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, 1994 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)
33 /* We need this for `regex.h', and perhaps for the Emacs include files. */
34 #include <sys/types.h>
36 /* This is for other GNU distributions with internationalized messages.
37 The GNU C Library itself does not yet support such messages. */
41 # define gettext(msgid) (msgid)
44 /* The `emacs' switch turns on certain matching commands
45 that make sense only in Emacs. */
62 /* We used to test for `BSTRING' here, but only GCC and Emacs define
63 `BSTRING', as far as I know, and neither of them use this code. */
64 #ifndef INHIBIT_STRING_HEADER
65 #if HAVE_STRING_H || STDC_HEADERS
68 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
71 #define bcopy(s, d, n) memcpy ((d), (s), (n))
74 #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. */
89 #ifdef SWITCH_ENUM_BUG
90 #define SWITCH_ENUM_CAST(x) ((int)(x))
92 #define SWITCH_ENUM_CAST(x) (x)
97 extern char *re_syntax_table;
99 #else /* not SYNTAX_TABLE */
101 /* How many characters in the character set. */
102 #define CHAR_SET_SIZE 256
104 static char re_syntax_table[CHAR_SET_SIZE];
115 bzero (re_syntax_table, sizeof re_syntax_table);
117 for (c = 'a'; c <= 'z'; c++)
118 re_syntax_table[c] = Sword;
120 for (c = 'A'; c <= 'Z'; c++)
121 re_syntax_table[c] = Sword;
123 for (c = '0'; c <= '9'; c++)
124 re_syntax_table[c] = Sword;
126 re_syntax_table['_'] = Sword;
131 #endif /* not SYNTAX_TABLE */
133 #define SYNTAX(c) re_syntax_table[c]
135 #endif /* not emacs */
137 /* Get the interface, including the syntax bits. */
140 /* isalpha etc. are used for the character classes. */
143 /* Jim Meyering writes:
145 "... Some ctype macros are valid only for character codes that
146 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
147 using /bin/cc or gcc but without giving an ansi option). So, all
148 ctype uses should be through macros like ISPRINT... If
149 STDC_HEADERS is defined, then autoconf has verified that the ctype
150 macros don't need to be guarded with references to isascii. ...
151 Defining isascii to 1 should let any compiler worth its salt
152 eliminate the && through constant folding." */
154 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
157 #define ISASCII(c) isascii(c)
161 #define ISBLANK(c) (ISASCII (c) && isblank (c))
163 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
166 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
168 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
171 #define ISPRINT(c) (ISASCII (c) && isprint (c))
172 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
173 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
174 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
175 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
176 #define ISLOWER(c) (ISASCII (c) && islower (c))
177 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
178 #define ISSPACE(c) (ISASCII (c) && isspace (c))
179 #define ISUPPER(c) (ISASCII (c) && isupper (c))
180 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
186 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
187 since ours (we hope) works properly with all combinations of
188 machines, compilers, `char' and `unsigned char' argument types.
189 (Per Bothner suggested the basic approach.) */
190 #undef SIGN_EXTEND_CHAR
192 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
193 #else /* not __STDC__ */
194 /* As in Harbison and Steele. */
195 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
198 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
199 use `alloca' instead of `malloc'. This is because using malloc in
200 re_search* or re_match* could cause memory leaks when C-g is used in
201 Emacs; also, malloc is slower and causes storage fragmentation. On
202 the other hand, malloc is more portable, and easier to debug.
204 Because we sometimes use alloca, some routines have to be macros,
205 not functions -- `alloca'-allocated space disappears at the end of the
206 function it is called in. */
210 #define REGEX_ALLOCATE malloc
211 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
212 #define REGEX_FREE free
214 #else /* not REGEX_MALLOC */
216 /* Emacs already defines alloca, sometimes. */
219 /* Make alloca work the best possible way. */
221 #define alloca __builtin_alloca
222 #else /* not __GNUC__ */
225 #else /* not __GNUC__ or HAVE_ALLOCA_H */
226 #ifndef _AIX /* Already did AIX, up at the top. */
228 #endif /* not _AIX */
229 #endif /* not HAVE_ALLOCA_H */
230 #endif /* not __GNUC__ */
232 #endif /* not alloca */
234 #define REGEX_ALLOCATE alloca
236 /* Assumes a `char *destination' variable. */
237 #define REGEX_REALLOCATE(source, osize, nsize) \
238 (destination = (char *) alloca (nsize), \
239 bcopy (source, destination, osize), \
242 /* No need to do anything to free, after alloca. */
243 #define REGEX_FREE(arg) while (0) /* empty */
245 #endif /* not REGEX_MALLOC */
247 /* Define how to allocate the failure stack. */
250 #define REGEX_ALLOCATE_STACK(size) \
251 r_alloc (&failure_stack_ptr, (size))
252 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
253 r_re_alloc (&failure_stack_ptr, (nsize))
254 #define REGEX_FREE_STACK(ptr) \
255 r_alloc_free (&failure_stack_ptr)
257 #else /* not REL_ALLOC */
261 #define REGEX_ALLOCATE_STACK malloc
262 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
263 #define REGEX_FREE_STACK free
265 #else /* not REGEX_MALLOC */
267 #define REGEX_ALLOCATE_STACK alloca
269 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
270 REGEX_REALLOCATE (source, osize, nsize)
271 /* No need to explicitly free anything. */
272 #define REGEX_FREE_STACK(arg)
274 #endif /* not REGEX_MALLOC */
275 #endif /* not REL_ALLOC */
278 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
279 `string1' or just past its end. This works if PTR is NULL, which is
281 #define FIRST_STRING_P(ptr) \
282 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
284 /* (Re)Allocate N items of type T using malloc, or fail. */
285 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
286 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
287 #define RETALLOC_IF(addr, n, t) \
288 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
289 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
291 #define BYTEWIDTH 8 /* In bits. */
293 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
297 #define MAX(a, b) ((a) > (b) ? (a) : (b))
298 #define MIN(a, b) ((a) < (b) ? (a) : (b))
300 typedef char boolean;
304 static int re_match_2_internal ();
306 /* These are the command codes that appear in compiled regular
307 expressions. Some opcodes are followed by argument bytes. A
308 command code can specify any interpretation whatsoever for its
309 arguments. Zero bytes may appear in the compiled regular expression. */
315 /* Succeed right away--no more backtracking. */
318 /* Followed by one byte giving n, then by n literal bytes. */
321 /* Matches any (more or less) character. */
324 /* Matches any one char belonging to specified set. First
325 following byte is number of bitmap bytes. Then come bytes
326 for a bitmap saying which chars are in. Bits in each byte
327 are ordered low-bit-first. A character is in the set if its
328 bit is 1. A character too large to have a bit in the map is
329 automatically not in the set. */
332 /* Same parameters as charset, but match any character that is
333 not one of those specified. */
336 /* Start remembering the text that is matched, for storing in a
337 register. Followed by one byte with the register number, in
338 the range 0 to one less than the pattern buffer's re_nsub
339 field. Then followed by one byte with the number of groups
340 inner to this one. (This last has to be part of the
341 start_memory only because we need it in the on_failure_jump
345 /* Stop remembering the text that is matched and store it in a
346 memory register. Followed by one byte with the register
347 number, in the range 0 to one less than `re_nsub' in the
348 pattern buffer, and one byte with the number of inner groups,
349 just like `start_memory'. (We need the number of inner
350 groups here because we don't have any easy way of finding the
351 corresponding start_memory when we're at a stop_memory.) */
354 /* Match a duplicate of something remembered. Followed by one
355 byte containing the register number. */
358 /* Fail unless at beginning of line. */
361 /* Fail unless at end of line. */
364 /* Succeeds if at beginning of buffer (if emacs) or at beginning
365 of string to be matched (if not). */
368 /* Analogously, for end of buffer/string. */
371 /* Followed by two byte relative address to which to jump. */
374 /* Same as jump, but marks the end of an alternative. */
377 /* Followed by two-byte relative address of place to resume at
378 in case of failure. */
381 /* Like on_failure_jump, but pushes a placeholder instead of the
382 current string position when executed. */
383 on_failure_keep_string_jump,
385 /* Throw away latest failure point and then jump to following
386 two-byte relative address. */
389 /* Change to pop_failure_jump if know won't have to backtrack to
390 match; otherwise change to jump. This is used to jump
391 back to the beginning of a repeat. If what follows this jump
392 clearly won't match what the repeat does, such that we can be
393 sure that there is no use backtracking out of repetitions
394 already matched, then we change it to a pop_failure_jump.
395 Followed by two-byte address. */
398 /* Jump to following two-byte address, and push a dummy failure
399 point. This failure point will be thrown away if an attempt
400 is made to use it for a failure. A `+' construct makes this
401 before the first repeat. Also used as an intermediary kind
402 of jump when compiling an alternative. */
405 /* Push a dummy failure point and continue. Used at the end of
409 /* Followed by two-byte relative address and two-byte number n.
410 After matching N times, jump to the address upon failure. */
413 /* Followed by two-byte relative address, and two-byte number n.
414 Jump to the address N times, then fail. */
417 /* Set the following two-byte relative address to the
418 subsequent two-byte number. The address *includes* the two
422 wordchar, /* Matches any word-constituent character. */
423 notwordchar, /* Matches any char that is not a word-constituent. */
425 wordbeg, /* Succeeds if at word beginning. */
426 wordend, /* Succeeds if at word end. */
428 wordbound, /* Succeeds if at a word boundary. */
429 notwordbound /* Succeeds if not at a word boundary. */
432 ,before_dot, /* Succeeds if before point. */
433 at_dot, /* Succeeds if at point. */
434 after_dot, /* Succeeds if after point. */
436 /* Matches any character whose syntax is specified. Followed by
437 a byte which contains a syntax code, e.g., Sword. */
440 /* Matches any character whose syntax is not that specified. */
445 /* Common operations on the compiled pattern. */
447 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
449 #define STORE_NUMBER(destination, number) \
451 (destination)[0] = (number) & 0377; \
452 (destination)[1] = (number) >> 8; \
455 /* Same as STORE_NUMBER, except increment DESTINATION to
456 the byte after where the number is stored. Therefore, DESTINATION
457 must be an lvalue. */
459 #define STORE_NUMBER_AND_INCR(destination, number) \
461 STORE_NUMBER (destination, number); \
462 (destination) += 2; \
465 /* Put into DESTINATION a number stored in two contiguous bytes starting
468 #define EXTRACT_NUMBER(destination, source) \
470 (destination) = *(source) & 0377; \
471 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
476 extract_number (dest, source)
478 unsigned char *source;
480 int temp = SIGN_EXTEND_CHAR (*(source + 1));
481 *dest = *source & 0377;
485 #ifndef EXTRACT_MACROS /* To debug the macros. */
486 #undef EXTRACT_NUMBER
487 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
488 #endif /* not EXTRACT_MACROS */
492 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
493 SOURCE must be an lvalue. */
495 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
497 EXTRACT_NUMBER (destination, source); \
503 extract_number_and_incr (destination, source)
505 unsigned char **source;
507 extract_number (destination, *source);
511 #ifndef EXTRACT_MACROS
512 #undef EXTRACT_NUMBER_AND_INCR
513 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
514 extract_number_and_incr (&dest, &src)
515 #endif /* not EXTRACT_MACROS */
519 /* If DEBUG is defined, Regex prints many voluminous messages about what
520 it is doing (if the variable `debug' is nonzero). If linked with the
521 main program in `iregex.c', you can enter patterns and strings
522 interactively. And if linked with the main program in `main.c' and
523 the other test files, you can run the already-written tests. */
527 /* We use standard I/O for debugging. */
530 /* It is useful to test things that ``must'' be true when debugging. */
533 static int debug = 0;
535 #define DEBUG_STATEMENT(e) e
536 #define DEBUG_PRINT1(x) if (debug) printf (x)
537 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
538 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
539 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
540 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
541 if (debug) print_partial_compiled_pattern (s, e)
542 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
543 if (debug) print_double_string (w, s1, sz1, s2, sz2)
546 /* Print the fastmap in human-readable form. */
549 print_fastmap (fastmap)
552 unsigned was_a_range = 0;
555 while (i < (1 << BYTEWIDTH))
561 while (i < (1 << BYTEWIDTH) && fastmap[i])
577 /* Print a compiled pattern string in human-readable form, starting at
578 the START pointer into it and ending just before the pointer END. */
581 print_partial_compiled_pattern (start, end)
582 unsigned char *start;
586 unsigned char *p = start;
587 unsigned char *pend = end;
595 /* Loop over pattern commands. */
598 printf ("%d:\t", p - start);
600 switch ((re_opcode_t) *p++)
608 printf ("/exactn/%d", mcnt);
619 printf ("/start_memory/%d/%d", mcnt, *p++);
624 printf ("/stop_memory/%d/%d", mcnt, *p++);
628 printf ("/duplicate/%d", *p++);
638 register int c, last = -100;
639 register int in_range = 0;
641 printf ("/charset [%s",
642 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
644 assert (p + *p < pend);
646 for (c = 0; c < 256; c++)
648 && (p[1 + (c/8)] & (1 << (c % 8))))
650 /* Are we starting a range? */
651 if (last + 1 == c && ! in_range)
656 /* Have we broken a range? */
657 else if (last + 1 != c && in_range)
686 case on_failure_jump:
687 extract_number_and_incr (&mcnt, &p);
688 printf ("/on_failure_jump to %d", p + mcnt - start);
691 case on_failure_keep_string_jump:
692 extract_number_and_incr (&mcnt, &p);
693 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
696 case dummy_failure_jump:
697 extract_number_and_incr (&mcnt, &p);
698 printf ("/dummy_failure_jump to %d", p + mcnt - start);
701 case push_dummy_failure:
702 printf ("/push_dummy_failure");
706 extract_number_and_incr (&mcnt, &p);
707 printf ("/maybe_pop_jump to %d", p + mcnt - start);
710 case pop_failure_jump:
711 extract_number_and_incr (&mcnt, &p);
712 printf ("/pop_failure_jump to %d", p + mcnt - start);
716 extract_number_and_incr (&mcnt, &p);
717 printf ("/jump_past_alt to %d", p + mcnt - start);
721 extract_number_and_incr (&mcnt, &p);
722 printf ("/jump to %d", p + mcnt - start);
726 extract_number_and_incr (&mcnt, &p);
727 extract_number_and_incr (&mcnt2, &p);
728 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
732 extract_number_and_incr (&mcnt, &p);
733 extract_number_and_incr (&mcnt2, &p);
734 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
738 extract_number_and_incr (&mcnt, &p);
739 extract_number_and_incr (&mcnt2, &p);
740 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
744 printf ("/wordbound");
748 printf ("/notwordbound");
760 printf ("/before_dot");
768 printf ("/after_dot");
772 printf ("/syntaxspec");
774 printf ("/%d", mcnt);
778 printf ("/notsyntaxspec");
780 printf ("/%d", mcnt);
785 printf ("/wordchar");
789 printf ("/notwordchar");
801 printf ("?%d", *(p-1));
807 printf ("%d:\tend of pattern.\n", p - start);
812 print_compiled_pattern (bufp)
813 struct re_pattern_buffer *bufp;
815 unsigned char *buffer = bufp->buffer;
817 print_partial_compiled_pattern (buffer, buffer + bufp->used);
818 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
820 if (bufp->fastmap_accurate && bufp->fastmap)
822 printf ("fastmap: ");
823 print_fastmap (bufp->fastmap);
826 printf ("re_nsub: %d\t", bufp->re_nsub);
827 printf ("regs_alloc: %d\t", bufp->regs_allocated);
828 printf ("can_be_null: %d\t", bufp->can_be_null);
829 printf ("newline_anchor: %d\n", bufp->newline_anchor);
830 printf ("no_sub: %d\t", bufp->no_sub);
831 printf ("not_bol: %d\t", bufp->not_bol);
832 printf ("not_eol: %d\t", bufp->not_eol);
833 printf ("syntax: %d\n", bufp->syntax);
834 /* Perhaps we should print the translate table? */
839 print_double_string (where, string1, size1, string2, size2)
852 if (FIRST_STRING_P (where))
854 for (this_char = where - string1; this_char < size1; this_char++)
855 putchar (string1[this_char]);
860 for (this_char = where - string2; this_char < size2; this_char++)
861 putchar (string2[this_char]);
865 #else /* not DEBUG */
870 #define DEBUG_STATEMENT(e)
871 #define DEBUG_PRINT1(x)
872 #define DEBUG_PRINT2(x1, x2)
873 #define DEBUG_PRINT3(x1, x2, x3)
874 #define DEBUG_PRINT4(x1, x2, x3, x4)
875 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
876 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
878 #endif /* not DEBUG */
880 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
881 also be assigned to arbitrarily: each pattern buffer stores its own
882 syntax, so it can be changed between regex compilations. */
883 /* This has no initializer because initialized variables in Emacs
884 become read-only after dumping. */
885 reg_syntax_t re_syntax_options;
888 /* Specify the precise syntax of regexps for compilation. This provides
889 for compatibility for various utilities which historically have
890 different, incompatible syntaxes.
892 The argument SYNTAX is a bit mask comprised of the various bits
893 defined in regex.h. We return the old syntax. */
896 re_set_syntax (syntax)
899 reg_syntax_t ret = re_syntax_options;
901 re_syntax_options = syntax;
905 /* This table gives an error message for each of the error codes listed
906 in regex.h. Obviously the order here has to be same as there.
907 POSIX doesn't require that we do anything for REG_NOERROR,
908 but why not be nice? */
910 static const char *re_error_msgid[] =
911 { "Success", /* REG_NOERROR */
912 "No match", /* REG_NOMATCH */
913 "Invalid regular expression", /* REG_BADPAT */
914 "Invalid collation character", /* REG_ECOLLATE */
915 "Invalid character class name", /* REG_ECTYPE */
916 "Trailing backslash", /* REG_EESCAPE */
917 "Invalid back reference", /* REG_ESUBREG */
918 "Unmatched [ or [^", /* REG_EBRACK */
919 "Unmatched ( or \\(", /* REG_EPAREN */
920 "Unmatched \\{", /* REG_EBRACE */
921 "Invalid content of \\{\\}", /* REG_BADBR */
922 "Invalid range end", /* REG_ERANGE */
923 "Memory exhausted", /* REG_ESPACE */
924 "Invalid preceding regular expression", /* REG_BADRPT */
925 "Premature end of regular expression", /* REG_EEND */
926 "Regular expression too big", /* REG_ESIZE */
927 "Unmatched ) or \\)", /* REG_ERPAREN */
930 /* Avoiding alloca during matching, to placate r_alloc. */
932 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
933 searching and matching functions should not call alloca. On some
934 systems, alloca is implemented in terms of malloc, and if we're
935 using the relocating allocator routines, then malloc could cause a
936 relocation, which might (if the strings being searched are in the
937 ralloc heap) shift the data out from underneath the regexp
940 Here's another reason to avoid allocation: Emacs
941 processes input from X in a signal handler; processing X input may
942 call malloc; if input arrives while a matching routine is calling
943 malloc, then we're scrod. But Emacs can't just block input while
944 calling matching routines; then we don't notice interrupts when
945 they come in. So, Emacs blocks input around all regexp calls
946 except the matching calls, which it leaves unprotected, in the
947 faith that they will not malloc. */
949 /* Normally, this is fine. */
950 #define MATCH_MAY_ALLOCATE
952 /* When using GNU C, we are not REALLY using the C alloca, no matter
953 what config.h may say. So don't take precautions for it. */
958 /* The match routines may not allocate if (1) they would do it with malloc
959 and (2) it's not safe for them to use malloc. */
960 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && (defined (emacs) || defined (REL_ALLOC))
961 #undef MATCH_MAY_ALLOCATE
965 /* Failure stack declarations and macros; both re_compile_fastmap and
966 re_match_2 use a failure stack. These have to be macros because of
967 REGEX_ALLOCATE_STACK. */
970 /* Number of failure points for which to initially allocate space
971 when matching. If this number is exceeded, we allocate more
972 space, so it is not a hard limit. */
973 #ifndef INIT_FAILURE_ALLOC
974 #define INIT_FAILURE_ALLOC 5
977 /* Roughly the maximum number of failure points on the stack. Would be
978 exactly that if always used MAX_FAILURE_SPACE each time we failed.
979 This is a variable only so users of regex can assign to it; we never
980 change it ourselves. */
982 int re_max_failures = 20000000;
984 int re_max_failures = 2000;
987 typedef unsigned char *fail_stack_elt_t;
991 fail_stack_elt_t *stack;
993 unsigned avail; /* Offset of next open position. */
996 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
997 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
998 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
999 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
1002 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
1004 #ifdef MATCH_MAY_ALLOCATE
1005 #define INIT_FAIL_STACK() \
1007 fail_stack.stack = (fail_stack_elt_t *) \
1008 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1010 if (fail_stack.stack == NULL) \
1013 fail_stack.size = INIT_FAILURE_ALLOC; \
1014 fail_stack.avail = 0; \
1017 #define INIT_FAIL_STACK() \
1019 fail_stack.avail = 0; \
1024 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1026 Return 1 if succeeds, and 0 if either ran out of memory
1027 allocating space for it or it was already too large.
1029 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1031 #define DOUBLE_FAIL_STACK(fail_stack) \
1032 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1034 : ((fail_stack).stack = (fail_stack_elt_t *) \
1035 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1036 (fail_stack).size * sizeof (fail_stack_elt_t), \
1037 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1039 (fail_stack).stack == NULL \
1041 : ((fail_stack).size <<= 1, \
1045 /* Push PATTERN_OP on FAIL_STACK.
1047 Return 1 if was able to do so and 0 if ran out of memory allocating
1049 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
1050 ((FAIL_STACK_FULL () \
1051 && !DOUBLE_FAIL_STACK (fail_stack)) \
1053 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
1056 /* Push a pointer value onto the failure stack.
1057 Assumes the variable `fail_stack'. Probably should only
1058 be called from within `PUSH_FAILURE_POINT'. */
1059 #define PUSH_FAILURE_POINTER(item) \
1060 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) (item)
1062 /* This pushes an integer-valued item onto the failure stack.
1063 Assumes the variable `fail_stack'. Probably should only
1064 be called from within `PUSH_FAILURE_POINT'. */
1065 #define PUSH_FAILURE_INT(item) \
1066 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) (EMACS_INT) (item)
1068 /* The complement operation. Assumes `fail_stack' is nonempty. */
1069 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail]
1071 /* The complement operation. Assumes `fail_stack' is nonempty. */
1072 #define POP_FAILURE_INT() (EMACS_INT) fail_stack.stack[--fail_stack.avail]
1074 /* Used to omit pushing failure point id's when we're not debugging. */
1076 #define DEBUG_PUSH PUSH_FAILURE_INT
1077 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1079 #define DEBUG_PUSH(item)
1080 #define DEBUG_POP(item_addr)
1084 /* Push the information about the state we will need
1085 if we ever fail back to it.
1087 Requires variables fail_stack, regstart, regend, reg_info, and
1088 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1091 Does `return FAILURE_CODE' if runs out of memory. */
1093 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1095 char *destination; \
1096 /* Must be int, so when we don't save any registers, the arithmetic \
1097 of 0 + -1 isn't done as unsigned. */ \
1100 DEBUG_STATEMENT (failure_id++); \
1101 DEBUG_STATEMENT (nfailure_points_pushed++); \
1102 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1103 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1104 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1106 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1107 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1109 /* Ensure we have enough space allocated for what we will push. */ \
1110 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1112 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1113 return failure_code; \
1115 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1116 (fail_stack).size); \
1117 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1120 /* Push the info, starting with the registers. */ \
1121 DEBUG_PRINT1 ("\n"); \
1123 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1126 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1127 DEBUG_STATEMENT (num_regs_pushed++); \
1129 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1130 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1132 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1133 PUSH_FAILURE_POINTER (regend[this_reg]); \
1135 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1136 DEBUG_PRINT2 (" match_null=%d", \
1137 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1138 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1139 DEBUG_PRINT2 (" matched_something=%d", \
1140 MATCHED_SOMETHING (reg_info[this_reg])); \
1141 DEBUG_PRINT2 (" ever_matched=%d", \
1142 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1143 DEBUG_PRINT1 ("\n"); \
1144 PUSH_FAILURE_POINTER (reg_info[this_reg].word); \
1147 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1148 PUSH_FAILURE_INT (lowest_active_reg); \
1150 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1151 PUSH_FAILURE_INT (highest_active_reg); \
1153 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1154 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1155 PUSH_FAILURE_POINTER (pattern_place); \
1157 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1158 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1160 DEBUG_PRINT1 ("'\n"); \
1161 PUSH_FAILURE_POINTER (string_place); \
1163 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1164 DEBUG_PUSH (failure_id); \
1167 /* This is the number of items that are pushed and popped on the stack
1168 for each register. */
1169 #define NUM_REG_ITEMS 3
1171 /* Individual items aside from the registers. */
1173 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1175 #define NUM_NONREG_ITEMS 4
1178 /* We push at most this many items on the stack. */
1179 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1181 /* We actually push this many items. */
1182 #define NUM_FAILURE_ITEMS \
1183 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1186 /* How many items can still be added to the stack without overflowing it. */
1187 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1190 /* Pops what PUSH_FAIL_STACK pushes.
1192 We restore into the parameters, all of which should be lvalues:
1193 STR -- the saved data position.
1194 PAT -- the saved pattern position.
1195 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1196 REGSTART, REGEND -- arrays of string positions.
1197 REG_INFO -- array of information about each subexpression.
1199 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1200 `pend', `string1', `size1', `string2', and `size2'. */
1202 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1204 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1206 const unsigned char *string_temp; \
1208 assert (!FAIL_STACK_EMPTY ()); \
1210 /* Remove failure points and point to how many regs pushed. */ \
1211 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1212 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1213 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1215 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1217 DEBUG_POP (&failure_id); \
1218 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1220 /* If the saved string location is NULL, it came from an \
1221 on_failure_keep_string_jump opcode, and we want to throw away the \
1222 saved NULL, thus retaining our current position in the string. */ \
1223 string_temp = POP_FAILURE_POINTER (); \
1224 if (string_temp != NULL) \
1225 str = (const char *) string_temp; \
1227 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1228 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1229 DEBUG_PRINT1 ("'\n"); \
1231 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1232 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1233 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1235 /* Restore register info. */ \
1236 high_reg = (unsigned) POP_FAILURE_INT (); \
1237 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1239 low_reg = (unsigned) POP_FAILURE_INT (); \
1240 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1242 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1244 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1246 reg_info[this_reg].word = POP_FAILURE_POINTER (); \
1247 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1249 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1250 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1252 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1253 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1256 set_regs_matched_done = 0; \
1257 DEBUG_STATEMENT (nfailure_points_popped++); \
1258 } /* POP_FAILURE_POINT */
1262 /* Structure for per-register (a.k.a. per-group) information.
1263 This must not be longer than one word, because we push this value
1264 onto the failure stack. Other register information, such as the
1265 starting and ending positions (which are addresses), and the list of
1266 inner groups (which is a bits list) are maintained in separate
1269 We are making a (strictly speaking) nonportable assumption here: that
1270 the compiler will pack our bit fields into something that fits into
1271 the type of `word', i.e., is something that fits into one item on the
1275 fail_stack_elt_t word;
1278 /* This field is one if this group can match the empty string,
1279 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1280 #define MATCH_NULL_UNSET_VALUE 3
1281 unsigned match_null_string_p : 2;
1282 unsigned is_active : 1;
1283 unsigned matched_something : 1;
1284 unsigned ever_matched_something : 1;
1286 } register_info_type;
1288 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1289 #define IS_ACTIVE(R) ((R).bits.is_active)
1290 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1291 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1294 /* Call this when have matched a real character; it sets `matched' flags
1295 for the subexpressions which we are currently inside. Also records
1296 that those subexprs have matched. */
1297 #define SET_REGS_MATCHED() \
1300 if (!set_regs_matched_done) \
1303 set_regs_matched_done = 1; \
1304 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1306 MATCHED_SOMETHING (reg_info[r]) \
1307 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1314 /* Registers are set to a sentinel when they haven't yet matched. */
1315 static char reg_unset_dummy;
1316 #define REG_UNSET_VALUE (®_unset_dummy)
1317 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1319 /* Subroutine declarations and macros for regex_compile. */
1321 static void store_op1 (), store_op2 ();
1322 static void insert_op1 (), insert_op2 ();
1323 static boolean at_begline_loc_p (), at_endline_loc_p ();
1324 static boolean group_in_compile_stack ();
1325 static reg_errcode_t compile_range ();
1327 /* Fetch the next character in the uncompiled pattern---translating it
1328 if necessary. Also cast from a signed character in the constant
1329 string passed to us by the user to an unsigned char that we can use
1330 as an array index (in, e.g., `translate'). */
1331 #define PATFETCH(c) \
1332 do {if (p == pend) return REG_EEND; \
1333 c = (unsigned char) *p++; \
1334 if (translate) c = translate[c]; \
1337 /* Fetch the next character in the uncompiled pattern, with no
1339 #define PATFETCH_RAW(c) \
1340 do {if (p == pend) return REG_EEND; \
1341 c = (unsigned char) *p++; \
1344 /* Go backwards one character in the pattern. */
1345 #define PATUNFETCH p--
1348 /* If `translate' is non-null, return translate[D], else just D. We
1349 cast the subscript to translate because some data is declared as
1350 `char *', to avoid warnings when a string constant is passed. But
1351 when we use a character as a subscript we must make it unsigned. */
1352 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1355 /* Macros for outputting the compiled pattern into `buffer'. */
1357 /* If the buffer isn't allocated when it comes in, use this. */
1358 #define INIT_BUF_SIZE 32
1360 /* Make sure we have at least N more bytes of space in buffer. */
1361 #define GET_BUFFER_SPACE(n) \
1362 while (b - bufp->buffer + (n) > bufp->allocated) \
1365 /* Make sure we have one more byte of buffer space and then add C to it. */
1366 #define BUF_PUSH(c) \
1368 GET_BUFFER_SPACE (1); \
1369 *b++ = (unsigned char) (c); \
1373 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1374 #define BUF_PUSH_2(c1, c2) \
1376 GET_BUFFER_SPACE (2); \
1377 *b++ = (unsigned char) (c1); \
1378 *b++ = (unsigned char) (c2); \
1382 /* As with BUF_PUSH_2, except for three bytes. */
1383 #define BUF_PUSH_3(c1, c2, c3) \
1385 GET_BUFFER_SPACE (3); \
1386 *b++ = (unsigned char) (c1); \
1387 *b++ = (unsigned char) (c2); \
1388 *b++ = (unsigned char) (c3); \
1392 /* Store a jump with opcode OP at LOC to location TO. We store a
1393 relative address offset by the three bytes the jump itself occupies. */
1394 #define STORE_JUMP(op, loc, to) \
1395 store_op1 (op, loc, (to) - (loc) - 3)
1397 /* Likewise, for a two-argument jump. */
1398 #define STORE_JUMP2(op, loc, to, arg) \
1399 store_op2 (op, loc, (to) - (loc) - 3, arg)
1401 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1402 #define INSERT_JUMP(op, loc, to) \
1403 insert_op1 (op, loc, (to) - (loc) - 3, b)
1405 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1406 #define INSERT_JUMP2(op, loc, to, arg) \
1407 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1410 /* This is not an arbitrary limit: the arguments which represent offsets
1411 into the pattern are two bytes long. So if 2^16 bytes turns out to
1412 be too small, many things would have to change. */
1413 #define MAX_BUF_SIZE (1L << 16)
1416 /* Extend the buffer by twice its current size via realloc and
1417 reset the pointers that pointed into the old block to point to the
1418 correct places in the new one. If extending the buffer results in it
1419 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1420 #define EXTEND_BUFFER() \
1422 unsigned char *old_buffer = bufp->buffer; \
1423 if (bufp->allocated == MAX_BUF_SIZE) \
1425 bufp->allocated <<= 1; \
1426 if (bufp->allocated > MAX_BUF_SIZE) \
1427 bufp->allocated = MAX_BUF_SIZE; \
1428 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1429 if (bufp->buffer == NULL) \
1430 return REG_ESPACE; \
1431 /* If the buffer moved, move all the pointers into it. */ \
1432 if (old_buffer != bufp->buffer) \
1434 b = (b - old_buffer) + bufp->buffer; \
1435 begalt = (begalt - old_buffer) + bufp->buffer; \
1436 if (fixup_alt_jump) \
1437 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1439 laststart = (laststart - old_buffer) + bufp->buffer; \
1440 if (pending_exact) \
1441 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1446 /* Since we have one byte reserved for the register number argument to
1447 {start,stop}_memory, the maximum number of groups we can report
1448 things about is what fits in that byte. */
1449 #define MAX_REGNUM 255
1451 /* But patterns can have more than `MAX_REGNUM' registers. We just
1452 ignore the excess. */
1453 typedef unsigned regnum_t;
1456 /* Macros for the compile stack. */
1458 /* Since offsets can go either forwards or backwards, this type needs to
1459 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1460 typedef int pattern_offset_t;
1464 pattern_offset_t begalt_offset;
1465 pattern_offset_t fixup_alt_jump;
1466 pattern_offset_t inner_group_offset;
1467 pattern_offset_t laststart_offset;
1469 } compile_stack_elt_t;
1474 compile_stack_elt_t *stack;
1476 unsigned avail; /* Offset of next open position. */
1477 } compile_stack_type;
1480 #define INIT_COMPILE_STACK_SIZE 32
1482 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1483 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1485 /* The next available element. */
1486 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1489 /* Set the bit for character C in a list. */
1490 #define SET_LIST_BIT(c) \
1491 (b[((unsigned char) (c)) / BYTEWIDTH] \
1492 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1495 /* Get the next unsigned number in the uncompiled pattern. */
1496 #define GET_UNSIGNED_NUMBER(num) \
1500 while (ISDIGIT (c)) \
1504 num = num * 10 + c - '0'; \
1512 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1514 #define IS_CHAR_CLASS(string) \
1515 (STREQ (string, "alpha") || STREQ (string, "upper") \
1516 || STREQ (string, "lower") || STREQ (string, "digit") \
1517 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1518 || STREQ (string, "space") || STREQ (string, "print") \
1519 || STREQ (string, "punct") || STREQ (string, "graph") \
1520 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1522 #ifndef MATCH_MAY_ALLOCATE
1524 /* If we cannot allocate large objects within re_match_2_internal,
1525 we make the fail stack and register vectors global.
1526 The fail stack, we grow to the maximum size when a regexp
1528 The register vectors, we adjust in size each time we
1529 compile a regexp, according to the number of registers it needs. */
1531 static fail_stack_type fail_stack;
1533 /* Size with which the following vectors are currently allocated.
1534 That is so we can make them bigger as needed,
1535 but never make them smaller. */
1536 static int regs_allocated_size;
1538 static const char ** regstart, ** regend;
1539 static const char ** old_regstart, ** old_regend;
1540 static const char **best_regstart, **best_regend;
1541 static register_info_type *reg_info;
1542 static const char **reg_dummy;
1543 static register_info_type *reg_info_dummy;
1545 /* Make the register vectors big enough for NUM_REGS registers,
1546 but don't make them smaller. */
1549 regex_grow_registers (num_regs)
1552 if (num_regs > regs_allocated_size)
1554 RETALLOC_IF (regstart, num_regs, const char *);
1555 RETALLOC_IF (regend, num_regs, const char *);
1556 RETALLOC_IF (old_regstart, num_regs, const char *);
1557 RETALLOC_IF (old_regend, num_regs, const char *);
1558 RETALLOC_IF (best_regstart, num_regs, const char *);
1559 RETALLOC_IF (best_regend, num_regs, const char *);
1560 RETALLOC_IF (reg_info, num_regs, register_info_type);
1561 RETALLOC_IF (reg_dummy, num_regs, const char *);
1562 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1564 regs_allocated_size = num_regs;
1568 #endif /* not MATCH_MAY_ALLOCATE */
1570 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1571 Returns one of error codes defined in `regex.h', or zero for success.
1573 Assumes the `allocated' (and perhaps `buffer') and `translate'
1574 fields are set in BUFP on entry.
1576 If it succeeds, results are put in BUFP (if it returns an error, the
1577 contents of BUFP are undefined):
1578 `buffer' is the compiled pattern;
1579 `syntax' is set to SYNTAX;
1580 `used' is set to the length of the compiled pattern;
1581 `fastmap_accurate' is zero;
1582 `re_nsub' is the number of subexpressions in PATTERN;
1583 `not_bol' and `not_eol' are zero;
1585 The `fastmap' and `newline_anchor' fields are neither
1586 examined nor set. */
1588 /* Return, freeing storage we allocated. */
1589 #define FREE_STACK_RETURN(value) \
1590 return (free (compile_stack.stack), value)
1592 static reg_errcode_t
1593 regex_compile (pattern, size, syntax, bufp)
1594 const char *pattern;
1596 reg_syntax_t syntax;
1597 struct re_pattern_buffer *bufp;
1599 /* We fetch characters from PATTERN here. Even though PATTERN is
1600 `char *' (i.e., signed), we declare these variables as unsigned, so
1601 they can be reliably used as array indices. */
1602 register unsigned char c, c1;
1604 /* A random temporary spot in PATTERN. */
1607 /* Points to the end of the buffer, where we should append. */
1608 register unsigned char *b;
1610 /* Keeps track of unclosed groups. */
1611 compile_stack_type compile_stack;
1613 /* Points to the current (ending) position in the pattern. */
1614 const char *p = pattern;
1615 const char *pend = pattern + size;
1617 /* How to translate the characters in the pattern. */
1618 char *translate = bufp->translate;
1620 /* Address of the count-byte of the most recently inserted `exactn'
1621 command. This makes it possible to tell if a new exact-match
1622 character can be added to that command or if the character requires
1623 a new `exactn' command. */
1624 unsigned char *pending_exact = 0;
1626 /* Address of start of the most recently finished expression.
1627 This tells, e.g., postfix * where to find the start of its
1628 operand. Reset at the beginning of groups and alternatives. */
1629 unsigned char *laststart = 0;
1631 /* Address of beginning of regexp, or inside of last group. */
1632 unsigned char *begalt;
1634 /* Place in the uncompiled pattern (i.e., the {) to
1635 which to go back if the interval is invalid. */
1636 const char *beg_interval;
1638 /* Address of the place where a forward jump should go to the end of
1639 the containing expression. Each alternative of an `or' -- except the
1640 last -- ends with a forward jump of this sort. */
1641 unsigned char *fixup_alt_jump = 0;
1643 /* Counts open-groups as they are encountered. Remembered for the
1644 matching close-group on the compile stack, so the same register
1645 number is put in the stop_memory as the start_memory. */
1646 regnum_t regnum = 0;
1649 DEBUG_PRINT1 ("\nCompiling pattern: ");
1652 unsigned debug_count;
1654 for (debug_count = 0; debug_count < size; debug_count++)
1655 putchar (pattern[debug_count]);
1660 /* Initialize the compile stack. */
1661 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1662 if (compile_stack.stack == NULL)
1665 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1666 compile_stack.avail = 0;
1668 /* Initialize the pattern buffer. */
1669 bufp->syntax = syntax;
1670 bufp->fastmap_accurate = 0;
1671 bufp->not_bol = bufp->not_eol = 0;
1673 /* Set `used' to zero, so that if we return an error, the pattern
1674 printer (for debugging) will think there's no pattern. We reset it
1678 /* Always count groups, whether or not bufp->no_sub is set. */
1681 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1682 /* Initialize the syntax table. */
1683 init_syntax_once ();
1686 if (bufp->allocated == 0)
1689 { /* If zero allocated, but buffer is non-null, try to realloc
1690 enough space. This loses if buffer's address is bogus, but
1691 that is the user's responsibility. */
1692 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1695 { /* Caller did not allocate a buffer. Do it for them. */
1696 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1698 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1700 bufp->allocated = INIT_BUF_SIZE;
1703 begalt = b = bufp->buffer;
1705 /* Loop through the uncompiled pattern until we're at the end. */
1714 if ( /* If at start of pattern, it's an operator. */
1716 /* If context independent, it's an operator. */
1717 || syntax & RE_CONTEXT_INDEP_ANCHORS
1718 /* Otherwise, depends on what's come before. */
1719 || at_begline_loc_p (pattern, p, syntax))
1729 if ( /* If at end of pattern, it's an operator. */
1731 /* If context independent, it's an operator. */
1732 || syntax & RE_CONTEXT_INDEP_ANCHORS
1733 /* Otherwise, depends on what's next. */
1734 || at_endline_loc_p (p, pend, syntax))
1744 if ((syntax & RE_BK_PLUS_QM)
1745 || (syntax & RE_LIMITED_OPS))
1749 /* If there is no previous pattern... */
1752 if (syntax & RE_CONTEXT_INVALID_OPS)
1753 FREE_STACK_RETURN (REG_BADRPT);
1754 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1759 /* Are we optimizing this jump? */
1760 boolean keep_string_p = false;
1762 /* 1 means zero (many) matches is allowed. */
1763 char zero_times_ok = 0, many_times_ok = 0;
1765 /* If there is a sequence of repetition chars, collapse it
1766 down to just one (the right one). We can't combine
1767 interval operators with these because of, e.g., `a{2}*',
1768 which should only match an even number of `a's. */
1772 zero_times_ok |= c != '+';
1773 many_times_ok |= c != '?';
1781 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1784 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1786 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1789 if (!(c1 == '+' || c1 == '?'))
1804 /* If we get here, we found another repeat character. */
1807 /* Star, etc. applied to an empty pattern is equivalent
1808 to an empty pattern. */
1812 /* Now we know whether or not zero matches is allowed
1813 and also whether or not two or more matches is allowed. */
1815 { /* More than one repetition is allowed, so put in at the
1816 end a backward relative jump from `b' to before the next
1817 jump we're going to put in below (which jumps from
1818 laststart to after this jump).
1820 But if we are at the `*' in the exact sequence `.*\n',
1821 insert an unconditional jump backwards to the .,
1822 instead of the beginning of the loop. This way we only
1823 push a failure point once, instead of every time
1824 through the loop. */
1825 assert (p - 1 > pattern);
1827 /* Allocate the space for the jump. */
1828 GET_BUFFER_SPACE (3);
1830 /* We know we are not at the first character of the pattern,
1831 because laststart was nonzero. And we've already
1832 incremented `p', by the way, to be the character after
1833 the `*'. Do we have to do something analogous here
1834 for null bytes, because of RE_DOT_NOT_NULL? */
1835 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1837 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1838 && !(syntax & RE_DOT_NEWLINE))
1839 { /* We have .*\n. */
1840 STORE_JUMP (jump, b, laststart);
1841 keep_string_p = true;
1844 /* Anything else. */
1845 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1847 /* We've added more stuff to the buffer. */
1851 /* On failure, jump from laststart to b + 3, which will be the
1852 end of the buffer after this jump is inserted. */
1853 GET_BUFFER_SPACE (3);
1854 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1862 /* At least one repetition is required, so insert a
1863 `dummy_failure_jump' before the initial
1864 `on_failure_jump' instruction of the loop. This
1865 effects a skip over that instruction the first time
1866 we hit that loop. */
1867 GET_BUFFER_SPACE (3);
1868 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1883 boolean had_char_class = false;
1885 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1887 /* Ensure that we have enough space to push a charset: the
1888 opcode, the length count, and the bitset; 34 bytes in all. */
1889 GET_BUFFER_SPACE (34);
1893 /* We test `*p == '^' twice, instead of using an if
1894 statement, so we only need one BUF_PUSH. */
1895 BUF_PUSH (*p == '^' ? charset_not : charset);
1899 /* Remember the first position in the bracket expression. */
1902 /* Push the number of bytes in the bitmap. */
1903 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1905 /* Clear the whole map. */
1906 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1908 /* charset_not matches newline according to a syntax bit. */
1909 if ((re_opcode_t) b[-2] == charset_not
1910 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1911 SET_LIST_BIT ('\n');
1913 /* Read in characters and ranges, setting map bits. */
1916 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1920 /* \ might escape characters inside [...] and [^...]. */
1921 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1923 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1930 /* Could be the end of the bracket expression. If it's
1931 not (i.e., when the bracket expression is `[]' so
1932 far), the ']' character bit gets set way below. */
1933 if (c == ']' && p != p1 + 1)
1936 /* Look ahead to see if it's a range when the last thing
1937 was a character class. */
1938 if (had_char_class && c == '-' && *p != ']')
1939 FREE_STACK_RETURN (REG_ERANGE);
1941 /* Look ahead to see if it's a range when the last thing
1942 was a character: if this is a hyphen not at the
1943 beginning or the end of a list, then it's the range
1946 && !(p - 2 >= pattern && p[-2] == '[')
1947 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1951 = compile_range (&p, pend, translate, syntax, b);
1952 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1955 else if (p[0] == '-' && p[1] != ']')
1956 { /* This handles ranges made up of characters only. */
1959 /* Move past the `-'. */
1962 ret = compile_range (&p, pend, translate, syntax, b);
1963 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1966 /* See if we're at the beginning of a possible character
1969 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1970 { /* Leave room for the null. */
1971 char str[CHAR_CLASS_MAX_LENGTH + 1];
1976 /* If pattern is `[[:'. */
1977 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1982 if (c == ':' || c == ']' || p == pend
1983 || c1 == CHAR_CLASS_MAX_LENGTH)
1989 /* If isn't a word bracketed by `[:' and:`]':
1990 undo the ending character, the letters, and leave
1991 the leading `:' and `[' (but set bits for them). */
1992 if (c == ':' && *p == ']')
1995 boolean is_alnum = STREQ (str, "alnum");
1996 boolean is_alpha = STREQ (str, "alpha");
1997 boolean is_blank = STREQ (str, "blank");
1998 boolean is_cntrl = STREQ (str, "cntrl");
1999 boolean is_digit = STREQ (str, "digit");
2000 boolean is_graph = STREQ (str, "graph");
2001 boolean is_lower = STREQ (str, "lower");
2002 boolean is_print = STREQ (str, "print");
2003 boolean is_punct = STREQ (str, "punct");
2004 boolean is_space = STREQ (str, "space");
2005 boolean is_upper = STREQ (str, "upper");
2006 boolean is_xdigit = STREQ (str, "xdigit");
2008 if (!IS_CHAR_CLASS (str))
2009 FREE_STACK_RETURN (REG_ECTYPE);
2011 /* Throw away the ] at the end of the character
2015 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2017 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2019 /* This was split into 3 if's to
2020 avoid an arbitrary limit in some compiler. */
2021 if ( (is_alnum && ISALNUM (ch))
2022 || (is_alpha && ISALPHA (ch))
2023 || (is_blank && ISBLANK (ch))
2024 || (is_cntrl && ISCNTRL (ch)))
2026 if ( (is_digit && ISDIGIT (ch))
2027 || (is_graph && ISGRAPH (ch))
2028 || (is_lower && ISLOWER (ch))
2029 || (is_print && ISPRINT (ch)))
2031 if ( (is_punct && ISPUNCT (ch))
2032 || (is_space && ISSPACE (ch))
2033 || (is_upper && ISUPPER (ch))
2034 || (is_xdigit && ISXDIGIT (ch)))
2037 had_char_class = true;
2046 had_char_class = false;
2051 had_char_class = false;
2056 /* Discard any (non)matching list bytes that are all 0 at the
2057 end of the map. Decrease the map-length byte too. */
2058 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2066 if (syntax & RE_NO_BK_PARENS)
2073 if (syntax & RE_NO_BK_PARENS)
2080 if (syntax & RE_NEWLINE_ALT)
2087 if (syntax & RE_NO_BK_VBAR)
2094 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2095 goto handle_interval;
2101 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2103 /* Do not translate the character after the \, so that we can
2104 distinguish, e.g., \B from \b, even if we normally would
2105 translate, e.g., B to b. */
2111 if (syntax & RE_NO_BK_PARENS)
2112 goto normal_backslash;
2118 if (COMPILE_STACK_FULL)
2120 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2121 compile_stack_elt_t);
2122 if (compile_stack.stack == NULL) return REG_ESPACE;
2124 compile_stack.size <<= 1;
2127 /* These are the values to restore when we hit end of this
2128 group. They are all relative offsets, so that if the
2129 whole pattern moves because of realloc, they will still
2131 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2132 COMPILE_STACK_TOP.fixup_alt_jump
2133 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2134 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2135 COMPILE_STACK_TOP.regnum = regnum;
2137 /* We will eventually replace the 0 with the number of
2138 groups inner to this one. But do not push a
2139 start_memory for groups beyond the last one we can
2140 represent in the compiled pattern. */
2141 if (regnum <= MAX_REGNUM)
2143 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2144 BUF_PUSH_3 (start_memory, regnum, 0);
2147 compile_stack.avail++;
2152 /* If we've reached MAX_REGNUM groups, then this open
2153 won't actually generate any code, so we'll have to
2154 clear pending_exact explicitly. */
2160 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2162 if (COMPILE_STACK_EMPTY)
2163 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2164 goto normal_backslash;
2166 FREE_STACK_RETURN (REG_ERPAREN);
2170 { /* Push a dummy failure point at the end of the
2171 alternative for a possible future
2172 `pop_failure_jump' to pop. See comments at
2173 `push_dummy_failure' in `re_match_2'. */
2174 BUF_PUSH (push_dummy_failure);
2176 /* We allocated space for this jump when we assigned
2177 to `fixup_alt_jump', in the `handle_alt' case below. */
2178 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2181 /* See similar code for backslashed left paren above. */
2182 if (COMPILE_STACK_EMPTY)
2183 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2186 FREE_STACK_RETURN (REG_ERPAREN);
2188 /* Since we just checked for an empty stack above, this
2189 ``can't happen''. */
2190 assert (compile_stack.avail != 0);
2192 /* We don't just want to restore into `regnum', because
2193 later groups should continue to be numbered higher,
2194 as in `(ab)c(de)' -- the second group is #2. */
2195 regnum_t this_group_regnum;
2197 compile_stack.avail--;
2198 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2200 = COMPILE_STACK_TOP.fixup_alt_jump
2201 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2203 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2204 this_group_regnum = COMPILE_STACK_TOP.regnum;
2205 /* If we've reached MAX_REGNUM groups, then this open
2206 won't actually generate any code, so we'll have to
2207 clear pending_exact explicitly. */
2210 /* We're at the end of the group, so now we know how many
2211 groups were inside this one. */
2212 if (this_group_regnum <= MAX_REGNUM)
2214 unsigned char *inner_group_loc
2215 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2217 *inner_group_loc = regnum - this_group_regnum;
2218 BUF_PUSH_3 (stop_memory, this_group_regnum,
2219 regnum - this_group_regnum);
2225 case '|': /* `\|'. */
2226 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2227 goto normal_backslash;
2229 if (syntax & RE_LIMITED_OPS)
2232 /* Insert before the previous alternative a jump which
2233 jumps to this alternative if the former fails. */
2234 GET_BUFFER_SPACE (3);
2235 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2239 /* The alternative before this one has a jump after it
2240 which gets executed if it gets matched. Adjust that
2241 jump so it will jump to this alternative's analogous
2242 jump (put in below, which in turn will jump to the next
2243 (if any) alternative's such jump, etc.). The last such
2244 jump jumps to the correct final destination. A picture:
2250 If we are at `b', then fixup_alt_jump right now points to a
2251 three-byte space after `a'. We'll put in the jump, set
2252 fixup_alt_jump to right after `b', and leave behind three
2253 bytes which we'll fill in when we get to after `c'. */
2256 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2258 /* Mark and leave space for a jump after this alternative,
2259 to be filled in later either by next alternative or
2260 when know we're at the end of a series of alternatives. */
2262 GET_BUFFER_SPACE (3);
2271 /* If \{ is a literal. */
2272 if (!(syntax & RE_INTERVALS)
2273 /* If we're at `\{' and it's not the open-interval
2275 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2276 || (p - 2 == pattern && p == pend))
2277 goto normal_backslash;
2281 /* If got here, then the syntax allows intervals. */
2283 /* At least (most) this many matches must be made. */
2284 int lower_bound = -1, upper_bound = -1;
2286 beg_interval = p - 1;
2290 if (syntax & RE_NO_BK_BRACES)
2291 goto unfetch_interval;
2293 FREE_STACK_RETURN (REG_EBRACE);
2296 GET_UNSIGNED_NUMBER (lower_bound);
2300 GET_UNSIGNED_NUMBER (upper_bound);
2301 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2304 /* Interval such as `{1}' => match exactly once. */
2305 upper_bound = lower_bound;
2307 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2308 || lower_bound > upper_bound)
2310 if (syntax & RE_NO_BK_BRACES)
2311 goto unfetch_interval;
2313 FREE_STACK_RETURN (REG_BADBR);
2316 if (!(syntax & RE_NO_BK_BRACES))
2318 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2325 if (syntax & RE_NO_BK_BRACES)
2326 goto unfetch_interval;
2328 FREE_STACK_RETURN (REG_BADBR);
2331 /* We just parsed a valid interval. */
2333 /* If it's invalid to have no preceding re. */
2336 if (syntax & RE_CONTEXT_INVALID_OPS)
2337 FREE_STACK_RETURN (REG_BADRPT);
2338 else if (syntax & RE_CONTEXT_INDEP_OPS)
2341 goto unfetch_interval;
2344 /* If the upper bound is zero, don't want to succeed at
2345 all; jump from `laststart' to `b + 3', which will be
2346 the end of the buffer after we insert the jump. */
2347 if (upper_bound == 0)
2349 GET_BUFFER_SPACE (3);
2350 INSERT_JUMP (jump, laststart, b + 3);
2354 /* Otherwise, we have a nontrivial interval. When
2355 we're all done, the pattern will look like:
2356 set_number_at <jump count> <upper bound>
2357 set_number_at <succeed_n count> <lower bound>
2358 succeed_n <after jump addr> <succeed_n count>
2360 jump_n <succeed_n addr> <jump count>
2361 (The upper bound and `jump_n' are omitted if
2362 `upper_bound' is 1, though.) */
2364 { /* If the upper bound is > 1, we need to insert
2365 more at the end of the loop. */
2366 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2368 GET_BUFFER_SPACE (nbytes);
2370 /* Initialize lower bound of the `succeed_n', even
2371 though it will be set during matching by its
2372 attendant `set_number_at' (inserted next),
2373 because `re_compile_fastmap' needs to know.
2374 Jump to the `jump_n' we might insert below. */
2375 INSERT_JUMP2 (succeed_n, laststart,
2376 b + 5 + (upper_bound > 1) * 5,
2380 /* Code to initialize the lower bound. Insert
2381 before the `succeed_n'. The `5' is the last two
2382 bytes of this `set_number_at', plus 3 bytes of
2383 the following `succeed_n'. */
2384 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2387 if (upper_bound > 1)
2388 { /* More than one repetition is allowed, so
2389 append a backward jump to the `succeed_n'
2390 that starts this interval.
2392 When we've reached this during matching,
2393 we'll have matched the interval once, so
2394 jump back only `upper_bound - 1' times. */
2395 STORE_JUMP2 (jump_n, b, laststart + 5,
2399 /* The location we want to set is the second
2400 parameter of the `jump_n'; that is `b-2' as
2401 an absolute address. `laststart' will be
2402 the `set_number_at' we're about to insert;
2403 `laststart+3' the number to set, the source
2404 for the relative address. But we are
2405 inserting into the middle of the pattern --
2406 so everything is getting moved up by 5.
2407 Conclusion: (b - 2) - (laststart + 3) + 5,
2408 i.e., b - laststart.
2410 We insert this at the beginning of the loop
2411 so that if we fail during matching, we'll
2412 reinitialize the bounds. */
2413 insert_op2 (set_number_at, laststart, b - laststart,
2414 upper_bound - 1, b);
2419 beg_interval = NULL;
2424 /* If an invalid interval, match the characters as literals. */
2425 assert (beg_interval);
2427 beg_interval = NULL;
2429 /* normal_char and normal_backslash need `c'. */
2432 if (!(syntax & RE_NO_BK_BRACES))
2434 if (p > pattern && p[-1] == '\\')
2435 goto normal_backslash;
2440 /* There is no way to specify the before_dot and after_dot
2441 operators. rms says this is ok. --karl */
2449 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2455 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2462 BUF_PUSH (wordchar);
2468 BUF_PUSH (notwordchar);
2481 BUF_PUSH (wordbound);
2485 BUF_PUSH (notwordbound);
2496 case '1': case '2': case '3': case '4': case '5':
2497 case '6': case '7': case '8': case '9':
2498 if (syntax & RE_NO_BK_REFS)
2504 FREE_STACK_RETURN (REG_ESUBREG);
2506 /* Can't back reference to a subexpression if inside of it. */
2507 if (group_in_compile_stack (compile_stack, c1))
2511 BUF_PUSH_2 (duplicate, c1);
2517 if (syntax & RE_BK_PLUS_QM)
2520 goto normal_backslash;
2524 /* You might think it would be useful for \ to mean
2525 not to translate; but if we don't translate it
2526 it will never match anything. */
2534 /* Expects the character in `c'. */
2536 /* If no exactn currently being built. */
2539 /* If last exactn not at current position. */
2540 || pending_exact + *pending_exact + 1 != b
2542 /* We have only one byte following the exactn for the count. */
2543 || *pending_exact == (1 << BYTEWIDTH) - 1
2545 /* If followed by a repetition operator. */
2546 || *p == '*' || *p == '^'
2547 || ((syntax & RE_BK_PLUS_QM)
2548 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2549 : (*p == '+' || *p == '?'))
2550 || ((syntax & RE_INTERVALS)
2551 && ((syntax & RE_NO_BK_BRACES)
2553 : (p[0] == '\\' && p[1] == '{'))))
2555 /* Start building a new exactn. */
2559 BUF_PUSH_2 (exactn, 0);
2560 pending_exact = b - 1;
2567 } /* while p != pend */
2570 /* Through the pattern now. */
2573 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2575 if (!COMPILE_STACK_EMPTY)
2576 FREE_STACK_RETURN (REG_EPAREN);
2578 /* If we don't want backtracking, force success
2579 the first time we reach the end of the compiled pattern. */
2580 if (syntax & RE_NO_POSIX_BACKTRACKING)
2583 free (compile_stack.stack);
2585 /* We have succeeded; set the length of the buffer. */
2586 bufp->used = b - bufp->buffer;
2591 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2592 print_compiled_pattern (bufp);
2596 #ifndef MATCH_MAY_ALLOCATE
2597 /* Initialize the failure stack to the largest possible stack. This
2598 isn't necessary unless we're trying to avoid calling alloca in
2599 the search and match routines. */
2601 int num_regs = bufp->re_nsub + 1;
2603 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2604 is strictly greater than re_max_failures, the largest possible stack
2605 is 2 * re_max_failures failure points. */
2606 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2608 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2611 if (! fail_stack.stack)
2613 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2614 * sizeof (fail_stack_elt_t));
2617 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2619 * sizeof (fail_stack_elt_t)));
2620 #else /* not emacs */
2621 if (! fail_stack.stack)
2623 = (fail_stack_elt_t *) malloc (fail_stack.size
2624 * sizeof (fail_stack_elt_t));
2627 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2629 * sizeof (fail_stack_elt_t)));
2630 #endif /* not emacs */
2633 regex_grow_registers (num_regs);
2635 #endif /* not MATCH_MAY_ALLOCATE */
2638 } /* regex_compile */
2640 /* Subroutines for `regex_compile'. */
2642 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2645 store_op1 (op, loc, arg)
2650 *loc = (unsigned char) op;
2651 STORE_NUMBER (loc + 1, arg);
2655 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2658 store_op2 (op, loc, arg1, arg2)
2663 *loc = (unsigned char) op;
2664 STORE_NUMBER (loc + 1, arg1);
2665 STORE_NUMBER (loc + 3, arg2);
2669 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2670 for OP followed by two-byte integer parameter ARG. */
2673 insert_op1 (op, loc, arg, end)
2679 register unsigned char *pfrom = end;
2680 register unsigned char *pto = end + 3;
2682 while (pfrom != loc)
2685 store_op1 (op, loc, arg);
2689 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2692 insert_op2 (op, loc, arg1, arg2, end)
2698 register unsigned char *pfrom = end;
2699 register unsigned char *pto = end + 5;
2701 while (pfrom != loc)
2704 store_op2 (op, loc, arg1, arg2);
2708 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2709 after an alternative or a begin-subexpression. We assume there is at
2710 least one character before the ^. */
2713 at_begline_loc_p (pattern, p, syntax)
2714 const char *pattern, *p;
2715 reg_syntax_t syntax;
2717 const char *prev = p - 2;
2718 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2721 /* After a subexpression? */
2722 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2723 /* After an alternative? */
2724 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2728 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2729 at least one character after the $, i.e., `P < PEND'. */
2732 at_endline_loc_p (p, pend, syntax)
2733 const char *p, *pend;
2736 const char *next = p;
2737 boolean next_backslash = *next == '\\';
2738 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2741 /* Before a subexpression? */
2742 (syntax & RE_NO_BK_PARENS ? *next == ')'
2743 : next_backslash && next_next && *next_next == ')')
2744 /* Before an alternative? */
2745 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2746 : next_backslash && next_next && *next_next == '|');
2750 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2751 false if it's not. */
2754 group_in_compile_stack (compile_stack, regnum)
2755 compile_stack_type compile_stack;
2760 for (this_element = compile_stack.avail - 1;
2763 if (compile_stack.stack[this_element].regnum == regnum)
2770 /* Read the ending character of a range (in a bracket expression) from the
2771 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2772 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2773 Then we set the translation of all bits between the starting and
2774 ending characters (inclusive) in the compiled pattern B.
2776 Return an error code.
2778 We use these short variable names so we can use the same macros as
2779 `regex_compile' itself. */
2781 static reg_errcode_t
2782 compile_range (p_ptr, pend, translate, syntax, b)
2783 const char **p_ptr, *pend;
2785 reg_syntax_t syntax;
2790 const char *p = *p_ptr;
2791 int range_start, range_end;
2796 /* Even though the pattern is a signed `char *', we need to fetch
2797 with unsigned char *'s; if the high bit of the pattern character
2798 is set, the range endpoints will be negative if we fetch using a
2801 We also want to fetch the endpoints without translating them; the
2802 appropriate translation is done in the bit-setting loop below. */
2803 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2804 range_start = ((const unsigned char *) p)[-2];
2805 range_end = ((const unsigned char *) p)[0];
2807 /* Have to increment the pointer into the pattern string, so the
2808 caller isn't still at the ending character. */
2811 /* If the start is after the end, the range is empty. */
2812 if (range_start > range_end)
2813 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2815 /* Here we see why `this_char' has to be larger than an `unsigned
2816 char' -- the range is inclusive, so if `range_end' == 0xff
2817 (assuming 8-bit characters), we would otherwise go into an infinite
2818 loop, since all characters <= 0xff. */
2819 for (this_char = range_start; this_char <= range_end; this_char++)
2821 SET_LIST_BIT (TRANSLATE (this_char));
2827 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2828 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2829 characters can start a string that matches the pattern. This fastmap
2830 is used by re_search to skip quickly over impossible starting points.
2832 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2833 area as BUFP->fastmap.
2835 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2838 Returns 0 if we succeed, -2 if an internal error. */
2841 re_compile_fastmap (bufp)
2842 struct re_pattern_buffer *bufp;
2845 #ifdef MATCH_MAY_ALLOCATE
2846 fail_stack_type fail_stack;
2848 #ifndef REGEX_MALLOC
2851 /* We don't push any register information onto the failure stack. */
2852 unsigned num_regs = 0;
2854 register char *fastmap = bufp->fastmap;
2855 unsigned char *pattern = bufp->buffer;
2856 unsigned long size = bufp->used;
2857 unsigned char *p = pattern;
2858 register unsigned char *pend = pattern + size;
2861 /* This holds the pointer to the failure stack, when
2862 it is allocated relocatably. */
2863 fail_stack_elt_t *failure_stack_ptr;
2866 /* Assume that each path through the pattern can be null until
2867 proven otherwise. We set this false at the bottom of switch
2868 statement, to which we get only if a particular path doesn't
2869 match the empty string. */
2870 boolean path_can_be_null = true;
2872 /* We aren't doing a `succeed_n' to begin with. */
2873 boolean succeed_n_p = false;
2875 assert (fastmap != NULL && p != NULL);
2878 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2879 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2880 bufp->can_be_null = 0;
2884 if (p == pend || *p == succeed)
2886 /* We have reached the (effective) end of pattern. */
2887 if (!FAIL_STACK_EMPTY ())
2889 bufp->can_be_null |= path_can_be_null;
2891 /* Reset for next path. */
2892 path_can_be_null = true;
2894 p = fail_stack.stack[--fail_stack.avail];
2902 /* We should never be about to go beyond the end of the pattern. */
2905 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2908 /* I guess the idea here is to simply not bother with a fastmap
2909 if a backreference is used, since it's too hard to figure out
2910 the fastmap for the corresponding group. Setting
2911 `can_be_null' stops `re_search_2' from using the fastmap, so
2912 that is all we do. */
2914 bufp->can_be_null = 1;
2918 /* Following are the cases which match a character. These end
2927 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2928 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2934 /* Chars beyond end of map must be allowed. */
2935 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2938 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2939 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2945 for (j = 0; j < (1 << BYTEWIDTH); j++)
2946 if (SYNTAX (j) == Sword)
2952 for (j = 0; j < (1 << BYTEWIDTH); j++)
2953 if (SYNTAX (j) != Sword)
2960 int fastmap_newline = fastmap['\n'];
2962 /* `.' matches anything ... */
2963 for (j = 0; j < (1 << BYTEWIDTH); j++)
2966 /* ... except perhaps newline. */
2967 if (!(bufp->syntax & RE_DOT_NEWLINE))
2968 fastmap['\n'] = fastmap_newline;
2970 /* Return if we have already set `can_be_null'; if we have,
2971 then the fastmap is irrelevant. Something's wrong here. */
2972 else if (bufp->can_be_null)
2975 /* Otherwise, have to check alternative paths. */
2982 for (j = 0; j < (1 << BYTEWIDTH); j++)
2983 if (SYNTAX (j) == (enum syntaxcode) k)
2990 for (j = 0; j < (1 << BYTEWIDTH); j++)
2991 if (SYNTAX (j) != (enum syntaxcode) k)
2996 /* All cases after this match the empty string. These end with
3004 #endif /* not emacs */
3016 case push_dummy_failure:
3021 case pop_failure_jump:
3022 case maybe_pop_jump:
3025 case dummy_failure_jump:
3026 EXTRACT_NUMBER_AND_INCR (j, p);
3031 /* Jump backward implies we just went through the body of a
3032 loop and matched nothing. Opcode jumped to should be
3033 `on_failure_jump' or `succeed_n'. Just treat it like an
3034 ordinary jump. For a * loop, it has pushed its failure
3035 point already; if so, discard that as redundant. */
3036 if ((re_opcode_t) *p != on_failure_jump
3037 && (re_opcode_t) *p != succeed_n)
3041 EXTRACT_NUMBER_AND_INCR (j, p);
3044 /* If what's on the stack is where we are now, pop it. */
3045 if (!FAIL_STACK_EMPTY ()
3046 && fail_stack.stack[fail_stack.avail - 1] == p)
3052 case on_failure_jump:
3053 case on_failure_keep_string_jump:
3054 handle_on_failure_jump:
3055 EXTRACT_NUMBER_AND_INCR (j, p);
3057 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3058 end of the pattern. We don't want to push such a point,
3059 since when we restore it above, entering the switch will
3060 increment `p' past the end of the pattern. We don't need
3061 to push such a point since we obviously won't find any more
3062 fastmap entries beyond `pend'. Such a pattern can match
3063 the null string, though. */
3066 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3068 REGEX_FREE_STACK (fail_stack.stack);
3073 bufp->can_be_null = 1;
3077 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3078 succeed_n_p = false;
3085 /* Get to the number of times to succeed. */
3088 /* Increment p past the n for when k != 0. */
3089 EXTRACT_NUMBER_AND_INCR (k, p);
3093 succeed_n_p = true; /* Spaghetti code alert. */
3094 goto handle_on_failure_jump;
3111 abort (); /* We have listed all the cases. */
3114 /* Getting here means we have found the possible starting
3115 characters for one path of the pattern -- and that the empty
3116 string does not match. We need not follow this path further.
3117 Instead, look at the next alternative (remembered on the
3118 stack), or quit if no more. The test at the top of the loop
3119 does these things. */
3120 path_can_be_null = false;
3124 /* Set `can_be_null' for the last path (also the first path, if the
3125 pattern is empty). */
3126 bufp->can_be_null |= path_can_be_null;
3129 REGEX_FREE_STACK (fail_stack.stack);
3131 } /* re_compile_fastmap */
3133 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3134 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3135 this memory for recording register information. STARTS and ENDS
3136 must be allocated using the malloc library routine, and must each
3137 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3139 If NUM_REGS == 0, then subsequent matches should allocate their own
3142 Unless this function is called, the first search or match using
3143 PATTERN_BUFFER will allocate its own register data, without
3144 freeing the old data. */
3147 re_set_registers (bufp, regs, num_regs, starts, ends)
3148 struct re_pattern_buffer *bufp;
3149 struct re_registers *regs;
3151 regoff_t *starts, *ends;
3155 bufp->regs_allocated = REGS_REALLOCATE;
3156 regs->num_regs = num_regs;
3157 regs->start = starts;
3162 bufp->regs_allocated = REGS_UNALLOCATED;
3164 regs->start = regs->end = (regoff_t *) 0;
3168 /* Searching routines. */
3170 /* Like re_search_2, below, but only one string is specified, and
3171 doesn't let you say where to stop matching. */
3174 re_search (bufp, string, size, startpos, range, regs)
3175 struct re_pattern_buffer *bufp;
3177 int size, startpos, range;
3178 struct re_registers *regs;
3180 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3185 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3186 virtual concatenation of STRING1 and STRING2, starting first at index
3187 STARTPOS, then at STARTPOS + 1, and so on.
3189 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3191 RANGE is how far to scan while trying to match. RANGE = 0 means try
3192 only at STARTPOS; in general, the last start tried is STARTPOS +
3195 In REGS, return the indices of the virtual concatenation of STRING1
3196 and STRING2 that matched the entire BUFP->buffer and its contained
3199 Do not consider matching one past the index STOP in the virtual
3200 concatenation of STRING1 and STRING2.
3202 We return either the position in the strings at which the match was
3203 found, -1 if no match, or -2 if error (such as failure
3207 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3208 struct re_pattern_buffer *bufp;
3209 const char *string1, *string2;
3213 struct re_registers *regs;
3217 register char *fastmap = bufp->fastmap;
3218 register char *translate = bufp->translate;
3219 int total_size = size1 + size2;
3220 int endpos = startpos + range;
3222 /* Check for out-of-range STARTPOS. */
3223 if (startpos < 0 || startpos > total_size)
3226 /* Fix up RANGE if it might eventually take us outside
3227 the virtual concatenation of STRING1 and STRING2. */
3229 range = -1 - startpos;
3230 else if (endpos > total_size)
3231 range = total_size - startpos;
3233 /* If the search isn't to be a backwards one, don't waste time in a
3234 search for a pattern that must be anchored. */
3235 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3243 /* Update the fastmap now if not correct already. */
3244 if (fastmap && !bufp->fastmap_accurate)
3245 if (re_compile_fastmap (bufp) == -2)
3248 /* Loop through the string, looking for a place to start matching. */
3251 /* If a fastmap is supplied, skip quickly over characters that
3252 cannot be the start of a match. If the pattern can match the
3253 null string, however, we don't need to skip characters; we want
3254 the first null string. */
3255 if (fastmap && startpos < total_size && !bufp->can_be_null)
3257 if (range > 0) /* Searching forwards. */
3259 register const char *d;
3260 register int lim = 0;
3263 if (startpos < size1 && startpos + range >= size1)
3264 lim = range - (size1 - startpos);
3266 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3268 /* Written out as an if-else to avoid testing `translate'
3272 && !fastmap[(unsigned char)
3273 translate[(unsigned char) *d++]])
3276 while (range > lim && !fastmap[(unsigned char) *d++])
3279 startpos += irange - range;
3281 else /* Searching backwards. */
3283 register char c = (size1 == 0 || startpos >= size1
3284 ? string2[startpos - size1]
3285 : string1[startpos]);
3287 if (!fastmap[(unsigned char) TRANSLATE (c)])
3292 /* If can't match the null string, and that's all we have left, fail. */
3293 if (range >= 0 && startpos == total_size && fastmap
3294 && !bufp->can_be_null)
3297 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3298 startpos, regs, stop);
3299 #ifndef REGEX_MALLOC
3328 /* Declarations and macros for re_match_2. */
3330 static int bcmp_translate ();
3331 static boolean alt_match_null_string_p (),
3332 common_op_match_null_string_p (),
3333 group_match_null_string_p ();
3335 /* This converts PTR, a pointer into one of the search strings `string1'
3336 and `string2' into an offset from the beginning of that string. */
3337 #define POINTER_TO_OFFSET(ptr) \
3338 (FIRST_STRING_P (ptr) \
3339 ? ((regoff_t) ((ptr) - string1)) \
3340 : ((regoff_t) ((ptr) - string2 + size1)))
3342 /* Macros for dealing with the split strings in re_match_2. */
3344 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3346 /* Call before fetching a character with *d. This switches over to
3347 string2 if necessary. */
3348 #define PREFETCH() \
3351 /* End of string2 => fail. */ \
3352 if (dend == end_match_2) \
3354 /* End of string1 => advance to string2. */ \
3356 dend = end_match_2; \
3360 /* Test if at very beginning or at very end of the virtual concatenation
3361 of `string1' and `string2'. If only one string, it's `string2'. */
3362 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3363 #define AT_STRINGS_END(d) ((d) == end2)
3366 /* Test if D points to a character which is word-constituent. We have
3367 two special cases to check for: if past the end of string1, look at
3368 the first character in string2; and if before the beginning of
3369 string2, look at the last character in string1. */
3370 #define WORDCHAR_P(d) \
3371 (SYNTAX ((d) == end1 ? *string2 \
3372 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3375 /* Test if the character before D and the one at D differ with respect
3376 to being word-constituent. */
3377 #define AT_WORD_BOUNDARY(d) \
3378 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3379 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3382 /* Free everything we malloc. */
3383 #ifdef MATCH_MAY_ALLOCATE
3384 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3385 #define FREE_VARIABLES() \
3387 REGEX_FREE_STACK (fail_stack.stack); \
3388 FREE_VAR (regstart); \
3389 FREE_VAR (regend); \
3390 FREE_VAR (old_regstart); \
3391 FREE_VAR (old_regend); \
3392 FREE_VAR (best_regstart); \
3393 FREE_VAR (best_regend); \
3394 FREE_VAR (reg_info); \
3395 FREE_VAR (reg_dummy); \
3396 FREE_VAR (reg_info_dummy); \
3399 #define FREE_VARIABLES() /* Do nothing! */
3400 #endif /* not MATCH_MAY_ALLOCATE */
3402 /* These values must meet several constraints. They must not be valid
3403 register values; since we have a limit of 255 registers (because
3404 we use only one byte in the pattern for the register number), we can
3405 use numbers larger than 255. They must differ by 1, because of
3406 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3407 be larger than the value for the highest register, so we do not try
3408 to actually save any registers when none are active. */
3409 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3410 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3412 /* Matching routines. */
3414 #ifndef emacs /* Emacs never uses this. */
3415 /* re_match is like re_match_2 except it takes only a single string. */
3418 re_match (bufp, string, size, pos, regs)
3419 struct re_pattern_buffer *bufp;
3422 struct re_registers *regs;
3424 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3429 #endif /* not emacs */
3432 /* re_match_2 matches the compiled pattern in BUFP against the
3433 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3434 and SIZE2, respectively). We start matching at POS, and stop
3437 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3438 store offsets for the substring each group matched in REGS. See the
3439 documentation for exactly how many groups we fill.
3441 We return -1 if no match, -2 if an internal error (such as the
3442 failure stack overflowing). Otherwise, we return the length of the
3443 matched substring. */
3446 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3447 struct re_pattern_buffer *bufp;
3448 const char *string1, *string2;
3451 struct re_registers *regs;
3454 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3460 /* This is a separate function so that we can force an alloca cleanup
3463 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3464 struct re_pattern_buffer *bufp;
3465 const char *string1, *string2;
3468 struct re_registers *regs;
3471 /* General temporaries. */
3475 /* Just past the end of the corresponding string. */
3476 const char *end1, *end2;
3478 /* Pointers into string1 and string2, just past the last characters in
3479 each to consider matching. */
3480 const char *end_match_1, *end_match_2;
3482 /* Where we are in the data, and the end of the current string. */
3483 const char *d, *dend;
3485 /* Where we are in the pattern, and the end of the pattern. */
3486 unsigned char *p = bufp->buffer;
3487 register unsigned char *pend = p + bufp->used;
3489 /* Mark the opcode just after a start_memory, so we can test for an
3490 empty subpattern when we get to the stop_memory. */
3491 unsigned char *just_past_start_mem = 0;
3493 /* We use this to map every character in the string. */
3494 char *translate = bufp->translate;
3496 /* Failure point stack. Each place that can handle a failure further
3497 down the line pushes a failure point on this stack. It consists of
3498 restart, regend, and reg_info for all registers corresponding to
3499 the subexpressions we're currently inside, plus the number of such
3500 registers, and, finally, two char *'s. The first char * is where
3501 to resume scanning the pattern; the second one is where to resume
3502 scanning the strings. If the latter is zero, the failure point is
3503 a ``dummy''; if a failure happens and the failure point is a dummy,
3504 it gets discarded and the next next one is tried. */
3505 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3506 fail_stack_type fail_stack;
3509 static unsigned failure_id = 0;
3510 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3514 /* This holds the pointer to the failure stack, when
3515 it is allocated relocatably. */
3516 fail_stack_elt_t *failure_stack_ptr;
3519 /* We fill all the registers internally, independent of what we
3520 return, for use in backreferences. The number here includes
3521 an element for register zero. */
3522 unsigned num_regs = bufp->re_nsub + 1;
3524 /* The currently active registers. */
3525 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3526 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3528 /* Information on the contents of registers. These are pointers into
3529 the input strings; they record just what was matched (on this
3530 attempt) by a subexpression part of the pattern, that is, the
3531 regnum-th regstart pointer points to where in the pattern we began
3532 matching and the regnum-th regend points to right after where we
3533 stopped matching the regnum-th subexpression. (The zeroth register
3534 keeps track of what the whole pattern matches.) */
3535 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3536 const char **regstart, **regend;
3539 /* If a group that's operated upon by a repetition operator fails to
3540 match anything, then the register for its start will need to be
3541 restored because it will have been set to wherever in the string we
3542 are when we last see its open-group operator. Similarly for a
3544 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3545 const char **old_regstart, **old_regend;
3548 /* The is_active field of reg_info helps us keep track of which (possibly
3549 nested) subexpressions we are currently in. The matched_something
3550 field of reg_info[reg_num] helps us tell whether or not we have
3551 matched any of the pattern so far this time through the reg_num-th
3552 subexpression. These two fields get reset each time through any
3553 loop their register is in. */
3554 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3555 register_info_type *reg_info;
3558 /* The following record the register info as found in the above
3559 variables when we find a match better than any we've seen before.
3560 This happens as we backtrack through the failure points, which in
3561 turn happens only if we have not yet matched the entire string. */
3562 unsigned best_regs_set = false;
3563 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3564 const char **best_regstart, **best_regend;
3567 /* Logically, this is `best_regend[0]'. But we don't want to have to
3568 allocate space for that if we're not allocating space for anything
3569 else (see below). Also, we never need info about register 0 for
3570 any of the other register vectors, and it seems rather a kludge to
3571 treat `best_regend' differently than the rest. So we keep track of
3572 the end of the best match so far in a separate variable. We
3573 initialize this to NULL so that when we backtrack the first time
3574 and need to test it, it's not garbage. */
3575 const char *match_end = NULL;
3577 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3578 int set_regs_matched_done = 0;
3580 /* Used when we pop values we don't care about. */
3581 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3582 const char **reg_dummy;
3583 register_info_type *reg_info_dummy;
3587 /* Counts the total number of registers pushed. */
3588 unsigned num_regs_pushed = 0;
3591 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3595 #ifdef MATCH_MAY_ALLOCATE
3596 /* Do not bother to initialize all the register variables if there are
3597 no groups in the pattern, as it takes a fair amount of time. If
3598 there are groups, we include space for register 0 (the whole
3599 pattern), even though we never use it, since it simplifies the
3600 array indexing. We should fix this. */
3603 regstart = REGEX_TALLOC (num_regs, const char *);
3604 regend = REGEX_TALLOC (num_regs, const char *);
3605 old_regstart = REGEX_TALLOC (num_regs, const char *);
3606 old_regend = REGEX_TALLOC (num_regs, const char *);
3607 best_regstart = REGEX_TALLOC (num_regs, const char *);
3608 best_regend = REGEX_TALLOC (num_regs, const char *);
3609 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3610 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3611 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3613 if (!(regstart && regend && old_regstart && old_regend && reg_info
3614 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3622 /* We must initialize all our variables to NULL, so that
3623 `FREE_VARIABLES' doesn't try to free them. */
3624 regstart = regend = old_regstart = old_regend = best_regstart
3625 = best_regend = reg_dummy = NULL;
3626 reg_info = reg_info_dummy = (register_info_type *) NULL;
3628 #endif /* MATCH_MAY_ALLOCATE */
3630 /* The starting position is bogus. */
3631 if (pos < 0 || pos > size1 + size2)
3637 /* Initialize subexpression text positions to -1 to mark ones that no
3638 start_memory/stop_memory has been seen for. Also initialize the
3639 register information struct. */
3640 for (mcnt = 1; mcnt < num_regs; mcnt++)
3642 regstart[mcnt] = regend[mcnt]
3643 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3645 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3646 IS_ACTIVE (reg_info[mcnt]) = 0;
3647 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3648 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3651 /* We move `string1' into `string2' if the latter's empty -- but not if
3652 `string1' is null. */
3653 if (size2 == 0 && string1 != NULL)
3660 end1 = string1 + size1;
3661 end2 = string2 + size2;
3663 /* Compute where to stop matching, within the two strings. */
3666 end_match_1 = string1 + stop;
3667 end_match_2 = string2;
3672 end_match_2 = string2 + stop - size1;
3675 /* `p' scans through the pattern as `d' scans through the data.
3676 `dend' is the end of the input string that `d' points within. `d'
3677 is advanced into the following input string whenever necessary, but
3678 this happens before fetching; therefore, at the beginning of the
3679 loop, `d' can be pointing at the end of a string, but it cannot
3681 if (size1 > 0 && pos <= size1)
3688 d = string2 + pos - size1;
3692 DEBUG_PRINT1 ("The compiled pattern is: ");
3693 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3694 DEBUG_PRINT1 ("The string to match is: `");
3695 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3696 DEBUG_PRINT1 ("'\n");
3698 /* This loops over pattern commands. It exits by returning from the
3699 function if the match is complete, or it drops through if the match
3700 fails at this starting point in the input data. */
3703 DEBUG_PRINT2 ("\n0x%x: ", p);
3706 { /* End of pattern means we might have succeeded. */
3707 DEBUG_PRINT1 ("end of pattern ... ");
3709 /* If we haven't matched the entire string, and we want the
3710 longest match, try backtracking. */
3711 if (d != end_match_2)
3713 /* 1 if this match ends in the same string (string1 or string2)
3714 as the best previous match. */
3715 boolean same_str_p = (FIRST_STRING_P (match_end)
3716 == MATCHING_IN_FIRST_STRING);
3717 /* 1 if this match is the best seen so far. */
3718 boolean best_match_p;
3720 /* AIX compiler got confused when this was combined
3721 with the previous declaration. */
3723 best_match_p = d > match_end;
3725 best_match_p = !MATCHING_IN_FIRST_STRING;
3727 DEBUG_PRINT1 ("backtracking.\n");
3729 if (!FAIL_STACK_EMPTY ())
3730 { /* More failure points to try. */
3732 /* If exceeds best match so far, save it. */
3733 if (!best_regs_set || best_match_p)
3735 best_regs_set = true;
3738 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3740 for (mcnt = 1; mcnt < num_regs; mcnt++)
3742 best_regstart[mcnt] = regstart[mcnt];
3743 best_regend[mcnt] = regend[mcnt];
3749 /* If no failure points, don't restore garbage. And if
3750 last match is real best match, don't restore second
3752 else if (best_regs_set && !best_match_p)
3755 /* Restore best match. It may happen that `dend ==
3756 end_match_1' while the restored d is in string2.
3757 For example, the pattern `x.*y.*z' against the
3758 strings `x-' and `y-z-', if the two strings are
3759 not consecutive in memory. */
3760 DEBUG_PRINT1 ("Restoring best registers.\n");
3763 dend = ((d >= string1 && d <= end1)
3764 ? end_match_1 : end_match_2);
3766 for (mcnt = 1; mcnt < num_regs; mcnt++)
3768 regstart[mcnt] = best_regstart[mcnt];
3769 regend[mcnt] = best_regend[mcnt];
3772 } /* d != end_match_2 */
3775 DEBUG_PRINT1 ("Accepting match.\n");
3777 /* If caller wants register contents data back, do it. */
3778 if (regs && !bufp->no_sub)
3780 /* Have the register data arrays been allocated? */
3781 if (bufp->regs_allocated == REGS_UNALLOCATED)
3782 { /* No. So allocate them with malloc. We need one
3783 extra element beyond `num_regs' for the `-1' marker
3785 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3786 regs->start = TALLOC (regs->num_regs, regoff_t);
3787 regs->end = TALLOC (regs->num_regs, regoff_t);
3788 if (regs->start == NULL || regs->end == NULL)
3793 bufp->regs_allocated = REGS_REALLOCATE;
3795 else if (bufp->regs_allocated == REGS_REALLOCATE)
3796 { /* Yes. If we need more elements than were already
3797 allocated, reallocate them. If we need fewer, just
3799 if (regs->num_regs < num_regs + 1)
3801 regs->num_regs = num_regs + 1;
3802 RETALLOC (regs->start, regs->num_regs, regoff_t);
3803 RETALLOC (regs->end, regs->num_regs, regoff_t);
3804 if (regs->start == NULL || regs->end == NULL)
3813 /* These braces fend off a "empty body in an else-statement"
3814 warning under GCC when assert expands to nothing. */
3815 assert (bufp->regs_allocated == REGS_FIXED);
3818 /* Convert the pointer data in `regstart' and `regend' to
3819 indices. Register zero has to be set differently,
3820 since we haven't kept track of any info for it. */
3821 if (regs->num_regs > 0)
3823 regs->start[0] = pos;
3824 regs->end[0] = (MATCHING_IN_FIRST_STRING
3825 ? ((regoff_t) (d - string1))
3826 : ((regoff_t) (d - string2 + size1)));
3829 /* Go through the first `min (num_regs, regs->num_regs)'
3830 registers, since that is all we initialized. */
3831 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3833 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3834 regs->start[mcnt] = regs->end[mcnt] = -1;
3838 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3840 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3844 /* If the regs structure we return has more elements than
3845 were in the pattern, set the extra elements to -1. If
3846 we (re)allocated the registers, this is the case,
3847 because we always allocate enough to have at least one
3849 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3850 regs->start[mcnt] = regs->end[mcnt] = -1;
3851 } /* regs && !bufp->no_sub */
3853 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3854 nfailure_points_pushed, nfailure_points_popped,
3855 nfailure_points_pushed - nfailure_points_popped);
3856 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3858 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3862 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3868 /* Otherwise match next pattern command. */
3869 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3871 /* Ignore these. Used to ignore the n of succeed_n's which
3872 currently have n == 0. */
3874 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3878 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3881 /* Match the next n pattern characters exactly. The following
3882 byte in the pattern defines n, and the n bytes after that
3883 are the characters to match. */
3886 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3888 /* This is written out as an if-else so we don't waste time
3889 testing `translate' inside the loop. */
3895 if (translate[(unsigned char) *d++] != (char) *p++)
3905 if (*d++ != (char) *p++) goto fail;
3909 SET_REGS_MATCHED ();
3913 /* Match any character except possibly a newline or a null. */
3915 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3919 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3920 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3923 SET_REGS_MATCHED ();
3924 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3932 register unsigned char c;
3933 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3935 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3938 c = TRANSLATE (*d); /* The character to match. */
3940 /* Cast to `unsigned' instead of `unsigned char' in case the
3941 bit list is a full 32 bytes long. */
3942 if (c < (unsigned) (*p * BYTEWIDTH)
3943 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3948 if (!not) goto fail;
3950 SET_REGS_MATCHED ();
3956 /* The beginning of a group is represented by start_memory.
3957 The arguments are the register number in the next byte, and the
3958 number of groups inner to this one in the next. The text
3959 matched within the group is recorded (in the internal
3960 registers data structure) under the register number. */
3962 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3964 /* Find out if this group can match the empty string. */
3965 p1 = p; /* To send to group_match_null_string_p. */
3967 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3968 REG_MATCH_NULL_STRING_P (reg_info[*p])
3969 = group_match_null_string_p (&p1, pend, reg_info);
3971 /* Save the position in the string where we were the last time
3972 we were at this open-group operator in case the group is
3973 operated upon by a repetition operator, e.g., with `(a*)*b'
3974 against `ab'; then we want to ignore where we are now in
3975 the string in case this attempt to match fails. */
3976 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3977 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3979 DEBUG_PRINT2 (" old_regstart: %d\n",
3980 POINTER_TO_OFFSET (old_regstart[*p]));
3983 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
3985 IS_ACTIVE (reg_info[*p]) = 1;
3986 MATCHED_SOMETHING (reg_info[*p]) = 0;
3988 /* Clear this whenever we change the register activity status. */
3989 set_regs_matched_done = 0;
3991 /* This is the new highest active register. */
3992 highest_active_reg = *p;
3994 /* If nothing was active before, this is the new lowest active
3996 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3997 lowest_active_reg = *p;
3999 /* Move past the register number and inner group count. */
4001 just_past_start_mem = p;
4006 /* The stop_memory opcode represents the end of a group. Its
4007 arguments are the same as start_memory's: the register
4008 number, and the number of inner groups. */
4010 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4012 /* We need to save the string position the last time we were at
4013 this close-group operator in case the group is operated
4014 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4015 against `aba'; then we want to ignore where we are now in
4016 the string in case this attempt to match fails. */
4017 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4018 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4020 DEBUG_PRINT2 (" old_regend: %d\n",
4021 POINTER_TO_OFFSET (old_regend[*p]));
4024 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4026 /* This register isn't active anymore. */
4027 IS_ACTIVE (reg_info[*p]) = 0;
4029 /* Clear this whenever we change the register activity status. */
4030 set_regs_matched_done = 0;
4032 /* If this was the only register active, nothing is active
4034 if (lowest_active_reg == highest_active_reg)
4036 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4037 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4040 { /* We must scan for the new highest active register, since
4041 it isn't necessarily one less than now: consider
4042 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4043 new highest active register is 1. */
4044 unsigned char r = *p - 1;
4045 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4048 /* If we end up at register zero, that means that we saved
4049 the registers as the result of an `on_failure_jump', not
4050 a `start_memory', and we jumped to past the innermost
4051 `stop_memory'. For example, in ((.)*) we save
4052 registers 1 and 2 as a result of the *, but when we pop
4053 back to the second ), we are at the stop_memory 1.
4054 Thus, nothing is active. */
4057 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4058 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4061 highest_active_reg = r;
4064 /* If just failed to match something this time around with a
4065 group that's operated on by a repetition operator, try to
4066 force exit from the ``loop'', and restore the register
4067 information for this group that we had before trying this
4069 if ((!MATCHED_SOMETHING (reg_info[*p])
4070 || just_past_start_mem == p - 1)
4073 boolean is_a_jump_n = false;
4077 switch ((re_opcode_t) *p1++)
4081 case pop_failure_jump:
4082 case maybe_pop_jump:
4084 case dummy_failure_jump:
4085 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4095 /* If the next operation is a jump backwards in the pattern
4096 to an on_failure_jump right before the start_memory
4097 corresponding to this stop_memory, exit from the loop
4098 by forcing a failure after pushing on the stack the
4099 on_failure_jump's jump in the pattern, and d. */
4100 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4101 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4103 /* If this group ever matched anything, then restore
4104 what its registers were before trying this last
4105 failed match, e.g., with `(a*)*b' against `ab' for
4106 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4107 against `aba' for regend[3].
4109 Also restore the registers for inner groups for,
4110 e.g., `((a*)(b*))*' against `aba' (register 3 would
4111 otherwise get trashed). */
4113 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4117 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4119 /* Restore this and inner groups' (if any) registers. */
4120 for (r = *p; r < *p + *(p + 1); r++)
4122 regstart[r] = old_regstart[r];
4124 /* xx why this test? */
4125 if (old_regend[r] >= regstart[r])
4126 regend[r] = old_regend[r];
4130 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4131 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4137 /* Move past the register number and the inner group count. */
4142 /* \<digit> has been turned into a `duplicate' command which is
4143 followed by the numeric value of <digit> as the register number. */
4146 register const char *d2, *dend2;
4147 int regno = *p++; /* Get which register to match against. */
4148 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4150 /* Can't back reference a group which we've never matched. */
4151 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4154 /* Where in input to try to start matching. */
4155 d2 = regstart[regno];
4157 /* Where to stop matching; if both the place to start and
4158 the place to stop matching are in the same string, then
4159 set to the place to stop, otherwise, for now have to use
4160 the end of the first string. */
4162 dend2 = ((FIRST_STRING_P (regstart[regno])
4163 == FIRST_STRING_P (regend[regno]))
4164 ? regend[regno] : end_match_1);
4167 /* If necessary, advance to next segment in register
4171 if (dend2 == end_match_2) break;
4172 if (dend2 == regend[regno]) break;
4174 /* End of string1 => advance to string2. */
4176 dend2 = regend[regno];
4178 /* At end of register contents => success */
4179 if (d2 == dend2) break;
4181 /* If necessary, advance to next segment in data. */
4184 /* How many characters left in this segment to match. */
4187 /* Want how many consecutive characters we can match in
4188 one shot, so, if necessary, adjust the count. */
4189 if (mcnt > dend2 - d2)
4192 /* Compare that many; failure if mismatch, else move
4195 ? bcmp_translate (d, d2, mcnt, translate)
4196 : bcmp (d, d2, mcnt))
4198 d += mcnt, d2 += mcnt;
4200 /* Do this because we've match some characters. */
4201 SET_REGS_MATCHED ();
4207 /* begline matches the empty string at the beginning of the string
4208 (unless `not_bol' is set in `bufp'), and, if
4209 `newline_anchor' is set, after newlines. */
4211 DEBUG_PRINT1 ("EXECUTING begline.\n");
4213 if (AT_STRINGS_BEG (d))
4215 if (!bufp->not_bol) break;
4217 else if (d[-1] == '\n' && bufp->newline_anchor)
4221 /* In all other cases, we fail. */
4225 /* endline is the dual of begline. */
4227 DEBUG_PRINT1 ("EXECUTING endline.\n");
4229 if (AT_STRINGS_END (d))
4231 if (!bufp->not_eol) break;
4234 /* We have to ``prefetch'' the next character. */
4235 else if ((d == end1 ? *string2 : *d) == '\n'
4236 && bufp->newline_anchor)
4243 /* Match at the very beginning of the data. */
4245 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4246 if (AT_STRINGS_BEG (d))
4251 /* Match at the very end of the data. */
4253 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4254 if (AT_STRINGS_END (d))
4259 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4260 pushes NULL as the value for the string on the stack. Then
4261 `pop_failure_point' will keep the current value for the
4262 string, instead of restoring it. To see why, consider
4263 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4264 then the . fails against the \n. But the next thing we want
4265 to do is match the \n against the \n; if we restored the
4266 string value, we would be back at the foo.
4268 Because this is used only in specific cases, we don't need to
4269 check all the things that `on_failure_jump' does, to make
4270 sure the right things get saved on the stack. Hence we don't
4271 share its code. The only reason to push anything on the
4272 stack at all is that otherwise we would have to change
4273 `anychar's code to do something besides goto fail in this
4274 case; that seems worse than this. */
4275 case on_failure_keep_string_jump:
4276 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4278 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4279 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4281 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4285 /* Uses of on_failure_jump:
4287 Each alternative starts with an on_failure_jump that points
4288 to the beginning of the next alternative. Each alternative
4289 except the last ends with a jump that in effect jumps past
4290 the rest of the alternatives. (They really jump to the
4291 ending jump of the following alternative, because tensioning
4292 these jumps is a hassle.)
4294 Repeats start with an on_failure_jump that points past both
4295 the repetition text and either the following jump or
4296 pop_failure_jump back to this on_failure_jump. */
4297 case on_failure_jump:
4299 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4301 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4302 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4304 /* If this on_failure_jump comes right before a group (i.e.,
4305 the original * applied to a group), save the information
4306 for that group and all inner ones, so that if we fail back
4307 to this point, the group's information will be correct.
4308 For example, in \(a*\)*\1, we need the preceding group,
4309 and in \(\(a*\)b*\)\2, we need the inner group. */
4311 /* We can't use `p' to check ahead because we push
4312 a failure point to `p + mcnt' after we do this. */
4315 /* We need to skip no_op's before we look for the
4316 start_memory in case this on_failure_jump is happening as
4317 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4319 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4322 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4324 /* We have a new highest active register now. This will
4325 get reset at the start_memory we are about to get to,
4326 but we will have saved all the registers relevant to
4327 this repetition op, as described above. */
4328 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4329 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4330 lowest_active_reg = *(p1 + 1);
4333 DEBUG_PRINT1 (":\n");
4334 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4338 /* A smart repeat ends with `maybe_pop_jump'.
4339 We change it to either `pop_failure_jump' or `jump'. */
4340 case maybe_pop_jump:
4341 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4342 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4344 register unsigned char *p2 = p;
4346 /* Compare the beginning of the repeat with what in the
4347 pattern follows its end. If we can establish that there
4348 is nothing that they would both match, i.e., that we
4349 would have to backtrack because of (as in, e.g., `a*a')
4350 then we can change to pop_failure_jump, because we'll
4351 never have to backtrack.
4353 This is not true in the case of alternatives: in
4354 `(a|ab)*' we do need to backtrack to the `ab' alternative
4355 (e.g., if the string was `ab'). But instead of trying to
4356 detect that here, the alternative has put on a dummy
4357 failure point which is what we will end up popping. */
4359 /* Skip over open/close-group commands.
4360 If what follows this loop is a ...+ construct,
4361 look at what begins its body, since we will have to
4362 match at least one of that. */
4366 && ((re_opcode_t) *p2 == stop_memory
4367 || (re_opcode_t) *p2 == start_memory))
4369 else if (p2 + 6 < pend
4370 && (re_opcode_t) *p2 == dummy_failure_jump)
4377 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4378 to the `maybe_finalize_jump' of this case. Examine what
4381 /* If we're at the end of the pattern, we can change. */
4384 /* Consider what happens when matching ":\(.*\)"
4385 against ":/". I don't really understand this code
4387 p[-3] = (unsigned char) pop_failure_jump;
4389 (" End of pattern: change to `pop_failure_jump'.\n");
4392 else if ((re_opcode_t) *p2 == exactn
4393 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4395 register unsigned char c
4396 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4398 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4400 p[-3] = (unsigned char) pop_failure_jump;
4401 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4405 else if ((re_opcode_t) p1[3] == charset
4406 || (re_opcode_t) p1[3] == charset_not)
4408 int not = (re_opcode_t) p1[3] == charset_not;
4410 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4411 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4414 /* `not' is equal to 1 if c would match, which means
4415 that we can't change to pop_failure_jump. */
4418 p[-3] = (unsigned char) pop_failure_jump;
4419 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4423 else if ((re_opcode_t) *p2 == charset)
4426 register unsigned char c
4427 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4430 if ((re_opcode_t) p1[3] == exactn
4431 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4432 && (p2[1 + p1[4] / BYTEWIDTH]
4433 & (1 << (p1[4] % BYTEWIDTH)))))
4435 p[-3] = (unsigned char) pop_failure_jump;
4436 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4440 else if ((re_opcode_t) p1[3] == charset_not)
4443 /* We win if the charset_not inside the loop
4444 lists every character listed in the charset after. */
4445 for (idx = 0; idx < (int) p2[1]; idx++)
4446 if (! (p2[2 + idx] == 0
4447 || (idx < (int) p1[4]
4448 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4453 p[-3] = (unsigned char) pop_failure_jump;
4454 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4457 else if ((re_opcode_t) p1[3] == charset)
4460 /* We win if the charset inside the loop
4461 has no overlap with the one after the loop. */
4463 idx < (int) p2[1] && idx < (int) p1[4];
4465 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4468 if (idx == p2[1] || idx == p1[4])
4470 p[-3] = (unsigned char) pop_failure_jump;
4471 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4476 p -= 2; /* Point at relative address again. */
4477 if ((re_opcode_t) p[-1] != pop_failure_jump)
4479 p[-1] = (unsigned char) jump;
4480 DEBUG_PRINT1 (" Match => jump.\n");
4481 goto unconditional_jump;
4483 /* Note fall through. */
4486 /* The end of a simple repeat has a pop_failure_jump back to
4487 its matching on_failure_jump, where the latter will push a
4488 failure point. The pop_failure_jump takes off failure
4489 points put on by this pop_failure_jump's matching
4490 on_failure_jump; we got through the pattern to here from the
4491 matching on_failure_jump, so didn't fail. */
4492 case pop_failure_jump:
4494 /* We need to pass separate storage for the lowest and
4495 highest registers, even though we don't care about the
4496 actual values. Otherwise, we will restore only one
4497 register from the stack, since lowest will == highest in
4498 `pop_failure_point'. */
4499 unsigned dummy_low_reg, dummy_high_reg;
4500 unsigned char *pdummy;
4503 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4504 POP_FAILURE_POINT (sdummy, pdummy,
4505 dummy_low_reg, dummy_high_reg,
4506 reg_dummy, reg_dummy, reg_info_dummy);
4508 /* Note fall through. */
4511 /* Unconditionally jump (without popping any failure points). */
4514 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4515 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4516 p += mcnt; /* Do the jump. */
4517 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4521 /* We need this opcode so we can detect where alternatives end
4522 in `group_match_null_string_p' et al. */
4524 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4525 goto unconditional_jump;
4528 /* Normally, the on_failure_jump pushes a failure point, which
4529 then gets popped at pop_failure_jump. We will end up at
4530 pop_failure_jump, also, and with a pattern of, say, `a+', we
4531 are skipping over the on_failure_jump, so we have to push
4532 something meaningless for pop_failure_jump to pop. */
4533 case dummy_failure_jump:
4534 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4535 /* It doesn't matter what we push for the string here. What
4536 the code at `fail' tests is the value for the pattern. */
4537 PUSH_FAILURE_POINT (0, 0, -2);
4538 goto unconditional_jump;
4541 /* At the end of an alternative, we need to push a dummy failure
4542 point in case we are followed by a `pop_failure_jump', because
4543 we don't want the failure point for the alternative to be
4544 popped. For example, matching `(a|ab)*' against `aab'
4545 requires that we match the `ab' alternative. */
4546 case push_dummy_failure:
4547 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4548 /* See comments just above at `dummy_failure_jump' about the
4550 PUSH_FAILURE_POINT (0, 0, -2);
4553 /* Have to succeed matching what follows at least n times.
4554 After that, handle like `on_failure_jump'. */
4556 EXTRACT_NUMBER (mcnt, p + 2);
4557 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4560 /* Originally, this is how many times we HAVE to succeed. */
4565 STORE_NUMBER_AND_INCR (p, mcnt);
4566 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4570 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4571 p[2] = (unsigned char) no_op;
4572 p[3] = (unsigned char) no_op;
4578 EXTRACT_NUMBER (mcnt, p + 2);
4579 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4581 /* Originally, this is how many times we CAN jump. */
4585 STORE_NUMBER (p + 2, mcnt);
4586 goto unconditional_jump;
4588 /* If don't have to jump any more, skip over the rest of command. */
4595 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4597 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4599 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4600 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4601 STORE_NUMBER (p1, mcnt);
4606 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4607 if (AT_WORD_BOUNDARY (d))
4612 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4613 if (AT_WORD_BOUNDARY (d))
4618 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4619 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4624 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4625 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4626 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4632 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4633 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4638 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4639 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4644 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4645 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4648 #if 0 /* not emacs19 */
4650 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4651 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4654 #endif /* not emacs19 */
4657 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4662 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4666 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4668 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4670 SET_REGS_MATCHED ();
4674 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4676 goto matchnotsyntax;
4679 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4683 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4685 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4687 SET_REGS_MATCHED ();
4690 #else /* not emacs */
4692 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4694 if (!WORDCHAR_P (d))
4696 SET_REGS_MATCHED ();
4701 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4705 SET_REGS_MATCHED ();
4708 #endif /* not emacs */
4713 continue; /* Successfully executed one pattern command; keep going. */
4716 /* We goto here if a matching operation fails. */
4718 if (!FAIL_STACK_EMPTY ())
4719 { /* A restart point is known. Restore to that state. */
4720 DEBUG_PRINT1 ("\nFAIL:\n");
4721 POP_FAILURE_POINT (d, p,
4722 lowest_active_reg, highest_active_reg,
4723 regstart, regend, reg_info);
4725 /* If this failure point is a dummy, try the next one. */
4729 /* If we failed to the end of the pattern, don't examine *p. */
4733 boolean is_a_jump_n = false;
4735 /* If failed to a backwards jump that's part of a repetition
4736 loop, need to pop this failure point and use the next one. */
4737 switch ((re_opcode_t) *p)
4741 case maybe_pop_jump:
4742 case pop_failure_jump:
4745 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4748 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4750 && (re_opcode_t) *p1 == on_failure_jump))
4758 if (d >= string1 && d <= end1)
4762 break; /* Matching at this starting point really fails. */
4766 goto restore_best_regs;
4770 return -1; /* Failure to match. */
4773 /* Subroutine definitions for re_match_2. */
4776 /* We are passed P pointing to a register number after a start_memory.
4778 Return true if the pattern up to the corresponding stop_memory can
4779 match the empty string, and false otherwise.
4781 If we find the matching stop_memory, sets P to point to one past its number.
4782 Otherwise, sets P to an undefined byte less than or equal to END.
4784 We don't handle duplicates properly (yet). */
4787 group_match_null_string_p (p, end, reg_info)
4788 unsigned char **p, *end;
4789 register_info_type *reg_info;
4792 /* Point to after the args to the start_memory. */
4793 unsigned char *p1 = *p + 2;
4797 /* Skip over opcodes that can match nothing, and return true or
4798 false, as appropriate, when we get to one that can't, or to the
4799 matching stop_memory. */
4801 switch ((re_opcode_t) *p1)
4803 /* Could be either a loop or a series of alternatives. */
4804 case on_failure_jump:
4806 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4808 /* If the next operation is not a jump backwards in the
4813 /* Go through the on_failure_jumps of the alternatives,
4814 seeing if any of the alternatives cannot match nothing.
4815 The last alternative starts with only a jump,
4816 whereas the rest start with on_failure_jump and end
4817 with a jump, e.g., here is the pattern for `a|b|c':
4819 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4820 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4823 So, we have to first go through the first (n-1)
4824 alternatives and then deal with the last one separately. */
4827 /* Deal with the first (n-1) alternatives, which start
4828 with an on_failure_jump (see above) that jumps to right
4829 past a jump_past_alt. */
4831 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4833 /* `mcnt' holds how many bytes long the alternative
4834 is, including the ending `jump_past_alt' and
4837 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4841 /* Move to right after this alternative, including the
4845 /* Break if it's the beginning of an n-th alternative
4846 that doesn't begin with an on_failure_jump. */
4847 if ((re_opcode_t) *p1 != on_failure_jump)
4850 /* Still have to check that it's not an n-th
4851 alternative that starts with an on_failure_jump. */
4853 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4854 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4856 /* Get to the beginning of the n-th alternative. */
4862 /* Deal with the last alternative: go back and get number
4863 of the `jump_past_alt' just before it. `mcnt' contains
4864 the length of the alternative. */
4865 EXTRACT_NUMBER (mcnt, p1 - 2);
4867 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4870 p1 += mcnt; /* Get past the n-th alternative. */
4876 assert (p1[1] == **p);
4882 if (!common_op_match_null_string_p (&p1, end, reg_info))
4885 } /* while p1 < end */
4888 } /* group_match_null_string_p */
4891 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4892 It expects P to be the first byte of a single alternative and END one
4893 byte past the last. The alternative can contain groups. */
4896 alt_match_null_string_p (p, end, reg_info)
4897 unsigned char *p, *end;
4898 register_info_type *reg_info;
4901 unsigned char *p1 = p;
4905 /* Skip over opcodes that can match nothing, and break when we get
4906 to one that can't. */
4908 switch ((re_opcode_t) *p1)
4911 case on_failure_jump:
4913 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4918 if (!common_op_match_null_string_p (&p1, end, reg_info))
4921 } /* while p1 < end */
4924 } /* alt_match_null_string_p */
4927 /* Deals with the ops common to group_match_null_string_p and
4928 alt_match_null_string_p.
4930 Sets P to one after the op and its arguments, if any. */
4933 common_op_match_null_string_p (p, end, reg_info)
4934 unsigned char **p, *end;
4935 register_info_type *reg_info;
4940 unsigned char *p1 = *p;
4942 switch ((re_opcode_t) *p1++)
4962 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4963 ret = group_match_null_string_p (&p1, end, reg_info);
4965 /* Have to set this here in case we're checking a group which
4966 contains a group and a back reference to it. */
4968 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4969 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4975 /* If this is an optimized succeed_n for zero times, make the jump. */
4977 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4985 /* Get to the number of times to succeed. */
4987 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4992 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5000 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5008 /* All other opcodes mean we cannot match the empty string. */
5014 } /* common_op_match_null_string_p */
5017 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5018 bytes; nonzero otherwise. */
5021 bcmp_translate (s1, s2, len, translate)
5022 unsigned char *s1, *s2;
5026 register unsigned char *p1 = s1, *p2 = s2;
5029 if (translate[*p1++] != translate[*p2++]) return 1;
5035 /* Entry points for GNU code. */
5037 /* re_compile_pattern is the GNU regular expression compiler: it
5038 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5039 Returns 0 if the pattern was valid, otherwise an error string.
5041 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5042 are set in BUFP on entry.
5044 We call regex_compile to do the actual compilation. */
5047 re_compile_pattern (pattern, length, bufp)
5048 const char *pattern;
5050 struct re_pattern_buffer *bufp;
5054 /* GNU code is written to assume at least RE_NREGS registers will be set
5055 (and at least one extra will be -1). */
5056 bufp->regs_allocated = REGS_UNALLOCATED;
5058 /* And GNU code determines whether or not to get register information
5059 by passing null for the REGS argument to re_match, etc., not by
5063 /* Match anchors at newline. */
5064 bufp->newline_anchor = 1;
5066 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5070 return gettext (re_error_msgid[(int) ret]);
5073 /* Entry points compatible with 4.2 BSD regex library. We don't define
5074 them unless specifically requested. */
5076 #ifdef _REGEX_RE_COMP
5078 /* BSD has one and only one pattern buffer. */
5079 static struct re_pattern_buffer re_comp_buf;
5089 if (!re_comp_buf.buffer)
5090 return gettext ("No previous regular expression");
5094 if (!re_comp_buf.buffer)
5096 re_comp_buf.buffer = (unsigned char *) malloc (200);
5097 if (re_comp_buf.buffer == NULL)
5098 return gettext (re_error_msgid[(int) REG_ESPACE]);
5099 re_comp_buf.allocated = 200;
5101 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5102 if (re_comp_buf.fastmap == NULL)
5103 return gettext (re_error_msgid[(int) REG_ESPACE]);
5106 /* Since `re_exec' always passes NULL for the `regs' argument, we
5107 don't need to initialize the pattern buffer fields which affect it. */
5109 /* Match anchors at newlines. */
5110 re_comp_buf.newline_anchor = 1;
5112 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5117 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5118 return (char *) gettext (re_error_msgid[(int) ret]);
5126 const int len = strlen (s);
5128 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5130 #endif /* _REGEX_RE_COMP */
5132 /* POSIX.2 functions. Don't define these for Emacs. */
5136 /* regcomp takes a regular expression as a string and compiles it.
5138 PREG is a regex_t *. We do not expect any fields to be initialized,
5139 since POSIX says we shouldn't. Thus, we set
5141 `buffer' to the compiled pattern;
5142 `used' to the length of the compiled pattern;
5143 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5144 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5145 RE_SYNTAX_POSIX_BASIC;
5146 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5147 `fastmap' and `fastmap_accurate' to zero;
5148 `re_nsub' to the number of subexpressions in PATTERN.
5150 PATTERN is the address of the pattern string.
5152 CFLAGS is a series of bits which affect compilation.
5154 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5155 use POSIX basic syntax.
5157 If REG_NEWLINE is set, then . and [^...] don't match newline.
5158 Also, regexec will try a match beginning after every newline.
5160 If REG_ICASE is set, then we considers upper- and lowercase
5161 versions of letters to be equivalent when matching.
5163 If REG_NOSUB is set, then when PREG is passed to regexec, that
5164 routine will report only success or failure, and nothing about the
5167 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5168 the return codes and their meanings.) */
5171 regcomp (preg, pattern, cflags)
5173 const char *pattern;
5178 = (cflags & REG_EXTENDED) ?
5179 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5181 /* regex_compile will allocate the space for the compiled pattern. */
5183 preg->allocated = 0;
5186 /* Don't bother to use a fastmap when searching. This simplifies the
5187 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5188 characters after newlines into the fastmap. This way, we just try
5192 if (cflags & REG_ICASE)
5196 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5197 if (preg->translate == NULL)
5198 return (int) REG_ESPACE;
5200 /* Map uppercase characters to corresponding lowercase ones. */
5201 for (i = 0; i < CHAR_SET_SIZE; i++)
5202 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5205 preg->translate = NULL;
5207 /* If REG_NEWLINE is set, newlines are treated differently. */
5208 if (cflags & REG_NEWLINE)
5209 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5210 syntax &= ~RE_DOT_NEWLINE;
5211 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5212 /* It also changes the matching behavior. */
5213 preg->newline_anchor = 1;
5216 preg->newline_anchor = 0;
5218 preg->no_sub = !!(cflags & REG_NOSUB);
5220 /* POSIX says a null character in the pattern terminates it, so we
5221 can use strlen here in compiling the pattern. */
5222 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5224 /* POSIX doesn't distinguish between an unmatched open-group and an
5225 unmatched close-group: both are REG_EPAREN. */
5226 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5232 /* regexec searches for a given pattern, specified by PREG, in the
5235 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5236 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5237 least NMATCH elements, and we set them to the offsets of the
5238 corresponding matched substrings.
5240 EFLAGS specifies `execution flags' which affect matching: if
5241 REG_NOTBOL is set, then ^ does not match at the beginning of the
5242 string; if REG_NOTEOL is set, then $ does not match at the end.
5244 We return 0 if we find a match and REG_NOMATCH if not. */
5247 regexec (preg, string, nmatch, pmatch, eflags)
5248 const regex_t *preg;
5251 regmatch_t pmatch[];
5255 struct re_registers regs;
5256 regex_t private_preg;
5257 int len = strlen (string);
5258 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5260 private_preg = *preg;
5262 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5263 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5265 /* The user has told us exactly how many registers to return
5266 information about, via `nmatch'. We have to pass that on to the
5267 matching routines. */
5268 private_preg.regs_allocated = REGS_FIXED;
5272 regs.num_regs = nmatch;
5273 regs.start = TALLOC (nmatch, regoff_t);
5274 regs.end = TALLOC (nmatch, regoff_t);
5275 if (regs.start == NULL || regs.end == NULL)
5276 return (int) REG_NOMATCH;
5279 /* Perform the searching operation. */
5280 ret = re_search (&private_preg, string, len,
5281 /* start: */ 0, /* range: */ len,
5282 want_reg_info ? ®s : (struct re_registers *) 0);
5284 /* Copy the register information to the POSIX structure. */
5291 for (r = 0; r < nmatch; r++)
5293 pmatch[r].rm_so = regs.start[r];
5294 pmatch[r].rm_eo = regs.end[r];
5298 /* If we needed the temporary register info, free the space now. */
5303 /* We want zero return to mean success, unlike `re_search'. */
5304 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5308 /* Returns a message corresponding to an error code, ERRCODE, returned
5309 from either regcomp or regexec. We don't use PREG here. */
5312 regerror (errcode, preg, errbuf, errbuf_size)
5314 const regex_t *preg;
5322 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5323 /* Only error codes returned by the rest of the code should be passed
5324 to this routine. If we are given anything else, or if other regex
5325 code generates an invalid error code, then the program has a bug.
5326 Dump core so we can fix it. */
5329 msg = gettext (re_error_msgid[errcode]);
5331 msg_size = strlen (msg) + 1; /* Includes the null. */
5333 if (errbuf_size != 0)
5335 if (msg_size > errbuf_size)
5337 strncpy (errbuf, msg, errbuf_size - 1);
5338 errbuf[errbuf_size - 1] = 0;
5341 strcpy (errbuf, msg);
5348 /* Free dynamically allocated space used by PREG. */
5354 if (preg->buffer != NULL)
5355 free (preg->buffer);
5356 preg->buffer = NULL;
5358 preg->allocated = 0;
5361 if (preg->fastmap != NULL)
5362 free (preg->fastmap);
5363 preg->fastmap = NULL;
5364 preg->fastmap_accurate = 0;
5366 if (preg->translate != NULL)
5367 free (preg->translate);
5368 preg->translate = NULL;
5371 #endif /* not emacs */
5375 make-backup-files: t
5377 trim-versions-without-asking: nil