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. */
135 #define ISBLANK(c) (isascii (c) && isblank (c))
137 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
140 #define ISGRAPH(c) (isascii (c) && isgraph (c))
142 #define ISGRAPH(c) (isascii (c) && isprint (c) && !isspace (c))
145 #define ISPRINT(c) (isascii (c) && isprint (c))
146 #define ISDIGIT(c) (isascii (c) && isdigit (c))
147 #define ISALNUM(c) (isascii (c) && isalnum (c))
148 #define ISALPHA(c) (isascii (c) && isalpha (c))
149 #define ISCNTRL(c) (isascii (c) && iscntrl (c))
150 #define ISLOWER(c) (isascii (c) && islower (c))
151 #define ISPUNCT(c) (isascii (c) && ispunct (c))
152 #define ISSPACE(c) (isascii (c) && isspace (c))
153 #define ISUPPER(c) (isascii (c) && isupper (c))
154 #define ISXDIGIT(c) (isascii (c) && isxdigit (c))
160 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
161 since ours (we hope) works properly with all combinations of
162 machines, compilers, `char' and `unsigned char' argument types.
163 (Per Bothner suggested the basic approach.) */
164 #undef SIGN_EXTEND_CHAR
166 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
167 #else /* not __STDC__ */
168 /* As in Harbison and Steele. */
169 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
172 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
173 use `alloca' instead of `malloc'. This is because using malloc in
174 re_search* or re_match* could cause memory leaks when C-g is used in
175 Emacs; also, malloc is slower and causes storage fragmentation. On
176 the other hand, malloc is more portable, and easier to debug.
178 Because we sometimes use alloca, some routines have to be macros,
179 not functions -- `alloca'-allocated space disappears at the end of the
180 function it is called in. */
184 #define REGEX_ALLOCATE malloc
185 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
187 #else /* not REGEX_MALLOC */
189 /* Emacs already defines alloca, sometimes. */
192 /* Make alloca work the best possible way. */
194 #define alloca __builtin_alloca
195 #else /* not __GNUC__ */
198 #else /* not __GNUC__ or HAVE_ALLOCA_H */
199 #ifndef _AIX /* Already did AIX, up at the top. */
201 #endif /* not _AIX */
202 #endif /* not HAVE_ALLOCA_H */
203 #endif /* not __GNUC__ */
205 #endif /* not alloca */
207 #define REGEX_ALLOCATE alloca
209 /* Assumes a `char *destination' variable. */
210 #define REGEX_REALLOCATE(source, osize, nsize) \
211 (destination = (char *) alloca (nsize), \
212 bcopy (source, destination, osize), \
215 #endif /* not REGEX_MALLOC */
218 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
219 `string1' or just past its end. This works if PTR is NULL, which is
221 #define FIRST_STRING_P(ptr) \
222 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
224 /* (Re)Allocate N items of type T using malloc, or fail. */
225 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
226 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
227 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
229 #define BYTEWIDTH 8 /* In bits. */
231 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
233 #define MAX(a, b) ((a) > (b) ? (a) : (b))
234 #define MIN(a, b) ((a) < (b) ? (a) : (b))
236 typedef char boolean
;
240 /* These are the command codes that appear in compiled regular
241 expressions. Some opcodes are followed by argument bytes. A
242 command code can specify any interpretation whatsoever for its
243 arguments. Zero bytes may appear in the compiled regular expression.
245 The value of `exactn' is needed in search.c (search_buffer) in Emacs.
246 So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of
247 `exactn' we use here must also be 1. */
253 /* Followed by one byte giving n, then by n literal bytes. */
256 /* Matches any (more or less) character. */
259 /* Matches any one char belonging to specified set. First
260 following byte is number of bitmap bytes. Then come bytes
261 for a bitmap saying which chars are in. Bits in each byte
262 are ordered low-bit-first. A character is in the set if its
263 bit is 1. A character too large to have a bit in the map is
264 automatically not in the set. */
267 /* Same parameters as charset, but match any character that is
268 not one of those specified. */
271 /* Start remembering the text that is matched, for storing in a
272 register. Followed by one byte with the register number, in
273 the range 0 to one less than the pattern buffer's re_nsub
274 field. Then followed by one byte with the number of groups
275 inner to this one. (This last has to be part of the
276 start_memory only because we need it in the on_failure_jump
280 /* Stop remembering the text that is matched and store it in a
281 memory register. Followed by one byte with the register
282 number, in the range 0 to one less than `re_nsub' in the
283 pattern buffer, and one byte with the number of inner groups,
284 just like `start_memory'. (We need the number of inner
285 groups here because we don't have any easy way of finding the
286 corresponding start_memory when we're at a stop_memory.) */
289 /* Match a duplicate of something remembered. Followed by one
290 byte containing the register number. */
293 /* Fail unless at beginning of line. */
296 /* Fail unless at end of line. */
299 /* Succeeds if at beginning of buffer (if emacs) or at beginning
300 of string to be matched (if not). */
303 /* Analogously, for end of buffer/string. */
306 /* Followed by two byte relative address to which to jump. */
309 /* Same as jump, but marks the end of an alternative. */
312 /* Followed by two-byte relative address of place to resume at
313 in case of failure. */
316 /* Like on_failure_jump, but pushes a placeholder instead of the
317 current string position when executed. */
318 on_failure_keep_string_jump
,
320 /* Throw away latest failure point and then jump to following
321 two-byte relative address. */
324 /* Change to pop_failure_jump if know won't have to backtrack to
325 match; otherwise change to jump. This is used to jump
326 back to the beginning of a repeat. If what follows this jump
327 clearly won't match what the repeat does, such that we can be
328 sure that there is no use backtracking out of repetitions
329 already matched, then we change it to a pop_failure_jump.
330 Followed by two-byte address. */
333 /* Jump to following two-byte address, and push a dummy failure
334 point. This failure point will be thrown away if an attempt
335 is made to use it for a failure. A `+' construct makes this
336 before the first repeat. Also used as an intermediary kind
337 of jump when compiling an alternative. */
340 /* Push a dummy failure point and continue. Used at the end of
344 /* Followed by two-byte relative address and two-byte number n.
345 After matching N times, jump to the address upon failure. */
348 /* Followed by two-byte relative address, and two-byte number n.
349 Jump to the address N times, then fail. */
352 /* Set the following two-byte relative address to the
353 subsequent two-byte number. The address *includes* the two
357 wordchar
, /* Matches any word-constituent character. */
358 notwordchar
, /* Matches any char that is not a word-constituent. */
360 wordbeg
, /* Succeeds if at word beginning. */
361 wordend
, /* Succeeds if at word end. */
363 wordbound
, /* Succeeds if at a word boundary. */
364 notwordbound
/* Succeeds if not at a word boundary. */
367 ,before_dot
, /* Succeeds if before point. */
368 at_dot
, /* Succeeds if at point. */
369 after_dot
, /* Succeeds if after point. */
371 /* Matches any character whose syntax is specified. Followed by
372 a byte which contains a syntax code, e.g., Sword. */
375 /* Matches any character whose syntax is not that specified. */
380 /* Common operations on the compiled pattern. */
382 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
384 #define STORE_NUMBER(destination, number) \
386 (destination)[0] = (number) & 0377; \
387 (destination)[1] = (number) >> 8; \
390 /* Same as STORE_NUMBER, except increment DESTINATION to
391 the byte after where the number is stored. Therefore, DESTINATION
392 must be an lvalue. */
394 #define STORE_NUMBER_AND_INCR(destination, number) \
396 STORE_NUMBER (destination, number); \
397 (destination) += 2; \
400 /* Put into DESTINATION a number stored in two contiguous bytes starting
403 #define EXTRACT_NUMBER(destination, source) \
405 (destination) = *(source) & 0377; \
406 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
411 extract_number (dest
, source
)
413 unsigned char *source
;
415 int temp
= SIGN_EXTEND_CHAR (*(source
+ 1));
416 *dest
= *source
& 0377;
420 #ifndef EXTRACT_MACROS /* To debug the macros. */
421 #undef EXTRACT_NUMBER
422 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
423 #endif /* not EXTRACT_MACROS */
427 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
428 SOURCE must be an lvalue. */
430 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
432 EXTRACT_NUMBER (destination, source); \
438 extract_number_and_incr (destination
, source
)
440 unsigned char **source
;
442 extract_number (destination
, *source
);
446 #ifndef EXTRACT_MACROS
447 #undef EXTRACT_NUMBER_AND_INCR
448 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
449 extract_number_and_incr (&dest, &src)
450 #endif /* not EXTRACT_MACROS */
454 /* If DEBUG is defined, Regex prints many voluminous messages about what
455 it is doing (if the variable `debug' is nonzero). If linked with the
456 main program in `iregex.c', you can enter patterns and strings
457 interactively. And if linked with the main program in `main.c' and
458 the other test files, you can run the already-written tests. */
462 /* We use standard I/O for debugging. */
465 /* It is useful to test things that ``must'' be true when debugging. */
468 static int debug
= 0;
470 #define DEBUG_STATEMENT(e) e
471 #define DEBUG_PRINT1(x) if (debug) printf (x)
472 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
473 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
474 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
475 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
476 if (debug) print_partial_compiled_pattern (s, e)
477 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
478 if (debug) print_double_string (w, s1, sz1, s2, sz2)
481 extern void printchar ();
483 /* Print the fastmap in human-readable form. */
486 print_fastmap (fastmap
)
489 unsigned was_a_range
= 0;
492 while (i
< (1 << BYTEWIDTH
))
498 while (i
< (1 << BYTEWIDTH
) && fastmap
[i
])
514 /* Print a compiled pattern string in human-readable form, starting at
515 the START pointer into it and ending just before the pointer END. */
518 print_partial_compiled_pattern (start
, end
)
519 unsigned char *start
;
523 unsigned char *p
= start
;
524 unsigned char *pend
= end
;
532 /* Loop over pattern commands. */
535 switch ((re_opcode_t
) *p
++)
543 printf ("/exactn/%d", mcnt
);
554 printf ("/start_memory/%d/%d", mcnt
, *p
++);
559 printf ("/stop_memory/%d/%d", mcnt
, *p
++);
563 printf ("/duplicate/%d", *p
++);
575 printf ("/charset%s",
576 (re_opcode_t
) *(p
- 1) == charset_not
? "_not" : "");
578 assert (p
+ *p
< pend
);
580 for (c
= 0; c
< *p
; c
++)
583 unsigned char map_byte
= p
[1 + c
];
587 for (bit
= 0; bit
< BYTEWIDTH
; bit
++)
588 if (map_byte
& (1 << bit
))
589 printchar (c
* BYTEWIDTH
+ bit
);
603 case on_failure_jump
:
604 extract_number_and_incr (&mcnt
, &p
);
605 printf ("/on_failure_jump/0/%d", mcnt
);
608 case on_failure_keep_string_jump
:
609 extract_number_and_incr (&mcnt
, &p
);
610 printf ("/on_failure_keep_string_jump/0/%d", mcnt
);
613 case dummy_failure_jump
:
614 extract_number_and_incr (&mcnt
, &p
);
615 printf ("/dummy_failure_jump/0/%d", mcnt
);
618 case push_dummy_failure
:
619 printf ("/push_dummy_failure");
623 extract_number_and_incr (&mcnt
, &p
);
624 printf ("/maybe_pop_jump/0/%d", mcnt
);
627 case pop_failure_jump
:
628 extract_number_and_incr (&mcnt
, &p
);
629 printf ("/pop_failure_jump/0/%d", mcnt
);
633 extract_number_and_incr (&mcnt
, &p
);
634 printf ("/jump_past_alt/0/%d", mcnt
);
638 extract_number_and_incr (&mcnt
, &p
);
639 printf ("/jump/0/%d", mcnt
);
643 extract_number_and_incr (&mcnt
, &p
);
644 extract_number_and_incr (&mcnt2
, &p
);
645 printf ("/succeed_n/0/%d/0/%d", mcnt
, mcnt2
);
649 extract_number_and_incr (&mcnt
, &p
);
650 extract_number_and_incr (&mcnt2
, &p
);
651 printf ("/jump_n/0/%d/0/%d", mcnt
, mcnt2
);
655 extract_number_and_incr (&mcnt
, &p
);
656 extract_number_and_incr (&mcnt2
, &p
);
657 printf ("/set_number_at/0/%d/0/%d", mcnt
, mcnt2
);
661 printf ("/wordbound");
665 printf ("/notwordbound");
677 printf ("/before_dot");
685 printf ("/after_dot");
689 printf ("/syntaxspec");
691 printf ("/%d", mcnt
);
695 printf ("/notsyntaxspec");
697 printf ("/%d", mcnt
);
702 printf ("/wordchar");
706 printf ("/notwordchar");
718 printf ("?%d", *(p
-1));
726 print_compiled_pattern (bufp
)
727 struct re_pattern_buffer
*bufp
;
729 unsigned char *buffer
= bufp
->buffer
;
731 print_partial_compiled_pattern (buffer
, buffer
+ bufp
->used
);
732 printf ("%d bytes used/%d bytes allocated.\n", bufp
->used
, bufp
->allocated
);
734 if (bufp
->fastmap_accurate
&& bufp
->fastmap
)
736 printf ("fastmap: ");
737 print_fastmap (bufp
->fastmap
);
740 printf ("re_nsub: %d\t", bufp
->re_nsub
);
741 printf ("regs_alloc: %d\t", bufp
->regs_allocated
);
742 printf ("can_be_null: %d\t", bufp
->can_be_null
);
743 printf ("newline_anchor: %d\n", bufp
->newline_anchor
);
744 printf ("no_sub: %d\t", bufp
->no_sub
);
745 printf ("not_bol: %d\t", bufp
->not_bol
);
746 printf ("not_eol: %d\t", bufp
->not_eol
);
747 printf ("syntax: %d\n", bufp
->syntax
);
748 /* Perhaps we should print the translate table? */
753 print_double_string (where
, string1
, size1
, string2
, size2
)
766 if (FIRST_STRING_P (where
))
768 for (this_char
= where
- string1
; this_char
< size1
; this_char
++)
769 printchar (string1
[this_char
]);
774 for (this_char
= where
- string2
; this_char
< size2
; this_char
++)
775 printchar (string2
[this_char
]);
779 #else /* not DEBUG */
784 #define DEBUG_STATEMENT(e)
785 #define DEBUG_PRINT1(x)
786 #define DEBUG_PRINT2(x1, x2)
787 #define DEBUG_PRINT3(x1, x2, x3)
788 #define DEBUG_PRINT4(x1, x2, x3, x4)
789 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
790 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
792 #endif /* not DEBUG */
794 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
795 also be assigned to arbitrarily: each pattern buffer stores its own
796 syntax, so it can be changed between regex compilations. */
797 reg_syntax_t re_syntax_options
= RE_SYNTAX_EMACS
;
800 /* Specify the precise syntax of regexps for compilation. This provides
801 for compatibility for various utilities which historically have
802 different, incompatible syntaxes.
804 The argument SYNTAX is a bit mask comprised of the various bits
805 defined in regex.h. We return the old syntax. */
808 re_set_syntax (syntax
)
811 reg_syntax_t ret
= re_syntax_options
;
813 re_syntax_options
= syntax
;
817 /* This table gives an error message for each of the error codes listed
818 in regex.h. Obviously the order here has to be same as there. */
820 static const char *re_error_msg
[] =
821 { NULL
, /* REG_NOERROR */
822 "No match", /* REG_NOMATCH */
823 "Invalid regular expression", /* REG_BADPAT */
824 "Invalid collation character", /* REG_ECOLLATE */
825 "Invalid character class name", /* REG_ECTYPE */
826 "Trailing backslash", /* REG_EESCAPE */
827 "Invalid back reference", /* REG_ESUBREG */
828 "Unmatched [ or [^", /* REG_EBRACK */
829 "Unmatched ( or \\(", /* REG_EPAREN */
830 "Unmatched \\{", /* REG_EBRACE */
831 "Invalid content of \\{\\}", /* REG_BADBR */
832 "Invalid range end", /* REG_ERANGE */
833 "Memory exhausted", /* REG_ESPACE */
834 "Invalid preceding regular expression", /* REG_BADRPT */
835 "Premature end of regular expression", /* REG_EEND */
836 "Regular expression too big", /* REG_ESIZE */
837 "Unmatched ) or \\)", /* REG_ERPAREN */
840 /* Subroutine declarations and macros for regex_compile. */
842 static void store_op1 (), store_op2 ();
843 static void insert_op1 (), insert_op2 ();
844 static boolean
at_begline_loc_p (), at_endline_loc_p ();
845 static boolean
group_in_compile_stack ();
846 static reg_errcode_t
compile_range ();
848 /* Fetch the next character in the uncompiled pattern---translating it
849 if necessary. Also cast from a signed character in the constant
850 string passed to us by the user to an unsigned char that we can use
851 as an array index (in, e.g., `translate'). */
852 #define PATFETCH(c) \
853 do {if (p == pend) return REG_EEND; \
854 c = (unsigned char) *p++; \
855 if (translate) c = translate[c]; \
858 /* Fetch the next character in the uncompiled pattern, with no
860 #define PATFETCH_RAW(c) \
861 do {if (p == pend) return REG_EEND; \
862 c = (unsigned char) *p++; \
865 /* Go backwards one character in the pattern. */
866 #define PATUNFETCH p--
869 /* If `translate' is non-null, return translate[D], else just D. We
870 cast the subscript to translate because some data is declared as
871 `char *', to avoid warnings when a string constant is passed. But
872 when we use a character as a subscript we must make it unsigned. */
873 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
876 /* Macros for outputting the compiled pattern into `buffer'. */
878 /* If the buffer isn't allocated when it comes in, use this. */
879 #define INIT_BUF_SIZE 32
881 /* Make sure we have at least N more bytes of space in buffer. */
882 #define GET_BUFFER_SPACE(n) \
883 while (b - bufp->buffer + (n) > bufp->allocated) \
886 /* Make sure we have one more byte of buffer space and then add C to it. */
887 #define BUF_PUSH(c) \
889 GET_BUFFER_SPACE (1); \
890 *b++ = (unsigned char) (c); \
894 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
895 #define BUF_PUSH_2(c1, c2) \
897 GET_BUFFER_SPACE (2); \
898 *b++ = (unsigned char) (c1); \
899 *b++ = (unsigned char) (c2); \
903 /* As with BUF_PUSH_2, except for three bytes. */
904 #define BUF_PUSH_3(c1, c2, c3) \
906 GET_BUFFER_SPACE (3); \
907 *b++ = (unsigned char) (c1); \
908 *b++ = (unsigned char) (c2); \
909 *b++ = (unsigned char) (c3); \
913 /* Store a jump with opcode OP at LOC to location TO. We store a
914 relative address offset by the three bytes the jump itself occupies. */
915 #define STORE_JUMP(op, loc, to) \
916 store_op1 (op, loc, (to) - (loc) - 3)
918 /* Likewise, for a two-argument jump. */
919 #define STORE_JUMP2(op, loc, to, arg) \
920 store_op2 (op, loc, (to) - (loc) - 3, arg)
922 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
923 #define INSERT_JUMP(op, loc, to) \
924 insert_op1 (op, loc, (to) - (loc) - 3, b)
926 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
927 #define INSERT_JUMP2(op, loc, to, arg) \
928 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
931 /* This is not an arbitrary limit: the arguments which represent offsets
932 into the pattern are two bytes long. So if 2^16 bytes turns out to
933 be too small, many things would have to change. */
934 #define MAX_BUF_SIZE (1L << 16)
937 /* Extend the buffer by twice its current size via realloc and
938 reset the pointers that pointed into the old block to point to the
939 correct places in the new one. If extending the buffer results in it
940 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
941 #define EXTEND_BUFFER() \
943 unsigned char *old_buffer = bufp->buffer; \
944 if (bufp->allocated == MAX_BUF_SIZE) \
946 bufp->allocated <<= 1; \
947 if (bufp->allocated > MAX_BUF_SIZE) \
948 bufp->allocated = MAX_BUF_SIZE; \
949 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
950 if (bufp->buffer == NULL) \
952 /* If the buffer moved, move all the pointers into it. */ \
953 if (old_buffer != bufp->buffer) \
955 b = (b - old_buffer) + bufp->buffer; \
956 begalt = (begalt - old_buffer) + bufp->buffer; \
957 if (fixup_alt_jump) \
958 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
960 laststart = (laststart - old_buffer) + bufp->buffer; \
962 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
967 /* Since we have one byte reserved for the register number argument to
968 {start,stop}_memory, the maximum number of groups we can report
969 things about is what fits in that byte. */
970 #define MAX_REGNUM 255
972 /* But patterns can have more than `MAX_REGNUM' registers. We just
973 ignore the excess. */
974 typedef unsigned regnum_t
;
977 /* Macros for the compile stack. */
979 /* Since offsets can go either forwards or backwards, this type needs to
980 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
981 typedef int pattern_offset_t
;
985 pattern_offset_t begalt_offset
;
986 pattern_offset_t fixup_alt_jump
;
987 pattern_offset_t inner_group_offset
;
988 pattern_offset_t laststart_offset
;
990 } compile_stack_elt_t
;
995 compile_stack_elt_t
*stack
;
997 unsigned avail
; /* Offset of next open position. */
998 } compile_stack_type
;
1001 #define INIT_COMPILE_STACK_SIZE 32
1003 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1004 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1006 /* The next available element. */
1007 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1010 /* Set the bit for character C in a list. */
1011 #define SET_LIST_BIT(c) \
1012 (b[((unsigned char) (c)) / BYTEWIDTH] \
1013 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1016 /* Get the next unsigned number in the uncompiled pattern. */
1017 #define GET_UNSIGNED_NUMBER(num) \
1021 while (ISDIGIT (c)) \
1025 num = num * 10 + c - '0'; \
1033 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1035 #define IS_CHAR_CLASS(string) \
1036 (STREQ (string, "alpha") || STREQ (string, "upper") \
1037 || STREQ (string, "lower") || STREQ (string, "digit") \
1038 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1039 || STREQ (string, "space") || STREQ (string, "print") \
1040 || STREQ (string, "punct") || STREQ (string, "graph") \
1041 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1043 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1044 Returns one of error codes defined in `regex.h', or zero for success.
1046 Assumes the `allocated' (and perhaps `buffer') and `translate'
1047 fields are set in BUFP on entry.
1049 If it succeeds, results are put in BUFP (if it returns an error, the
1050 contents of BUFP are undefined):
1051 `buffer' is the compiled pattern;
1052 `syntax' is set to SYNTAX;
1053 `used' is set to the length of the compiled pattern;
1054 `fastmap_accurate' is zero;
1055 `re_nsub' is the number of subexpressions in PATTERN;
1056 `not_bol' and `not_eol' are zero;
1058 The `fastmap' and `newline_anchor' fields are neither
1059 examined nor set. */
1061 static reg_errcode_t
1062 regex_compile (pattern
, size
, syntax
, bufp
)
1063 const char *pattern
;
1065 reg_syntax_t syntax
;
1066 struct re_pattern_buffer
*bufp
;
1068 /* We fetch characters from PATTERN here. Even though PATTERN is
1069 `char *' (i.e., signed), we declare these variables as unsigned, so
1070 they can be reliably used as array indices. */
1071 register unsigned char c
, c1
;
1073 /* A random tempory spot in PATTERN. */
1076 /* Points to the end of the buffer, where we should append. */
1077 register unsigned char *b
;
1079 /* Keeps track of unclosed groups. */
1080 compile_stack_type compile_stack
;
1082 /* Points to the current (ending) position in the pattern. */
1083 const char *p
= pattern
;
1084 const char *pend
= pattern
+ size
;
1086 /* How to translate the characters in the pattern. */
1087 char *translate
= bufp
->translate
;
1089 /* Address of the count-byte of the most recently inserted `exactn'
1090 command. This makes it possible to tell if a new exact-match
1091 character can be added to that command or if the character requires
1092 a new `exactn' command. */
1093 unsigned char *pending_exact
= 0;
1095 /* Address of start of the most recently finished expression.
1096 This tells, e.g., postfix * where to find the start of its
1097 operand. Reset at the beginning of groups and alternatives. */
1098 unsigned char *laststart
= 0;
1100 /* Address of beginning of regexp, or inside of last group. */
1101 unsigned char *begalt
;
1103 /* Place in the uncompiled pattern (i.e., the {) to
1104 which to go back if the interval is invalid. */
1105 const char *beg_interval
;
1107 /* Address of the place where a forward jump should go to the end of
1108 the containing expression. Each alternative of an `or' -- except the
1109 last -- ends with a forward jump of this sort. */
1110 unsigned char *fixup_alt_jump
= 0;
1112 /* Counts open-groups as they are encountered. Remembered for the
1113 matching close-group on the compile stack, so the same register
1114 number is put in the stop_memory as the start_memory. */
1115 regnum_t regnum
= 0;
1118 DEBUG_PRINT1 ("\nCompiling pattern: ");
1121 unsigned debug_count
;
1123 for (debug_count
= 0; debug_count
< size
; debug_count
++)
1124 printchar (pattern
[debug_count
]);
1129 /* Initialize the compile stack. */
1130 compile_stack
.stack
= TALLOC (INIT_COMPILE_STACK_SIZE
, compile_stack_elt_t
);
1131 if (compile_stack
.stack
== NULL
)
1134 compile_stack
.size
= INIT_COMPILE_STACK_SIZE
;
1135 compile_stack
.avail
= 0;
1137 /* Initialize the pattern buffer. */
1138 bufp
->syntax
= syntax
;
1139 bufp
->fastmap_accurate
= 0;
1140 bufp
->not_bol
= bufp
->not_eol
= 0;
1142 /* Set `used' to zero, so that if we return an error, the pattern
1143 printer (for debugging) will think there's no pattern. We reset it
1147 /* Always count groups, whether or not bufp->no_sub is set. */
1150 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1151 /* Initialize the syntax table. */
1152 init_syntax_once ();
1155 if (bufp
->allocated
== 0)
1158 { /* If zero allocated, but buffer is non-null, try to realloc
1159 enough space. This loses if buffer's address is bogus, but
1160 that is the user's responsibility. */
1161 RETALLOC (bufp
->buffer
, INIT_BUF_SIZE
, unsigned char);
1164 { /* Caller did not allocate a buffer. Do it for them. */
1165 bufp
->buffer
= TALLOC (INIT_BUF_SIZE
, unsigned char);
1167 if (!bufp
->buffer
) return REG_ESPACE
;
1169 bufp
->allocated
= INIT_BUF_SIZE
;
1172 begalt
= b
= bufp
->buffer
;
1174 /* Loop through the uncompiled pattern until we're at the end. */
1183 if ( /* If at start of pattern, it's an operator. */
1185 /* If context independent, it's an operator. */
1186 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1187 /* Otherwise, depends on what's come before. */
1188 || at_begline_loc_p (pattern
, p
, syntax
))
1198 if ( /* If at end of pattern, it's an operator. */
1200 /* If context independent, it's an operator. */
1201 || syntax
& RE_CONTEXT_INDEP_ANCHORS
1202 /* Otherwise, depends on what's next. */
1203 || at_endline_loc_p (p
, pend
, syntax
))
1213 if ((syntax
& RE_BK_PLUS_QM
)
1214 || (syntax
& RE_LIMITED_OPS
))
1218 /* If there is no previous pattern... */
1221 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1223 else if (!(syntax
& RE_CONTEXT_INDEP_OPS
))
1228 /* Are we optimizing this jump? */
1229 boolean keep_string_p
= false;
1231 /* 1 means zero (many) matches is allowed. */
1232 char zero_times_ok
= 0, many_times_ok
= 0;
1234 /* If there is a sequence of repetition chars, collapse it
1235 down to just one (the right one). We can't combine
1236 interval operators with these because of, e.g., `a{2}*',
1237 which should only match an even number of `a's. */
1241 zero_times_ok
|= c
!= '+';
1242 many_times_ok
|= c
!= '?';
1250 || (!(syntax
& RE_BK_PLUS_QM
) && (c
== '+' || c
== '?')))
1253 else if (syntax
& RE_BK_PLUS_QM
&& c
== '\\')
1255 if (p
== pend
) return REG_EESCAPE
;
1258 if (!(c1
== '+' || c1
== '?'))
1273 /* If we get here, we found another repeat character. */
1276 /* Star, etc. applied to an empty pattern is equivalent
1277 to an empty pattern. */
1281 /* Now we know whether or not zero matches is allowed
1282 and also whether or not two or more matches is allowed. */
1284 { /* More than one repetition is allowed, so put in at the
1285 end a backward relative jump from `b' to before the next
1286 jump we're going to put in below (which jumps from
1287 laststart to after this jump).
1289 But if we are at the `*' in the exact sequence `.*\n',
1290 insert an unconditional jump backwards to the .,
1291 instead of the beginning of the loop. This way we only
1292 push a failure point once, instead of every time
1293 through the loop. */
1294 assert (p
- 1 > pattern
);
1296 /* Allocate the space for the jump. */
1297 GET_BUFFER_SPACE (3);
1299 /* We know we are not at the first character of the pattern,
1300 because laststart was nonzero. And we've already
1301 incremented `p', by the way, to be the character after
1302 the `*'. Do we have to do something analogous here
1303 for null bytes, because of RE_DOT_NOT_NULL? */
1304 if (TRANSLATE (*(p
- 2)) == TRANSLATE ('.')
1305 && p
< pend
&& TRANSLATE (*p
) == TRANSLATE ('\n')
1306 && !(syntax
& RE_DOT_NEWLINE
))
1307 { /* We have .*\n. */
1308 STORE_JUMP (jump
, b
, laststart
);
1309 keep_string_p
= true;
1312 /* Anything else. */
1313 STORE_JUMP (maybe_pop_jump
, b
, laststart
- 3);
1315 /* We've added more stuff to the buffer. */
1319 /* On failure, jump from laststart to b + 3, which will be the
1320 end of the buffer after this jump is inserted. */
1321 GET_BUFFER_SPACE (3);
1322 INSERT_JUMP (keep_string_p
? on_failure_keep_string_jump
1330 /* At least one repetition is required, so insert a
1331 `dummy_failure_jump' before the initial
1332 `on_failure_jump' instruction of the loop. This
1333 effects a skip over that instruction the first time
1334 we hit that loop. */
1335 GET_BUFFER_SPACE (3);
1336 INSERT_JUMP (dummy_failure_jump
, laststart
, laststart
+ 6);
1351 boolean had_char_class
= false;
1353 if (p
== pend
) return REG_EBRACK
;
1355 /* Ensure that we have enough space to push a charset: the
1356 opcode, the length count, and the bitset; 34 bytes in all. */
1357 GET_BUFFER_SPACE (34);
1361 /* We test `*p == '^' twice, instead of using an if
1362 statement, so we only need one BUF_PUSH. */
1363 BUF_PUSH (*p
== '^' ? charset_not
: charset
);
1367 /* Remember the first position in the bracket expression. */
1370 /* Push the number of bytes in the bitmap. */
1371 BUF_PUSH ((1 << BYTEWIDTH
) / BYTEWIDTH
);
1373 /* Clear the whole map. */
1374 bzero (b
, (1 << BYTEWIDTH
) / BYTEWIDTH
);
1376 /* charset_not matches newline according to a syntax bit. */
1377 if ((re_opcode_t
) b
[-2] == charset_not
1378 && (syntax
& RE_HAT_LISTS_NOT_NEWLINE
))
1379 SET_LIST_BIT ('\n');
1381 /* Read in characters and ranges, setting map bits. */
1384 if (p
== pend
) return REG_EBRACK
;
1388 /* \ might escape characters inside [...] and [^...]. */
1389 if ((syntax
& RE_BACKSLASH_ESCAPE_IN_LISTS
) && c
== '\\')
1391 if (p
== pend
) return REG_EESCAPE
;
1398 /* Could be the end of the bracket expression. If it's
1399 not (i.e., when the bracket expression is `[]' so
1400 far), the ']' character bit gets set way below. */
1401 if (c
== ']' && p
!= p1
+ 1)
1404 /* Look ahead to see if it's a range when the last thing
1405 was a character class. */
1406 if (had_char_class
&& c
== '-' && *p
!= ']')
1409 /* Look ahead to see if it's a range when the last thing
1410 was a character: if this is a hyphen not at the
1411 beginning or the end of a list, then it's the range
1414 && !(p
- 2 >= pattern
&& p
[-2] == '[')
1415 && !(p
- 3 >= pattern
&& p
[-3] == '[' && p
[-2] == '^')
1419 = compile_range (&p
, pend
, translate
, syntax
, b
);
1420 if (ret
!= REG_NOERROR
) return ret
;
1423 else if (p
[0] == '-' && p
[1] != ']')
1424 { /* This handles ranges made up of characters only. */
1427 /* Move past the `-'. */
1430 ret
= compile_range (&p
, pend
, translate
, syntax
, b
);
1431 if (ret
!= REG_NOERROR
) return ret
;
1434 /* See if we're at the beginning of a possible character
1437 else if (syntax
& RE_CHAR_CLASSES
&& c
== '[' && *p
== ':')
1438 { /* Leave room for the null. */
1439 char str
[CHAR_CLASS_MAX_LENGTH
+ 1];
1444 /* If pattern is `[[:'. */
1445 if (p
== pend
) return REG_EBRACK
;
1450 if (c
== ':' || c
== ']' || p
== pend
1451 || c1
== CHAR_CLASS_MAX_LENGTH
)
1457 /* If isn't a word bracketed by `[:' and:`]':
1458 undo the ending character, the letters, and leave
1459 the leading `:' and `[' (but set bits for them). */
1460 if (c
== ':' && *p
== ']')
1463 boolean is_alnum
= STREQ (str
, "alnum");
1464 boolean is_alpha
= STREQ (str
, "alpha");
1465 boolean is_blank
= STREQ (str
, "blank");
1466 boolean is_cntrl
= STREQ (str
, "cntrl");
1467 boolean is_digit
= STREQ (str
, "digit");
1468 boolean is_graph
= STREQ (str
, "graph");
1469 boolean is_lower
= STREQ (str
, "lower");
1470 boolean is_print
= STREQ (str
, "print");
1471 boolean is_punct
= STREQ (str
, "punct");
1472 boolean is_space
= STREQ (str
, "space");
1473 boolean is_upper
= STREQ (str
, "upper");
1474 boolean is_xdigit
= STREQ (str
, "xdigit");
1476 if (!IS_CHAR_CLASS (str
)) return REG_ECTYPE
;
1478 /* Throw away the ] at the end of the character
1482 if (p
== pend
) return REG_EBRACK
;
1484 for (ch
= 0; ch
< 1 << BYTEWIDTH
; ch
++)
1486 if ( (is_alnum
&& ISALNUM (ch
))
1487 || (is_alpha
&& ISALPHA (ch
))
1488 || (is_blank
&& ISBLANK (ch
))
1489 || (is_cntrl
&& ISCNTRL (ch
))
1490 || (is_digit
&& ISDIGIT (ch
))
1491 || (is_graph
&& ISGRAPH (ch
))
1492 || (is_lower
&& ISLOWER (ch
))
1493 || (is_print
&& ISPRINT (ch
))
1494 || (is_punct
&& ISPUNCT (ch
))
1495 || (is_space
&& ISSPACE (ch
))
1496 || (is_upper
&& ISUPPER (ch
))
1497 || (is_xdigit
&& ISXDIGIT (ch
)))
1500 had_char_class
= true;
1509 had_char_class
= false;
1514 had_char_class
= false;
1519 /* Discard any (non)matching list bytes that are all 0 at the
1520 end of the map. Decrease the map-length byte too. */
1521 while ((int) b
[-1] > 0 && b
[b
[-1] - 1] == 0)
1529 if (syntax
& RE_NO_BK_PARENS
)
1536 if (syntax
& RE_NO_BK_PARENS
)
1543 if (syntax
& RE_NEWLINE_ALT
)
1550 if (syntax
& RE_NO_BK_VBAR
)
1557 if (syntax
& RE_INTERVALS
&& syntax
& RE_NO_BK_BRACES
)
1558 goto handle_interval
;
1564 if (p
== pend
) return REG_EESCAPE
;
1566 /* Do not translate the character after the \, so that we can
1567 distinguish, e.g., \B from \b, even if we normally would
1568 translate, e.g., B to b. */
1574 if (syntax
& RE_NO_BK_PARENS
)
1575 goto normal_backslash
;
1581 if (COMPILE_STACK_FULL
)
1583 RETALLOC (compile_stack
.stack
, compile_stack
.size
<< 1,
1584 compile_stack_elt_t
);
1585 if (compile_stack
.stack
== NULL
) return REG_ESPACE
;
1587 compile_stack
.size
<<= 1;
1590 /* These are the values to restore when we hit end of this
1591 group. They are all relative offsets, so that if the
1592 whole pattern moves because of realloc, they will still
1594 COMPILE_STACK_TOP
.begalt_offset
= begalt
- bufp
->buffer
;
1595 COMPILE_STACK_TOP
.fixup_alt_jump
1596 = fixup_alt_jump
? fixup_alt_jump
- bufp
->buffer
+ 1 : 0;
1597 COMPILE_STACK_TOP
.laststart_offset
= b
- bufp
->buffer
;
1598 COMPILE_STACK_TOP
.regnum
= regnum
;
1600 /* We will eventually replace the 0 with the number of
1601 groups inner to this one. But do not push a
1602 start_memory for groups beyond the last one we can
1603 represent in the compiled pattern. */
1604 if (regnum
<= MAX_REGNUM
)
1606 COMPILE_STACK_TOP
.inner_group_offset
= b
- bufp
->buffer
+ 2;
1607 BUF_PUSH_3 (start_memory
, regnum
, 0);
1610 compile_stack
.avail
++;
1619 if (syntax
& RE_NO_BK_PARENS
) goto normal_backslash
;
1621 if (COMPILE_STACK_EMPTY
)
1622 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
1623 goto normal_backslash
;
1629 { /* Push a dummy failure point at the end of the
1630 alternative for a possible future
1631 `pop_failure_jump' to pop. See comments at
1632 `push_dummy_failure' in `re_match_2'. */
1633 BUF_PUSH (push_dummy_failure
);
1635 /* We allocated space for this jump when we assigned
1636 to `fixup_alt_jump', in the `handle_alt' case below. */
1637 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
- 1);
1640 /* See similar code for backslashed left paren above. */
1641 if (COMPILE_STACK_EMPTY
)
1642 if (syntax
& RE_UNMATCHED_RIGHT_PAREN_ORD
)
1647 /* Since we just checked for an empty stack above, this
1648 ``can't happen''. */
1649 assert (compile_stack
.avail
!= 0);
1651 /* We don't just want to restore into `regnum', because
1652 later groups should continue to be numbered higher,
1653 as in `(ab)c(de)' -- the second group is #2. */
1654 regnum_t this_group_regnum
;
1656 compile_stack
.avail
--;
1657 begalt
= bufp
->buffer
+ COMPILE_STACK_TOP
.begalt_offset
;
1659 = COMPILE_STACK_TOP
.fixup_alt_jump
1660 ? bufp
->buffer
+ COMPILE_STACK_TOP
.fixup_alt_jump
- 1
1662 laststart
= bufp
->buffer
+ COMPILE_STACK_TOP
.laststart_offset
;
1663 this_group_regnum
= COMPILE_STACK_TOP
.regnum
;
1665 /* We're at the end of the group, so now we know how many
1666 groups were inside this one. */
1667 if (this_group_regnum
<= MAX_REGNUM
)
1669 unsigned char *inner_group_loc
1670 = bufp
->buffer
+ COMPILE_STACK_TOP
.inner_group_offset
;
1672 *inner_group_loc
= regnum
- this_group_regnum
;
1673 BUF_PUSH_3 (stop_memory
, this_group_regnum
,
1674 regnum
- this_group_regnum
);
1680 case '|': /* `\|'. */
1681 if (syntax
& RE_LIMITED_OPS
|| syntax
& RE_NO_BK_VBAR
)
1682 goto normal_backslash
;
1684 if (syntax
& RE_LIMITED_OPS
)
1687 /* Insert before the previous alternative a jump which
1688 jumps to this alternative if the former fails. */
1689 GET_BUFFER_SPACE (3);
1690 INSERT_JUMP (on_failure_jump
, begalt
, b
+ 6);
1694 /* The alternative before this one has a jump after it
1695 which gets executed if it gets matched. Adjust that
1696 jump so it will jump to this alternative's analogous
1697 jump (put in below, which in turn will jump to the next
1698 (if any) alternative's such jump, etc.). The last such
1699 jump jumps to the correct final destination. A picture:
1705 If we are at `b', then fixup_alt_jump right now points to a
1706 three-byte space after `a'. We'll put in the jump, set
1707 fixup_alt_jump to right after `b', and leave behind three
1708 bytes which we'll fill in when we get to after `c'. */
1711 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
1713 /* Mark and leave space for a jump after this alternative,
1714 to be filled in later either by next alternative or
1715 when know we're at the end of a series of alternatives. */
1717 GET_BUFFER_SPACE (3);
1726 /* If \{ is a literal. */
1727 if (!(syntax
& RE_INTERVALS
)
1728 /* If we're at `\{' and it's not the open-interval
1730 || ((syntax
& RE_INTERVALS
) && (syntax
& RE_NO_BK_BRACES
))
1731 || (p
- 2 == pattern
&& p
== pend
))
1732 goto normal_backslash
;
1736 /* If got here, then the syntax allows intervals. */
1738 /* At least (most) this many matches must be made. */
1739 int lower_bound
= -1, upper_bound
= -1;
1741 beg_interval
= p
- 1;
1745 if (syntax
& RE_NO_BK_BRACES
)
1746 goto unfetch_interval
;
1751 GET_UNSIGNED_NUMBER (lower_bound
);
1755 GET_UNSIGNED_NUMBER (upper_bound
);
1756 if (upper_bound
< 0) upper_bound
= RE_DUP_MAX
;
1759 /* Interval such as `{1}' => match exactly once. */
1760 upper_bound
= lower_bound
;
1762 if (lower_bound
< 0 || upper_bound
> RE_DUP_MAX
1763 || lower_bound
> upper_bound
)
1765 if (syntax
& RE_NO_BK_BRACES
)
1766 goto unfetch_interval
;
1771 if (!(syntax
& RE_NO_BK_BRACES
))
1773 if (c
!= '\\') return REG_EBRACE
;
1780 if (syntax
& RE_NO_BK_BRACES
)
1781 goto unfetch_interval
;
1786 /* We just parsed a valid interval. */
1788 /* If it's invalid to have no preceding re. */
1791 if (syntax
& RE_CONTEXT_INVALID_OPS
)
1793 else if (syntax
& RE_CONTEXT_INDEP_OPS
)
1796 goto unfetch_interval
;
1799 /* If the upper bound is zero, don't want to succeed at
1800 all; jump from `laststart' to `b + 3', which will be
1801 the end of the buffer after we insert the jump. */
1802 if (upper_bound
== 0)
1804 GET_BUFFER_SPACE (3);
1805 INSERT_JUMP (jump
, laststart
, b
+ 3);
1809 /* Otherwise, we have a nontrivial interval. When
1810 we're all done, the pattern will look like:
1811 set_number_at <jump count> <upper bound>
1812 set_number_at <succeed_n count> <lower bound>
1813 succeed_n <after jump addr> <succed_n count>
1815 jump_n <succeed_n addr> <jump count>
1816 (The upper bound and `jump_n' are omitted if
1817 `upper_bound' is 1, though.) */
1819 { /* If the upper bound is > 1, we need to insert
1820 more at the end of the loop. */
1821 unsigned nbytes
= 10 + (upper_bound
> 1) * 10;
1823 GET_BUFFER_SPACE (nbytes
);
1825 /* Initialize lower bound of the `succeed_n', even
1826 though it will be set during matching by its
1827 attendant `set_number_at' (inserted next),
1828 because `re_compile_fastmap' needs to know.
1829 Jump to the `jump_n' we might insert below. */
1830 INSERT_JUMP2 (succeed_n
, laststart
,
1831 b
+ 5 + (upper_bound
> 1) * 5,
1835 /* Code to initialize the lower bound. Insert
1836 before the `succeed_n'. The `5' is the last two
1837 bytes of this `set_number_at', plus 3 bytes of
1838 the following `succeed_n'. */
1839 insert_op2 (set_number_at
, laststart
, 5, lower_bound
, b
);
1842 if (upper_bound
> 1)
1843 { /* More than one repetition is allowed, so
1844 append a backward jump to the `succeed_n'
1845 that starts this interval.
1847 When we've reached this during matching,
1848 we'll have matched the interval once, so
1849 jump back only `upper_bound - 1' times. */
1850 STORE_JUMP2 (jump_n
, b
, laststart
+ 5,
1854 /* The location we want to set is the second
1855 parameter of the `jump_n'; that is `b-2' as
1856 an absolute address. `laststart' will be
1857 the `set_number_at' we're about to insert;
1858 `laststart+3' the number to set, the source
1859 for the relative address. But we are
1860 inserting into the middle of the pattern --
1861 so everything is getting moved up by 5.
1862 Conclusion: (b - 2) - (laststart + 3) + 5,
1863 i.e., b - laststart.
1865 We insert this at the beginning of the loop
1866 so that if we fail during matching, we'll
1867 reinitialize the bounds. */
1868 insert_op2 (set_number_at
, laststart
, b
- laststart
,
1869 upper_bound
- 1, b
);
1874 beg_interval
= NULL
;
1879 /* If an invalid interval, match the characters as literals. */
1880 assert (beg_interval
);
1882 beg_interval
= NULL
;
1884 /* normal_char and normal_backslash need `c'. */
1887 if (!(syntax
& RE_NO_BK_BRACES
))
1889 if (p
> pattern
&& p
[-1] == '\\')
1890 goto normal_backslash
;
1895 /* There is no way to specify the before_dot and after_dot
1896 operators. rms says this is ok. --karl */
1904 BUF_PUSH_2 (syntaxspec
, syntax_spec_code
[c
]);
1910 BUF_PUSH_2 (notsyntaxspec
, syntax_spec_code
[c
]);
1917 BUF_PUSH (wordchar
);
1923 BUF_PUSH (notwordchar
);
1936 BUF_PUSH (wordbound
);
1940 BUF_PUSH (notwordbound
);
1951 case '1': case '2': case '3': case '4': case '5':
1952 case '6': case '7': case '8': case '9':
1953 if (syntax
& RE_NO_BK_REFS
)
1961 /* Can't back reference to a subexpression if inside of it. */
1962 if (group_in_compile_stack (compile_stack
, c1
))
1966 BUF_PUSH_2 (duplicate
, c1
);
1972 if (syntax
& RE_BK_PLUS_QM
)
1975 goto normal_backslash
;
1979 /* You might think it would be useful for \ to mean
1980 not to translate; but if we don't translate it
1981 it will never match anything. */
1989 /* Expects the character in `c'. */
1991 /* If no exactn currently being built. */
1994 /* If last exactn not at current position. */
1995 || pending_exact
+ *pending_exact
+ 1 != b
1997 /* We have only one byte following the exactn for the count. */
1998 || *pending_exact
== (1 << BYTEWIDTH
) - 1
2000 /* If followed by a repetition operator. */
2001 || *p
== '*' || *p
== '^'
2002 || ((syntax
& RE_BK_PLUS_QM
)
2003 ? *p
== '\\' && (p
[1] == '+' || p
[1] == '?')
2004 : (*p
== '+' || *p
== '?'))
2005 || ((syntax
& RE_INTERVALS
)
2006 && ((syntax
& RE_NO_BK_BRACES
)
2008 : (p
[0] == '\\' && p
[1] == '{'))))
2010 /* Start building a new exactn. */
2014 BUF_PUSH_2 (exactn
, 0);
2015 pending_exact
= b
- 1;
2022 } /* while p != pend */
2025 /* Through the pattern now. */
2028 STORE_JUMP (jump_past_alt
, fixup_alt_jump
, b
);
2030 if (!COMPILE_STACK_EMPTY
)
2033 free (compile_stack
.stack
);
2035 /* We have succeeded; set the length of the buffer. */
2036 bufp
->used
= b
- bufp
->buffer
;
2041 DEBUG_PRINT1 ("\nCompiled pattern: ");
2042 print_compiled_pattern (bufp
);
2047 } /* regex_compile */
2049 /* Subroutines for `regex_compile'. */
2051 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2054 store_op1 (op
, loc
, arg
)
2059 *loc
= (unsigned char) op
;
2060 STORE_NUMBER (loc
+ 1, arg
);
2064 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2067 store_op2 (op
, loc
, arg1
, arg2
)
2072 *loc
= (unsigned char) op
;
2073 STORE_NUMBER (loc
+ 1, arg1
);
2074 STORE_NUMBER (loc
+ 3, arg2
);
2078 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2079 for OP followed by two-byte integer parameter ARG. */
2082 insert_op1 (op
, loc
, arg
, end
)
2088 register unsigned char *pfrom
= end
;
2089 register unsigned char *pto
= end
+ 3;
2091 while (pfrom
!= loc
)
2094 store_op1 (op
, loc
, arg
);
2098 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2101 insert_op2 (op
, loc
, arg1
, arg2
, end
)
2107 register unsigned char *pfrom
= end
;
2108 register unsigned char *pto
= end
+ 5;
2110 while (pfrom
!= loc
)
2113 store_op2 (op
, loc
, arg1
, arg2
);
2117 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2118 after an alternative or a begin-subexpression. We assume there is at
2119 least one character before the ^. */
2122 at_begline_loc_p (pattern
, p
, syntax
)
2123 const char *pattern
, *p
;
2124 reg_syntax_t syntax
;
2126 const char *prev
= p
- 2;
2127 boolean prev_prev_backslash
= prev
> pattern
&& prev
[-1] == '\\';
2130 /* After a subexpression? */
2131 (*prev
== '(' && (syntax
& RE_NO_BK_PARENS
|| prev_prev_backslash
))
2132 /* After an alternative? */
2133 || (*prev
== '|' && (syntax
& RE_NO_BK_VBAR
|| prev_prev_backslash
));
2137 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2138 at least one character after the $, i.e., `P < PEND'. */
2141 at_endline_loc_p (p
, pend
, syntax
)
2142 const char *p
, *pend
;
2145 const char *next
= p
;
2146 boolean next_backslash
= *next
== '\\';
2147 const char *next_next
= p
+ 1 < pend
? p
+ 1 : NULL
;
2150 /* Before a subexpression? */
2151 (syntax
& RE_NO_BK_PARENS
? *next
== ')'
2152 : next_backslash
&& next_next
&& *next_next
== ')')
2153 /* Before an alternative? */
2154 || (syntax
& RE_NO_BK_VBAR
? *next
== '|'
2155 : next_backslash
&& next_next
&& *next_next
== '|');
2159 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2160 false if it's not. */
2163 group_in_compile_stack (compile_stack
, regnum
)
2164 compile_stack_type compile_stack
;
2169 for (this_element
= compile_stack
.avail
- 1;
2172 if (compile_stack
.stack
[this_element
].regnum
== regnum
)
2179 /* Read the ending character of a range (in a bracket expression) from the
2180 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2181 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2182 Then we set the translation of all bits between the starting and
2183 ending characters (inclusive) in the compiled pattern B.
2185 Return an error code.
2187 We use these short variable names so we can use the same macros as
2188 `regex_compile' itself. */
2190 static reg_errcode_t
2191 compile_range (p_ptr
, pend
, translate
, syntax
, b
)
2192 const char **p_ptr
, *pend
;
2194 reg_syntax_t syntax
;
2199 const char *p
= *p_ptr
;
2200 int range_start
, range_end
;
2205 /* Even though the pattern is a signed `char *', we need to fetch
2206 with unsigned char *'s; if the high bit of the pattern character
2207 is set, the range endpoints will be negative if we fetch using a
2210 We also want to fetch the endpoints without translating them; the
2211 appropriate translation is done in the bit-setting loop below. */
2212 range_start
= ((unsigned char *) p
)[-2];
2213 range_end
= ((unsigned char *) p
)[0];
2215 /* Have to increment the pointer into the pattern string, so the
2216 caller isn't still at the ending character. */
2219 /* If the start is after the end, the range is empty. */
2220 if (range_start
> range_end
)
2221 return syntax
& RE_NO_EMPTY_RANGES
? REG_ERANGE
: REG_NOERROR
;
2223 /* Here we see why `this_char' has to be larger than an `unsigned
2224 char' -- the range is inclusive, so if `range_end' == 0xff
2225 (assuming 8-bit characters), we would otherwise go into an infinite
2226 loop, since all characters <= 0xff. */
2227 for (this_char
= range_start
; this_char
<= range_end
; this_char
++)
2229 SET_LIST_BIT (TRANSLATE (this_char
));
2235 /* Failure stack declarations and macros; both re_compile_fastmap and
2236 re_match_2 use a failure stack. These have to be macros because of
2240 /* Number of failure points for which to initially allocate space
2241 when matching. If this number is exceeded, we allocate more
2242 space, so it is not a hard limit. */
2243 #ifndef INIT_FAILURE_ALLOC
2244 #define INIT_FAILURE_ALLOC 5
2247 /* Roughly the maximum number of failure points on the stack. Would be
2248 exactly that if always used MAX_FAILURE_SPACE each time we failed.
2249 This is a variable only so users of regex can assign to it; we never
2250 change it ourselves. */
2251 int re_max_failures
= 2000;
2253 typedef const unsigned char *fail_stack_elt_t
;
2257 fail_stack_elt_t
*stack
;
2259 unsigned avail
; /* Offset of next open position. */
2262 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
2263 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
2264 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
2265 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
2268 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
2270 #define INIT_FAIL_STACK() \
2272 fail_stack.stack = (fail_stack_elt_t *) \
2273 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
2275 if (fail_stack.stack == NULL) \
2278 fail_stack.size = INIT_FAILURE_ALLOC; \
2279 fail_stack.avail = 0; \
2283 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
2285 Return 1 if succeeds, and 0 if either ran out of memory
2286 allocating space for it or it was already too large.
2288 REGEX_REALLOCATE requires `destination' be declared. */
2290 #define DOUBLE_FAIL_STACK(fail_stack) \
2291 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
2293 : ((fail_stack).stack = (fail_stack_elt_t *) \
2294 REGEX_REALLOCATE ((fail_stack).stack, \
2295 (fail_stack).size * sizeof (fail_stack_elt_t), \
2296 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
2298 (fail_stack).stack == NULL \
2300 : ((fail_stack).size <<= 1, \
2304 /* Push PATTERN_OP on FAIL_STACK.
2306 Return 1 if was able to do so and 0 if ran out of memory allocating
2308 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
2309 ((FAIL_STACK_FULL () \
2310 && !DOUBLE_FAIL_STACK (fail_stack)) \
2312 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
2315 /* This pushes an item onto the failure stack. Must be a four-byte
2316 value. Assumes the variable `fail_stack'. Probably should only
2317 be called from within `PUSH_FAILURE_POINT'. */
2318 #define PUSH_FAILURE_ITEM(item) \
2319 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
2321 /* The complement operation. Assumes `fail_stack' is nonempty. */
2322 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
2324 /* Used to omit pushing failure point id's when we're not debugging. */
2326 #define DEBUG_PUSH PUSH_FAILURE_ITEM
2327 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
2329 #define DEBUG_PUSH(item)
2330 #define DEBUG_POP(item_addr)
2334 /* Push the information about the state we will need
2335 if we ever fail back to it.
2337 Requires variables fail_stack, regstart, regend, reg_info, and
2338 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
2341 Does `return FAILURE_CODE' if runs out of memory. */
2343 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
2345 char *destination; \
2346 /* Must be int, so when we don't save any registers, the arithmetic \
2347 of 0 + -1 isn't done as unsigned. */ \
2350 DEBUG_STATEMENT (failure_id++); \
2351 DEBUG_STATEMENT (nfailure_points_pushed++); \
2352 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
2353 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
2354 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
2356 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
2357 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
2359 /* Ensure we have enough space allocated for what we will push. */ \
2360 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
2362 if (!DOUBLE_FAIL_STACK (fail_stack)) \
2363 return failure_code; \
2365 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
2366 (fail_stack).size); \
2367 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
2370 /* Push the info, starting with the registers. */ \
2371 DEBUG_PRINT1 ("\n"); \
2373 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
2376 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
2377 DEBUG_STATEMENT (num_regs_pushed++); \
2379 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
2380 PUSH_FAILURE_ITEM (regstart[this_reg]); \
2382 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
2383 PUSH_FAILURE_ITEM (regend[this_reg]); \
2385 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
2386 DEBUG_PRINT2 (" match_null=%d", \
2387 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
2388 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
2389 DEBUG_PRINT2 (" matched_something=%d", \
2390 MATCHED_SOMETHING (reg_info[this_reg])); \
2391 DEBUG_PRINT2 (" ever_matched=%d", \
2392 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
2393 DEBUG_PRINT1 ("\n"); \
2394 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
2397 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
2398 PUSH_FAILURE_ITEM (lowest_active_reg); \
2400 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
2401 PUSH_FAILURE_ITEM (highest_active_reg); \
2403 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
2404 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
2405 PUSH_FAILURE_ITEM (pattern_place); \
2407 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
2408 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
2410 DEBUG_PRINT1 ("'\n"); \
2411 PUSH_FAILURE_ITEM (string_place); \
2413 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
2414 DEBUG_PUSH (failure_id); \
2417 /* This is the number of items that are pushed and popped on the stack
2418 for each register. */
2419 #define NUM_REG_ITEMS 3
2421 /* Individual items aside from the registers. */
2423 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
2425 #define NUM_NONREG_ITEMS 4
2428 /* We push at most this many items on the stack. */
2429 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
2431 /* We actually push this many items. */
2432 #define NUM_FAILURE_ITEMS \
2433 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
2436 /* How many items can still be added to the stack without overflowing it. */
2437 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
2440 /* Pops what PUSH_FAIL_STACK pushes.
2442 We restore into the parameters, all of which should be lvalues:
2443 STR -- the saved data position.
2444 PAT -- the saved pattern position.
2445 LOW_REG, HIGH_REG -- the highest and lowest active registers.
2446 REGSTART, REGEND -- arrays of string positions.
2447 REG_INFO -- array of information about each subexpression.
2449 Also assumes the variables `fail_stack' and (if debugging), `bufp',
2450 `pend', `string1', `size1', `string2', and `size2'. */
2452 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
2454 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
2456 const unsigned char *string_temp; \
2458 assert (!FAIL_STACK_EMPTY ()); \
2460 /* Remove failure points and point to how many regs pushed. */ \
2461 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
2462 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
2463 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
2465 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
2467 DEBUG_POP (&failure_id); \
2468 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
2470 /* If the saved string location is NULL, it came from an \
2471 on_failure_keep_string_jump opcode, and we want to throw away the \
2472 saved NULL, thus retaining our current position in the string. */ \
2473 string_temp = POP_FAILURE_ITEM (); \
2474 if (string_temp != NULL) \
2475 str = (const char *) string_temp; \
2477 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
2478 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
2479 DEBUG_PRINT1 ("'\n"); \
2481 pat = (unsigned char *) POP_FAILURE_ITEM (); \
2482 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
2483 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
2485 /* Restore register info. */ \
2486 high_reg = (unsigned) POP_FAILURE_ITEM (); \
2487 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
2489 low_reg = (unsigned) POP_FAILURE_ITEM (); \
2490 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
2492 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
2494 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
2496 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
2497 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
2499 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
2500 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
2502 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
2503 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
2506 DEBUG_STATEMENT (nfailure_points_popped++); \
2507 } /* POP_FAILURE_POINT */
2509 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2510 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2511 characters can start a string that matches the pattern. This fastmap
2512 is used by re_search to skip quickly over impossible starting points.
2514 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2515 area as BUFP->fastmap.
2517 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2520 Returns 0 if we succeed, -2 if an internal error. */
2523 re_compile_fastmap (bufp
)
2524 struct re_pattern_buffer
*bufp
;
2527 fail_stack_type fail_stack
;
2528 #ifndef REGEX_MALLOC
2531 /* We don't push any register information onto the failure stack. */
2532 unsigned num_regs
= 0;
2534 register char *fastmap
= bufp
->fastmap
;
2535 unsigned char *pattern
= bufp
->buffer
;
2536 unsigned long size
= bufp
->used
;
2537 const unsigned char *p
= pattern
;
2538 register unsigned char *pend
= pattern
+ size
;
2540 /* Assume that each path through the pattern can be null until
2541 proven otherwise. We set this false at the bottom of switch
2542 statement, to which we get only if a particular path doesn't
2543 match the empty string. */
2544 boolean path_can_be_null
= true;
2546 /* We aren't doing a `succeed_n' to begin with. */
2547 boolean succeed_n_p
= false;
2549 assert (fastmap
!= NULL
&& p
!= NULL
);
2552 bzero (fastmap
, 1 << BYTEWIDTH
); /* Assume nothing's valid. */
2553 bufp
->fastmap_accurate
= 1; /* It will be when we're done. */
2554 bufp
->can_be_null
= 0;
2556 while (p
!= pend
|| !FAIL_STACK_EMPTY ())
2560 bufp
->can_be_null
|= path_can_be_null
;
2562 /* Reset for next path. */
2563 path_can_be_null
= true;
2565 p
= fail_stack
.stack
[--fail_stack
.avail
];
2568 /* We should never be about to go beyond the end of the pattern. */
2571 #ifdef SWITCH_ENUM_BUG
2572 switch ((int) ((re_opcode_t
) *p
++))
2574 switch ((re_opcode_t
) *p
++)
2578 /* I guess the idea here is to simply not bother with a fastmap
2579 if a backreference is used, since it's too hard to figure out
2580 the fastmap for the corresponding group. Setting
2581 `can_be_null' stops `re_search_2' from using the fastmap, so
2582 that is all we do. */
2584 bufp
->can_be_null
= 1;
2588 /* Following are the cases which match a character. These end
2597 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2598 if (p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
)))
2604 /* Chars beyond end of map must be allowed. */
2605 for (j
= *p
* BYTEWIDTH
; j
< (1 << BYTEWIDTH
); j
++)
2608 for (j
= *p
++ * BYTEWIDTH
- 1; j
>= 0; j
--)
2609 if (!(p
[j
/ BYTEWIDTH
] & (1 << (j
% BYTEWIDTH
))))
2615 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2616 if (SYNTAX (j
) == Sword
)
2622 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2623 if (SYNTAX (j
) != Sword
)
2629 /* `.' matches anything ... */
2630 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2633 /* ... except perhaps newline. */
2634 if (!(bufp
->syntax
& RE_DOT_NEWLINE
))
2637 /* Return if we have already set `can_be_null'; if we have,
2638 then the fastmap is irrelevant. Something's wrong here. */
2639 else if (bufp
->can_be_null
)
2642 /* Otherwise, have to check alternative paths. */
2649 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2650 if (SYNTAX (j
) == (enum syntaxcode
) k
)
2657 for (j
= 0; j
< (1 << BYTEWIDTH
); j
++)
2658 if (SYNTAX (j
) != (enum syntaxcode
) k
)
2663 /* All cases after this match the empty string. These end with
2671 #endif /* not emacs */
2683 case push_dummy_failure
:
2688 case pop_failure_jump
:
2689 case maybe_pop_jump
:
2692 case dummy_failure_jump
:
2693 EXTRACT_NUMBER_AND_INCR (j
, p
);
2698 /* Jump backward implies we just went through the body of a
2699 loop and matched nothing. Opcode jumped to should be
2700 `on_failure_jump' or `succeed_n'. Just treat it like an
2701 ordinary jump. For a * loop, it has pushed its failure
2702 point already; if so, discard that as redundant. */
2703 if ((re_opcode_t
) *p
!= on_failure_jump
2704 && (re_opcode_t
) *p
!= succeed_n
)
2708 EXTRACT_NUMBER_AND_INCR (j
, p
);
2711 /* If what's on the stack is where we are now, pop it. */
2712 if (!FAIL_STACK_EMPTY ()
2713 && fail_stack
.stack
[fail_stack
.avail
- 1] == p
)
2719 case on_failure_jump
:
2720 case on_failure_keep_string_jump
:
2721 handle_on_failure_jump
:
2722 EXTRACT_NUMBER_AND_INCR (j
, p
);
2724 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2725 end of the pattern. We don't want to push such a point,
2726 since when we restore it above, entering the switch will
2727 increment `p' past the end of the pattern. We don't need
2728 to push such a point since we obviously won't find any more
2729 fastmap entries beyond `pend'. Such a pattern can match
2730 the null string, though. */
2733 if (!PUSH_PATTERN_OP (p
+ j
, fail_stack
))
2737 bufp
->can_be_null
= 1;
2741 EXTRACT_NUMBER_AND_INCR (k
, p
); /* Skip the n. */
2742 succeed_n_p
= false;
2749 /* Get to the number of times to succeed. */
2752 /* Increment p past the n for when k != 0. */
2753 EXTRACT_NUMBER_AND_INCR (k
, p
);
2757 succeed_n_p
= true; /* Spaghetti code alert. */
2758 goto handle_on_failure_jump
;
2775 abort (); /* We have listed all the cases. */
2778 /* Getting here means we have found the possible starting
2779 characters for one path of the pattern -- and that the empty
2780 string does not match. We need not follow this path further.
2781 Instead, look at the next alternative (remembered on the
2782 stack), or quit if no more. The test at the top of the loop
2783 does these things. */
2784 path_can_be_null
= false;
2788 /* Set `can_be_null' for the last path (also the first path, if the
2789 pattern is empty). */
2790 bufp
->can_be_null
|= path_can_be_null
;
2792 } /* re_compile_fastmap */
2794 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
2795 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
2796 this memory for recording register information. STARTS and ENDS
2797 must be allocated using the malloc library routine, and must each
2798 be at least NUM_REGS * sizeof (regoff_t) bytes long.
2800 If NUM_REGS == 0, then subsequent matches should allocate their own
2803 Unless this function is called, the first search or match using
2804 PATTERN_BUFFER will allocate its own register data, without
2805 freeing the old data. */
2808 re_set_registers (bufp
, regs
, num_regs
, starts
, ends
)
2809 struct re_pattern_buffer
*bufp
;
2810 struct re_registers
*regs
;
2812 regoff_t
*starts
, *ends
;
2816 bufp
->regs_allocated
= REGS_REALLOCATE
;
2817 regs
->num_regs
= num_regs
;
2818 regs
->start
= starts
;
2823 bufp
->regs_allocated
= REGS_UNALLOCATED
;
2825 regs
->start
= regs
->end
= (regoff_t
) 0;
2829 /* Searching routines. */
2831 /* Like re_search_2, below, but only one string is specified, and
2832 doesn't let you say where to stop matching. */
2835 re_search (bufp
, string
, size
, startpos
, range
, regs
)
2836 struct re_pattern_buffer
*bufp
;
2838 int size
, startpos
, range
;
2839 struct re_registers
*regs
;
2841 return re_search_2 (bufp
, NULL
, 0, string
, size
, startpos
, range
,
2846 /* Using the compiled pattern in BUFP->buffer, first tries to match the
2847 virtual concatenation of STRING1 and STRING2, starting first at index
2848 STARTPOS, then at STARTPOS + 1, and so on.
2850 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
2852 RANGE is how far to scan while trying to match. RANGE = 0 means try
2853 only at STARTPOS; in general, the last start tried is STARTPOS +
2856 In REGS, return the indices of the virtual concatenation of STRING1
2857 and STRING2 that matched the entire BUFP->buffer and its contained
2860 Do not consider matching one past the index STOP in the virtual
2861 concatenation of STRING1 and STRING2.
2863 We return either the position in the strings at which the match was
2864 found, -1 if no match, or -2 if error (such as failure
2868 re_search_2 (bufp
, string1
, size1
, string2
, size2
, startpos
, range
, regs
, stop
)
2869 struct re_pattern_buffer
*bufp
;
2870 const char *string1
, *string2
;
2874 struct re_registers
*regs
;
2878 register char *fastmap
= bufp
->fastmap
;
2879 register char *translate
= bufp
->translate
;
2880 int total_size
= size1
+ size2
;
2881 int endpos
= startpos
+ range
;
2883 /* Check for out-of-range STARTPOS. */
2884 if (startpos
< 0 || startpos
> total_size
)
2887 /* Fix up RANGE if it might eventually take us outside
2888 the virtual concatenation of STRING1 and STRING2. */
2890 range
= -1 - startpos
;
2891 else if (endpos
> total_size
)
2892 range
= total_size
- startpos
;
2894 /* If the search isn't to be a backwards one, don't waste time in a
2895 search for a pattern that must be anchored. */
2896 if (bufp
->used
> 0 && (re_opcode_t
) bufp
->buffer
[0] == begbuf
&& range
> 0)
2904 /* Update the fastmap now if not correct already. */
2905 if (fastmap
&& !bufp
->fastmap_accurate
)
2906 if (re_compile_fastmap (bufp
) == -2)
2909 /* Loop through the string, looking for a place to start matching. */
2912 /* If a fastmap is supplied, skip quickly over characters that
2913 cannot be the start of a match. If the pattern can match the
2914 null string, however, we don't need to skip characters; we want
2915 the first null string. */
2916 if (fastmap
&& startpos
< total_size
&& !bufp
->can_be_null
)
2918 if (range
> 0) /* Searching forwards. */
2920 register const char *d
;
2921 register int lim
= 0;
2924 if (startpos
< size1
&& startpos
+ range
>= size1
)
2925 lim
= range
- (size1
- startpos
);
2927 d
= (startpos
>= size1
? string2
- size1
: string1
) + startpos
;
2929 /* Written out as an if-else to avoid testing `translate'
2933 && !fastmap
[(unsigned char)
2934 translate
[(unsigned char) *d
++]])
2937 while (range
> lim
&& !fastmap
[(unsigned char) *d
++])
2940 startpos
+= irange
- range
;
2942 else /* Searching backwards. */
2944 register char c
= (size1
== 0 || startpos
>= size1
2945 ? string2
[startpos
- size1
]
2946 : string1
[startpos
]);
2948 if (!fastmap
[(unsigned char) TRANSLATE (c
)])
2953 /* If can't match the null string, and that's all we have left, fail. */
2954 if (range
>= 0 && startpos
== total_size
&& fastmap
2955 && !bufp
->can_be_null
)
2958 val
= re_match_2 (bufp
, string1
, size1
, string2
, size2
,
2959 startpos
, regs
, stop
);
2983 /* Declarations and macros for re_match_2. */
2985 static int bcmp_translate ();
2986 static boolean
alt_match_null_string_p (),
2987 common_op_match_null_string_p (),
2988 group_match_null_string_p ();
2990 /* Structure for per-register (a.k.a. per-group) information.
2991 This must not be longer than one word, because we push this value
2992 onto the failure stack. Other register information, such as the
2993 starting and ending positions (which are addresses), and the list of
2994 inner groups (which is a bits list) are maintained in separate
2997 We are making a (strictly speaking) nonportable assumption here: that
2998 the compiler will pack our bit fields into something that fits into
2999 the type of `word', i.e., is something that fits into one item on the
3003 fail_stack_elt_t word
;
3006 /* This field is one if this group can match the empty string,
3007 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
3008 #define MATCH_NULL_UNSET_VALUE 3
3009 unsigned match_null_string_p
: 2;
3010 unsigned is_active
: 1;
3011 unsigned matched_something
: 1;
3012 unsigned ever_matched_something
: 1;
3014 } register_info_type
;
3016 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
3017 #define IS_ACTIVE(R) ((R).bits.is_active)
3018 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
3019 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
3022 /* Call this when have matched a real character; it sets `matched' flags
3023 for the subexpressions which we are currently inside. Also records
3024 that those subexprs have matched. */
3025 #define SET_REGS_MATCHED() \
3029 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
3031 MATCHED_SOMETHING (reg_info[r]) \
3032 = EVER_MATCHED_SOMETHING (reg_info[r]) \
3039 /* This converts PTR, a pointer into one of the search strings `string1'
3040 and `string2' into an offset from the beginning of that string. */
3041 #define POINTER_TO_OFFSET(ptr) \
3042 (FIRST_STRING_P (ptr) ? (ptr) - string1 : (ptr) - string2 + size1)
3044 /* Registers are set to a sentinel when they haven't yet matched. */
3045 #define REG_UNSET_VALUE ((char *) -1)
3046 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
3049 /* Macros for dealing with the split strings in re_match_2. */
3051 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3053 /* Call before fetching a character with *d. This switches over to
3054 string2 if necessary. */
3055 #define PREFETCH() \
3058 /* End of string2 => fail. */ \
3059 if (dend == end_match_2) \
3061 /* End of string1 => advance to string2. */ \
3063 dend = end_match_2; \
3067 /* Test if at very beginning or at very end of the virtual concatenation
3068 of `string1' and `string2'. If only one string, it's `string2'. */
3069 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3070 #define AT_STRINGS_END(d) ((d) == end2)
3073 /* Test if D points to a character which is word-constituent. We have
3074 two special cases to check for: if past the end of string1, look at
3075 the first character in string2; and if before the beginning of
3076 string2, look at the last character in string1. */
3077 #define WORDCHAR_P(d) \
3078 (SYNTAX ((d) == end1 ? *string2 \
3079 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3082 /* Test if the character before D and the one at D differ with respect
3083 to being word-constituent. */
3084 #define AT_WORD_BOUNDARY(d) \
3085 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3086 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3089 /* Free everything we malloc. */
3091 #define FREE_VAR(var) if (var) free (var); var = NULL
3092 #define FREE_VARIABLES() \
3094 FREE_VAR (fail_stack.stack); \
3095 FREE_VAR (regstart); \
3096 FREE_VAR (regend); \
3097 FREE_VAR (old_regstart); \
3098 FREE_VAR (old_regend); \
3099 FREE_VAR (best_regstart); \
3100 FREE_VAR (best_regend); \
3101 FREE_VAR (reg_info); \
3102 FREE_VAR (reg_dummy); \
3103 FREE_VAR (reg_info_dummy); \
3105 #else /* not REGEX_MALLOC */
3106 /* Some MIPS systems (at least) want this to free alloca'd storage. */
3107 #define FREE_VARIABLES() alloca (0)
3108 #endif /* not REGEX_MALLOC */
3111 /* These values must meet several constraints. They must not be valid
3112 register values; since we have a limit of 255 registers (because
3113 we use only one byte in the pattern for the register number), we can
3114 use numbers larger than 255. They must differ by 1, because of
3115 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3116 be larger than the value for the highest register, so we do not try
3117 to actually save any registers when none are active. */
3118 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3119 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3121 /* Matching routines. */
3123 #ifndef emacs /* Emacs never uses this. */
3124 /* re_match is like re_match_2 except it takes only a single string. */
3127 re_match (bufp
, string
, size
, pos
, regs
)
3128 struct re_pattern_buffer
*bufp
;
3131 struct re_registers
*regs
;
3133 return re_match_2 (bufp
, NULL
, 0, string
, size
, pos
, regs
, size
);
3135 #endif /* not emacs */
3138 /* re_match_2 matches the compiled pattern in BUFP against the
3139 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3140 and SIZE2, respectively). We start matching at POS, and stop
3143 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3144 store offsets for the substring each group matched in REGS. See the
3145 documentation for exactly how many groups we fill.
3147 We return -1 if no match, -2 if an internal error (such as the
3148 failure stack overflowing). Otherwise, we return the length of the
3149 matched substring. */
3152 re_match_2 (bufp
, string1
, size1
, string2
, size2
, pos
, regs
, stop
)
3153 struct re_pattern_buffer
*bufp
;
3154 const char *string1
, *string2
;
3157 struct re_registers
*regs
;
3160 /* General temporaries. */
3164 /* Just past the end of the corresponding string. */
3165 const char *end1
, *end2
;
3167 /* Pointers into string1 and string2, just past the last characters in
3168 each to consider matching. */
3169 const char *end_match_1
, *end_match_2
;
3171 /* Where we are in the data, and the end of the current string. */
3172 const char *d
, *dend
;
3174 /* Where we are in the pattern, and the end of the pattern. */
3175 unsigned char *p
= bufp
->buffer
;
3176 register unsigned char *pend
= p
+ bufp
->used
;
3178 /* We use this to map every character in the string. */
3179 char *translate
= bufp
->translate
;
3181 /* Failure point stack. Each place that can handle a failure further
3182 down the line pushes a failure point on this stack. It consists of
3183 restart, regend, and reg_info for all registers corresponding to
3184 the subexpressions we're currently inside, plus the number of such
3185 registers, and, finally, two char *'s. The first char * is where
3186 to resume scanning the pattern; the second one is where to resume
3187 scanning the strings. If the latter is zero, the failure point is
3188 a ``dummy''; if a failure happens and the failure point is a dummy,
3189 it gets discarded and the next next one is tried. */
3190 fail_stack_type fail_stack
;
3192 static unsigned failure_id
= 0;
3193 unsigned nfailure_points_pushed
= 0, nfailure_points_popped
= 0;
3196 /* We fill all the registers internally, independent of what we
3197 return, for use in backreferences. The number here includes
3198 an element for register zero. */
3199 unsigned num_regs
= bufp
->re_nsub
+ 1;
3201 /* The currently active registers. */
3202 unsigned lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3203 unsigned highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3205 /* Information on the contents of registers. These are pointers into
3206 the input strings; they record just what was matched (on this
3207 attempt) by a subexpression part of the pattern, that is, the
3208 regnum-th regstart pointer points to where in the pattern we began
3209 matching and the regnum-th regend points to right after where we
3210 stopped matching the regnum-th subexpression. (The zeroth register
3211 keeps track of what the whole pattern matches.) */
3212 const char **regstart
, **regend
;
3214 /* If a group that's operated upon by a repetition operator fails to
3215 match anything, then the register for its start will need to be
3216 restored because it will have been set to wherever in the string we
3217 are when we last see its open-group operator. Similarly for a
3219 const char **old_regstart
, **old_regend
;
3221 /* The is_active field of reg_info helps us keep track of which (possibly
3222 nested) subexpressions we are currently in. The matched_something
3223 field of reg_info[reg_num] helps us tell whether or not we have
3224 matched any of the pattern so far this time through the reg_num-th
3225 subexpression. These two fields get reset each time through any
3226 loop their register is in. */
3227 register_info_type
*reg_info
;
3229 /* The following record the register info as found in the above
3230 variables when we find a match better than any we've seen before.
3231 This happens as we backtrack through the failure points, which in
3232 turn happens only if we have not yet matched the entire string. */
3233 unsigned best_regs_set
= false;
3234 const char **best_regstart
, **best_regend
;
3236 /* Logically, this is `best_regend[0]'. But we don't want to have to
3237 allocate space for that if we're not allocating space for anything
3238 else (see below). Also, we never need info about register 0 for
3239 any of the other register vectors, and it seems rather a kludge to
3240 treat `best_regend' differently than the rest. So we keep track of
3241 the end of the best match so far in a separate variable. We
3242 initialize this to NULL so that when we backtrack the first time
3243 and need to test it, it's not garbage. */
3244 const char *match_end
= NULL
;
3246 /* Used when we pop values we don't care about. */
3247 const char **reg_dummy
;
3248 register_info_type
*reg_info_dummy
;
3251 /* Counts the total number of registers pushed. */
3252 unsigned num_regs_pushed
= 0;
3255 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3259 /* Do not bother to initialize all the register variables if there are
3260 no groups in the pattern, as it takes a fair amount of time. If
3261 there are groups, we include space for register 0 (the whole
3262 pattern), even though we never use it, since it simplifies the
3263 array indexing. We should fix this. */
3266 regstart
= REGEX_TALLOC (num_regs
, const char *);
3267 regend
= REGEX_TALLOC (num_regs
, const char *);
3268 old_regstart
= REGEX_TALLOC (num_regs
, const char *);
3269 old_regend
= REGEX_TALLOC (num_regs
, const char *);
3270 best_regstart
= REGEX_TALLOC (num_regs
, const char *);
3271 best_regend
= REGEX_TALLOC (num_regs
, const char *);
3272 reg_info
= REGEX_TALLOC (num_regs
, register_info_type
);
3273 reg_dummy
= REGEX_TALLOC (num_regs
, const char *);
3274 reg_info_dummy
= REGEX_TALLOC (num_regs
, register_info_type
);
3276 if (!(regstart
&& regend
&& old_regstart
&& old_regend
&& reg_info
3277 && best_regstart
&& best_regend
&& reg_dummy
&& reg_info_dummy
))
3286 /* We must initialize all our variables to NULL, so that
3287 `FREE_VARIABLES' doesn't try to free them. */
3288 regstart
= regend
= old_regstart
= old_regend
= best_regstart
3289 = best_regend
= reg_dummy
= NULL
;
3290 reg_info
= reg_info_dummy
= (register_info_type
*) NULL
;
3292 #endif /* REGEX_MALLOC */
3294 /* The starting position is bogus. */
3295 if (pos
< 0 || pos
> size1
+ size2
)
3301 /* Initialize subexpression text positions to -1 to mark ones that no
3302 start_memory/stop_memory has been seen for. Also initialize the
3303 register information struct. */
3304 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3306 regstart
[mcnt
] = regend
[mcnt
]
3307 = old_regstart
[mcnt
] = old_regend
[mcnt
] = REG_UNSET_VALUE
;
3309 REG_MATCH_NULL_STRING_P (reg_info
[mcnt
]) = MATCH_NULL_UNSET_VALUE
;
3310 IS_ACTIVE (reg_info
[mcnt
]) = 0;
3311 MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3312 EVER_MATCHED_SOMETHING (reg_info
[mcnt
]) = 0;
3315 /* We move `string1' into `string2' if the latter's empty -- but not if
3316 `string1' is null. */
3317 if (size2
== 0 && string1
!= NULL
)
3324 end1
= string1
+ size1
;
3325 end2
= string2
+ size2
;
3327 /* Compute where to stop matching, within the two strings. */
3330 end_match_1
= string1
+ stop
;
3331 end_match_2
= string2
;
3336 end_match_2
= string2
+ stop
- size1
;
3339 /* `p' scans through the pattern as `d' scans through the data.
3340 `dend' is the end of the input string that `d' points within. `d'
3341 is advanced into the following input string whenever necessary, but
3342 this happens before fetching; therefore, at the beginning of the
3343 loop, `d' can be pointing at the end of a string, but it cannot
3345 if (size1
> 0 && pos
<= size1
)
3352 d
= string2
+ pos
- size1
;
3356 DEBUG_PRINT1 ("The compiled pattern is: ");
3357 DEBUG_PRINT_COMPILED_PATTERN (bufp
, p
, pend
);
3358 DEBUG_PRINT1 ("The string to match is: `");
3359 DEBUG_PRINT_DOUBLE_STRING (d
, string1
, size1
, string2
, size2
);
3360 DEBUG_PRINT1 ("'\n");
3362 /* This loops over pattern commands. It exits by returning from the
3363 function if the match is complete, or it drops through if the match
3364 fails at this starting point in the input data. */
3367 DEBUG_PRINT2 ("\n0x%x: ", p
);
3370 { /* End of pattern means we might have succeeded. */
3371 DEBUG_PRINT1 ("end of pattern ... ");
3373 /* If we haven't matched the entire string, and we want the
3374 longest match, try backtracking. */
3375 if (d
!= end_match_2
)
3377 DEBUG_PRINT1 ("backtracking.\n");
3379 if (!FAIL_STACK_EMPTY ())
3380 { /* More failure points to try. */
3381 boolean same_str_p
= (FIRST_STRING_P (match_end
)
3382 == MATCHING_IN_FIRST_STRING
);
3384 /* If exceeds best match so far, save it. */
3386 || (same_str_p
&& d
> match_end
)
3387 || (!same_str_p
&& !MATCHING_IN_FIRST_STRING
))
3389 best_regs_set
= true;
3392 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3394 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3396 best_regstart
[mcnt
] = regstart
[mcnt
];
3397 best_regend
[mcnt
] = regend
[mcnt
];
3403 /* If no failure points, don't restore garbage. */
3404 else if (best_regs_set
)
3407 /* Restore best match. It may happen that `dend ==
3408 end_match_1' while the restored d is in string2.
3409 For example, the pattern `x.*y.*z' against the
3410 strings `x-' and `y-z-', if the two strings are
3411 not consecutive in memory. */
3412 DEBUG_PRINT1 ("Restoring best registers.\n");
3415 dend
= ((d
>= string1
&& d
<= end1
)
3416 ? end_match_1
: end_match_2
);
3418 for (mcnt
= 1; mcnt
< num_regs
; mcnt
++)
3420 regstart
[mcnt
] = best_regstart
[mcnt
];
3421 regend
[mcnt
] = best_regend
[mcnt
];
3424 } /* d != end_match_2 */
3426 DEBUG_PRINT1 ("Accepting match.\n");
3428 /* If caller wants register contents data back, do it. */
3429 if (regs
&& !bufp
->no_sub
)
3431 /* Have the register data arrays been allocated? */
3432 if (bufp
->regs_allocated
== REGS_UNALLOCATED
)
3433 { /* No. So allocate them with malloc. We need one
3434 extra element beyond `num_regs' for the `-1' marker
3436 regs
->num_regs
= MAX (RE_NREGS
, num_regs
+ 1);
3437 regs
->start
= TALLOC (regs
->num_regs
, regoff_t
);
3438 regs
->end
= TALLOC (regs
->num_regs
, regoff_t
);
3439 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3441 bufp
->regs_allocated
= REGS_REALLOCATE
;
3443 else if (bufp
->regs_allocated
== REGS_REALLOCATE
)
3444 { /* Yes. If we need more elements than were already
3445 allocated, reallocate them. If we need fewer, just
3447 if (regs
->num_regs
< num_regs
+ 1)
3449 regs
->num_regs
= num_regs
+ 1;
3450 RETALLOC (regs
->start
, regs
->num_regs
, regoff_t
);
3451 RETALLOC (regs
->end
, regs
->num_regs
, regoff_t
);
3452 if (regs
->start
== NULL
|| regs
->end
== NULL
)
3457 assert (bufp
->regs_allocated
== REGS_FIXED
);
3459 /* Convert the pointer data in `regstart' and `regend' to
3460 indices. Register zero has to be set differently,
3461 since we haven't kept track of any info for it. */
3462 if (regs
->num_regs
> 0)
3464 regs
->start
[0] = pos
;
3465 regs
->end
[0] = (MATCHING_IN_FIRST_STRING
? d
- string1
3466 : d
- string2
+ size1
);
3469 /* Go through the first `min (num_regs, regs->num_regs)'
3470 registers, since that is all we initialized. */
3471 for (mcnt
= 1; mcnt
< MIN (num_regs
, regs
->num_regs
); mcnt
++)
3473 if (REG_UNSET (regstart
[mcnt
]) || REG_UNSET (regend
[mcnt
]))
3474 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3477 regs
->start
[mcnt
] = POINTER_TO_OFFSET (regstart
[mcnt
]);
3478 regs
->end
[mcnt
] = POINTER_TO_OFFSET (regend
[mcnt
]);
3482 /* If the regs structure we return has more elements than
3483 were in the pattern, set the extra elements to -1. If
3484 we (re)allocated the registers, this is the case,
3485 because we always allocate enough to have at least one
3487 for (mcnt
= num_regs
; mcnt
< regs
->num_regs
; mcnt
++)
3488 regs
->start
[mcnt
] = regs
->end
[mcnt
] = -1;
3489 } /* regs && !bufp->no_sub */
3492 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3493 nfailure_points_pushed
, nfailure_points_popped
,
3494 nfailure_points_pushed
- nfailure_points_popped
);
3495 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed
);
3497 mcnt
= d
- pos
- (MATCHING_IN_FIRST_STRING
3501 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt
);
3506 /* Otherwise match next pattern command. */
3507 #ifdef SWITCH_ENUM_BUG
3508 switch ((int) ((re_opcode_t
) *p
++))
3510 switch ((re_opcode_t
) *p
++)
3513 /* Ignore these. Used to ignore the n of succeed_n's which
3514 currently have n == 0. */
3516 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3520 /* Match the next n pattern characters exactly. The following
3521 byte in the pattern defines n, and the n bytes after that
3522 are the characters to match. */
3525 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt
);
3527 /* This is written out as an if-else so we don't waste time
3528 testing `translate' inside the loop. */
3534 if (translate
[(unsigned char) *d
++] != (char) *p
++)
3544 if (*d
++ != (char) *p
++) goto fail
;
3548 SET_REGS_MATCHED ();
3552 /* Match any character except possibly a newline or a null. */
3554 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3558 if ((!(bufp
->syntax
& RE_DOT_NEWLINE
) && TRANSLATE (*d
) == '\n')
3559 || (bufp
->syntax
& RE_DOT_NOT_NULL
&& TRANSLATE (*d
) == '\000'))
3562 SET_REGS_MATCHED ();
3563 DEBUG_PRINT2 (" Matched `%d'.\n", *d
);
3571 register unsigned char c
;
3572 boolean
not = (re_opcode_t
) *(p
- 1) == charset_not
;
3574 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3577 c
= TRANSLATE (*d
); /* The character to match. */
3579 /* Cast to `unsigned' instead of `unsigned char' in case the
3580 bit list is a full 32 bytes long. */
3581 if (c
< (unsigned) (*p
* BYTEWIDTH
)
3582 && p
[1 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
3587 if (!not) goto fail
;
3589 SET_REGS_MATCHED ();
3595 /* The beginning of a group is represented by start_memory.
3596 The arguments are the register number in the next byte, and the
3597 number of groups inner to this one in the next. The text
3598 matched within the group is recorded (in the internal
3599 registers data structure) under the register number. */
3601 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p
, p
[1]);
3603 /* Find out if this group can match the empty string. */
3604 p1
= p
; /* To send to group_match_null_string_p. */
3606 if (REG_MATCH_NULL_STRING_P (reg_info
[*p
]) == MATCH_NULL_UNSET_VALUE
)
3607 REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3608 = group_match_null_string_p (&p1
, pend
, reg_info
);
3610 /* Save the position in the string where we were the last time
3611 we were at this open-group operator in case the group is
3612 operated upon by a repetition operator, e.g., with `(a*)*b'
3613 against `ab'; then we want to ignore where we are now in
3614 the string in case this attempt to match fails. */
3615 old_regstart
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3616 ? REG_UNSET (regstart
[*p
]) ? d
: regstart
[*p
]
3618 DEBUG_PRINT2 (" old_regstart: %d\n",
3619 POINTER_TO_OFFSET (old_regstart
[*p
]));
3622 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart
[*p
]));
3624 IS_ACTIVE (reg_info
[*p
]) = 1;
3625 MATCHED_SOMETHING (reg_info
[*p
]) = 0;
3627 /* This is the new highest active register. */
3628 highest_active_reg
= *p
;
3630 /* If nothing was active before, this is the new lowest active
3632 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
3633 lowest_active_reg
= *p
;
3635 /* Move past the register number and inner group count. */
3640 /* The stop_memory opcode represents the end of a group. Its
3641 arguments are the same as start_memory's: the register
3642 number, and the number of inner groups. */
3644 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p
, p
[1]);
3646 /* We need to save the string position the last time we were at
3647 this close-group operator in case the group is operated
3648 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3649 against `aba'; then we want to ignore where we are now in
3650 the string in case this attempt to match fails. */
3651 old_regend
[*p
] = REG_MATCH_NULL_STRING_P (reg_info
[*p
])
3652 ? REG_UNSET (regend
[*p
]) ? d
: regend
[*p
]
3654 DEBUG_PRINT2 (" old_regend: %d\n",
3655 POINTER_TO_OFFSET (old_regend
[*p
]));
3658 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend
[*p
]));
3660 /* This register isn't active anymore. */
3661 IS_ACTIVE (reg_info
[*p
]) = 0;
3663 /* If this was the only register active, nothing is active
3665 if (lowest_active_reg
== highest_active_reg
)
3667 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3668 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3671 { /* We must scan for the new highest active register, since
3672 it isn't necessarily one less than now: consider
3673 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3674 new highest active register is 1. */
3675 unsigned char r
= *p
- 1;
3676 while (r
> 0 && !IS_ACTIVE (reg_info
[r
]))
3679 /* If we end up at register zero, that means that we saved
3680 the registers as the result of an `on_failure_jump', not
3681 a `start_memory', and we jumped to past the innermost
3682 `stop_memory'. For example, in ((.)*) we save
3683 registers 1 and 2 as a result of the *, but when we pop
3684 back to the second ), we are at the stop_memory 1.
3685 Thus, nothing is active. */
3688 lowest_active_reg
= NO_LOWEST_ACTIVE_REG
;
3689 highest_active_reg
= NO_HIGHEST_ACTIVE_REG
;
3692 highest_active_reg
= r
;
3695 /* If just failed to match something this time around with a
3696 group that's operated on by a repetition operator, try to
3697 force exit from the ``loop'', and restore the register
3698 information for this group that we had before trying this
3700 if ((!MATCHED_SOMETHING (reg_info
[*p
])
3701 || (re_opcode_t
) p
[-3] == start_memory
)
3704 boolean is_a_jump_n
= false;
3708 switch ((re_opcode_t
) *p1
++)
3712 case pop_failure_jump
:
3713 case maybe_pop_jump
:
3715 case dummy_failure_jump
:
3716 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
3726 /* If the next operation is a jump backwards in the pattern
3727 to an on_failure_jump right before the start_memory
3728 corresponding to this stop_memory, exit from the loop
3729 by forcing a failure after pushing on the stack the
3730 on_failure_jump's jump in the pattern, and d. */
3731 if (mcnt
< 0 && (re_opcode_t
) *p1
== on_failure_jump
3732 && (re_opcode_t
) p1
[3] == start_memory
&& p1
[4] == *p
)
3734 /* If this group ever matched anything, then restore
3735 what its registers were before trying this last
3736 failed match, e.g., with `(a*)*b' against `ab' for
3737 regstart[1], and, e.g., with `((a*)*(b*)*)*'
3738 against `aba' for regend[3].
3740 Also restore the registers for inner groups for,
3741 e.g., `((a*)(b*))*' against `aba' (register 3 would
3742 otherwise get trashed). */
3744 if (EVER_MATCHED_SOMETHING (reg_info
[*p
]))
3748 EVER_MATCHED_SOMETHING (reg_info
[*p
]) = 0;
3750 /* Restore this and inner groups' (if any) registers. */
3751 for (r
= *p
; r
< *p
+ *(p
+ 1); r
++)
3753 regstart
[r
] = old_regstart
[r
];
3755 /* xx why this test? */
3756 if ((int) old_regend
[r
] >= (int) regstart
[r
])
3757 regend
[r
] = old_regend
[r
];
3761 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
3762 PUSH_FAILURE_POINT (p1
+ mcnt
, d
, -2);
3768 /* Move past the register number and the inner group count. */
3773 /* \<digit> has been turned into a `duplicate' command which is
3774 followed by the numeric value of <digit> as the register number. */
3777 register const char *d2
, *dend2
;
3778 int regno
= *p
++; /* Get which register to match against. */
3779 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno
);
3781 /* Can't back reference a group which we've never matched. */
3782 if (REG_UNSET (regstart
[regno
]) || REG_UNSET (regend
[regno
]))
3785 /* Where in input to try to start matching. */
3786 d2
= regstart
[regno
];
3788 /* Where to stop matching; if both the place to start and
3789 the place to stop matching are in the same string, then
3790 set to the place to stop, otherwise, for now have to use
3791 the end of the first string. */
3793 dend2
= ((FIRST_STRING_P (regstart
[regno
])
3794 == FIRST_STRING_P (regend
[regno
]))
3795 ? regend
[regno
] : end_match_1
);
3798 /* If necessary, advance to next segment in register
3802 if (dend2
== end_match_2
) break;
3803 if (dend2
== regend
[regno
]) break;
3805 /* End of string1 => advance to string2. */
3807 dend2
= regend
[regno
];
3809 /* At end of register contents => success */
3810 if (d2
== dend2
) break;
3812 /* If necessary, advance to next segment in data. */
3815 /* How many characters left in this segment to match. */
3818 /* Want how many consecutive characters we can match in
3819 one shot, so, if necessary, adjust the count. */
3820 if (mcnt
> dend2
- d2
)
3823 /* Compare that many; failure if mismatch, else move
3826 ? bcmp_translate (d
, d2
, mcnt
, translate
)
3827 : bcmp (d
, d2
, mcnt
))
3829 d
+= mcnt
, d2
+= mcnt
;
3835 /* begline matches the empty string at the beginning of the string
3836 (unless `not_bol' is set in `bufp'), and, if
3837 `newline_anchor' is set, after newlines. */
3839 DEBUG_PRINT1 ("EXECUTING begline.\n");
3841 if (AT_STRINGS_BEG (d
))
3843 if (!bufp
->not_bol
) break;
3845 else if (d
[-1] == '\n' && bufp
->newline_anchor
)
3849 /* In all other cases, we fail. */
3853 /* endline is the dual of begline. */
3855 DEBUG_PRINT1 ("EXECUTING endline.\n");
3857 if (AT_STRINGS_END (d
))
3859 if (!bufp
->not_eol
) break;
3862 /* We have to ``prefetch'' the next character. */
3863 else if ((d
== end1
? *string2
: *d
) == '\n'
3864 && bufp
->newline_anchor
)
3871 /* Match at the very beginning of the data. */
3873 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
3874 if (AT_STRINGS_BEG (d
))
3879 /* Match at the very end of the data. */
3881 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
3882 if (AT_STRINGS_END (d
))
3887 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
3888 pushes NULL as the value for the string on the stack. Then
3889 `pop_failure_point' will keep the current value for the
3890 string, instead of restoring it. To see why, consider
3891 matching `foo\nbar' against `.*\n'. The .* matches the foo;
3892 then the . fails against the \n. But the next thing we want
3893 to do is match the \n against the \n; if we restored the
3894 string value, we would be back at the foo.
3896 Because this is used only in specific cases, we don't need to
3897 check all the things that `on_failure_jump' does, to make
3898 sure the right things get saved on the stack. Hence we don't
3899 share its code. The only reason to push anything on the
3900 stack at all is that otherwise we would have to change
3901 `anychar's code to do something besides goto fail in this
3902 case; that seems worse than this. */
3903 case on_failure_keep_string_jump
:
3904 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
3906 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
3907 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt
, p
+ mcnt
);
3909 PUSH_FAILURE_POINT (p
+ mcnt
, NULL
, -2);
3913 /* Uses of on_failure_jump:
3915 Each alternative starts with an on_failure_jump that points
3916 to the beginning of the next alternative. Each alternative
3917 except the last ends with a jump that in effect jumps past
3918 the rest of the alternatives. (They really jump to the
3919 ending jump of the following alternative, because tensioning
3920 these jumps is a hassle.)
3922 Repeats start with an on_failure_jump that points past both
3923 the repetition text and either the following jump or
3924 pop_failure_jump back to this on_failure_jump. */
3925 case on_failure_jump
:
3927 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
3929 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
3930 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt
, p
+ mcnt
);
3932 /* If this on_failure_jump comes right before a group (i.e.,
3933 the original * applied to a group), save the information
3934 for that group and all inner ones, so that if we fail back
3935 to this point, the group's information will be correct.
3936 For example, in \(a*\)*\1, we need the preceding group,
3937 and in \(\(a*\)b*\)\2, we need the inner group. */
3939 /* We can't use `p' to check ahead because we push
3940 a failure point to `p + mcnt' after we do this. */
3943 /* We need to skip no_op's before we look for the
3944 start_memory in case this on_failure_jump is happening as
3945 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
3947 while (p1
< pend
&& (re_opcode_t
) *p1
== no_op
)
3950 if (p1
< pend
&& (re_opcode_t
) *p1
== start_memory
)
3952 /* We have a new highest active register now. This will
3953 get reset at the start_memory we are about to get to,
3954 but we will have saved all the registers relevant to
3955 this repetition op, as described above. */
3956 highest_active_reg
= *(p1
+ 1) + *(p1
+ 2);
3957 if (lowest_active_reg
== NO_LOWEST_ACTIVE_REG
)
3958 lowest_active_reg
= *(p1
+ 1);
3961 DEBUG_PRINT1 (":\n");
3962 PUSH_FAILURE_POINT (p
+ mcnt
, d
, -2);
3966 /* A smart repeat ends with `maybe_pop_jump'.
3967 We change it to either `pop_failure_jump' or `jump'. */
3968 case maybe_pop_jump
:
3969 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
3970 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt
);
3972 register unsigned char *p2
= p
;
3974 /* Compare the beginning of the repeat with what in the
3975 pattern follows its end. If we can establish that there
3976 is nothing that they would both match, i.e., that we
3977 would have to backtrack because of (as in, e.g., `a*a')
3978 then we can change to pop_failure_jump, because we'll
3979 never have to backtrack.
3981 This is not true in the case of alternatives: in
3982 `(a|ab)*' we do need to backtrack to the `ab' alternative
3983 (e.g., if the string was `ab'). But instead of trying to
3984 detect that here, the alternative has put on a dummy
3985 failure point which is what we will end up popping. */
3987 /* Skip over open/close-group commands. */
3988 while (p2
+ 2 < pend
3989 && ((re_opcode_t
) *p2
== stop_memory
3990 || (re_opcode_t
) *p2
== start_memory
))
3991 p2
+= 3; /* Skip over args, too. */
3993 /* If we're at the end of the pattern, we can change. */
3996 /* Consider what happens when matching ":\(.*\)"
3997 against ":/". I don't really understand this code
3999 p
[-3] = (unsigned char) pop_failure_jump
;
4001 (" End of pattern: change to `pop_failure_jump'.\n");
4004 else if ((re_opcode_t
) *p2
== exactn
4005 || (bufp
->newline_anchor
&& (re_opcode_t
) *p2
== endline
))
4007 register unsigned char c
4008 = *p2
== (unsigned char) endline
? '\n' : p2
[2];
4011 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4012 to the `maybe_finalize_jump' of this case. Examine what
4014 if ((re_opcode_t
) p1
[3] == exactn
&& p1
[5] != c
)
4016 p
[-3] = (unsigned char) pop_failure_jump
;
4017 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4021 else if ((re_opcode_t
) p1
[3] == charset
4022 || (re_opcode_t
) p1
[3] == charset_not
)
4024 int not = (re_opcode_t
) p1
[3] == charset_not
;
4026 if (c
< (unsigned char) (p1
[4] * BYTEWIDTH
)
4027 && p1
[5 + c
/ BYTEWIDTH
] & (1 << (c
% BYTEWIDTH
)))
4030 /* `not' is equal to 1 if c would match, which means
4031 that we can't change to pop_failure_jump. */
4034 p
[-3] = (unsigned char) pop_failure_jump
;
4035 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4040 p
-= 2; /* Point at relative address again. */
4041 if ((re_opcode_t
) p
[-1] != pop_failure_jump
)
4043 p
[-1] = (unsigned char) jump
;
4044 DEBUG_PRINT1 (" Match => jump.\n");
4045 goto unconditional_jump
;
4047 /* Note fall through. */
4050 /* The end of a simple repeat has a pop_failure_jump back to
4051 its matching on_failure_jump, where the latter will push a
4052 failure point. The pop_failure_jump takes off failure
4053 points put on by this pop_failure_jump's matching
4054 on_failure_jump; we got through the pattern to here from the
4055 matching on_failure_jump, so didn't fail. */
4056 case pop_failure_jump
:
4058 /* We need to pass separate storage for the lowest and
4059 highest registers, even though we don't care about the
4060 actual values. Otherwise, we will restore only one
4061 register from the stack, since lowest will == highest in
4062 `pop_failure_point'. */
4063 unsigned dummy_low_reg
, dummy_high_reg
;
4064 unsigned char *pdummy
;
4067 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4068 POP_FAILURE_POINT (sdummy
, pdummy
,
4069 dummy_low_reg
, dummy_high_reg
,
4070 reg_dummy
, reg_dummy
, reg_info_dummy
);
4072 /* Note fall through. */
4075 /* Unconditionally jump (without popping any failure points). */
4078 EXTRACT_NUMBER_AND_INCR (mcnt
, p
); /* Get the amount to jump. */
4079 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt
);
4080 p
+= mcnt
; /* Do the jump. */
4081 DEBUG_PRINT2 ("(to 0x%x).\n", p
);
4085 /* We need this opcode so we can detect where alternatives end
4086 in `group_match_null_string_p' et al. */
4088 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4089 goto unconditional_jump
;
4092 /* Normally, the on_failure_jump pushes a failure point, which
4093 then gets popped at pop_failure_jump. We will end up at
4094 pop_failure_jump, also, and with a pattern of, say, `a+', we
4095 are skipping over the on_failure_jump, so we have to push
4096 something meaningless for pop_failure_jump to pop. */
4097 case dummy_failure_jump
:
4098 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4099 /* It doesn't matter what we push for the string here. What
4100 the code at `fail' tests is the value for the pattern. */
4101 PUSH_FAILURE_POINT (0, 0, -2);
4102 goto unconditional_jump
;
4105 /* At the end of an alternative, we need to push a dummy failure
4106 point in case we are followed by a `pop_failure_jump', because
4107 we don't want the failure point for the alternative to be
4108 popped. For example, matching `(a|ab)*' against `aab'
4109 requires that we match the `ab' alternative. */
4110 case push_dummy_failure
:
4111 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4112 /* See comments just above at `dummy_failure_jump' about the
4114 PUSH_FAILURE_POINT (0, 0, -2);
4117 /* Have to succeed matching what follows at least n times.
4118 After that, handle like `on_failure_jump'. */
4120 EXTRACT_NUMBER (mcnt
, p
+ 2);
4121 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt
);
4124 /* Originally, this is how many times we HAVE to succeed. */
4129 STORE_NUMBER_AND_INCR (p
, mcnt
);
4130 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p
, mcnt
);
4134 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p
+2);
4135 p
[2] = (unsigned char) no_op
;
4136 p
[3] = (unsigned char) no_op
;
4142 EXTRACT_NUMBER (mcnt
, p
+ 2);
4143 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt
);
4145 /* Originally, this is how many times we CAN jump. */
4149 STORE_NUMBER (p
+ 2, mcnt
);
4150 goto unconditional_jump
;
4152 /* If don't have to jump any more, skip over the rest of command. */
4159 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4161 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4163 EXTRACT_NUMBER_AND_INCR (mcnt
, p
);
4164 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1
, mcnt
);
4165 STORE_NUMBER (p1
, mcnt
);
4170 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4171 if (AT_WORD_BOUNDARY (d
))
4176 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4177 if (AT_WORD_BOUNDARY (d
))
4182 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4183 if (WORDCHAR_P (d
) && (AT_STRINGS_BEG (d
) || !WORDCHAR_P (d
- 1)))
4188 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4189 if (!AT_STRINGS_BEG (d
) && WORDCHAR_P (d
- 1)
4190 && (!WORDCHAR_P (d
) || AT_STRINGS_END (d
)))
4197 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4198 if (PTR_CHAR_POS ((unsigned char *) d
) >= point
)
4203 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4204 if (PTR_CHAR_POS ((unsigned char *) d
) != point
)
4209 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4210 if (PTR_CHAR_POS ((unsigned char *) d
) <= point
)
4213 #else /* not emacs19 */
4215 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4216 if (PTR_CHAR_POS ((unsigned char *) d
) + 1 != point
)
4219 #endif /* not emacs19 */
4222 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt
);
4227 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4231 if (SYNTAX (*d
++) != (enum syntaxcode
) mcnt
)
4233 SET_REGS_MATCHED ();
4237 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt
);
4239 goto matchnotsyntax
;
4242 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4246 if (SYNTAX (*d
++) == (enum syntaxcode
) mcnt
)
4248 SET_REGS_MATCHED ();
4251 #else /* not emacs */
4253 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4255 if (!WORDCHAR_P (d
))
4257 SET_REGS_MATCHED ();
4262 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4266 SET_REGS_MATCHED ();
4269 #endif /* not emacs */
4274 continue; /* Successfully executed one pattern command; keep going. */
4277 /* We goto here if a matching operation fails. */
4279 if (!FAIL_STACK_EMPTY ())
4280 { /* A restart point is known. Restore to that state. */
4281 DEBUG_PRINT1 ("\nFAIL:\n");
4282 POP_FAILURE_POINT (d
, p
,
4283 lowest_active_reg
, highest_active_reg
,
4284 regstart
, regend
, reg_info
);
4286 /* If this failure point is a dummy, try the next one. */
4290 /* If we failed to the end of the pattern, don't examine *p. */
4294 boolean is_a_jump_n
= false;
4296 /* If failed to a backwards jump that's part of a repetition
4297 loop, need to pop this failure point and use the next one. */
4298 switch ((re_opcode_t
) *p
)
4302 case maybe_pop_jump
:
4303 case pop_failure_jump
:
4306 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4309 if ((is_a_jump_n
&& (re_opcode_t
) *p1
== succeed_n
)
4311 && (re_opcode_t
) *p1
== on_failure_jump
))
4319 if (d
>= string1
&& d
<= end1
)
4323 break; /* Matching at this starting point really fails. */
4327 goto restore_best_regs
;
4331 return -1; /* Failure to match. */
4334 /* Subroutine definitions for re_match_2. */
4337 /* We are passed P pointing to a register number after a start_memory.
4339 Return true if the pattern up to the corresponding stop_memory can
4340 match the empty string, and false otherwise.
4342 If we find the matching stop_memory, sets P to point to one past its number.
4343 Otherwise, sets P to an undefined byte less than or equal to END.
4345 We don't handle duplicates properly (yet). */
4348 group_match_null_string_p (p
, end
, reg_info
)
4349 unsigned char **p
, *end
;
4350 register_info_type
*reg_info
;
4353 /* Point to after the args to the start_memory. */
4354 unsigned char *p1
= *p
+ 2;
4358 /* Skip over opcodes that can match nothing, and return true or
4359 false, as appropriate, when we get to one that can't, or to the
4360 matching stop_memory. */
4362 switch ((re_opcode_t
) *p1
)
4364 /* Could be either a loop or a series of alternatives. */
4365 case on_failure_jump
:
4367 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4369 /* If the next operation is not a jump backwards in the
4374 /* Go through the on_failure_jumps of the alternatives,
4375 seeing if any of the alternatives cannot match nothing.
4376 The last alternative starts with only a jump,
4377 whereas the rest start with on_failure_jump and end
4378 with a jump, e.g., here is the pattern for `a|b|c':
4380 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4381 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4384 So, we have to first go through the first (n-1)
4385 alternatives and then deal with the last one separately. */
4388 /* Deal with the first (n-1) alternatives, which start
4389 with an on_failure_jump (see above) that jumps to right
4390 past a jump_past_alt. */
4392 while ((re_opcode_t
) p1
[mcnt
-3] == jump_past_alt
)
4394 /* `mcnt' holds how many bytes long the alternative
4395 is, including the ending `jump_past_alt' and
4398 if (!alt_match_null_string_p (p1
, p1
+ mcnt
- 3,
4402 /* Move to right after this alternative, including the
4406 /* Break if it's the beginning of an n-th alternative
4407 that doesn't begin with an on_failure_jump. */
4408 if ((re_opcode_t
) *p1
!= on_failure_jump
)
4411 /* Still have to check that it's not an n-th
4412 alternative that starts with an on_failure_jump. */
4414 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4415 if ((re_opcode_t
) p1
[mcnt
-3] != jump_past_alt
)
4417 /* Get to the beginning of the n-th alternative. */
4423 /* Deal with the last alternative: go back and get number
4424 of the `jump_past_alt' just before it. `mcnt' contains
4425 the length of the alternative. */
4426 EXTRACT_NUMBER (mcnt
, p1
- 2);
4428 if (!alt_match_null_string_p (p1
, p1
+ mcnt
, reg_info
))
4431 p1
+= mcnt
; /* Get past the n-th alternative. */
4437 assert (p1
[1] == **p
);
4443 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4446 } /* while p1 < end */
4449 } /* group_match_null_string_p */
4452 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4453 It expects P to be the first byte of a single alternative and END one
4454 byte past the last. The alternative can contain groups. */
4457 alt_match_null_string_p (p
, end
, reg_info
)
4458 unsigned char *p
, *end
;
4459 register_info_type
*reg_info
;
4462 unsigned char *p1
= p
;
4466 /* Skip over opcodes that can match nothing, and break when we get
4467 to one that can't. */
4469 switch ((re_opcode_t
) *p1
)
4472 case on_failure_jump
:
4474 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4479 if (!common_op_match_null_string_p (&p1
, end
, reg_info
))
4482 } /* while p1 < end */
4485 } /* alt_match_null_string_p */
4488 /* Deals with the ops common to group_match_null_string_p and
4489 alt_match_null_string_p.
4491 Sets P to one after the op and its arguments, if any. */
4494 common_op_match_null_string_p (p
, end
, reg_info
)
4495 unsigned char **p
, *end
;
4496 register_info_type
*reg_info
;
4501 unsigned char *p1
= *p
;
4503 switch ((re_opcode_t
) *p1
++)
4523 assert (reg_no
> 0 && reg_no
<= MAX_REGNUM
);
4524 ret
= group_match_null_string_p (&p1
, end
, reg_info
);
4526 /* Have to set this here in case we're checking a group which
4527 contains a group and a back reference to it. */
4529 if (REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) == MATCH_NULL_UNSET_VALUE
)
4530 REG_MATCH_NULL_STRING_P (reg_info
[reg_no
]) = ret
;
4536 /* If this is an optimized succeed_n for zero times, make the jump. */
4538 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4546 /* Get to the number of times to succeed. */
4548 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4553 EXTRACT_NUMBER_AND_INCR (mcnt
, p1
);
4561 if (!REG_MATCH_NULL_STRING_P (reg_info
[*p1
]))
4569 /* All other opcodes mean we cannot match the empty string. */
4575 } /* common_op_match_null_string_p */
4578 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4579 bytes; nonzero otherwise. */
4582 bcmp_translate (s1
, s2
, len
, translate
)
4583 unsigned char *s1
, *s2
;
4587 register unsigned char *p1
= s1
, *p2
= s2
;
4590 if (translate
[*p1
++] != translate
[*p2
++]) return 1;
4596 /* Entry points for GNU code. */
4598 /* re_compile_pattern is the GNU regular expression compiler: it
4599 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4600 Returns 0 if the pattern was valid, otherwise an error string.
4602 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4603 are set in BUFP on entry.
4605 We call regex_compile to do the actual compilation. */
4608 re_compile_pattern (pattern
, length
, bufp
)
4609 const char *pattern
;
4611 struct re_pattern_buffer
*bufp
;
4615 /* GNU code is written to assume at least RE_NREGS registers will be set
4616 (and at least one extra will be -1). */
4617 bufp
->regs_allocated
= REGS_UNALLOCATED
;
4619 /* And GNU code determines whether or not to get register information
4620 by passing null for the REGS argument to re_match, etc., not by
4624 /* Match anchors at newline. */
4625 bufp
->newline_anchor
= 1;
4627 ret
= regex_compile (pattern
, length
, re_syntax_options
, bufp
);
4629 return re_error_msg
[(int) ret
];
4632 /* Entry points compatible with 4.2 BSD regex library. We don't define
4633 them if this is an Emacs or POSIX compilation. */
4635 #if !defined (emacs) && !defined (_POSIX_SOURCE)
4637 /* BSD has one and only one pattern buffer. */
4638 static struct re_pattern_buffer re_comp_buf
;
4648 if (!re_comp_buf
.buffer
)
4649 return "No previous regular expression";
4653 if (!re_comp_buf
.buffer
)
4655 re_comp_buf
.buffer
= (unsigned char *) malloc (200);
4656 if (re_comp_buf
.buffer
== NULL
)
4657 return "Memory exhausted";
4658 re_comp_buf
.allocated
= 200;
4660 re_comp_buf
.fastmap
= (char *) malloc (1 << BYTEWIDTH
);
4661 if (re_comp_buf
.fastmap
== NULL
)
4662 return "Memory exhausted";
4665 /* Since `re_exec' always passes NULL for the `regs' argument, we
4666 don't need to initialize the pattern buffer fields which affect it. */
4668 /* Match anchors at newlines. */
4669 re_comp_buf
.newline_anchor
= 1;
4671 ret
= regex_compile (s
, strlen (s
), re_syntax_options
, &re_comp_buf
);
4673 /* Yes, we're discarding `const' here. */
4674 return (char *) re_error_msg
[(int) ret
];
4682 const int len
= strlen (s
);
4684 0 <= re_search (&re_comp_buf
, s
, len
, 0, len
, (struct re_registers
*) 0);
4686 #endif /* not emacs and not _POSIX_SOURCE */
4688 /* POSIX.2 functions. Don't define these for Emacs. */
4692 /* regcomp takes a regular expression as a string and compiles it.
4694 PREG is a regex_t *. We do not expect any fields to be initialized,
4695 since POSIX says we shouldn't. Thus, we set
4697 `buffer' to the compiled pattern;
4698 `used' to the length of the compiled pattern;
4699 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
4700 REG_EXTENDED bit in CFLAGS is set; otherwise, to
4701 RE_SYNTAX_POSIX_BASIC;
4702 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
4703 `fastmap' and `fastmap_accurate' to zero;
4704 `re_nsub' to the number of subexpressions in PATTERN.
4706 PATTERN is the address of the pattern string.
4708 CFLAGS is a series of bits which affect compilation.
4710 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
4711 use POSIX basic syntax.
4713 If REG_NEWLINE is set, then . and [^...] don't match newline.
4714 Also, regexec will try a match beginning after every newline.
4716 If REG_ICASE is set, then we considers upper- and lowercase
4717 versions of letters to be equivalent when matching.
4719 If REG_NOSUB is set, then when PREG is passed to regexec, that
4720 routine will report only success or failure, and nothing about the
4723 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
4724 the return codes and their meanings.) */
4727 regcomp (preg
, pattern
, cflags
)
4729 const char *pattern
;
4734 = (cflags
& REG_EXTENDED
) ?
4735 RE_SYNTAX_POSIX_EXTENDED
: RE_SYNTAX_POSIX_BASIC
;
4737 /* regex_compile will allocate the space for the compiled pattern. */
4739 preg
->allocated
= 0;
4741 /* Don't bother to use a fastmap when searching. This simplifies the
4742 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
4743 characters after newlines into the fastmap. This way, we just try
4747 if (cflags
& REG_ICASE
)
4751 preg
->translate
= (char *) malloc (CHAR_SET_SIZE
);
4752 if (preg
->translate
== NULL
)
4753 return (int) REG_ESPACE
;
4755 /* Map uppercase characters to corresponding lowercase ones. */
4756 for (i
= 0; i
< CHAR_SET_SIZE
; i
++)
4757 preg
->translate
[i
] = ISUPPER (i
) ? tolower (i
) : i
;
4760 preg
->translate
= NULL
;
4762 /* If REG_NEWLINE is set, newlines are treated differently. */
4763 if (cflags
& REG_NEWLINE
)
4764 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
4765 syntax
&= ~RE_DOT_NEWLINE
;
4766 syntax
|= RE_HAT_LISTS_NOT_NEWLINE
;
4767 /* It also changes the matching behavior. */
4768 preg
->newline_anchor
= 1;
4771 preg
->newline_anchor
= 0;
4773 preg
->no_sub
= !!(cflags
& REG_NOSUB
);
4775 /* POSIX says a null character in the pattern terminates it, so we
4776 can use strlen here in compiling the pattern. */
4777 ret
= regex_compile (pattern
, strlen (pattern
), syntax
, preg
);
4779 /* POSIX doesn't distinguish between an unmatched open-group and an
4780 unmatched close-group: both are REG_EPAREN. */
4781 if (ret
== REG_ERPAREN
) ret
= REG_EPAREN
;
4787 /* regexec searches for a given pattern, specified by PREG, in the
4790 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
4791 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
4792 least NMATCH elements, and we set them to the offsets of the
4793 corresponding matched substrings.
4795 EFLAGS specifies `execution flags' which affect matching: if
4796 REG_NOTBOL is set, then ^ does not match at the beginning of the
4797 string; if REG_NOTEOL is set, then $ does not match at the end.
4799 We return 0 if we find a match and REG_NOMATCH if not. */
4802 regexec (preg
, string
, nmatch
, pmatch
, eflags
)
4803 const regex_t
*preg
;
4806 regmatch_t pmatch
[];
4810 struct re_registers regs
;
4811 regex_t private_preg
;
4812 int len
= strlen (string
);
4813 boolean want_reg_info
= !preg
->no_sub
&& nmatch
> 0;
4815 private_preg
= *preg
;
4817 private_preg
.not_bol
= !!(eflags
& REG_NOTBOL
);
4818 private_preg
.not_eol
= !!(eflags
& REG_NOTEOL
);
4820 /* The user has told us exactly how many registers to return
4821 information about, via `nmatch'. We have to pass that on to the
4822 matching routines. */
4823 private_preg
.regs_allocated
= REGS_FIXED
;
4827 regs
.num_regs
= nmatch
;
4828 regs
.start
= TALLOC (nmatch
, regoff_t
);
4829 regs
.end
= TALLOC (nmatch
, regoff_t
);
4830 if (regs
.start
== NULL
|| regs
.end
== NULL
)
4831 return (int) REG_NOMATCH
;
4834 /* Perform the searching operation. */
4835 ret
= re_search (&private_preg
, string
, len
,
4836 /* start: */ 0, /* range: */ len
,
4837 want_reg_info
? ®s
: (struct re_registers
*) 0);
4839 /* Copy the register information to the POSIX structure. */
4846 for (r
= 0; r
< nmatch
; r
++)
4848 pmatch
[r
].rm_so
= regs
.start
[r
];
4849 pmatch
[r
].rm_eo
= regs
.end
[r
];
4853 /* If we needed the temporary register info, free the space now. */
4858 /* We want zero return to mean success, unlike `re_search'. */
4859 return ret
>= 0 ? (int) REG_NOERROR
: (int) REG_NOMATCH
;
4863 /* Returns a message corresponding to an error code, ERRCODE, returned
4864 from either regcomp or regexec. We don't use PREG here. */
4867 regerror (errcode
, preg
, errbuf
, errbuf_size
)
4869 const regex_t
*preg
;
4877 || errcode
>= (sizeof (re_error_msg
) / sizeof (re_error_msg
[0])))
4878 /* Only error codes returned by the rest of the code should be passed
4879 to this routine. If we are given anything else, or if other regex
4880 code generates an invalid error code, then the program has a bug.
4881 Dump core so we can fix it. */
4884 msg_size
= strlen (msg
) + 1; /* Includes the null. */
4886 if (errbuf_size
!= 0)
4888 if (msg_size
> errbuf_size
)
4890 strncpy (errbuf
, msg
, errbuf_size
- 1);
4891 errbuf
[errbuf_size
- 1] = 0;
4894 strcpy (errbuf
, msg
);
4901 /* Free dynamically allocated space used by PREG. */
4907 if (preg
->buffer
!= NULL
)
4908 free (preg
->buffer
);
4909 preg
->buffer
= NULL
;
4911 preg
->allocated
= 0;
4914 if (preg
->fastmap
!= NULL
)
4915 free (preg
->fastmap
);
4916 preg
->fastmap
= NULL
;
4917 preg
->fastmap_accurate
= 0;
4919 if (preg
->translate
!= NULL
)
4920 free (preg
->translate
);
4921 preg
->translate
= NULL
;
4924 #endif /* not emacs */
4928 make-backup-files: t
4930 trim-versions-without-asking: nil