1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993 Free Software Foundation, Inc.
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
29 /* We need this for `regex.h', and perhaps for the Emacs include files. */
30 #include <sys/types.h>
36 /* The `emacs' switch turns on certain matching commands
37 that make sense only in Emacs. */
44 /* Emacs uses `NULL' as a predicate. */
49 /* We used to test for `BSTRING' here, but only GCC and Emacs define
50 `BSTRING', as far as I know, and neither of them use this code. */
51 #if HAVE_STRING_H || STDC_HEADERS
54 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
57 #define bcopy(s, d, n) memcpy ((d), (s), (n))
60 #define bzero(s, n) memset ((s), 0, (n))
74 /* Define the syntax stuff for \<, \>, etc. */
76 /* This must be nonzero for the wordchar and notwordchar pattern
77 commands in re_match_2. */
84 extern char *re_syntax_table;
86 #else /* not SYNTAX_TABLE */
88 /* How many characters in the character set. */
89 #define CHAR_SET_SIZE 256
91 static char re_syntax_table[CHAR_SET_SIZE];
102 bzero (re_syntax_table, sizeof re_syntax_table);
104 for (c = 'a'; c <= 'z'; c++)
105 re_syntax_table[c] = Sword;
107 for (c = 'A'; c <= 'Z'; c++)
108 re_syntax_table[c] = Sword;
110 for (c = '0'; c <= '9'; c++)
111 re_syntax_table[c] = Sword;
113 re_syntax_table['_'] = Sword;
118 #endif /* not SYNTAX_TABLE */
120 #define SYNTAX(c) re_syntax_table[c]
122 #endif /* not emacs */
124 /* Get the interface, including the syntax bits. */
127 /* isalpha etc. are used for the character classes. */
130 /* Jim Meyering writes:
132 "... Some ctype macros are valid only for character codes that
133 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
134 using /bin/cc or gcc but without giving an ansi option). So, all
135 ctype uses should be through macros like ISPRINT... If
136 STDC_HEADERS is defined, then autoconf has verified that the ctype
137 macros don't need to be guarded with references to isascii. ...
138 Defining isascii to 1 should let any compiler worth its salt
139 eliminate the && through constant folding." */
140 #if ! defined (isascii) || defined (STDC_HEADERS)
146 #define ISBLANK(c) (isascii (c) && isblank (c))
148 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
151 #define ISGRAPH(c) (isascii (c) && isgraph (c))
153 #define ISGRAPH(c) (isascii (c) && isprint (c) && !isspace (c))
156 #define ISPRINT(c) (isascii (c) && isprint (c))
157 #define ISDIGIT(c) (isascii (c) && isdigit (c))
158 #define ISALNUM(c) (isascii (c) && isalnum (c))
159 #define ISALPHA(c) (isascii (c) && isalpha (c))
160 #define ISCNTRL(c) (isascii (c) && iscntrl (c))
161 #define ISLOWER(c) (isascii (c) && islower (c))
162 #define ISPUNCT(c) (isascii (c) && ispunct (c))
163 #define ISSPACE(c) (isascii (c) && isspace (c))
164 #define ISUPPER(c) (isascii (c) && isupper (c))
165 #define ISXDIGIT(c) (isascii (c) && isxdigit (c))
171 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
172 since ours (we hope) works properly with all combinations of
173 machines, compilers, `char' and `unsigned char' argument types.
174 (Per Bothner suggested the basic approach.) */
175 #undef SIGN_EXTEND_CHAR
177 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
178 #else /* not __STDC__ */
179 /* As in Harbison and Steele. */
180 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
183 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
184 use `alloca' instead of `malloc'. This is because using malloc in
185 re_search* or re_match* could cause memory leaks when C-g is used in
186 Emacs; also, malloc is slower and causes storage fragmentation. On
187 the other hand, malloc is more portable, and easier to debug.
189 Because we sometimes use alloca, some routines have to be macros,
190 not functions -- `alloca'-allocated space disappears at the end of the
191 function it is called in. */
195 #define REGEX_ALLOCATE malloc
196 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
198 #else /* not REGEX_MALLOC */
200 /* Emacs already defines alloca, sometimes. */
203 /* Make alloca work the best possible way. */
205 #define alloca __builtin_alloca
206 #else /* not __GNUC__ */
209 #else /* not __GNUC__ or HAVE_ALLOCA_H */
210 #ifndef _AIX /* Already did AIX, up at the top. */
212 #endif /* not _AIX */
213 #endif /* not HAVE_ALLOCA_H */
214 #endif /* not __GNUC__ */
216 #endif /* not alloca */
218 #define REGEX_ALLOCATE alloca
220 /* Assumes a `char *destination' variable. */
221 #define REGEX_REALLOCATE(source, osize, nsize) \
222 (destination = (char *) alloca (nsize), \
223 bcopy (source, destination, osize), \
226 #endif /* not REGEX_MALLOC */
229 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
230 `string1' or just past its end. This works if PTR is NULL, which is
232 #define FIRST_STRING_P(ptr) \
233 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
235 /* (Re)Allocate N items of type T using malloc, or fail. */
236 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
237 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
238 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
240 #define BYTEWIDTH 8 /* In bits. */
242 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
244 #define MAX(a, b) ((a) > (b) ? (a) : (b))
245 #define MIN(a, b) ((a) < (b) ? (a) : (b))
247 typedef char boolean;
251 /* These are the command codes that appear in compiled regular
252 expressions. Some opcodes are followed by argument bytes. A
253 command code can specify any interpretation whatsoever for its
254 arguments. Zero bytes may appear in the compiled regular expression.
256 The value of `exactn' is needed in search.c (search_buffer) in Emacs.
257 So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of
258 `exactn' we use here must also be 1. */
264 /* Followed by one byte giving n, then by n literal bytes. */
267 /* Matches any (more or less) character. */
270 /* Matches any one char belonging to specified set. First
271 following byte is number of bitmap bytes. Then come bytes
272 for a bitmap saying which chars are in. Bits in each byte
273 are ordered low-bit-first. A character is in the set if its
274 bit is 1. A character too large to have a bit in the map is
275 automatically not in the set. */
278 /* Same parameters as charset, but match any character that is
279 not one of those specified. */
282 /* Start remembering the text that is matched, for storing in a
283 register. Followed by one byte with the register number, in
284 the range 0 to one less than the pattern buffer's re_nsub
285 field. Then followed by one byte with the number of groups
286 inner to this one. (This last has to be part of the
287 start_memory only because we need it in the on_failure_jump
291 /* Stop remembering the text that is matched and store it in a
292 memory register. Followed by one byte with the register
293 number, in the range 0 to one less than `re_nsub' in the
294 pattern buffer, and one byte with the number of inner groups,
295 just like `start_memory'. (We need the number of inner
296 groups here because we don't have any easy way of finding the
297 corresponding start_memory when we're at a stop_memory.) */
300 /* Match a duplicate of something remembered. Followed by one
301 byte containing the register number. */
304 /* Fail unless at beginning of line. */
307 /* Fail unless at end of line. */
310 /* Succeeds if at beginning of buffer (if emacs) or at beginning
311 of string to be matched (if not). */
314 /* Analogously, for end of buffer/string. */
317 /* Followed by two byte relative address to which to jump. */
320 /* Same as jump, but marks the end of an alternative. */
323 /* Followed by two-byte relative address of place to resume at
324 in case of failure. */
327 /* Like on_failure_jump, but pushes a placeholder instead of the
328 current string position when executed. */
329 on_failure_keep_string_jump,
331 /* Throw away latest failure point and then jump to following
332 two-byte relative address. */
335 /* Change to pop_failure_jump if know won't have to backtrack to
336 match; otherwise change to jump. This is used to jump
337 back to the beginning of a repeat. If what follows this jump
338 clearly won't match what the repeat does, such that we can be
339 sure that there is no use backtracking out of repetitions
340 already matched, then we change it to a pop_failure_jump.
341 Followed by two-byte address. */
344 /* Jump to following two-byte address, and push a dummy failure
345 point. This failure point will be thrown away if an attempt
346 is made to use it for a failure. A `+' construct makes this
347 before the first repeat. Also used as an intermediary kind
348 of jump when compiling an alternative. */
351 /* Push a dummy failure point and continue. Used at the end of
355 /* Followed by two-byte relative address and two-byte number n.
356 After matching N times, jump to the address upon failure. */
359 /* Followed by two-byte relative address, and two-byte number n.
360 Jump to the address N times, then fail. */
363 /* Set the following two-byte relative address to the
364 subsequent two-byte number. The address *includes* the two
368 wordchar, /* Matches any word-constituent character. */
369 notwordchar, /* Matches any char that is not a word-constituent. */
371 wordbeg, /* Succeeds if at word beginning. */
372 wordend, /* Succeeds if at word end. */
374 wordbound, /* Succeeds if at a word boundary. */
375 notwordbound /* Succeeds if not at a word boundary. */
378 ,before_dot, /* Succeeds if before point. */
379 at_dot, /* Succeeds if at point. */
380 after_dot, /* Succeeds if after point. */
382 /* Matches any character whose syntax is specified. Followed by
383 a byte which contains a syntax code, e.g., Sword. */
386 /* Matches any character whose syntax is not that specified. */
391 /* Common operations on the compiled pattern. */
393 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
395 #define STORE_NUMBER(destination, number) \
397 (destination)[0] = (number) & 0377; \
398 (destination)[1] = (number) >> 8; \
401 /* Same as STORE_NUMBER, except increment DESTINATION to
402 the byte after where the number is stored. Therefore, DESTINATION
403 must be an lvalue. */
405 #define STORE_NUMBER_AND_INCR(destination, number) \
407 STORE_NUMBER (destination, number); \
408 (destination) += 2; \
411 /* Put into DESTINATION a number stored in two contiguous bytes starting
414 #define EXTRACT_NUMBER(destination, source) \
416 (destination) = *(source) & 0377; \
417 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
422 extract_number (dest, source)
424 unsigned char *source;
426 int temp = SIGN_EXTEND_CHAR (*(source + 1));
427 *dest = *source & 0377;
431 #ifndef EXTRACT_MACROS /* To debug the macros. */
432 #undef EXTRACT_NUMBER
433 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
434 #endif /* not EXTRACT_MACROS */
438 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
439 SOURCE must be an lvalue. */
441 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
443 EXTRACT_NUMBER (destination, source); \
449 extract_number_and_incr (destination, source)
451 unsigned char **source;
453 extract_number (destination, *source);
457 #ifndef EXTRACT_MACROS
458 #undef EXTRACT_NUMBER_AND_INCR
459 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
460 extract_number_and_incr (&dest, &src)
461 #endif /* not EXTRACT_MACROS */
465 /* If DEBUG is defined, Regex prints many voluminous messages about what
466 it is doing (if the variable `debug' is nonzero). If linked with the
467 main program in `iregex.c', you can enter patterns and strings
468 interactively. And if linked with the main program in `main.c' and
469 the other test files, you can run the already-written tests. */
473 /* We use standard I/O for debugging. */
476 /* It is useful to test things that ``must'' be true when debugging. */
479 static int debug = 0;
481 #define DEBUG_STATEMENT(e) e
482 #define DEBUG_PRINT1(x) if (debug) printf (x)
483 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
484 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
485 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
486 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
487 if (debug) print_partial_compiled_pattern (s, e)
488 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
489 if (debug) print_double_string (w, s1, sz1, s2, sz2)
492 extern void printchar ();
494 /* Print the fastmap in human-readable form. */
497 print_fastmap (fastmap)
500 unsigned was_a_range = 0;
503 while (i < (1 << BYTEWIDTH))
509 while (i < (1 << BYTEWIDTH) && fastmap[i])
525 /* Print a compiled pattern string in human-readable form, starting at
526 the START pointer into it and ending just before the pointer END. */
529 print_partial_compiled_pattern (start, end)
530 unsigned char *start;
534 unsigned char *p = start;
535 unsigned char *pend = end;
543 /* Loop over pattern commands. */
546 printf ("%d:\t", p - start);
548 switch ((re_opcode_t) *p++)
556 printf ("/exactn/%d", mcnt);
567 printf ("/start_memory/%d/%d", mcnt, *p++);
572 printf ("/stop_memory/%d/%d", mcnt, *p++);
576 printf ("/duplicate/%d", *p++);
586 register int c, last = -100;
587 register int in_range = 0;
589 printf ("/charset [%s",
590 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
592 assert (p + *p < pend);
594 for (c = 0; c < 256; c++)
596 && (p[1 + (c/8)] & (1 << (c % 8))))
598 /* Are we starting a range? */
599 if (last + 1 == c && ! in_range)
604 /* Have we broken a range? */
605 else if (last + 1 != c && in_range)
634 case on_failure_jump:
635 extract_number_and_incr (&mcnt, &p);
636 printf ("/on_failure_jump to %d", p + mcnt - start);
639 case on_failure_keep_string_jump:
640 extract_number_and_incr (&mcnt, &p);
641 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
644 case dummy_failure_jump:
645 extract_number_and_incr (&mcnt, &p);
646 printf ("/dummy_failure_jump to %d", p + mcnt - start);
649 case push_dummy_failure:
650 printf ("/push_dummy_failure");
654 extract_number_and_incr (&mcnt, &p);
655 printf ("/maybe_pop_jump to %d", p + mcnt - start);
658 case pop_failure_jump:
659 extract_number_and_incr (&mcnt, &p);
660 printf ("/pop_failure_jump to %d", p + mcnt - start);
664 extract_number_and_incr (&mcnt, &p);
665 printf ("/jump_past_alt to %d", p + mcnt - start);
669 extract_number_and_incr (&mcnt, &p);
670 printf ("/jump to %d", p + mcnt - start);
674 extract_number_and_incr (&mcnt, &p);
675 extract_number_and_incr (&mcnt2, &p);
676 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
680 extract_number_and_incr (&mcnt, &p);
681 extract_number_and_incr (&mcnt2, &p);
682 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
686 extract_number_and_incr (&mcnt, &p);
687 extract_number_and_incr (&mcnt2, &p);
688 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
692 printf ("/wordbound");
696 printf ("/notwordbound");
708 printf ("/before_dot");
716 printf ("/after_dot");
720 printf ("/syntaxspec");
722 printf ("/%d", mcnt);
726 printf ("/notsyntaxspec");
728 printf ("/%d", mcnt);
733 printf ("/wordchar");
737 printf ("/notwordchar");
749 printf ("?%d", *(p-1));
755 printf ("%d:\tend of pattern.\n", p - start);
760 print_compiled_pattern (bufp)
761 struct re_pattern_buffer *bufp;
763 unsigned char *buffer = bufp->buffer;
765 print_partial_compiled_pattern (buffer, buffer + bufp->used);
766 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
768 if (bufp->fastmap_accurate && bufp->fastmap)
770 printf ("fastmap: ");
771 print_fastmap (bufp->fastmap);
774 printf ("re_nsub: %d\t", bufp->re_nsub);
775 printf ("regs_alloc: %d\t", bufp->regs_allocated);
776 printf ("can_be_null: %d\t", bufp->can_be_null);
777 printf ("newline_anchor: %d\n", bufp->newline_anchor);
778 printf ("no_sub: %d\t", bufp->no_sub);
779 printf ("not_bol: %d\t", bufp->not_bol);
780 printf ("not_eol: %d\t", bufp->not_eol);
781 printf ("syntax: %d\n", bufp->syntax);
782 /* Perhaps we should print the translate table? */
787 print_double_string (where, string1, size1, string2, size2)
800 if (FIRST_STRING_P (where))
802 for (this_char = where - string1; this_char < size1; this_char++)
803 printchar (string1[this_char]);
808 for (this_char = where - string2; this_char < size2; this_char++)
809 printchar (string2[this_char]);
813 #else /* not DEBUG */
818 #define DEBUG_STATEMENT(e)
819 #define DEBUG_PRINT1(x)
820 #define DEBUG_PRINT2(x1, x2)
821 #define DEBUG_PRINT3(x1, x2, x3)
822 #define DEBUG_PRINT4(x1, x2, x3, x4)
823 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
824 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
826 #endif /* not DEBUG */
828 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
829 also be assigned to arbitrarily: each pattern buffer stores its own
830 syntax, so it can be changed between regex compilations. */
831 reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS;
834 /* Specify the precise syntax of regexps for compilation. This provides
835 for compatibility for various utilities which historically have
836 different, incompatible syntaxes.
838 The argument SYNTAX is a bit mask comprised of the various bits
839 defined in regex.h. We return the old syntax. */
842 re_set_syntax (syntax)
845 reg_syntax_t ret = re_syntax_options;
847 re_syntax_options = syntax;
851 /* This table gives an error message for each of the error codes listed
852 in regex.h. Obviously the order here has to be same as there. */
854 static const char *re_error_msg[] =
855 { NULL, /* REG_NOERROR */
856 "No match", /* REG_NOMATCH */
857 "Invalid regular expression", /* REG_BADPAT */
858 "Invalid collation character", /* REG_ECOLLATE */
859 "Invalid character class name", /* REG_ECTYPE */
860 "Trailing backslash", /* REG_EESCAPE */
861 "Invalid back reference", /* REG_ESUBREG */
862 "Unmatched [ or [^", /* REG_EBRACK */
863 "Unmatched ( or \\(", /* REG_EPAREN */
864 "Unmatched \\{", /* REG_EBRACE */
865 "Invalid content of \\{\\}", /* REG_BADBR */
866 "Invalid range end", /* REG_ERANGE */
867 "Memory exhausted", /* REG_ESPACE */
868 "Invalid preceding regular expression", /* REG_BADRPT */
869 "Premature end of regular expression", /* REG_EEND */
870 "Regular expression too big", /* REG_ESIZE */
871 "Unmatched ) or \\)", /* REG_ERPAREN */
874 /* Subroutine declarations and macros for regex_compile. */
876 static void store_op1 (), store_op2 ();
877 static void insert_op1 (), insert_op2 ();
878 static boolean at_begline_loc_p (), at_endline_loc_p ();
879 static boolean group_in_compile_stack ();
880 static reg_errcode_t compile_range ();
882 /* Fetch the next character in the uncompiled pattern---translating it
883 if necessary. Also cast from a signed character in the constant
884 string passed to us by the user to an unsigned char that we can use
885 as an array index (in, e.g., `translate'). */
886 #define PATFETCH(c) \
887 do {if (p == pend) return REG_EEND; \
888 c = (unsigned char) *p++; \
889 if (translate) c = translate[c]; \
892 /* Fetch the next character in the uncompiled pattern, with no
894 #define PATFETCH_RAW(c) \
895 do {if (p == pend) return REG_EEND; \
896 c = (unsigned char) *p++; \
899 /* Go backwards one character in the pattern. */
900 #define PATUNFETCH p--
903 /* If `translate' is non-null, return translate[D], else just D. We
904 cast the subscript to translate because some data is declared as
905 `char *', to avoid warnings when a string constant is passed. But
906 when we use a character as a subscript we must make it unsigned. */
907 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
910 /* Macros for outputting the compiled pattern into `buffer'. */
912 /* If the buffer isn't allocated when it comes in, use this. */
913 #define INIT_BUF_SIZE 32
915 /* Make sure we have at least N more bytes of space in buffer. */
916 #define GET_BUFFER_SPACE(n) \
917 while (b - bufp->buffer + (n) > bufp->allocated) \
920 /* Make sure we have one more byte of buffer space and then add C to it. */
921 #define BUF_PUSH(c) \
923 GET_BUFFER_SPACE (1); \
924 *b++ = (unsigned char) (c); \
928 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
929 #define BUF_PUSH_2(c1, c2) \
931 GET_BUFFER_SPACE (2); \
932 *b++ = (unsigned char) (c1); \
933 *b++ = (unsigned char) (c2); \
937 /* As with BUF_PUSH_2, except for three bytes. */
938 #define BUF_PUSH_3(c1, c2, c3) \
940 GET_BUFFER_SPACE (3); \
941 *b++ = (unsigned char) (c1); \
942 *b++ = (unsigned char) (c2); \
943 *b++ = (unsigned char) (c3); \
947 /* Store a jump with opcode OP at LOC to location TO. We store a
948 relative address offset by the three bytes the jump itself occupies. */
949 #define STORE_JUMP(op, loc, to) \
950 store_op1 (op, loc, (to) - (loc) - 3)
952 /* Likewise, for a two-argument jump. */
953 #define STORE_JUMP2(op, loc, to, arg) \
954 store_op2 (op, loc, (to) - (loc) - 3, arg)
956 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
957 #define INSERT_JUMP(op, loc, to) \
958 insert_op1 (op, loc, (to) - (loc) - 3, b)
960 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
961 #define INSERT_JUMP2(op, loc, to, arg) \
962 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
965 /* This is not an arbitrary limit: the arguments which represent offsets
966 into the pattern are two bytes long. So if 2^16 bytes turns out to
967 be too small, many things would have to change. */
968 #define MAX_BUF_SIZE (1L << 16)
971 /* Extend the buffer by twice its current size via realloc and
972 reset the pointers that pointed into the old block to point to the
973 correct places in the new one. If extending the buffer results in it
974 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
975 #define EXTEND_BUFFER() \
977 unsigned char *old_buffer = bufp->buffer; \
978 if (bufp->allocated == MAX_BUF_SIZE) \
980 bufp->allocated <<= 1; \
981 if (bufp->allocated > MAX_BUF_SIZE) \
982 bufp->allocated = MAX_BUF_SIZE; \
983 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
984 if (bufp->buffer == NULL) \
986 /* If the buffer moved, move all the pointers into it. */ \
987 if (old_buffer != bufp->buffer) \
989 b = (b - old_buffer) + bufp->buffer; \
990 begalt = (begalt - old_buffer) + bufp->buffer; \
991 if (fixup_alt_jump) \
992 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
994 laststart = (laststart - old_buffer) + bufp->buffer; \
996 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1001 /* Since we have one byte reserved for the register number argument to
1002 {start,stop}_memory, the maximum number of groups we can report
1003 things about is what fits in that byte. */
1004 #define MAX_REGNUM 255
1006 /* But patterns can have more than `MAX_REGNUM' registers. We just
1007 ignore the excess. */
1008 typedef unsigned regnum_t;
1011 /* Macros for the compile stack. */
1013 /* Since offsets can go either forwards or backwards, this type needs to
1014 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1015 typedef int pattern_offset_t;
1019 pattern_offset_t begalt_offset;
1020 pattern_offset_t fixup_alt_jump;
1021 pattern_offset_t inner_group_offset;
1022 pattern_offset_t laststart_offset;
1024 } compile_stack_elt_t;
1029 compile_stack_elt_t *stack;
1031 unsigned avail; /* Offset of next open position. */
1032 } compile_stack_type;
1035 #define INIT_COMPILE_STACK_SIZE 32
1037 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1038 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1040 /* The next available element. */
1041 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1044 /* Set the bit for character C in a list. */
1045 #define SET_LIST_BIT(c) \
1046 (b[((unsigned char) (c)) / BYTEWIDTH] \
1047 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1050 /* Get the next unsigned number in the uncompiled pattern. */
1051 #define GET_UNSIGNED_NUMBER(num) \
1055 while (ISDIGIT (c)) \
1059 num = num * 10 + c - '0'; \
1067 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1069 #define IS_CHAR_CLASS(string) \
1070 (STREQ (string, "alpha") || STREQ (string, "upper") \
1071 || STREQ (string, "lower") || STREQ (string, "digit") \
1072 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1073 || STREQ (string, "space") || STREQ (string, "print") \
1074 || STREQ (string, "punct") || STREQ (string, "graph") \
1075 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1077 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1078 Returns one of error codes defined in `regex.h', or zero for success.
1080 Assumes the `allocated' (and perhaps `buffer') and `translate'
1081 fields are set in BUFP on entry.
1083 If it succeeds, results are put in BUFP (if it returns an error, the
1084 contents of BUFP are undefined):
1085 `buffer' is the compiled pattern;
1086 `syntax' is set to SYNTAX;
1087 `used' is set to the length of the compiled pattern;
1088 `fastmap_accurate' is zero;
1089 `re_nsub' is the number of subexpressions in PATTERN;
1090 `not_bol' and `not_eol' are zero;
1092 The `fastmap' and `newline_anchor' fields are neither
1093 examined nor set. */
1095 static reg_errcode_t
1096 regex_compile (pattern, size, syntax, bufp)
1097 const char *pattern;
1099 reg_syntax_t syntax;
1100 struct re_pattern_buffer *bufp;
1102 /* We fetch characters from PATTERN here. Even though PATTERN is
1103 `char *' (i.e., signed), we declare these variables as unsigned, so
1104 they can be reliably used as array indices. */
1105 register unsigned char c, c1;
1107 /* A random tempory spot in PATTERN. */
1110 /* Points to the end of the buffer, where we should append. */
1111 register unsigned char *b;
1113 /* Keeps track of unclosed groups. */
1114 compile_stack_type compile_stack;
1116 /* Points to the current (ending) position in the pattern. */
1117 const char *p = pattern;
1118 const char *pend = pattern + size;
1120 /* How to translate the characters in the pattern. */
1121 char *translate = bufp->translate;
1123 /* Address of the count-byte of the most recently inserted `exactn'
1124 command. This makes it possible to tell if a new exact-match
1125 character can be added to that command or if the character requires
1126 a new `exactn' command. */
1127 unsigned char *pending_exact = 0;
1129 /* Address of start of the most recently finished expression.
1130 This tells, e.g., postfix * where to find the start of its
1131 operand. Reset at the beginning of groups and alternatives. */
1132 unsigned char *laststart = 0;
1134 /* Address of beginning of regexp, or inside of last group. */
1135 unsigned char *begalt;
1137 /* Place in the uncompiled pattern (i.e., the {) to
1138 which to go back if the interval is invalid. */
1139 const char *beg_interval;
1141 /* Address of the place where a forward jump should go to the end of
1142 the containing expression. Each alternative of an `or' -- except the
1143 last -- ends with a forward jump of this sort. */
1144 unsigned char *fixup_alt_jump = 0;
1146 /* Counts open-groups as they are encountered. Remembered for the
1147 matching close-group on the compile stack, so the same register
1148 number is put in the stop_memory as the start_memory. */
1149 regnum_t regnum = 0;
1152 DEBUG_PRINT1 ("\nCompiling pattern: ");
1155 unsigned debug_count;
1157 for (debug_count = 0; debug_count < size; debug_count++)
1158 printchar (pattern[debug_count]);
1163 /* Initialize the compile stack. */
1164 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1165 if (compile_stack.stack == NULL)
1168 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1169 compile_stack.avail = 0;
1171 /* Initialize the pattern buffer. */
1172 bufp->syntax = syntax;
1173 bufp->fastmap_accurate = 0;
1174 bufp->not_bol = bufp->not_eol = 0;
1176 /* Set `used' to zero, so that if we return an error, the pattern
1177 printer (for debugging) will think there's no pattern. We reset it
1181 /* Always count groups, whether or not bufp->no_sub is set. */
1184 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1185 /* Initialize the syntax table. */
1186 init_syntax_once ();
1189 if (bufp->allocated == 0)
1192 { /* If zero allocated, but buffer is non-null, try to realloc
1193 enough space. This loses if buffer's address is bogus, but
1194 that is the user's responsibility. */
1195 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1198 { /* Caller did not allocate a buffer. Do it for them. */
1199 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1201 if (!bufp->buffer) return REG_ESPACE;
1203 bufp->allocated = INIT_BUF_SIZE;
1206 begalt = b = bufp->buffer;
1208 /* Loop through the uncompiled pattern until we're at the end. */
1217 if ( /* If at start of pattern, it's an operator. */
1219 /* If context independent, it's an operator. */
1220 || syntax & RE_CONTEXT_INDEP_ANCHORS
1221 /* Otherwise, depends on what's come before. */
1222 || at_begline_loc_p (pattern, p, syntax))
1232 if ( /* If at end of pattern, it's an operator. */
1234 /* If context independent, it's an operator. */
1235 || syntax & RE_CONTEXT_INDEP_ANCHORS
1236 /* Otherwise, depends on what's next. */
1237 || at_endline_loc_p (p, pend, syntax))
1247 if ((syntax & RE_BK_PLUS_QM)
1248 || (syntax & RE_LIMITED_OPS))
1252 /* If there is no previous pattern... */
1255 if (syntax & RE_CONTEXT_INVALID_OPS)
1257 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1262 /* Are we optimizing this jump? */
1263 boolean keep_string_p = false;
1265 /* 1 means zero (many) matches is allowed. */
1266 char zero_times_ok = 0, many_times_ok = 0;
1268 /* If there is a sequence of repetition chars, collapse it
1269 down to just one (the right one). We can't combine
1270 interval operators with these because of, e.g., `a{2}*',
1271 which should only match an even number of `a's. */
1275 zero_times_ok |= c != '+';
1276 many_times_ok |= c != '?';
1284 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1287 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1289 if (p == pend) return REG_EESCAPE;
1292 if (!(c1 == '+' || c1 == '?'))
1307 /* If we get here, we found another repeat character. */
1310 /* Star, etc. applied to an empty pattern is equivalent
1311 to an empty pattern. */
1315 /* Now we know whether or not zero matches is allowed
1316 and also whether or not two or more matches is allowed. */
1318 { /* More than one repetition is allowed, so put in at the
1319 end a backward relative jump from `b' to before the next
1320 jump we're going to put in below (which jumps from
1321 laststart to after this jump).
1323 But if we are at the `*' in the exact sequence `.*\n',
1324 insert an unconditional jump backwards to the .,
1325 instead of the beginning of the loop. This way we only
1326 push a failure point once, instead of every time
1327 through the loop. */
1328 assert (p - 1 > pattern);
1330 /* Allocate the space for the jump. */
1331 GET_BUFFER_SPACE (3);
1333 /* We know we are not at the first character of the pattern,
1334 because laststart was nonzero. And we've already
1335 incremented `p', by the way, to be the character after
1336 the `*'. Do we have to do something analogous here
1337 for null bytes, because of RE_DOT_NOT_NULL? */
1338 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1340 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1341 && !(syntax & RE_DOT_NEWLINE))
1342 { /* We have .*\n. */
1343 STORE_JUMP (jump, b, laststart);
1344 keep_string_p = true;
1347 /* Anything else. */
1348 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1350 /* We've added more stuff to the buffer. */
1354 /* On failure, jump from laststart to b + 3, which will be the
1355 end of the buffer after this jump is inserted. */
1356 GET_BUFFER_SPACE (3);
1357 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1365 /* At least one repetition is required, so insert a
1366 `dummy_failure_jump' before the initial
1367 `on_failure_jump' instruction of the loop. This
1368 effects a skip over that instruction the first time
1369 we hit that loop. */
1370 GET_BUFFER_SPACE (3);
1371 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1386 boolean had_char_class = false;
1388 if (p == pend) return REG_EBRACK;
1390 /* Ensure that we have enough space to push a charset: the
1391 opcode, the length count, and the bitset; 34 bytes in all. */
1392 GET_BUFFER_SPACE (34);
1396 /* We test `*p == '^' twice, instead of using an if
1397 statement, so we only need one BUF_PUSH. */
1398 BUF_PUSH (*p == '^' ? charset_not : charset);
1402 /* Remember the first position in the bracket expression. */
1405 /* Push the number of bytes in the bitmap. */
1406 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1408 /* Clear the whole map. */
1409 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1411 /* charset_not matches newline according to a syntax bit. */
1412 if ((re_opcode_t) b[-2] == charset_not
1413 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1414 SET_LIST_BIT ('\n');
1416 /* Read in characters and ranges, setting map bits. */
1419 if (p == pend) return REG_EBRACK;
1423 /* \ might escape characters inside [...] and [^...]. */
1424 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1426 if (p == pend) return REG_EESCAPE;
1433 /* Could be the end of the bracket expression. If it's
1434 not (i.e., when the bracket expression is `[]' so
1435 far), the ']' character bit gets set way below. */
1436 if (c == ']' && p != p1 + 1)
1439 /* Look ahead to see if it's a range when the last thing
1440 was a character class. */
1441 if (had_char_class && c == '-' && *p != ']')
1444 /* Look ahead to see if it's a range when the last thing
1445 was a character: if this is a hyphen not at the
1446 beginning or the end of a list, then it's the range
1449 && !(p - 2 >= pattern && p[-2] == '[')
1450 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1454 = compile_range (&p, pend, translate, syntax, b);
1455 if (ret != REG_NOERROR) return ret;
1458 else if (p[0] == '-' && p[1] != ']')
1459 { /* This handles ranges made up of characters only. */
1462 /* Move past the `-'. */
1465 ret = compile_range (&p, pend, translate, syntax, b);
1466 if (ret != REG_NOERROR) return ret;
1469 /* See if we're at the beginning of a possible character
1472 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1473 { /* Leave room for the null. */
1474 char str[CHAR_CLASS_MAX_LENGTH + 1];
1479 /* If pattern is `[[:'. */
1480 if (p == pend) return REG_EBRACK;
1485 if (c == ':' || c == ']' || p == pend
1486 || c1 == CHAR_CLASS_MAX_LENGTH)
1492 /* If isn't a word bracketed by `[:' and:`]':
1493 undo the ending character, the letters, and leave
1494 the leading `:' and `[' (but set bits for them). */
1495 if (c == ':' && *p == ']')
1498 boolean is_alnum = STREQ (str, "alnum");
1499 boolean is_alpha = STREQ (str, "alpha");
1500 boolean is_blank = STREQ (str, "blank");
1501 boolean is_cntrl = STREQ (str, "cntrl");
1502 boolean is_digit = STREQ (str, "digit");
1503 boolean is_graph = STREQ (str, "graph");
1504 boolean is_lower = STREQ (str, "lower");
1505 boolean is_print = STREQ (str, "print");
1506 boolean is_punct = STREQ (str, "punct");
1507 boolean is_space = STREQ (str, "space");
1508 boolean is_upper = STREQ (str, "upper");
1509 boolean is_xdigit = STREQ (str, "xdigit");
1511 if (!IS_CHAR_CLASS (str)) return REG_ECTYPE;
1513 /* Throw away the ] at the end of the character
1517 if (p == pend) return REG_EBRACK;
1519 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
1521 if ( (is_alnum && ISALNUM (ch))
1522 || (is_alpha && ISALPHA (ch))
1523 || (is_blank && ISBLANK (ch))
1524 || (is_cntrl && ISCNTRL (ch))
1525 || (is_digit && ISDIGIT (ch))
1526 || (is_graph && ISGRAPH (ch))
1527 || (is_lower && ISLOWER (ch))
1528 || (is_print && ISPRINT (ch))
1529 || (is_punct && ISPUNCT (ch))
1530 || (is_space && ISSPACE (ch))
1531 || (is_upper && ISUPPER (ch))
1532 || (is_xdigit && ISXDIGIT (ch)))
1535 had_char_class = true;
1544 had_char_class = false;
1549 had_char_class = false;
1554 /* Discard any (non)matching list bytes that are all 0 at the
1555 end of the map. Decrease the map-length byte too. */
1556 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
1564 if (syntax & RE_NO_BK_PARENS)
1571 if (syntax & RE_NO_BK_PARENS)
1578 if (syntax & RE_NEWLINE_ALT)
1585 if (syntax & RE_NO_BK_VBAR)
1592 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
1593 goto handle_interval;
1599 if (p == pend) return REG_EESCAPE;
1601 /* Do not translate the character after the \, so that we can
1602 distinguish, e.g., \B from \b, even if we normally would
1603 translate, e.g., B to b. */
1609 if (syntax & RE_NO_BK_PARENS)
1610 goto normal_backslash;
1616 if (COMPILE_STACK_FULL)
1618 RETALLOC (compile_stack.stack, compile_stack.size << 1,
1619 compile_stack_elt_t);
1620 if (compile_stack.stack == NULL) return REG_ESPACE;
1622 compile_stack.size <<= 1;
1625 /* These are the values to restore when we hit end of this
1626 group. They are all relative offsets, so that if the
1627 whole pattern moves because of realloc, they will still
1629 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
1630 COMPILE_STACK_TOP.fixup_alt_jump
1631 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
1632 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
1633 COMPILE_STACK_TOP.regnum = regnum;
1635 /* We will eventually replace the 0 with the number of
1636 groups inner to this one. But do not push a
1637 start_memory for groups beyond the last one we can
1638 represent in the compiled pattern. */
1639 if (regnum <= MAX_REGNUM)
1641 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
1642 BUF_PUSH_3 (start_memory, regnum, 0);
1645 compile_stack.avail++;
1650 /* If we've reached MAX_REGNUM groups, then this open
1651 won't actually generate any code, so we'll have to
1652 clear pending_exact explicitly. */
1658 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
1660 if (COMPILE_STACK_EMPTY)
1661 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
1662 goto normal_backslash;
1668 { /* Push a dummy failure point at the end of the
1669 alternative for a possible future
1670 `pop_failure_jump' to pop. See comments at
1671 `push_dummy_failure' in `re_match_2'. */
1672 BUF_PUSH (push_dummy_failure);
1674 /* We allocated space for this jump when we assigned
1675 to `fixup_alt_jump', in the `handle_alt' case below. */
1676 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
1679 /* See similar code for backslashed left paren above. */
1680 if (COMPILE_STACK_EMPTY)
1681 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
1686 /* Since we just checked for an empty stack above, this
1687 ``can't happen''. */
1688 assert (compile_stack.avail != 0);
1690 /* We don't just want to restore into `regnum', because
1691 later groups should continue to be numbered higher,
1692 as in `(ab)c(de)' -- the second group is #2. */
1693 regnum_t this_group_regnum;
1695 compile_stack.avail--;
1696 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
1698 = COMPILE_STACK_TOP.fixup_alt_jump
1699 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
1701 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
1702 this_group_regnum = COMPILE_STACK_TOP.regnum;
1703 /* If we've reached MAX_REGNUM groups, then this open
1704 won't actually generate any code, so we'll have to
1705 clear pending_exact explicitly. */
1708 /* We're at the end of the group, so now we know how many
1709 groups were inside this one. */
1710 if (this_group_regnum <= MAX_REGNUM)
1712 unsigned char *inner_group_loc
1713 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
1715 *inner_group_loc = regnum - this_group_regnum;
1716 BUF_PUSH_3 (stop_memory, this_group_regnum,
1717 regnum - this_group_regnum);
1723 case '|': /* `\|'. */
1724 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
1725 goto normal_backslash;
1727 if (syntax & RE_LIMITED_OPS)
1730 /* Insert before the previous alternative a jump which
1731 jumps to this alternative if the former fails. */
1732 GET_BUFFER_SPACE (3);
1733 INSERT_JUMP (on_failure_jump, begalt, b + 6);
1737 /* The alternative before this one has a jump after it
1738 which gets executed if it gets matched. Adjust that
1739 jump so it will jump to this alternative's analogous
1740 jump (put in below, which in turn will jump to the next
1741 (if any) alternative's such jump, etc.). The last such
1742 jump jumps to the correct final destination. A picture:
1748 If we are at `b', then fixup_alt_jump right now points to a
1749 three-byte space after `a'. We'll put in the jump, set
1750 fixup_alt_jump to right after `b', and leave behind three
1751 bytes which we'll fill in when we get to after `c'. */
1754 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
1756 /* Mark and leave space for a jump after this alternative,
1757 to be filled in later either by next alternative or
1758 when know we're at the end of a series of alternatives. */
1760 GET_BUFFER_SPACE (3);
1769 /* If \{ is a literal. */
1770 if (!(syntax & RE_INTERVALS)
1771 /* If we're at `\{' and it's not the open-interval
1773 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
1774 || (p - 2 == pattern && p == pend))
1775 goto normal_backslash;
1779 /* If got here, then the syntax allows intervals. */
1781 /* At least (most) this many matches must be made. */
1782 int lower_bound = -1, upper_bound = -1;
1784 beg_interval = p - 1;
1788 if (syntax & RE_NO_BK_BRACES)
1789 goto unfetch_interval;
1794 GET_UNSIGNED_NUMBER (lower_bound);
1798 GET_UNSIGNED_NUMBER (upper_bound);
1799 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
1802 /* Interval such as `{1}' => match exactly once. */
1803 upper_bound = lower_bound;
1805 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
1806 || lower_bound > upper_bound)
1808 if (syntax & RE_NO_BK_BRACES)
1809 goto unfetch_interval;
1814 if (!(syntax & RE_NO_BK_BRACES))
1816 if (c != '\\') return REG_EBRACE;
1823 if (syntax & RE_NO_BK_BRACES)
1824 goto unfetch_interval;
1829 /* We just parsed a valid interval. */
1831 /* If it's invalid to have no preceding re. */
1834 if (syntax & RE_CONTEXT_INVALID_OPS)
1836 else if (syntax & RE_CONTEXT_INDEP_OPS)
1839 goto unfetch_interval;
1842 /* If the upper bound is zero, don't want to succeed at
1843 all; jump from `laststart' to `b + 3', which will be
1844 the end of the buffer after we insert the jump. */
1845 if (upper_bound == 0)
1847 GET_BUFFER_SPACE (3);
1848 INSERT_JUMP (jump, laststart, b + 3);
1852 /* Otherwise, we have a nontrivial interval. When
1853 we're all done, the pattern will look like:
1854 set_number_at <jump count> <upper bound>
1855 set_number_at <succeed_n count> <lower bound>
1856 succeed_n <after jump addr> <succed_n count>
1858 jump_n <succeed_n addr> <jump count>
1859 (The upper bound and `jump_n' are omitted if
1860 `upper_bound' is 1, though.) */
1862 { /* If the upper bound is > 1, we need to insert
1863 more at the end of the loop. */
1864 unsigned nbytes = 10 + (upper_bound > 1) * 10;
1866 GET_BUFFER_SPACE (nbytes);
1868 /* Initialize lower bound of the `succeed_n', even
1869 though it will be set during matching by its
1870 attendant `set_number_at' (inserted next),
1871 because `re_compile_fastmap' needs to know.
1872 Jump to the `jump_n' we might insert below. */
1873 INSERT_JUMP2 (succeed_n, laststart,
1874 b + 5 + (upper_bound > 1) * 5,
1878 /* Code to initialize the lower bound. Insert
1879 before the `succeed_n'. The `5' is the last two
1880 bytes of this `set_number_at', plus 3 bytes of
1881 the following `succeed_n'. */
1882 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
1885 if (upper_bound > 1)
1886 { /* More than one repetition is allowed, so
1887 append a backward jump to the `succeed_n'
1888 that starts this interval.
1890 When we've reached this during matching,
1891 we'll have matched the interval once, so
1892 jump back only `upper_bound - 1' times. */
1893 STORE_JUMP2 (jump_n, b, laststart + 5,
1897 /* The location we want to set is the second
1898 parameter of the `jump_n'; that is `b-2' as
1899 an absolute address. `laststart' will be
1900 the `set_number_at' we're about to insert;
1901 `laststart+3' the number to set, the source
1902 for the relative address. But we are
1903 inserting into the middle of the pattern --
1904 so everything is getting moved up by 5.
1905 Conclusion: (b - 2) - (laststart + 3) + 5,
1906 i.e., b - laststart.
1908 We insert this at the beginning of the loop
1909 so that if we fail during matching, we'll
1910 reinitialize the bounds. */
1911 insert_op2 (set_number_at, laststart, b - laststart,
1912 upper_bound - 1, b);
1917 beg_interval = NULL;
1922 /* If an invalid interval, match the characters as literals. */
1923 assert (beg_interval);
1925 beg_interval = NULL;
1927 /* normal_char and normal_backslash need `c'. */
1930 if (!(syntax & RE_NO_BK_BRACES))
1932 if (p > pattern && p[-1] == '\\')
1933 goto normal_backslash;
1938 /* There is no way to specify the before_dot and after_dot
1939 operators. rms says this is ok. --karl */
1947 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
1953 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
1960 BUF_PUSH (wordchar);
1966 BUF_PUSH (notwordchar);
1979 BUF_PUSH (wordbound);
1983 BUF_PUSH (notwordbound);
1994 case '1': case '2': case '3': case '4': case '5':
1995 case '6': case '7': case '8': case '9':
1996 if (syntax & RE_NO_BK_REFS)
2004 /* Can't back reference to a subexpression if inside of it. */
2005 if (group_in_compile_stack (compile_stack, c1))
2009 BUF_PUSH_2 (duplicate, c1);
2015 if (syntax & RE_BK_PLUS_QM)
2018 goto normal_backslash;
2022 /* You might think it would be useful for \ to mean
2023 not to translate; but if we don't translate it
2024 it will never match anything. */
2032 /* Expects the character in `c'. */
2034 /* If no exactn currently being built. */
2037 /* If last exactn not at current position. */
2038 || pending_exact + *pending_exact + 1 != b
2040 /* We have only one byte following the exactn for the count. */
2041 || *pending_exact == (1 << BYTEWIDTH) - 1
2043 /* If followed by a repetition operator. */
2044 || *p == '*' || *p == '^'
2045 || ((syntax & RE_BK_PLUS_QM)
2046 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2047 : (*p == '+' || *p == '?'))
2048 || ((syntax & RE_INTERVALS)
2049 && ((syntax & RE_NO_BK_BRACES)
2051 : (p[0] == '\\' && p[1] == '{'))))
2053 /* Start building a new exactn. */
2057 BUF_PUSH_2 (exactn, 0);
2058 pending_exact = b - 1;
2065 } /* while p != pend */
2068 /* Through the pattern now. */
2071 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2073 if (!COMPILE_STACK_EMPTY)
2076 free (compile_stack.stack);
2078 /* We have succeeded; set the length of the buffer. */
2079 bufp->used = b - bufp->buffer;
2084 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2085 print_compiled_pattern (bufp);
2090 } /* regex_compile */
2092 /* Subroutines for `regex_compile'. */
2094 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2097 store_op1 (op, loc, arg)
2102 *loc = (unsigned char) op;
2103 STORE_NUMBER (loc + 1, arg);
2107 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2110 store_op2 (op, loc, arg1, arg2)
2115 *loc = (unsigned char) op;
2116 STORE_NUMBER (loc + 1, arg1);
2117 STORE_NUMBER (loc + 3, arg2);
2121 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2122 for OP followed by two-byte integer parameter ARG. */
2125 insert_op1 (op, loc, arg, end)
2131 register unsigned char *pfrom = end;
2132 register unsigned char *pto = end + 3;
2134 while (pfrom != loc)
2137 store_op1 (op, loc, arg);
2141 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2144 insert_op2 (op, loc, arg1, arg2, end)
2150 register unsigned char *pfrom = end;
2151 register unsigned char *pto = end + 5;
2153 while (pfrom != loc)
2156 store_op2 (op, loc, arg1, arg2);
2160 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2161 after an alternative or a begin-subexpression. We assume there is at
2162 least one character before the ^. */
2165 at_begline_loc_p (pattern, p, syntax)
2166 const char *pattern, *p;
2167 reg_syntax_t syntax;
2169 const char *prev = p - 2;
2170 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2173 /* After a subexpression? */
2174 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2175 /* After an alternative? */
2176 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2180 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2181 at least one character after the $, i.e., `P < PEND'. */
2184 at_endline_loc_p (p, pend, syntax)
2185 const char *p, *pend;
2188 const char *next = p;
2189 boolean next_backslash = *next == '\\';
2190 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2193 /* Before a subexpression? */
2194 (syntax & RE_NO_BK_PARENS ? *next == ')'
2195 : next_backslash && next_next && *next_next == ')')
2196 /* Before an alternative? */
2197 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2198 : next_backslash && next_next && *next_next == '|');
2202 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2203 false if it's not. */
2206 group_in_compile_stack (compile_stack, regnum)
2207 compile_stack_type compile_stack;
2212 for (this_element = compile_stack.avail - 1;
2215 if (compile_stack.stack[this_element].regnum == regnum)
2222 /* Read the ending character of a range (in a bracket expression) from the
2223 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2224 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2225 Then we set the translation of all bits between the starting and
2226 ending characters (inclusive) in the compiled pattern B.
2228 Return an error code.
2230 We use these short variable names so we can use the same macros as
2231 `regex_compile' itself. */
2233 static reg_errcode_t
2234 compile_range (p_ptr, pend, translate, syntax, b)
2235 const char **p_ptr, *pend;
2237 reg_syntax_t syntax;
2242 const char *p = *p_ptr;
2243 int range_start, range_end;
2248 /* Even though the pattern is a signed `char *', we need to fetch
2249 with unsigned char *'s; if the high bit of the pattern character
2250 is set, the range endpoints will be negative if we fetch using a
2253 We also want to fetch the endpoints without translating them; the
2254 appropriate translation is done in the bit-setting loop below. */
2255 range_start = ((unsigned char *) p)[-2];
2256 range_end = ((unsigned char *) p)[0];
2258 /* Have to increment the pointer into the pattern string, so the
2259 caller isn't still at the ending character. */
2262 /* If the start is after the end, the range is empty. */
2263 if (range_start > range_end)
2264 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2266 /* Here we see why `this_char' has to be larger than an `unsigned
2267 char' -- the range is inclusive, so if `range_end' == 0xff
2268 (assuming 8-bit characters), we would otherwise go into an infinite
2269 loop, since all characters <= 0xff. */
2270 for (this_char = range_start; this_char <= range_end; this_char++)
2272 SET_LIST_BIT (TRANSLATE (this_char));
2278 /* Failure stack declarations and macros; both re_compile_fastmap and
2279 re_match_2 use a failure stack. These have to be macros because of
2283 /* Number of failure points for which to initially allocate space
2284 when matching. If this number is exceeded, we allocate more
2285 space, so it is not a hard limit. */
2286 #ifndef INIT_FAILURE_ALLOC
2287 #define INIT_FAILURE_ALLOC 5
2290 /* Roughly the maximum number of failure points on the stack. Would be
2291 exactly that if always used MAX_FAILURE_SPACE each time we failed.
2292 This is a variable only so users of regex can assign to it; we never
2293 change it ourselves. */
2294 int re_max_failures = 2000;
2296 typedef const unsigned char *fail_stack_elt_t;
2300 fail_stack_elt_t *stack;
2302 unsigned avail; /* Offset of next open position. */
2305 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
2306 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
2307 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
2308 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
2311 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
2313 #define INIT_FAIL_STACK() \
2315 fail_stack.stack = (fail_stack_elt_t *) \
2316 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
2318 if (fail_stack.stack == NULL) \
2321 fail_stack.size = INIT_FAILURE_ALLOC; \
2322 fail_stack.avail = 0; \
2326 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
2328 Return 1 if succeeds, and 0 if either ran out of memory
2329 allocating space for it or it was already too large.
2331 REGEX_REALLOCATE requires `destination' be declared. */
2333 #define DOUBLE_FAIL_STACK(fail_stack) \
2334 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
2336 : ((fail_stack).stack = (fail_stack_elt_t *) \
2337 REGEX_REALLOCATE ((fail_stack).stack, \
2338 (fail_stack).size * sizeof (fail_stack_elt_t), \
2339 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
2341 (fail_stack).stack == NULL \
2343 : ((fail_stack).size <<= 1, \
2347 /* Push PATTERN_OP on FAIL_STACK.
2349 Return 1 if was able to do so and 0 if ran out of memory allocating
2351 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
2352 ((FAIL_STACK_FULL () \
2353 && !DOUBLE_FAIL_STACK (fail_stack)) \
2355 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
2358 /* This pushes an item onto the failure stack. Must be a four-byte
2359 value. Assumes the variable `fail_stack'. Probably should only
2360 be called from within `PUSH_FAILURE_POINT'. */
2361 #define PUSH_FAILURE_ITEM(item) \
2362 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
2364 /* The complement operation. Assumes `fail_stack' is nonempty. */
2365 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
2367 /* Used to omit pushing failure point id's when we're not debugging. */
2369 #define DEBUG_PUSH PUSH_FAILURE_ITEM
2370 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
2372 #define DEBUG_PUSH(item)
2373 #define DEBUG_POP(item_addr)
2377 /* Push the information about the state we will need
2378 if we ever fail back to it.
2380 Requires variables fail_stack, regstart, regend, reg_info, and
2381 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
2384 Does `return FAILURE_CODE' if runs out of memory. */
2386 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
2388 char *destination; \
2389 /* Must be int, so when we don't save any registers, the arithmetic \
2390 of 0 + -1 isn't done as unsigned. */ \
2393 DEBUG_STATEMENT (failure_id++); \
2394 DEBUG_STATEMENT (nfailure_points_pushed++); \
2395 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
2396 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
2397 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
2399 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
2400 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
2402 /* Ensure we have enough space allocated for what we will push. */ \
2403 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
2405 if (!DOUBLE_FAIL_STACK (fail_stack)) \
2406 return failure_code; \
2408 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
2409 (fail_stack).size); \
2410 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
2413 /* Push the info, starting with the registers. */ \
2414 DEBUG_PRINT1 ("\n"); \
2416 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
2419 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
2420 DEBUG_STATEMENT (num_regs_pushed++); \
2422 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
2423 PUSH_FAILURE_ITEM (regstart[this_reg]); \
2425 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
2426 PUSH_FAILURE_ITEM (regend[this_reg]); \
2428 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
2429 DEBUG_PRINT2 (" match_null=%d", \
2430 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
2431 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
2432 DEBUG_PRINT2 (" matched_something=%d", \
2433 MATCHED_SOMETHING (reg_info[this_reg])); \
2434 DEBUG_PRINT2 (" ever_matched=%d", \
2435 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
2436 DEBUG_PRINT1 ("\n"); \
2437 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
2440 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
2441 PUSH_FAILURE_ITEM (lowest_active_reg); \
2443 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
2444 PUSH_FAILURE_ITEM (highest_active_reg); \
2446 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
2447 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
2448 PUSH_FAILURE_ITEM (pattern_place); \
2450 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
2451 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
2453 DEBUG_PRINT1 ("'\n"); \
2454 PUSH_FAILURE_ITEM (string_place); \
2456 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
2457 DEBUG_PUSH (failure_id); \
2460 /* This is the number of items that are pushed and popped on the stack
2461 for each register. */
2462 #define NUM_REG_ITEMS 3
2464 /* Individual items aside from the registers. */
2466 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
2468 #define NUM_NONREG_ITEMS 4
2471 /* We push at most this many items on the stack. */
2472 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
2474 /* We actually push this many items. */
2475 #define NUM_FAILURE_ITEMS \
2476 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
2479 /* How many items can still be added to the stack without overflowing it. */
2480 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
2483 /* Pops what PUSH_FAIL_STACK pushes.
2485 We restore into the parameters, all of which should be lvalues:
2486 STR -- the saved data position.
2487 PAT -- the saved pattern position.
2488 LOW_REG, HIGH_REG -- the highest and lowest active registers.
2489 REGSTART, REGEND -- arrays of string positions.
2490 REG_INFO -- array of information about each subexpression.
2492 Also assumes the variables `fail_stack' and (if debugging), `bufp',
2493 `pend', `string1', `size1', `string2', and `size2'. */
2495 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
2497 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
2499 const unsigned char *string_temp; \
2501 assert (!FAIL_STACK_EMPTY ()); \
2503 /* Remove failure points and point to how many regs pushed. */ \
2504 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
2505 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
2506 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
2508 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
2510 DEBUG_POP (&failure_id); \
2511 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
2513 /* If the saved string location is NULL, it came from an \
2514 on_failure_keep_string_jump opcode, and we want to throw away the \
2515 saved NULL, thus retaining our current position in the string. */ \
2516 string_temp = POP_FAILURE_ITEM (); \
2517 if (string_temp != NULL) \
2518 str = (const char *) string_temp; \
2520 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
2521 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
2522 DEBUG_PRINT1 ("'\n"); \
2524 pat = (unsigned char *) POP_FAILURE_ITEM (); \
2525 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
2526 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
2528 /* Restore register info. */ \
2529 high_reg = (unsigned) POP_FAILURE_ITEM (); \
2530 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
2532 low_reg = (unsigned) POP_FAILURE_ITEM (); \
2533 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
2535 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
2537 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
2539 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
2540 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
2542 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
2543 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
2545 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
2546 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
2549 DEBUG_STATEMENT (nfailure_points_popped++); \
2550 } /* POP_FAILURE_POINT */
2552 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2553 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2554 characters can start a string that matches the pattern. This fastmap
2555 is used by re_search to skip quickly over impossible starting points.
2557 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2558 area as BUFP->fastmap.
2560 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2563 Returns 0 if we succeed, -2 if an internal error. */
2566 re_compile_fastmap (bufp)
2567 struct re_pattern_buffer *bufp;
2570 fail_stack_type fail_stack;
2571 #ifndef REGEX_MALLOC
2574 /* We don't push any register information onto the failure stack. */
2575 unsigned num_regs = 0;
2577 register char *fastmap = bufp->fastmap;
2578 unsigned char *pattern = bufp->buffer;
2579 unsigned long size = bufp->used;
2580 const unsigned char *p = pattern;
2581 register unsigned char *pend = pattern + size;
2583 /* Assume that each path through the pattern can be null until
2584 proven otherwise. We set this false at the bottom of switch
2585 statement, to which we get only if a particular path doesn't
2586 match the empty string. */
2587 boolean path_can_be_null = true;
2589 /* We aren't doing a `succeed_n' to begin with. */
2590 boolean succeed_n_p = false;
2592 assert (fastmap != NULL && p != NULL);
2595 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2596 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2597 bufp->can_be_null = 0;
2599 while (p != pend || !FAIL_STACK_EMPTY ())
2603 bufp->can_be_null |= path_can_be_null;
2605 /* Reset for next path. */
2606 path_can_be_null = true;
2608 p = fail_stack.stack[--fail_stack.avail];
2611 /* We should never be about to go beyond the end of the pattern. */
2614 #ifdef SWITCH_ENUM_BUG
2615 switch ((int) ((re_opcode_t) *p++))
2617 switch ((re_opcode_t) *p++)
2621 /* I guess the idea here is to simply not bother with a fastmap
2622 if a backreference is used, since it's too hard to figure out
2623 the fastmap for the corresponding group. Setting
2624 `can_be_null' stops `re_search_2' from using the fastmap, so
2625 that is all we do. */
2627 bufp->can_be_null = 1;
2631 /* Following are the cases which match a character. These end
2640 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2641 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2647 /* Chars beyond end of map must be allowed. */
2648 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2651 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2652 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2658 for (j = 0; j < (1 << BYTEWIDTH); j++)
2659 if (SYNTAX (j) == Sword)
2665 for (j = 0; j < (1 << BYTEWIDTH); j++)
2666 if (SYNTAX (j) != Sword)
2672 /* `.' matches anything ... */
2673 for (j = 0; j < (1 << BYTEWIDTH); j++)
2676 /* ... except perhaps newline. */
2677 if (!(bufp->syntax & RE_DOT_NEWLINE))
2680 /* Return if we have already set `can_be_null'; if we have,
2681 then the fastmap is irrelevant. Something's wrong here. */
2682 else if (bufp->can_be_null)
2685 /* Otherwise, have to check alternative paths. */
2692 for (j = 0; j < (1 << BYTEWIDTH); j++)
2693 if (SYNTAX (j) == (enum syntaxcode) k)
2700 for (j = 0; j < (1 << BYTEWIDTH); j++)
2701 if (SYNTAX (j) != (enum syntaxcode) k)
2706 /* All cases after this match the empty string. These end with
2714 #endif /* not emacs */
2726 case push_dummy_failure:
2731 case pop_failure_jump:
2732 case maybe_pop_jump:
2735 case dummy_failure_jump:
2736 EXTRACT_NUMBER_AND_INCR (j, p);
2741 /* Jump backward implies we just went through the body of a
2742 loop and matched nothing. Opcode jumped to should be
2743 `on_failure_jump' or `succeed_n'. Just treat it like an
2744 ordinary jump. For a * loop, it has pushed its failure
2745 point already; if so, discard that as redundant. */
2746 if ((re_opcode_t) *p != on_failure_jump
2747 && (re_opcode_t) *p != succeed_n)
2751 EXTRACT_NUMBER_AND_INCR (j, p);
2754 /* If what's on the stack is where we are now, pop it. */
2755 if (!FAIL_STACK_EMPTY ()
2756 && fail_stack.stack[fail_stack.avail - 1] == p)
2762 case on_failure_jump:
2763 case on_failure_keep_string_jump:
2764 handle_on_failure_jump:
2765 EXTRACT_NUMBER_AND_INCR (j, p);
2767 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2768 end of the pattern. We don't want to push such a point,
2769 since when we restore it above, entering the switch will
2770 increment `p' past the end of the pattern. We don't need
2771 to push such a point since we obviously won't find any more
2772 fastmap entries beyond `pend'. Such a pattern can match
2773 the null string, though. */
2776 if (!PUSH_PATTERN_OP (p + j, fail_stack))
2780 bufp->can_be_null = 1;
2784 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
2785 succeed_n_p = false;
2792 /* Get to the number of times to succeed. */
2795 /* Increment p past the n for when k != 0. */
2796 EXTRACT_NUMBER_AND_INCR (k, p);
2800 succeed_n_p = true; /* Spaghetti code alert. */
2801 goto handle_on_failure_jump;
2818 abort (); /* We have listed all the cases. */
2821 /* Getting here means we have found the possible starting
2822 characters for one path of the pattern -- and that the empty
2823 string does not match. We need not follow this path further.
2824 Instead, look at the next alternative (remembered on the
2825 stack), or quit if no more. The test at the top of the loop
2826 does these things. */
2827 path_can_be_null = false;
2831 /* Set `can_be_null' for the last path (also the first path, if the
2832 pattern is empty). */
2833 bufp->can_be_null |= path_can_be_null;
2835 } /* re_compile_fastmap */
2837 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
2838 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
2839 this memory for recording register information. STARTS and ENDS
2840 must be allocated using the malloc library routine, and must each
2841 be at least NUM_REGS * sizeof (regoff_t) bytes long.
2843 If NUM_REGS == 0, then subsequent matches should allocate their own
2846 Unless this function is called, the first search or match using
2847 PATTERN_BUFFER will allocate its own register data, without
2848 freeing the old data. */
2851 re_set_registers (bufp, regs, num_regs, starts, ends)
2852 struct re_pattern_buffer *bufp;
2853 struct re_registers *regs;
2855 regoff_t *starts, *ends;
2859 bufp->regs_allocated = REGS_REALLOCATE;
2860 regs->num_regs = num_regs;
2861 regs->start = starts;
2866 bufp->regs_allocated = REGS_UNALLOCATED;
2868 regs->start = regs->end = (regoff_t) 0;
2872 /* Searching routines. */
2874 /* Like re_search_2, below, but only one string is specified, and
2875 doesn't let you say where to stop matching. */
2878 re_search (bufp, string, size, startpos, range, regs)
2879 struct re_pattern_buffer *bufp;
2881 int size, startpos, range;
2882 struct re_registers *regs;
2884 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
2889 /* Using the compiled pattern in BUFP->buffer, first tries to match the
2890 virtual concatenation of STRING1 and STRING2, starting first at index
2891 STARTPOS, then at STARTPOS + 1, and so on.
2893 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
2895 RANGE is how far to scan while trying to match. RANGE = 0 means try
2896 only at STARTPOS; in general, the last start tried is STARTPOS +
2899 In REGS, return the indices of the virtual concatenation of STRING1
2900 and STRING2 that matched the entire BUFP->buffer and its contained
2903 Do not consider matching one past the index STOP in the virtual
2904 concatenation of STRING1 and STRING2.
2906 We return either the position in the strings at which the match was
2907 found, -1 if no match, or -2 if error (such as failure
2911 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
2912 struct re_pattern_buffer *bufp;
2913 const char *string1, *string2;
2917 struct re_registers *regs;
2921 register char *fastmap = bufp->fastmap;
2922 register char *translate = bufp->translate;
2923 int total_size = size1 + size2;
2924 int endpos = startpos + range;
2926 /* Check for out-of-range STARTPOS. */
2927 if (startpos < 0 || startpos > total_size)
2930 /* Fix up RANGE if it might eventually take us outside
2931 the virtual concatenation of STRING1 and STRING2. */
2933 range = -1 - startpos;
2934 else if (endpos > total_size)
2935 range = total_size - startpos;
2937 /* If the search isn't to be a backwards one, don't waste time in a
2938 search for a pattern that must be anchored. */
2939 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
2947 /* Update the fastmap now if not correct already. */
2948 if (fastmap && !bufp->fastmap_accurate)
2949 if (re_compile_fastmap (bufp) == -2)
2952 /* Loop through the string, looking for a place to start matching. */
2955 /* If a fastmap is supplied, skip quickly over characters that
2956 cannot be the start of a match. If the pattern can match the
2957 null string, however, we don't need to skip characters; we want
2958 the first null string. */
2959 if (fastmap && startpos < total_size && !bufp->can_be_null)
2961 if (range > 0) /* Searching forwards. */
2963 register const char *d;
2964 register int lim = 0;
2967 if (startpos < size1 && startpos + range >= size1)
2968 lim = range - (size1 - startpos);
2970 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
2972 /* Written out as an if-else to avoid testing `translate'
2976 && !fastmap[(unsigned char)
2977 translate[(unsigned char) *d++]])
2980 while (range > lim && !fastmap[(unsigned char) *d++])
2983 startpos += irange - range;
2985 else /* Searching backwards. */
2987 register char c = (size1 == 0 || startpos >= size1
2988 ? string2[startpos - size1]
2989 : string1[startpos]);
2991 if (!fastmap[(unsigned char) TRANSLATE (c)])
2996 /* If can't match the null string, and that's all we have left, fail. */
2997 if (range >= 0 && startpos == total_size && fastmap
2998 && !bufp->can_be_null)
3001 val = re_match_2 (bufp, string1, size1, string2, size2,
3002 startpos, regs, stop);
3026 /* Declarations and macros for re_match_2. */
3028 static int bcmp_translate ();
3029 static boolean alt_match_null_string_p (),
3030 common_op_match_null_string_p (),
3031 group_match_null_string_p ();
3033 /* Structure for per-register (a.k.a. per-group) information.
3034 This must not be longer than one word, because we push this value
3035 onto the failure stack. Other register information, such as the
3036 starting and ending positions (which are addresses), and the list of
3037 inner groups (which is a bits list) are maintained in separate
3040 We are making a (strictly speaking) nonportable assumption here: that
3041 the compiler will pack our bit fields into something that fits into
3042 the type of `word', i.e., is something that fits into one item on the
3046 fail_stack_elt_t word;
3049 /* This field is one if this group can match the empty string,
3050 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
3051 #define MATCH_NULL_UNSET_VALUE 3
3052 unsigned match_null_string_p : 2;
3053 unsigned is_active : 1;
3054 unsigned matched_something : 1;
3055 unsigned ever_matched_something : 1;
3057 } register_info_type;
3059 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
3060 #define IS_ACTIVE(R) ((R).bits.is_active)
3061 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
3062 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
3065 /* Call this when have matched a real character; it sets `matched' flags
3066 for the subexpressions which we are currently inside. Also records
3067 that those subexprs have matched. */
3068 #define SET_REGS_MATCHED() \
3072 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
3074 MATCHED_SOMETHING (reg_info[r]) \
3075 = EVER_MATCHED_SOMETHING (reg_info[r]) \
3082 /* This converts PTR, a pointer into one of the search strings `string1'
3083 and `string2' into an offset from the beginning of that string. */
3084 #define POINTER_TO_OFFSET(ptr) \
3085 (FIRST_STRING_P (ptr) ? (ptr) - string1 : (ptr) - string2 + size1)
3087 /* Registers are set to a sentinel when they haven't yet matched. */
3088 #define REG_UNSET_VALUE ((char *) -1)
3089 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
3092 /* Macros for dealing with the split strings in re_match_2. */
3094 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3096 /* Call before fetching a character with *d. This switches over to
3097 string2 if necessary. */
3098 #define PREFETCH() \
3101 /* End of string2 => fail. */ \
3102 if (dend == end_match_2) \
3104 /* End of string1 => advance to string2. */ \
3106 dend = end_match_2; \
3110 /* Test if at very beginning or at very end of the virtual concatenation
3111 of `string1' and `string2'. If only one string, it's `string2'. */
3112 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3113 #define AT_STRINGS_END(d) ((d) == end2)
3116 /* Test if D points to a character which is word-constituent. We have
3117 two special cases to check for: if past the end of string1, look at
3118 the first character in string2; and if before the beginning of
3119 string2, look at the last character in string1. */
3120 #define WORDCHAR_P(d) \
3121 (SYNTAX ((d) == end1 ? *string2 \
3122 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3125 /* Test if the character before D and the one at D differ with respect
3126 to being word-constituent. */
3127 #define AT_WORD_BOUNDARY(d) \
3128 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3129 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3132 /* Free everything we malloc. */
3134 #define FREE_VAR(var) if (var) free (var); var = NULL
3135 #define FREE_VARIABLES() \
3137 FREE_VAR (fail_stack.stack); \
3138 FREE_VAR (regstart); \
3139 FREE_VAR (regend); \
3140 FREE_VAR (old_regstart); \
3141 FREE_VAR (old_regend); \
3142 FREE_VAR (best_regstart); \
3143 FREE_VAR (best_regend); \
3144 FREE_VAR (reg_info); \
3145 FREE_VAR (reg_dummy); \
3146 FREE_VAR (reg_info_dummy); \
3148 #else /* not REGEX_MALLOC */
3149 /* Some MIPS systems (at least) want this to free alloca'd storage. */
3150 #define FREE_VARIABLES() alloca (0)
3151 #endif /* not REGEX_MALLOC */
3154 /* These values must meet several constraints. They must not be valid
3155 register values; since we have a limit of 255 registers (because
3156 we use only one byte in the pattern for the register number), we can
3157 use numbers larger than 255. They must differ by 1, because of
3158 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3159 be larger than the value for the highest register, so we do not try
3160 to actually save any registers when none are active. */
3161 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3162 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3164 /* Matching routines. */
3166 #ifndef emacs /* Emacs never uses this. */
3167 /* re_match is like re_match_2 except it takes only a single string. */
3170 re_match (bufp, string, size, pos, regs)
3171 struct re_pattern_buffer *bufp;
3174 struct re_registers *regs;
3176 return re_match_2 (bufp, NULL, 0, string, size, pos, regs, size);
3178 #endif /* not emacs */
3181 /* re_match_2 matches the compiled pattern in BUFP against the
3182 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3183 and SIZE2, respectively). We start matching at POS, and stop
3186 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3187 store offsets for the substring each group matched in REGS. See the
3188 documentation for exactly how many groups we fill.
3190 We return -1 if no match, -2 if an internal error (such as the
3191 failure stack overflowing). Otherwise, we return the length of the
3192 matched substring. */
3195 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3196 struct re_pattern_buffer *bufp;
3197 const char *string1, *string2;
3200 struct re_registers *regs;
3203 /* General temporaries. */
3207 /* Just past the end of the corresponding string. */
3208 const char *end1, *end2;
3210 /* Pointers into string1 and string2, just past the last characters in
3211 each to consider matching. */
3212 const char *end_match_1, *end_match_2;
3214 /* Where we are in the data, and the end of the current string. */
3215 const char *d, *dend;
3217 /* Where we are in the pattern, and the end of the pattern. */
3218 unsigned char *p = bufp->buffer;
3219 register unsigned char *pend = p + bufp->used;
3221 /* We use this to map every character in the string. */
3222 char *translate = bufp->translate;
3224 /* Failure point stack. Each place that can handle a failure further
3225 down the line pushes a failure point on this stack. It consists of
3226 restart, regend, and reg_info for all registers corresponding to
3227 the subexpressions we're currently inside, plus the number of such
3228 registers, and, finally, two char *'s. The first char * is where
3229 to resume scanning the pattern; the second one is where to resume
3230 scanning the strings. If the latter is zero, the failure point is
3231 a ``dummy''; if a failure happens and the failure point is a dummy,
3232 it gets discarded and the next next one is tried. */
3233 fail_stack_type fail_stack;
3235 static unsigned failure_id = 0;
3236 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3239 /* We fill all the registers internally, independent of what we
3240 return, for use in backreferences. The number here includes
3241 an element for register zero. */
3242 unsigned num_regs = bufp->re_nsub + 1;
3244 /* The currently active registers. */
3245 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3246 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3248 /* Information on the contents of registers. These are pointers into
3249 the input strings; they record just what was matched (on this
3250 attempt) by a subexpression part of the pattern, that is, the
3251 regnum-th regstart pointer points to where in the pattern we began
3252 matching and the regnum-th regend points to right after where we
3253 stopped matching the regnum-th subexpression. (The zeroth register
3254 keeps track of what the whole pattern matches.) */
3255 const char **regstart, **regend;
3257 /* If a group that's operated upon by a repetition operator fails to
3258 match anything, then the register for its start will need to be
3259 restored because it will have been set to wherever in the string we
3260 are when we last see its open-group operator. Similarly for a
3262 const char **old_regstart, **old_regend;
3264 /* The is_active field of reg_info helps us keep track of which (possibly
3265 nested) subexpressions we are currently in. The matched_something
3266 field of reg_info[reg_num] helps us tell whether or not we have
3267 matched any of the pattern so far this time through the reg_num-th
3268 subexpression. These two fields get reset each time through any
3269 loop their register is in. */
3270 register_info_type *reg_info;
3272 /* The following record the register info as found in the above
3273 variables when we find a match better than any we've seen before.
3274 This happens as we backtrack through the failure points, which in
3275 turn happens only if we have not yet matched the entire string. */
3276 unsigned best_regs_set = false;
3277 const char **best_regstart, **best_regend;
3279 /* Logically, this is `best_regend[0]'. But we don't want to have to
3280 allocate space for that if we're not allocating space for anything
3281 else (see below). Also, we never need info about register 0 for
3282 any of the other register vectors, and it seems rather a kludge to
3283 treat `best_regend' differently than the rest. So we keep track of
3284 the end of the best match so far in a separate variable. We
3285 initialize this to NULL so that when we backtrack the first time
3286 and need to test it, it's not garbage. */
3287 const char *match_end = NULL;
3289 /* Used when we pop values we don't care about. */
3290 const char **reg_dummy;
3291 register_info_type *reg_info_dummy;
3294 /* Counts the total number of registers pushed. */
3295 unsigned num_regs_pushed = 0;
3298 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3302 /* Do not bother to initialize all the register variables if there are
3303 no groups in the pattern, as it takes a fair amount of time. If
3304 there are groups, we include space for register 0 (the whole
3305 pattern), even though we never use it, since it simplifies the
3306 array indexing. We should fix this. */
3309 regstart = REGEX_TALLOC (num_regs, const char *);
3310 regend = REGEX_TALLOC (num_regs, const char *);
3311 old_regstart = REGEX_TALLOC (num_regs, const char *);
3312 old_regend = REGEX_TALLOC (num_regs, const char *);
3313 best_regstart = REGEX_TALLOC (num_regs, const char *);
3314 best_regend = REGEX_TALLOC (num_regs, const char *);
3315 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3316 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3317 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3319 if (!(regstart && regend && old_regstart && old_regend && reg_info
3320 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3329 /* We must initialize all our variables to NULL, so that
3330 `FREE_VARIABLES' doesn't try to free them. */
3331 regstart = regend = old_regstart = old_regend = best_regstart
3332 = best_regend = reg_dummy = NULL;
3333 reg_info = reg_info_dummy = (register_info_type *) NULL;
3335 #endif /* REGEX_MALLOC */
3337 /* The starting position is bogus. */
3338 if (pos < 0 || pos > size1 + size2)
3344 /* Initialize subexpression text positions to -1 to mark ones that no
3345 start_memory/stop_memory has been seen for. Also initialize the
3346 register information struct. */
3347 for (mcnt = 1; mcnt < num_regs; mcnt++)
3349 regstart[mcnt] = regend[mcnt]
3350 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3352 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3353 IS_ACTIVE (reg_info[mcnt]) = 0;
3354 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3355 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3358 /* We move `string1' into `string2' if the latter's empty -- but not if
3359 `string1' is null. */
3360 if (size2 == 0 && string1 != NULL)
3367 end1 = string1 + size1;
3368 end2 = string2 + size2;
3370 /* Compute where to stop matching, within the two strings. */
3373 end_match_1 = string1 + stop;
3374 end_match_2 = string2;
3379 end_match_2 = string2 + stop - size1;
3382 /* `p' scans through the pattern as `d' scans through the data.
3383 `dend' is the end of the input string that `d' points within. `d'
3384 is advanced into the following input string whenever necessary, but
3385 this happens before fetching; therefore, at the beginning of the
3386 loop, `d' can be pointing at the end of a string, but it cannot
3388 if (size1 > 0 && pos <= size1)
3395 d = string2 + pos - size1;
3399 DEBUG_PRINT1 ("The compiled pattern is: ");
3400 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3401 DEBUG_PRINT1 ("The string to match is: `");
3402 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3403 DEBUG_PRINT1 ("'\n");
3405 /* This loops over pattern commands. It exits by returning from the
3406 function if the match is complete, or it drops through if the match
3407 fails at this starting point in the input data. */
3410 DEBUG_PRINT2 ("\n0x%x: ", p);
3413 { /* End of pattern means we might have succeeded. */
3414 DEBUG_PRINT1 ("end of pattern ... ");
3416 /* If we haven't matched the entire string, and we want the
3417 longest match, try backtracking. */
3418 if (d != end_match_2)
3420 DEBUG_PRINT1 ("backtracking.\n");
3422 if (!FAIL_STACK_EMPTY ())
3423 { /* More failure points to try. */
3424 boolean same_str_p = (FIRST_STRING_P (match_end)
3425 == MATCHING_IN_FIRST_STRING);
3427 /* If exceeds best match so far, save it. */
3429 || (same_str_p && d > match_end)
3430 || (!same_str_p && !MATCHING_IN_FIRST_STRING))
3432 best_regs_set = true;
3435 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3437 for (mcnt = 1; mcnt < num_regs; mcnt++)
3439 best_regstart[mcnt] = regstart[mcnt];
3440 best_regend[mcnt] = regend[mcnt];
3446 /* If no failure points, don't restore garbage. */
3447 else if (best_regs_set)
3450 /* Restore best match. It may happen that `dend ==
3451 end_match_1' while the restored d is in string2.
3452 For example, the pattern `x.*y.*z' against the
3453 strings `x-' and `y-z-', if the two strings are
3454 not consecutive in memory. */
3455 DEBUG_PRINT1 ("Restoring best registers.\n");
3458 dend = ((d >= string1 && d <= end1)
3459 ? end_match_1 : end_match_2);
3461 for (mcnt = 1; mcnt < num_regs; mcnt++)
3463 regstart[mcnt] = best_regstart[mcnt];
3464 regend[mcnt] = best_regend[mcnt];
3467 } /* d != end_match_2 */
3469 DEBUG_PRINT1 ("Accepting match.\n");
3471 /* If caller wants register contents data back, do it. */
3472 if (regs && !bufp->no_sub)
3474 /* Have the register data arrays been allocated? */
3475 if (bufp->regs_allocated == REGS_UNALLOCATED)
3476 { /* No. So allocate them with malloc. We need one
3477 extra element beyond `num_regs' for the `-1' marker
3479 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3480 regs->start = TALLOC (regs->num_regs, regoff_t);
3481 regs->end = TALLOC (regs->num_regs, regoff_t);
3482 if (regs->start == NULL || regs->end == NULL)
3484 bufp->regs_allocated = REGS_REALLOCATE;
3486 else if (bufp->regs_allocated == REGS_REALLOCATE)
3487 { /* Yes. If we need more elements than were already
3488 allocated, reallocate them. If we need fewer, just
3490 if (regs->num_regs < num_regs + 1)
3492 regs->num_regs = num_regs + 1;
3493 RETALLOC (regs->start, regs->num_regs, regoff_t);
3494 RETALLOC (regs->end, regs->num_regs, regoff_t);
3495 if (regs->start == NULL || regs->end == NULL)
3501 /* These braces fend off a "empty body in an else-statement"
3502 warning under GCC when assert expands to nothing. */
3503 assert (bufp->regs_allocated == REGS_FIXED);
3506 /* Convert the pointer data in `regstart' and `regend' to
3507 indices. Register zero has to be set differently,
3508 since we haven't kept track of any info for it. */
3509 if (regs->num_regs > 0)
3511 regs->start[0] = pos;
3512 regs->end[0] = (MATCHING_IN_FIRST_STRING ? d - string1
3513 : d - string2 + size1);
3516 /* Go through the first `min (num_regs, regs->num_regs)'
3517 registers, since that is all we initialized. */
3518 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3520 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3521 regs->start[mcnt] = regs->end[mcnt] = -1;
3524 regs->start[mcnt] = POINTER_TO_OFFSET (regstart[mcnt]);
3525 regs->end[mcnt] = POINTER_TO_OFFSET (regend[mcnt]);
3529 /* If the regs structure we return has more elements than
3530 were in the pattern, set the extra elements to -1. If
3531 we (re)allocated the registers, this is the case,
3532 because we always allocate enough to have at least one
3534 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3535 regs->start[mcnt] = regs->end[mcnt] = -1;
3536 } /* regs && !bufp->no_sub */
3539 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3540 nfailure_points_pushed, nfailure_points_popped,
3541 nfailure_points_pushed - nfailure_points_popped);
3542 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3544 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3548 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3553 /* Otherwise match next pattern command. */
3554 #ifdef SWITCH_ENUM_BUG
3555 switch ((int) ((re_opcode_t) *p++))
3557 switch ((re_opcode_t) *p++)
3560 /* Ignore these. Used to ignore the n of succeed_n's which
3561 currently have n == 0. */
3563 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3567 /* Match the next n pattern characters exactly. The following
3568 byte in the pattern defines n, and the n bytes after that
3569 are the characters to match. */
3572 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3574 /* This is written out as an if-else so we don't waste time
3575 testing `translate' inside the loop. */
3581 if (translate[(unsigned char) *d++] != (char) *p++)
3591 if (*d++ != (char) *p++) goto fail;
3595 SET_REGS_MATCHED ();
3599 /* Match any character except possibly a newline or a null. */
3601 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3605 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3606 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3609 SET_REGS_MATCHED ();
3610 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3618 register unsigned char c;
3619 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3621 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3624 c = TRANSLATE (*d); /* The character to match. */
3626 /* Cast to `unsigned' instead of `unsigned char' in case the
3627 bit list is a full 32 bytes long. */
3628 if (c < (unsigned) (*p * BYTEWIDTH)
3629 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3634 if (!not) goto fail;
3636 SET_REGS_MATCHED ();
3642 /* The beginning of a group is represented by start_memory.
3643 The arguments are the register number in the next byte, and the
3644 number of groups inner to this one in the next. The text
3645 matched within the group is recorded (in the internal
3646 registers data structure) under the register number. */
3648 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3650 /* Find out if this group can match the empty string. */
3651 p1 = p; /* To send to group_match_null_string_p. */
3653 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3654 REG_MATCH_NULL_STRING_P (reg_info[*p])
3655 = group_match_null_string_p (&p1, pend, reg_info);
3657 /* Save the position in the string where we were the last time
3658 we were at this open-group operator in case the group is
3659 operated upon by a repetition operator, e.g., with `(a*)*b'
3660 against `ab'; then we want to ignore where we are now in
3661 the string in case this attempt to match fails. */
3662 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3663 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3665 DEBUG_PRINT2 (" old_regstart: %d\n",
3666 POINTER_TO_OFFSET (old_regstart[*p]));
3669 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
3671 IS_ACTIVE (reg_info[*p]) = 1;
3672 MATCHED_SOMETHING (reg_info[*p]) = 0;
3674 /* This is the new highest active register. */
3675 highest_active_reg = *p;
3677 /* If nothing was active before, this is the new lowest active
3679 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3680 lowest_active_reg = *p;
3682 /* Move past the register number and inner group count. */
3687 /* The stop_memory opcode represents the end of a group. Its
3688 arguments are the same as start_memory's: the register
3689 number, and the number of inner groups. */
3691 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
3693 /* We need to save the string position the last time we were at
3694 this close-group operator in case the group is operated
3695 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3696 against `aba'; then we want to ignore where we are now in
3697 the string in case this attempt to match fails. */
3698 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3699 ? REG_UNSET (regend[*p]) ? d : regend[*p]
3701 DEBUG_PRINT2 (" old_regend: %d\n",
3702 POINTER_TO_OFFSET (old_regend[*p]));
3705 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
3707 /* This register isn't active anymore. */
3708 IS_ACTIVE (reg_info[*p]) = 0;
3710 /* If this was the only register active, nothing is active
3712 if (lowest_active_reg == highest_active_reg)
3714 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3715 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3718 { /* We must scan for the new highest active register, since
3719 it isn't necessarily one less than now: consider
3720 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3721 new highest active register is 1. */
3722 unsigned char r = *p - 1;
3723 while (r > 0 && !IS_ACTIVE (reg_info[r]))
3726 /* If we end up at register zero, that means that we saved
3727 the registers as the result of an `on_failure_jump', not
3728 a `start_memory', and we jumped to past the innermost
3729 `stop_memory'. For example, in ((.)*) we save
3730 registers 1 and 2 as a result of the *, but when we pop
3731 back to the second ), we are at the stop_memory 1.
3732 Thus, nothing is active. */
3735 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3736 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3739 highest_active_reg = r;
3742 /* If just failed to match something this time around with a
3743 group that's operated on by a repetition operator, try to
3744 force exit from the ``loop'', and restore the register
3745 information for this group that we had before trying this
3747 if ((!MATCHED_SOMETHING (reg_info[*p])
3748 || (re_opcode_t) p[-3] == start_memory)
3751 boolean is_a_jump_n = false;
3755 switch ((re_opcode_t) *p1++)
3759 case pop_failure_jump:
3760 case maybe_pop_jump:
3762 case dummy_failure_jump:
3763 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3773 /* If the next operation is a jump backwards in the pattern
3774 to an on_failure_jump right before the start_memory
3775 corresponding to this stop_memory, exit from the loop
3776 by forcing a failure after pushing on the stack the
3777 on_failure_jump's jump in the pattern, and d. */
3778 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
3779 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
3781 /* If this group ever matched anything, then restore
3782 what its registers were before trying this last
3783 failed match, e.g., with `(a*)*b' against `ab' for
3784 regstart[1], and, e.g., with `((a*)*(b*)*)*'
3785 against `aba' for regend[3].
3787 Also restore the registers for inner groups for,
3788 e.g., `((a*)(b*))*' against `aba' (register 3 would
3789 otherwise get trashed). */
3791 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
3795 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
3797 /* Restore this and inner groups' (if any) registers. */
3798 for (r = *p; r < *p + *(p + 1); r++)
3800 regstart[r] = old_regstart[r];
3802 /* xx why this test? */
3803 if ((int) old_regend[r] >= (int) regstart[r])
3804 regend[r] = old_regend[r];
3808 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3809 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
3815 /* Move past the register number and the inner group count. */
3820 /* \<digit> has been turned into a `duplicate' command which is
3821 followed by the numeric value of <digit> as the register number. */
3824 register const char *d2, *dend2;
3825 int regno = *p++; /* Get which register to match against. */
3826 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
3828 /* Can't back reference a group which we've never matched. */
3829 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
3832 /* Where in input to try to start matching. */
3833 d2 = regstart[regno];
3835 /* Where to stop matching; if both the place to start and
3836 the place to stop matching are in the same string, then
3837 set to the place to stop, otherwise, for now have to use
3838 the end of the first string. */
3840 dend2 = ((FIRST_STRING_P (regstart[regno])
3841 == FIRST_STRING_P (regend[regno]))
3842 ? regend[regno] : end_match_1);
3845 /* If necessary, advance to next segment in register
3849 if (dend2 == end_match_2) break;
3850 if (dend2 == regend[regno]) break;
3852 /* End of string1 => advance to string2. */
3854 dend2 = regend[regno];
3856 /* At end of register contents => success */
3857 if (d2 == dend2) break;
3859 /* If necessary, advance to next segment in data. */
3862 /* How many characters left in this segment to match. */
3865 /* Want how many consecutive characters we can match in
3866 one shot, so, if necessary, adjust the count. */
3867 if (mcnt > dend2 - d2)
3870 /* Compare that many; failure if mismatch, else move
3873 ? bcmp_translate (d, d2, mcnt, translate)
3874 : bcmp (d, d2, mcnt))
3876 d += mcnt, d2 += mcnt;
3882 /* begline matches the empty string at the beginning of the string
3883 (unless `not_bol' is set in `bufp'), and, if
3884 `newline_anchor' is set, after newlines. */
3886 DEBUG_PRINT1 ("EXECUTING begline.\n");
3888 if (AT_STRINGS_BEG (d))
3890 if (!bufp->not_bol) break;
3892 else if (d[-1] == '\n' && bufp->newline_anchor)
3896 /* In all other cases, we fail. */
3900 /* endline is the dual of begline. */
3902 DEBUG_PRINT1 ("EXECUTING endline.\n");
3904 if (AT_STRINGS_END (d))
3906 if (!bufp->not_eol) break;
3909 /* We have to ``prefetch'' the next character. */
3910 else if ((d == end1 ? *string2 : *d) == '\n'
3911 && bufp->newline_anchor)
3918 /* Match at the very beginning of the data. */
3920 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
3921 if (AT_STRINGS_BEG (d))
3926 /* Match at the very end of the data. */
3928 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
3929 if (AT_STRINGS_END (d))
3934 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
3935 pushes NULL as the value for the string on the stack. Then
3936 `pop_failure_point' will keep the current value for the
3937 string, instead of restoring it. To see why, consider
3938 matching `foo\nbar' against `.*\n'. The .* matches the foo;
3939 then the . fails against the \n. But the next thing we want
3940 to do is match the \n against the \n; if we restored the
3941 string value, we would be back at the foo.
3943 Because this is used only in specific cases, we don't need to
3944 check all the things that `on_failure_jump' does, to make
3945 sure the right things get saved on the stack. Hence we don't
3946 share its code. The only reason to push anything on the
3947 stack at all is that otherwise we would have to change
3948 `anychar's code to do something besides goto fail in this
3949 case; that seems worse than this. */
3950 case on_failure_keep_string_jump:
3951 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
3953 EXTRACT_NUMBER_AND_INCR (mcnt, p);
3954 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
3956 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
3960 /* Uses of on_failure_jump:
3962 Each alternative starts with an on_failure_jump that points
3963 to the beginning of the next alternative. Each alternative
3964 except the last ends with a jump that in effect jumps past
3965 the rest of the alternatives. (They really jump to the
3966 ending jump of the following alternative, because tensioning
3967 these jumps is a hassle.)
3969 Repeats start with an on_failure_jump that points past both
3970 the repetition text and either the following jump or
3971 pop_failure_jump back to this on_failure_jump. */
3972 case on_failure_jump:
3974 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
3976 EXTRACT_NUMBER_AND_INCR (mcnt, p);
3977 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
3979 /* If this on_failure_jump comes right before a group (i.e.,
3980 the original * applied to a group), save the information
3981 for that group and all inner ones, so that if we fail back
3982 to this point, the group's information will be correct.
3983 For example, in \(a*\)*\1, we need the preceding group,
3984 and in \(\(a*\)b*\)\2, we need the inner group. */
3986 /* We can't use `p' to check ahead because we push
3987 a failure point to `p + mcnt' after we do this. */
3990 /* We need to skip no_op's before we look for the
3991 start_memory in case this on_failure_jump is happening as
3992 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
3994 while (p1 < pend && (re_opcode_t) *p1 == no_op)
3997 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
3999 /* We have a new highest active register now. This will
4000 get reset at the start_memory we are about to get to,
4001 but we will have saved all the registers relevant to
4002 this repetition op, as described above. */
4003 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4004 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4005 lowest_active_reg = *(p1 + 1);
4008 DEBUG_PRINT1 (":\n");
4009 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4013 /* A smart repeat ends with `maybe_pop_jump'.
4014 We change it to either `pop_failure_jump' or `jump'. */
4015 case maybe_pop_jump:
4016 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4017 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4019 register unsigned char *p2 = p;
4021 /* Compare the beginning of the repeat with what in the
4022 pattern follows its end. If we can establish that there
4023 is nothing that they would both match, i.e., that we
4024 would have to backtrack because of (as in, e.g., `a*a')
4025 then we can change to pop_failure_jump, because we'll
4026 never have to backtrack.
4028 This is not true in the case of alternatives: in
4029 `(a|ab)*' we do need to backtrack to the `ab' alternative
4030 (e.g., if the string was `ab'). But instead of trying to
4031 detect that here, the alternative has put on a dummy
4032 failure point which is what we will end up popping. */
4034 /* Skip over open/close-group commands. */
4035 while (p2 + 2 < pend
4036 && ((re_opcode_t) *p2 == stop_memory
4037 || (re_opcode_t) *p2 == start_memory))
4038 p2 += 3; /* Skip over args, too. */
4040 /* If we're at the end of the pattern, we can change. */
4043 /* Consider what happens when matching ":\(.*\)"
4044 against ":/". I don't really understand this code
4046 p[-3] = (unsigned char) pop_failure_jump;
4048 (" End of pattern: change to `pop_failure_jump'.\n");
4051 else if ((re_opcode_t) *p2 == exactn
4052 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4054 register unsigned char c
4055 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4058 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4059 to the `maybe_finalize_jump' of this case. Examine what
4061 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4063 p[-3] = (unsigned char) pop_failure_jump;
4064 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4068 else if ((re_opcode_t) p1[3] == charset
4069 || (re_opcode_t) p1[3] == charset_not)
4071 int not = (re_opcode_t) p1[3] == charset_not;
4073 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4074 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4077 /* `not' is equal to 1 if c would match, which means
4078 that we can't change to pop_failure_jump. */
4081 p[-3] = (unsigned char) pop_failure_jump;
4082 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4087 p -= 2; /* Point at relative address again. */
4088 if ((re_opcode_t) p[-1] != pop_failure_jump)
4090 p[-1] = (unsigned char) jump;
4091 DEBUG_PRINT1 (" Match => jump.\n");
4092 goto unconditional_jump;
4094 /* Note fall through. */
4097 /* The end of a simple repeat has a pop_failure_jump back to
4098 its matching on_failure_jump, where the latter will push a
4099 failure point. The pop_failure_jump takes off failure
4100 points put on by this pop_failure_jump's matching
4101 on_failure_jump; we got through the pattern to here from the
4102 matching on_failure_jump, so didn't fail. */
4103 case pop_failure_jump:
4105 /* We need to pass separate storage for the lowest and
4106 highest registers, even though we don't care about the
4107 actual values. Otherwise, we will restore only one
4108 register from the stack, since lowest will == highest in
4109 `pop_failure_point'. */
4110 unsigned dummy_low_reg, dummy_high_reg;
4111 unsigned char *pdummy;
4114 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4115 POP_FAILURE_POINT (sdummy, pdummy,
4116 dummy_low_reg, dummy_high_reg,
4117 reg_dummy, reg_dummy, reg_info_dummy);
4119 /* Note fall through. */
4122 /* Unconditionally jump (without popping any failure points). */
4125 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4126 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4127 p += mcnt; /* Do the jump. */
4128 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4132 /* We need this opcode so we can detect where alternatives end
4133 in `group_match_null_string_p' et al. */
4135 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4136 goto unconditional_jump;
4139 /* Normally, the on_failure_jump pushes a failure point, which
4140 then gets popped at pop_failure_jump. We will end up at
4141 pop_failure_jump, also, and with a pattern of, say, `a+', we
4142 are skipping over the on_failure_jump, so we have to push
4143 something meaningless for pop_failure_jump to pop. */
4144 case dummy_failure_jump:
4145 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4146 /* It doesn't matter what we push for the string here. What
4147 the code at `fail' tests is the value for the pattern. */
4148 PUSH_FAILURE_POINT (0, 0, -2);
4149 goto unconditional_jump;
4152 /* At the end of an alternative, we need to push a dummy failure
4153 point in case we are followed by a `pop_failure_jump', because
4154 we don't want the failure point for the alternative to be
4155 popped. For example, matching `(a|ab)*' against `aab'
4156 requires that we match the `ab' alternative. */
4157 case push_dummy_failure:
4158 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4159 /* See comments just above at `dummy_failure_jump' about the
4161 PUSH_FAILURE_POINT (0, 0, -2);
4164 /* Have to succeed matching what follows at least n times.
4165 After that, handle like `on_failure_jump'. */
4167 EXTRACT_NUMBER (mcnt, p + 2);
4168 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4171 /* Originally, this is how many times we HAVE to succeed. */
4176 STORE_NUMBER_AND_INCR (p, mcnt);
4177 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4181 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4182 p[2] = (unsigned char) no_op;
4183 p[3] = (unsigned char) no_op;
4189 EXTRACT_NUMBER (mcnt, p + 2);
4190 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4192 /* Originally, this is how many times we CAN jump. */
4196 STORE_NUMBER (p + 2, mcnt);
4197 goto unconditional_jump;
4199 /* If don't have to jump any more, skip over the rest of command. */
4206 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4208 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4210 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4211 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4212 STORE_NUMBER (p1, mcnt);
4217 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4218 if (AT_WORD_BOUNDARY (d))
4223 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4224 if (AT_WORD_BOUNDARY (d))
4229 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4230 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4235 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4236 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4237 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4244 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4245 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4250 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4251 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4256 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4257 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4260 #else /* not emacs19 */
4262 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4263 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4266 #endif /* not emacs19 */
4269 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4274 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4278 if (SYNTAX (*d++) != (enum syntaxcode) mcnt)
4280 SET_REGS_MATCHED ();
4284 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4286 goto matchnotsyntax;
4289 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4293 if (SYNTAX (*d++) == (enum syntaxcode) mcnt)
4295 SET_REGS_MATCHED ();
4298 #else /* not emacs */
4300 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4302 if (!WORDCHAR_P (d))
4304 SET_REGS_MATCHED ();
4309 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4313 SET_REGS_MATCHED ();
4316 #endif /* not emacs */
4321 continue; /* Successfully executed one pattern command; keep going. */
4324 /* We goto here if a matching operation fails. */
4326 if (!FAIL_STACK_EMPTY ())
4327 { /* A restart point is known. Restore to that state. */
4328 DEBUG_PRINT1 ("\nFAIL:\n");
4329 POP_FAILURE_POINT (d, p,
4330 lowest_active_reg, highest_active_reg,
4331 regstart, regend, reg_info);
4333 /* If this failure point is a dummy, try the next one. */
4337 /* If we failed to the end of the pattern, don't examine *p. */
4341 boolean is_a_jump_n = false;
4343 /* If failed to a backwards jump that's part of a repetition
4344 loop, need to pop this failure point and use the next one. */
4345 switch ((re_opcode_t) *p)
4349 case maybe_pop_jump:
4350 case pop_failure_jump:
4353 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4356 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4358 && (re_opcode_t) *p1 == on_failure_jump))
4366 if (d >= string1 && d <= end1)
4370 break; /* Matching at this starting point really fails. */
4374 goto restore_best_regs;
4378 return -1; /* Failure to match. */
4381 /* Subroutine definitions for re_match_2. */
4384 /* We are passed P pointing to a register number after a start_memory.
4386 Return true if the pattern up to the corresponding stop_memory can
4387 match the empty string, and false otherwise.
4389 If we find the matching stop_memory, sets P to point to one past its number.
4390 Otherwise, sets P to an undefined byte less than or equal to END.
4392 We don't handle duplicates properly (yet). */
4395 group_match_null_string_p (p, end, reg_info)
4396 unsigned char **p, *end;
4397 register_info_type *reg_info;
4400 /* Point to after the args to the start_memory. */
4401 unsigned char *p1 = *p + 2;
4405 /* Skip over opcodes that can match nothing, and return true or
4406 false, as appropriate, when we get to one that can't, or to the
4407 matching stop_memory. */
4409 switch ((re_opcode_t) *p1)
4411 /* Could be either a loop or a series of alternatives. */
4412 case on_failure_jump:
4414 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4416 /* If the next operation is not a jump backwards in the
4421 /* Go through the on_failure_jumps of the alternatives,
4422 seeing if any of the alternatives cannot match nothing.
4423 The last alternative starts with only a jump,
4424 whereas the rest start with on_failure_jump and end
4425 with a jump, e.g., here is the pattern for `a|b|c':
4427 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4428 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4431 So, we have to first go through the first (n-1)
4432 alternatives and then deal with the last one separately. */
4435 /* Deal with the first (n-1) alternatives, which start
4436 with an on_failure_jump (see above) that jumps to right
4437 past a jump_past_alt. */
4439 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4441 /* `mcnt' holds how many bytes long the alternative
4442 is, including the ending `jump_past_alt' and
4445 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4449 /* Move to right after this alternative, including the
4453 /* Break if it's the beginning of an n-th alternative
4454 that doesn't begin with an on_failure_jump. */
4455 if ((re_opcode_t) *p1 != on_failure_jump)
4458 /* Still have to check that it's not an n-th
4459 alternative that starts with an on_failure_jump. */
4461 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4462 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4464 /* Get to the beginning of the n-th alternative. */
4470 /* Deal with the last alternative: go back and get number
4471 of the `jump_past_alt' just before it. `mcnt' contains
4472 the length of the alternative. */
4473 EXTRACT_NUMBER (mcnt, p1 - 2);
4475 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4478 p1 += mcnt; /* Get past the n-th alternative. */
4484 assert (p1[1] == **p);
4490 if (!common_op_match_null_string_p (&p1, end, reg_info))
4493 } /* while p1 < end */
4496 } /* group_match_null_string_p */
4499 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4500 It expects P to be the first byte of a single alternative and END one
4501 byte past the last. The alternative can contain groups. */
4504 alt_match_null_string_p (p, end, reg_info)
4505 unsigned char *p, *end;
4506 register_info_type *reg_info;
4509 unsigned char *p1 = p;
4513 /* Skip over opcodes that can match nothing, and break when we get
4514 to one that can't. */
4516 switch ((re_opcode_t) *p1)
4519 case on_failure_jump:
4521 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4526 if (!common_op_match_null_string_p (&p1, end, reg_info))
4529 } /* while p1 < end */
4532 } /* alt_match_null_string_p */
4535 /* Deals with the ops common to group_match_null_string_p and
4536 alt_match_null_string_p.
4538 Sets P to one after the op and its arguments, if any. */
4541 common_op_match_null_string_p (p, end, reg_info)
4542 unsigned char **p, *end;
4543 register_info_type *reg_info;
4548 unsigned char *p1 = *p;
4550 switch ((re_opcode_t) *p1++)
4570 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4571 ret = group_match_null_string_p (&p1, end, reg_info);
4573 /* Have to set this here in case we're checking a group which
4574 contains a group and a back reference to it. */
4576 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4577 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4583 /* If this is an optimized succeed_n for zero times, make the jump. */
4585 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4593 /* Get to the number of times to succeed. */
4595 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4600 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4608 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
4616 /* All other opcodes mean we cannot match the empty string. */
4622 } /* common_op_match_null_string_p */
4625 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4626 bytes; nonzero otherwise. */
4629 bcmp_translate (s1, s2, len, translate)
4630 unsigned char *s1, *s2;
4634 register unsigned char *p1 = s1, *p2 = s2;
4637 if (translate[*p1++] != translate[*p2++]) return 1;
4643 /* Entry points for GNU code. */
4645 /* re_compile_pattern is the GNU regular expression compiler: it
4646 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4647 Returns 0 if the pattern was valid, otherwise an error string.
4649 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4650 are set in BUFP on entry.
4652 We call regex_compile to do the actual compilation. */
4655 re_compile_pattern (pattern, length, bufp)
4656 const char *pattern;
4658 struct re_pattern_buffer *bufp;
4662 /* GNU code is written to assume at least RE_NREGS registers will be set
4663 (and at least one extra will be -1). */
4664 bufp->regs_allocated = REGS_UNALLOCATED;
4666 /* And GNU code determines whether or not to get register information
4667 by passing null for the REGS argument to re_match, etc., not by
4671 /* Match anchors at newline. */
4672 bufp->newline_anchor = 1;
4674 ret = regex_compile (pattern, length, re_syntax_options, bufp);
4676 return re_error_msg[(int) ret];
4679 /* Entry points compatible with 4.2 BSD regex library. We don't define
4680 them if this is an Emacs or POSIX compilation. */
4682 #if !defined (emacs) && !defined (_POSIX_SOURCE)
4684 /* BSD has one and only one pattern buffer. */
4685 static struct re_pattern_buffer re_comp_buf;
4695 if (!re_comp_buf.buffer)
4696 return "No previous regular expression";
4700 if (!re_comp_buf.buffer)
4702 re_comp_buf.buffer = (unsigned char *) malloc (200);
4703 if (re_comp_buf.buffer == NULL)
4704 return "Memory exhausted";
4705 re_comp_buf.allocated = 200;
4707 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
4708 if (re_comp_buf.fastmap == NULL)
4709 return "Memory exhausted";
4712 /* Since `re_exec' always passes NULL for the `regs' argument, we
4713 don't need to initialize the pattern buffer fields which affect it. */
4715 /* Match anchors at newlines. */
4716 re_comp_buf.newline_anchor = 1;
4718 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
4720 /* Yes, we're discarding `const' here. */
4721 return (char *) re_error_msg[(int) ret];
4729 const int len = strlen (s);
4731 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
4733 #endif /* not emacs and not _POSIX_SOURCE */
4735 /* POSIX.2 functions. Don't define these for Emacs. */
4739 /* regcomp takes a regular expression as a string and compiles it.
4741 PREG is a regex_t *. We do not expect any fields to be initialized,
4742 since POSIX says we shouldn't. Thus, we set
4744 `buffer' to the compiled pattern;
4745 `used' to the length of the compiled pattern;
4746 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
4747 REG_EXTENDED bit in CFLAGS is set; otherwise, to
4748 RE_SYNTAX_POSIX_BASIC;
4749 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
4750 `fastmap' and `fastmap_accurate' to zero;
4751 `re_nsub' to the number of subexpressions in PATTERN.
4753 PATTERN is the address of the pattern string.
4755 CFLAGS is a series of bits which affect compilation.
4757 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
4758 use POSIX basic syntax.
4760 If REG_NEWLINE is set, then . and [^...] don't match newline.
4761 Also, regexec will try a match beginning after every newline.
4763 If REG_ICASE is set, then we considers upper- and lowercase
4764 versions of letters to be equivalent when matching.
4766 If REG_NOSUB is set, then when PREG is passed to regexec, that
4767 routine will report only success or failure, and nothing about the
4770 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
4771 the return codes and their meanings.) */
4774 regcomp (preg, pattern, cflags)
4776 const char *pattern;
4781 = (cflags & REG_EXTENDED) ?
4782 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
4784 /* regex_compile will allocate the space for the compiled pattern. */
4786 preg->allocated = 0;
4788 /* Don't bother to use a fastmap when searching. This simplifies the
4789 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
4790 characters after newlines into the fastmap. This way, we just try
4794 if (cflags & REG_ICASE)
4798 preg->translate = (char *) malloc (CHAR_SET_SIZE);
4799 if (preg->translate == NULL)
4800 return (int) REG_ESPACE;
4802 /* Map uppercase characters to corresponding lowercase ones. */
4803 for (i = 0; i < CHAR_SET_SIZE; i++)
4804 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
4807 preg->translate = NULL;
4809 /* If REG_NEWLINE is set, newlines are treated differently. */
4810 if (cflags & REG_NEWLINE)
4811 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
4812 syntax &= ~RE_DOT_NEWLINE;
4813 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
4814 /* It also changes the matching behavior. */
4815 preg->newline_anchor = 1;
4818 preg->newline_anchor = 0;
4820 preg->no_sub = !!(cflags & REG_NOSUB);
4822 /* POSIX says a null character in the pattern terminates it, so we
4823 can use strlen here in compiling the pattern. */
4824 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
4826 /* POSIX doesn't distinguish between an unmatched open-group and an
4827 unmatched close-group: both are REG_EPAREN. */
4828 if (ret == REG_ERPAREN) ret = REG_EPAREN;
4834 /* regexec searches for a given pattern, specified by PREG, in the
4837 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
4838 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
4839 least NMATCH elements, and we set them to the offsets of the
4840 corresponding matched substrings.
4842 EFLAGS specifies `execution flags' which affect matching: if
4843 REG_NOTBOL is set, then ^ does not match at the beginning of the
4844 string; if REG_NOTEOL is set, then $ does not match at the end.
4846 We return 0 if we find a match and REG_NOMATCH if not. */
4849 regexec (preg, string, nmatch, pmatch, eflags)
4850 const regex_t *preg;
4853 regmatch_t pmatch[];
4857 struct re_registers regs;
4858 regex_t private_preg;
4859 int len = strlen (string);
4860 boolean want_reg_info = !preg->no_sub && nmatch > 0;
4862 private_preg = *preg;
4864 private_preg.not_bol = !!(eflags & REG_NOTBOL);
4865 private_preg.not_eol = !!(eflags & REG_NOTEOL);
4867 /* The user has told us exactly how many registers to return
4868 information about, via `nmatch'. We have to pass that on to the
4869 matching routines. */
4870 private_preg.regs_allocated = REGS_FIXED;
4874 regs.num_regs = nmatch;
4875 regs.start = TALLOC (nmatch, regoff_t);
4876 regs.end = TALLOC (nmatch, regoff_t);
4877 if (regs.start == NULL || regs.end == NULL)
4878 return (int) REG_NOMATCH;
4881 /* Perform the searching operation. */
4882 ret = re_search (&private_preg, string, len,
4883 /* start: */ 0, /* range: */ len,
4884 want_reg_info ? ®s : (struct re_registers *) 0);
4886 /* Copy the register information to the POSIX structure. */
4893 for (r = 0; r < nmatch; r++)
4895 pmatch[r].rm_so = regs.start[r];
4896 pmatch[r].rm_eo = regs.end[r];
4900 /* If we needed the temporary register info, free the space now. */
4905 /* We want zero return to mean success, unlike `re_search'. */
4906 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
4910 /* Returns a message corresponding to an error code, ERRCODE, returned
4911 from either regcomp or regexec. We don't use PREG here. */
4914 regerror (errcode, preg, errbuf, errbuf_size)
4916 const regex_t *preg;
4924 || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0])))
4925 /* Only error codes returned by the rest of the code should be passed
4926 to this routine. If we are given anything else, or if other regex
4927 code generates an invalid error code, then the program has a bug.
4928 Dump core so we can fix it. */
4931 msg = re_error_msg[errcode];
4933 /* POSIX doesn't require that we do anything in this case, but why
4938 msg_size = strlen (msg) + 1; /* Includes the null. */
4940 if (errbuf_size != 0)
4942 if (msg_size > errbuf_size)
4944 strncpy (errbuf, msg, errbuf_size - 1);
4945 errbuf[errbuf_size - 1] = 0;
4948 strcpy (errbuf, msg);
4955 /* Free dynamically allocated space used by PREG. */
4961 if (preg->buffer != NULL)
4962 free (preg->buffer);
4963 preg->buffer = NULL;
4965 preg->allocated = 0;
4968 if (preg->fastmap != NULL)
4969 free (preg->fastmap);
4970 preg->fastmap = NULL;
4971 preg->fastmap_accurate = 0;
4973 if (preg->translate != NULL)
4974 free (preg->translate);
4975 preg->translate = NULL;
4978 #endif /* not emacs */
4982 make-backup-files: t
4984 trim-versions-without-asking: nil