1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
30 #if defined (emacs) || defined (CONFIG_BROKETS)
31 /* We use <config.h> instead of "config.h" so that a compilation
32 using -I. -I$srcdir will use ./config.h rather than $srcdir/config.h
33 (which it would do because it found this file in $srcdir). */
40 /* We need this for `regex.h', and perhaps for the Emacs include files. */
41 #include <sys/types.h>
43 /* The `emacs' switch turns on certain matching commands
44 that make sense only in Emacs. */
51 /* Emacs uses `NULL' as a predicate. */
64 /* We used to test for `BSTRING' here, but only GCC and Emacs define
65 `BSTRING', as far as I know, and neither of them use this code. */
66 #if HAVE_STRING_H || STDC_HEADERS
69 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
72 #define bcopy(s, d, n) memcpy ((d), (s), (n))
75 #define bzero(s, n) memset ((s), 0, (n))
81 /* Define the syntax stuff for \<, \>, etc. */
83 /* This must be nonzero for the wordchar and notwordchar pattern
84 commands in re_match_2. */
91 extern char *re_syntax_table;
93 #else /* not SYNTAX_TABLE */
95 /* How many characters in the character set. */
96 #define CHAR_SET_SIZE 256
98 static char re_syntax_table[CHAR_SET_SIZE];
109 bzero (re_syntax_table, sizeof re_syntax_table);
111 for (c = 'a'; c <= 'z'; c++)
112 re_syntax_table[c] = Sword;
114 for (c = 'A'; c <= 'Z'; c++)
115 re_syntax_table[c] = Sword;
117 for (c = '0'; c <= '9'; c++)
118 re_syntax_table[c] = Sword;
120 re_syntax_table['_'] = Sword;
125 #endif /* not SYNTAX_TABLE */
127 #define SYNTAX(c) re_syntax_table[c]
129 #endif /* not emacs */
131 /* Get the interface, including the syntax bits. */
134 /* isalpha etc. are used for the character classes. */
137 /* Jim Meyering writes:
139 "... Some ctype macros are valid only for character codes that
140 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
141 using /bin/cc or gcc but without giving an ansi option). So, all
142 ctype uses should be through macros like ISPRINT... If
143 STDC_HEADERS is defined, then autoconf has verified that the ctype
144 macros don't need to be guarded with references to isascii. ...
145 Defining isascii to 1 should let any compiler worth its salt
146 eliminate the && through constant folding." */
148 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
151 #define ISASCII(c) isascii(c)
155 #define ISBLANK(c) (ISASCII (c) && isblank (c))
157 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
160 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
162 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
165 #define ISPRINT(c) (ISASCII (c) && isprint (c))
166 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
167 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
168 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
169 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
170 #define ISLOWER(c) (ISASCII (c) && islower (c))
171 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
172 #define ISSPACE(c) (ISASCII (c) && isspace (c))
173 #define ISUPPER(c) (ISASCII (c) && isupper (c))
174 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
180 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
181 since ours (we hope) works properly with all combinations of
182 machines, compilers, `char' and `unsigned char' argument types.
183 (Per Bothner suggested the basic approach.) */
184 #undef SIGN_EXTEND_CHAR
186 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
187 #else /* not __STDC__ */
188 /* As in Harbison and Steele. */
189 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
192 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
193 use `alloca' instead of `malloc'. This is because using malloc in
194 re_search* or re_match* could cause memory leaks when C-g is used in
195 Emacs; also, malloc is slower and causes storage fragmentation. On
196 the other hand, malloc is more portable, and easier to debug.
198 Because we sometimes use alloca, some routines have to be macros,
199 not functions -- `alloca'-allocated space disappears at the end of the
200 function it is called in. */
204 #define REGEX_ALLOCATE malloc
205 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
207 #else /* not REGEX_MALLOC */
209 /* Emacs already defines alloca, sometimes. */
212 /* Make alloca work the best possible way. */
214 #define alloca __builtin_alloca
215 #else /* not __GNUC__ */
218 #else /* not __GNUC__ or HAVE_ALLOCA_H */
219 #ifndef _AIX /* Already did AIX, up at the top. */
221 #endif /* not _AIX */
222 #endif /* not HAVE_ALLOCA_H */
223 #endif /* not __GNUC__ */
225 #endif /* not alloca */
227 #define REGEX_ALLOCATE alloca
229 /* Assumes a `char *destination' variable. */
230 #define REGEX_REALLOCATE(source, osize, nsize) \
231 (destination = (char *) alloca (nsize), \
232 bcopy (source, destination, osize), \
235 #endif /* not REGEX_MALLOC */
238 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
239 `string1' or just past its end. This works if PTR is NULL, which is
241 #define FIRST_STRING_P(ptr) \
242 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
244 /* (Re)Allocate N items of type T using malloc, or fail. */
245 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
246 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
247 #define RETALLOC_IF(addr, n, t) \
248 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
249 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
251 #define BYTEWIDTH 8 /* In bits. */
253 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
257 #define MAX(a, b) ((a) > (b) ? (a) : (b))
258 #define MIN(a, b) ((a) < (b) ? (a) : (b))
260 typedef char boolean;
264 static int re_match_2_internal ();
266 /* These are the command codes that appear in compiled regular
267 expressions. Some opcodes are followed by argument bytes. A
268 command code can specify any interpretation whatsoever for its
269 arguments. Zero bytes may appear in the compiled regular expression.
271 The value of `exactn' is needed in search.c (search_buffer) in Emacs.
272 So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of
273 `exactn' we use here must also be 1. */
279 /* Followed by one byte giving n, then by n literal bytes. */
282 /* Matches any (more or less) character. */
285 /* Matches any one char belonging to specified set. First
286 following byte is number of bitmap bytes. Then come bytes
287 for a bitmap saying which chars are in. Bits in each byte
288 are ordered low-bit-first. A character is in the set if its
289 bit is 1. A character too large to have a bit in the map is
290 automatically not in the set. */
293 /* Same parameters as charset, but match any character that is
294 not one of those specified. */
297 /* Start remembering the text that is matched, for storing in a
298 register. Followed by one byte with the register number, in
299 the range 0 to one less than the pattern buffer's re_nsub
300 field. Then followed by one byte with the number of groups
301 inner to this one. (This last has to be part of the
302 start_memory only because we need it in the on_failure_jump
306 /* Stop remembering the text that is matched and store it in a
307 memory register. Followed by one byte with the register
308 number, in the range 0 to one less than `re_nsub' in the
309 pattern buffer, and one byte with the number of inner groups,
310 just like `start_memory'. (We need the number of inner
311 groups here because we don't have any easy way of finding the
312 corresponding start_memory when we're at a stop_memory.) */
315 /* Match a duplicate of something remembered. Followed by one
316 byte containing the register number. */
319 /* Fail unless at beginning of line. */
322 /* Fail unless at end of line. */
325 /* Succeeds if at beginning of buffer (if emacs) or at beginning
326 of string to be matched (if not). */
329 /* Analogously, for end of buffer/string. */
332 /* Followed by two byte relative address to which to jump. */
335 /* Same as jump, but marks the end of an alternative. */
338 /* Followed by two-byte relative address of place to resume at
339 in case of failure. */
342 /* Like on_failure_jump, but pushes a placeholder instead of the
343 current string position when executed. */
344 on_failure_keep_string_jump,
346 /* Throw away latest failure point and then jump to following
347 two-byte relative address. */
350 /* Change to pop_failure_jump if know won't have to backtrack to
351 match; otherwise change to jump. This is used to jump
352 back to the beginning of a repeat. If what follows this jump
353 clearly won't match what the repeat does, such that we can be
354 sure that there is no use backtracking out of repetitions
355 already matched, then we change it to a pop_failure_jump.
356 Followed by two-byte address. */
359 /* Jump to following two-byte address, and push a dummy failure
360 point. This failure point will be thrown away if an attempt
361 is made to use it for a failure. A `+' construct makes this
362 before the first repeat. Also used as an intermediary kind
363 of jump when compiling an alternative. */
366 /* Push a dummy failure point and continue. Used at the end of
370 /* Followed by two-byte relative address and two-byte number n.
371 After matching N times, jump to the address upon failure. */
374 /* Followed by two-byte relative address, and two-byte number n.
375 Jump to the address N times, then fail. */
378 /* Set the following two-byte relative address to the
379 subsequent two-byte number. The address *includes* the two
383 wordchar, /* Matches any word-constituent character. */
384 notwordchar, /* Matches any char that is not a word-constituent. */
386 wordbeg, /* Succeeds if at word beginning. */
387 wordend, /* Succeeds if at word end. */
389 wordbound, /* Succeeds if at a word boundary. */
390 notwordbound /* Succeeds if not at a word boundary. */
393 ,before_dot, /* Succeeds if before point. */
394 at_dot, /* Succeeds if at point. */
395 after_dot, /* Succeeds if after point. */
397 /* Matches any character whose syntax is specified. Followed by
398 a byte which contains a syntax code, e.g., Sword. */
401 /* Matches any character whose syntax is not that specified. */
406 /* Common operations on the compiled pattern. */
408 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
410 #define STORE_NUMBER(destination, number) \
412 (destination)[0] = (number) & 0377; \
413 (destination)[1] = (number) >> 8; \
416 /* Same as STORE_NUMBER, except increment DESTINATION to
417 the byte after where the number is stored. Therefore, DESTINATION
418 must be an lvalue. */
420 #define STORE_NUMBER_AND_INCR(destination, number) \
422 STORE_NUMBER (destination, number); \
423 (destination) += 2; \
426 /* Put into DESTINATION a number stored in two contiguous bytes starting
429 #define EXTRACT_NUMBER(destination, source) \
431 (destination) = *(source) & 0377; \
432 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
437 extract_number (dest, source)
439 unsigned char *source;
441 int temp = SIGN_EXTEND_CHAR (*(source + 1));
442 *dest = *source & 0377;
446 #ifndef EXTRACT_MACROS /* To debug the macros. */
447 #undef EXTRACT_NUMBER
448 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
449 #endif /* not EXTRACT_MACROS */
453 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
454 SOURCE must be an lvalue. */
456 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
458 EXTRACT_NUMBER (destination, source); \
464 extract_number_and_incr (destination, source)
466 unsigned char **source;
468 extract_number (destination, *source);
472 #ifndef EXTRACT_MACROS
473 #undef EXTRACT_NUMBER_AND_INCR
474 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
475 extract_number_and_incr (&dest, &src)
476 #endif /* not EXTRACT_MACROS */
480 /* If DEBUG is defined, Regex prints many voluminous messages about what
481 it is doing (if the variable `debug' is nonzero). If linked with the
482 main program in `iregex.c', you can enter patterns and strings
483 interactively. And if linked with the main program in `main.c' and
484 the other test files, you can run the already-written tests. */
488 /* We use standard I/O for debugging. */
491 /* It is useful to test things that ``must'' be true when debugging. */
494 static int debug = 0;
496 #define DEBUG_STATEMENT(e) e
497 #define DEBUG_PRINT1(x) if (debug) printf (x)
498 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
499 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
500 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
501 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
502 if (debug) print_partial_compiled_pattern (s, e)
503 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
504 if (debug) print_double_string (w, s1, sz1, s2, sz2)
507 extern void printchar ();
509 /* Print the fastmap in human-readable form. */
512 print_fastmap (fastmap)
515 unsigned was_a_range = 0;
518 while (i < (1 << BYTEWIDTH))
524 while (i < (1 << BYTEWIDTH) && fastmap[i])
540 /* Print a compiled pattern string in human-readable form, starting at
541 the START pointer into it and ending just before the pointer END. */
544 print_partial_compiled_pattern (start, end)
545 unsigned char *start;
549 unsigned char *p = start;
550 unsigned char *pend = end;
558 /* Loop over pattern commands. */
561 printf ("%d:\t", p - start);
563 switch ((re_opcode_t) *p++)
571 printf ("/exactn/%d", mcnt);
582 printf ("/start_memory/%d/%d", mcnt, *p++);
587 printf ("/stop_memory/%d/%d", mcnt, *p++);
591 printf ("/duplicate/%d", *p++);
601 register int c, last = -100;
602 register int in_range = 0;
604 printf ("/charset [%s",
605 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
607 assert (p + *p < pend);
609 for (c = 0; c < 256; c++)
611 && (p[1 + (c/8)] & (1 << (c % 8))))
613 /* Are we starting a range? */
614 if (last + 1 == c && ! in_range)
619 /* Have we broken a range? */
620 else if (last + 1 != c && in_range)
649 case on_failure_jump:
650 extract_number_and_incr (&mcnt, &p);
651 printf ("/on_failure_jump to %d", p + mcnt - start);
654 case on_failure_keep_string_jump:
655 extract_number_and_incr (&mcnt, &p);
656 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
659 case dummy_failure_jump:
660 extract_number_and_incr (&mcnt, &p);
661 printf ("/dummy_failure_jump to %d", p + mcnt - start);
664 case push_dummy_failure:
665 printf ("/push_dummy_failure");
669 extract_number_and_incr (&mcnt, &p);
670 printf ("/maybe_pop_jump to %d", p + mcnt - start);
673 case pop_failure_jump:
674 extract_number_and_incr (&mcnt, &p);
675 printf ("/pop_failure_jump to %d", p + mcnt - start);
679 extract_number_and_incr (&mcnt, &p);
680 printf ("/jump_past_alt to %d", p + mcnt - start);
684 extract_number_and_incr (&mcnt, &p);
685 printf ("/jump to %d", p + mcnt - start);
689 extract_number_and_incr (&mcnt, &p);
690 extract_number_and_incr (&mcnt2, &p);
691 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
695 extract_number_and_incr (&mcnt, &p);
696 extract_number_and_incr (&mcnt2, &p);
697 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
701 extract_number_and_incr (&mcnt, &p);
702 extract_number_and_incr (&mcnt2, &p);
703 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
707 printf ("/wordbound");
711 printf ("/notwordbound");
723 printf ("/before_dot");
731 printf ("/after_dot");
735 printf ("/syntaxspec");
737 printf ("/%d", mcnt);
741 printf ("/notsyntaxspec");
743 printf ("/%d", mcnt);
748 printf ("/wordchar");
752 printf ("/notwordchar");
764 printf ("?%d", *(p-1));
770 printf ("%d:\tend of pattern.\n", p - start);
775 print_compiled_pattern (bufp)
776 struct re_pattern_buffer *bufp;
778 unsigned char *buffer = bufp->buffer;
780 print_partial_compiled_pattern (buffer, buffer + bufp->used);
781 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
783 if (bufp->fastmap_accurate && bufp->fastmap)
785 printf ("fastmap: ");
786 print_fastmap (bufp->fastmap);
789 printf ("re_nsub: %d\t", bufp->re_nsub);
790 printf ("regs_alloc: %d\t", bufp->regs_allocated);
791 printf ("can_be_null: %d\t", bufp->can_be_null);
792 printf ("newline_anchor: %d\n", bufp->newline_anchor);
793 printf ("no_sub: %d\t", bufp->no_sub);
794 printf ("not_bol: %d\t", bufp->not_bol);
795 printf ("not_eol: %d\t", bufp->not_eol);
796 printf ("syntax: %d\n", bufp->syntax);
797 /* Perhaps we should print the translate table? */
802 print_double_string (where, string1, size1, string2, size2)
815 if (FIRST_STRING_P (where))
817 for (this_char = where - string1; this_char < size1; this_char++)
818 printchar (string1[this_char]);
823 for (this_char = where - string2; this_char < size2; this_char++)
824 printchar (string2[this_char]);
828 #else /* not DEBUG */
833 #define DEBUG_STATEMENT(e)
834 #define DEBUG_PRINT1(x)
835 #define DEBUG_PRINT2(x1, x2)
836 #define DEBUG_PRINT3(x1, x2, x3)
837 #define DEBUG_PRINT4(x1, x2, x3, x4)
838 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
839 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
841 #endif /* not DEBUG */
843 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
844 also be assigned to arbitrarily: each pattern buffer stores its own
845 syntax, so it can be changed between regex compilations. */
846 reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS;
849 /* Specify the precise syntax of regexps for compilation. This provides
850 for compatibility for various utilities which historically have
851 different, incompatible syntaxes.
853 The argument SYNTAX is a bit mask comprised of the various bits
854 defined in regex.h. We return the old syntax. */
857 re_set_syntax (syntax)
860 reg_syntax_t ret = re_syntax_options;
862 re_syntax_options = syntax;
866 /* This table gives an error message for each of the error codes listed
867 in regex.h. Obviously the order here has to be same as there. */
869 static const char *re_error_msg[] =
870 { NULL, /* REG_NOERROR */
871 "No match", /* REG_NOMATCH */
872 "Invalid regular expression", /* REG_BADPAT */
873 "Invalid collation character", /* REG_ECOLLATE */
874 "Invalid character class name", /* REG_ECTYPE */
875 "Trailing backslash", /* REG_EESCAPE */
876 "Invalid back reference", /* REG_ESUBREG */
877 "Unmatched [ or [^", /* REG_EBRACK */
878 "Unmatched ( or \\(", /* REG_EPAREN */
879 "Unmatched \\{", /* REG_EBRACE */
880 "Invalid content of \\{\\}", /* REG_BADBR */
881 "Invalid range end", /* REG_ERANGE */
882 "Memory exhausted", /* REG_ESPACE */
883 "Invalid preceding regular expression", /* REG_BADRPT */
884 "Premature end of regular expression", /* REG_EEND */
885 "Regular expression too big", /* REG_ESIZE */
886 "Unmatched ) or \\)", /* REG_ERPAREN */
889 /* Avoiding alloca during matching, to placate r_alloc. */
891 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
892 searching and matching functions should not call alloca. On some
893 systems, alloca is implemented in terms of malloc, and if we're
894 using the relocating allocator routines, then malloc could cause a
895 relocation, which might (if the strings being searched are in the
896 ralloc heap) shift the data out from underneath the regexp
899 Here's another reason to avoid allocation: Emacs insists on
900 processing input from X in a signal handler; processing X input may
901 call malloc; if input arrives while a matching routine is calling
902 malloc, then we're scrod. But Emacs can't just block input while
903 calling matching routines; then we don't notice interrupts when
904 they come in. So, Emacs blocks input around all regexp calls
905 except the matching calls, which it leaves unprotected, in the
906 faith that they will not malloc. */
908 /* Normally, this is fine. */
909 #define MATCH_MAY_ALLOCATE
911 /* But under some circumstances, it's not. */
912 #if defined (emacs) || (defined (REL_ALLOC) && defined (C_ALLOCA))
913 #undef MATCH_MAY_ALLOCATE
917 /* Failure stack declarations and macros; both re_compile_fastmap and
918 re_match_2 use a failure stack. These have to be macros because of
922 /* Number of failure points for which to initially allocate space
923 when matching. If this number is exceeded, we allocate more
924 space, so it is not a hard limit. */
925 #ifndef INIT_FAILURE_ALLOC
926 #define INIT_FAILURE_ALLOC 5
929 /* Roughly the maximum number of failure points on the stack. Would be
930 exactly that if always used MAX_FAILURE_SPACE each time we failed.
931 This is a variable only so users of regex can assign to it; we never
932 change it ourselves. */
933 int re_max_failures = 2000;
935 typedef unsigned char *fail_stack_elt_t;
939 fail_stack_elt_t *stack;
941 unsigned avail; /* Offset of next open position. */
944 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
945 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
946 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
947 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
950 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
952 #ifdef MATCH_MAY_ALLOCATE
953 #define INIT_FAIL_STACK() \
955 fail_stack.stack = (fail_stack_elt_t *) \
956 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
958 if (fail_stack.stack == NULL) \
961 fail_stack.size = INIT_FAILURE_ALLOC; \
962 fail_stack.avail = 0; \
965 #define INIT_FAIL_STACK() \
967 fail_stack.avail = 0; \
972 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
974 Return 1 if succeeds, and 0 if either ran out of memory
975 allocating space for it or it was already too large.
977 REGEX_REALLOCATE requires `destination' be declared. */
979 #define DOUBLE_FAIL_STACK(fail_stack) \
980 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
982 : ((fail_stack).stack = (fail_stack_elt_t *) \
983 REGEX_REALLOCATE ((fail_stack).stack, \
984 (fail_stack).size * sizeof (fail_stack_elt_t), \
985 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
987 (fail_stack).stack == NULL \
989 : ((fail_stack).size <<= 1, \
993 /* Push PATTERN_OP on FAIL_STACK.
995 Return 1 if was able to do so and 0 if ran out of memory allocating
997 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
998 ((FAIL_STACK_FULL () \
999 && !DOUBLE_FAIL_STACK (fail_stack)) \
1001 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
1004 /* This pushes an item onto the failure stack. Must be a four-byte
1005 value. Assumes the variable `fail_stack'. Probably should only
1006 be called from within `PUSH_FAILURE_POINT'. */
1007 #define PUSH_FAILURE_ITEM(item) \
1008 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
1010 /* The complement operation. Assumes `fail_stack' is nonempty. */
1011 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
1013 /* Used to omit pushing failure point id's when we're not debugging. */
1015 #define DEBUG_PUSH PUSH_FAILURE_ITEM
1016 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
1018 #define DEBUG_PUSH(item)
1019 #define DEBUG_POP(item_addr)
1023 /* Push the information about the state we will need
1024 if we ever fail back to it.
1026 Requires variables fail_stack, regstart, regend, reg_info, and
1027 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1030 Does `return FAILURE_CODE' if runs out of memory. */
1032 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1034 char *destination; \
1035 /* Must be int, so when we don't save any registers, the arithmetic \
1036 of 0 + -1 isn't done as unsigned. */ \
1039 DEBUG_STATEMENT (failure_id++); \
1040 DEBUG_STATEMENT (nfailure_points_pushed++); \
1041 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1042 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1043 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1045 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1046 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1048 /* Ensure we have enough space allocated for what we will push. */ \
1049 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1051 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1052 return failure_code; \
1054 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1055 (fail_stack).size); \
1056 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1059 /* Push the info, starting with the registers. */ \
1060 DEBUG_PRINT1 ("\n"); \
1062 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1065 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1066 DEBUG_STATEMENT (num_regs_pushed++); \
1068 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1069 PUSH_FAILURE_ITEM (regstart[this_reg]); \
1071 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1072 PUSH_FAILURE_ITEM (regend[this_reg]); \
1074 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1075 DEBUG_PRINT2 (" match_null=%d", \
1076 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1077 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1078 DEBUG_PRINT2 (" matched_something=%d", \
1079 MATCHED_SOMETHING (reg_info[this_reg])); \
1080 DEBUG_PRINT2 (" ever_matched=%d", \
1081 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1082 DEBUG_PRINT1 ("\n"); \
1083 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
1086 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1087 PUSH_FAILURE_ITEM (lowest_active_reg); \
1089 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1090 PUSH_FAILURE_ITEM (highest_active_reg); \
1092 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1093 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1094 PUSH_FAILURE_ITEM (pattern_place); \
1096 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1097 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1099 DEBUG_PRINT1 ("'\n"); \
1100 PUSH_FAILURE_ITEM (string_place); \
1102 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1103 DEBUG_PUSH (failure_id); \
1106 /* This is the number of items that are pushed and popped on the stack
1107 for each register. */
1108 #define NUM_REG_ITEMS 3
1110 /* Individual items aside from the registers. */
1112 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1114 #define NUM_NONREG_ITEMS 4
1117 /* We push at most this many items on the stack. */
1118 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1120 /* We actually push this many items. */
1121 #define NUM_FAILURE_ITEMS \
1122 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1125 /* How many items can still be added to the stack without overflowing it. */
1126 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1129 /* Pops what PUSH_FAIL_STACK pushes.
1131 We restore into the parameters, all of which should be lvalues:
1132 STR -- the saved data position.
1133 PAT -- the saved pattern position.
1134 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1135 REGSTART, REGEND -- arrays of string positions.
1136 REG_INFO -- array of information about each subexpression.
1138 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1139 `pend', `string1', `size1', `string2', and `size2'. */
1141 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1143 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1145 const unsigned char *string_temp; \
1147 assert (!FAIL_STACK_EMPTY ()); \
1149 /* Remove failure points and point to how many regs pushed. */ \
1150 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1151 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1152 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1154 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1156 DEBUG_POP (&failure_id); \
1157 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1159 /* If the saved string location is NULL, it came from an \
1160 on_failure_keep_string_jump opcode, and we want to throw away the \
1161 saved NULL, thus retaining our current position in the string. */ \
1162 string_temp = POP_FAILURE_ITEM (); \
1163 if (string_temp != NULL) \
1164 str = (const char *) string_temp; \
1166 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1167 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1168 DEBUG_PRINT1 ("'\n"); \
1170 pat = (unsigned char *) POP_FAILURE_ITEM (); \
1171 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1172 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1174 /* Restore register info. */ \
1175 high_reg = (unsigned) POP_FAILURE_ITEM (); \
1176 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1178 low_reg = (unsigned) POP_FAILURE_ITEM (); \
1179 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1181 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1183 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1185 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
1186 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1188 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1189 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1191 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
1192 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1195 DEBUG_STATEMENT (nfailure_points_popped++); \
1196 } /* POP_FAILURE_POINT */
1200 /* Structure for per-register (a.k.a. per-group) information.
1201 This must not be longer than one word, because we push this value
1202 onto the failure stack. Other register information, such as the
1203 starting and ending positions (which are addresses), and the list of
1204 inner groups (which is a bits list) are maintained in separate
1207 We are making a (strictly speaking) nonportable assumption here: that
1208 the compiler will pack our bit fields into something that fits into
1209 the type of `word', i.e., is something that fits into one item on the
1213 fail_stack_elt_t word;
1216 /* This field is one if this group can match the empty string,
1217 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1218 #define MATCH_NULL_UNSET_VALUE 3
1219 unsigned match_null_string_p : 2;
1220 unsigned is_active : 1;
1221 unsigned matched_something : 1;
1222 unsigned ever_matched_something : 1;
1224 } register_info_type;
1226 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1227 #define IS_ACTIVE(R) ((R).bits.is_active)
1228 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1229 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1232 /* Call this when have matched a real character; it sets `matched' flags
1233 for the subexpressions which we are currently inside. Also records
1234 that those subexprs have matched. */
1235 #define SET_REGS_MATCHED() \
1239 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1241 MATCHED_SOMETHING (reg_info[r]) \
1242 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1249 /* Registers are set to a sentinel when they haven't yet matched. */
1250 #define REG_UNSET_VALUE ((char *) -1)
1251 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1255 /* How do we implement a missing MATCH_MAY_ALLOCATE?
1256 We make the fail stack a global thing, and then grow it to
1257 re_max_failures when we compile. */
1258 #ifndef MATCH_MAY_ALLOCATE
1259 static int fail_stack_allocated;
1260 static fail_stack_type fail_stack;
1262 static const char ** regstart, ** regend;
1263 static const char ** old_regstart, ** old_regend;
1264 static const char **best_regstart, **best_regend;
1265 static register_info_type *reg_info;
1266 static const char **reg_dummy;
1267 static register_info_type *reg_info_dummy;
1271 /* Subroutine declarations and macros for regex_compile. */
1273 static void store_op1 (), store_op2 ();
1274 static void insert_op1 (), insert_op2 ();
1275 static boolean at_begline_loc_p (), at_endline_loc_p ();
1276 static boolean group_in_compile_stack ();
1277 static reg_errcode_t compile_range ();
1279 /* Fetch the next character in the uncompiled pattern---translating it
1280 if necessary. Also cast from a signed character in the constant
1281 string passed to us by the user to an unsigned char that we can use
1282 as an array index (in, e.g., `translate'). */
1283 #define PATFETCH(c) \
1284 do {if (p == pend) return REG_EEND; \
1285 c = (unsigned char) *p++; \
1286 if (translate) c = translate[c]; \
1289 /* Fetch the next character in the uncompiled pattern, with no
1291 #define PATFETCH_RAW(c) \
1292 do {if (p == pend) return REG_EEND; \
1293 c = (unsigned char) *p++; \
1296 /* Go backwards one character in the pattern. */
1297 #define PATUNFETCH p--
1300 /* If `translate' is non-null, return translate[D], else just D. We
1301 cast the subscript to translate because some data is declared as
1302 `char *', to avoid warnings when a string constant is passed. But
1303 when we use a character as a subscript we must make it unsigned. */
1304 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1307 /* Macros for outputting the compiled pattern into `buffer'. */
1309 /* If the buffer isn't allocated when it comes in, use this. */
1310 #define INIT_BUF_SIZE 32
1312 /* Make sure we have at least N more bytes of space in buffer. */
1313 #define GET_BUFFER_SPACE(n) \
1314 while (b - bufp->buffer + (n) > bufp->allocated) \
1317 /* Make sure we have one more byte of buffer space and then add C to it. */
1318 #define BUF_PUSH(c) \
1320 GET_BUFFER_SPACE (1); \
1321 *b++ = (unsigned char) (c); \
1325 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1326 #define BUF_PUSH_2(c1, c2) \
1328 GET_BUFFER_SPACE (2); \
1329 *b++ = (unsigned char) (c1); \
1330 *b++ = (unsigned char) (c2); \
1334 /* As with BUF_PUSH_2, except for three bytes. */
1335 #define BUF_PUSH_3(c1, c2, c3) \
1337 GET_BUFFER_SPACE (3); \
1338 *b++ = (unsigned char) (c1); \
1339 *b++ = (unsigned char) (c2); \
1340 *b++ = (unsigned char) (c3); \
1344 /* Store a jump with opcode OP at LOC to location TO. We store a
1345 relative address offset by the three bytes the jump itself occupies. */
1346 #define STORE_JUMP(op, loc, to) \
1347 store_op1 (op, loc, (to) - (loc) - 3)
1349 /* Likewise, for a two-argument jump. */
1350 #define STORE_JUMP2(op, loc, to, arg) \
1351 store_op2 (op, loc, (to) - (loc) - 3, arg)
1353 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1354 #define INSERT_JUMP(op, loc, to) \
1355 insert_op1 (op, loc, (to) - (loc) - 3, b)
1357 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1358 #define INSERT_JUMP2(op, loc, to, arg) \
1359 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1362 /* This is not an arbitrary limit: the arguments which represent offsets
1363 into the pattern are two bytes long. So if 2^16 bytes turns out to
1364 be too small, many things would have to change. */
1365 #define MAX_BUF_SIZE (1L << 16)
1368 /* Extend the buffer by twice its current size via realloc and
1369 reset the pointers that pointed into the old block to point to the
1370 correct places in the new one. If extending the buffer results in it
1371 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1372 #define EXTEND_BUFFER() \
1374 unsigned char *old_buffer = bufp->buffer; \
1375 if (bufp->allocated == MAX_BUF_SIZE) \
1377 bufp->allocated <<= 1; \
1378 if (bufp->allocated > MAX_BUF_SIZE) \
1379 bufp->allocated = MAX_BUF_SIZE; \
1380 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1381 if (bufp->buffer == NULL) \
1382 return REG_ESPACE; \
1383 /* If the buffer moved, move all the pointers into it. */ \
1384 if (old_buffer != bufp->buffer) \
1386 b = (b - old_buffer) + bufp->buffer; \
1387 begalt = (begalt - old_buffer) + bufp->buffer; \
1388 if (fixup_alt_jump) \
1389 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1391 laststart = (laststart - old_buffer) + bufp->buffer; \
1392 if (pending_exact) \
1393 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1398 /* Since we have one byte reserved for the register number argument to
1399 {start,stop}_memory, the maximum number of groups we can report
1400 things about is what fits in that byte. */
1401 #define MAX_REGNUM 255
1403 /* But patterns can have more than `MAX_REGNUM' registers. We just
1404 ignore the excess. */
1405 typedef unsigned regnum_t;
1408 /* Macros for the compile stack. */
1410 /* Since offsets can go either forwards or backwards, this type needs to
1411 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1412 typedef int pattern_offset_t;
1416 pattern_offset_t begalt_offset;
1417 pattern_offset_t fixup_alt_jump;
1418 pattern_offset_t inner_group_offset;
1419 pattern_offset_t laststart_offset;
1421 } compile_stack_elt_t;
1426 compile_stack_elt_t *stack;
1428 unsigned avail; /* Offset of next open position. */
1429 } compile_stack_type;
1432 #define INIT_COMPILE_STACK_SIZE 32
1434 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1435 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1437 /* The next available element. */
1438 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1441 /* Set the bit for character C in a list. */
1442 #define SET_LIST_BIT(c) \
1443 (b[((unsigned char) (c)) / BYTEWIDTH] \
1444 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1447 /* Get the next unsigned number in the uncompiled pattern. */
1448 #define GET_UNSIGNED_NUMBER(num) \
1452 while (ISDIGIT (c)) \
1456 num = num * 10 + c - '0'; \
1464 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1466 #define IS_CHAR_CLASS(string) \
1467 (STREQ (string, "alpha") || STREQ (string, "upper") \
1468 || STREQ (string, "lower") || STREQ (string, "digit") \
1469 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1470 || STREQ (string, "space") || STREQ (string, "print") \
1471 || STREQ (string, "punct") || STREQ (string, "graph") \
1472 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1474 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1475 Returns one of error codes defined in `regex.h', or zero for success.
1477 Assumes the `allocated' (and perhaps `buffer') and `translate'
1478 fields are set in BUFP on entry.
1480 If it succeeds, results are put in BUFP (if it returns an error, the
1481 contents of BUFP are undefined):
1482 `buffer' is the compiled pattern;
1483 `syntax' is set to SYNTAX;
1484 `used' is set to the length of the compiled pattern;
1485 `fastmap_accurate' is zero;
1486 `re_nsub' is the number of subexpressions in PATTERN;
1487 `not_bol' and `not_eol' are zero;
1489 The `fastmap' and `newline_anchor' fields are neither
1490 examined nor set. */
1492 static reg_errcode_t
1493 regex_compile (pattern, size, syntax, bufp)
1494 const char *pattern;
1496 reg_syntax_t syntax;
1497 struct re_pattern_buffer *bufp;
1499 /* We fetch characters from PATTERN here. Even though PATTERN is
1500 `char *' (i.e., signed), we declare these variables as unsigned, so
1501 they can be reliably used as array indices. */
1502 register unsigned char c, c1;
1504 /* A random temporary spot in PATTERN. */
1507 /* Points to the end of the buffer, where we should append. */
1508 register unsigned char *b;
1510 /* Keeps track of unclosed groups. */
1511 compile_stack_type compile_stack;
1513 /* Points to the current (ending) position in the pattern. */
1514 const char *p = pattern;
1515 const char *pend = pattern + size;
1517 /* How to translate the characters in the pattern. */
1518 char *translate = bufp->translate;
1520 /* Address of the count-byte of the most recently inserted `exactn'
1521 command. This makes it possible to tell if a new exact-match
1522 character can be added to that command or if the character requires
1523 a new `exactn' command. */
1524 unsigned char *pending_exact = 0;
1526 /* Address of start of the most recently finished expression.
1527 This tells, e.g., postfix * where to find the start of its
1528 operand. Reset at the beginning of groups and alternatives. */
1529 unsigned char *laststart = 0;
1531 /* Address of beginning of regexp, or inside of last group. */
1532 unsigned char *begalt;
1534 /* Place in the uncompiled pattern (i.e., the {) to
1535 which to go back if the interval is invalid. */
1536 const char *beg_interval;
1538 /* Address of the place where a forward jump should go to the end of
1539 the containing expression. Each alternative of an `or' -- except the
1540 last -- ends with a forward jump of this sort. */
1541 unsigned char *fixup_alt_jump = 0;
1543 /* Counts open-groups as they are encountered. Remembered for the
1544 matching close-group on the compile stack, so the same register
1545 number is put in the stop_memory as the start_memory. */
1546 regnum_t regnum = 0;
1549 DEBUG_PRINT1 ("\nCompiling pattern: ");
1552 unsigned debug_count;
1554 for (debug_count = 0; debug_count < size; debug_count++)
1555 printchar (pattern[debug_count]);
1560 /* Initialize the compile stack. */
1561 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1562 if (compile_stack.stack == NULL)
1565 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1566 compile_stack.avail = 0;
1568 /* Initialize the pattern buffer. */
1569 bufp->syntax = syntax;
1570 bufp->fastmap_accurate = 0;
1571 bufp->not_bol = bufp->not_eol = 0;
1573 /* Set `used' to zero, so that if we return an error, the pattern
1574 printer (for debugging) will think there's no pattern. We reset it
1578 /* Always count groups, whether or not bufp->no_sub is set. */
1581 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1582 /* Initialize the syntax table. */
1583 init_syntax_once ();
1586 if (bufp->allocated == 0)
1589 { /* If zero allocated, but buffer is non-null, try to realloc
1590 enough space. This loses if buffer's address is bogus, but
1591 that is the user's responsibility. */
1592 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1595 { /* Caller did not allocate a buffer. Do it for them. */
1596 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1598 if (!bufp->buffer) return REG_ESPACE;
1600 bufp->allocated = INIT_BUF_SIZE;
1603 begalt = b = bufp->buffer;
1605 /* Loop through the uncompiled pattern until we're at the end. */
1614 if ( /* If at start of pattern, it's an operator. */
1616 /* If context independent, it's an operator. */
1617 || syntax & RE_CONTEXT_INDEP_ANCHORS
1618 /* Otherwise, depends on what's come before. */
1619 || at_begline_loc_p (pattern, p, syntax))
1629 if ( /* If at end of pattern, it's an operator. */
1631 /* If context independent, it's an operator. */
1632 || syntax & RE_CONTEXT_INDEP_ANCHORS
1633 /* Otherwise, depends on what's next. */
1634 || at_endline_loc_p (p, pend, syntax))
1644 if ((syntax & RE_BK_PLUS_QM)
1645 || (syntax & RE_LIMITED_OPS))
1649 /* If there is no previous pattern... */
1652 if (syntax & RE_CONTEXT_INVALID_OPS)
1654 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1659 /* Are we optimizing this jump? */
1660 boolean keep_string_p = false;
1662 /* 1 means zero (many) matches is allowed. */
1663 char zero_times_ok = 0, many_times_ok = 0;
1665 /* If there is a sequence of repetition chars, collapse it
1666 down to just one (the right one). We can't combine
1667 interval operators with these because of, e.g., `a{2}*',
1668 which should only match an even number of `a's. */
1672 zero_times_ok |= c != '+';
1673 many_times_ok |= c != '?';
1681 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1684 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1686 if (p == pend) return REG_EESCAPE;
1689 if (!(c1 == '+' || c1 == '?'))
1704 /* If we get here, we found another repeat character. */
1707 /* Star, etc. applied to an empty pattern is equivalent
1708 to an empty pattern. */
1712 /* Now we know whether or not zero matches is allowed
1713 and also whether or not two or more matches is allowed. */
1715 { /* More than one repetition is allowed, so put in at the
1716 end a backward relative jump from `b' to before the next
1717 jump we're going to put in below (which jumps from
1718 laststart to after this jump).
1720 But if we are at the `*' in the exact sequence `.*\n',
1721 insert an unconditional jump backwards to the .,
1722 instead of the beginning of the loop. This way we only
1723 push a failure point once, instead of every time
1724 through the loop. */
1725 assert (p - 1 > pattern);
1727 /* Allocate the space for the jump. */
1728 GET_BUFFER_SPACE (3);
1730 /* We know we are not at the first character of the pattern,
1731 because laststart was nonzero. And we've already
1732 incremented `p', by the way, to be the character after
1733 the `*'. Do we have to do something analogous here
1734 for null bytes, because of RE_DOT_NOT_NULL? */
1735 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1737 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1738 && !(syntax & RE_DOT_NEWLINE))
1739 { /* We have .*\n. */
1740 STORE_JUMP (jump, b, laststart);
1741 keep_string_p = true;
1744 /* Anything else. */
1745 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1747 /* We've added more stuff to the buffer. */
1751 /* On failure, jump from laststart to b + 3, which will be the
1752 end of the buffer after this jump is inserted. */
1753 GET_BUFFER_SPACE (3);
1754 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1762 /* At least one repetition is required, so insert a
1763 `dummy_failure_jump' before the initial
1764 `on_failure_jump' instruction of the loop. This
1765 effects a skip over that instruction the first time
1766 we hit that loop. */
1767 GET_BUFFER_SPACE (3);
1768 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1783 boolean had_char_class = false;
1785 if (p == pend) return REG_EBRACK;
1787 /* Ensure that we have enough space to push a charset: the
1788 opcode, the length count, and the bitset; 34 bytes in all. */
1789 GET_BUFFER_SPACE (34);
1793 /* We test `*p == '^' twice, instead of using an if
1794 statement, so we only need one BUF_PUSH. */
1795 BUF_PUSH (*p == '^' ? charset_not : charset);
1799 /* Remember the first position in the bracket expression. */
1802 /* Push the number of bytes in the bitmap. */
1803 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1805 /* Clear the whole map. */
1806 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1808 /* charset_not matches newline according to a syntax bit. */
1809 if ((re_opcode_t) b[-2] == charset_not
1810 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1811 SET_LIST_BIT ('\n');
1813 /* Read in characters and ranges, setting map bits. */
1816 if (p == pend) return REG_EBRACK;
1820 /* \ might escape characters inside [...] and [^...]. */
1821 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1823 if (p == pend) return REG_EESCAPE;
1830 /* Could be the end of the bracket expression. If it's
1831 not (i.e., when the bracket expression is `[]' so
1832 far), the ']' character bit gets set way below. */
1833 if (c == ']' && p != p1 + 1)
1836 /* Look ahead to see if it's a range when the last thing
1837 was a character class. */
1838 if (had_char_class && c == '-' && *p != ']')
1841 /* Look ahead to see if it's a range when the last thing
1842 was a character: if this is a hyphen not at the
1843 beginning or the end of a list, then it's the range
1846 && !(p - 2 >= pattern && p[-2] == '[')
1847 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1851 = compile_range (&p, pend, translate, syntax, b);
1852 if (ret != REG_NOERROR) return ret;
1855 else if (p[0] == '-' && p[1] != ']')
1856 { /* This handles ranges made up of characters only. */
1859 /* Move past the `-'. */
1862 ret = compile_range (&p, pend, translate, syntax, b);
1863 if (ret != REG_NOERROR) return ret;
1866 /* See if we're at the beginning of a possible character
1869 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1870 { /* Leave room for the null. */
1871 char str[CHAR_CLASS_MAX_LENGTH + 1];
1876 /* If pattern is `[[:'. */
1877 if (p == pend) return REG_EBRACK;
1882 if (c == ':' || c == ']' || p == pend
1883 || c1 == CHAR_CLASS_MAX_LENGTH)
1889 /* If isn't a word bracketed by `[:' and:`]':
1890 undo the ending character, the letters, and leave
1891 the leading `:' and `[' (but set bits for them). */
1892 if (c == ':' && *p == ']')
1895 boolean is_alnum = STREQ (str, "alnum");
1896 boolean is_alpha = STREQ (str, "alpha");
1897 boolean is_blank = STREQ (str, "blank");
1898 boolean is_cntrl = STREQ (str, "cntrl");
1899 boolean is_digit = STREQ (str, "digit");
1900 boolean is_graph = STREQ (str, "graph");
1901 boolean is_lower = STREQ (str, "lower");
1902 boolean is_print = STREQ (str, "print");
1903 boolean is_punct = STREQ (str, "punct");
1904 boolean is_space = STREQ (str, "space");
1905 boolean is_upper = STREQ (str, "upper");
1906 boolean is_xdigit = STREQ (str, "xdigit");
1908 if (!IS_CHAR_CLASS (str)) return REG_ECTYPE;
1910 /* Throw away the ] at the end of the character
1914 if (p == pend) return REG_EBRACK;
1916 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
1918 if ( (is_alnum && ISALNUM (ch))
1919 || (is_alpha && ISALPHA (ch))
1920 || (is_blank && ISBLANK (ch))
1921 || (is_cntrl && ISCNTRL (ch))
1922 || (is_digit && ISDIGIT (ch))
1923 || (is_graph && ISGRAPH (ch))
1924 || (is_lower && ISLOWER (ch))
1925 || (is_print && ISPRINT (ch))
1926 || (is_punct && ISPUNCT (ch))
1927 || (is_space && ISSPACE (ch))
1928 || (is_upper && ISUPPER (ch))
1929 || (is_xdigit && ISXDIGIT (ch)))
1932 had_char_class = true;
1941 had_char_class = false;
1946 had_char_class = false;
1951 /* Discard any (non)matching list bytes that are all 0 at the
1952 end of the map. Decrease the map-length byte too. */
1953 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
1961 if (syntax & RE_NO_BK_PARENS)
1968 if (syntax & RE_NO_BK_PARENS)
1975 if (syntax & RE_NEWLINE_ALT)
1982 if (syntax & RE_NO_BK_VBAR)
1989 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
1990 goto handle_interval;
1996 if (p == pend) return REG_EESCAPE;
1998 /* Do not translate the character after the \, so that we can
1999 distinguish, e.g., \B from \b, even if we normally would
2000 translate, e.g., B to b. */
2006 if (syntax & RE_NO_BK_PARENS)
2007 goto normal_backslash;
2013 if (COMPILE_STACK_FULL)
2015 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2016 compile_stack_elt_t);
2017 if (compile_stack.stack == NULL) return REG_ESPACE;
2019 compile_stack.size <<= 1;
2022 /* These are the values to restore when we hit end of this
2023 group. They are all relative offsets, so that if the
2024 whole pattern moves because of realloc, they will still
2026 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2027 COMPILE_STACK_TOP.fixup_alt_jump
2028 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2029 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2030 COMPILE_STACK_TOP.regnum = regnum;
2032 /* We will eventually replace the 0 with the number of
2033 groups inner to this one. But do not push a
2034 start_memory for groups beyond the last one we can
2035 represent in the compiled pattern. */
2036 if (regnum <= MAX_REGNUM)
2038 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2039 BUF_PUSH_3 (start_memory, regnum, 0);
2042 compile_stack.avail++;
2047 /* If we've reached MAX_REGNUM groups, then this open
2048 won't actually generate any code, so we'll have to
2049 clear pending_exact explicitly. */
2055 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2057 if (COMPILE_STACK_EMPTY)
2058 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2059 goto normal_backslash;
2065 { /* Push a dummy failure point at the end of the
2066 alternative for a possible future
2067 `pop_failure_jump' to pop. See comments at
2068 `push_dummy_failure' in `re_match_2'. */
2069 BUF_PUSH (push_dummy_failure);
2071 /* We allocated space for this jump when we assigned
2072 to `fixup_alt_jump', in the `handle_alt' case below. */
2073 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2076 /* See similar code for backslashed left paren above. */
2077 if (COMPILE_STACK_EMPTY)
2078 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2083 /* Since we just checked for an empty stack above, this
2084 ``can't happen''. */
2085 assert (compile_stack.avail != 0);
2087 /* We don't just want to restore into `regnum', because
2088 later groups should continue to be numbered higher,
2089 as in `(ab)c(de)' -- the second group is #2. */
2090 regnum_t this_group_regnum;
2092 compile_stack.avail--;
2093 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2095 = COMPILE_STACK_TOP.fixup_alt_jump
2096 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2098 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2099 this_group_regnum = COMPILE_STACK_TOP.regnum;
2100 /* If we've reached MAX_REGNUM groups, then this open
2101 won't actually generate any code, so we'll have to
2102 clear pending_exact explicitly. */
2105 /* We're at the end of the group, so now we know how many
2106 groups were inside this one. */
2107 if (this_group_regnum <= MAX_REGNUM)
2109 unsigned char *inner_group_loc
2110 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2112 *inner_group_loc = regnum - this_group_regnum;
2113 BUF_PUSH_3 (stop_memory, this_group_regnum,
2114 regnum - this_group_regnum);
2120 case '|': /* `\|'. */
2121 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2122 goto normal_backslash;
2124 if (syntax & RE_LIMITED_OPS)
2127 /* Insert before the previous alternative a jump which
2128 jumps to this alternative if the former fails. */
2129 GET_BUFFER_SPACE (3);
2130 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2134 /* The alternative before this one has a jump after it
2135 which gets executed if it gets matched. Adjust that
2136 jump so it will jump to this alternative's analogous
2137 jump (put in below, which in turn will jump to the next
2138 (if any) alternative's such jump, etc.). The last such
2139 jump jumps to the correct final destination. A picture:
2145 If we are at `b', then fixup_alt_jump right now points to a
2146 three-byte space after `a'. We'll put in the jump, set
2147 fixup_alt_jump to right after `b', and leave behind three
2148 bytes which we'll fill in when we get to after `c'. */
2151 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2153 /* Mark and leave space for a jump after this alternative,
2154 to be filled in later either by next alternative or
2155 when know we're at the end of a series of alternatives. */
2157 GET_BUFFER_SPACE (3);
2166 /* If \{ is a literal. */
2167 if (!(syntax & RE_INTERVALS)
2168 /* If we're at `\{' and it's not the open-interval
2170 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2171 || (p - 2 == pattern && p == pend))
2172 goto normal_backslash;
2176 /* If got here, then the syntax allows intervals. */
2178 /* At least (most) this many matches must be made. */
2179 int lower_bound = -1, upper_bound = -1;
2181 beg_interval = p - 1;
2185 if (syntax & RE_NO_BK_BRACES)
2186 goto unfetch_interval;
2191 GET_UNSIGNED_NUMBER (lower_bound);
2195 GET_UNSIGNED_NUMBER (upper_bound);
2196 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2199 /* Interval such as `{1}' => match exactly once. */
2200 upper_bound = lower_bound;
2202 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2203 || lower_bound > upper_bound)
2205 if (syntax & RE_NO_BK_BRACES)
2206 goto unfetch_interval;
2211 if (!(syntax & RE_NO_BK_BRACES))
2213 if (c != '\\') return REG_EBRACE;
2220 if (syntax & RE_NO_BK_BRACES)
2221 goto unfetch_interval;
2226 /* We just parsed a valid interval. */
2228 /* If it's invalid to have no preceding re. */
2231 if (syntax & RE_CONTEXT_INVALID_OPS)
2233 else if (syntax & RE_CONTEXT_INDEP_OPS)
2236 goto unfetch_interval;
2239 /* If the upper bound is zero, don't want to succeed at
2240 all; jump from `laststart' to `b + 3', which will be
2241 the end of the buffer after we insert the jump. */
2242 if (upper_bound == 0)
2244 GET_BUFFER_SPACE (3);
2245 INSERT_JUMP (jump, laststart, b + 3);
2249 /* Otherwise, we have a nontrivial interval. When
2250 we're all done, the pattern will look like:
2251 set_number_at <jump count> <upper bound>
2252 set_number_at <succeed_n count> <lower bound>
2253 succeed_n <after jump addr> <succeed_n count>
2255 jump_n <succeed_n addr> <jump count>
2256 (The upper bound and `jump_n' are omitted if
2257 `upper_bound' is 1, though.) */
2259 { /* If the upper bound is > 1, we need to insert
2260 more at the end of the loop. */
2261 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2263 GET_BUFFER_SPACE (nbytes);
2265 /* Initialize lower bound of the `succeed_n', even
2266 though it will be set during matching by its
2267 attendant `set_number_at' (inserted next),
2268 because `re_compile_fastmap' needs to know.
2269 Jump to the `jump_n' we might insert below. */
2270 INSERT_JUMP2 (succeed_n, laststart,
2271 b + 5 + (upper_bound > 1) * 5,
2275 /* Code to initialize the lower bound. Insert
2276 before the `succeed_n'. The `5' is the last two
2277 bytes of this `set_number_at', plus 3 bytes of
2278 the following `succeed_n'. */
2279 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2282 if (upper_bound > 1)
2283 { /* More than one repetition is allowed, so
2284 append a backward jump to the `succeed_n'
2285 that starts this interval.
2287 When we've reached this during matching,
2288 we'll have matched the interval once, so
2289 jump back only `upper_bound - 1' times. */
2290 STORE_JUMP2 (jump_n, b, laststart + 5,
2294 /* The location we want to set is the second
2295 parameter of the `jump_n'; that is `b-2' as
2296 an absolute address. `laststart' will be
2297 the `set_number_at' we're about to insert;
2298 `laststart+3' the number to set, the source
2299 for the relative address. But we are
2300 inserting into the middle of the pattern --
2301 so everything is getting moved up by 5.
2302 Conclusion: (b - 2) - (laststart + 3) + 5,
2303 i.e., b - laststart.
2305 We insert this at the beginning of the loop
2306 so that if we fail during matching, we'll
2307 reinitialize the bounds. */
2308 insert_op2 (set_number_at, laststart, b - laststart,
2309 upper_bound - 1, b);
2314 beg_interval = NULL;
2319 /* If an invalid interval, match the characters as literals. */
2320 assert (beg_interval);
2322 beg_interval = NULL;
2324 /* normal_char and normal_backslash need `c'. */
2327 if (!(syntax & RE_NO_BK_BRACES))
2329 if (p > pattern && p[-1] == '\\')
2330 goto normal_backslash;
2335 /* There is no way to specify the before_dot and after_dot
2336 operators. rms says this is ok. --karl */
2344 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2350 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2357 BUF_PUSH (wordchar);
2363 BUF_PUSH (notwordchar);
2376 BUF_PUSH (wordbound);
2380 BUF_PUSH (notwordbound);
2391 case '1': case '2': case '3': case '4': case '5':
2392 case '6': case '7': case '8': case '9':
2393 if (syntax & RE_NO_BK_REFS)
2401 /* Can't back reference to a subexpression if inside of it. */
2402 if (group_in_compile_stack (compile_stack, c1))
2406 BUF_PUSH_2 (duplicate, c1);
2412 if (syntax & RE_BK_PLUS_QM)
2415 goto normal_backslash;
2419 /* You might think it would be useful for \ to mean
2420 not to translate; but if we don't translate it
2421 it will never match anything. */
2429 /* Expects the character in `c'. */
2431 /* If no exactn currently being built. */
2434 /* If last exactn not at current position. */
2435 || pending_exact + *pending_exact + 1 != b
2437 /* We have only one byte following the exactn for the count. */
2438 || *pending_exact == (1 << BYTEWIDTH) - 1
2440 /* If followed by a repetition operator. */
2441 || *p == '*' || *p == '^'
2442 || ((syntax & RE_BK_PLUS_QM)
2443 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2444 : (*p == '+' || *p == '?'))
2445 || ((syntax & RE_INTERVALS)
2446 && ((syntax & RE_NO_BK_BRACES)
2448 : (p[0] == '\\' && p[1] == '{'))))
2450 /* Start building a new exactn. */
2454 BUF_PUSH_2 (exactn, 0);
2455 pending_exact = b - 1;
2462 } /* while p != pend */
2465 /* Through the pattern now. */
2468 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2470 if (!COMPILE_STACK_EMPTY)
2473 free (compile_stack.stack);
2475 /* We have succeeded; set the length of the buffer. */
2476 bufp->used = b - bufp->buffer;
2481 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2482 print_compiled_pattern (bufp);
2486 #ifndef MATCH_MAY_ALLOCATE
2487 /* Initialize the failure stack to the largest possible stack. This
2488 isn't necessary unless we're trying to avoid calling alloca in
2489 the search and match routines. */
2491 int num_regs = bufp->re_nsub + 1;
2493 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2494 is strictly greater than re_max_failures, the largest possible stack
2495 is 2 * re_max_failures failure points. */
2496 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2497 if (fail_stack.size > fail_stack_allocated)
2499 if (! fail_stack.stack)
2501 (fail_stack_elt_t *) malloc (fail_stack.size
2502 * sizeof (fail_stack_elt_t));
2505 (fail_stack_elt_t *) realloc (fail_stack.stack,
2507 * sizeof (fail_stack_elt_t)));
2508 fail_stack_allocated = fail_stack.size;
2511 /* Initialize some other variables the matcher uses. */
2512 RETALLOC_IF (regstart, num_regs, const char *);
2513 RETALLOC_IF (regend, num_regs, const char *);
2514 RETALLOC_IF (old_regstart, num_regs, const char *);
2515 RETALLOC_IF (old_regend, num_regs, const char *);
2516 RETALLOC_IF (best_regstart, num_regs, const char *);
2517 RETALLOC_IF (best_regend, num_regs, const char *);
2518 RETALLOC_IF (reg_info, num_regs, register_info_type);
2519 RETALLOC_IF (reg_dummy, num_regs, const char *);
2520 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
2525 } /* regex_compile */
2527 /* Subroutines for `regex_compile'. */
2529 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2532 store_op1 (op, loc, arg)
2537 *loc = (unsigned char) op;
2538 STORE_NUMBER (loc + 1, arg);
2542 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2545 store_op2 (op, loc, arg1, arg2)
2550 *loc = (unsigned char) op;
2551 STORE_NUMBER (loc + 1, arg1);
2552 STORE_NUMBER (loc + 3, arg2);
2556 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2557 for OP followed by two-byte integer parameter ARG. */
2560 insert_op1 (op, loc, arg, end)
2566 register unsigned char *pfrom = end;
2567 register unsigned char *pto = end + 3;
2569 while (pfrom != loc)
2572 store_op1 (op, loc, arg);
2576 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2579 insert_op2 (op, loc, arg1, arg2, end)
2585 register unsigned char *pfrom = end;
2586 register unsigned char *pto = end + 5;
2588 while (pfrom != loc)
2591 store_op2 (op, loc, arg1, arg2);
2595 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2596 after an alternative or a begin-subexpression. We assume there is at
2597 least one character before the ^. */
2600 at_begline_loc_p (pattern, p, syntax)
2601 const char *pattern, *p;
2602 reg_syntax_t syntax;
2604 const char *prev = p - 2;
2605 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2608 /* After a subexpression? */
2609 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2610 /* After an alternative? */
2611 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2615 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2616 at least one character after the $, i.e., `P < PEND'. */
2619 at_endline_loc_p (p, pend, syntax)
2620 const char *p, *pend;
2623 const char *next = p;
2624 boolean next_backslash = *next == '\\';
2625 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2628 /* Before a subexpression? */
2629 (syntax & RE_NO_BK_PARENS ? *next == ')'
2630 : next_backslash && next_next && *next_next == ')')
2631 /* Before an alternative? */
2632 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2633 : next_backslash && next_next && *next_next == '|');
2637 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2638 false if it's not. */
2641 group_in_compile_stack (compile_stack, regnum)
2642 compile_stack_type compile_stack;
2647 for (this_element = compile_stack.avail - 1;
2650 if (compile_stack.stack[this_element].regnum == regnum)
2657 /* Read the ending character of a range (in a bracket expression) from the
2658 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2659 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2660 Then we set the translation of all bits between the starting and
2661 ending characters (inclusive) in the compiled pattern B.
2663 Return an error code.
2665 We use these short variable names so we can use the same macros as
2666 `regex_compile' itself. */
2668 static reg_errcode_t
2669 compile_range (p_ptr, pend, translate, syntax, b)
2670 const char **p_ptr, *pend;
2672 reg_syntax_t syntax;
2677 const char *p = *p_ptr;
2678 int range_start, range_end;
2683 /* Even though the pattern is a signed `char *', we need to fetch
2684 with unsigned char *'s; if the high bit of the pattern character
2685 is set, the range endpoints will be negative if we fetch using a
2688 We also want to fetch the endpoints without translating them; the
2689 appropriate translation is done in the bit-setting loop below. */
2690 range_start = ((unsigned char *) p)[-2];
2691 range_end = ((unsigned char *) p)[0];
2693 /* Have to increment the pointer into the pattern string, so the
2694 caller isn't still at the ending character. */
2697 /* If the start is after the end, the range is empty. */
2698 if (range_start > range_end)
2699 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2701 /* Here we see why `this_char' has to be larger than an `unsigned
2702 char' -- the range is inclusive, so if `range_end' == 0xff
2703 (assuming 8-bit characters), we would otherwise go into an infinite
2704 loop, since all characters <= 0xff. */
2705 for (this_char = range_start; this_char <= range_end; this_char++)
2707 SET_LIST_BIT (TRANSLATE (this_char));
2713 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2714 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2715 characters can start a string that matches the pattern. This fastmap
2716 is used by re_search to skip quickly over impossible starting points.
2718 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2719 area as BUFP->fastmap.
2721 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2724 Returns 0 if we succeed, -2 if an internal error. */
2727 re_compile_fastmap (bufp)
2728 struct re_pattern_buffer *bufp;
2731 #ifdef MATCH_MAY_ALLOCATE
2732 fail_stack_type fail_stack;
2734 #ifndef REGEX_MALLOC
2737 /* We don't push any register information onto the failure stack. */
2738 unsigned num_regs = 0;
2740 register char *fastmap = bufp->fastmap;
2741 unsigned char *pattern = bufp->buffer;
2742 unsigned long size = bufp->used;
2743 unsigned char *p = pattern;
2744 register unsigned char *pend = pattern + size;
2746 /* Assume that each path through the pattern can be null until
2747 proven otherwise. We set this false at the bottom of switch
2748 statement, to which we get only if a particular path doesn't
2749 match the empty string. */
2750 boolean path_can_be_null = true;
2752 /* We aren't doing a `succeed_n' to begin with. */
2753 boolean succeed_n_p = false;
2755 assert (fastmap != NULL && p != NULL);
2758 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2759 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2760 bufp->can_be_null = 0;
2762 while (p != pend || !FAIL_STACK_EMPTY ())
2766 bufp->can_be_null |= path_can_be_null;
2768 /* Reset for next path. */
2769 path_can_be_null = true;
2771 p = fail_stack.stack[--fail_stack.avail];
2774 /* We should never be about to go beyond the end of the pattern. */
2777 #ifdef SWITCH_ENUM_BUG
2778 switch ((int) ((re_opcode_t) *p++))
2780 switch ((re_opcode_t) *p++)
2784 /* I guess the idea here is to simply not bother with a fastmap
2785 if a backreference is used, since it's too hard to figure out
2786 the fastmap for the corresponding group. Setting
2787 `can_be_null' stops `re_search_2' from using the fastmap, so
2788 that is all we do. */
2790 bufp->can_be_null = 1;
2794 /* Following are the cases which match a character. These end
2803 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2804 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2810 /* Chars beyond end of map must be allowed. */
2811 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2814 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2815 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2821 for (j = 0; j < (1 << BYTEWIDTH); j++)
2822 if (SYNTAX (j) == Sword)
2828 for (j = 0; j < (1 << BYTEWIDTH); j++)
2829 if (SYNTAX (j) != Sword)
2835 /* `.' matches anything ... */
2836 for (j = 0; j < (1 << BYTEWIDTH); j++)
2839 /* ... except perhaps newline. */
2840 if (!(bufp->syntax & RE_DOT_NEWLINE))
2843 /* Return if we have already set `can_be_null'; if we have,
2844 then the fastmap is irrelevant. Something's wrong here. */
2845 else if (bufp->can_be_null)
2848 /* Otherwise, have to check alternative paths. */
2855 for (j = 0; j < (1 << BYTEWIDTH); j++)
2856 if (SYNTAX (j) == (enum syntaxcode) k)
2863 for (j = 0; j < (1 << BYTEWIDTH); j++)
2864 if (SYNTAX (j) != (enum syntaxcode) k)
2869 /* All cases after this match the empty string. These end with
2877 #endif /* not emacs */
2889 case push_dummy_failure:
2894 case pop_failure_jump:
2895 case maybe_pop_jump:
2898 case dummy_failure_jump:
2899 EXTRACT_NUMBER_AND_INCR (j, p);
2904 /* Jump backward implies we just went through the body of a
2905 loop and matched nothing. Opcode jumped to should be
2906 `on_failure_jump' or `succeed_n'. Just treat it like an
2907 ordinary jump. For a * loop, it has pushed its failure
2908 point already; if so, discard that as redundant. */
2909 if ((re_opcode_t) *p != on_failure_jump
2910 && (re_opcode_t) *p != succeed_n)
2914 EXTRACT_NUMBER_AND_INCR (j, p);
2917 /* If what's on the stack is where we are now, pop it. */
2918 if (!FAIL_STACK_EMPTY ()
2919 && fail_stack.stack[fail_stack.avail - 1] == p)
2925 case on_failure_jump:
2926 case on_failure_keep_string_jump:
2927 handle_on_failure_jump:
2928 EXTRACT_NUMBER_AND_INCR (j, p);
2930 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2931 end of the pattern. We don't want to push such a point,
2932 since when we restore it above, entering the switch will
2933 increment `p' past the end of the pattern. We don't need
2934 to push such a point since we obviously won't find any more
2935 fastmap entries beyond `pend'. Such a pattern can match
2936 the null string, though. */
2939 if (!PUSH_PATTERN_OP (p + j, fail_stack))
2943 bufp->can_be_null = 1;
2947 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
2948 succeed_n_p = false;
2955 /* Get to the number of times to succeed. */
2958 /* Increment p past the n for when k != 0. */
2959 EXTRACT_NUMBER_AND_INCR (k, p);
2963 succeed_n_p = true; /* Spaghetti code alert. */
2964 goto handle_on_failure_jump;
2981 abort (); /* We have listed all the cases. */
2984 /* Getting here means we have found the possible starting
2985 characters for one path of the pattern -- and that the empty
2986 string does not match. We need not follow this path further.
2987 Instead, look at the next alternative (remembered on the
2988 stack), or quit if no more. The test at the top of the loop
2989 does these things. */
2990 path_can_be_null = false;
2994 /* Set `can_be_null' for the last path (also the first path, if the
2995 pattern is empty). */
2996 bufp->can_be_null |= path_can_be_null;
2998 } /* re_compile_fastmap */
3000 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3001 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3002 this memory for recording register information. STARTS and ENDS
3003 must be allocated using the malloc library routine, and must each
3004 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3006 If NUM_REGS == 0, then subsequent matches should allocate their own
3009 Unless this function is called, the first search or match using
3010 PATTERN_BUFFER will allocate its own register data, without
3011 freeing the old data. */
3014 re_set_registers (bufp, regs, num_regs, starts, ends)
3015 struct re_pattern_buffer *bufp;
3016 struct re_registers *regs;
3018 regoff_t *starts, *ends;
3022 bufp->regs_allocated = REGS_REALLOCATE;
3023 regs->num_regs = num_regs;
3024 regs->start = starts;
3029 bufp->regs_allocated = REGS_UNALLOCATED;
3031 regs->start = regs->end = (regoff_t *) 0;
3035 /* Searching routines. */
3037 /* Like re_search_2, below, but only one string is specified, and
3038 doesn't let you say where to stop matching. */
3041 re_search (bufp, string, size, startpos, range, regs)
3042 struct re_pattern_buffer *bufp;
3044 int size, startpos, range;
3045 struct re_registers *regs;
3047 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3052 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3053 virtual concatenation of STRING1 and STRING2, starting first at index
3054 STARTPOS, then at STARTPOS + 1, and so on.
3056 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3058 RANGE is how far to scan while trying to match. RANGE = 0 means try
3059 only at STARTPOS; in general, the last start tried is STARTPOS +
3062 In REGS, return the indices of the virtual concatenation of STRING1
3063 and STRING2 that matched the entire BUFP->buffer and its contained
3066 Do not consider matching one past the index STOP in the virtual
3067 concatenation of STRING1 and STRING2.
3069 We return either the position in the strings at which the match was
3070 found, -1 if no match, or -2 if error (such as failure
3074 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3075 struct re_pattern_buffer *bufp;
3076 const char *string1, *string2;
3080 struct re_registers *regs;
3084 register char *fastmap = bufp->fastmap;
3085 register char *translate = bufp->translate;
3086 int total_size = size1 + size2;
3087 int endpos = startpos + range;
3089 /* Check for out-of-range STARTPOS. */
3090 if (startpos < 0 || startpos > total_size)
3093 /* Fix up RANGE if it might eventually take us outside
3094 the virtual concatenation of STRING1 and STRING2. */
3096 range = -1 - startpos;
3097 else if (endpos > total_size)
3098 range = total_size - startpos;
3100 /* If the search isn't to be a backwards one, don't waste time in a
3101 search for a pattern that must be anchored. */
3102 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3110 /* Update the fastmap now if not correct already. */
3111 if (fastmap && !bufp->fastmap_accurate)
3112 if (re_compile_fastmap (bufp) == -2)
3115 /* Loop through the string, looking for a place to start matching. */
3118 /* If a fastmap is supplied, skip quickly over characters that
3119 cannot be the start of a match. If the pattern can match the
3120 null string, however, we don't need to skip characters; we want
3121 the first null string. */
3122 if (fastmap && startpos < total_size && !bufp->can_be_null)
3124 if (range > 0) /* Searching forwards. */
3126 register const char *d;
3127 register int lim = 0;
3130 if (startpos < size1 && startpos + range >= size1)
3131 lim = range - (size1 - startpos);
3133 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3135 /* Written out as an if-else to avoid testing `translate'
3139 && !fastmap[(unsigned char)
3140 translate[(unsigned char) *d++]])
3143 while (range > lim && !fastmap[(unsigned char) *d++])
3146 startpos += irange - range;
3148 else /* Searching backwards. */
3150 register char c = (size1 == 0 || startpos >= size1
3151 ? string2[startpos - size1]
3152 : string1[startpos]);
3154 if (!fastmap[(unsigned char) TRANSLATE (c)])
3159 /* If can't match the null string, and that's all we have left, fail. */
3160 if (range >= 0 && startpos == total_size && fastmap
3161 && !bufp->can_be_null)
3164 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3165 startpos, regs, stop);
3191 /* Declarations and macros for re_match_2. */
3193 static int bcmp_translate ();
3194 static boolean alt_match_null_string_p (),
3195 common_op_match_null_string_p (),
3196 group_match_null_string_p ();
3198 /* This converts PTR, a pointer into one of the search strings `string1'
3199 and `string2' into an offset from the beginning of that string. */
3200 #define POINTER_TO_OFFSET(ptr) \
3201 (FIRST_STRING_P (ptr) \
3202 ? ((regoff_t) ((ptr) - string1)) \
3203 : ((regoff_t) ((ptr) - string2 + size1)))
3205 /* Macros for dealing with the split strings in re_match_2. */
3207 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3209 /* Call before fetching a character with *d. This switches over to
3210 string2 if necessary. */
3211 #define PREFETCH() \
3214 /* End of string2 => fail. */ \
3215 if (dend == end_match_2) \
3217 /* End of string1 => advance to string2. */ \
3219 dend = end_match_2; \
3223 /* Test if at very beginning or at very end of the virtual concatenation
3224 of `string1' and `string2'. If only one string, it's `string2'. */
3225 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3226 #define AT_STRINGS_END(d) ((d) == end2)
3229 /* Test if D points to a character which is word-constituent. We have
3230 two special cases to check for: if past the end of string1, look at
3231 the first character in string2; and if before the beginning of
3232 string2, look at the last character in string1. */
3233 #define WORDCHAR_P(d) \
3234 (SYNTAX ((d) == end1 ? *string2 \
3235 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3238 /* Test if the character before D and the one at D differ with respect
3239 to being word-constituent. */
3240 #define AT_WORD_BOUNDARY(d) \
3241 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3242 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3245 /* Free everything we malloc. */
3246 #ifdef MATCH_MAY_ALLOCATE
3248 #define FREE_VAR(var) if (var) free (var); var = NULL
3249 #define FREE_VARIABLES() \
3251 FREE_VAR (fail_stack.stack); \
3252 FREE_VAR (regstart); \
3253 FREE_VAR (regend); \
3254 FREE_VAR (old_regstart); \
3255 FREE_VAR (old_regend); \
3256 FREE_VAR (best_regstart); \
3257 FREE_VAR (best_regend); \
3258 FREE_VAR (reg_info); \
3259 FREE_VAR (reg_dummy); \
3260 FREE_VAR (reg_info_dummy); \
3262 #else /* not REGEX_MALLOC */
3263 /* This used to do alloca (0), but now we do that in the caller. */
3264 #define FREE_VARIABLES() /* Nothing */
3265 #endif /* not REGEX_MALLOC */
3267 #define FREE_VARIABLES() /* Do nothing! */
3268 #endif /* not MATCH_MAY_ALLOCATE */
3270 /* These values must meet several constraints. They must not be valid
3271 register values; since we have a limit of 255 registers (because
3272 we use only one byte in the pattern for the register number), we can
3273 use numbers larger than 255. They must differ by 1, because of
3274 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3275 be larger than the value for the highest register, so we do not try
3276 to actually save any registers when none are active. */
3277 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3278 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3280 /* Matching routines. */
3282 #ifndef emacs /* Emacs never uses this. */
3283 /* re_match is like re_match_2 except it takes only a single string. */
3286 re_match (bufp, string, size, pos, regs)
3287 struct re_pattern_buffer *bufp;
3290 struct re_registers *regs;
3292 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3297 #endif /* not emacs */
3300 /* re_match_2 matches the compiled pattern in BUFP against the
3301 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3302 and SIZE2, respectively). We start matching at POS, and stop
3305 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3306 store offsets for the substring each group matched in REGS. See the
3307 documentation for exactly how many groups we fill.
3309 We return -1 if no match, -2 if an internal error (such as the
3310 failure stack overflowing). Otherwise, we return the length of the
3311 matched substring. */
3314 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3315 struct re_pattern_buffer *bufp;
3316 const char *string1, *string2;
3319 struct re_registers *regs;
3322 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3328 /* This is a separate function so that we can force an alloca cleanup
3331 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3332 struct re_pattern_buffer *bufp;
3333 const char *string1, *string2;
3336 struct re_registers *regs;
3339 /* General temporaries. */
3343 /* Just past the end of the corresponding string. */
3344 const char *end1, *end2;
3346 /* Pointers into string1 and string2, just past the last characters in
3347 each to consider matching. */
3348 const char *end_match_1, *end_match_2;
3350 /* Where we are in the data, and the end of the current string. */
3351 const char *d, *dend;
3353 /* Where we are in the pattern, and the end of the pattern. */
3354 unsigned char *p = bufp->buffer;
3355 register unsigned char *pend = p + bufp->used;
3357 /* Mark the opcode just after a start_memory, so we can test for an
3358 empty subpattern when we get to the stop_memory. */
3359 unsigned char *just_past_start_mem = 0;
3361 /* We use this to map every character in the string. */
3362 char *translate = bufp->translate;
3364 /* Failure point stack. Each place that can handle a failure further
3365 down the line pushes a failure point on this stack. It consists of
3366 restart, regend, and reg_info for all registers corresponding to
3367 the subexpressions we're currently inside, plus the number of such
3368 registers, and, finally, two char *'s. The first char * is where
3369 to resume scanning the pattern; the second one is where to resume
3370 scanning the strings. If the latter is zero, the failure point is
3371 a ``dummy''; if a failure happens and the failure point is a dummy,
3372 it gets discarded and the next next one is tried. */
3373 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3374 fail_stack_type fail_stack;
3377 static unsigned failure_id = 0;
3378 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3381 /* We fill all the registers internally, independent of what we
3382 return, for use in backreferences. The number here includes
3383 an element for register zero. */
3384 unsigned num_regs = bufp->re_nsub + 1;
3386 /* The currently active registers. */
3387 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3388 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3390 /* Information on the contents of registers. These are pointers into
3391 the input strings; they record just what was matched (on this
3392 attempt) by a subexpression part of the pattern, that is, the
3393 regnum-th regstart pointer points to where in the pattern we began
3394 matching and the regnum-th regend points to right after where we
3395 stopped matching the regnum-th subexpression. (The zeroth register
3396 keeps track of what the whole pattern matches.) */
3397 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3398 const char **regstart, **regend;
3401 /* If a group that's operated upon by a repetition operator fails to
3402 match anything, then the register for its start will need to be
3403 restored because it will have been set to wherever in the string we
3404 are when we last see its open-group operator. Similarly for a
3406 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3407 const char **old_regstart, **old_regend;
3410 /* The is_active field of reg_info helps us keep track of which (possibly
3411 nested) subexpressions we are currently in. The matched_something
3412 field of reg_info[reg_num] helps us tell whether or not we have
3413 matched any of the pattern so far this time through the reg_num-th
3414 subexpression. These two fields get reset each time through any
3415 loop their register is in. */
3416 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3417 register_info_type *reg_info;
3420 /* The following record the register info as found in the above
3421 variables when we find a match better than any we've seen before.
3422 This happens as we backtrack through the failure points, which in
3423 turn happens only if we have not yet matched the entire string. */
3424 unsigned best_regs_set = false;
3425 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3426 const char **best_regstart, **best_regend;
3429 /* Logically, this is `best_regend[0]'. But we don't want to have to
3430 allocate space for that if we're not allocating space for anything
3431 else (see below). Also, we never need info about register 0 for
3432 any of the other register vectors, and it seems rather a kludge to
3433 treat `best_regend' differently than the rest. So we keep track of
3434 the end of the best match so far in a separate variable. We
3435 initialize this to NULL so that when we backtrack the first time
3436 and need to test it, it's not garbage. */
3437 const char *match_end = NULL;
3439 /* Used when we pop values we don't care about. */
3440 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3441 const char **reg_dummy;
3442 register_info_type *reg_info_dummy;
3446 /* Counts the total number of registers pushed. */
3447 unsigned num_regs_pushed = 0;
3450 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3454 #ifdef MATCH_MAY_ALLOCATE
3455 /* Do not bother to initialize all the register variables if there are
3456 no groups in the pattern, as it takes a fair amount of time. If
3457 there are groups, we include space for register 0 (the whole
3458 pattern), even though we never use it, since it simplifies the
3459 array indexing. We should fix this. */
3462 regstart = REGEX_TALLOC (num_regs, const char *);
3463 regend = REGEX_TALLOC (num_regs, const char *);
3464 old_regstart = REGEX_TALLOC (num_regs, const char *);
3465 old_regend = REGEX_TALLOC (num_regs, const char *);
3466 best_regstart = REGEX_TALLOC (num_regs, const char *);
3467 best_regend = REGEX_TALLOC (num_regs, const char *);
3468 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3469 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3470 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3472 if (!(regstart && regend && old_regstart && old_regend && reg_info
3473 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3479 #if defined (REGEX_MALLOC)
3482 /* We must initialize all our variables to NULL, so that
3483 `FREE_VARIABLES' doesn't try to free them. */
3484 regstart = regend = old_regstart = old_regend = best_regstart
3485 = best_regend = reg_dummy = NULL;
3486 reg_info = reg_info_dummy = (register_info_type *) NULL;
3488 #endif /* REGEX_MALLOC */
3489 #endif /* MATCH_MAY_ALLOCATE */
3491 /* The starting position is bogus. */
3492 if (pos < 0 || pos > size1 + size2)
3498 /* Initialize subexpression text positions to -1 to mark ones that no
3499 start_memory/stop_memory has been seen for. Also initialize the
3500 register information struct. */
3501 for (mcnt = 1; mcnt < num_regs; mcnt++)
3503 regstart[mcnt] = regend[mcnt]
3504 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3506 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3507 IS_ACTIVE (reg_info[mcnt]) = 0;
3508 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3509 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3512 /* We move `string1' into `string2' if the latter's empty -- but not if
3513 `string1' is null. */
3514 if (size2 == 0 && string1 != NULL)
3521 end1 = string1 + size1;
3522 end2 = string2 + size2;
3524 /* Compute where to stop matching, within the two strings. */
3527 end_match_1 = string1 + stop;
3528 end_match_2 = string2;
3533 end_match_2 = string2 + stop - size1;
3536 /* `p' scans through the pattern as `d' scans through the data.
3537 `dend' is the end of the input string that `d' points within. `d'
3538 is advanced into the following input string whenever necessary, but
3539 this happens before fetching; therefore, at the beginning of the
3540 loop, `d' can be pointing at the end of a string, but it cannot
3542 if (size1 > 0 && pos <= size1)
3549 d = string2 + pos - size1;
3553 DEBUG_PRINT1 ("The compiled pattern is: ");
3554 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3555 DEBUG_PRINT1 ("The string to match is: `");
3556 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3557 DEBUG_PRINT1 ("'\n");
3559 /* This loops over pattern commands. It exits by returning from the
3560 function if the match is complete, or it drops through if the match
3561 fails at this starting point in the input data. */
3564 DEBUG_PRINT2 ("\n0x%x: ", p);
3567 { /* End of pattern means we might have succeeded. */
3568 DEBUG_PRINT1 ("end of pattern ... ");
3570 /* If we haven't matched the entire string, and we want the
3571 longest match, try backtracking. */
3572 if (d != end_match_2)
3574 DEBUG_PRINT1 ("backtracking.\n");
3576 if (!FAIL_STACK_EMPTY ())
3577 { /* More failure points to try. */
3578 boolean same_str_p = (FIRST_STRING_P (match_end)
3579 == MATCHING_IN_FIRST_STRING);
3581 /* If exceeds best match so far, save it. */
3583 || (same_str_p && d > match_end)
3584 || (!same_str_p && !MATCHING_IN_FIRST_STRING))
3586 best_regs_set = true;
3589 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3591 for (mcnt = 1; mcnt < num_regs; mcnt++)
3593 best_regstart[mcnt] = regstart[mcnt];
3594 best_regend[mcnt] = regend[mcnt];
3600 /* If no failure points, don't restore garbage. */
3601 else if (best_regs_set)
3604 /* Restore best match. It may happen that `dend ==
3605 end_match_1' while the restored d is in string2.
3606 For example, the pattern `x.*y.*z' against the
3607 strings `x-' and `y-z-', if the two strings are
3608 not consecutive in memory. */
3609 DEBUG_PRINT1 ("Restoring best registers.\n");
3612 dend = ((d >= string1 && d <= end1)
3613 ? end_match_1 : end_match_2);
3615 for (mcnt = 1; mcnt < num_regs; mcnt++)
3617 regstart[mcnt] = best_regstart[mcnt];
3618 regend[mcnt] = best_regend[mcnt];
3621 } /* d != end_match_2 */
3623 DEBUG_PRINT1 ("Accepting match.\n");
3625 /* If caller wants register contents data back, do it. */
3626 if (regs && !bufp->no_sub)
3628 /* Have the register data arrays been allocated? */
3629 if (bufp->regs_allocated == REGS_UNALLOCATED)
3630 { /* No. So allocate them with malloc. We need one
3631 extra element beyond `num_regs' for the `-1' marker
3633 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3634 regs->start = TALLOC (regs->num_regs, regoff_t);
3635 regs->end = TALLOC (regs->num_regs, regoff_t);
3636 if (regs->start == NULL || regs->end == NULL)
3638 bufp->regs_allocated = REGS_REALLOCATE;
3640 else if (bufp->regs_allocated == REGS_REALLOCATE)
3641 { /* Yes. If we need more elements than were already
3642 allocated, reallocate them. If we need fewer, just
3644 if (regs->num_regs < num_regs + 1)
3646 regs->num_regs = num_regs + 1;
3647 RETALLOC (regs->start, regs->num_regs, regoff_t);
3648 RETALLOC (regs->end, regs->num_regs, regoff_t);
3649 if (regs->start == NULL || regs->end == NULL)
3655 /* These braces fend off a "empty body in an else-statement"
3656 warning under GCC when assert expands to nothing. */
3657 assert (bufp->regs_allocated == REGS_FIXED);
3660 /* Convert the pointer data in `regstart' and `regend' to
3661 indices. Register zero has to be set differently,
3662 since we haven't kept track of any info for it. */
3663 if (regs->num_regs > 0)
3665 regs->start[0] = pos;
3666 regs->end[0] = (MATCHING_IN_FIRST_STRING
3667 ? ((regoff_t) (d - string1))
3668 : ((regoff_t) (d - string2 + size1)));
3671 /* Go through the first `min (num_regs, regs->num_regs)'
3672 registers, since that is all we initialized. */
3673 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3675 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3676 regs->start[mcnt] = regs->end[mcnt] = -1;
3680 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3682 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3686 /* If the regs structure we return has more elements than
3687 were in the pattern, set the extra elements to -1. If
3688 we (re)allocated the registers, this is the case,
3689 because we always allocate enough to have at least one
3691 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3692 regs->start[mcnt] = regs->end[mcnt] = -1;
3693 } /* regs && !bufp->no_sub */
3696 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3697 nfailure_points_pushed, nfailure_points_popped,
3698 nfailure_points_pushed - nfailure_points_popped);
3699 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3701 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3705 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3710 /* Otherwise match next pattern command. */
3711 #ifdef SWITCH_ENUM_BUG
3712 switch ((int) ((re_opcode_t) *p++))
3714 switch ((re_opcode_t) *p++)
3717 /* Ignore these. Used to ignore the n of succeed_n's which
3718 currently have n == 0. */
3720 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3724 /* Match the next n pattern characters exactly. The following
3725 byte in the pattern defines n, and the n bytes after that
3726 are the characters to match. */
3729 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3731 /* This is written out as an if-else so we don't waste time
3732 testing `translate' inside the loop. */
3738 if (translate[(unsigned char) *d++] != (char) *p++)
3748 if (*d++ != (char) *p++) goto fail;
3752 SET_REGS_MATCHED ();
3756 /* Match any character except possibly a newline or a null. */
3758 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3762 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3763 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3766 SET_REGS_MATCHED ();
3767 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3775 register unsigned char c;
3776 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3778 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3781 c = TRANSLATE (*d); /* The character to match. */
3783 /* Cast to `unsigned' instead of `unsigned char' in case the
3784 bit list is a full 32 bytes long. */
3785 if (c < (unsigned) (*p * BYTEWIDTH)
3786 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3791 if (!not) goto fail;
3793 SET_REGS_MATCHED ();
3799 /* The beginning of a group is represented by start_memory.
3800 The arguments are the register number in the next byte, and the
3801 number of groups inner to this one in the next. The text
3802 matched within the group is recorded (in the internal
3803 registers data structure) under the register number. */
3805 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3807 /* Find out if this group can match the empty string. */
3808 p1 = p; /* To send to group_match_null_string_p. */
3810 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3811 REG_MATCH_NULL_STRING_P (reg_info[*p])
3812 = group_match_null_string_p (&p1, pend, reg_info);
3814 /* Save the position in the string where we were the last time
3815 we were at this open-group operator in case the group is
3816 operated upon by a repetition operator, e.g., with `(a*)*b'
3817 against `ab'; then we want to ignore where we are now in
3818 the string in case this attempt to match fails. */
3819 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3820 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3822 DEBUG_PRINT2 (" old_regstart: %d\n",
3823 POINTER_TO_OFFSET (old_regstart[*p]));
3826 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
3828 IS_ACTIVE (reg_info[*p]) = 1;
3829 MATCHED_SOMETHING (reg_info[*p]) = 0;
3831 /* This is the new highest active register. */
3832 highest_active_reg = *p;
3834 /* If nothing was active before, this is the new lowest active
3836 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3837 lowest_active_reg = *p;
3839 /* Move past the register number and inner group count. */
3841 just_past_start_mem = p;
3845 /* The stop_memory opcode represents the end of a group. Its
3846 arguments are the same as start_memory's: the register
3847 number, and the number of inner groups. */
3849 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
3851 /* We need to save the string position the last time we were at
3852 this close-group operator in case the group is operated
3853 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3854 against `aba'; then we want to ignore where we are now in
3855 the string in case this attempt to match fails. */
3856 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3857 ? REG_UNSET (regend[*p]) ? d : regend[*p]
3859 DEBUG_PRINT2 (" old_regend: %d\n",
3860 POINTER_TO_OFFSET (old_regend[*p]));
3863 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
3865 /* This register isn't active anymore. */
3866 IS_ACTIVE (reg_info[*p]) = 0;
3868 /* If this was the only register active, nothing is active
3870 if (lowest_active_reg == highest_active_reg)
3872 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3873 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3876 { /* We must scan for the new highest active register, since
3877 it isn't necessarily one less than now: consider
3878 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3879 new highest active register is 1. */
3880 unsigned char r = *p - 1;
3881 while (r > 0 && !IS_ACTIVE (reg_info[r]))
3884 /* If we end up at register zero, that means that we saved
3885 the registers as the result of an `on_failure_jump', not
3886 a `start_memory', and we jumped to past the innermost
3887 `stop_memory'. For example, in ((.)*) we save
3888 registers 1 and 2 as a result of the *, but when we pop
3889 back to the second ), we are at the stop_memory 1.
3890 Thus, nothing is active. */
3893 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3894 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3897 highest_active_reg = r;
3900 /* If just failed to match something this time around with a
3901 group that's operated on by a repetition operator, try to
3902 force exit from the ``loop'', and restore the register
3903 information for this group that we had before trying this
3905 if ((!MATCHED_SOMETHING (reg_info[*p])
3906 || just_past_start_mem == p - 1)
3909 boolean is_a_jump_n = false;
3913 switch ((re_opcode_t) *p1++)
3917 case pop_failure_jump:
3918 case maybe_pop_jump:
3920 case dummy_failure_jump:
3921 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3931 /* If the next operation is a jump backwards in the pattern
3932 to an on_failure_jump right before the start_memory
3933 corresponding to this stop_memory, exit from the loop
3934 by forcing a failure after pushing on the stack the
3935 on_failure_jump's jump in the pattern, and d. */
3936 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
3937 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
3939 /* If this group ever matched anything, then restore
3940 what its registers were before trying this last
3941 failed match, e.g., with `(a*)*b' against `ab' for
3942 regstart[1], and, e.g., with `((a*)*(b*)*)*'
3943 against `aba' for regend[3].
3945 Also restore the registers for inner groups for,
3946 e.g., `((a*)(b*))*' against `aba' (register 3 would
3947 otherwise get trashed). */
3949 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
3953 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
3955 /* Restore this and inner groups' (if any) registers. */
3956 for (r = *p; r < *p + *(p + 1); r++)
3958 regstart[r] = old_regstart[r];
3960 /* xx why this test? */
3961 if ((int) old_regend[r] >= (int) regstart[r])
3962 regend[r] = old_regend[r];
3966 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3967 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
3973 /* Move past the register number and the inner group count. */
3978 /* \<digit> has been turned into a `duplicate' command which is
3979 followed by the numeric value of <digit> as the register number. */
3982 register const char *d2, *dend2;
3983 int regno = *p++; /* Get which register to match against. */
3984 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
3986 /* Can't back reference a group which we've never matched. */
3987 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
3990 /* Where in input to try to start matching. */
3991 d2 = regstart[regno];
3993 /* Where to stop matching; if both the place to start and
3994 the place to stop matching are in the same string, then
3995 set to the place to stop, otherwise, for now have to use
3996 the end of the first string. */
3998 dend2 = ((FIRST_STRING_P (regstart[regno])
3999 == FIRST_STRING_P (regend[regno]))
4000 ? regend[regno] : end_match_1);
4003 /* If necessary, advance to next segment in register
4007 if (dend2 == end_match_2) break;
4008 if (dend2 == regend[regno]) break;
4010 /* End of string1 => advance to string2. */
4012 dend2 = regend[regno];
4014 /* At end of register contents => success */
4015 if (d2 == dend2) break;
4017 /* If necessary, advance to next segment in data. */
4020 /* How many characters left in this segment to match. */
4023 /* Want how many consecutive characters we can match in
4024 one shot, so, if necessary, adjust the count. */
4025 if (mcnt > dend2 - d2)
4028 /* Compare that many; failure if mismatch, else move
4031 ? bcmp_translate (d, d2, mcnt, translate)
4032 : bcmp (d, d2, mcnt))
4034 d += mcnt, d2 += mcnt;
4040 /* begline matches the empty string at the beginning of the string
4041 (unless `not_bol' is set in `bufp'), and, if
4042 `newline_anchor' is set, after newlines. */
4044 DEBUG_PRINT1 ("EXECUTING begline.\n");
4046 if (AT_STRINGS_BEG (d))
4048 if (!bufp->not_bol) break;
4050 else if (d[-1] == '\n' && bufp->newline_anchor)
4054 /* In all other cases, we fail. */
4058 /* endline is the dual of begline. */
4060 DEBUG_PRINT1 ("EXECUTING endline.\n");
4062 if (AT_STRINGS_END (d))
4064 if (!bufp->not_eol) break;
4067 /* We have to ``prefetch'' the next character. */
4068 else if ((d == end1 ? *string2 : *d) == '\n'
4069 && bufp->newline_anchor)
4076 /* Match at the very beginning of the data. */
4078 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4079 if (AT_STRINGS_BEG (d))
4084 /* Match at the very end of the data. */
4086 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4087 if (AT_STRINGS_END (d))
4092 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4093 pushes NULL as the value for the string on the stack. Then
4094 `pop_failure_point' will keep the current value for the
4095 string, instead of restoring it. To see why, consider
4096 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4097 then the . fails against the \n. But the next thing we want
4098 to do is match the \n against the \n; if we restored the
4099 string value, we would be back at the foo.
4101 Because this is used only in specific cases, we don't need to
4102 check all the things that `on_failure_jump' does, to make
4103 sure the right things get saved on the stack. Hence we don't
4104 share its code. The only reason to push anything on the
4105 stack at all is that otherwise we would have to change
4106 `anychar's code to do something besides goto fail in this
4107 case; that seems worse than this. */
4108 case on_failure_keep_string_jump:
4109 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4111 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4112 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4114 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4118 /* Uses of on_failure_jump:
4120 Each alternative starts with an on_failure_jump that points
4121 to the beginning of the next alternative. Each alternative
4122 except the last ends with a jump that in effect jumps past
4123 the rest of the alternatives. (They really jump to the
4124 ending jump of the following alternative, because tensioning
4125 these jumps is a hassle.)
4127 Repeats start with an on_failure_jump that points past both
4128 the repetition text and either the following jump or
4129 pop_failure_jump back to this on_failure_jump. */
4130 case on_failure_jump:
4132 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4134 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4135 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4137 /* If this on_failure_jump comes right before a group (i.e.,
4138 the original * applied to a group), save the information
4139 for that group and all inner ones, so that if we fail back
4140 to this point, the group's information will be correct.
4141 For example, in \(a*\)*\1, we need the preceding group,
4142 and in \(\(a*\)b*\)\2, we need the inner group. */
4144 /* We can't use `p' to check ahead because we push
4145 a failure point to `p + mcnt' after we do this. */
4148 /* We need to skip no_op's before we look for the
4149 start_memory in case this on_failure_jump is happening as
4150 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4152 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4155 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4157 /* We have a new highest active register now. This will
4158 get reset at the start_memory we are about to get to,
4159 but we will have saved all the registers relevant to
4160 this repetition op, as described above. */
4161 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4162 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4163 lowest_active_reg = *(p1 + 1);
4166 DEBUG_PRINT1 (":\n");
4167 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4171 /* A smart repeat ends with `maybe_pop_jump'.
4172 We change it to either `pop_failure_jump' or `jump'. */
4173 case maybe_pop_jump:
4174 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4175 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4177 register unsigned char *p2 = p;
4179 /* Compare the beginning of the repeat with what in the
4180 pattern follows its end. If we can establish that there
4181 is nothing that they would both match, i.e., that we
4182 would have to backtrack because of (as in, e.g., `a*a')
4183 then we can change to pop_failure_jump, because we'll
4184 never have to backtrack.
4186 This is not true in the case of alternatives: in
4187 `(a|ab)*' we do need to backtrack to the `ab' alternative
4188 (e.g., if the string was `ab'). But instead of trying to
4189 detect that here, the alternative has put on a dummy
4190 failure point which is what we will end up popping. */
4192 /* Skip over open/close-group commands.
4193 If what follows this loop is a ...+ construct,
4194 look at what begins its body, since we will have to
4195 match at least one of that. */
4199 && ((re_opcode_t) *p2 == stop_memory
4200 || (re_opcode_t) *p2 == start_memory))
4202 else if (p2 + 6 < pend
4203 && (re_opcode_t) *p2 == dummy_failure_jump)
4210 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4211 to the `maybe_finalize_jump' of this case. Examine what
4214 /* If we're at the end of the pattern, we can change. */
4217 /* Consider what happens when matching ":\(.*\)"
4218 against ":/". I don't really understand this code
4220 p[-3] = (unsigned char) pop_failure_jump;
4222 (" End of pattern: change to `pop_failure_jump'.\n");
4225 else if ((re_opcode_t) *p2 == exactn
4226 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4228 register unsigned char c
4229 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4231 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4233 p[-3] = (unsigned char) pop_failure_jump;
4234 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4238 else if ((re_opcode_t) p1[3] == charset
4239 || (re_opcode_t) p1[3] == charset_not)
4241 int not = (re_opcode_t) p1[3] == charset_not;
4243 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4244 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4247 /* `not' is equal to 1 if c would match, which means
4248 that we can't change to pop_failure_jump. */
4251 p[-3] = (unsigned char) pop_failure_jump;
4252 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4256 else if ((re_opcode_t) *p2 == charset)
4258 register unsigned char c
4259 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4261 if ((re_opcode_t) p1[3] == exactn
4262 && ! (p2[1] * BYTEWIDTH > p1[4]
4263 && (p2[1 + p1[4] / BYTEWIDTH]
4264 & (1 << (p1[4] % BYTEWIDTH)))))
4266 p[-3] = (unsigned char) pop_failure_jump;
4267 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4271 else if ((re_opcode_t) p1[3] == charset_not)
4274 /* We win if the charset_not inside the loop
4275 lists every character listed in the charset after. */
4276 for (idx = 0; idx < p2[1]; idx++)
4277 if (! (p2[2 + idx] == 0
4279 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4284 p[-3] = (unsigned char) pop_failure_jump;
4285 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4288 else if ((re_opcode_t) p1[3] == charset)
4291 /* We win if the charset inside the loop
4292 has no overlap with the one after the loop. */
4293 for (idx = 0; idx < p2[1] && idx < p1[4]; idx++)
4294 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4297 if (idx == p2[1] || idx == p1[4])
4299 p[-3] = (unsigned char) pop_failure_jump;
4300 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4305 p -= 2; /* Point at relative address again. */
4306 if ((re_opcode_t) p[-1] != pop_failure_jump)
4308 p[-1] = (unsigned char) jump;
4309 DEBUG_PRINT1 (" Match => jump.\n");
4310 goto unconditional_jump;
4312 /* Note fall through. */
4315 /* The end of a simple repeat has a pop_failure_jump back to
4316 its matching on_failure_jump, where the latter will push a
4317 failure point. The pop_failure_jump takes off failure
4318 points put on by this pop_failure_jump's matching
4319 on_failure_jump; we got through the pattern to here from the
4320 matching on_failure_jump, so didn't fail. */
4321 case pop_failure_jump:
4323 /* We need to pass separate storage for the lowest and
4324 highest registers, even though we don't care about the
4325 actual values. Otherwise, we will restore only one
4326 register from the stack, since lowest will == highest in
4327 `pop_failure_point'. */
4328 unsigned dummy_low_reg, dummy_high_reg;
4329 unsigned char *pdummy;
4332 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4333 POP_FAILURE_POINT (sdummy, pdummy,
4334 dummy_low_reg, dummy_high_reg,
4335 reg_dummy, reg_dummy, reg_info_dummy);
4337 /* Note fall through. */
4340 /* Unconditionally jump (without popping any failure points). */
4343 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4344 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4345 p += mcnt; /* Do the jump. */
4346 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4350 /* We need this opcode so we can detect where alternatives end
4351 in `group_match_null_string_p' et al. */
4353 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4354 goto unconditional_jump;
4357 /* Normally, the on_failure_jump pushes a failure point, which
4358 then gets popped at pop_failure_jump. We will end up at
4359 pop_failure_jump, also, and with a pattern of, say, `a+', we
4360 are skipping over the on_failure_jump, so we have to push
4361 something meaningless for pop_failure_jump to pop. */
4362 case dummy_failure_jump:
4363 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4364 /* It doesn't matter what we push for the string here. What
4365 the code at `fail' tests is the value for the pattern. */
4366 PUSH_FAILURE_POINT (0, 0, -2);
4367 goto unconditional_jump;
4370 /* At the end of an alternative, we need to push a dummy failure
4371 point in case we are followed by a `pop_failure_jump', because
4372 we don't want the failure point for the alternative to be
4373 popped. For example, matching `(a|ab)*' against `aab'
4374 requires that we match the `ab' alternative. */
4375 case push_dummy_failure:
4376 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4377 /* See comments just above at `dummy_failure_jump' about the
4379 PUSH_FAILURE_POINT (0, 0, -2);
4382 /* Have to succeed matching what follows at least n times.
4383 After that, handle like `on_failure_jump'. */
4385 EXTRACT_NUMBER (mcnt, p + 2);
4386 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4389 /* Originally, this is how many times we HAVE to succeed. */
4394 STORE_NUMBER_AND_INCR (p, mcnt);
4395 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4399 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4400 p[2] = (unsigned char) no_op;
4401 p[3] = (unsigned char) no_op;
4407 EXTRACT_NUMBER (mcnt, p + 2);
4408 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4410 /* Originally, this is how many times we CAN jump. */
4414 STORE_NUMBER (p + 2, mcnt);
4415 goto unconditional_jump;
4417 /* If don't have to jump any more, skip over the rest of command. */
4424 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4426 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4428 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4429 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4430 STORE_NUMBER (p1, mcnt);
4435 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4436 if (AT_WORD_BOUNDARY (d))
4441 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4442 if (AT_WORD_BOUNDARY (d))
4447 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4448 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4453 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4454 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4455 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4461 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4462 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4467 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4468 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4473 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4474 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4477 #if 0 /* not emacs19 */
4479 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4480 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4483 #endif /* not emacs19 */
4486 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4491 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4495 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4497 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4499 SET_REGS_MATCHED ();
4503 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4505 goto matchnotsyntax;
4508 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4512 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4514 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4516 SET_REGS_MATCHED ();
4519 #else /* not emacs */
4521 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4523 if (!WORDCHAR_P (d))
4525 SET_REGS_MATCHED ();
4530 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4534 SET_REGS_MATCHED ();
4537 #endif /* not emacs */
4542 continue; /* Successfully executed one pattern command; keep going. */
4545 /* We goto here if a matching operation fails. */
4547 if (!FAIL_STACK_EMPTY ())
4548 { /* A restart point is known. Restore to that state. */
4549 DEBUG_PRINT1 ("\nFAIL:\n");
4550 POP_FAILURE_POINT (d, p,
4551 lowest_active_reg, highest_active_reg,
4552 regstart, regend, reg_info);
4554 /* If this failure point is a dummy, try the next one. */
4558 /* If we failed to the end of the pattern, don't examine *p. */
4562 boolean is_a_jump_n = false;
4564 /* If failed to a backwards jump that's part of a repetition
4565 loop, need to pop this failure point and use the next one. */
4566 switch ((re_opcode_t) *p)
4570 case maybe_pop_jump:
4571 case pop_failure_jump:
4574 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4577 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4579 && (re_opcode_t) *p1 == on_failure_jump))
4587 if (d >= string1 && d <= end1)
4591 break; /* Matching at this starting point really fails. */
4595 goto restore_best_regs;
4599 return -1; /* Failure to match. */
4602 /* Subroutine definitions for re_match_2. */
4605 /* We are passed P pointing to a register number after a start_memory.
4607 Return true if the pattern up to the corresponding stop_memory can
4608 match the empty string, and false otherwise.
4610 If we find the matching stop_memory, sets P to point to one past its number.
4611 Otherwise, sets P to an undefined byte less than or equal to END.
4613 We don't handle duplicates properly (yet). */
4616 group_match_null_string_p (p, end, reg_info)
4617 unsigned char **p, *end;
4618 register_info_type *reg_info;
4621 /* Point to after the args to the start_memory. */
4622 unsigned char *p1 = *p + 2;
4626 /* Skip over opcodes that can match nothing, and return true or
4627 false, as appropriate, when we get to one that can't, or to the
4628 matching stop_memory. */
4630 switch ((re_opcode_t) *p1)
4632 /* Could be either a loop or a series of alternatives. */
4633 case on_failure_jump:
4635 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4637 /* If the next operation is not a jump backwards in the
4642 /* Go through the on_failure_jumps of the alternatives,
4643 seeing if any of the alternatives cannot match nothing.
4644 The last alternative starts with only a jump,
4645 whereas the rest start with on_failure_jump and end
4646 with a jump, e.g., here is the pattern for `a|b|c':
4648 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4649 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4652 So, we have to first go through the first (n-1)
4653 alternatives and then deal with the last one separately. */
4656 /* Deal with the first (n-1) alternatives, which start
4657 with an on_failure_jump (see above) that jumps to right
4658 past a jump_past_alt. */
4660 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4662 /* `mcnt' holds how many bytes long the alternative
4663 is, including the ending `jump_past_alt' and
4666 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4670 /* Move to right after this alternative, including the
4674 /* Break if it's the beginning of an n-th alternative
4675 that doesn't begin with an on_failure_jump. */
4676 if ((re_opcode_t) *p1 != on_failure_jump)
4679 /* Still have to check that it's not an n-th
4680 alternative that starts with an on_failure_jump. */
4682 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4683 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4685 /* Get to the beginning of the n-th alternative. */
4691 /* Deal with the last alternative: go back and get number
4692 of the `jump_past_alt' just before it. `mcnt' contains
4693 the length of the alternative. */
4694 EXTRACT_NUMBER (mcnt, p1 - 2);
4696 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4699 p1 += mcnt; /* Get past the n-th alternative. */
4705 assert (p1[1] == **p);
4711 if (!common_op_match_null_string_p (&p1, end, reg_info))
4714 } /* while p1 < end */
4717 } /* group_match_null_string_p */
4720 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4721 It expects P to be the first byte of a single alternative and END one
4722 byte past the last. The alternative can contain groups. */
4725 alt_match_null_string_p (p, end, reg_info)
4726 unsigned char *p, *end;
4727 register_info_type *reg_info;
4730 unsigned char *p1 = p;
4734 /* Skip over opcodes that can match nothing, and break when we get
4735 to one that can't. */
4737 switch ((re_opcode_t) *p1)
4740 case on_failure_jump:
4742 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4747 if (!common_op_match_null_string_p (&p1, end, reg_info))
4750 } /* while p1 < end */
4753 } /* alt_match_null_string_p */
4756 /* Deals with the ops common to group_match_null_string_p and
4757 alt_match_null_string_p.
4759 Sets P to one after the op and its arguments, if any. */
4762 common_op_match_null_string_p (p, end, reg_info)
4763 unsigned char **p, *end;
4764 register_info_type *reg_info;
4769 unsigned char *p1 = *p;
4771 switch ((re_opcode_t) *p1++)
4791 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4792 ret = group_match_null_string_p (&p1, end, reg_info);
4794 /* Have to set this here in case we're checking a group which
4795 contains a group and a back reference to it. */
4797 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4798 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4804 /* If this is an optimized succeed_n for zero times, make the jump. */
4806 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4814 /* Get to the number of times to succeed. */
4816 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4821 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4829 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
4837 /* All other opcodes mean we cannot match the empty string. */
4843 } /* common_op_match_null_string_p */
4846 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4847 bytes; nonzero otherwise. */
4850 bcmp_translate (s1, s2, len, translate)
4851 unsigned char *s1, *s2;
4855 register unsigned char *p1 = s1, *p2 = s2;
4858 if (translate[*p1++] != translate[*p2++]) return 1;
4864 /* Entry points for GNU code. */
4866 /* re_compile_pattern is the GNU regular expression compiler: it
4867 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4868 Returns 0 if the pattern was valid, otherwise an error string.
4870 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4871 are set in BUFP on entry.
4873 We call regex_compile to do the actual compilation. */
4876 re_compile_pattern (pattern, length, bufp)
4877 const char *pattern;
4879 struct re_pattern_buffer *bufp;
4883 /* GNU code is written to assume at least RE_NREGS registers will be set
4884 (and at least one extra will be -1). */
4885 bufp->regs_allocated = REGS_UNALLOCATED;
4887 /* And GNU code determines whether or not to get register information
4888 by passing null for the REGS argument to re_match, etc., not by
4892 /* Match anchors at newline. */
4893 bufp->newline_anchor = 1;
4895 ret = regex_compile (pattern, length, re_syntax_options, bufp);
4897 return re_error_msg[(int) ret];
4900 /* Entry points compatible with 4.2 BSD regex library. We don't define
4901 them if this is an Emacs or POSIX compilation. */
4903 #if !defined (emacs) && !defined (_POSIX_SOURCE)
4905 /* BSD has one and only one pattern buffer. */
4906 static struct re_pattern_buffer re_comp_buf;
4916 if (!re_comp_buf.buffer)
4917 return "No previous regular expression";
4921 if (!re_comp_buf.buffer)
4923 re_comp_buf.buffer = (unsigned char *) malloc (200);
4924 if (re_comp_buf.buffer == NULL)
4925 return "Memory exhausted";
4926 re_comp_buf.allocated = 200;
4928 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
4929 if (re_comp_buf.fastmap == NULL)
4930 return "Memory exhausted";
4933 /* Since `re_exec' always passes NULL for the `regs' argument, we
4934 don't need to initialize the pattern buffer fields which affect it. */
4936 /* Match anchors at newlines. */
4937 re_comp_buf.newline_anchor = 1;
4939 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
4941 /* Yes, we're discarding `const' here. */
4942 return (char *) re_error_msg[(int) ret];
4950 const int len = strlen (s);
4952 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
4954 #endif /* not emacs and not _POSIX_SOURCE */
4956 /* POSIX.2 functions. Don't define these for Emacs. */
4960 /* regcomp takes a regular expression as a string and compiles it.
4962 PREG is a regex_t *. We do not expect any fields to be initialized,
4963 since POSIX says we shouldn't. Thus, we set
4965 `buffer' to the compiled pattern;
4966 `used' to the length of the compiled pattern;
4967 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
4968 REG_EXTENDED bit in CFLAGS is set; otherwise, to
4969 RE_SYNTAX_POSIX_BASIC;
4970 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
4971 `fastmap' and `fastmap_accurate' to zero;
4972 `re_nsub' to the number of subexpressions in PATTERN.
4974 PATTERN is the address of the pattern string.
4976 CFLAGS is a series of bits which affect compilation.
4978 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
4979 use POSIX basic syntax.
4981 If REG_NEWLINE is set, then . and [^...] don't match newline.
4982 Also, regexec will try a match beginning after every newline.
4984 If REG_ICASE is set, then we considers upper- and lowercase
4985 versions of letters to be equivalent when matching.
4987 If REG_NOSUB is set, then when PREG is passed to regexec, that
4988 routine will report only success or failure, and nothing about the
4991 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
4992 the return codes and their meanings.) */
4995 regcomp (preg, pattern, cflags)
4997 const char *pattern;
5002 = (cflags & REG_EXTENDED) ?
5003 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5005 /* regex_compile will allocate the space for the compiled pattern. */
5007 preg->allocated = 0;
5010 /* Don't bother to use a fastmap when searching. This simplifies the
5011 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5012 characters after newlines into the fastmap. This way, we just try
5016 if (cflags & REG_ICASE)
5020 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5021 if (preg->translate == NULL)
5022 return (int) REG_ESPACE;
5024 /* Map uppercase characters to corresponding lowercase ones. */
5025 for (i = 0; i < CHAR_SET_SIZE; i++)
5026 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5029 preg->translate = NULL;
5031 /* If REG_NEWLINE is set, newlines are treated differently. */
5032 if (cflags & REG_NEWLINE)
5033 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5034 syntax &= ~RE_DOT_NEWLINE;
5035 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5036 /* It also changes the matching behavior. */
5037 preg->newline_anchor = 1;
5040 preg->newline_anchor = 0;
5042 preg->no_sub = !!(cflags & REG_NOSUB);
5044 /* POSIX says a null character in the pattern terminates it, so we
5045 can use strlen here in compiling the pattern. */
5046 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5048 /* POSIX doesn't distinguish between an unmatched open-group and an
5049 unmatched close-group: both are REG_EPAREN. */
5050 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5056 /* regexec searches for a given pattern, specified by PREG, in the
5059 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5060 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5061 least NMATCH elements, and we set them to the offsets of the
5062 corresponding matched substrings.
5064 EFLAGS specifies `execution flags' which affect matching: if
5065 REG_NOTBOL is set, then ^ does not match at the beginning of the
5066 string; if REG_NOTEOL is set, then $ does not match at the end.
5068 We return 0 if we find a match and REG_NOMATCH if not. */
5071 regexec (preg, string, nmatch, pmatch, eflags)
5072 const regex_t *preg;
5075 regmatch_t pmatch[];
5079 struct re_registers regs;
5080 regex_t private_preg;
5081 int len = strlen (string);
5082 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5084 private_preg = *preg;
5086 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5087 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5089 /* The user has told us exactly how many registers to return
5090 information about, via `nmatch'. We have to pass that on to the
5091 matching routines. */
5092 private_preg.regs_allocated = REGS_FIXED;
5096 regs.num_regs = nmatch;
5097 regs.start = TALLOC (nmatch, regoff_t);
5098 regs.end = TALLOC (nmatch, regoff_t);
5099 if (regs.start == NULL || regs.end == NULL)
5100 return (int) REG_NOMATCH;
5103 /* Perform the searching operation. */
5104 ret = re_search (&private_preg, string, len,
5105 /* start: */ 0, /* range: */ len,
5106 want_reg_info ? ®s : (struct re_registers *) 0);
5108 /* Copy the register information to the POSIX structure. */
5115 for (r = 0; r < nmatch; r++)
5117 pmatch[r].rm_so = regs.start[r];
5118 pmatch[r].rm_eo = regs.end[r];
5122 /* If we needed the temporary register info, free the space now. */
5127 /* We want zero return to mean success, unlike `re_search'. */
5128 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5132 /* Returns a message corresponding to an error code, ERRCODE, returned
5133 from either regcomp or regexec. We don't use PREG here. */
5136 regerror (errcode, preg, errbuf, errbuf_size)
5138 const regex_t *preg;
5146 || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0])))
5147 /* Only error codes returned by the rest of the code should be passed
5148 to this routine. If we are given anything else, or if other regex
5149 code generates an invalid error code, then the program has a bug.
5150 Dump core so we can fix it. */
5153 msg = re_error_msg[errcode];
5155 /* POSIX doesn't require that we do anything in this case, but why
5160 msg_size = strlen (msg) + 1; /* Includes the null. */
5162 if (errbuf_size != 0)
5164 if (msg_size > errbuf_size)
5166 strncpy (errbuf, msg, errbuf_size - 1);
5167 errbuf[errbuf_size - 1] = 0;
5170 strcpy (errbuf, msg);
5177 /* Free dynamically allocated space used by PREG. */
5183 if (preg->buffer != NULL)
5184 free (preg->buffer);
5185 preg->buffer = NULL;
5187 preg->allocated = 0;
5190 if (preg->fastmap != NULL)
5191 free (preg->fastmap);
5192 preg->fastmap = NULL;
5193 preg->fastmap_accurate = 0;
5195 if (preg->translate != NULL)
5196 free (preg->translate);
5197 preg->translate = NULL;
5200 #endif /* not emacs */
5204 make-backup-files: t
5206 trim-versions-without-asking: nil