1 /* Extended regular expression matching and search library, version
2 0.12. (Implements POSIX draft P10003.2/D11.2, except for
3 internationalization features.)
5 Copyright (C) 1993, 1994, 1995, 1996, 1997 Free Software Foundation, Inc.
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
12 This program 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
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
30 /* Converts the pointer to the char to BEG-based offset from the start. */
31 #define PTR_TO_OFFSET(d) \
32 POS_AS_IN_BUFFER (MATCHING_IN_FIRST_STRING \
33 ? (d) - string1 : (d) - (string2 - size1))
34 #define POS_AS_IN_BUFFER(p) ((p) + 1)
40 /* We need this for `regex.h', and perhaps for the Emacs include files. */
41 #include <sys/types.h>
43 /* This is for other GNU distributions with internationalized messages. */
44 #if HAVE_LIBINTL_H || defined (_LIBC)
47 # define gettext(msgid) (msgid)
51 /* This define is so xgettext can find the internationalizable
53 #define gettext_noop(String) String
56 /* The `emacs' switch turns on certain matching commands
57 that make sense only in Emacs. */
63 /* Make syntax table lookup grant data in gl_state. */
64 #define SYNTAX_ENTRY_VIA_PROPERTY
70 #define malloc xmalloc
75 /* If we are not linking with Emacs proper,
76 we can't use the relocating allocator
77 even if config.h says that we can. */
80 #if defined (STDC_HEADERS) || defined (_LIBC)
87 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
88 If nothing else has been done, use the method below. */
89 #ifdef INHIBIT_STRING_HEADER
90 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
91 #if !defined (bzero) && !defined (bcopy)
92 #undef INHIBIT_STRING_HEADER
97 /* This is the normal way of making sure we have a bcopy and a bzero.
98 This is used in most programs--a few other programs avoid this
99 by defining INHIBIT_STRING_HEADER. */
100 #ifndef INHIBIT_STRING_HEADER
101 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
104 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
107 #define bcopy(s, d, n) memcpy ((d), (s), (n))
110 #define bzero(s, n) memset ((s), 0, (n))
117 /* Define the syntax stuff for \<, \>, etc. */
119 /* This must be nonzero for the wordchar and notwordchar pattern
120 commands in re_match_2. */
125 #ifdef SWITCH_ENUM_BUG
126 #define SWITCH_ENUM_CAST(x) ((int)(x))
128 #define SWITCH_ENUM_CAST(x) (x)
133 extern char *re_syntax_table;
135 #else /* not SYNTAX_TABLE */
137 /* How many characters in the character set. */
138 #define CHAR_SET_SIZE 256
140 static char re_syntax_table[CHAR_SET_SIZE];
151 bzero (re_syntax_table, sizeof re_syntax_table);
153 for (c = 'a'; c <= 'z'; c++)
154 re_syntax_table[c] = Sword;
156 for (c = 'A'; c <= 'Z'; c++)
157 re_syntax_table[c] = Sword;
159 for (c = '0'; c <= '9'; c++)
160 re_syntax_table[c] = Sword;
162 re_syntax_table['_'] = Sword;
167 #endif /* not SYNTAX_TABLE */
169 #define SYNTAX(c) re_syntax_table[c]
171 /* Dummy macro for non emacs environments. */
172 #define BASE_LEADING_CODE_P(c) (0)
173 #define WORD_BOUNDARY_P(c1, c2) (0)
174 #define CHAR_HEAD_P(p) (1)
175 #define SINGLE_BYTE_CHAR_P(c) (1)
176 #define SAME_CHARSET_P(c1, c2) (1)
177 #define MULTIBYTE_FORM_LENGTH(p, s) (1)
178 #define STRING_CHAR(p, s) (*(p))
179 #define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
180 #define GET_CHAR_AFTER_2(c, p, str1, end1, str2, end2) \
181 (c = ((p) == (end1) ? *(str2) : *(p)))
182 #define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
183 (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
184 #endif /* not emacs */
186 /* Get the interface, including the syntax bits. */
189 /* isalpha etc. are used for the character classes. */
192 /* Jim Meyering writes:
194 "... Some ctype macros are valid only for character codes that
195 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
196 using /bin/cc or gcc but without giving an ansi option). So, all
197 ctype uses should be through macros like ISPRINT... If
198 STDC_HEADERS is defined, then autoconf has verified that the ctype
199 macros don't need to be guarded with references to isascii. ...
200 Defining isascii to 1 should let any compiler worth its salt
201 eliminate the && through constant folding." */
203 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
206 #define ISASCII(c) isascii(c)
210 #define ISBLANK(c) (ISASCII (c) && isblank (c))
212 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
215 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
217 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
220 #define ISPRINT(c) (ISASCII (c) && isprint (c))
221 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
222 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
223 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
224 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
225 #define ISLOWER(c) (ISASCII (c) && islower (c))
226 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
227 #define ISSPACE(c) (ISASCII (c) && isspace (c))
228 #define ISUPPER(c) (ISASCII (c) && isupper (c))
229 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
232 #define NULL (void *)0
235 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
236 since ours (we hope) works properly with all combinations of
237 machines, compilers, `char' and `unsigned char' argument types.
238 (Per Bothner suggested the basic approach.) */
239 #undef SIGN_EXTEND_CHAR
241 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
242 #else /* not __STDC__ */
243 /* As in Harbison and Steele. */
244 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
247 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
248 use `alloca' instead of `malloc'. This is because using malloc in
249 re_search* or re_match* could cause memory leaks when C-g is used in
250 Emacs; also, malloc is slower and causes storage fragmentation. On
251 the other hand, malloc is more portable, and easier to debug.
253 Because we sometimes use alloca, some routines have to be macros,
254 not functions -- `alloca'-allocated space disappears at the end of the
255 function it is called in. */
259 #define REGEX_ALLOCATE malloc
260 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
261 #define REGEX_FREE free
263 #else /* not REGEX_MALLOC */
265 /* Emacs already defines alloca, sometimes. */
268 /* Make alloca work the best possible way. */
270 #define alloca __builtin_alloca
271 #else /* not __GNUC__ */
274 #else /* not __GNUC__ or HAVE_ALLOCA_H */
275 #if 0 /* It is a bad idea to declare alloca. We always cast the result. */
276 #ifndef _AIX /* Already did AIX, up at the top. */
278 #endif /* not _AIX */
280 #endif /* not HAVE_ALLOCA_H */
281 #endif /* not __GNUC__ */
283 #endif /* not alloca */
285 #define REGEX_ALLOCATE alloca
287 /* Assumes a `char *destination' variable. */
288 #define REGEX_REALLOCATE(source, osize, nsize) \
289 (destination = (char *) alloca (nsize), \
290 bcopy (source, destination, osize), \
293 /* No need to do anything to free, after alloca. */
294 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
296 #endif /* not REGEX_MALLOC */
298 /* Define how to allocate the failure stack. */
300 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
302 #define REGEX_ALLOCATE_STACK(size) \
303 r_alloc (&failure_stack_ptr, (size))
304 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
305 r_re_alloc (&failure_stack_ptr, (nsize))
306 #define REGEX_FREE_STACK(ptr) \
307 r_alloc_free (&failure_stack_ptr)
309 #else /* not using relocating allocator */
313 #define REGEX_ALLOCATE_STACK malloc
314 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
315 #define REGEX_FREE_STACK free
317 #else /* not REGEX_MALLOC */
319 #define REGEX_ALLOCATE_STACK alloca
321 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
322 REGEX_REALLOCATE (source, osize, nsize)
323 /* No need to explicitly free anything. */
324 #define REGEX_FREE_STACK(arg)
326 #endif /* not REGEX_MALLOC */
327 #endif /* not using relocating allocator */
330 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
331 `string1' or just past its end. This works if PTR is NULL, which is
333 #define FIRST_STRING_P(ptr) \
334 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
336 /* (Re)Allocate N items of type T using malloc, or fail. */
337 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
338 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
339 #define RETALLOC_IF(addr, n, t) \
340 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
341 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
343 #define BYTEWIDTH 8 /* In bits. */
345 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
349 #define MAX(a, b) ((a) > (b) ? (a) : (b))
350 #define MIN(a, b) ((a) < (b) ? (a) : (b))
352 typedef char boolean;
356 static int re_match_2_internal ();
358 /* These are the command codes that appear in compiled regular
359 expressions. Some opcodes are followed by argument bytes. A
360 command code can specify any interpretation whatsoever for its
361 arguments. Zero bytes may appear in the compiled regular expression. */
367 /* Succeed right away--no more backtracking. */
370 /* Followed by one byte giving n, then by n literal bytes. */
373 /* Matches any (more or less) character. */
376 /* Matches any one char belonging to specified set. First
377 following byte is number of bitmap bytes. Then come bytes
378 for a bitmap saying which chars are in. Bits in each byte
379 are ordered low-bit-first. A character is in the set if its
380 bit is 1. A character too large to have a bit in the map is
381 automatically not in the set. */
384 /* Same parameters as charset, but match any character that is
385 not one of those specified. */
388 /* Start remembering the text that is matched, for storing in a
389 register. Followed by one byte with the register number, in
390 the range 0 to one less than the pattern buffer's re_nsub
391 field. Then followed by one byte with the number of groups
392 inner to this one. (This last has to be part of the
393 start_memory only because we need it in the on_failure_jump
397 /* Stop remembering the text that is matched and store it in a
398 memory register. Followed by one byte with the register
399 number, in the range 0 to one less than `re_nsub' in the
400 pattern buffer, and one byte with the number of inner groups,
401 just like `start_memory'. (We need the number of inner
402 groups here because we don't have any easy way of finding the
403 corresponding start_memory when we're at a stop_memory.) */
406 /* Match a duplicate of something remembered. Followed by one
407 byte containing the register number. */
410 /* Fail unless at beginning of line. */
413 /* Fail unless at end of line. */
416 /* Succeeds if at beginning of buffer (if emacs) or at beginning
417 of string to be matched (if not). */
420 /* Analogously, for end of buffer/string. */
423 /* Followed by two byte relative address to which to jump. */
426 /* Same as jump, but marks the end of an alternative. */
429 /* Followed by two-byte relative address of place to resume at
430 in case of failure. */
433 /* Like on_failure_jump, but pushes a placeholder instead of the
434 current string position when executed. */
435 on_failure_keep_string_jump,
437 /* Throw away latest failure point and then jump to following
438 two-byte relative address. */
441 /* Change to pop_failure_jump if know won't have to backtrack to
442 match; otherwise change to jump. This is used to jump
443 back to the beginning of a repeat. If what follows this jump
444 clearly won't match what the repeat does, such that we can be
445 sure that there is no use backtracking out of repetitions
446 already matched, then we change it to a pop_failure_jump.
447 Followed by two-byte address. */
450 /* Jump to following two-byte address, and push a dummy failure
451 point. This failure point will be thrown away if an attempt
452 is made to use it for a failure. A `+' construct makes this
453 before the first repeat. Also used as an intermediary kind
454 of jump when compiling an alternative. */
457 /* Push a dummy failure point and continue. Used at the end of
461 /* Followed by two-byte relative address and two-byte number n.
462 After matching N times, jump to the address upon failure. */
465 /* Followed by two-byte relative address, and two-byte number n.
466 Jump to the address N times, then fail. */
469 /* Set the following two-byte relative address to the
470 subsequent two-byte number. The address *includes* the two
474 wordchar, /* Matches any word-constituent character. */
475 notwordchar, /* Matches any char that is not a word-constituent. */
477 wordbeg, /* Succeeds if at word beginning. */
478 wordend, /* Succeeds if at word end. */
480 wordbound, /* Succeeds if at a word boundary. */
481 notwordbound /* Succeeds if not at a word boundary. */
484 ,before_dot, /* Succeeds if before point. */
485 at_dot, /* Succeeds if at point. */
486 after_dot, /* Succeeds if after point. */
488 /* Matches any character whose syntax is specified. Followed by
489 a byte which contains a syntax code, e.g., Sword. */
492 /* Matches any character whose syntax is not that specified. */
495 /* Matches any character whose category-set contains the specified
496 category. The operator is followed by a byte which contains a
497 category code (mnemonic ASCII character). */
500 /* Matches any character whose category-set does not contain the
501 specified category. The operator is followed by a byte which
502 contains the category code (mnemonic ASCII character). */
507 /* Common operations on the compiled pattern. */
509 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
511 #define STORE_NUMBER(destination, number) \
513 (destination)[0] = (number) & 0377; \
514 (destination)[1] = (number) >> 8; \
517 /* Same as STORE_NUMBER, except increment DESTINATION to
518 the byte after where the number is stored. Therefore, DESTINATION
519 must be an lvalue. */
521 #define STORE_NUMBER_AND_INCR(destination, number) \
523 STORE_NUMBER (destination, number); \
524 (destination) += 2; \
527 /* Put into DESTINATION a number stored in two contiguous bytes starting
530 #define EXTRACT_NUMBER(destination, source) \
532 (destination) = *(source) & 0377; \
533 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
538 extract_number (dest, source)
540 unsigned char *source;
542 int temp = SIGN_EXTEND_CHAR (*(source + 1));
543 *dest = *source & 0377;
547 #ifndef EXTRACT_MACROS /* To debug the macros. */
548 #undef EXTRACT_NUMBER
549 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
550 #endif /* not EXTRACT_MACROS */
554 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
555 SOURCE must be an lvalue. */
557 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
559 EXTRACT_NUMBER (destination, source); \
565 extract_number_and_incr (destination, source)
567 unsigned char **source;
569 extract_number (destination, *source);
573 #ifndef EXTRACT_MACROS
574 #undef EXTRACT_NUMBER_AND_INCR
575 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
576 extract_number_and_incr (&dest, &src)
577 #endif /* not EXTRACT_MACROS */
581 /* Store a multibyte character in three contiguous bytes starting
582 DESTINATION, and increment DESTINATION to the byte after where the
583 character is stored. Therefore, DESTINATION must be an lvalue. */
585 #define STORE_CHARACTER_AND_INCR(destination, character) \
587 (destination)[0] = (character) & 0377; \
588 (destination)[1] = ((character) >> 8) & 0377; \
589 (destination)[2] = (character) >> 16; \
590 (destination) += 3; \
593 /* Put into DESTINATION a character stored in three contiguous bytes
594 starting at SOURCE. */
596 #define EXTRACT_CHARACTER(destination, source) \
598 (destination) = ((source)[0] \
599 | ((source)[1] << 8) \
600 | ((source)[2] << 16)); \
604 /* Macros for charset. */
606 /* Size of bitmap of charset P in bytes. P is a start of charset,
607 i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
608 #define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
610 /* Nonzero if charset P has range table. */
611 #define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
613 /* Return the address of range table of charset P. But not the start
614 of table itself, but the before where the number of ranges is
615 stored. `2 +' means to skip re_opcode_t and size of bitmap. */
616 #define CHARSET_RANGE_TABLE(p) (&(p)[2 + CHARSET_BITMAP_SIZE (p)])
618 /* Test if C is listed in the bitmap of charset P. */
619 #define CHARSET_LOOKUP_BITMAP(p, c) \
620 ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
621 && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
623 /* Return the address of end of RANGE_TABLE. COUNT is number of
624 ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
625 is start of range and end of range. `* 3' is size of each start
627 #define CHARSET_RANGE_TABLE_END(range_table, count) \
628 ((range_table) + (count) * 2 * 3)
630 /* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
631 COUNT is number of ranges in RANGE_TABLE. */
632 #define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
635 int range_start, range_end; \
637 unsigned char *range_table_end \
638 = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
640 for (p = (range_table); p < range_table_end; p += 2 * 3) \
642 EXTRACT_CHARACTER (range_start, p); \
643 EXTRACT_CHARACTER (range_end, p + 3); \
645 if (range_start <= (c) && (c) <= range_end) \
654 /* Test if C is in range table of CHARSET. The flag NOT is negated if
655 C is listed in it. */
656 #define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
659 /* Number of ranges in range table. */ \
661 unsigned char *range_table = CHARSET_RANGE_TABLE (charset); \
663 EXTRACT_NUMBER_AND_INCR (count, range_table); \
664 CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
668 /* If DEBUG is defined, Regex prints many voluminous messages about what
669 it is doing (if the variable `debug' is nonzero). If linked with the
670 main program in `iregex.c', you can enter patterns and strings
671 interactively. And if linked with the main program in `main.c' and
672 the other test files, you can run the already-written tests. */
676 /* We use standard I/O for debugging. */
679 /* It is useful to test things that ``must'' be true when debugging. */
682 static int debug = 0;
684 #define DEBUG_STATEMENT(e) e
685 #define DEBUG_PRINT1(x) if (debug) printf (x)
686 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
687 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
688 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
689 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
690 if (debug) print_partial_compiled_pattern (s, e)
691 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
692 if (debug) print_double_string (w, s1, sz1, s2, sz2)
695 /* Print the fastmap in human-readable form. */
698 print_fastmap (fastmap)
701 unsigned was_a_range = 0;
704 while (i < (1 << BYTEWIDTH))
710 while (i < (1 << BYTEWIDTH) && fastmap[i])
726 /* Print a compiled pattern string in human-readable form, starting at
727 the START pointer into it and ending just before the pointer END. */
730 print_partial_compiled_pattern (start, end)
731 unsigned char *start;
735 unsigned char *p = start;
736 unsigned char *pend = end;
744 /* Loop over pattern commands. */
747 printf ("%d:\t", p - start);
749 switch ((re_opcode_t) *p++)
757 printf ("/exactn/%d", mcnt);
768 printf ("/start_memory/%d/%d", mcnt, *p++);
773 printf ("/stop_memory/%d/%d", mcnt, *p++);
777 printf ("/duplicate/%d", *p++);
787 register int c, last = -100;
788 register int in_range = 0;
790 printf ("/charset [%s",
791 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
793 assert (p + *p < pend);
795 for (c = 0; c < 256; c++)
797 && (p[1 + (c/8)] & (1 << (c % 8))))
799 /* Are we starting a range? */
800 if (last + 1 == c && ! in_range)
805 /* Have we broken a range? */
806 else if (last + 1 != c && in_range)
835 case on_failure_jump:
836 extract_number_and_incr (&mcnt, &p);
837 printf ("/on_failure_jump to %d", p + mcnt - start);
840 case on_failure_keep_string_jump:
841 extract_number_and_incr (&mcnt, &p);
842 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
845 case dummy_failure_jump:
846 extract_number_and_incr (&mcnt, &p);
847 printf ("/dummy_failure_jump to %d", p + mcnt - start);
850 case push_dummy_failure:
851 printf ("/push_dummy_failure");
855 extract_number_and_incr (&mcnt, &p);
856 printf ("/maybe_pop_jump to %d", p + mcnt - start);
859 case pop_failure_jump:
860 extract_number_and_incr (&mcnt, &p);
861 printf ("/pop_failure_jump to %d", p + mcnt - start);
865 extract_number_and_incr (&mcnt, &p);
866 printf ("/jump_past_alt to %d", p + mcnt - start);
870 extract_number_and_incr (&mcnt, &p);
871 printf ("/jump to %d", p + mcnt - start);
875 extract_number_and_incr (&mcnt, &p);
876 extract_number_and_incr (&mcnt2, &p);
877 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
881 extract_number_and_incr (&mcnt, &p);
882 extract_number_and_incr (&mcnt2, &p);
883 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
887 extract_number_and_incr (&mcnt, &p);
888 extract_number_and_incr (&mcnt2, &p);
889 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
893 printf ("/wordbound");
897 printf ("/notwordbound");
909 printf ("/before_dot");
917 printf ("/after_dot");
921 printf ("/syntaxspec");
923 printf ("/%d", mcnt);
927 printf ("/notsyntaxspec");
929 printf ("/%d", mcnt);
934 printf ("/wordchar");
938 printf ("/notwordchar");
950 printf ("?%d", *(p-1));
956 printf ("%d:\tend of pattern.\n", p - start);
961 print_compiled_pattern (bufp)
962 struct re_pattern_buffer *bufp;
964 unsigned char *buffer = bufp->buffer;
966 print_partial_compiled_pattern (buffer, buffer + bufp->used);
967 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
969 if (bufp->fastmap_accurate && bufp->fastmap)
971 printf ("fastmap: ");
972 print_fastmap (bufp->fastmap);
975 printf ("re_nsub: %d\t", bufp->re_nsub);
976 printf ("regs_alloc: %d\t", bufp->regs_allocated);
977 printf ("can_be_null: %d\t", bufp->can_be_null);
978 printf ("newline_anchor: %d\n", bufp->newline_anchor);
979 printf ("no_sub: %d\t", bufp->no_sub);
980 printf ("not_bol: %d\t", bufp->not_bol);
981 printf ("not_eol: %d\t", bufp->not_eol);
982 printf ("syntax: %d\n", bufp->syntax);
983 /* Perhaps we should print the translate table? */
988 print_double_string (where, string1, size1, string2, size2)
1001 if (FIRST_STRING_P (where))
1003 for (this_char = where - string1; this_char < size1; this_char++)
1004 putchar (string1[this_char]);
1009 for (this_char = where - string2; this_char < size2; this_char++)
1010 putchar (string2[this_char]);
1014 #else /* not DEBUG */
1019 #define DEBUG_STATEMENT(e)
1020 #define DEBUG_PRINT1(x)
1021 #define DEBUG_PRINT2(x1, x2)
1022 #define DEBUG_PRINT3(x1, x2, x3)
1023 #define DEBUG_PRINT4(x1, x2, x3, x4)
1024 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1025 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1027 #endif /* not DEBUG */
1029 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1030 also be assigned to arbitrarily: each pattern buffer stores its own
1031 syntax, so it can be changed between regex compilations. */
1032 /* This has no initializer because initialized variables in Emacs
1033 become read-only after dumping. */
1034 reg_syntax_t re_syntax_options;
1037 /* Specify the precise syntax of regexps for compilation. This provides
1038 for compatibility for various utilities which historically have
1039 different, incompatible syntaxes.
1041 The argument SYNTAX is a bit mask comprised of the various bits
1042 defined in regex.h. We return the old syntax. */
1045 re_set_syntax (syntax)
1046 reg_syntax_t syntax;
1048 reg_syntax_t ret = re_syntax_options;
1050 re_syntax_options = syntax;
1054 /* This table gives an error message for each of the error codes listed
1055 in regex.h. Obviously the order here has to be same as there.
1056 POSIX doesn't require that we do anything for REG_NOERROR,
1057 but why not be nice? */
1059 static const char *re_error_msgid[] =
1061 gettext_noop ("Success"), /* REG_NOERROR */
1062 gettext_noop ("No match"), /* REG_NOMATCH */
1063 gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
1064 gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
1065 gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
1066 gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
1067 gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
1068 gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
1069 gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
1070 gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
1071 gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
1072 gettext_noop ("Invalid range end"), /* REG_ERANGE */
1073 gettext_noop ("Memory exhausted"), /* REG_ESPACE */
1074 gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
1075 gettext_noop ("Premature end of regular expression"), /* REG_EEND */
1076 gettext_noop ("Regular expression too big"), /* REG_ESIZE */
1077 gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
1080 /* Avoiding alloca during matching, to placate r_alloc. */
1082 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1083 searching and matching functions should not call alloca. On some
1084 systems, alloca is implemented in terms of malloc, and if we're
1085 using the relocating allocator routines, then malloc could cause a
1086 relocation, which might (if the strings being searched are in the
1087 ralloc heap) shift the data out from underneath the regexp
1090 Here's another reason to avoid allocation: Emacs
1091 processes input from X in a signal handler; processing X input may
1092 call malloc; if input arrives while a matching routine is calling
1093 malloc, then we're scrod. But Emacs can't just block input while
1094 calling matching routines; then we don't notice interrupts when
1095 they come in. So, Emacs blocks input around all regexp calls
1096 except the matching calls, which it leaves unprotected, in the
1097 faith that they will not malloc. */
1099 /* Normally, this is fine. */
1100 #define MATCH_MAY_ALLOCATE
1102 /* When using GNU C, we are not REALLY using the C alloca, no matter
1103 what config.h may say. So don't take precautions for it. */
1108 /* The match routines may not allocate if (1) they would do it with malloc
1109 and (2) it's not safe for them to use malloc.
1110 Note that if REL_ALLOC is defined, matching would not use malloc for the
1111 failure stack, but we would still use it for the register vectors;
1112 so REL_ALLOC should not affect this. */
1113 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
1114 #undef MATCH_MAY_ALLOCATE
1118 /* Failure stack declarations and macros; both re_compile_fastmap and
1119 re_match_2 use a failure stack. These have to be macros because of
1120 REGEX_ALLOCATE_STACK. */
1123 /* Approximate number of failure points for which to initially allocate space
1124 when matching. If this number is exceeded, we allocate more
1125 space, so it is not a hard limit. */
1126 #ifndef INIT_FAILURE_ALLOC
1127 #define INIT_FAILURE_ALLOC 20
1130 /* Roughly the maximum number of failure points on the stack. Would be
1131 exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
1132 This is a variable only so users of regex can assign to it; we never
1133 change it ourselves. */
1134 #if defined (MATCH_MAY_ALLOCATE)
1135 /* Note that 4400 is enough to cause a crash on Alpha OSF/1,
1136 whose default stack limit is 2mb. In order for a larger
1137 value to work reliably, you have to try to make it accord
1138 with the process stack limit. */
1139 int re_max_failures = 40000;
1141 int re_max_failures = 4000;
1144 union fail_stack_elt
1146 unsigned char *pointer;
1150 typedef union fail_stack_elt fail_stack_elt_t;
1154 fail_stack_elt_t *stack;
1156 unsigned avail; /* Offset of next open position. */
1159 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1160 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1161 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1164 /* Define macros to initialize and free the failure stack.
1165 Do `return -2' if the alloc fails. */
1167 #ifdef MATCH_MAY_ALLOCATE
1168 #define INIT_FAIL_STACK() \
1170 fail_stack.stack = (fail_stack_elt_t *) \
1171 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE \
1172 * sizeof (fail_stack_elt_t)); \
1174 if (fail_stack.stack == NULL) \
1177 fail_stack.size = INIT_FAILURE_ALLOC; \
1178 fail_stack.avail = 0; \
1181 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1183 #define INIT_FAIL_STACK() \
1185 fail_stack.avail = 0; \
1188 #define RESET_FAIL_STACK()
1192 /* Double the size of FAIL_STACK, up to a limit
1193 which allows approximately `re_max_failures' items.
1195 Return 1 if succeeds, and 0 if either ran out of memory
1196 allocating space for it or it was already too large.
1198 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1200 /* Factor to increase the failure stack size by
1201 when we increase it.
1202 This used to be 2, but 2 was too wasteful
1203 because the old discarded stacks added up to as much space
1204 were as ultimate, maximum-size stack. */
1205 #define FAIL_STACK_GROWTH_FACTOR 4
1207 #define GROW_FAIL_STACK(fail_stack) \
1208 ((fail_stack).size >= re_max_failures * TYPICAL_FAILURE_SIZE \
1210 : ((fail_stack).stack \
1211 = (fail_stack_elt_t *) \
1212 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1213 (fail_stack).size * sizeof (fail_stack_elt_t), \
1214 MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1215 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1216 * FAIL_STACK_GROWTH_FACTOR))), \
1218 (fail_stack).stack == NULL \
1220 : ((fail_stack).size \
1221 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE, \
1222 ((fail_stack).size * sizeof (fail_stack_elt_t) \
1223 * FAIL_STACK_GROWTH_FACTOR)) \
1224 / sizeof (fail_stack_elt_t)), \
1228 /* Push pointer POINTER on FAIL_STACK.
1229 Return 1 if was able to do so and 0 if ran out of memory allocating
1231 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1232 ((FAIL_STACK_FULL () \
1233 && !GROW_FAIL_STACK (FAIL_STACK)) \
1235 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1238 /* Push a pointer value onto the failure stack.
1239 Assumes the variable `fail_stack'. Probably should only
1240 be called from within `PUSH_FAILURE_POINT'. */
1241 #define PUSH_FAILURE_POINTER(item) \
1242 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1244 /* This pushes an integer-valued item onto the failure stack.
1245 Assumes the variable `fail_stack'. Probably should only
1246 be called from within `PUSH_FAILURE_POINT'. */
1247 #define PUSH_FAILURE_INT(item) \
1248 fail_stack.stack[fail_stack.avail++].integer = (item)
1250 /* Push a fail_stack_elt_t value onto the failure stack.
1251 Assumes the variable `fail_stack'. Probably should only
1252 be called from within `PUSH_FAILURE_POINT'. */
1253 #define PUSH_FAILURE_ELT(item) \
1254 fail_stack.stack[fail_stack.avail++] = (item)
1256 /* These three POP... operations complement the three PUSH... operations.
1257 All assume that `fail_stack' is nonempty. */
1258 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1259 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1260 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1262 /* Used to omit pushing failure point id's when we're not debugging. */
1264 #define DEBUG_PUSH PUSH_FAILURE_INT
1265 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1267 #define DEBUG_PUSH(item)
1268 #define DEBUG_POP(item_addr)
1272 /* Push the information about the state we will need
1273 if we ever fail back to it.
1275 Requires variables fail_stack, regstart, regend, reg_info, and
1276 num_regs be declared. GROW_FAIL_STACK requires `destination' be
1279 Does `return FAILURE_CODE' if runs out of memory. */
1281 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1283 char *destination; \
1284 /* Must be int, so when we don't save any registers, the arithmetic \
1285 of 0 + -1 isn't done as unsigned. */ \
1288 DEBUG_STATEMENT (failure_id++); \
1289 DEBUG_STATEMENT (nfailure_points_pushed++); \
1290 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1291 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1292 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1294 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1295 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1297 /* Ensure we have enough space allocated for what we will push. */ \
1298 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1300 if (!GROW_FAIL_STACK (fail_stack)) \
1301 return failure_code; \
1303 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1304 (fail_stack).size); \
1305 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1308 /* Push the info, starting with the registers. */ \
1309 DEBUG_PRINT1 ("\n"); \
1312 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1315 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1316 DEBUG_STATEMENT (num_regs_pushed++); \
1318 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1319 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1321 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1322 PUSH_FAILURE_POINTER (regend[this_reg]); \
1324 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1325 DEBUG_PRINT2 (" match_null=%d", \
1326 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1327 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1328 DEBUG_PRINT2 (" matched_something=%d", \
1329 MATCHED_SOMETHING (reg_info[this_reg])); \
1330 DEBUG_PRINT2 (" ever_matched=%d", \
1331 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1332 DEBUG_PRINT1 ("\n"); \
1333 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1336 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1337 PUSH_FAILURE_INT (lowest_active_reg); \
1339 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1340 PUSH_FAILURE_INT (highest_active_reg); \
1342 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1343 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1344 PUSH_FAILURE_POINTER (pattern_place); \
1346 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1347 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1349 DEBUG_PRINT1 ("'\n"); \
1350 PUSH_FAILURE_POINTER (string_place); \
1352 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1353 DEBUG_PUSH (failure_id); \
1356 /* This is the number of items that are pushed and popped on the stack
1357 for each register. */
1358 #define NUM_REG_ITEMS 3
1360 /* Individual items aside from the registers. */
1362 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1364 #define NUM_NONREG_ITEMS 4
1367 /* Estimate the size of data pushed by a typical failure stack entry.
1368 An estimate is all we need, because all we use this for
1369 is to choose a limit for how big to make the failure stack. */
1371 #define TYPICAL_FAILURE_SIZE 20
1373 /* This is how many items we actually use for a failure point.
1374 It depends on the regexp. */
1375 #define NUM_FAILURE_ITEMS \
1377 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1381 /* How many items can still be added to the stack without overflowing it. */
1382 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1385 /* Pops what PUSH_FAIL_STACK pushes.
1387 We restore into the parameters, all of which should be lvalues:
1388 STR -- the saved data position.
1389 PAT -- the saved pattern position.
1390 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1391 REGSTART, REGEND -- arrays of string positions.
1392 REG_INFO -- array of information about each subexpression.
1394 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1395 `pend', `string1', `size1', `string2', and `size2'. */
1397 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1399 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1401 const unsigned char *string_temp; \
1403 assert (!FAIL_STACK_EMPTY ()); \
1405 /* Remove failure points and point to how many regs pushed. */ \
1406 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1407 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1408 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1410 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1412 DEBUG_POP (&failure_id); \
1413 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1415 /* If the saved string location is NULL, it came from an \
1416 on_failure_keep_string_jump opcode, and we want to throw away the \
1417 saved NULL, thus retaining our current position in the string. */ \
1418 string_temp = POP_FAILURE_POINTER (); \
1419 if (string_temp != NULL) \
1420 str = (const char *) string_temp; \
1422 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1423 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1424 DEBUG_PRINT1 ("'\n"); \
1426 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1427 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1428 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1430 /* Restore register info. */ \
1431 high_reg = (unsigned) POP_FAILURE_INT (); \
1432 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1434 low_reg = (unsigned) POP_FAILURE_INT (); \
1435 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1438 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1440 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1442 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1443 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1445 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1446 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1448 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1449 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1453 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1455 reg_info[this_reg].word.integer = 0; \
1456 regend[this_reg] = 0; \
1457 regstart[this_reg] = 0; \
1459 highest_active_reg = high_reg; \
1462 set_regs_matched_done = 0; \
1463 DEBUG_STATEMENT (nfailure_points_popped++); \
1464 } /* POP_FAILURE_POINT */
1468 /* Structure for per-register (a.k.a. per-group) information.
1469 Other register information, such as the
1470 starting and ending positions (which are addresses), and the list of
1471 inner groups (which is a bits list) are maintained in separate
1474 We are making a (strictly speaking) nonportable assumption here: that
1475 the compiler will pack our bit fields into something that fits into
1476 the type of `word', i.e., is something that fits into one item on the
1481 fail_stack_elt_t word;
1484 /* This field is one if this group can match the empty string,
1485 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1486 #define MATCH_NULL_UNSET_VALUE 3
1487 unsigned match_null_string_p : 2;
1488 unsigned is_active : 1;
1489 unsigned matched_something : 1;
1490 unsigned ever_matched_something : 1;
1492 } register_info_type;
1494 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1495 #define IS_ACTIVE(R) ((R).bits.is_active)
1496 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1497 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1500 /* Call this when have matched a real character; it sets `matched' flags
1501 for the subexpressions which we are currently inside. Also records
1502 that those subexprs have matched. */
1503 #define SET_REGS_MATCHED() \
1506 if (!set_regs_matched_done) \
1509 set_regs_matched_done = 1; \
1510 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1512 MATCHED_SOMETHING (reg_info[r]) \
1513 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1520 /* Registers are set to a sentinel when they haven't yet matched. */
1521 static char reg_unset_dummy;
1522 #define REG_UNSET_VALUE (®_unset_dummy)
1523 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1525 /* Subroutine declarations and macros for regex_compile. */
1527 static void store_op1 (), store_op2 ();
1528 static void insert_op1 (), insert_op2 ();
1529 static boolean at_begline_loc_p (), at_endline_loc_p ();
1530 static boolean group_in_compile_stack ();
1531 static reg_errcode_t compile_range ();
1533 /* Fetch the next character in the uncompiled pattern---translating it
1534 if necessary. Also cast from a signed character in the constant
1535 string passed to us by the user to an unsigned char that we can use
1536 as an array index (in, e.g., `translate'). */
1538 #define PATFETCH(c) \
1539 do {if (p == pend) return REG_EEND; \
1540 c = (unsigned char) *p++; \
1541 if (translate) c = (unsigned char) translate[c]; \
1545 /* Fetch the next character in the uncompiled pattern, with no
1547 #define PATFETCH_RAW(c) \
1548 do {if (p == pend) return REG_EEND; \
1549 c = (unsigned char) *p++; \
1552 /* Go backwards one character in the pattern. */
1553 #define PATUNFETCH p--
1556 /* If `translate' is non-null, return translate[D], else just D. We
1557 cast the subscript to translate because some data is declared as
1558 `char *', to avoid warnings when a string constant is passed. But
1559 when we use a character as a subscript we must make it unsigned. */
1561 #define TRANSLATE(d) \
1562 (translate ? (unsigned char) RE_TRANSLATE (translate, (unsigned char) (d)) : (d))
1566 /* Macros for outputting the compiled pattern into `buffer'. */
1568 /* If the buffer isn't allocated when it comes in, use this. */
1569 #define INIT_BUF_SIZE 32
1571 /* Make sure we have at least N more bytes of space in buffer. */
1572 #define GET_BUFFER_SPACE(n) \
1573 while (b - bufp->buffer + (n) > bufp->allocated) \
1576 /* Make sure we have one more byte of buffer space and then add C to it. */
1577 #define BUF_PUSH(c) \
1579 GET_BUFFER_SPACE (1); \
1580 *b++ = (unsigned char) (c); \
1584 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1585 #define BUF_PUSH_2(c1, c2) \
1587 GET_BUFFER_SPACE (2); \
1588 *b++ = (unsigned char) (c1); \
1589 *b++ = (unsigned char) (c2); \
1593 /* As with BUF_PUSH_2, except for three bytes. */
1594 #define BUF_PUSH_3(c1, c2, c3) \
1596 GET_BUFFER_SPACE (3); \
1597 *b++ = (unsigned char) (c1); \
1598 *b++ = (unsigned char) (c2); \
1599 *b++ = (unsigned char) (c3); \
1603 /* Store a jump with opcode OP at LOC to location TO. We store a
1604 relative address offset by the three bytes the jump itself occupies. */
1605 #define STORE_JUMP(op, loc, to) \
1606 store_op1 (op, loc, (to) - (loc) - 3)
1608 /* Likewise, for a two-argument jump. */
1609 #define STORE_JUMP2(op, loc, to, arg) \
1610 store_op2 (op, loc, (to) - (loc) - 3, arg)
1612 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1613 #define INSERT_JUMP(op, loc, to) \
1614 insert_op1 (op, loc, (to) - (loc) - 3, b)
1616 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1617 #define INSERT_JUMP2(op, loc, to, arg) \
1618 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1621 /* This is not an arbitrary limit: the arguments which represent offsets
1622 into the pattern are two bytes long. So if 2^16 bytes turns out to
1623 be too small, many things would have to change. */
1624 #define MAX_BUF_SIZE (1L << 16)
1627 /* Extend the buffer by twice its current size via realloc and
1628 reset the pointers that pointed into the old block to point to the
1629 correct places in the new one. If extending the buffer results in it
1630 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1631 #define EXTEND_BUFFER() \
1633 unsigned char *old_buffer = bufp->buffer; \
1634 if (bufp->allocated == MAX_BUF_SIZE) \
1636 bufp->allocated <<= 1; \
1637 if (bufp->allocated > MAX_BUF_SIZE) \
1638 bufp->allocated = MAX_BUF_SIZE; \
1639 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1640 if (bufp->buffer == NULL) \
1641 return REG_ESPACE; \
1642 /* If the buffer moved, move all the pointers into it. */ \
1643 if (old_buffer != bufp->buffer) \
1645 b = (b - old_buffer) + bufp->buffer; \
1646 begalt = (begalt - old_buffer) + bufp->buffer; \
1647 if (fixup_alt_jump) \
1648 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1650 laststart = (laststart - old_buffer) + bufp->buffer; \
1651 if (pending_exact) \
1652 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1657 /* Since we have one byte reserved for the register number argument to
1658 {start,stop}_memory, the maximum number of groups we can report
1659 things about is what fits in that byte. */
1660 #define MAX_REGNUM 255
1662 /* But patterns can have more than `MAX_REGNUM' registers. We just
1663 ignore the excess. */
1664 typedef unsigned regnum_t;
1667 /* Macros for the compile stack. */
1669 /* Since offsets can go either forwards or backwards, this type needs to
1670 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1671 typedef int pattern_offset_t;
1675 pattern_offset_t begalt_offset;
1676 pattern_offset_t fixup_alt_jump;
1677 pattern_offset_t inner_group_offset;
1678 pattern_offset_t laststart_offset;
1680 } compile_stack_elt_t;
1685 compile_stack_elt_t *stack;
1687 unsigned avail; /* Offset of next open position. */
1688 } compile_stack_type;
1691 #define INIT_COMPILE_STACK_SIZE 32
1693 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1694 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1696 /* The next available element. */
1697 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1700 /* Structure to manage work area for range table. */
1701 struct range_table_work_area
1703 int *table; /* actual work area. */
1704 int allocated; /* allocated size for work area in bytes. */
1705 int used; /* actually used size in words. */
1708 /* Make sure that WORK_AREA can hold more N multibyte characters. */
1709 #define EXTEND_RANGE_TABLE_WORK_AREA(work_area, n) \
1711 if (((work_area).used + (n)) * sizeof (int) > (work_area).allocated) \
1713 (work_area).allocated += 16 * sizeof (int); \
1714 if ((work_area).table) \
1716 = (int *) realloc ((work_area).table, (work_area).allocated); \
1719 = (int *) malloc ((work_area).allocated); \
1720 if ((work_area).table == 0) \
1721 FREE_STACK_RETURN (REG_ESPACE); \
1725 /* Set a range (RANGE_START, RANGE_END) to WORK_AREA. */
1726 #define SET_RANGE_TABLE_WORK_AREA(work_area, range_start, range_end) \
1728 EXTEND_RANGE_TABLE_WORK_AREA ((work_area), 2); \
1729 (work_area).table[(work_area).used++] = (range_start); \
1730 (work_area).table[(work_area).used++] = (range_end); \
1733 /* Free allocated memory for WORK_AREA. */
1734 #define FREE_RANGE_TABLE_WORK_AREA(work_area) \
1736 if ((work_area).table) \
1737 free ((work_area).table); \
1740 #define CLEAR_RANGE_TABLE_WORK_USED(work_area) ((work_area).used = 0)
1741 #define RANGE_TABLE_WORK_USED(work_area) ((work_area).used)
1742 #define RANGE_TABLE_WORK_ELT(work_area, i) ((work_area).table[i])
1745 /* Set the bit for character C in a list. */
1746 #define SET_LIST_BIT(c) \
1747 (b[((unsigned char) (c)) / BYTEWIDTH] \
1748 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1751 /* Get the next unsigned number in the uncompiled pattern. */
1752 #define GET_UNSIGNED_NUMBER(num) \
1756 while (ISDIGIT (c)) \
1760 num = num * 10 + c - '0'; \
1768 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1770 #define IS_CHAR_CLASS(string) \
1771 (STREQ (string, "alpha") || STREQ (string, "upper") \
1772 || STREQ (string, "lower") || STREQ (string, "digit") \
1773 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1774 || STREQ (string, "space") || STREQ (string, "print") \
1775 || STREQ (string, "punct") || STREQ (string, "graph") \
1776 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1778 #ifndef MATCH_MAY_ALLOCATE
1780 /* If we cannot allocate large objects within re_match_2_internal,
1781 we make the fail stack and register vectors global.
1782 The fail stack, we grow to the maximum size when a regexp
1784 The register vectors, we adjust in size each time we
1785 compile a regexp, according to the number of registers it needs. */
1787 static fail_stack_type fail_stack;
1789 /* Size with which the following vectors are currently allocated.
1790 That is so we can make them bigger as needed,
1791 but never make them smaller. */
1792 static int regs_allocated_size;
1794 static const char ** regstart, ** regend;
1795 static const char ** old_regstart, ** old_regend;
1796 static const char **best_regstart, **best_regend;
1797 static register_info_type *reg_info;
1798 static const char **reg_dummy;
1799 static register_info_type *reg_info_dummy;
1801 /* Make the register vectors big enough for NUM_REGS registers,
1802 but don't make them smaller. */
1805 regex_grow_registers (num_regs)
1808 if (num_regs > regs_allocated_size)
1810 RETALLOC_IF (regstart, num_regs, const char *);
1811 RETALLOC_IF (regend, num_regs, const char *);
1812 RETALLOC_IF (old_regstart, num_regs, const char *);
1813 RETALLOC_IF (old_regend, num_regs, const char *);
1814 RETALLOC_IF (best_regstart, num_regs, const char *);
1815 RETALLOC_IF (best_regend, num_regs, const char *);
1816 RETALLOC_IF (reg_info, num_regs, register_info_type);
1817 RETALLOC_IF (reg_dummy, num_regs, const char *);
1818 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1820 regs_allocated_size = num_regs;
1824 #endif /* not MATCH_MAY_ALLOCATE */
1826 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1827 Returns one of error codes defined in `regex.h', or zero for success.
1829 Assumes the `allocated' (and perhaps `buffer') and `translate'
1830 fields are set in BUFP on entry.
1832 If it succeeds, results are put in BUFP (if it returns an error, the
1833 contents of BUFP are undefined):
1834 `buffer' is the compiled pattern;
1835 `syntax' is set to SYNTAX;
1836 `used' is set to the length of the compiled pattern;
1837 `fastmap_accurate' is zero;
1838 `re_nsub' is the number of subexpressions in PATTERN;
1839 `not_bol' and `not_eol' are zero;
1841 The `fastmap' and `newline_anchor' fields are neither
1842 examined nor set. */
1844 /* Return, freeing storage we allocated. */
1845 #define FREE_STACK_RETURN(value) \
1847 FREE_RANGE_TABLE_WORK_AREA (range_table_work); \
1848 free (compile_stack.stack); \
1852 static reg_errcode_t
1853 regex_compile (pattern, size, syntax, bufp)
1854 const char *pattern;
1856 reg_syntax_t syntax;
1857 struct re_pattern_buffer *bufp;
1859 /* We fetch characters from PATTERN here. Even though PATTERN is
1860 `char *' (i.e., signed), we declare these variables as unsigned, so
1861 they can be reliably used as array indices. */
1862 register unsigned int c, c1;
1864 /* A random temporary spot in PATTERN. */
1867 /* Points to the end of the buffer, where we should append. */
1868 register unsigned char *b;
1870 /* Keeps track of unclosed groups. */
1871 compile_stack_type compile_stack;
1873 /* Points to the current (ending) position in the pattern. */
1874 const char *p = pattern;
1875 const char *pend = pattern + size;
1877 /* How to translate the characters in the pattern. */
1878 RE_TRANSLATE_TYPE translate = bufp->translate;
1880 /* Address of the count-byte of the most recently inserted `exactn'
1881 command. This makes it possible to tell if a new exact-match
1882 character can be added to that command or if the character requires
1883 a new `exactn' command. */
1884 unsigned char *pending_exact = 0;
1886 /* Address of start of the most recently finished expression.
1887 This tells, e.g., postfix * where to find the start of its
1888 operand. Reset at the beginning of groups and alternatives. */
1889 unsigned char *laststart = 0;
1891 /* Address of beginning of regexp, or inside of last group. */
1892 unsigned char *begalt;
1894 /* Place in the uncompiled pattern (i.e., the {) to
1895 which to go back if the interval is invalid. */
1896 const char *beg_interval;
1898 /* Address of the place where a forward jump should go to the end of
1899 the containing expression. Each alternative of an `or' -- except the
1900 last -- ends with a forward jump of this sort. */
1901 unsigned char *fixup_alt_jump = 0;
1903 /* Counts open-groups as they are encountered. Remembered for the
1904 matching close-group on the compile stack, so the same register
1905 number is put in the stop_memory as the start_memory. */
1906 regnum_t regnum = 0;
1908 /* Work area for range table of charset. */
1909 struct range_table_work_area range_table_work;
1912 DEBUG_PRINT1 ("\nCompiling pattern: ");
1915 unsigned debug_count;
1917 for (debug_count = 0; debug_count < size; debug_count++)
1918 putchar (pattern[debug_count]);
1923 /* Initialize the compile stack. */
1924 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1925 if (compile_stack.stack == NULL)
1928 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1929 compile_stack.avail = 0;
1931 range_table_work.table = 0;
1932 range_table_work.allocated = 0;
1934 /* Initialize the pattern buffer. */
1935 bufp->syntax = syntax;
1936 bufp->fastmap_accurate = 0;
1937 bufp->not_bol = bufp->not_eol = 0;
1939 /* Set `used' to zero, so that if we return an error, the pattern
1940 printer (for debugging) will think there's no pattern. We reset it
1944 /* Always count groups, whether or not bufp->no_sub is set. */
1948 /* bufp->multibyte is set before regex_compile is called, so don't alter
1950 #else /* not emacs */
1951 /* Nothing is recognized as a multibyte character. */
1952 bufp->multibyte = 0;
1955 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1956 /* Initialize the syntax table. */
1957 init_syntax_once ();
1960 if (bufp->allocated == 0)
1963 { /* If zero allocated, but buffer is non-null, try to realloc
1964 enough space. This loses if buffer's address is bogus, but
1965 that is the user's responsibility. */
1966 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1969 { /* Caller did not allocate a buffer. Do it for them. */
1970 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1972 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1974 bufp->allocated = INIT_BUF_SIZE;
1977 begalt = b = bufp->buffer;
1979 /* Loop through the uncompiled pattern until we're at the end. */
1988 if ( /* If at start of pattern, it's an operator. */
1990 /* If context independent, it's an operator. */
1991 || syntax & RE_CONTEXT_INDEP_ANCHORS
1992 /* Otherwise, depends on what's come before. */
1993 || at_begline_loc_p (pattern, p, syntax))
2003 if ( /* If at end of pattern, it's an operator. */
2005 /* If context independent, it's an operator. */
2006 || syntax & RE_CONTEXT_INDEP_ANCHORS
2007 /* Otherwise, depends on what's next. */
2008 || at_endline_loc_p (p, pend, syntax))
2018 if ((syntax & RE_BK_PLUS_QM)
2019 || (syntax & RE_LIMITED_OPS))
2023 /* If there is no previous pattern... */
2026 if (syntax & RE_CONTEXT_INVALID_OPS)
2027 FREE_STACK_RETURN (REG_BADRPT);
2028 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2033 /* Are we optimizing this jump? */
2034 boolean keep_string_p = false;
2036 /* 1 means zero (many) matches is allowed. */
2037 char zero_times_ok = 0, many_times_ok = 0;
2039 /* If there is a sequence of repetition chars, collapse it
2040 down to just one (the right one). We can't combine
2041 interval operators with these because of, e.g., `a{2}*',
2042 which should only match an even number of `a's. */
2046 zero_times_ok |= c != '+';
2047 many_times_ok |= c != '?';
2055 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2058 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2060 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2063 if (!(c1 == '+' || c1 == '?'))
2078 /* If we get here, we found another repeat character. */
2081 /* Star, etc. applied to an empty pattern is equivalent
2082 to an empty pattern. */
2086 /* Now we know whether or not zero matches is allowed
2087 and also whether or not two or more matches is allowed. */
2089 { /* More than one repetition is allowed, so put in at the
2090 end a backward relative jump from `b' to before the next
2091 jump we're going to put in below (which jumps from
2092 laststart to after this jump).
2094 But if we are at the `*' in the exact sequence `.*\n',
2095 insert an unconditional jump backwards to the .,
2096 instead of the beginning of the loop. This way we only
2097 push a failure point once, instead of every time
2098 through the loop. */
2099 assert (p - 1 > pattern);
2101 /* Allocate the space for the jump. */
2102 GET_BUFFER_SPACE (3);
2104 /* We know we are not at the first character of the pattern,
2105 because laststart was nonzero. And we've already
2106 incremented `p', by the way, to be the character after
2107 the `*'. Do we have to do something analogous here
2108 for null bytes, because of RE_DOT_NOT_NULL? */
2109 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2111 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2112 && !(syntax & RE_DOT_NEWLINE))
2113 { /* We have .*\n. */
2114 STORE_JUMP (jump, b, laststart);
2115 keep_string_p = true;
2118 /* Anything else. */
2119 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
2121 /* We've added more stuff to the buffer. */
2125 /* On failure, jump from laststart to b + 3, which will be the
2126 end of the buffer after this jump is inserted. */
2127 GET_BUFFER_SPACE (3);
2128 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2136 /* At least one repetition is required, so insert a
2137 `dummy_failure_jump' before the initial
2138 `on_failure_jump' instruction of the loop. This
2139 effects a skip over that instruction the first time
2140 we hit that loop. */
2141 GET_BUFFER_SPACE (3);
2142 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
2157 CLEAR_RANGE_TABLE_WORK_USED (range_table_work);
2159 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2161 /* Ensure that we have enough space to push a charset: the
2162 opcode, the length count, and the bitset; 34 bytes in all. */
2163 GET_BUFFER_SPACE (34);
2167 /* We test `*p == '^' twice, instead of using an if
2168 statement, so we only need one BUF_PUSH. */
2169 BUF_PUSH (*p == '^' ? charset_not : charset);
2173 /* Remember the first position in the bracket expression. */
2176 /* Push the number of bytes in the bitmap. */
2177 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
2179 /* Clear the whole map. */
2180 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
2182 /* charset_not matches newline according to a syntax bit. */
2183 if ((re_opcode_t) b[-2] == charset_not
2184 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2185 SET_LIST_BIT ('\n');
2187 /* Read in characters and ranges, setting map bits. */
2191 boolean escaped_char = false;
2193 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2197 /* \ might escape characters inside [...] and [^...]. */
2198 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2200 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2203 escaped_char = true;
2207 /* Could be the end of the bracket expression. If it's
2208 not (i.e., when the bracket expression is `[]' so
2209 far), the ']' character bit gets set way below. */
2210 if (c == ']' && p != p1 + 1)
2214 /* If C indicates start of multibyte char, get the
2215 actual character code in C, and set the pattern
2216 pointer P to the next character boundary. */
2217 if (bufp->multibyte && BASE_LEADING_CODE_P (c))
2220 c = STRING_CHAR_AND_LENGTH (p, pend - p, len);
2223 /* What should we do for the character which is
2224 greater than 0x7F, but not BASE_LEADING_CODE_P?
2227 /* See if we're at the beginning of a possible character
2230 else if (!escaped_char &&
2231 syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2233 /* Leave room for the null. */
2234 char str[CHAR_CLASS_MAX_LENGTH + 1];
2239 /* If pattern is `[[:'. */
2240 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2245 if (c == ':' || c == ']' || p == pend
2246 || c1 == CHAR_CLASS_MAX_LENGTH)
2252 /* If isn't a word bracketed by `[:' and `:]':
2253 undo the ending character, the letters, and
2254 leave the leading `:' and `[' (but set bits for
2256 if (c == ':' && *p == ']')
2259 boolean is_alnum = STREQ (str, "alnum");
2260 boolean is_alpha = STREQ (str, "alpha");
2261 boolean is_blank = STREQ (str, "blank");
2262 boolean is_cntrl = STREQ (str, "cntrl");
2263 boolean is_digit = STREQ (str, "digit");
2264 boolean is_graph = STREQ (str, "graph");
2265 boolean is_lower = STREQ (str, "lower");
2266 boolean is_print = STREQ (str, "print");
2267 boolean is_punct = STREQ (str, "punct");
2268 boolean is_space = STREQ (str, "space");
2269 boolean is_upper = STREQ (str, "upper");
2270 boolean is_xdigit = STREQ (str, "xdigit");
2272 if (!IS_CHAR_CLASS (str))
2273 FREE_STACK_RETURN (REG_ECTYPE);
2275 /* Throw away the ] at the end of the character
2279 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2281 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2283 int translated = TRANSLATE (ch);
2284 /* This was split into 3 if's to
2285 avoid an arbitrary limit in some compiler. */
2286 if ( (is_alnum && ISALNUM (ch))
2287 || (is_alpha && ISALPHA (ch))
2288 || (is_blank && ISBLANK (ch))
2289 || (is_cntrl && ISCNTRL (ch)))
2290 SET_LIST_BIT (translated);
2291 if ( (is_digit && ISDIGIT (ch))
2292 || (is_graph && ISGRAPH (ch))
2293 || (is_lower && ISLOWER (ch))
2294 || (is_print && ISPRINT (ch)))
2295 SET_LIST_BIT (translated);
2296 if ( (is_punct && ISPUNCT (ch))
2297 || (is_space && ISSPACE (ch))
2298 || (is_upper && ISUPPER (ch))
2299 || (is_xdigit && ISXDIGIT (ch)))
2300 SET_LIST_BIT (translated);
2303 /* Repeat the loop. */
2313 /* Because the `:' may starts the range, we
2314 can't simply set bit and repeat the loop.
2315 Instead, just set it to C and handle below. */
2320 if (p < pend && p[0] == '-' && p[1] != ']')
2323 /* Discard the `-'. */
2326 /* Fetch the character which ends the range. */
2328 if (bufp->multibyte && BASE_LEADING_CODE_P (c1))
2331 c1 = STRING_CHAR_AND_LENGTH (p, pend - p, len);
2335 if (!SAME_CHARSET_P (c, c1))
2336 FREE_STACK_RETURN (REG_ERANGE);
2339 /* Range from C to C. */
2342 /* Set the range ... */
2343 if (SINGLE_BYTE_CHAR_P (c))
2344 /* ... into bitmap. */
2347 int range_start = c, range_end = c1;
2349 /* If the start is after the end, the range is empty. */
2350 if (range_start > range_end)
2352 if (syntax & RE_NO_EMPTY_RANGES)
2353 FREE_STACK_RETURN (REG_ERANGE);
2354 /* Else, repeat the loop. */
2358 for (this_char = range_start; this_char <= range_end;
2360 SET_LIST_BIT (TRANSLATE (this_char));
2364 /* ... into range table. */
2365 SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1);
2368 /* Discard any (non)matching list bytes that are all 0 at the
2369 end of the map. Decrease the map-length byte too. */
2370 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2374 /* Build real range table from work area. */
2375 if (RANGE_TABLE_WORK_USED (range_table_work))
2378 int used = RANGE_TABLE_WORK_USED (range_table_work);
2380 /* Allocate space for COUNT + RANGE_TABLE. Needs two
2381 bytes for COUNT and three bytes for each character. */
2382 GET_BUFFER_SPACE (2 + used * 3);
2384 /* Indicate the existence of range table. */
2385 laststart[1] |= 0x80;
2387 STORE_NUMBER_AND_INCR (b, used / 2);
2388 for (i = 0; i < used; i++)
2389 STORE_CHARACTER_AND_INCR
2390 (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
2397 if (syntax & RE_NO_BK_PARENS)
2404 if (syntax & RE_NO_BK_PARENS)
2411 if (syntax & RE_NEWLINE_ALT)
2418 if (syntax & RE_NO_BK_VBAR)
2425 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2426 goto handle_interval;
2432 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2434 /* Do not translate the character after the \, so that we can
2435 distinguish, e.g., \B from \b, even if we normally would
2436 translate, e.g., B to b. */
2442 if (syntax & RE_NO_BK_PARENS)
2443 goto normal_backslash;
2449 if (COMPILE_STACK_FULL)
2451 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2452 compile_stack_elt_t);
2453 if (compile_stack.stack == NULL) return REG_ESPACE;
2455 compile_stack.size <<= 1;
2458 /* These are the values to restore when we hit end of this
2459 group. They are all relative offsets, so that if the
2460 whole pattern moves because of realloc, they will still
2462 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2463 COMPILE_STACK_TOP.fixup_alt_jump
2464 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2465 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2466 COMPILE_STACK_TOP.regnum = regnum;
2468 /* We will eventually replace the 0 with the number of
2469 groups inner to this one. But do not push a
2470 start_memory for groups beyond the last one we can
2471 represent in the compiled pattern. */
2472 if (regnum <= MAX_REGNUM)
2474 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2475 BUF_PUSH_3 (start_memory, regnum, 0);
2478 compile_stack.avail++;
2483 /* If we've reached MAX_REGNUM groups, then this open
2484 won't actually generate any code, so we'll have to
2485 clear pending_exact explicitly. */
2491 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2493 if (COMPILE_STACK_EMPTY)
2494 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2495 goto normal_backslash;
2497 FREE_STACK_RETURN (REG_ERPAREN);
2501 { /* Push a dummy failure point at the end of the
2502 alternative for a possible future
2503 `pop_failure_jump' to pop. See comments at
2504 `push_dummy_failure' in `re_match_2'. */
2505 BUF_PUSH (push_dummy_failure);
2507 /* We allocated space for this jump when we assigned
2508 to `fixup_alt_jump', in the `handle_alt' case below. */
2509 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2512 /* See similar code for backslashed left paren above. */
2513 if (COMPILE_STACK_EMPTY)
2514 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2517 FREE_STACK_RETURN (REG_ERPAREN);
2519 /* Since we just checked for an empty stack above, this
2520 ``can't happen''. */
2521 assert (compile_stack.avail != 0);
2523 /* We don't just want to restore into `regnum', because
2524 later groups should continue to be numbered higher,
2525 as in `(ab)c(de)' -- the second group is #2. */
2526 regnum_t this_group_regnum;
2528 compile_stack.avail--;
2529 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2531 = COMPILE_STACK_TOP.fixup_alt_jump
2532 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2534 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2535 this_group_regnum = COMPILE_STACK_TOP.regnum;
2536 /* If we've reached MAX_REGNUM groups, then this open
2537 won't actually generate any code, so we'll have to
2538 clear pending_exact explicitly. */
2541 /* We're at the end of the group, so now we know how many
2542 groups were inside this one. */
2543 if (this_group_regnum <= MAX_REGNUM)
2545 unsigned char *inner_group_loc
2546 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2548 *inner_group_loc = regnum - this_group_regnum;
2549 BUF_PUSH_3 (stop_memory, this_group_regnum,
2550 regnum - this_group_regnum);
2556 case '|': /* `\|'. */
2557 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2558 goto normal_backslash;
2560 if (syntax & RE_LIMITED_OPS)
2563 /* Insert before the previous alternative a jump which
2564 jumps to this alternative if the former fails. */
2565 GET_BUFFER_SPACE (3);
2566 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2570 /* The alternative before this one has a jump after it
2571 which gets executed if it gets matched. Adjust that
2572 jump so it will jump to this alternative's analogous
2573 jump (put in below, which in turn will jump to the next
2574 (if any) alternative's such jump, etc.). The last such
2575 jump jumps to the correct final destination. A picture:
2581 If we are at `b', then fixup_alt_jump right now points to a
2582 three-byte space after `a'. We'll put in the jump, set
2583 fixup_alt_jump to right after `b', and leave behind three
2584 bytes which we'll fill in when we get to after `c'. */
2587 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2589 /* Mark and leave space for a jump after this alternative,
2590 to be filled in later either by next alternative or
2591 when know we're at the end of a series of alternatives. */
2593 GET_BUFFER_SPACE (3);
2602 /* If \{ is a literal. */
2603 if (!(syntax & RE_INTERVALS)
2604 /* If we're at `\{' and it's not the open-interval
2606 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2607 || (p - 2 == pattern && p == pend))
2608 goto normal_backslash;
2612 /* If got here, then the syntax allows intervals. */
2614 /* At least (most) this many matches must be made. */
2615 int lower_bound = -1, upper_bound = -1;
2617 beg_interval = p - 1;
2621 if (syntax & RE_NO_BK_BRACES)
2622 goto unfetch_interval;
2624 FREE_STACK_RETURN (REG_EBRACE);
2627 GET_UNSIGNED_NUMBER (lower_bound);
2631 GET_UNSIGNED_NUMBER (upper_bound);
2632 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2635 /* Interval such as `{1}' => match exactly once. */
2636 upper_bound = lower_bound;
2638 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2639 || lower_bound > upper_bound)
2641 if (syntax & RE_NO_BK_BRACES)
2642 goto unfetch_interval;
2644 FREE_STACK_RETURN (REG_BADBR);
2647 if (!(syntax & RE_NO_BK_BRACES))
2649 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2656 if (syntax & RE_NO_BK_BRACES)
2657 goto unfetch_interval;
2659 FREE_STACK_RETURN (REG_BADBR);
2662 /* We just parsed a valid interval. */
2664 /* If it's invalid to have no preceding re. */
2667 if (syntax & RE_CONTEXT_INVALID_OPS)
2668 FREE_STACK_RETURN (REG_BADRPT);
2669 else if (syntax & RE_CONTEXT_INDEP_OPS)
2672 goto unfetch_interval;
2675 /* If the upper bound is zero, don't want to succeed at
2676 all; jump from `laststart' to `b + 3', which will be
2677 the end of the buffer after we insert the jump. */
2678 if (upper_bound == 0)
2680 GET_BUFFER_SPACE (3);
2681 INSERT_JUMP (jump, laststart, b + 3);
2685 /* Otherwise, we have a nontrivial interval. When
2686 we're all done, the pattern will look like:
2687 set_number_at <jump count> <upper bound>
2688 set_number_at <succeed_n count> <lower bound>
2689 succeed_n <after jump addr> <succeed_n count>
2691 jump_n <succeed_n addr> <jump count>
2692 (The upper bound and `jump_n' are omitted if
2693 `upper_bound' is 1, though.) */
2695 { /* If the upper bound is > 1, we need to insert
2696 more at the end of the loop. */
2697 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2699 GET_BUFFER_SPACE (nbytes);
2701 /* Initialize lower bound of the `succeed_n', even
2702 though it will be set during matching by its
2703 attendant `set_number_at' (inserted next),
2704 because `re_compile_fastmap' needs to know.
2705 Jump to the `jump_n' we might insert below. */
2706 INSERT_JUMP2 (succeed_n, laststart,
2707 b + 5 + (upper_bound > 1) * 5,
2711 /* Code to initialize the lower bound. Insert
2712 before the `succeed_n'. The `5' is the last two
2713 bytes of this `set_number_at', plus 3 bytes of
2714 the following `succeed_n'. */
2715 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2718 if (upper_bound > 1)
2719 { /* More than one repetition is allowed, so
2720 append a backward jump to the `succeed_n'
2721 that starts this interval.
2723 When we've reached this during matching,
2724 we'll have matched the interval once, so
2725 jump back only `upper_bound - 1' times. */
2726 STORE_JUMP2 (jump_n, b, laststart + 5,
2730 /* The location we want to set is the second
2731 parameter of the `jump_n'; that is `b-2' as
2732 an absolute address. `laststart' will be
2733 the `set_number_at' we're about to insert;
2734 `laststart+3' the number to set, the source
2735 for the relative address. But we are
2736 inserting into the middle of the pattern --
2737 so everything is getting moved up by 5.
2738 Conclusion: (b - 2) - (laststart + 3) + 5,
2739 i.e., b - laststart.
2741 We insert this at the beginning of the loop
2742 so that if we fail during matching, we'll
2743 reinitialize the bounds. */
2744 insert_op2 (set_number_at, laststart, b - laststart,
2745 upper_bound - 1, b);
2750 beg_interval = NULL;
2755 /* If an invalid interval, match the characters as literals. */
2756 assert (beg_interval);
2758 beg_interval = NULL;
2760 /* normal_char and normal_backslash need `c'. */
2763 if (!(syntax & RE_NO_BK_BRACES))
2765 if (p > pattern && p[-1] == '\\')
2766 goto normal_backslash;
2771 /* There is no way to specify the before_dot and after_dot
2772 operators. rms says this is ok. --karl */
2780 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2786 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2792 BUF_PUSH_2 (categoryspec, c);
2798 BUF_PUSH_2 (notcategoryspec, c);
2805 BUF_PUSH (wordchar);
2811 BUF_PUSH (notwordchar);
2824 BUF_PUSH (wordbound);
2828 BUF_PUSH (notwordbound);
2839 case '1': case '2': case '3': case '4': case '5':
2840 case '6': case '7': case '8': case '9':
2841 if (syntax & RE_NO_BK_REFS)
2847 FREE_STACK_RETURN (REG_ESUBREG);
2849 /* Can't back reference to a subexpression if inside of it. */
2850 if (group_in_compile_stack (compile_stack, c1))
2854 BUF_PUSH_2 (duplicate, c1);
2860 if (syntax & RE_BK_PLUS_QM)
2863 goto normal_backslash;
2867 /* You might think it would be useful for \ to mean
2868 not to translate; but if we don't translate it
2869 it will never match anything. */
2877 /* Expects the character in `c'. */
2879 p1 = p - 1; /* P1 points the head of C. */
2881 if (bufp->multibyte)
2882 /* Set P to the next character boundary. */
2883 p += MULTIBYTE_FORM_LENGTH (p1, pend - p1) - 1;
2885 /* If no exactn currently being built. */
2888 /* If last exactn not at current position. */
2889 || pending_exact + *pending_exact + 1 != b
2891 /* We have only one byte following the exactn for the count. */
2892 || *pending_exact >= (1 << BYTEWIDTH) - (p - p1)
2894 /* If followed by a repetition operator. */
2895 || *p == '*' || *p == '^'
2896 || ((syntax & RE_BK_PLUS_QM)
2897 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2898 : (*p == '+' || *p == '?'))
2899 || ((syntax & RE_INTERVALS)
2900 && ((syntax & RE_NO_BK_BRACES)
2902 : (p[0] == '\\' && p[1] == '{'))))
2904 /* Start building a new exactn. */
2908 BUF_PUSH_2 (exactn, 0);
2909 pending_exact = b - 1;
2912 /* Here, C may translated, therefore C may not equal to *P1. */
2920 /* Rest of multibyte form should be copied literally. */
2921 c = *(unsigned char *)p1;
2925 } /* while p != pend */
2928 /* Through the pattern now. */
2931 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2933 if (!COMPILE_STACK_EMPTY)
2934 FREE_STACK_RETURN (REG_EPAREN);
2936 /* If we don't want backtracking, force success
2937 the first time we reach the end of the compiled pattern. */
2938 if (syntax & RE_NO_POSIX_BACKTRACKING)
2941 free (compile_stack.stack);
2943 /* We have succeeded; set the length of the buffer. */
2944 bufp->used = b - bufp->buffer;
2949 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2950 print_compiled_pattern (bufp);
2954 #ifndef MATCH_MAY_ALLOCATE
2955 /* Initialize the failure stack to the largest possible stack. This
2956 isn't necessary unless we're trying to avoid calling alloca in
2957 the search and match routines. */
2959 int num_regs = bufp->re_nsub + 1;
2961 if (fail_stack.size < re_max_failures * TYPICAL_FAILURE_SIZE)
2963 fail_stack.size = re_max_failures * TYPICAL_FAILURE_SIZE);
2966 if (! fail_stack.stack)
2968 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2969 * sizeof (fail_stack_elt_t));
2972 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2974 * sizeof (fail_stack_elt_t)));
2975 #else /* not emacs */
2976 if (! fail_stack.stack)
2978 = (fail_stack_elt_t *) malloc (fail_stack.size
2979 * sizeof (fail_stack_elt_t));
2982 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2984 * sizeof (fail_stack_elt_t)));
2985 #endif /* not emacs */
2988 regex_grow_registers (num_regs);
2990 #endif /* not MATCH_MAY_ALLOCATE */
2993 } /* regex_compile */
2995 /* Subroutines for `regex_compile'. */
2997 /* Store OP at LOC followed by two-byte integer parameter ARG. */
3000 store_op1 (op, loc, arg)
3005 *loc = (unsigned char) op;
3006 STORE_NUMBER (loc + 1, arg);
3010 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
3013 store_op2 (op, loc, arg1, arg2)
3018 *loc = (unsigned char) op;
3019 STORE_NUMBER (loc + 1, arg1);
3020 STORE_NUMBER (loc + 3, arg2);
3024 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
3025 for OP followed by two-byte integer parameter ARG. */
3028 insert_op1 (op, loc, arg, end)
3034 register unsigned char *pfrom = end;
3035 register unsigned char *pto = end + 3;
3037 while (pfrom != loc)
3040 store_op1 (op, loc, arg);
3044 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
3047 insert_op2 (op, loc, arg1, arg2, end)
3053 register unsigned char *pfrom = end;
3054 register unsigned char *pto = end + 5;
3056 while (pfrom != loc)
3059 store_op2 (op, loc, arg1, arg2);
3063 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
3064 after an alternative or a begin-subexpression. We assume there is at
3065 least one character before the ^. */
3068 at_begline_loc_p (pattern, p, syntax)
3069 const char *pattern, *p;
3070 reg_syntax_t syntax;
3072 const char *prev = p - 2;
3073 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
3076 /* After a subexpression? */
3077 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
3078 /* After an alternative? */
3079 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
3083 /* The dual of at_begline_loc_p. This one is for $. We assume there is
3084 at least one character after the $, i.e., `P < PEND'. */
3087 at_endline_loc_p (p, pend, syntax)
3088 const char *p, *pend;
3091 const char *next = p;
3092 boolean next_backslash = *next == '\\';
3093 const char *next_next = p + 1 < pend ? p + 1 : 0;
3096 /* Before a subexpression? */
3097 (syntax & RE_NO_BK_PARENS ? *next == ')'
3098 : next_backslash && next_next && *next_next == ')')
3099 /* Before an alternative? */
3100 || (syntax & RE_NO_BK_VBAR ? *next == '|'
3101 : next_backslash && next_next && *next_next == '|');
3105 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
3106 false if it's not. */
3109 group_in_compile_stack (compile_stack, regnum)
3110 compile_stack_type compile_stack;
3115 for (this_element = compile_stack.avail - 1;
3118 if (compile_stack.stack[this_element].regnum == regnum)
3125 /* Read the ending character of a range (in a bracket expression) from the
3126 uncompiled pattern *P_PTR (which ends at PEND). We assume the
3127 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
3128 Then we set the translation of all bits between the starting and
3129 ending characters (inclusive) in the compiled pattern B.
3131 Return an error code.
3133 We use these short variable names so we can use the same macros as
3134 `regex_compile' itself. */
3136 static reg_errcode_t
3137 compile_range (p_ptr, pend, translate, syntax, b)
3138 const char **p_ptr, *pend;
3139 RE_TRANSLATE_TYPE translate;
3140 reg_syntax_t syntax;
3145 const char *p = *p_ptr;
3146 int range_start, range_end;
3151 /* Even though the pattern is a signed `char *', we need to fetch
3152 with unsigned char *'s; if the high bit of the pattern character
3153 is set, the range endpoints will be negative if we fetch using a
3156 We also want to fetch the endpoints without translating them; the
3157 appropriate translation is done in the bit-setting loop below. */
3158 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
3159 range_start = ((const unsigned char *) p)[-2];
3160 range_end = ((const unsigned char *) p)[0];
3162 /* Have to increment the pointer into the pattern string, so the
3163 caller isn't still at the ending character. */
3166 /* If the start is after the end, the range is empty. */
3167 if (range_start > range_end)
3168 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
3170 /* Here we see why `this_char' has to be larger than an `unsigned
3171 char' -- the range is inclusive, so if `range_end' == 0xff
3172 (assuming 8-bit characters), we would otherwise go into an infinite
3173 loop, since all characters <= 0xff. */
3174 for (this_char = range_start; this_char <= range_end; this_char++)
3176 SET_LIST_BIT (TRANSLATE (this_char));
3182 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
3183 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
3184 characters can start a string that matches the pattern. This fastmap
3185 is used by re_search to skip quickly over impossible starting points.
3187 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
3188 area as BUFP->fastmap.
3190 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
3193 Returns 0 if we succeed, -2 if an internal error. */
3196 re_compile_fastmap (bufp)
3197 struct re_pattern_buffer *bufp;
3200 #ifdef MATCH_MAY_ALLOCATE
3201 fail_stack_type fail_stack;
3203 #ifndef REGEX_MALLOC
3206 /* We don't push any register information onto the failure stack. */
3207 unsigned num_regs = 0;
3209 register char *fastmap = bufp->fastmap;
3210 unsigned char *pattern = bufp->buffer;
3211 unsigned long size = bufp->used;
3212 unsigned char *p = pattern;
3213 register unsigned char *pend = pattern + size;
3215 /* This holds the pointer to the failure stack, when
3216 it is allocated relocatably. */
3217 fail_stack_elt_t *failure_stack_ptr;
3219 /* Assume that each path through the pattern can be null until
3220 proven otherwise. We set this false at the bottom of switch
3221 statement, to which we get only if a particular path doesn't
3222 match the empty string. */
3223 boolean path_can_be_null = true;
3225 /* We aren't doing a `succeed_n' to begin with. */
3226 boolean succeed_n_p = false;
3228 /* If all elements for base leading-codes in fastmap is set, this
3229 flag is set true. */
3230 boolean match_any_multibyte_characters = false;
3232 /* Maximum code of simple (single byte) character. */
3233 int simple_char_max;
3235 assert (fastmap != NULL && p != NULL);
3238 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
3239 bufp->fastmap_accurate = 1; /* It will be when we're done. */
3240 bufp->can_be_null = 0;
3244 if (p == pend || *p == succeed)
3246 /* We have reached the (effective) end of pattern. */
3247 if (!FAIL_STACK_EMPTY ())
3249 bufp->can_be_null |= path_can_be_null;
3251 /* Reset for next path. */
3252 path_can_be_null = true;
3254 p = fail_stack.stack[--fail_stack.avail].pointer;
3262 /* We should never be about to go beyond the end of the pattern. */
3265 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3268 /* I guess the idea here is to simply not bother with a fastmap
3269 if a backreference is used, since it's too hard to figure out
3270 the fastmap for the corresponding group. Setting
3271 `can_be_null' stops `re_search_2' from using the fastmap, so
3272 that is all we do. */
3274 bufp->can_be_null = 1;
3278 /* Following are the cases which match a character. These end
3288 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3289 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
3295 /* Chars beyond end of map must be allowed. */
3296 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
3299 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
3300 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
3306 for (j = 0; j < (1 << BYTEWIDTH); j++)
3307 if (SYNTAX (j) == Sword)
3313 for (j = 0; j < (1 << BYTEWIDTH); j++)
3314 if (SYNTAX (j) != Sword)
3319 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3321 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
3324 if (CHARSET_RANGE_TABLE_EXISTS_P (&p[-2])
3325 && match_any_multibyte_characters == false)
3327 /* Set fastmap[I] 1 where I is a base leading code of each
3328 multibyte character in the range table. */
3331 /* Make P points the range table. */
3332 p += CHARSET_BITMAP_SIZE (&p[-2]);
3334 /* Extract the number of ranges in range table into
3336 EXTRACT_NUMBER_AND_INCR (count, p);
3337 for (; count > 0; count--, p += 2 * 3) /* XXX */
3339 /* Extract the start of each range. */
3340 EXTRACT_CHARACTER (c, p);
3341 j = CHAR_CHARSET (c);
3342 fastmap[CHARSET_LEADING_CODE_BASE (j)] = 1;
3349 /* Chars beyond end of map must be allowed. End of map is
3350 `127' if bufp->multibyte is nonzero. */
3351 simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH);
3352 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH;
3353 j < simple_char_max; j++)
3356 for (j = CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH - 1, p++;
3358 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
3361 if (bufp->multibyte)
3362 /* Any character set can possibly contain a character
3363 which doesn't match the specified set of characters. */
3365 set_fastmap_for_multibyte_characters:
3366 if (match_any_multibyte_characters == false)
3368 for (j = 0x80; j < 0xA0; j++) /* XXX */
3369 if (BASE_LEADING_CODE_P (j))
3371 match_any_multibyte_characters = true;
3378 simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH);
3379 for (j = 0; j < simple_char_max; j++)
3380 if (SYNTAX (j) == Sword)
3383 if (bufp->multibyte)
3384 /* Any character set can possibly contain a character
3385 whose syntax is `Sword'. */
3386 goto set_fastmap_for_multibyte_characters;
3391 simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH);
3392 for (j = 0; j < simple_char_max; j++)
3393 if (SYNTAX (j) != Sword)
3396 if (bufp->multibyte)
3397 /* Any character set can possibly contain a character
3398 whose syntax is not `Sword'. */
3399 goto set_fastmap_for_multibyte_characters;
3405 int fastmap_newline = fastmap['\n'];
3407 /* `.' matches anything (but if bufp->multibyte is
3408 nonzero, matches `\000' .. `\127' and possible multibyte
3410 if (bufp->multibyte)
3412 simple_char_max = 0x80;
3414 for (j = 0x80; j < 0xA0; j++)
3415 if (BASE_LEADING_CODE_P (j))
3417 match_any_multibyte_characters = true;
3420 simple_char_max = (1 << BYTEWIDTH);
3422 for (j = 0; j < simple_char_max; j++)
3425 /* ... except perhaps newline. */
3426 if (!(bufp->syntax & RE_DOT_NEWLINE))
3427 fastmap['\n'] = fastmap_newline;
3429 /* Return if we have already set `can_be_null'; if we have,
3430 then the fastmap is irrelevant. Something's wrong here. */
3431 else if (bufp->can_be_null)
3434 /* Otherwise, have to check alternative paths. */
3445 /* This match depends on text properties. These end with
3446 aborting optimizations. */
3447 bufp->can_be_null = 1;
3451 simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH);
3452 for (j = 0; j < simple_char_max; j++)
3453 if (SYNTAX (j) == (enum syntaxcode) k)
3456 if (bufp->multibyte)
3457 /* Any character set can possibly contain a character
3458 whose syntax is K. */
3459 goto set_fastmap_for_multibyte_characters;
3464 simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH);
3465 for (j = 0; j < simple_char_max; j++)
3466 if (SYNTAX (j) != (enum syntaxcode) k)
3469 if (bufp->multibyte)
3470 /* Any character set can possibly contain a character
3471 whose syntax is not K. */
3472 goto set_fastmap_for_multibyte_characters;
3479 simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH);
3480 for (j = 0; j < simple_char_max; j++)
3481 if (CHAR_HAS_CATEGORY (j, k))
3484 if (bufp->multibyte)
3485 /* Any character set can possibly contain a character
3486 whose category is K. */
3487 goto set_fastmap_for_multibyte_characters;
3491 case notcategoryspec:
3493 simple_char_max = bufp->multibyte ? 0x80 : (1 << BYTEWIDTH);
3494 for (j = 0; j < simple_char_max; j++)
3495 if (!CHAR_HAS_CATEGORY (j, k))
3498 if (bufp->multibyte)
3499 /* Any character set can possibly contain a character
3500 whose category is not K. */
3501 goto set_fastmap_for_multibyte_characters;
3504 /* All cases after this match the empty string. These end with
3526 case push_dummy_failure:
3531 case pop_failure_jump:
3532 case maybe_pop_jump:
3535 case dummy_failure_jump:
3536 EXTRACT_NUMBER_AND_INCR (j, p);
3541 /* Jump backward implies we just went through the body of a
3542 loop and matched nothing. Opcode jumped to should be
3543 `on_failure_jump' or `succeed_n'. Just treat it like an
3544 ordinary jump. For a * loop, it has pushed its failure
3545 point already; if so, discard that as redundant. */
3546 if ((re_opcode_t) *p != on_failure_jump
3547 && (re_opcode_t) *p != succeed_n)
3551 EXTRACT_NUMBER_AND_INCR (j, p);
3554 /* If what's on the stack is where we are now, pop it. */
3555 if (!FAIL_STACK_EMPTY ()
3556 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3562 case on_failure_jump:
3563 case on_failure_keep_string_jump:
3564 handle_on_failure_jump:
3565 EXTRACT_NUMBER_AND_INCR (j, p);
3567 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3568 end of the pattern. We don't want to push such a point,
3569 since when we restore it above, entering the switch will
3570 increment `p' past the end of the pattern. We don't need
3571 to push such a point since we obviously won't find any more
3572 fastmap entries beyond `pend'. Such a pattern can match
3573 the null string, though. */
3576 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3578 RESET_FAIL_STACK ();
3583 bufp->can_be_null = 1;
3587 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3588 succeed_n_p = false;
3595 /* Get to the number of times to succeed. */
3598 /* Increment p past the n for when k != 0. */
3599 EXTRACT_NUMBER_AND_INCR (k, p);
3603 succeed_n_p = true; /* Spaghetti code alert. */
3604 goto handle_on_failure_jump;
3621 abort (); /* We have listed all the cases. */
3624 /* Getting here means we have found the possible starting
3625 characters for one path of the pattern -- and that the empty
3626 string does not match. We need not follow this path further.
3627 Instead, look at the next alternative (remembered on the
3628 stack), or quit if no more. The test at the top of the loop
3629 does these things. */
3630 path_can_be_null = false;
3634 /* Set `can_be_null' for the last path (also the first path, if the
3635 pattern is empty). */
3636 bufp->can_be_null |= path_can_be_null;
3639 RESET_FAIL_STACK ();
3641 } /* re_compile_fastmap */
3643 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3644 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3645 this memory for recording register information. STARTS and ENDS
3646 must be allocated using the malloc library routine, and must each
3647 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3649 If NUM_REGS == 0, then subsequent matches should allocate their own
3652 Unless this function is called, the first search or match using
3653 PATTERN_BUFFER will allocate its own register data, without
3654 freeing the old data. */
3657 re_set_registers (bufp, regs, num_regs, starts, ends)
3658 struct re_pattern_buffer *bufp;
3659 struct re_registers *regs;
3661 regoff_t *starts, *ends;
3665 bufp->regs_allocated = REGS_REALLOCATE;
3666 regs->num_regs = num_regs;
3667 regs->start = starts;
3672 bufp->regs_allocated = REGS_UNALLOCATED;
3674 regs->start = regs->end = (regoff_t *) 0;
3678 /* Searching routines. */
3680 /* Like re_search_2, below, but only one string is specified, and
3681 doesn't let you say where to stop matching. */
3684 re_search (bufp, string, size, startpos, range, regs)
3685 struct re_pattern_buffer *bufp;
3687 int size, startpos, range;
3688 struct re_registers *regs;
3690 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3694 /* End address of virtual concatenation of string. */
3695 #define STOP_ADDR_VSTRING(P) \
3696 (((P) >= size1 ? string2 + size2 : string1 + size1))
3698 /* Address of POS in the concatenation of virtual string. */
3699 #define POS_ADDR_VSTRING(POS) \
3700 (((POS) >= size1 ? string2 - size1 : string1) + (POS))
3702 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3703 virtual concatenation of STRING1 and STRING2, starting first at index
3704 STARTPOS, then at STARTPOS + 1, and so on.
3706 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3708 RANGE is how far to scan while trying to match. RANGE = 0 means try
3709 only at STARTPOS; in general, the last start tried is STARTPOS +
3712 In REGS, return the indices of the virtual concatenation of STRING1
3713 and STRING2 that matched the entire BUFP->buffer and its contained
3716 Do not consider matching one past the index STOP in the virtual
3717 concatenation of STRING1 and STRING2.
3719 We return either the position in the strings at which the match was
3720 found, -1 if no match, or -2 if error (such as failure
3724 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3725 struct re_pattern_buffer *bufp;
3726 const char *string1, *string2;
3730 struct re_registers *regs;
3734 register char *fastmap = bufp->fastmap;
3735 register RE_TRANSLATE_TYPE translate = bufp->translate;
3736 int total_size = size1 + size2;
3737 int endpos = startpos + range;
3738 int anchored_start = 0;
3740 /* Nonzero if we have to concern multibyte character. */
3741 int multibyte = bufp->multibyte;
3743 /* Check for out-of-range STARTPOS. */
3744 if (startpos < 0 || startpos > total_size)
3747 /* Fix up RANGE if it might eventually take us outside
3748 the virtual concatenation of STRING1 and STRING2.
3749 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3751 range = 0 - startpos;
3752 else if (endpos > total_size)
3753 range = total_size - startpos;
3755 /* If the search isn't to be a backwards one, don't waste time in a
3756 search for a pattern that must be anchored. */
3757 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3766 /* In a forward search for something that starts with \=.
3767 don't keep searching past point. */
3768 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3770 range = PT - startpos;
3776 /* Update the fastmap now if not correct already. */
3777 if (fastmap && !bufp->fastmap_accurate)
3778 if (re_compile_fastmap (bufp) == -2)
3781 /* See whether the pattern is anchored. */
3782 if (bufp->buffer[0] == begline)
3786 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object,
3787 POS_AS_IN_BUFFER (startpos > 0
3788 ? startpos - 1 : startpos),
3792 /* Loop through the string, looking for a place to start matching. */
3795 /* If the pattern is anchored,
3796 skip quickly past places we cannot match.
3797 We don't bother to treat startpos == 0 specially
3798 because that case doesn't repeat. */
3799 if (anchored_start && startpos > 0)
3801 if (! (bufp->newline_anchor
3802 && ((startpos <= size1 ? string1[startpos - 1]
3803 : string2[startpos - size1 - 1])
3808 /* If a fastmap is supplied, skip quickly over characters that
3809 cannot be the start of a match. If the pattern can match the
3810 null string, however, we don't need to skip characters; we want
3811 the first null string. */
3812 if (fastmap && startpos < total_size && !bufp->can_be_null)
3814 if (range > 0) /* Searching forwards. */
3816 register const char *d;
3817 register int lim = 0;
3820 if (startpos < size1 && startpos + range >= size1)
3821 lim = range - (size1 - startpos);
3823 d = POS_ADDR_VSTRING (startpos);
3825 /* Written out as an if-else to avoid testing `translate'
3829 && !fastmap[(unsigned char)
3830 RE_TRANSLATE (translate, (unsigned char) *d++)])
3833 while (range > lim && !fastmap[(unsigned char) *d++])
3836 startpos += irange - range;
3838 else /* Searching backwards. */
3840 register char c = (size1 == 0 || startpos >= size1
3841 ? string2[startpos - size1]
3842 : string1[startpos]);
3844 if (!fastmap[(unsigned char) TRANSLATE (c)])
3849 /* If can't match the null string, and that's all we have left, fail. */
3850 if (range >= 0 && startpos == total_size && fastmap
3851 && !bufp->can_be_null)
3854 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3855 startpos, regs, stop);
3856 #ifndef REGEX_MALLOC
3873 /* Update STARTPOS to the next character boundary. */
3876 const unsigned char *p
3877 = (const unsigned char *) POS_ADDR_VSTRING (startpos);
3878 const unsigned char *pend
3879 = (const unsigned char *) STOP_ADDR_VSTRING (startpos);
3880 int len = MULTIBYTE_FORM_LENGTH (p, pend - p);
3898 /* Update STARTPOS to the previous character boundary. */
3901 const unsigned char *p
3902 = (const unsigned char *) POS_ADDR_VSTRING (startpos);
3905 /* Find the head of multibyte form. */
3906 while (!CHAR_HEAD_P (p))
3911 if (MULTIBYTE_FORM_LENGTH (p, len + 1) != (len + 1))
3928 /* Declarations and macros for re_match_2. */
3930 static int bcmp_translate ();
3931 static boolean alt_match_null_string_p (),
3932 common_op_match_null_string_p (),
3933 group_match_null_string_p ();
3935 /* This converts PTR, a pointer into one of the search strings `string1'
3936 and `string2' into an offset from the beginning of that string. */
3937 #define POINTER_TO_OFFSET(ptr) \
3938 (FIRST_STRING_P (ptr) \
3939 ? ((regoff_t) ((ptr) - string1)) \
3940 : ((regoff_t) ((ptr) - string2 + size1)))
3942 /* Macros for dealing with the split strings in re_match_2. */
3944 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3946 /* Call before fetching a character with *d. This switches over to
3947 string2 if necessary. */
3948 #define PREFETCH() \
3951 /* End of string2 => fail. */ \
3952 if (dend == end_match_2) \
3954 /* End of string1 => advance to string2. */ \
3956 dend = end_match_2; \
3960 /* Test if at very beginning or at very end of the virtual concatenation
3961 of `string1' and `string2'. If only one string, it's `string2'. */
3962 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3963 #define AT_STRINGS_END(d) ((d) == end2)
3966 /* Test if D points to a character which is word-constituent. We have
3967 two special cases to check for: if past the end of string1, look at
3968 the first character in string2; and if before the beginning of
3969 string2, look at the last character in string1. */
3970 #define WORDCHAR_P(d) \
3971 (SYNTAX ((d) == end1 ? *string2 \
3972 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3975 /* Disabled due to a compiler bug -- see comment at case wordbound */
3977 /* The comment at case wordbound is following one, but we don't use
3978 AT_WORD_BOUNDARY anymore to support multibyte form.
3980 The DEC Alpha C compiler 3.x generates incorrect code for the
3981 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
3982 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
3983 macro and introducing temporary variables works around the bug. */
3986 /* Test if the character before D and the one at D differ with respect
3987 to being word-constituent. */
3988 #define AT_WORD_BOUNDARY(d) \
3989 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3990 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3993 /* Free everything we malloc. */
3994 #ifdef MATCH_MAY_ALLOCATE
3995 #define FREE_VAR(var) if (var) { REGEX_FREE (var); var = NULL; } else
3996 #define FREE_VARIABLES() \
3998 REGEX_FREE_STACK (fail_stack.stack); \
3999 FREE_VAR (regstart); \
4000 FREE_VAR (regend); \
4001 FREE_VAR (old_regstart); \
4002 FREE_VAR (old_regend); \
4003 FREE_VAR (best_regstart); \
4004 FREE_VAR (best_regend); \
4005 FREE_VAR (reg_info); \
4006 FREE_VAR (reg_dummy); \
4007 FREE_VAR (reg_info_dummy); \
4010 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
4011 #endif /* not MATCH_MAY_ALLOCATE */
4013 /* These values must meet several constraints. They must not be valid
4014 register values; since we have a limit of 255 registers (because
4015 we use only one byte in the pattern for the register number), we can
4016 use numbers larger than 255. They must differ by 1, because of
4017 NUM_FAILURE_ITEMS above. And the value for the lowest register must
4018 be larger than the value for the highest register, so we do not try
4019 to actually save any registers when none are active. */
4020 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
4021 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
4023 /* Matching routines. */
4025 #ifndef emacs /* Emacs never uses this. */
4026 /* re_match is like re_match_2 except it takes only a single string. */
4029 re_match (bufp, string, size, pos, regs)
4030 struct re_pattern_buffer *bufp;
4033 struct re_registers *regs;
4035 int result = re_match_2_internal (bufp, NULL, 0, string, size,
4040 #endif /* not emacs */
4043 /* In Emacs, this is the string or buffer in which we
4044 are matching. It is used for looking up syntax properties. */
4045 Lisp_Object re_match_object;
4048 /* re_match_2 matches the compiled pattern in BUFP against the
4049 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
4050 and SIZE2, respectively). We start matching at POS, and stop
4053 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
4054 store offsets for the substring each group matched in REGS. See the
4055 documentation for exactly how many groups we fill.
4057 We return -1 if no match, -2 if an internal error (such as the
4058 failure stack overflowing). Otherwise, we return the length of the
4059 matched substring. */
4062 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
4063 struct re_pattern_buffer *bufp;
4064 const char *string1, *string2;
4067 struct re_registers *regs;
4073 SETUP_SYNTAX_TABLE_FOR_OBJECT (re_match_object,
4074 POS_AS_IN_BUFFER (pos > 0 ? pos - 1 : pos),
4078 result = re_match_2_internal (bufp, string1, size1, string2, size2,
4084 /* This is a separate function so that we can force an alloca cleanup
4087 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
4088 struct re_pattern_buffer *bufp;
4089 const char *string1, *string2;
4092 struct re_registers *regs;
4095 /* General temporaries. */
4099 /* Just past the end of the corresponding string. */
4100 const char *end1, *end2;
4102 /* Pointers into string1 and string2, just past the last characters in
4103 each to consider matching. */
4104 const char *end_match_1, *end_match_2;
4106 /* Where we are in the data, and the end of the current string. */
4107 const char *d, *dend;
4109 /* Where we are in the pattern, and the end of the pattern. */
4110 unsigned char *p = bufp->buffer;
4111 register unsigned char *pend = p + bufp->used;
4113 /* Mark the opcode just after a start_memory, so we can test for an
4114 empty subpattern when we get to the stop_memory. */
4115 unsigned char *just_past_start_mem = 0;
4117 /* We use this to map every character in the string. */
4118 RE_TRANSLATE_TYPE translate = bufp->translate;
4120 /* Nonzero if we have to concern multibyte character. */
4121 int multibyte = bufp->multibyte;
4123 /* Failure point stack. Each place that can handle a failure further
4124 down the line pushes a failure point on this stack. It consists of
4125 restart, regend, and reg_info for all registers corresponding to
4126 the subexpressions we're currently inside, plus the number of such
4127 registers, and, finally, two char *'s. The first char * is where
4128 to resume scanning the pattern; the second one is where to resume
4129 scanning the strings. If the latter is zero, the failure point is
4130 a ``dummy''; if a failure happens and the failure point is a dummy,
4131 it gets discarded and the next next one is tried. */
4132 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4133 fail_stack_type fail_stack;
4136 static unsigned failure_id = 0;
4137 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
4140 /* This holds the pointer to the failure stack, when
4141 it is allocated relocatably. */
4142 fail_stack_elt_t *failure_stack_ptr;
4144 /* We fill all the registers internally, independent of what we
4145 return, for use in backreferences. The number here includes
4146 an element for register zero. */
4147 unsigned num_regs = bufp->re_nsub + 1;
4149 /* The currently active registers. */
4150 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4151 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4153 /* Information on the contents of registers. These are pointers into
4154 the input strings; they record just what was matched (on this
4155 attempt) by a subexpression part of the pattern, that is, the
4156 regnum-th regstart pointer points to where in the pattern we began
4157 matching and the regnum-th regend points to right after where we
4158 stopped matching the regnum-th subexpression. (The zeroth register
4159 keeps track of what the whole pattern matches.) */
4160 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4161 const char **regstart, **regend;
4164 /* If a group that's operated upon by a repetition operator fails to
4165 match anything, then the register for its start will need to be
4166 restored because it will have been set to wherever in the string we
4167 are when we last see its open-group operator. Similarly for a
4169 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4170 const char **old_regstart, **old_regend;
4173 /* The is_active field of reg_info helps us keep track of which (possibly
4174 nested) subexpressions we are currently in. The matched_something
4175 field of reg_info[reg_num] helps us tell whether or not we have
4176 matched any of the pattern so far this time through the reg_num-th
4177 subexpression. These two fields get reset each time through any
4178 loop their register is in. */
4179 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
4180 register_info_type *reg_info;
4183 /* The following record the register info as found in the above
4184 variables when we find a match better than any we've seen before.
4185 This happens as we backtrack through the failure points, which in
4186 turn happens only if we have not yet matched the entire string. */
4187 unsigned best_regs_set = false;
4188 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4189 const char **best_regstart, **best_regend;
4192 /* Logically, this is `best_regend[0]'. But we don't want to have to
4193 allocate space for that if we're not allocating space for anything
4194 else (see below). Also, we never need info about register 0 for
4195 any of the other register vectors, and it seems rather a kludge to
4196 treat `best_regend' differently than the rest. So we keep track of
4197 the end of the best match so far in a separate variable. We
4198 initialize this to NULL so that when we backtrack the first time
4199 and need to test it, it's not garbage. */
4200 const char *match_end = NULL;
4202 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
4203 int set_regs_matched_done = 0;
4205 /* Used when we pop values we don't care about. */
4206 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
4207 const char **reg_dummy;
4208 register_info_type *reg_info_dummy;
4212 /* Counts the total number of registers pushed. */
4213 unsigned num_regs_pushed = 0;
4216 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
4220 #ifdef MATCH_MAY_ALLOCATE
4221 /* Do not bother to initialize all the register variables if there are
4222 no groups in the pattern, as it takes a fair amount of time. If
4223 there are groups, we include space for register 0 (the whole
4224 pattern), even though we never use it, since it simplifies the
4225 array indexing. We should fix this. */
4228 regstart = REGEX_TALLOC (num_regs, const char *);
4229 regend = REGEX_TALLOC (num_regs, const char *);
4230 old_regstart = REGEX_TALLOC (num_regs, const char *);
4231 old_regend = REGEX_TALLOC (num_regs, const char *);
4232 best_regstart = REGEX_TALLOC (num_regs, const char *);
4233 best_regend = REGEX_TALLOC (num_regs, const char *);
4234 reg_info = REGEX_TALLOC (num_regs, register_info_type);
4235 reg_dummy = REGEX_TALLOC (num_regs, const char *);
4236 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
4238 if (!(regstart && regend && old_regstart && old_regend && reg_info
4239 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
4247 /* We must initialize all our variables to NULL, so that
4248 `FREE_VARIABLES' doesn't try to free them. */
4249 regstart = regend = old_regstart = old_regend = best_regstart
4250 = best_regend = reg_dummy = NULL;
4251 reg_info = reg_info_dummy = (register_info_type *) NULL;
4253 #endif /* MATCH_MAY_ALLOCATE */
4255 /* The starting position is bogus. */
4256 if (pos < 0 || pos > size1 + size2)
4262 /* Initialize subexpression text positions to -1 to mark ones that no
4263 start_memory/stop_memory has been seen for. Also initialize the
4264 register information struct. */
4265 for (mcnt = 1; mcnt < num_regs; mcnt++)
4267 regstart[mcnt] = regend[mcnt]
4268 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
4270 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
4271 IS_ACTIVE (reg_info[mcnt]) = 0;
4272 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
4273 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
4276 /* We move `string1' into `string2' if the latter's empty -- but not if
4277 `string1' is null. */
4278 if (size2 == 0 && string1 != NULL)
4285 end1 = string1 + size1;
4286 end2 = string2 + size2;
4288 /* Compute where to stop matching, within the two strings. */
4291 end_match_1 = string1 + stop;
4292 end_match_2 = string2;
4297 end_match_2 = string2 + stop - size1;
4300 /* `p' scans through the pattern as `d' scans through the data.
4301 `dend' is the end of the input string that `d' points within. `d'
4302 is advanced into the following input string whenever necessary, but
4303 this happens before fetching; therefore, at the beginning of the
4304 loop, `d' can be pointing at the end of a string, but it cannot
4306 if (size1 > 0 && pos <= size1)
4313 d = string2 + pos - size1;
4317 DEBUG_PRINT1 ("The compiled pattern is: ");
4318 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
4319 DEBUG_PRINT1 ("The string to match is: `");
4320 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
4321 DEBUG_PRINT1 ("'\n");
4323 /* This loops over pattern commands. It exits by returning from the
4324 function if the match is complete, or it drops through if the match
4325 fails at this starting point in the input data. */
4328 DEBUG_PRINT2 ("\n0x%x: ", p);
4331 { /* End of pattern means we might have succeeded. */
4332 DEBUG_PRINT1 ("end of pattern ... ");
4334 /* If we haven't matched the entire string, and we want the
4335 longest match, try backtracking. */
4336 if (d != end_match_2)
4338 /* 1 if this match ends in the same string (string1 or string2)
4339 as the best previous match. */
4340 boolean same_str_p = (FIRST_STRING_P (match_end)
4341 == MATCHING_IN_FIRST_STRING);
4342 /* 1 if this match is the best seen so far. */
4343 boolean best_match_p;
4345 /* AIX compiler got confused when this was combined
4346 with the previous declaration. */
4348 best_match_p = d > match_end;
4350 best_match_p = !MATCHING_IN_FIRST_STRING;
4352 DEBUG_PRINT1 ("backtracking.\n");
4354 if (!FAIL_STACK_EMPTY ())
4355 { /* More failure points to try. */
4357 /* If exceeds best match so far, save it. */
4358 if (!best_regs_set || best_match_p)
4360 best_regs_set = true;
4363 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
4365 for (mcnt = 1; mcnt < num_regs; mcnt++)
4367 best_regstart[mcnt] = regstart[mcnt];
4368 best_regend[mcnt] = regend[mcnt];
4374 /* If no failure points, don't restore garbage. And if
4375 last match is real best match, don't restore second
4377 else if (best_regs_set && !best_match_p)
4380 /* Restore best match. It may happen that `dend ==
4381 end_match_1' while the restored d is in string2.
4382 For example, the pattern `x.*y.*z' against the
4383 strings `x-' and `y-z-', if the two strings are
4384 not consecutive in memory. */
4385 DEBUG_PRINT1 ("Restoring best registers.\n");
4388 dend = ((d >= string1 && d <= end1)
4389 ? end_match_1 : end_match_2);
4391 for (mcnt = 1; mcnt < num_regs; mcnt++)
4393 regstart[mcnt] = best_regstart[mcnt];
4394 regend[mcnt] = best_regend[mcnt];
4397 } /* d != end_match_2 */
4400 DEBUG_PRINT1 ("Accepting match.\n");
4402 /* If caller wants register contents data back, do it. */
4403 if (regs && !bufp->no_sub)
4405 /* Have the register data arrays been allocated? */
4406 if (bufp->regs_allocated == REGS_UNALLOCATED)
4407 { /* No. So allocate them with malloc. We need one
4408 extra element beyond `num_regs' for the `-1' marker
4410 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
4411 regs->start = TALLOC (regs->num_regs, regoff_t);
4412 regs->end = TALLOC (regs->num_regs, regoff_t);
4413 if (regs->start == NULL || regs->end == NULL)
4418 bufp->regs_allocated = REGS_REALLOCATE;
4420 else if (bufp->regs_allocated == REGS_REALLOCATE)
4421 { /* Yes. If we need more elements than were already
4422 allocated, reallocate them. If we need fewer, just
4424 if (regs->num_regs < num_regs + 1)
4426 regs->num_regs = num_regs + 1;
4427 RETALLOC (regs->start, regs->num_regs, regoff_t);
4428 RETALLOC (regs->end, regs->num_regs, regoff_t);
4429 if (regs->start == NULL || regs->end == NULL)
4438 /* These braces fend off a "empty body in an else-statement"
4439 warning under GCC when assert expands to nothing. */
4440 assert (bufp->regs_allocated == REGS_FIXED);
4443 /* Convert the pointer data in `regstart' and `regend' to
4444 indices. Register zero has to be set differently,
4445 since we haven't kept track of any info for it. */
4446 if (regs->num_regs > 0)
4448 regs->start[0] = pos;
4449 regs->end[0] = (MATCHING_IN_FIRST_STRING
4450 ? ((regoff_t) (d - string1))
4451 : ((regoff_t) (d - string2 + size1)));
4454 /* Go through the first `min (num_regs, regs->num_regs)'
4455 registers, since that is all we initialized. */
4456 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
4458 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
4459 regs->start[mcnt] = regs->end[mcnt] = -1;
4463 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
4465 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
4469 /* If the regs structure we return has more elements than
4470 were in the pattern, set the extra elements to -1. If
4471 we (re)allocated the registers, this is the case,
4472 because we always allocate enough to have at least one
4474 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
4475 regs->start[mcnt] = regs->end[mcnt] = -1;
4476 } /* regs && !bufp->no_sub */
4478 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
4479 nfailure_points_pushed, nfailure_points_popped,
4480 nfailure_points_pushed - nfailure_points_popped);
4481 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
4483 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
4487 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
4493 /* Otherwise match next pattern command. */
4494 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4496 /* Ignore these. Used to ignore the n of succeed_n's which
4497 currently have n == 0. */
4499 DEBUG_PRINT1 ("EXECUTING no_op.\n");
4503 DEBUG_PRINT1 ("EXECUTING succeed.\n");
4506 /* Match the next n pattern characters exactly. The following
4507 byte in the pattern defines n, and the n bytes after that
4508 are the characters to match. */
4511 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
4513 /* This is written out as an if-else so we don't waste time
4514 testing `translate' inside the loop. */
4520 if ((unsigned char) RE_TRANSLATE (translate, (unsigned char) *d++)
4521 != (unsigned char) *p++)
4531 if (*d++ != (char) *p++) goto fail;
4535 SET_REGS_MATCHED ();
4539 /* Match any character except possibly a newline or a null. */
4541 DEBUG_PRINT1 ("EXECUTING anychar.\n");
4545 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
4546 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
4549 SET_REGS_MATCHED ();
4550 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
4551 d += multibyte ? MULTIBYTE_FORM_LENGTH (d, dend - d) : 1;
4558 register unsigned int c;
4559 boolean not = (re_opcode_t) *(p - 1) == charset_not;
4562 /* Start of actual range_table, or end of bitmap if there is no
4564 unsigned char *range_table;
4566 /* Nonzero if there is range table. */
4567 int range_table_exists;
4569 /* Number of ranges of range table. Not in bytes. */
4572 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4575 c = (unsigned char) *d;
4577 range_table = CHARSET_RANGE_TABLE (&p[-1]); /* Past the bitmap. */
4578 range_table_exists = CHARSET_RANGE_TABLE_EXISTS_P (&p[-1]);
4579 if (range_table_exists)
4580 EXTRACT_NUMBER_AND_INCR (count, range_table);
4584 if (multibyte && BASE_LEADING_CODE_P (c))
4585 c = STRING_CHAR_AND_LENGTH (d, dend - d, len);
4587 if (SINGLE_BYTE_CHAR_P (c))
4588 { /* Lookup bitmap. */
4589 c = TRANSLATE (c); /* The character to match. */
4592 /* Cast to `unsigned' instead of `unsigned char' in
4593 case the bit list is a full 32 bytes long. */
4594 if (c < (unsigned) (CHARSET_BITMAP_SIZE (&p[-1]) * BYTEWIDTH)
4595 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4598 else if (range_table_exists)
4599 CHARSET_LOOKUP_RANGE_TABLE_RAW (not, c, range_table, count);
4601 p = CHARSET_RANGE_TABLE_END (range_table, count);
4603 if (!not) goto fail;
4605 SET_REGS_MATCHED ();
4611 /* The beginning of a group is represented by start_memory.
4612 The arguments are the register number in the next byte, and the
4613 number of groups inner to this one in the next. The text
4614 matched within the group is recorded (in the internal
4615 registers data structure) under the register number. */
4617 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4619 /* Find out if this group can match the empty string. */
4620 p1 = p; /* To send to group_match_null_string_p. */
4622 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4623 REG_MATCH_NULL_STRING_P (reg_info[*p])
4624 = group_match_null_string_p (&p1, pend, reg_info);
4626 /* Save the position in the string where we were the last time
4627 we were at this open-group operator in case the group is
4628 operated upon by a repetition operator, e.g., with `(a*)*b'
4629 against `ab'; then we want to ignore where we are now in
4630 the string in case this attempt to match fails. */
4631 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4632 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4634 DEBUG_PRINT2 (" old_regstart: %d\n",
4635 POINTER_TO_OFFSET (old_regstart[*p]));
4638 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4640 IS_ACTIVE (reg_info[*p]) = 1;
4641 MATCHED_SOMETHING (reg_info[*p]) = 0;
4643 /* Clear this whenever we change the register activity status. */
4644 set_regs_matched_done = 0;
4646 /* This is the new highest active register. */
4647 highest_active_reg = *p;
4649 /* If nothing was active before, this is the new lowest active
4651 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4652 lowest_active_reg = *p;
4654 /* Move past the register number and inner group count. */
4656 just_past_start_mem = p;
4661 /* The stop_memory opcode represents the end of a group. Its
4662 arguments are the same as start_memory's: the register
4663 number, and the number of inner groups. */
4665 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4667 /* We need to save the string position the last time we were at
4668 this close-group operator in case the group is operated
4669 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4670 against `aba'; then we want to ignore where we are now in
4671 the string in case this attempt to match fails. */
4672 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4673 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4675 DEBUG_PRINT2 (" old_regend: %d\n",
4676 POINTER_TO_OFFSET (old_regend[*p]));
4679 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4681 /* This register isn't active anymore. */
4682 IS_ACTIVE (reg_info[*p]) = 0;
4684 /* Clear this whenever we change the register activity status. */
4685 set_regs_matched_done = 0;
4687 /* If this was the only register active, nothing is active
4689 if (lowest_active_reg == highest_active_reg)
4691 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4692 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4695 { /* We must scan for the new highest active register, since
4696 it isn't necessarily one less than now: consider
4697 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4698 new highest active register is 1. */
4699 unsigned char r = *p - 1;
4700 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4703 /* If we end up at register zero, that means that we saved
4704 the registers as the result of an `on_failure_jump', not
4705 a `start_memory', and we jumped to past the innermost
4706 `stop_memory'. For example, in ((.)*) we save
4707 registers 1 and 2 as a result of the *, but when we pop
4708 back to the second ), we are at the stop_memory 1.
4709 Thus, nothing is active. */
4712 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4713 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4716 highest_active_reg = r;
4719 /* If just failed to match something this time around with a
4720 group that's operated on by a repetition operator, try to
4721 force exit from the ``loop'', and restore the register
4722 information for this group that we had before trying this
4724 if ((!MATCHED_SOMETHING (reg_info[*p])
4725 || just_past_start_mem == p - 1)
4728 boolean is_a_jump_n = false;
4732 switch ((re_opcode_t) *p1++)
4736 case pop_failure_jump:
4737 case maybe_pop_jump:
4739 case dummy_failure_jump:
4740 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4750 /* If the next operation is a jump backwards in the pattern
4751 to an on_failure_jump right before the start_memory
4752 corresponding to this stop_memory, exit from the loop
4753 by forcing a failure after pushing on the stack the
4754 on_failure_jump's jump in the pattern, and d. */
4755 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4756 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4758 /* If this group ever matched anything, then restore
4759 what its registers were before trying this last
4760 failed match, e.g., with `(a*)*b' against `ab' for
4761 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4762 against `aba' for regend[3].
4764 Also restore the registers for inner groups for,
4765 e.g., `((a*)(b*))*' against `aba' (register 3 would
4766 otherwise get trashed). */
4768 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4772 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4774 /* Restore this and inner groups' (if any) registers. */
4775 for (r = *p; r < *p + *(p + 1); r++)
4777 regstart[r] = old_regstart[r];
4779 /* xx why this test? */
4780 if (old_regend[r] >= regstart[r])
4781 regend[r] = old_regend[r];
4785 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4786 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4792 /* Move past the register number and the inner group count. */
4797 /* \<digit> has been turned into a `duplicate' command which is
4798 followed by the numeric value of <digit> as the register number. */
4801 register const char *d2, *dend2;
4802 int regno = *p++; /* Get which register to match against. */
4803 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4805 /* Can't back reference a group which we've never matched. */
4806 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4809 /* Where in input to try to start matching. */
4810 d2 = regstart[regno];
4812 /* Where to stop matching; if both the place to start and
4813 the place to stop matching are in the same string, then
4814 set to the place to stop, otherwise, for now have to use
4815 the end of the first string. */
4817 dend2 = ((FIRST_STRING_P (regstart[regno])
4818 == FIRST_STRING_P (regend[regno]))
4819 ? regend[regno] : end_match_1);
4822 /* If necessary, advance to next segment in register
4826 if (dend2 == end_match_2) break;
4827 if (dend2 == regend[regno]) break;
4829 /* End of string1 => advance to string2. */
4831 dend2 = regend[regno];
4833 /* At end of register contents => success */
4834 if (d2 == dend2) break;
4836 /* If necessary, advance to next segment in data. */
4839 /* How many characters left in this segment to match. */
4842 /* Want how many consecutive characters we can match in
4843 one shot, so, if necessary, adjust the count. */
4844 if (mcnt > dend2 - d2)
4847 /* Compare that many; failure if mismatch, else move
4850 ? bcmp_translate (d, d2, mcnt, translate)
4851 : bcmp (d, d2, mcnt))
4853 d += mcnt, d2 += mcnt;
4855 /* Do this because we've match some characters. */
4856 SET_REGS_MATCHED ();
4862 /* begline matches the empty string at the beginning of the string
4863 (unless `not_bol' is set in `bufp'), and, if
4864 `newline_anchor' is set, after newlines. */
4866 DEBUG_PRINT1 ("EXECUTING begline.\n");
4868 if (AT_STRINGS_BEG (d))
4870 if (!bufp->not_bol) break;
4872 else if (d[-1] == '\n' && bufp->newline_anchor)
4876 /* In all other cases, we fail. */
4880 /* endline is the dual of begline. */
4882 DEBUG_PRINT1 ("EXECUTING endline.\n");
4884 if (AT_STRINGS_END (d))
4886 if (!bufp->not_eol) break;
4889 /* We have to ``prefetch'' the next character. */
4890 else if ((d == end1 ? *string2 : *d) == '\n'
4891 && bufp->newline_anchor)
4898 /* Match at the very beginning of the data. */
4900 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4901 if (AT_STRINGS_BEG (d))
4906 /* Match at the very end of the data. */
4908 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4909 if (AT_STRINGS_END (d))
4914 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4915 pushes NULL as the value for the string on the stack. Then
4916 `pop_failure_point' will keep the current value for the
4917 string, instead of restoring it. To see why, consider
4918 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4919 then the . fails against the \n. But the next thing we want
4920 to do is match the \n against the \n; if we restored the
4921 string value, we would be back at the foo.
4923 Because this is used only in specific cases, we don't need to
4924 check all the things that `on_failure_jump' does, to make
4925 sure the right things get saved on the stack. Hence we don't
4926 share its code. The only reason to push anything on the
4927 stack at all is that otherwise we would have to change
4928 `anychar's code to do something besides goto fail in this
4929 case; that seems worse than this. */
4930 case on_failure_keep_string_jump:
4931 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4933 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4934 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4936 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4940 /* Uses of on_failure_jump:
4942 Each alternative starts with an on_failure_jump that points
4943 to the beginning of the next alternative. Each alternative
4944 except the last ends with a jump that in effect jumps past
4945 the rest of the alternatives. (They really jump to the
4946 ending jump of the following alternative, because tensioning
4947 these jumps is a hassle.)
4949 Repeats start with an on_failure_jump that points past both
4950 the repetition text and either the following jump or
4951 pop_failure_jump back to this on_failure_jump. */
4952 case on_failure_jump:
4954 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4956 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4957 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4959 /* If this on_failure_jump comes right before a group (i.e.,
4960 the original * applied to a group), save the information
4961 for that group and all inner ones, so that if we fail back
4962 to this point, the group's information will be correct.
4963 For example, in \(a*\)*\1, we need the preceding group,
4964 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4966 /* We can't use `p' to check ahead because we push
4967 a failure point to `p + mcnt' after we do this. */
4970 /* We need to skip no_op's before we look for the
4971 start_memory in case this on_failure_jump is happening as
4972 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4974 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4977 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4979 /* We have a new highest active register now. This will
4980 get reset at the start_memory we are about to get to,
4981 but we will have saved all the registers relevant to
4982 this repetition op, as described above. */
4983 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4984 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4985 lowest_active_reg = *(p1 + 1);
4988 DEBUG_PRINT1 (":\n");
4989 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4993 /* A smart repeat ends with `maybe_pop_jump'.
4994 We change it to either `pop_failure_jump' or `jump'. */
4995 case maybe_pop_jump:
4996 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4997 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4999 register unsigned char *p2 = p;
5001 /* Compare the beginning of the repeat with what in the
5002 pattern follows its end. If we can establish that there
5003 is nothing that they would both match, i.e., that we
5004 would have to backtrack because of (as in, e.g., `a*a')
5005 then we can change to pop_failure_jump, because we'll
5006 never have to backtrack.
5008 This is not true in the case of alternatives: in
5009 `(a|ab)*' we do need to backtrack to the `ab' alternative
5010 (e.g., if the string was `ab'). But instead of trying to
5011 detect that here, the alternative has put on a dummy
5012 failure point which is what we will end up popping. */
5014 /* Skip over open/close-group commands.
5015 If what follows this loop is a ...+ construct,
5016 look at what begins its body, since we will have to
5017 match at least one of that. */
5021 && ((re_opcode_t) *p2 == stop_memory
5022 || (re_opcode_t) *p2 == start_memory))
5024 else if (p2 + 6 < pend
5025 && (re_opcode_t) *p2 == dummy_failure_jump)
5032 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
5033 to the `maybe_finalize_jump' of this case. Examine what
5036 /* If we're at the end of the pattern, we can change. */
5039 /* Consider what happens when matching ":\(.*\)"
5040 against ":/". I don't really understand this code
5042 p[-3] = (unsigned char) pop_failure_jump;
5044 (" End of pattern: change to `pop_failure_jump'.\n");
5047 else if ((re_opcode_t) *p2 == exactn
5048 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
5050 register unsigned int c
5051 = *p2 == (unsigned char) endline ? '\n' : p2[2];
5053 if ((re_opcode_t) p1[3] == exactn)
5055 if (!(multibyte /* && (c != '\n') */
5056 && BASE_LEADING_CODE_P (c))
5058 : (STRING_CHAR (&p2[2], pend - &p2[2])
5059 != STRING_CHAR (&p1[5], pend - &p1[5])))
5061 p[-3] = (unsigned char) pop_failure_jump;
5062 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
5067 else if ((re_opcode_t) p1[3] == charset
5068 || (re_opcode_t) p1[3] == charset_not)
5070 int not = (re_opcode_t) p1[3] == charset_not;
5072 if (multibyte /* && (c != '\n') */
5073 && BASE_LEADING_CODE_P (c))
5074 c = STRING_CHAR (&p2[2], pend - &p2[2]);
5076 /* Test if C is listed in charset (or charset_not)
5078 if (SINGLE_BYTE_CHAR_P (c))
5080 if (c < CHARSET_BITMAP_SIZE (&p1[3]) * BYTEWIDTH
5081 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
5084 else if (CHARSET_RANGE_TABLE_EXISTS_P (&p1[3]))
5085 CHARSET_LOOKUP_RANGE_TABLE (not, c, &p1[3]);
5087 /* `not' is equal to 1 if c would match, which means
5088 that we can't change to pop_failure_jump. */
5091 p[-3] = (unsigned char) pop_failure_jump;
5092 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5096 else if ((re_opcode_t) *p2 == charset)
5098 if ((re_opcode_t) p1[3] == exactn)
5100 register unsigned int c = p1[5];
5103 if (multibyte && BASE_LEADING_CODE_P (c))
5104 c = STRING_CHAR (&p1[5], pend - &p1[5]);
5106 /* Test if C is listed in charset at `p2'. */
5107 if (SINGLE_BYTE_CHAR_P (c))
5109 if (c < CHARSET_BITMAP_SIZE (p2) * BYTEWIDTH
5110 && (p2[2 + c / BYTEWIDTH]
5111 & (1 << (c % BYTEWIDTH))))
5114 else if (CHARSET_RANGE_TABLE_EXISTS_P (p2))
5115 CHARSET_LOOKUP_RANGE_TABLE (not, c, p2);
5119 p[-3] = (unsigned char) pop_failure_jump;
5120 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5124 /* It is hard to list up all the character in charset
5125 P2 if it includes multibyte character. Give up in
5127 else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
5129 /* Now, we are sure that P2 has no range table.
5130 So, for the size of bitmap in P2, `p2[1]' is
5131 enough. But P1 may have range table, so the
5132 size of bitmap table of P1 is extracted by
5133 using macro `CHARSET_BITMAP_SIZE'.
5135 Since we know that all the character listed in
5136 P2 is ASCII, it is enough to test only bitmap
5139 if ((re_opcode_t) p1[3] == charset_not)
5142 /* We win if the charset_not inside the loop lists
5143 every character listed in the charset after. */
5144 for (idx = 0; idx < (int) p2[1]; idx++)
5145 if (! (p2[2 + idx] == 0
5146 || (idx < CHARSET_BITMAP_SIZE (&p1[3])
5147 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
5152 p[-3] = (unsigned char) pop_failure_jump;
5153 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5156 else if ((re_opcode_t) p1[3] == charset)
5159 /* We win if the charset inside the loop
5160 has no overlap with the one after the loop. */
5163 && idx < CHARSET_BITMAP_SIZE (&p1[3]));
5165 if ((p2[2 + idx] & p1[5 + idx]) != 0)
5169 || idx == CHARSET_BITMAP_SIZE (&p1[3]))
5171 p[-3] = (unsigned char) pop_failure_jump;
5172 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
5178 p -= 2; /* Point at relative address again. */
5179 if ((re_opcode_t) p[-1] != pop_failure_jump)
5181 p[-1] = (unsigned char) jump;
5182 DEBUG_PRINT1 (" Match => jump.\n");
5183 goto unconditional_jump;
5185 /* Note fall through. */
5188 /* The end of a simple repeat has a pop_failure_jump back to
5189 its matching on_failure_jump, where the latter will push a
5190 failure point. The pop_failure_jump takes off failure
5191 points put on by this pop_failure_jump's matching
5192 on_failure_jump; we got through the pattern to here from the
5193 matching on_failure_jump, so didn't fail. */
5194 case pop_failure_jump:
5196 /* We need to pass separate storage for the lowest and
5197 highest registers, even though we don't care about the
5198 actual values. Otherwise, we will restore only one
5199 register from the stack, since lowest will == highest in
5200 `pop_failure_point'. */
5201 unsigned dummy_low_reg, dummy_high_reg;
5202 unsigned char *pdummy;
5205 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
5206 POP_FAILURE_POINT (sdummy, pdummy,
5207 dummy_low_reg, dummy_high_reg,
5208 reg_dummy, reg_dummy, reg_info_dummy);
5210 /* Note fall through. */
5213 /* Unconditionally jump (without popping any failure points). */
5216 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
5217 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
5218 p += mcnt; /* Do the jump. */
5219 DEBUG_PRINT2 ("(to 0x%x).\n", p);
5223 /* We need this opcode so we can detect where alternatives end
5224 in `group_match_null_string_p' et al. */
5226 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
5227 goto unconditional_jump;
5230 /* Normally, the on_failure_jump pushes a failure point, which
5231 then gets popped at pop_failure_jump. We will end up at
5232 pop_failure_jump, also, and with a pattern of, say, `a+', we
5233 are skipping over the on_failure_jump, so we have to push
5234 something meaningless for pop_failure_jump to pop. */
5235 case dummy_failure_jump:
5236 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
5237 /* It doesn't matter what we push for the string here. What
5238 the code at `fail' tests is the value for the pattern. */
5239 PUSH_FAILURE_POINT (0, 0, -2);
5240 goto unconditional_jump;
5243 /* At the end of an alternative, we need to push a dummy failure
5244 point in case we are followed by a `pop_failure_jump', because
5245 we don't want the failure point for the alternative to be
5246 popped. For example, matching `(a|ab)*' against `aab'
5247 requires that we match the `ab' alternative. */
5248 case push_dummy_failure:
5249 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
5250 /* See comments just above at `dummy_failure_jump' about the
5252 PUSH_FAILURE_POINT (0, 0, -2);
5255 /* Have to succeed matching what follows at least n times.
5256 After that, handle like `on_failure_jump'. */
5258 EXTRACT_NUMBER (mcnt, p + 2);
5259 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
5262 /* Originally, this is how many times we HAVE to succeed. */
5267 STORE_NUMBER_AND_INCR (p, mcnt);
5268 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
5272 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
5273 p[2] = (unsigned char) no_op;
5274 p[3] = (unsigned char) no_op;
5280 EXTRACT_NUMBER (mcnt, p + 2);
5281 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
5283 /* Originally, this is how many times we CAN jump. */
5287 STORE_NUMBER (p + 2, mcnt);
5288 goto unconditional_jump;
5290 /* If don't have to jump any more, skip over the rest of command. */
5297 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
5299 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5301 EXTRACT_NUMBER_AND_INCR (mcnt, p);
5302 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
5303 STORE_NUMBER (p1, mcnt);
5308 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
5310 /* We SUCCEED in one of the following cases: */
5312 /* Case 1: D is at the beginning or the end of string. */
5313 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5317 /* C1 is the character before D, S1 is the syntax of C1, C2
5318 is the character at D, and S2 is the syntax of C2. */
5320 int pos1 = PTR_TO_OFFSET (d - 1);
5322 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5323 GET_CHAR_AFTER_2 (c2, d, string1, end1, string2, end2);
5325 UPDATE_SYNTAX_TABLE (pos1 ? pos1 : 1);
5329 UPDATE_SYNTAX_TABLE_FORWARD (pos1 + 1);
5333 if (/* Case 2: Only one of S1 and S2 is Sword. */
5334 ((s1 == Sword) != (s2 == Sword))
5335 /* Case 3: Both of S1 and S2 are Sword, and macro
5336 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5337 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5343 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
5345 /* We FAIL in one of the following cases: */
5347 /* Case 1: D is at the beginning or the end of string. */
5348 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
5352 /* C1 is the character before D, S1 is the syntax of C1, C2
5353 is the character at D, and S2 is the syntax of C2. */
5355 int pos1 = PTR_TO_OFFSET (d - 1);
5357 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5358 GET_CHAR_AFTER_2 (c2, d, string1, end1, string2, end2);
5360 UPDATE_SYNTAX_TABLE (pos1);
5364 UPDATE_SYNTAX_TABLE_FORWARD (pos1 + 1);
5368 if (/* Case 2: Only one of S1 and S2 is Sword. */
5369 ((s1 == Sword) != (s2 == Sword))
5370 /* Case 3: Both of S1 and S2 are Sword, and macro
5371 WORD_BOUNDARY_P (C1, C2) returns nonzero. */
5372 || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
5378 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
5380 /* We FAIL in one of the following cases: */
5382 /* Case 1: D is at the end of string. */
5383 if (AT_STRINGS_END (d))
5387 /* C1 is the character before D, S1 is the syntax of C1, C2
5388 is the character at D, and S2 is the syntax of C2. */
5390 int pos1 = PTR_TO_OFFSET (d);
5392 GET_CHAR_AFTER_2 (c2, d, string1, end1, string2, end2);
5394 UPDATE_SYNTAX_TABLE (pos1);
5398 /* Case 2: S2 is not Sword. */
5402 /* Case 3: D is not at the beginning of string ... */
5403 if (!AT_STRINGS_BEG (d))
5405 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5407 UPDATE_SYNTAX_TABLE_BACKWARD (pos1 - 1);
5411 /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
5413 if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
5420 DEBUG_PRINT1 ("EXECUTING wordend.\n");
5422 /* We FAIL in one of the following cases: */
5424 /* Case 1: D is at the beginning of string. */
5425 if (AT_STRINGS_BEG (d))
5429 /* C1 is the character before D, S1 is the syntax of C1, C2
5430 is the character at D, and S2 is the syntax of C2. */
5433 GET_CHAR_BEFORE_2 (c1, d, string1, end1, string2, end2);
5436 /* Case 2: S1 is not Sword. */
5440 /* Case 3: D is not at the end of string ... */
5441 if (!AT_STRINGS_END (d))
5443 GET_CHAR_AFTER_2 (c2, d, string1, end1, string2, end2);
5446 /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
5448 if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
5456 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
5457 if (PTR_CHAR_POS ((unsigned char *) d) >= PT)
5462 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
5463 if (PTR_CHAR_POS ((unsigned char *) d) != PT)
5468 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
5469 if (PTR_CHAR_POS ((unsigned char *) d) <= PT)
5474 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
5479 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
5485 int pos1 = PTR_TO_OFFSET (d);
5486 UPDATE_SYNTAX_TABLE (pos1);
5493 /* we must concern about multibyte form, ... */
5494 c = STRING_CHAR_AND_LENGTH (d, dend - d, len);
5496 /* everything should be handled as ASCII, even though it
5497 looks like multibyte form. */
5500 if (SYNTAX (c) != (enum syntaxcode) mcnt)
5504 SET_REGS_MATCHED ();
5508 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
5510 goto matchnotsyntax;
5513 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
5519 int pos1 = PTR_TO_OFFSET (d);
5520 UPDATE_SYNTAX_TABLE (pos1);
5527 c = STRING_CHAR_AND_LENGTH (d, dend - d, len);
5531 if (SYNTAX (c) == (enum syntaxcode) mcnt)
5535 SET_REGS_MATCHED ();
5539 DEBUG_PRINT2 ("EXECUTING categoryspec %d.\n", *p);
5546 c = STRING_CHAR_AND_LENGTH (d, dend - d, len);
5550 if (!CHAR_HAS_CATEGORY (c, mcnt))
5554 SET_REGS_MATCHED ();
5557 case notcategoryspec:
5558 DEBUG_PRINT2 ("EXECUTING notcategoryspec %d.\n", *p);
5565 c = STRING_CHAR_AND_LENGTH (d, dend - d, len);
5569 if (CHAR_HAS_CATEGORY (c, mcnt))
5573 SET_REGS_MATCHED ();
5576 #else /* not emacs */
5578 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
5580 if (!WORDCHAR_P (d))
5582 SET_REGS_MATCHED ();
5587 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
5591 SET_REGS_MATCHED ();
5594 #endif /* not emacs */
5599 continue; /* Successfully executed one pattern command; keep going. */
5602 /* We goto here if a matching operation fails. */
5604 if (!FAIL_STACK_EMPTY ())
5605 { /* A restart point is known. Restore to that state. */
5606 DEBUG_PRINT1 ("\nFAIL:\n");
5607 POP_FAILURE_POINT (d, p,
5608 lowest_active_reg, highest_active_reg,
5609 regstart, regend, reg_info);
5611 /* If this failure point is a dummy, try the next one. */
5615 /* If we failed to the end of the pattern, don't examine *p. */
5619 boolean is_a_jump_n = false;
5621 /* If failed to a backwards jump that's part of a repetition
5622 loop, need to pop this failure point and use the next one. */
5623 switch ((re_opcode_t) *p)
5627 case maybe_pop_jump:
5628 case pop_failure_jump:
5631 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5634 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
5636 && (re_opcode_t) *p1 == on_failure_jump))
5644 if (d >= string1 && d <= end1)
5648 break; /* Matching at this starting point really fails. */
5652 goto restore_best_regs;
5656 return -1; /* Failure to match. */
5659 /* Subroutine definitions for re_match_2. */
5662 /* We are passed P pointing to a register number after a start_memory.
5664 Return true if the pattern up to the corresponding stop_memory can
5665 match the empty string, and false otherwise.
5667 If we find the matching stop_memory, sets P to point to one past its number.
5668 Otherwise, sets P to an undefined byte less than or equal to END.
5670 We don't handle duplicates properly (yet). */
5673 group_match_null_string_p (p, end, reg_info)
5674 unsigned char **p, *end;
5675 register_info_type *reg_info;
5678 /* Point to after the args to the start_memory. */
5679 unsigned char *p1 = *p + 2;
5683 /* Skip over opcodes that can match nothing, and return true or
5684 false, as appropriate, when we get to one that can't, or to the
5685 matching stop_memory. */
5687 switch ((re_opcode_t) *p1)
5689 /* Could be either a loop or a series of alternatives. */
5690 case on_failure_jump:
5692 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5694 /* If the next operation is not a jump backwards in the
5699 /* Go through the on_failure_jumps of the alternatives,
5700 seeing if any of the alternatives cannot match nothing.
5701 The last alternative starts with only a jump,
5702 whereas the rest start with on_failure_jump and end
5703 with a jump, e.g., here is the pattern for `a|b|c':
5705 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
5706 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
5709 So, we have to first go through the first (n-1)
5710 alternatives and then deal with the last one separately. */
5713 /* Deal with the first (n-1) alternatives, which start
5714 with an on_failure_jump (see above) that jumps to right
5715 past a jump_past_alt. */
5717 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
5719 /* `mcnt' holds how many bytes long the alternative
5720 is, including the ending `jump_past_alt' and
5723 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
5727 /* Move to right after this alternative, including the
5731 /* Break if it's the beginning of an n-th alternative
5732 that doesn't begin with an on_failure_jump. */
5733 if ((re_opcode_t) *p1 != on_failure_jump)
5736 /* Still have to check that it's not an n-th
5737 alternative that starts with an on_failure_jump. */
5739 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5740 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
5742 /* Get to the beginning of the n-th alternative. */
5748 /* Deal with the last alternative: go back and get number
5749 of the `jump_past_alt' just before it. `mcnt' contains
5750 the length of the alternative. */
5751 EXTRACT_NUMBER (mcnt, p1 - 2);
5753 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
5756 p1 += mcnt; /* Get past the n-th alternative. */
5762 assert (p1[1] == **p);
5768 if (!common_op_match_null_string_p (&p1, end, reg_info))
5771 } /* while p1 < end */
5774 } /* group_match_null_string_p */
5777 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
5778 It expects P to be the first byte of a single alternative and END one
5779 byte past the last. The alternative can contain groups. */
5782 alt_match_null_string_p (p, end, reg_info)
5783 unsigned char *p, *end;
5784 register_info_type *reg_info;
5787 unsigned char *p1 = p;
5791 /* Skip over opcodes that can match nothing, and break when we get
5792 to one that can't. */
5794 switch ((re_opcode_t) *p1)
5797 case on_failure_jump:
5799 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5804 if (!common_op_match_null_string_p (&p1, end, reg_info))
5807 } /* while p1 < end */
5810 } /* alt_match_null_string_p */
5813 /* Deals with the ops common to group_match_null_string_p and
5814 alt_match_null_string_p.
5816 Sets P to one after the op and its arguments, if any. */
5819 common_op_match_null_string_p (p, end, reg_info)
5820 unsigned char **p, *end;
5821 register_info_type *reg_info;
5826 unsigned char *p1 = *p;
5828 switch ((re_opcode_t) *p1++)
5848 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5849 ret = group_match_null_string_p (&p1, end, reg_info);
5851 /* Have to set this here in case we're checking a group which
5852 contains a group and a back reference to it. */
5854 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5855 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5861 /* If this is an optimized succeed_n for zero times, make the jump. */
5863 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5871 /* Get to the number of times to succeed. */
5873 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5878 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5886 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5894 /* All other opcodes mean we cannot match the empty string. */
5900 } /* common_op_match_null_string_p */
5903 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5904 bytes; nonzero otherwise. */
5907 bcmp_translate (s1, s2, len, translate)
5908 unsigned char *s1, *s2;
5910 RE_TRANSLATE_TYPE translate;
5912 register unsigned char *p1 = s1, *p2 = s2;
5915 if (RE_TRANSLATE (translate, *p1++) != RE_TRANSLATE (translate, *p2++))
5922 /* Entry points for GNU code. */
5924 /* re_compile_pattern is the GNU regular expression compiler: it
5925 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5926 Returns 0 if the pattern was valid, otherwise an error string.
5928 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5929 are set in BUFP on entry.
5931 We call regex_compile to do the actual compilation. */
5934 re_compile_pattern (pattern, length, bufp)
5935 const char *pattern;
5937 struct re_pattern_buffer *bufp;
5941 /* GNU code is written to assume at least RE_NREGS registers will be set
5942 (and at least one extra will be -1). */
5943 bufp->regs_allocated = REGS_UNALLOCATED;
5945 /* And GNU code determines whether or not to get register information
5946 by passing null for the REGS argument to re_match, etc., not by
5950 /* Match anchors at newline. */
5951 bufp->newline_anchor = 1;
5953 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5957 return gettext (re_error_msgid[(int) ret]);
5960 /* Entry points compatible with 4.2 BSD regex library. We don't define
5961 them unless specifically requested. */
5963 #if defined (_REGEX_RE_COMP) || defined (_LIBC)
5965 /* BSD has one and only one pattern buffer. */
5966 static struct re_pattern_buffer re_comp_buf;
5970 /* Make these definitions weak in libc, so POSIX programs can redefine
5971 these names if they don't use our functions, and still use
5972 regcomp/regexec below without link errors. */
5982 if (!re_comp_buf.buffer)
5983 return gettext ("No previous regular expression");
5987 if (!re_comp_buf.buffer)
5989 re_comp_buf.buffer = (unsigned char *) malloc (200);
5990 if (re_comp_buf.buffer == NULL)
5991 return gettext (re_error_msgid[(int) REG_ESPACE]);
5992 re_comp_buf.allocated = 200;
5994 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5995 if (re_comp_buf.fastmap == NULL)
5996 return gettext (re_error_msgid[(int) REG_ESPACE]);
5999 /* Since `re_exec' always passes NULL for the `regs' argument, we
6000 don't need to initialize the pattern buffer fields which affect it. */
6002 /* Match anchors at newlines. */
6003 re_comp_buf.newline_anchor = 1;
6005 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
6010 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
6011 return (char *) gettext (re_error_msgid[(int) ret]);
6022 const int len = strlen (s);
6024 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
6026 #endif /* _REGEX_RE_COMP */
6028 /* POSIX.2 functions. Don't define these for Emacs. */
6032 /* regcomp takes a regular expression as a string and compiles it.
6034 PREG is a regex_t *. We do not expect any fields to be initialized,
6035 since POSIX says we shouldn't. Thus, we set
6037 `buffer' to the compiled pattern;
6038 `used' to the length of the compiled pattern;
6039 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
6040 REG_EXTENDED bit in CFLAGS is set; otherwise, to
6041 RE_SYNTAX_POSIX_BASIC;
6042 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
6043 `fastmap' and `fastmap_accurate' to zero;
6044 `re_nsub' to the number of subexpressions in PATTERN.
6046 PATTERN is the address of the pattern string.
6048 CFLAGS is a series of bits which affect compilation.
6050 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
6051 use POSIX basic syntax.
6053 If REG_NEWLINE is set, then . and [^...] don't match newline.
6054 Also, regexec will try a match beginning after every newline.
6056 If REG_ICASE is set, then we considers upper- and lowercase
6057 versions of letters to be equivalent when matching.
6059 If REG_NOSUB is set, then when PREG is passed to regexec, that
6060 routine will report only success or failure, and nothing about the
6063 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
6064 the return codes and their meanings.) */
6067 regcomp (preg, pattern, cflags)
6069 const char *pattern;
6074 = (cflags & REG_EXTENDED) ?
6075 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
6077 /* regex_compile will allocate the space for the compiled pattern. */
6079 preg->allocated = 0;
6082 /* Don't bother to use a fastmap when searching. This simplifies the
6083 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
6084 characters after newlines into the fastmap. This way, we just try
6088 if (cflags & REG_ICASE)
6093 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
6094 * sizeof (*(RE_TRANSLATE_TYPE)0));
6095 if (preg->translate == NULL)
6096 return (int) REG_ESPACE;
6098 /* Map uppercase characters to corresponding lowercase ones. */
6099 for (i = 0; i < CHAR_SET_SIZE; i++)
6100 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
6103 preg->translate = NULL;
6105 /* If REG_NEWLINE is set, newlines are treated differently. */
6106 if (cflags & REG_NEWLINE)
6107 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
6108 syntax &= ~RE_DOT_NEWLINE;
6109 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
6110 /* It also changes the matching behavior. */
6111 preg->newline_anchor = 1;
6114 preg->newline_anchor = 0;
6116 preg->no_sub = !!(cflags & REG_NOSUB);
6118 /* POSIX says a null character in the pattern terminates it, so we
6119 can use strlen here in compiling the pattern. */
6120 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
6122 /* POSIX doesn't distinguish between an unmatched open-group and an
6123 unmatched close-group: both are REG_EPAREN. */
6124 if (ret == REG_ERPAREN) ret = REG_EPAREN;
6130 /* regexec searches for a given pattern, specified by PREG, in the
6133 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
6134 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
6135 least NMATCH elements, and we set them to the offsets of the
6136 corresponding matched substrings.
6138 EFLAGS specifies `execution flags' which affect matching: if
6139 REG_NOTBOL is set, then ^ does not match at the beginning of the
6140 string; if REG_NOTEOL is set, then $ does not match at the end.
6142 We return 0 if we find a match and REG_NOMATCH if not. */
6145 regexec (preg, string, nmatch, pmatch, eflags)
6146 const regex_t *preg;
6149 regmatch_t pmatch[];
6153 struct re_registers regs;
6154 regex_t private_preg;
6155 int len = strlen (string);
6156 boolean want_reg_info = !preg->no_sub && nmatch > 0;
6158 private_preg = *preg;
6160 private_preg.not_bol = !!(eflags & REG_NOTBOL);
6161 private_preg.not_eol = !!(eflags & REG_NOTEOL);
6163 /* The user has told us exactly how many registers to return
6164 information about, via `nmatch'. We have to pass that on to the
6165 matching routines. */
6166 private_preg.regs_allocated = REGS_FIXED;
6170 regs.num_regs = nmatch;
6171 regs.start = TALLOC (nmatch, regoff_t);
6172 regs.end = TALLOC (nmatch, regoff_t);
6173 if (regs.start == NULL || regs.end == NULL)
6174 return (int) REG_NOMATCH;
6177 /* Perform the searching operation. */
6178 ret = re_search (&private_preg, string, len,
6179 /* start: */ 0, /* range: */ len,
6180 want_reg_info ? ®s : (struct re_registers *) 0);
6182 /* Copy the register information to the POSIX structure. */
6189 for (r = 0; r < nmatch; r++)
6191 pmatch[r].rm_so = regs.start[r];
6192 pmatch[r].rm_eo = regs.end[r];
6196 /* If we needed the temporary register info, free the space now. */
6201 /* We want zero return to mean success, unlike `re_search'. */
6202 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
6206 /* Returns a message corresponding to an error code, ERRCODE, returned
6207 from either regcomp or regexec. We don't use PREG here. */
6210 regerror (errcode, preg, errbuf, errbuf_size)
6212 const regex_t *preg;
6220 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
6221 /* Only error codes returned by the rest of the code should be passed
6222 to this routine. If we are given anything else, or if other regex
6223 code generates an invalid error code, then the program has a bug.
6224 Dump core so we can fix it. */
6227 msg = gettext (re_error_msgid[errcode]);
6229 msg_size = strlen (msg) + 1; /* Includes the null. */
6231 if (errbuf_size != 0)
6233 if (msg_size > errbuf_size)
6235 strncpy (errbuf, msg, errbuf_size - 1);
6236 errbuf[errbuf_size - 1] = 0;
6239 strcpy (errbuf, msg);
6246 /* Free dynamically allocated space used by PREG. */
6252 if (preg->buffer != NULL)
6253 free (preg->buffer);
6254 preg->buffer = NULL;
6256 preg->allocated = 0;
6259 if (preg->fastmap != NULL)
6260 free (preg->fastmap);
6261 preg->fastmap = NULL;
6262 preg->fastmap_accurate = 0;
6264 if (preg->translate != NULL)
6265 free (preg->translate);
6266 preg->translate = NULL;
6269 #endif /* not emacs */