space for the compiled pattern. */
7952 preg->buffer = 0;
7953 preg->allocated = 0;
7954 preg->used = 0;
7955
7956 /* Try to allocate space for the fastmap. */
7957 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
7958
7959 if (cflags & REG_ICASE)
7960 {
7961 unsigned i;
7962
7963 preg->translate
7964 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
7965 * sizeof (*(RE_TRANSLATE_TYPE)0));
7966 if (preg->translate == NULL)
7967 return (int) REG_ESPACE;
7968
7969 /* Map uppercase characters to corresponding lowercase ones. */
7970 for (i = 0; i < CHAR_SET_SIZE; i++)
7971 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
7972 }
7973 else
7974 preg->translate = NULL;
7975
7976 /* If REG_NEWLINE is set, newlines are treated differently. */
7977 if (cflags & REG_NEWLINE)
7978 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
7979 syntax &= ~RE_DOT_NEWLINE;
7980 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
7981 /* It also changes the matching behavior. */
7982 preg->newline_anchor = 1;
7983 }
7984 else
7985 preg->newline_anchor = 0;
7986
7987 preg->no_sub = !!(cflags & REG_NOSUB);
7988
7989 /* POSIX says a null character in the pattern terminates it, so we
7990 can use strlen here in compiling the pattern. */
7991# ifdef MBS_SUPPORT
7992 if (MB_CUR_MAX != 1)
7993 ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg);
7994 else
7995# endif
7996 ret = byte_regex_compile (pattern, strlen (pattern), syntax, preg);
7997
7998 /* POSIX doesn't distinguish between an unmatched open-group and an
7999 unmatched close-group: both are REG_EPAREN. */
8000 if (ret == REG_ERPAREN) ret = REG_EPAREN;
8001
8002 if (ret == REG_NOERROR && preg->fastmap)
8003 {
8004 /* Compute the fastmap now, since regexec cannot modify the pattern
8005 buffer. */
8006 if (re_compile_fastmap (preg) == -2)
8007 {
8008 /* Some error occurred while computing the fastmap, just forget
8009 about it. */
8010 free (preg->fastmap);
8011 preg->fastmap = NULL;
8012 }
8013 }
8014
8015 return (int) ret;
8016}
8017#ifdef _LIBC
8018weak_alias (__regcomp, regcomp)
8019#endif
8020
8021
8022/* regexec searches for a given pattern, specified by PREG, in the
8023 string STRING.
8024
8025 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
8026 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
8027 least NMATCH elements, and we set them to the offsets of the
8028 corresponding matched substrings.
8029
8030 EFLAGS specifies `execution flags' which affect matching: if
8031 REG_NOTBOL is set, then ^ does not match at the beginning of the
8032 string; if REG_NOTEOL is set, then $ does not match at the end.
8033
8034 We return 0 if we find a match and REG_NOMATCH if not. */
8035
8036int
8037regexec (preg, string, nmatch, pmatch, eflags)
8038 const regex_t *preg;
8039 const char *string;
8040 size_t nmatch;
8041 regmatch_t pmatch[];
8042 int eflags;
8043{
8044 int ret;
8045 struct re_registers regs;
8046 regex_t private_preg;
8047 int len = strlen (string);
8048 boolean want_reg_info = !preg->no_sub && nmatch > 0;
8049
8050 private_preg = *preg;
8051
8052 private_preg.not_bol = !!(eflags & REG_NOTBOL);
8053 private_preg.not_eol = !!(eflags & REG_NOTEOL);
8054
8055 /* The user has told us exactly how many registers to return
8056 information about, via `nmatch'. We have to pass that on to the
8057 matching routines. */
8058 private_preg.regs_allocated = REGS_FIXED;
8059
8060 if (want_reg_info)
8061 {
8062 regs.num_regs = nmatch;
8063 regs.start = TALLOC (nmatch * 2, regoff_t);
8064 if (regs.start == NULL)
8065 return (int) REG_NOMATCH;
8066 regs.end = regs.start + nmatch;
8067 }
8068
8069 /* Perform the searching operation. */
8070 ret = re_search (&private_preg, string, len,
8071 /* start: */ 0, /* range: */ len,
8072 want_reg_info ? &regs : (struct re_registers *) 0);
8073
8074 /* Copy the register information to the POSIX structure. */
8075 if (want_reg_info)
8076 {
8077 if (ret >= 0)
8078 {
8079 unsigned r;
8080
8081 for (r = 0; r < nmatch; r++)
8082 {
8083 pmatch[r].rm_so = regs.start[r];
8084 pmatch[r].rm_eo = regs.end[r];
8085 }
8086 }
8087
8088 /* If we needed the temporary register info, free the space now. */
8089 free (regs.start);
8090 }
8091
8092 /* We want zero return to mean success, unlike `re_search'. */
8093 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
8094}
8095#ifdef _LIBC
8096weak_alias (__regexec, regexec)
8097#endif
8098
8099
8100/* Returns a message corresponding to an error code, ERRCODE, returned
8101 from either regcomp or regexec. We don't use PREG here. */
8102
8103size_t
8104regerror (errcode, preg, errbuf, errbuf_size)
8105 int errcode;
8106 const regex_t *preg;
8107 char *errbuf;
8108 size_t errbuf_size;
8109{
8110 const char *msg;
8111 size_t msg_size;
8112
8113 if (errcode < 0
8114 || errcode >= (int) (sizeof (re_error_msgid_idx)
8115 / sizeof (re_error_msgid_idx[0])))
8116 /* Only error codes returned by the rest of the code should be passed
8117 to this routine. If we are given anything else, or if other regex
8118 code generates an invalid error code, then the program has a bug.
8119 Dump core so we can fix it. */
8120 abort ();
8121
8122 msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
8123
8124 msg_size = strlen (msg) + 1; /* Includes the null. */
8125
8126 if (errbuf_size != 0)
8127 {
8128 if (msg_size > errbuf_size)
8129 {
8130#if defined HAVE_MEMPCPY || defined _LIBC
8131 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
8132#else
8133 memcpy (errbuf, msg, errbuf_size - 1);
8134 errbuf[errbuf_size - 1] = 0;
8135#endif
8136 }
8137 else
8138 memcpy (errbuf, msg, msg_size);
8139 }
8140
8141 return msg_size;
8142}
8143#ifdef _LIBC
8144weak_alias (__regerror, regerror)
8145#endif
8146
8147
8148/* Free dynamically allocated space used by PREG. */
8149
8150void
8151regfree (preg)
8152 regex_t *preg;
8153{
8154 if (preg->buffer != NULL)
8155 free (preg->buffer);
8156 preg->buffer = NULL;
8157
8158 preg->allocated = 0;
8159 preg->used = 0;
8160
8161 if (preg->fastmap != NULL)
8162 free (preg->fastmap);
8163 preg->fastmap = NULL;
8164 preg->fastmap_accurate = 0;
8165
8166 if (preg->translate != NULL)
8167 free (preg->translate);
8168 preg->translate = NULL;
8169}
8170#ifdef _LIBC
8171weak_alias (__regfree, regfree)
8172#endif
8173
8174#endif /* not emacs */
8175
8176#endif /* not INSIDE_RECURSION */
8177
8178\f
8179#undef STORE_NUMBER
8180#undef STORE_NUMBER_AND_INCR
8181#undef EXTRACT_NUMBER
8182#undef EXTRACT_NUMBER_AND_INCR
8183
8184#undef DEBUG_PRINT_COMPILED_PATTERN
8185#undef DEBUG_PRINT_DOUBLE_STRING
8186
8187#undef INIT_FAIL_STACK
8188#undef RESET_FAIL_STACK
8189#undef DOUBLE_FAIL_STACK
8190#undef PUSH_PATTERN_OP
8191#undef PUSH_FAILURE_POINTER
8192#undef PUSH_FAILURE_INT
8193#undef PUSH_FAILURE_ELT
8194#undef POP_FAILURE_POINTER
8195#undef POP_FAILURE_INT
8196#undef POP_FAILURE_ELT
8197#undef DEBUG_PUSH
8198#undef DEBUG_POP
8199#undef PUSH_FAILURE_POINT
8200#undef POP_FAILURE_POINT
8201
8202#undef REG_UNSET_VALUE
8203#undef REG_UNSET
8204
8205#undef PATFETCH
8206#undef PATFETCH_RAW
8207#undef PATUNFETCH
8208#undef TRANSLATE
8209
8210#undef INIT_BUF_SIZE
8211#undef GET_BUFFER_SPACE
8212#undef BUF_PUSH
8213#undef BUF_PUSH_2
8214#undef BUF_PUSH_3
8215#undef STORE_JUMP
8216#undef STORE_JUMP2
8217#undef INSERT_JUMP
8218#undef INSERT_JUMP2
8219#undef EXTEND_BUFFER
8220#undef GET_UNSIGNED_NUMBER
8221#undef FREE_STACK_RETURN
8222
8223# undef POINTER_TO_OFFSET
8224# undef MATCHING_IN_FRST_STRING
8225# undef PREFETCH
8226# undef AT_STRINGS_BEG
8227# undef AT_STRINGS_END
8228# undef WORDCHAR_P
8229# undef FREE_VAR
8230# undef FREE_VARIABLES
8231# undef NO_HIGHEST_ACTIVE_REG
8232# undef NO_LOWEST_ACTIVE_REG
8233
8234# undef CHAR_T
8235# undef UCHAR_T
8236# undef COMPILED_BUFFER_VAR
8237# undef OFFSET_ADDRESS_SIZE
8238# undef CHAR_CLASS_SIZE
8239# undef PREFIX
8240# undef ARG_PREFIX
8241# undef PUT_CHAR
8242# undef BYTE
8243# undef WCHAR
8244
8245# define DEFINED_ONCE
GDB Binutils - Deliverables
This page took 0.369042 seconds and 4 git commands to generate.