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)
33 /* We need this for `regex.h', and perhaps for the Emacs include files. */
34 #include <sys/types.h>
45 /* The `emacs' switch turns on certain matching commands
46 that make sense only in Emacs. */
53 /* Emacs uses `NULL' as a predicate. */
58 /* We used to test for `BSTRING' here, but only GCC and Emacs define
59 `BSTRING', as far as I know, and neither of them use this code. */
60 #if HAVE_STRING_H || STDC_HEADERS
63 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
66 #define bcopy(s, d, n) memcpy ((d), (s), (n))
69 #define bzero(s, n) memset ((s), 0, (n))
83 /* Define the syntax stuff for \<, \>, etc. */
85 /* This must be nonzero for the wordchar and notwordchar pattern
86 commands in re_match_2. */
93 extern char *re_syntax_table;
95 #else /* not SYNTAX_TABLE */
97 /* How many characters in the character set. */
98 #define CHAR_SET_SIZE 256
100 static char re_syntax_table[CHAR_SET_SIZE];
111 bzero (re_syntax_table, sizeof re_syntax_table);
113 for (c = 'a'; c <= 'z'; c++)
114 re_syntax_table[c] = Sword;
116 for (c = 'A'; c <= 'Z'; c++)
117 re_syntax_table[c] = Sword;
119 for (c = '0'; c <= '9'; c++)
120 re_syntax_table[c] = Sword;
122 re_syntax_table['_'] = Sword;
127 #endif /* not SYNTAX_TABLE */
129 #define SYNTAX(c) re_syntax_table[c]
131 #endif /* not emacs */
133 /* Get the interface, including the syntax bits. */
136 /* isalpha etc. are used for the character classes. */
144 #define ISBLANK(c) (isascii (c) && isblank (c))
146 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
149 #define ISGRAPH(c) (isascii (c) && isgraph (c))
151 #define ISGRAPH(c) (isascii (c) && isprint (c) && !isspace (c))
154 #define ISPRINT(c) (isascii (c) && isprint (c))
155 #define ISDIGIT(c) (isascii (c) && isdigit (c))
156 #define ISALNUM(c) (isascii (c) && isalnum (c))
157 #define ISALPHA(c) (isascii (c) && isalpha (c))
158 #define ISCNTRL(c) (isascii (c) && iscntrl (c))
159 #define ISLOWER(c) (isascii (c) && islower (c))
160 #define ISPUNCT(c) (isascii (c) && ispunct (c))
161 #define ISSPACE(c) (isascii (c) && isspace (c))
162 #define ISUPPER(c) (isascii (c) && isupper (c))
163 #define ISXDIGIT(c) (isascii (c) && isxdigit (c))
169 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
170 since ours (we hope) works properly with all combinations of
171 machines, compilers, `char' and `unsigned char' argument types.
172 (Per Bothner suggested the basic approach.) */
173 #undef SIGN_EXTEND_CHAR
175 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
176 #else /* not __STDC__ */
177 /* As in Harbison and Steele. */
178 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
181 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
182 use `alloca' instead of `malloc'. This is because using malloc in
183 re_search* or re_match* could cause memory leaks when C-g is used in
184 Emacs; also, malloc is slower and causes storage fragmentation. On
185 the other hand, malloc is more portable, and easier to debug.
187 Because we sometimes use alloca, some routines have to be macros,
188 not functions -- `alloca'-allocated space disappears at the end of the
189 function it is called in. */
193 #define REGEX_ALLOCATE malloc
194 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
196 #else /* not REGEX_MALLOC */
198 /* Emacs already defines alloca, sometimes. */
201 /* Make alloca work the best possible way. */
203 #define alloca __builtin_alloca
204 #else /* not __GNUC__ */
207 #else /* not __GNUC__ or HAVE_ALLOCA_H */
208 #ifndef _AIX /* Already did AIX, up at the top. */
210 #endif /* not _AIX */
211 #endif /* not HAVE_ALLOCA_H */
212 #endif /* not __GNUC__ */
214 #endif /* not alloca */
216 #define REGEX_ALLOCATE alloca
218 /* Assumes a `char *destination' variable. */
219 #define REGEX_REALLOCATE(source, osize, nsize) \
220 (destination = (char *) alloca (nsize), \
221 bcopy (source, destination, osize), \
224 #endif /* not REGEX_MALLOC */
227 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
228 `string1' or just past its end. This works if PTR is NULL, which is
230 #define FIRST_STRING_P(ptr) \
231 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
233 /* (Re)Allocate N items of type T using malloc, or fail. */
234 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
235 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
236 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
238 #define BYTEWIDTH 8 /* In bits. */
240 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
242 #define MAX(a, b) ((a) > (b) ? (a) : (b))
243 #define MIN(a, b) ((a) < (b) ? (a) : (b))
245 typedef char boolean;
249 /* These are the command codes that appear in compiled regular
250 expressions. Some opcodes are followed by argument bytes. A
251 command code can specify any interpretation whatsoever for its
252 arguments. Zero bytes may appear in the compiled regular expression.
254 The value of `exactn' is needed in search.c (search_buffer) in Emacs.
255 So regex.h defines a symbol `RE_EXACTN_VALUE' to be 1; the value of
256 `exactn' we use here must also be 1. */
262 /* Followed by one byte giving n, then by n literal bytes. */
265 /* Matches any (more or less) character. */
268 /* Matches any one char belonging to specified set. First
269 following byte is number of bitmap bytes. Then come bytes
270 for a bitmap saying which chars are in. Bits in each byte
271 are ordered low-bit-first. A character is in the set if its
272 bit is 1. A character too large to have a bit in the map is
273 automatically not in the set. */
276 /* Same parameters as charset, but match any character that is
277 not one of those specified. */
280 /* Start remembering the text that is matched, for storing in a
281 register. Followed by one byte with the register number, in
282 the range 0 to one less than the pattern buffer's re_nsub
283 field. Then followed by one byte with the number of groups
284 inner to this one. (This last has to be part of the
285 start_memory only because we need it in the on_failure_jump
289 /* Stop remembering the text that is matched and store it in a
290 memory register. Followed by one byte with the register
291 number, in the range 0 to one less than `re_nsub' in the
292 pattern buffer, and one byte with the number of inner groups,
293 just like `start_memory'. (We need the number of inner
294 groups here because we don't have any easy way of finding the
295 corresponding start_memory when we're at a stop_memory.) */
298 /* Match a duplicate of something remembered. Followed by one
299 byte containing the register number. */
302 /* Fail unless at beginning of line. */
305 /* Fail unless at end of line. */
308 /* Succeeds if at beginning of buffer (if emacs) or at beginning
309 of string to be matched (if not). */
312 /* Analogously, for end of buffer/string. */
315 /* Followed by two byte relative address to which to jump. */
318 /* Same as jump, but marks the end of an alternative. */
321 /* Followed by two-byte relative address of place to resume at
322 in case of failure. */
325 /* Like on_failure_jump, but pushes a placeholder instead of the
326 current string position when executed. */
327 on_failure_keep_string_jump,
329 /* Throw away latest failure point and then jump to following
330 two-byte relative address. */
333 /* Change to pop_failure_jump if know won't have to backtrack to
334 match; otherwise change to jump. This is used to jump
335 back to the beginning of a repeat. If what follows this jump
336 clearly won't match what the repeat does, such that we can be
337 sure that there is no use backtracking out of repetitions
338 already matched, then we change it to a pop_failure_jump.
339 Followed by two-byte address. */
342 /* Jump to following two-byte address, and push a dummy failure
343 point. This failure point will be thrown away if an attempt
344 is made to use it for a failure. A `+' construct makes this
345 before the first repeat. Also used as an intermediary kind
346 of jump when compiling an alternative. */
349 /* Push a dummy failure point and continue. Used at the end of
353 /* Followed by two-byte relative address and two-byte number n.
354 After matching N times, jump to the address upon failure. */
357 /* Followed by two-byte relative address, and two-byte number n.
358 Jump to the address N times, then fail. */
361 /* Set the following two-byte relative address to the
362 subsequent two-byte number. The address *includes* the two
366 wordchar, /* Matches any word-constituent character. */
367 notwordchar, /* Matches any char that is not a word-constituent. */
369 wordbeg, /* Succeeds if at word beginning. */
370 wordend, /* Succeeds if at word end. */
372 wordbound, /* Succeeds if at a word boundary. */
373 notwordbound /* Succeeds if not at a word boundary. */
376 ,before_dot, /* Succeeds if before point. */
377 at_dot, /* Succeeds if at point. */
378 after_dot, /* Succeeds if after point. */
380 /* Matches any character whose syntax is specified. Followed by
381 a byte which contains a syntax code, e.g., Sword. */
384 /* Matches any character whose syntax is not that specified. */
389 /* Common operations on the compiled pattern. */
391 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
393 #define STORE_NUMBER(destination, number) \
395 (destination)[0] = (number) & 0377; \
396 (destination)[1] = (number) >> 8; \
399 /* Same as STORE_NUMBER, except increment DESTINATION to
400 the byte after where the number is stored. Therefore, DESTINATION
401 must be an lvalue. */
403 #define STORE_NUMBER_AND_INCR(destination, number) \
405 STORE_NUMBER (destination, number); \
406 (destination) += 2; \
409 /* Put into DESTINATION a number stored in two contiguous bytes starting
412 #define EXTRACT_NUMBER(destination, source) \
414 (destination) = *(source) & 0377; \
415 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
420 extract_number (dest, source)
422 unsigned char *source;
424 int temp = SIGN_EXTEND_CHAR (*(source + 1));
425 *dest = *source & 0377;
429 #ifndef EXTRACT_MACROS /* To debug the macros. */
430 #undef EXTRACT_NUMBER
431 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
432 #endif /* not EXTRACT_MACROS */
436 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
437 SOURCE must be an lvalue. */
439 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
441 EXTRACT_NUMBER (destination, source); \
447 extract_number_and_incr (destination, source)
449 unsigned char **source;
451 extract_number (destination, *source);
455 #ifndef EXTRACT_MACROS
456 #undef EXTRACT_NUMBER_AND_INCR
457 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
458 extract_number_and_incr (&dest, &src)
459 #endif /* not EXTRACT_MACROS */
463 /* If DEBUG is defined, Regex prints many voluminous messages about what
464 it is doing (if the variable `debug' is nonzero). If linked with the
465 main program in `iregex.c', you can enter patterns and strings
466 interactively. And if linked with the main program in `main.c' and
467 the other test files, you can run the already-written tests. */
471 /* We use standard I/O for debugging. */
474 /* It is useful to test things that ``must'' be true when debugging. */
477 static int debug = 0;
479 #define DEBUG_STATEMENT(e) e
480 #define DEBUG_PRINT1(x) if (debug) printf (x)
481 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
482 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
483 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
484 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
485 if (debug) print_partial_compiled_pattern (s, e)
486 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
487 if (debug) print_double_string (w, s1, sz1, s2, sz2)
490 extern void printchar ();
492 /* Print the fastmap in human-readable form. */
495 print_fastmap (fastmap)
498 unsigned was_a_range = 0;
501 while (i < (1 << BYTEWIDTH))
507 while (i < (1 << BYTEWIDTH) && fastmap[i])
523 /* Print a compiled pattern string in human-readable form, starting at
524 the START pointer into it and ending just before the pointer END. */
527 print_partial_compiled_pattern (start, end)
528 unsigned char *start;
532 unsigned char *p = start;
533 unsigned char *pend = end;
541 /* Loop over pattern commands. */
544 switch ((re_opcode_t) *p++)
552 printf ("/exactn/%d", mcnt);
563 printf ("/start_memory/%d/%d", mcnt, *p++);
568 printf ("/stop_memory/%d/%d", mcnt, *p++);
572 printf ("/duplicate/%d", *p++);
584 printf ("/charset%s",
585 (re_opcode_t) *(p - 1) == charset_not ? "_not" : "");
587 assert (p + *p < pend);
589 for (c = 0; c < *p; c++)
592 unsigned char map_byte = p[1 + c];
596 for (bit = 0; bit < BYTEWIDTH; bit++)
597 if (map_byte & (1 << bit))
598 printchar (c * BYTEWIDTH + bit);
612 case on_failure_jump:
613 extract_number_and_incr (&mcnt, &p);
614 printf ("/on_failure_jump/0/%d", mcnt);
617 case on_failure_keep_string_jump:
618 extract_number_and_incr (&mcnt, &p);
619 printf ("/on_failure_keep_string_jump/0/%d", mcnt);
622 case dummy_failure_jump:
623 extract_number_and_incr (&mcnt, &p);
624 printf ("/dummy_failure_jump/0/%d", mcnt);
627 case push_dummy_failure:
628 printf ("/push_dummy_failure");
632 extract_number_and_incr (&mcnt, &p);
633 printf ("/maybe_pop_jump/0/%d", mcnt);
636 case pop_failure_jump:
637 extract_number_and_incr (&mcnt, &p);
638 printf ("/pop_failure_jump/0/%d", mcnt);
642 extract_number_and_incr (&mcnt, &p);
643 printf ("/jump_past_alt/0/%d", mcnt);
647 extract_number_and_incr (&mcnt, &p);
648 printf ("/jump/0/%d", mcnt);
652 extract_number_and_incr (&mcnt, &p);
653 extract_number_and_incr (&mcnt2, &p);
654 printf ("/succeed_n/0/%d/0/%d", mcnt, mcnt2);
658 extract_number_and_incr (&mcnt, &p);
659 extract_number_and_incr (&mcnt2, &p);
660 printf ("/jump_n/0/%d/0/%d", mcnt, mcnt2);
664 extract_number_and_incr (&mcnt, &p);
665 extract_number_and_incr (&mcnt2, &p);
666 printf ("/set_number_at/0/%d/0/%d", mcnt, mcnt2);
670 printf ("/wordbound");
674 printf ("/notwordbound");
686 printf ("/before_dot");
694 printf ("/after_dot");
698 printf ("/syntaxspec");
700 printf ("/%d", mcnt);
704 printf ("/notsyntaxspec");
706 printf ("/%d", mcnt);
711 printf ("/wordchar");
715 printf ("/notwordchar");
727 printf ("?%d", *(p-1));
735 print_compiled_pattern (bufp)
736 struct re_pattern_buffer *bufp;
738 unsigned char *buffer = bufp->buffer;
740 print_partial_compiled_pattern (buffer, buffer + bufp->used);
741 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
743 if (bufp->fastmap_accurate && bufp->fastmap)
745 printf ("fastmap: ");
746 print_fastmap (bufp->fastmap);
749 printf ("re_nsub: %d\t", bufp->re_nsub);
750 printf ("regs_alloc: %d\t", bufp->regs_allocated);
751 printf ("can_be_null: %d\t", bufp->can_be_null);
752 printf ("newline_anchor: %d\n", bufp->newline_anchor);
753 printf ("no_sub: %d\t", bufp->no_sub);
754 printf ("not_bol: %d\t", bufp->not_bol);
755 printf ("not_eol: %d\t", bufp->not_eol);
756 printf ("syntax: %d\n", bufp->syntax);
757 /* Perhaps we should print the translate table? */
762 print_double_string (where, string1, size1, string2, size2)
775 if (FIRST_STRING_P (where))
777 for (this_char = where - string1; this_char < size1; this_char++)
778 printchar (string1[this_char]);
783 for (this_char = where - string2; this_char < size2; this_char++)
784 printchar (string2[this_char]);
788 #else /* not DEBUG */
793 #define DEBUG_STATEMENT(e)
794 #define DEBUG_PRINT1(x)
795 #define DEBUG_PRINT2(x1, x2)
796 #define DEBUG_PRINT3(x1, x2, x3)
797 #define DEBUG_PRINT4(x1, x2, x3, x4)
798 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
799 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
801 #endif /* not DEBUG */
803 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
804 also be assigned to arbitrarily: each pattern buffer stores its own
805 syntax, so it can be changed between regex compilations. */
806 reg_syntax_t re_syntax_options = RE_SYNTAX_EMACS;
809 /* Specify the precise syntax of regexps for compilation. This provides
810 for compatibility for various utilities which historically have
811 different, incompatible syntaxes.
813 The argument SYNTAX is a bit mask comprised of the various bits
814 defined in regex.h. We return the old syntax. */
817 re_set_syntax (syntax)
820 reg_syntax_t ret = re_syntax_options;
822 re_syntax_options = syntax;
826 /* This table gives an error message for each of the error codes listed
827 in regex.h. Obviously the order here has to be same as there. */
829 static const char *re_error_msg[] =
830 { NULL, /* REG_NOERROR */
831 "No match", /* REG_NOMATCH */
832 "Invalid regular expression", /* REG_BADPAT */
833 "Invalid collation character", /* REG_ECOLLATE */
834 "Invalid character class name", /* REG_ECTYPE */
835 "Trailing backslash", /* REG_EESCAPE */
836 "Invalid back reference", /* REG_ESUBREG */
837 "Unmatched [ or [^", /* REG_EBRACK */
838 "Unmatched ( or \\(", /* REG_EPAREN */
839 "Unmatched \\{", /* REG_EBRACE */
840 "Invalid content of \\{\\}", /* REG_BADBR */
841 "Invalid range end", /* REG_ERANGE */
842 "Memory exhausted", /* REG_ESPACE */
843 "Invalid preceding regular expression", /* REG_BADRPT */
844 "Premature end of regular expression", /* REG_EEND */
845 "Regular expression too big", /* REG_ESIZE */
846 "Unmatched ) or \\)", /* REG_ERPAREN */
849 /* Subroutine declarations and macros for regex_compile. */
851 static void store_op1 (), store_op2 ();
852 static void insert_op1 (), insert_op2 ();
853 static boolean at_begline_loc_p (), at_endline_loc_p ();
854 static boolean group_in_compile_stack ();
855 static reg_errcode_t compile_range ();
857 /* Fetch the next character in the uncompiled pattern---translating it
858 if necessary. Also cast from a signed character in the constant
859 string passed to us by the user to an unsigned char that we can use
860 as an array index (in, e.g., `translate'). */
861 #define PATFETCH(c) \
862 do {if (p == pend) return REG_EEND; \
863 c = (unsigned char) *p++; \
864 if (translate) c = translate[c]; \
867 /* Fetch the next character in the uncompiled pattern, with no
869 #define PATFETCH_RAW(c) \
870 do {if (p == pend) return REG_EEND; \
871 c = (unsigned char) *p++; \
874 /* Go backwards one character in the pattern. */
875 #define PATUNFETCH p--
878 /* If `translate' is non-null, return translate[D], else just D. We
879 cast the subscript to translate because some data is declared as
880 `char *', to avoid warnings when a string constant is passed. But
881 when we use a character as a subscript we must make it unsigned. */
882 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
885 /* Macros for outputting the compiled pattern into `buffer'. */
887 /* If the buffer isn't allocated when it comes in, use this. */
888 #define INIT_BUF_SIZE 32
890 /* Make sure we have at least N more bytes of space in buffer. */
891 #define GET_BUFFER_SPACE(n) \
892 while (b - bufp->buffer + (n) > bufp->allocated) \
895 /* Make sure we have one more byte of buffer space and then add C to it. */
896 #define BUF_PUSH(c) \
898 GET_BUFFER_SPACE (1); \
899 *b++ = (unsigned char) (c); \
903 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
904 #define BUF_PUSH_2(c1, c2) \
906 GET_BUFFER_SPACE (2); \
907 *b++ = (unsigned char) (c1); \
908 *b++ = (unsigned char) (c2); \
912 /* As with BUF_PUSH_2, except for three bytes. */
913 #define BUF_PUSH_3(c1, c2, c3) \
915 GET_BUFFER_SPACE (3); \
916 *b++ = (unsigned char) (c1); \
917 *b++ = (unsigned char) (c2); \
918 *b++ = (unsigned char) (c3); \
922 /* Store a jump with opcode OP at LOC to location TO. We store a
923 relative address offset by the three bytes the jump itself occupies. */
924 #define STORE_JUMP(op, loc, to) \
925 store_op1 (op, loc, (to) - (loc) - 3)
927 /* Likewise, for a two-argument jump. */
928 #define STORE_JUMP2(op, loc, to, arg) \
929 store_op2 (op, loc, (to) - (loc) - 3, arg)
931 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
932 #define INSERT_JUMP(op, loc, to) \
933 insert_op1 (op, loc, (to) - (loc) - 3, b)
935 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
936 #define INSERT_JUMP2(op, loc, to, arg) \
937 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
940 /* This is not an arbitrary limit: the arguments which represent offsets
941 into the pattern are two bytes long. So if 2^16 bytes turns out to
942 be too small, many things would have to change. */
943 #define MAX_BUF_SIZE (1L << 16)
946 /* Extend the buffer by twice its current size via realloc and
947 reset the pointers that pointed into the old block to point to the
948 correct places in the new one. If extending the buffer results in it
949 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
950 #define EXTEND_BUFFER() \
952 unsigned char *old_buffer = bufp->buffer; \
953 if (bufp->allocated == MAX_BUF_SIZE) \
955 bufp->allocated <<= 1; \
956 if (bufp->allocated > MAX_BUF_SIZE) \
957 bufp->allocated = MAX_BUF_SIZE; \
958 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
959 if (bufp->buffer == NULL) \
961 /* If the buffer moved, move all the pointers into it. */ \
962 if (old_buffer != bufp->buffer) \
964 b = (b - old_buffer) + bufp->buffer; \
965 begalt = (begalt - old_buffer) + bufp->buffer; \
966 if (fixup_alt_jump) \
967 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
969 laststart = (laststart - old_buffer) + bufp->buffer; \
971 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
976 /* Since we have one byte reserved for the register number argument to
977 {start,stop}_memory, the maximum number of groups we can report
978 things about is what fits in that byte. */
979 #define MAX_REGNUM 255
981 /* But patterns can have more than `MAX_REGNUM' registers. We just
982 ignore the excess. */
983 typedef unsigned regnum_t;
986 /* Macros for the compile stack. */
988 /* Since offsets can go either forwards or backwards, this type needs to
989 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
990 typedef int pattern_offset_t;
994 pattern_offset_t begalt_offset;
995 pattern_offset_t fixup_alt_jump;
996 pattern_offset_t inner_group_offset;
997 pattern_offset_t laststart_offset;
999 } compile_stack_elt_t;
1004 compile_stack_elt_t *stack;
1006 unsigned avail; /* Offset of next open position. */
1007 } compile_stack_type;
1010 #define INIT_COMPILE_STACK_SIZE 32
1012 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1013 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1015 /* The next available element. */
1016 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1019 /* Set the bit for character C in a list. */
1020 #define SET_LIST_BIT(c) \
1021 (b[((unsigned char) (c)) / BYTEWIDTH] \
1022 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1025 /* Get the next unsigned number in the uncompiled pattern. */
1026 #define GET_UNSIGNED_NUMBER(num) \
1030 while (ISDIGIT (c)) \
1034 num = num * 10 + c - '0'; \
1042 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1044 #define IS_CHAR_CLASS(string) \
1045 (STREQ (string, "alpha") || STREQ (string, "upper") \
1046 || STREQ (string, "lower") || STREQ (string, "digit") \
1047 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1048 || STREQ (string, "space") || STREQ (string, "print") \
1049 || STREQ (string, "punct") || STREQ (string, "graph") \
1050 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1052 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1053 Returns one of error codes defined in `regex.h', or zero for success.
1055 Assumes the `allocated' (and perhaps `buffer') and `translate'
1056 fields are set in BUFP on entry.
1058 If it succeeds, results are put in BUFP (if it returns an error, the
1059 contents of BUFP are undefined):
1060 `buffer' is the compiled pattern;
1061 `syntax' is set to SYNTAX;
1062 `used' is set to the length of the compiled pattern;
1063 `fastmap_accurate' is zero;
1064 `re_nsub' is the number of subexpressions in PATTERN;
1065 `not_bol' and `not_eol' are zero;
1067 The `fastmap' and `newline_anchor' fields are neither
1068 examined nor set. */
1070 static reg_errcode_t
1071 regex_compile (pattern, size, syntax, bufp)
1072 const char *pattern;
1074 reg_syntax_t syntax;
1075 struct re_pattern_buffer *bufp;
1077 /* We fetch characters from PATTERN here. Even though PATTERN is
1078 `char *' (i.e., signed), we declare these variables as unsigned, so
1079 they can be reliably used as array indices. */
1080 register unsigned char c, c1;
1082 /* A random tempory spot in PATTERN. */
1085 /* Points to the end of the buffer, where we should append. */
1086 register unsigned char *b;
1088 /* Keeps track of unclosed groups. */
1089 compile_stack_type compile_stack;
1091 /* Points to the current (ending) position in the pattern. */
1092 const char *p = pattern;
1093 const char *pend = pattern + size;
1095 /* How to translate the characters in the pattern. */
1096 char *translate = bufp->translate;
1098 /* Address of the count-byte of the most recently inserted `exactn'
1099 command. This makes it possible to tell if a new exact-match
1100 character can be added to that command or if the character requires
1101 a new `exactn' command. */
1102 unsigned char *pending_exact = 0;
1104 /* Address of start of the most recently finished expression.
1105 This tells, e.g., postfix * where to find the start of its
1106 operand. Reset at the beginning of groups and alternatives. */
1107 unsigned char *laststart = 0;
1109 /* Address of beginning of regexp, or inside of last group. */
1110 unsigned char *begalt;
1112 /* Place in the uncompiled pattern (i.e., the {) to
1113 which to go back if the interval is invalid. */
1114 const char *beg_interval;
1116 /* Address of the place where a forward jump should go to the end of
1117 the containing expression. Each alternative of an `or' -- except the
1118 last -- ends with a forward jump of this sort. */
1119 unsigned char *fixup_alt_jump = 0;
1121 /* Counts open-groups as they are encountered. Remembered for the
1122 matching close-group on the compile stack, so the same register
1123 number is put in the stop_memory as the start_memory. */
1124 regnum_t regnum = 0;
1127 DEBUG_PRINT1 ("\nCompiling pattern: ");
1130 unsigned debug_count;
1132 for (debug_count = 0; debug_count < size; debug_count++)
1133 printchar (pattern[debug_count]);
1138 /* Initialize the compile stack. */
1139 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1140 if (compile_stack.stack == NULL)
1143 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1144 compile_stack.avail = 0;
1146 /* Initialize the pattern buffer. */
1147 bufp->syntax = syntax;
1148 bufp->fastmap_accurate = 0;
1149 bufp->not_bol = bufp->not_eol = 0;
1151 /* Set `used' to zero, so that if we return an error, the pattern
1152 printer (for debugging) will think there's no pattern. We reset it
1156 /* Always count groups, whether or not bufp->no_sub is set. */
1159 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1160 /* Initialize the syntax table. */
1161 init_syntax_once ();
1164 if (bufp->allocated == 0)
1167 { /* If zero allocated, but buffer is non-null, try to realloc
1168 enough space. This loses if buffer's address is bogus, but
1169 that is the user's responsibility. */
1170 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1173 { /* Caller did not allocate a buffer. Do it for them. */
1174 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1176 if (!bufp->buffer) return REG_ESPACE;
1178 bufp->allocated = INIT_BUF_SIZE;
1181 begalt = b = bufp->buffer;
1183 /* Loop through the uncompiled pattern until we're at the end. */
1192 if ( /* If at start of pattern, it's an operator. */
1194 /* If context independent, it's an operator. */
1195 || syntax & RE_CONTEXT_INDEP_ANCHORS
1196 /* Otherwise, depends on what's come before. */
1197 || at_begline_loc_p (pattern, p, syntax))
1207 if ( /* If at end of pattern, it's an operator. */
1209 /* If context independent, it's an operator. */
1210 || syntax & RE_CONTEXT_INDEP_ANCHORS
1211 /* Otherwise, depends on what's next. */
1212 || at_endline_loc_p (p, pend, syntax))
1222 if ((syntax & RE_BK_PLUS_QM)
1223 || (syntax & RE_LIMITED_OPS))
1227 /* If there is no previous pattern... */
1230 if (syntax & RE_CONTEXT_INVALID_OPS)
1232 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1237 /* Are we optimizing this jump? */
1238 boolean keep_string_p = false;
1240 /* 1 means zero (many) matches is allowed. */
1241 char zero_times_ok = 0, many_times_ok = 0;
1243 /* If there is a sequence of repetition chars, collapse it
1244 down to just one (the right one). We can't combine
1245 interval operators with these because of, e.g., `a{2}*',
1246 which should only match an even number of `a's. */
1250 zero_times_ok |= c != '+';
1251 many_times_ok |= c != '?';
1259 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1262 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1264 if (p == pend) return REG_EESCAPE;
1267 if (!(c1 == '+' || c1 == '?'))
1282 /* If we get here, we found another repeat character. */
1285 /* Star, etc. applied to an empty pattern is equivalent
1286 to an empty pattern. */
1290 /* Now we know whether or not zero matches is allowed
1291 and also whether or not two or more matches is allowed. */
1293 { /* More than one repetition is allowed, so put in at the
1294 end a backward relative jump from `b' to before the next
1295 jump we're going to put in below (which jumps from
1296 laststart to after this jump).
1298 But if we are at the `*' in the exact sequence `.*\n',
1299 insert an unconditional jump backwards to the .,
1300 instead of the beginning of the loop. This way we only
1301 push a failure point once, instead of every time
1302 through the loop. */
1303 assert (p - 1 > pattern);
1305 /* Allocate the space for the jump. */
1306 GET_BUFFER_SPACE (3);
1308 /* We know we are not at the first character of the pattern,
1309 because laststart was nonzero. And we've already
1310 incremented `p', by the way, to be the character after
1311 the `*'. Do we have to do something analogous here
1312 for null bytes, because of RE_DOT_NOT_NULL? */
1313 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1315 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1316 && !(syntax & RE_DOT_NEWLINE))
1317 { /* We have .*\n. */
1318 STORE_JUMP (jump, b, laststart);
1319 keep_string_p = true;
1322 /* Anything else. */
1323 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1325 /* We've added more stuff to the buffer. */
1329 /* On failure, jump from laststart to b + 3, which will be the
1330 end of the buffer after this jump is inserted. */
1331 GET_BUFFER_SPACE (3);
1332 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1340 /* At least one repetition is required, so insert a
1341 `dummy_failure_jump' before the initial
1342 `on_failure_jump' instruction of the loop. This
1343 effects a skip over that instruction the first time
1344 we hit that loop. */
1345 GET_BUFFER_SPACE (3);
1346 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1361 boolean had_char_class = false;
1363 if (p == pend) return REG_EBRACK;
1365 /* Ensure that we have enough space to push a charset: the
1366 opcode, the length count, and the bitset; 34 bytes in all. */
1367 GET_BUFFER_SPACE (34);
1371 /* We test `*p == '^' twice, instead of using an if
1372 statement, so we only need one BUF_PUSH. */
1373 BUF_PUSH (*p == '^' ? charset_not : charset);
1377 /* Remember the first position in the bracket expression. */
1380 /* Push the number of bytes in the bitmap. */
1381 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1383 /* Clear the whole map. */
1384 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1386 /* charset_not matches newline according to a syntax bit. */
1387 if ((re_opcode_t) b[-2] == charset_not
1388 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1389 SET_LIST_BIT ('\n');
1391 /* Read in characters and ranges, setting map bits. */
1394 if (p == pend) return REG_EBRACK;
1398 /* \ might escape characters inside [...] and [^...]. */
1399 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1401 if (p == pend) return REG_EESCAPE;
1408 /* Could be the end of the bracket expression. If it's
1409 not (i.e., when the bracket expression is `[]' so
1410 far), the ']' character bit gets set way below. */
1411 if (c == ']' && p != p1 + 1)
1414 /* Look ahead to see if it's a range when the last thing
1415 was a character class. */
1416 if (had_char_class && c == '-' && *p != ']')
1419 /* Look ahead to see if it's a range when the last thing
1420 was a character: if this is a hyphen not at the
1421 beginning or the end of a list, then it's the range
1424 && !(p - 2 >= pattern && p[-2] == '[')
1425 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1429 = compile_range (&p, pend, translate, syntax, b);
1430 if (ret != REG_NOERROR) return ret;
1433 else if (p[0] == '-' && p[1] != ']')
1434 { /* This handles ranges made up of characters only. */
1437 /* Move past the `-'. */
1440 ret = compile_range (&p, pend, translate, syntax, b);
1441 if (ret != REG_NOERROR) return ret;
1444 /* See if we're at the beginning of a possible character
1447 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1448 { /* Leave room for the null. */
1449 char str[CHAR_CLASS_MAX_LENGTH + 1];
1454 /* If pattern is `[[:'. */
1455 if (p == pend) return REG_EBRACK;
1460 if (c == ':' || c == ']' || p == pend
1461 || c1 == CHAR_CLASS_MAX_LENGTH)
1467 /* If isn't a word bracketed by `[:' and:`]':
1468 undo the ending character, the letters, and leave
1469 the leading `:' and `[' (but set bits for them). */
1470 if (c == ':' && *p == ']')
1473 boolean is_alnum = STREQ (str, "alnum");
1474 boolean is_alpha = STREQ (str, "alpha");
1475 boolean is_blank = STREQ (str, "blank");
1476 boolean is_cntrl = STREQ (str, "cntrl");
1477 boolean is_digit = STREQ (str, "digit");
1478 boolean is_graph = STREQ (str, "graph");
1479 boolean is_lower = STREQ (str, "lower");
1480 boolean is_print = STREQ (str, "print");
1481 boolean is_punct = STREQ (str, "punct");
1482 boolean is_space = STREQ (str, "space");
1483 boolean is_upper = STREQ (str, "upper");
1484 boolean is_xdigit = STREQ (str, "xdigit");
1486 if (!IS_CHAR_CLASS (str)) return REG_ECTYPE;
1488 /* Throw away the ] at the end of the character
1492 if (p == pend) return REG_EBRACK;
1494 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
1496 if ( (is_alnum && ISALNUM (ch))
1497 || (is_alpha && ISALPHA (ch))
1498 || (is_blank && ISBLANK (ch))
1499 || (is_cntrl && ISCNTRL (ch))
1500 || (is_digit && ISDIGIT (ch))
1501 || (is_graph && ISGRAPH (ch))
1502 || (is_lower && ISLOWER (ch))
1503 || (is_print && ISPRINT (ch))
1504 || (is_punct && ISPUNCT (ch))
1505 || (is_space && ISSPACE (ch))
1506 || (is_upper && ISUPPER (ch))
1507 || (is_xdigit && ISXDIGIT (ch)))
1510 had_char_class = true;
1519 had_char_class = false;
1524 had_char_class = false;
1529 /* Discard any (non)matching list bytes that are all 0 at the
1530 end of the map. Decrease the map-length byte too. */
1531 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
1539 if (syntax & RE_NO_BK_PARENS)
1546 if (syntax & RE_NO_BK_PARENS)
1553 if (syntax & RE_NEWLINE_ALT)
1560 if (syntax & RE_NO_BK_VBAR)
1567 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
1568 goto handle_interval;
1574 if (p == pend) return REG_EESCAPE;
1576 /* Do not translate the character after the \, so that we can
1577 distinguish, e.g., \B from \b, even if we normally would
1578 translate, e.g., B to b. */
1584 if (syntax & RE_NO_BK_PARENS)
1585 goto normal_backslash;
1591 if (COMPILE_STACK_FULL)
1593 RETALLOC (compile_stack.stack, compile_stack.size << 1,
1594 compile_stack_elt_t);
1595 if (compile_stack.stack == NULL) return REG_ESPACE;
1597 compile_stack.size <<= 1;
1600 /* These are the values to restore when we hit end of this
1601 group. They are all relative offsets, so that if the
1602 whole pattern moves because of realloc, they will still
1604 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
1605 COMPILE_STACK_TOP.fixup_alt_jump
1606 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
1607 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
1608 COMPILE_STACK_TOP.regnum = regnum;
1610 /* We will eventually replace the 0 with the number of
1611 groups inner to this one. But do not push a
1612 start_memory for groups beyond the last one we can
1613 represent in the compiled pattern. */
1614 if (regnum <= MAX_REGNUM)
1616 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
1617 BUF_PUSH_3 (start_memory, regnum, 0);
1620 compile_stack.avail++;
1625 /* If we've reached MAX_REGNUM groups, then this open
1626 won't actually generate any code, so we'll have to
1627 clear pending_exact explicitly. */
1633 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
1635 if (COMPILE_STACK_EMPTY)
1637 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
1638 goto normal_backslash;
1645 { /* Push a dummy failure point at the end of the
1646 alternative for a possible future
1647 `pop_failure_jump' to pop. See comments at
1648 `push_dummy_failure' in `re_match_2'. */
1649 BUF_PUSH (push_dummy_failure);
1651 /* We allocated space for this jump when we assigned
1652 to `fixup_alt_jump', in the `handle_alt' case below. */
1653 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
1656 /* See similar code for backslashed left paren above. */
1657 if (COMPILE_STACK_EMPTY)
1659 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
1665 /* Since we just checked for an empty stack above, this
1666 ``can't happen''. */
1667 assert (compile_stack.avail != 0);
1669 /* We don't just want to restore into `regnum', because
1670 later groups should continue to be numbered higher,
1671 as in `(ab)c(de)' -- the second group is #2. */
1672 regnum_t this_group_regnum;
1674 compile_stack.avail--;
1675 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
1677 = COMPILE_STACK_TOP.fixup_alt_jump
1678 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
1680 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
1681 this_group_regnum = COMPILE_STACK_TOP.regnum;
1682 /* If we've reached MAX_REGNUM groups, then this open
1683 won't actually generate any code, so we'll have to
1684 clear pending_exact explicitly. */
1687 /* We're at the end of the group, so now we know how many
1688 groups were inside this one. */
1689 if (this_group_regnum <= MAX_REGNUM)
1691 unsigned char *inner_group_loc
1692 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
1694 *inner_group_loc = regnum - this_group_regnum;
1695 BUF_PUSH_3 (stop_memory, this_group_regnum,
1696 regnum - this_group_regnum);
1702 case '|': /* `\|'. */
1703 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
1704 goto normal_backslash;
1706 if (syntax & RE_LIMITED_OPS)
1709 /* Insert before the previous alternative a jump which
1710 jumps to this alternative if the former fails. */
1711 GET_BUFFER_SPACE (3);
1712 INSERT_JUMP (on_failure_jump, begalt, b + 6);
1716 /* The alternative before this one has a jump after it
1717 which gets executed if it gets matched. Adjust that
1718 jump so it will jump to this alternative's analogous
1719 jump (put in below, which in turn will jump to the next
1720 (if any) alternative's such jump, etc.). The last such
1721 jump jumps to the correct final destination. A picture:
1727 If we are at `b', then fixup_alt_jump right now points to a
1728 three-byte space after `a'. We'll put in the jump, set
1729 fixup_alt_jump to right after `b', and leave behind three
1730 bytes which we'll fill in when we get to after `c'. */
1733 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
1735 /* Mark and leave space for a jump after this alternative,
1736 to be filled in later either by next alternative or
1737 when know we're at the end of a series of alternatives. */
1739 GET_BUFFER_SPACE (3);
1748 /* If \{ is a literal. */
1749 if (!(syntax & RE_INTERVALS)
1750 /* If we're at `\{' and it's not the open-interval
1752 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
1753 || (p - 2 == pattern && p == pend))
1754 goto normal_backslash;
1758 /* If got here, then the syntax allows intervals. */
1760 /* At least (most) this many matches must be made. */
1761 int lower_bound = -1, upper_bound = -1;
1763 beg_interval = p - 1;
1767 if (syntax & RE_NO_BK_BRACES)
1768 goto unfetch_interval;
1773 GET_UNSIGNED_NUMBER (lower_bound);
1777 GET_UNSIGNED_NUMBER (upper_bound);
1778 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
1781 /* Interval such as `{1}' => match exactly once. */
1782 upper_bound = lower_bound;
1784 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
1785 || lower_bound > upper_bound)
1787 if (syntax & RE_NO_BK_BRACES)
1788 goto unfetch_interval;
1793 if (!(syntax & RE_NO_BK_BRACES))
1795 if (c != '\\') return REG_EBRACE;
1802 if (syntax & RE_NO_BK_BRACES)
1803 goto unfetch_interval;
1808 /* We just parsed a valid interval. */
1810 /* If it's invalid to have no preceding re. */
1813 if (syntax & RE_CONTEXT_INVALID_OPS)
1815 else if (syntax & RE_CONTEXT_INDEP_OPS)
1818 goto unfetch_interval;
1821 /* If the upper bound is zero, don't want to succeed at
1822 all; jump from `laststart' to `b + 3', which will be
1823 the end of the buffer after we insert the jump. */
1824 if (upper_bound == 0)
1826 GET_BUFFER_SPACE (3);
1827 INSERT_JUMP (jump, laststart, b + 3);
1831 /* Otherwise, we have a nontrivial interval. When
1832 we're all done, the pattern will look like:
1833 set_number_at <jump count> <upper bound>
1834 set_number_at <succeed_n count> <lower bound>
1835 succeed_n <after jump addr> <succed_n count>
1837 jump_n <succeed_n addr> <jump count>
1838 (The upper bound and `jump_n' are omitted if
1839 `upper_bound' is 1, though.) */
1841 { /* If the upper bound is > 1, we need to insert
1842 more at the end of the loop. */
1843 unsigned nbytes = 10 + (upper_bound > 1) * 10;
1845 GET_BUFFER_SPACE (nbytes);
1847 /* Initialize lower bound of the `succeed_n', even
1848 though it will be set during matching by its
1849 attendant `set_number_at' (inserted next),
1850 because `re_compile_fastmap' needs to know.
1851 Jump to the `jump_n' we might insert below. */
1852 INSERT_JUMP2 (succeed_n, laststart,
1853 b + 5 + (upper_bound > 1) * 5,
1857 /* Code to initialize the lower bound. Insert
1858 before the `succeed_n'. The `5' is the last two
1859 bytes of this `set_number_at', plus 3 bytes of
1860 the following `succeed_n'. */
1861 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
1864 if (upper_bound > 1)
1865 { /* More than one repetition is allowed, so
1866 append a backward jump to the `succeed_n'
1867 that starts this interval.
1869 When we've reached this during matching,
1870 we'll have matched the interval once, so
1871 jump back only `upper_bound - 1' times. */
1872 STORE_JUMP2 (jump_n, b, laststart + 5,
1876 /* The location we want to set is the second
1877 parameter of the `jump_n'; that is `b-2' as
1878 an absolute address. `laststart' will be
1879 the `set_number_at' we're about to insert;
1880 `laststart+3' the number to set, the source
1881 for the relative address. But we are
1882 inserting into the middle of the pattern --
1883 so everything is getting moved up by 5.
1884 Conclusion: (b - 2) - (laststart + 3) + 5,
1885 i.e., b - laststart.
1887 We insert this at the beginning of the loop
1888 so that if we fail during matching, we'll
1889 reinitialize the bounds. */
1890 insert_op2 (set_number_at, laststart, b - laststart,
1891 upper_bound - 1, b);
1896 beg_interval = NULL;
1901 /* If an invalid interval, match the characters as literals. */
1902 assert (beg_interval);
1904 beg_interval = NULL;
1906 /* normal_char and normal_backslash need `c'. */
1909 if (!(syntax & RE_NO_BK_BRACES))
1911 if (p > pattern && p[-1] == '\\')
1912 goto normal_backslash;
1917 /* There is no way to specify the before_dot and after_dot
1918 operators. rms says this is ok. --karl */
1926 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
1932 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
1939 BUF_PUSH (wordchar);
1945 BUF_PUSH (notwordchar);
1958 BUF_PUSH (wordbound);
1962 BUF_PUSH (notwordbound);
1973 case '1': case '2': case '3': case '4': case '5':
1974 case '6': case '7': case '8': case '9':
1975 if (syntax & RE_NO_BK_REFS)
1983 /* Can't back reference to a subexpression if inside of it. */
1984 if (group_in_compile_stack (compile_stack, c1))
1988 BUF_PUSH_2 (duplicate, c1);
1994 if (syntax & RE_BK_PLUS_QM)
1997 goto normal_backslash;
2001 /* You might think it would be useful for \ to mean
2002 not to translate; but if we don't translate it
2003 it will never match anything. */
2011 /* Expects the character in `c'. */
2013 /* If no exactn currently being built. */
2016 /* If last exactn not at current position. */
2017 || pending_exact + *pending_exact + 1 != b
2019 /* We have only one byte following the exactn for the count. */
2020 || *pending_exact == (1 << BYTEWIDTH) - 1
2022 /* If followed by a repetition operator. */
2023 || *p == '*' || *p == '^'
2024 || ((syntax & RE_BK_PLUS_QM)
2025 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2026 : (*p == '+' || *p == '?'))
2027 || ((syntax & RE_INTERVALS)
2028 && ((syntax & RE_NO_BK_BRACES)
2030 : (p[0] == '\\' && p[1] == '{'))))
2032 /* Start building a new exactn. */
2036 BUF_PUSH_2 (exactn, 0);
2037 pending_exact = b - 1;
2044 } /* while p != pend */
2047 /* Through the pattern now. */
2050 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2052 if (!COMPILE_STACK_EMPTY)
2055 free (compile_stack.stack);
2057 /* We have succeeded; set the length of the buffer. */
2058 bufp->used = b - bufp->buffer;
2063 DEBUG_PRINT1 ("\nCompiled pattern: ");
2064 print_compiled_pattern (bufp);
2069 } /* regex_compile */
2071 /* Subroutines for `regex_compile'. */
2073 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2076 store_op1 (op, loc, arg)
2081 *loc = (unsigned char) op;
2082 STORE_NUMBER (loc + 1, arg);
2086 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2089 store_op2 (op, loc, arg1, arg2)
2094 *loc = (unsigned char) op;
2095 STORE_NUMBER (loc + 1, arg1);
2096 STORE_NUMBER (loc + 3, arg2);
2100 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2101 for OP followed by two-byte integer parameter ARG. */
2104 insert_op1 (op, loc, arg, end)
2110 register unsigned char *pfrom = end;
2111 register unsigned char *pto = end + 3;
2113 while (pfrom != loc)
2116 store_op1 (op, loc, arg);
2120 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2123 insert_op2 (op, loc, arg1, arg2, end)
2129 register unsigned char *pfrom = end;
2130 register unsigned char *pto = end + 5;
2132 while (pfrom != loc)
2135 store_op2 (op, loc, arg1, arg2);
2139 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2140 after an alternative or a begin-subexpression. We assume there is at
2141 least one character before the ^. */
2144 at_begline_loc_p (pattern, p, syntax)
2145 const char *pattern, *p;
2146 reg_syntax_t syntax;
2148 const char *prev = p - 2;
2149 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2152 /* After a subexpression? */
2153 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2154 /* After an alternative? */
2155 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2159 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2160 at least one character after the $, i.e., `P < PEND'. */
2163 at_endline_loc_p (p, pend, syntax)
2164 const char *p, *pend;
2167 const char *next = p;
2168 boolean next_backslash = *next == '\\';
2169 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2172 /* Before a subexpression? */
2173 (syntax & RE_NO_BK_PARENS ? *next == ')'
2174 : next_backslash && next_next && *next_next == ')')
2175 /* Before an alternative? */
2176 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2177 : next_backslash && next_next && *next_next == '|');
2181 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2182 false if it's not. */
2185 group_in_compile_stack (compile_stack, regnum)
2186 compile_stack_type compile_stack;
2191 for (this_element = compile_stack.avail - 1;
2194 if (compile_stack.stack[this_element].regnum == regnum)
2201 /* Read the ending character of a range (in a bracket expression) from the
2202 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2203 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2204 Then we set the translation of all bits between the starting and
2205 ending characters (inclusive) in the compiled pattern B.
2207 Return an error code.
2209 We use these short variable names so we can use the same macros as
2210 `regex_compile' itself. */
2212 static reg_errcode_t
2213 compile_range (p_ptr, pend, translate, syntax, b)
2214 const char **p_ptr, *pend;
2216 reg_syntax_t syntax;
2221 const char *p = *p_ptr;
2222 int range_start, range_end;
2227 /* Even though the pattern is a signed `char *', we need to fetch
2228 with unsigned char *'s; if the high bit of the pattern character
2229 is set, the range endpoints will be negative if we fetch using a
2232 We also want to fetch the endpoints without translating them; the
2233 appropriate translation is done in the bit-setting loop below. */
2234 range_start = ((unsigned char *) p)[-2];
2235 range_end = ((unsigned char *) p)[0];
2237 /* Have to increment the pointer into the pattern string, so the
2238 caller isn't still at the ending character. */
2241 /* If the start is after the end, the range is empty. */
2242 if (range_start > range_end)
2243 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2245 /* Here we see why `this_char' has to be larger than an `unsigned
2246 char' -- the range is inclusive, so if `range_end' == 0xff
2247 (assuming 8-bit characters), we would otherwise go into an infinite
2248 loop, since all characters <= 0xff. */
2249 for (this_char = range_start; this_char <= range_end; this_char++)
2251 SET_LIST_BIT (TRANSLATE (this_char));
2257 /* Failure stack declarations and macros; both re_compile_fastmap and
2258 re_match_2 use a failure stack. These have to be macros because of
2262 /* Number of failure points for which to initially allocate space
2263 when matching. If this number is exceeded, we allocate more
2264 space, so it is not a hard limit. */
2265 #ifndef INIT_FAILURE_ALLOC
2266 #define INIT_FAILURE_ALLOC 5
2269 /* Roughly the maximum number of failure points on the stack. Would be
2270 exactly that if always used MAX_FAILURE_SPACE each time we failed.
2271 This is a variable only so users of regex can assign to it; we never
2272 change it ourselves. */
2273 int re_max_failures = 2000;
2275 typedef const unsigned char *fail_stack_elt_t;
2279 fail_stack_elt_t *stack;
2281 unsigned avail; /* Offset of next open position. */
2284 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
2285 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
2286 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
2287 #define FAIL_STACK_TOP() (fail_stack.stack[fail_stack.avail])
2290 /* Initialize `fail_stack'. Do `return -2' if the alloc fails. */
2292 #define INIT_FAIL_STACK() \
2294 fail_stack.stack = (fail_stack_elt_t *) \
2295 REGEX_ALLOCATE (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
2297 if (fail_stack.stack == NULL) \
2300 fail_stack.size = INIT_FAILURE_ALLOC; \
2301 fail_stack.avail = 0; \
2305 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
2307 Return 1 if succeeds, and 0 if either ran out of memory
2308 allocating space for it or it was already too large.
2310 REGEX_REALLOCATE requires `destination' be declared. */
2312 #define DOUBLE_FAIL_STACK(fail_stack) \
2313 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
2315 : ((fail_stack).stack = (fail_stack_elt_t *) \
2316 REGEX_REALLOCATE ((fail_stack).stack, \
2317 (fail_stack).size * sizeof (fail_stack_elt_t), \
2318 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
2320 (fail_stack).stack == NULL \
2322 : ((fail_stack).size <<= 1, \
2326 /* Push PATTERN_OP on FAIL_STACK.
2328 Return 1 if was able to do so and 0 if ran out of memory allocating
2330 #define PUSH_PATTERN_OP(pattern_op, fail_stack) \
2331 ((FAIL_STACK_FULL () \
2332 && !DOUBLE_FAIL_STACK (fail_stack)) \
2334 : ((fail_stack).stack[(fail_stack).avail++] = pattern_op, \
2337 /* This pushes an item onto the failure stack. Must be a four-byte
2338 value. Assumes the variable `fail_stack'. Probably should only
2339 be called from within `PUSH_FAILURE_POINT'. */
2340 #define PUSH_FAILURE_ITEM(item) \
2341 fail_stack.stack[fail_stack.avail++] = (fail_stack_elt_t) item
2343 /* The complement operation. Assumes `fail_stack' is nonempty. */
2344 #define POP_FAILURE_ITEM() fail_stack.stack[--fail_stack.avail]
2346 /* Used to omit pushing failure point id's when we're not debugging. */
2348 #define DEBUG_PUSH PUSH_FAILURE_ITEM
2349 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_ITEM ()
2351 #define DEBUG_PUSH(item)
2352 #define DEBUG_POP(item_addr)
2356 /* Push the information about the state we will need
2357 if we ever fail back to it.
2359 Requires variables fail_stack, regstart, regend, reg_info, and
2360 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
2363 Does `return FAILURE_CODE' if runs out of memory. */
2365 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
2367 char *destination; \
2368 /* Must be int, so when we don't save any registers, the arithmetic \
2369 of 0 + -1 isn't done as unsigned. */ \
2372 DEBUG_STATEMENT (failure_id++); \
2373 DEBUG_STATEMENT (nfailure_points_pushed++); \
2374 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
2375 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
2376 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
2378 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
2379 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
2381 /* Ensure we have enough space allocated for what we will push. */ \
2382 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
2384 if (!DOUBLE_FAIL_STACK (fail_stack)) \
2385 return failure_code; \
2387 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
2388 (fail_stack).size); \
2389 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
2392 /* Push the info, starting with the registers. */ \
2393 DEBUG_PRINT1 ("\n"); \
2395 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
2398 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
2399 DEBUG_STATEMENT (num_regs_pushed++); \
2401 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
2402 PUSH_FAILURE_ITEM (regstart[this_reg]); \
2404 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
2405 PUSH_FAILURE_ITEM (regend[this_reg]); \
2407 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
2408 DEBUG_PRINT2 (" match_null=%d", \
2409 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
2410 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
2411 DEBUG_PRINT2 (" matched_something=%d", \
2412 MATCHED_SOMETHING (reg_info[this_reg])); \
2413 DEBUG_PRINT2 (" ever_matched=%d", \
2414 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
2415 DEBUG_PRINT1 ("\n"); \
2416 PUSH_FAILURE_ITEM (reg_info[this_reg].word); \
2419 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
2420 PUSH_FAILURE_ITEM (lowest_active_reg); \
2422 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
2423 PUSH_FAILURE_ITEM (highest_active_reg); \
2425 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
2426 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
2427 PUSH_FAILURE_ITEM (pattern_place); \
2429 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
2430 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
2432 DEBUG_PRINT1 ("'\n"); \
2433 PUSH_FAILURE_ITEM (string_place); \
2435 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
2436 DEBUG_PUSH (failure_id); \
2439 /* This is the number of items that are pushed and popped on the stack
2440 for each register. */
2441 #define NUM_REG_ITEMS 3
2443 /* Individual items aside from the registers. */
2445 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
2447 #define NUM_NONREG_ITEMS 4
2450 /* We push at most this many items on the stack. */
2451 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
2453 /* We actually push this many items. */
2454 #define NUM_FAILURE_ITEMS \
2455 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
2458 /* How many items can still be added to the stack without overflowing it. */
2459 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
2462 /* Pops what PUSH_FAIL_STACK pushes.
2464 We restore into the parameters, all of which should be lvalues:
2465 STR -- the saved data position.
2466 PAT -- the saved pattern position.
2467 LOW_REG, HIGH_REG -- the highest and lowest active registers.
2468 REGSTART, REGEND -- arrays of string positions.
2469 REG_INFO -- array of information about each subexpression.
2471 Also assumes the variables `fail_stack' and (if debugging), `bufp',
2472 `pend', `string1', `size1', `string2', and `size2'. */
2474 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
2476 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
2478 const unsigned char *string_temp; \
2480 assert (!FAIL_STACK_EMPTY ()); \
2482 /* Remove failure points and point to how many regs pushed. */ \
2483 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
2484 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
2485 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
2487 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
2489 DEBUG_POP (&failure_id); \
2490 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
2492 /* If the saved string location is NULL, it came from an \
2493 on_failure_keep_string_jump opcode, and we want to throw away the \
2494 saved NULL, thus retaining our current position in the string. */ \
2495 string_temp = POP_FAILURE_ITEM (); \
2496 if (string_temp != NULL) \
2497 str = (const char *) string_temp; \
2499 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
2500 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
2501 DEBUG_PRINT1 ("'\n"); \
2503 pat = (unsigned char *) POP_FAILURE_ITEM (); \
2504 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
2505 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
2507 /* Restore register info. */ \
2508 high_reg = (unsigned) POP_FAILURE_ITEM (); \
2509 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
2511 low_reg = (unsigned) POP_FAILURE_ITEM (); \
2512 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
2514 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
2516 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
2518 reg_info[this_reg].word = POP_FAILURE_ITEM (); \
2519 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
2521 regend[this_reg] = (const char *) POP_FAILURE_ITEM (); \
2522 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
2524 regstart[this_reg] = (const char *) POP_FAILURE_ITEM (); \
2525 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
2528 DEBUG_STATEMENT (nfailure_points_popped++); \
2529 } /* POP_FAILURE_POINT */
2531 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2532 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2533 characters can start a string that matches the pattern. This fastmap
2534 is used by re_search to skip quickly over impossible starting points.
2536 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2537 area as BUFP->fastmap.
2539 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2542 Returns 0 if we succeed, -2 if an internal error. */
2545 re_compile_fastmap (bufp)
2546 struct re_pattern_buffer *bufp;
2549 fail_stack_type fail_stack;
2550 #ifndef REGEX_MALLOC
2553 /* We don't push any register information onto the failure stack. */
2554 unsigned num_regs = 0;
2556 register char *fastmap = bufp->fastmap;
2557 unsigned char *pattern = bufp->buffer;
2558 unsigned long size = bufp->used;
2559 const unsigned char *p = pattern;
2560 register unsigned char *pend = pattern + size;
2562 /* Assume that each path through the pattern can be null until
2563 proven otherwise. We set this false at the bottom of switch
2564 statement, to which we get only if a particular path doesn't
2565 match the empty string. */
2566 boolean path_can_be_null = true;
2568 /* We aren't doing a `succeed_n' to begin with. */
2569 boolean succeed_n_p = false;
2571 assert (fastmap != NULL && p != NULL);
2574 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2575 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2576 bufp->can_be_null = 0;
2578 while (p != pend || !FAIL_STACK_EMPTY ())
2582 bufp->can_be_null |= path_can_be_null;
2584 /* Reset for next path. */
2585 path_can_be_null = true;
2587 p = fail_stack.stack[--fail_stack.avail];
2590 /* We should never be about to go beyond the end of the pattern. */
2593 #ifdef SWITCH_ENUM_BUG
2594 switch ((int) ((re_opcode_t) *p++))
2596 switch ((re_opcode_t) *p++)
2600 /* I guess the idea here is to simply not bother with a fastmap
2601 if a backreference is used, since it's too hard to figure out
2602 the fastmap for the corresponding group. Setting
2603 `can_be_null' stops `re_search_2' from using the fastmap, so
2604 that is all we do. */
2606 bufp->can_be_null = 1;
2610 /* Following are the cases which match a character. These end
2619 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2620 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2626 /* Chars beyond end of map must be allowed. */
2627 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2630 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2631 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2637 for (j = 0; j < (1 << BYTEWIDTH); j++)
2638 if (SYNTAX (j) == Sword)
2644 for (j = 0; j < (1 << BYTEWIDTH); j++)
2645 if (SYNTAX (j) != Sword)
2651 /* `.' matches anything ... */
2652 for (j = 0; j < (1 << BYTEWIDTH); j++)
2655 /* ... except perhaps newline. */
2656 if (!(bufp->syntax & RE_DOT_NEWLINE))
2659 /* Return if we have already set `can_be_null'; if we have,
2660 then the fastmap is irrelevant. Something's wrong here. */
2661 else if (bufp->can_be_null)
2664 /* Otherwise, have to check alternative paths. */
2671 for (j = 0; j < (1 << BYTEWIDTH); j++)
2672 if (SYNTAX (j) == (enum syntaxcode) k)
2679 for (j = 0; j < (1 << BYTEWIDTH); j++)
2680 if (SYNTAX (j) != (enum syntaxcode) k)
2685 /* All cases after this match the empty string. These end with
2693 #endif /* not emacs */
2705 case push_dummy_failure:
2710 case pop_failure_jump:
2711 case maybe_pop_jump:
2714 case dummy_failure_jump:
2715 EXTRACT_NUMBER_AND_INCR (j, p);
2720 /* Jump backward implies we just went through the body of a
2721 loop and matched nothing. Opcode jumped to should be
2722 `on_failure_jump' or `succeed_n'. Just treat it like an
2723 ordinary jump. For a * loop, it has pushed its failure
2724 point already; if so, discard that as redundant. */
2725 if ((re_opcode_t) *p != on_failure_jump
2726 && (re_opcode_t) *p != succeed_n)
2730 EXTRACT_NUMBER_AND_INCR (j, p);
2733 /* If what's on the stack is where we are now, pop it. */
2734 if (!FAIL_STACK_EMPTY ()
2735 && fail_stack.stack[fail_stack.avail - 1] == p)
2741 case on_failure_jump:
2742 case on_failure_keep_string_jump:
2743 handle_on_failure_jump:
2744 EXTRACT_NUMBER_AND_INCR (j, p);
2746 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
2747 end of the pattern. We don't want to push such a point,
2748 since when we restore it above, entering the switch will
2749 increment `p' past the end of the pattern. We don't need
2750 to push such a point since we obviously won't find any more
2751 fastmap entries beyond `pend'. Such a pattern can match
2752 the null string, though. */
2755 if (!PUSH_PATTERN_OP (p + j, fail_stack))
2759 bufp->can_be_null = 1;
2763 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
2764 succeed_n_p = false;
2771 /* Get to the number of times to succeed. */
2774 /* Increment p past the n for when k != 0. */
2775 EXTRACT_NUMBER_AND_INCR (k, p);
2779 succeed_n_p = true; /* Spaghetti code alert. */
2780 goto handle_on_failure_jump;
2797 abort (); /* We have listed all the cases. */
2800 /* Getting here means we have found the possible starting
2801 characters for one path of the pattern -- and that the empty
2802 string does not match. We need not follow this path further.
2803 Instead, look at the next alternative (remembered on the
2804 stack), or quit if no more. The test at the top of the loop
2805 does these things. */
2806 path_can_be_null = false;
2810 /* Set `can_be_null' for the last path (also the first path, if the
2811 pattern is empty). */
2812 bufp->can_be_null |= path_can_be_null;
2814 } /* re_compile_fastmap */
2816 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
2817 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
2818 this memory for recording register information. STARTS and ENDS
2819 must be allocated using the malloc library routine, and must each
2820 be at least NUM_REGS * sizeof (regoff_t) bytes long.
2822 If NUM_REGS == 0, then subsequent matches should allocate their own
2825 Unless this function is called, the first search or match using
2826 PATTERN_BUFFER will allocate its own register data, without
2827 freeing the old data. */
2830 re_set_registers (bufp, regs, num_regs, starts, ends)
2831 struct re_pattern_buffer *bufp;
2832 struct re_registers *regs;
2834 regoff_t *starts, *ends;
2838 bufp->regs_allocated = REGS_REALLOCATE;
2839 regs->num_regs = num_regs;
2840 regs->start = starts;
2845 bufp->regs_allocated = REGS_UNALLOCATED;
2847 regs->start = regs->end = (regoff_t) 0;
2851 /* Searching routines. */
2853 /* Like re_search_2, below, but only one string is specified, and
2854 doesn't let you say where to stop matching. */
2857 re_search (bufp, string, size, startpos, range, regs)
2858 struct re_pattern_buffer *bufp;
2860 int size, startpos, range;
2861 struct re_registers *regs;
2863 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
2868 /* Using the compiled pattern in BUFP->buffer, first tries to match the
2869 virtual concatenation of STRING1 and STRING2, starting first at index
2870 STARTPOS, then at STARTPOS + 1, and so on.
2872 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
2874 RANGE is how far to scan while trying to match. RANGE = 0 means try
2875 only at STARTPOS; in general, the last start tried is STARTPOS +
2878 In REGS, return the indices of the virtual concatenation of STRING1
2879 and STRING2 that matched the entire BUFP->buffer and its contained
2882 Do not consider matching one past the index STOP in the virtual
2883 concatenation of STRING1 and STRING2.
2885 We return either the position in the strings at which the match was
2886 found, -1 if no match, or -2 if error (such as failure
2890 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
2891 struct re_pattern_buffer *bufp;
2892 const char *string1, *string2;
2896 struct re_registers *regs;
2900 register char *fastmap = bufp->fastmap;
2901 register char *translate = bufp->translate;
2902 int total_size = size1 + size2;
2903 int endpos = startpos + range;
2905 /* Check for out-of-range STARTPOS. */
2906 if (startpos < 0 || startpos > total_size)
2909 /* Fix up RANGE if it might eventually take us outside
2910 the virtual concatenation of STRING1 and STRING2. */
2912 range = -1 - startpos;
2913 else if (endpos > total_size)
2914 range = total_size - startpos;
2916 /* If the search isn't to be a backwards one, don't waste time in a
2917 search for a pattern that must be anchored. */
2918 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
2926 /* Update the fastmap now if not correct already. */
2927 if (fastmap && !bufp->fastmap_accurate)
2928 if (re_compile_fastmap (bufp) == -2)
2931 /* Loop through the string, looking for a place to start matching. */
2934 /* If a fastmap is supplied, skip quickly over characters that
2935 cannot be the start of a match. If the pattern can match the
2936 null string, however, we don't need to skip characters; we want
2937 the first null string. */
2938 if (fastmap && startpos < total_size && !bufp->can_be_null)
2940 if (range > 0) /* Searching forwards. */
2942 register const char *d;
2943 register int lim = 0;
2946 if (startpos < size1 && startpos + range >= size1)
2947 lim = range - (size1 - startpos);
2949 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
2951 /* Written out as an if-else to avoid testing `translate'
2955 && !fastmap[(unsigned char)
2956 translate[(unsigned char) *d++]])
2959 while (range > lim && !fastmap[(unsigned char) *d++])
2962 startpos += irange - range;
2964 else /* Searching backwards. */
2966 register char c = (size1 == 0 || startpos >= size1
2967 ? string2[startpos - size1]
2968 : string1[startpos]);
2970 if (!fastmap[(unsigned char) TRANSLATE (c)])
2975 /* If can't match the null string, and that's all we have left, fail. */
2976 if (range >= 0 && startpos == total_size && fastmap
2977 && !bufp->can_be_null)
2980 val = re_match_2 (bufp, string1, size1, string2, size2,
2981 startpos, regs, stop);
3005 /* Declarations and macros for re_match_2. */
3007 static int bcmp_translate ();
3008 static boolean alt_match_null_string_p (),
3009 common_op_match_null_string_p (),
3010 group_match_null_string_p ();
3012 /* Structure for per-register (a.k.a. per-group) information.
3013 This must not be longer than one word, because we push this value
3014 onto the failure stack. Other register information, such as the
3015 starting and ending positions (which are addresses), and the list of
3016 inner groups (which is a bits list) are maintained in separate
3019 We are making a (strictly speaking) nonportable assumption here: that
3020 the compiler will pack our bit fields into something that fits into
3021 the type of `word', i.e., is something that fits into one item on the
3025 fail_stack_elt_t word;
3028 /* This field is one if this group can match the empty string,
3029 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
3030 #define MATCH_NULL_UNSET_VALUE 3
3031 unsigned match_null_string_p : 2;
3032 unsigned is_active : 1;
3033 unsigned matched_something : 1;
3034 unsigned ever_matched_something : 1;
3036 } register_info_type;
3038 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
3039 #define IS_ACTIVE(R) ((R).bits.is_active)
3040 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
3041 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
3044 /* Call this when have matched a real character; it sets `matched' flags
3045 for the subexpressions which we are currently inside. Also records
3046 that those subexprs have matched. */
3047 #define SET_REGS_MATCHED() \
3051 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
3053 MATCHED_SOMETHING (reg_info[r]) \
3054 = EVER_MATCHED_SOMETHING (reg_info[r]) \
3061 /* This converts PTR, a pointer into one of the search strings `string1'
3062 and `string2' into an offset from the beginning of that string. */
3063 #define POINTER_TO_OFFSET(ptr) \
3064 (FIRST_STRING_P (ptr) ? (ptr) - string1 : (ptr) - string2 + size1)
3066 /* Registers are set to a sentinel when they haven't yet matched. */
3067 #define REG_UNSET_VALUE ((char *) -1)
3068 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
3071 /* Macros for dealing with the split strings in re_match_2. */
3073 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3075 /* Call before fetching a character with *d. This switches over to
3076 string2 if necessary. */
3077 #define PREFETCH() \
3080 /* End of string2 => fail. */ \
3081 if (dend == end_match_2) \
3083 /* End of string1 => advance to string2. */ \
3085 dend = end_match_2; \
3089 /* Test if at very beginning or at very end of the virtual concatenation
3090 of `string1' and `string2'. If only one string, it's `string2'. */
3091 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3092 #define AT_STRINGS_END(d) ((d) == end2)
3095 /* Test if D points to a character which is word-constituent. We have
3096 two special cases to check for: if past the end of string1, look at
3097 the first character in string2; and if before the beginning of
3098 string2, look at the last character in string1. */
3099 #define WORDCHAR_P(d) \
3100 (SYNTAX ((d) == end1 ? *string2 \
3101 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3104 /* Test if the character before D and the one at D differ with respect
3105 to being word-constituent. */
3106 #define AT_WORD_BOUNDARY(d) \
3107 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3108 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3111 /* Free everything we malloc. */
3113 #define FREE_VAR(var) if (var) free (var); var = NULL
3114 #define FREE_VARIABLES() \
3116 FREE_VAR (fail_stack.stack); \
3117 FREE_VAR (regstart); \
3118 FREE_VAR (regend); \
3119 FREE_VAR (old_regstart); \
3120 FREE_VAR (old_regend); \
3121 FREE_VAR (best_regstart); \
3122 FREE_VAR (best_regend); \
3123 FREE_VAR (reg_info); \
3124 FREE_VAR (reg_dummy); \
3125 FREE_VAR (reg_info_dummy); \
3127 #else /* not REGEX_MALLOC */
3128 /* Some MIPS systems (at least) want this to free alloca'd storage. */
3129 #define FREE_VARIABLES() alloca (0)
3130 #endif /* not REGEX_MALLOC */
3133 /* These values must meet several constraints. They must not be valid
3134 register values; since we have a limit of 255 registers (because
3135 we use only one byte in the pattern for the register number), we can
3136 use numbers larger than 255. They must differ by 1, because of
3137 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3138 be larger than the value for the highest register, so we do not try
3139 to actually save any registers when none are active. */
3140 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3141 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3143 /* Matching routines. */
3145 #ifndef emacs /* Emacs never uses this. */
3146 /* re_match is like re_match_2 except it takes only a single string. */
3149 re_match (bufp, string, size, pos, regs)
3150 struct re_pattern_buffer *bufp;
3153 struct re_registers *regs;
3155 return re_match_2 (bufp, NULL, 0, string, size, pos, regs, size);
3157 #endif /* not emacs */
3160 /* re_match_2 matches the compiled pattern in BUFP against the
3161 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3162 and SIZE2, respectively). We start matching at POS, and stop
3165 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3166 store offsets for the substring each group matched in REGS. See the
3167 documentation for exactly how many groups we fill.
3169 We return -1 if no match, -2 if an internal error (such as the
3170 failure stack overflowing). Otherwise, we return the length of the
3171 matched substring. */
3174 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3175 struct re_pattern_buffer *bufp;
3176 const char *string1, *string2;
3179 struct re_registers *regs;
3182 /* General temporaries. */
3186 /* Just past the end of the corresponding string. */
3187 const char *end1, *end2;
3189 /* Pointers into string1 and string2, just past the last characters in
3190 each to consider matching. */
3191 const char *end_match_1, *end_match_2;
3193 /* Where we are in the data, and the end of the current string. */
3194 const char *d, *dend;
3196 /* Where we are in the pattern, and the end of the pattern. */
3197 unsigned char *p = bufp->buffer;
3198 register unsigned char *pend = p + bufp->used;
3200 /* We use this to map every character in the string. */
3201 char *translate = bufp->translate;
3203 /* Failure point stack. Each place that can handle a failure further
3204 down the line pushes a failure point on this stack. It consists of
3205 restart, regend, and reg_info for all registers corresponding to
3206 the subexpressions we're currently inside, plus the number of such
3207 registers, and, finally, two char *'s. The first char * is where
3208 to resume scanning the pattern; the second one is where to resume
3209 scanning the strings. If the latter is zero, the failure point is
3210 a ``dummy''; if a failure happens and the failure point is a dummy,
3211 it gets discarded and the next next one is tried. */
3212 fail_stack_type fail_stack;
3214 static unsigned failure_id = 0;
3215 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3218 /* We fill all the registers internally, independent of what we
3219 return, for use in backreferences. The number here includes
3220 an element for register zero. */
3221 unsigned num_regs = bufp->re_nsub + 1;
3223 /* The currently active registers. */
3224 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3225 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3227 /* Information on the contents of registers. These are pointers into
3228 the input strings; they record just what was matched (on this
3229 attempt) by a subexpression part of the pattern, that is, the
3230 regnum-th regstart pointer points to where in the pattern we began
3231 matching and the regnum-th regend points to right after where we
3232 stopped matching the regnum-th subexpression. (The zeroth register
3233 keeps track of what the whole pattern matches.) */
3234 const char **regstart = NULL, **regend = NULL;
3236 /* If a group that's operated upon by a repetition operator fails to
3237 match anything, then the register for its start will need to be
3238 restored because it will have been set to wherever in the string we
3239 are when we last see its open-group operator. Similarly for a
3241 const char **old_regstart = NULL, **old_regend = NULL;
3243 /* The is_active field of reg_info helps us keep track of which (possibly
3244 nested) subexpressions we are currently in. The matched_something
3245 field of reg_info[reg_num] helps us tell whether or not we have
3246 matched any of the pattern so far this time through the reg_num-th
3247 subexpression. These two fields get reset each time through any
3248 loop their register is in. */
3249 register_info_type *reg_info = NULL;
3251 /* The following record the register info as found in the above
3252 variables when we find a match better than any we've seen before.
3253 This happens as we backtrack through the failure points, which in
3254 turn happens only if we have not yet matched the entire string. */
3255 unsigned best_regs_set = false;
3256 const char **best_regstart = NULL, **best_regend = NULL;
3258 /* Logically, this is `best_regend[0]'. But we don't want to have to
3259 allocate space for that if we're not allocating space for anything
3260 else (see below). Also, we never need info about register 0 for
3261 any of the other register vectors, and it seems rather a kludge to
3262 treat `best_regend' differently than the rest. So we keep track of
3263 the end of the best match so far in a separate variable. We
3264 initialize this to NULL so that when we backtrack the first time
3265 and need to test it, it's not garbage. */
3266 const char *match_end = NULL;
3268 /* Used when we pop values we don't care about. */
3269 const char **reg_dummy = NULL;
3270 register_info_type *reg_info_dummy = NULL;
3273 /* Counts the total number of registers pushed. */
3274 unsigned num_regs_pushed = 0;
3277 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3281 /* Do not bother to initialize all the register variables if there are
3282 no groups in the pattern, as it takes a fair amount of time. If
3283 there are groups, we include space for register 0 (the whole
3284 pattern), even though we never use it, since it simplifies the
3285 array indexing. We should fix this. */
3288 regstart = REGEX_TALLOC (num_regs, const char *);
3289 regend = REGEX_TALLOC (num_regs, const char *);
3290 old_regstart = REGEX_TALLOC (num_regs, const char *);
3291 old_regend = REGEX_TALLOC (num_regs, const char *);
3292 best_regstart = REGEX_TALLOC (num_regs, const char *);
3293 best_regend = REGEX_TALLOC (num_regs, const char *);
3294 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3295 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3296 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3298 if (!(regstart && regend && old_regstart && old_regend && reg_info
3299 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3308 /* We must initialize all our variables to NULL, so that
3309 `FREE_VARIABLES' doesn't try to free them. */
3310 regstart = regend = old_regstart = old_regend = best_regstart
3311 = best_regend = reg_dummy = NULL;
3312 reg_info = reg_info_dummy = (register_info_type *) NULL;
3314 #endif /* REGEX_MALLOC */
3316 /* The starting position is bogus. */
3317 if (pos < 0 || pos > size1 + size2)
3323 /* Initialize subexpression text positions to -1 to mark ones that no
3324 start_memory/stop_memory has been seen for. Also initialize the
3325 register information struct. */
3326 for (mcnt = 1; mcnt < num_regs; mcnt++)
3328 regstart[mcnt] = regend[mcnt]
3329 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3331 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3332 IS_ACTIVE (reg_info[mcnt]) = 0;
3333 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3334 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3337 /* We move `string1' into `string2' if the latter's empty -- but not if
3338 `string1' is null. */
3339 if (size2 == 0 && string1 != NULL)
3346 end1 = string1 + size1;
3347 end2 = string2 + size2;
3349 /* Compute where to stop matching, within the two strings. */
3352 end_match_1 = string1 + stop;
3353 end_match_2 = string2;
3358 end_match_2 = string2 + stop - size1;
3361 /* `p' scans through the pattern as `d' scans through the data.
3362 `dend' is the end of the input string that `d' points within. `d'
3363 is advanced into the following input string whenever necessary, but
3364 this happens before fetching; therefore, at the beginning of the
3365 loop, `d' can be pointing at the end of a string, but it cannot
3367 if (size1 > 0 && pos <= size1)
3374 d = string2 + pos - size1;
3378 DEBUG_PRINT1 ("The compiled pattern is: ");
3379 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3380 DEBUG_PRINT1 ("The string to match is: `");
3381 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3382 DEBUG_PRINT1 ("'\n");
3384 /* This loops over pattern commands. It exits by returning from the
3385 function if the match is complete, or it drops through if the match
3386 fails at this starting point in the input data. */
3389 DEBUG_PRINT2 ("\n0x%x: ", p);
3392 { /* End of pattern means we might have succeeded. */
3393 DEBUG_PRINT1 ("end of pattern ... ");
3395 /* If we haven't matched the entire string, and we want the
3396 longest match, try backtracking. */
3397 if (d != end_match_2)
3399 DEBUG_PRINT1 ("backtracking.\n");
3401 if (!FAIL_STACK_EMPTY ())
3402 { /* More failure points to try. */
3403 boolean same_str_p = (FIRST_STRING_P (match_end)
3404 == MATCHING_IN_FIRST_STRING);
3406 /* If exceeds best match so far, save it. */
3408 || (same_str_p && d > match_end)
3409 || (!same_str_p && !MATCHING_IN_FIRST_STRING))
3411 best_regs_set = true;
3414 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3416 for (mcnt = 1; mcnt < num_regs; mcnt++)
3418 best_regstart[mcnt] = regstart[mcnt];
3419 best_regend[mcnt] = regend[mcnt];
3425 /* If no failure points, don't restore garbage. */
3426 else if (best_regs_set)
3429 /* Restore best match. It may happen that `dend ==
3430 end_match_1' while the restored d is in string2.
3431 For example, the pattern `x.*y.*z' against the
3432 strings `x-' and `y-z-', if the two strings are
3433 not consecutive in memory. */
3434 DEBUG_PRINT1 ("Restoring best registers.\n");
3437 dend = ((d >= string1 && d <= end1)
3438 ? end_match_1 : end_match_2);
3440 for (mcnt = 1; mcnt < num_regs; mcnt++)
3442 regstart[mcnt] = best_regstart[mcnt];
3443 regend[mcnt] = best_regend[mcnt];
3446 } /* d != end_match_2 */
3448 DEBUG_PRINT1 ("Accepting match.\n");
3450 /* If caller wants register contents data back, do it. */
3451 if (regs && !bufp->no_sub)
3453 /* Have the register data arrays been allocated? */
3454 if (bufp->regs_allocated == REGS_UNALLOCATED)
3455 { /* No. So allocate them with malloc. We need one
3456 extra element beyond `num_regs' for the `-1' marker
3458 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3459 regs->start = TALLOC (regs->num_regs, regoff_t);
3460 regs->end = TALLOC (regs->num_regs, regoff_t);
3461 if (regs->start == NULL || regs->end == NULL)
3463 bufp->regs_allocated = REGS_REALLOCATE;
3465 else if (bufp->regs_allocated == REGS_REALLOCATE)
3466 { /* Yes. If we need more elements than were already
3467 allocated, reallocate them. If we need fewer, just
3469 if (regs->num_regs < num_regs + 1)
3471 regs->num_regs = num_regs + 1;
3472 RETALLOC (regs->start, regs->num_regs, regoff_t);
3473 RETALLOC (regs->end, regs->num_regs, regoff_t);
3474 if (regs->start == NULL || regs->end == NULL)
3479 assert (bufp->regs_allocated == REGS_FIXED);
3481 /* Convert the pointer data in `regstart' and `regend' to
3482 indices. Register zero has to be set differently,
3483 since we haven't kept track of any info for it. */
3484 if (regs->num_regs > 0)
3486 regs->start[0] = pos;
3487 regs->end[0] = (MATCHING_IN_FIRST_STRING ? d - string1
3488 : d - string2 + size1);
3491 /* Go through the first `min (num_regs, regs->num_regs)'
3492 registers, since that is all we initialized. */
3493 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3495 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3496 regs->start[mcnt] = regs->end[mcnt] = -1;
3499 regs->start[mcnt] = POINTER_TO_OFFSET (regstart[mcnt]);
3500 regs->end[mcnt] = POINTER_TO_OFFSET (regend[mcnt]);
3504 /* If the regs structure we return has more elements than
3505 were in the pattern, set the extra elements to -1. If
3506 we (re)allocated the registers, this is the case,
3507 because we always allocate enough to have at least one
3509 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3510 regs->start[mcnt] = regs->end[mcnt] = -1;
3511 } /* regs && !bufp->no_sub */
3514 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3515 nfailure_points_pushed, nfailure_points_popped,
3516 nfailure_points_pushed - nfailure_points_popped);
3517 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3519 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3523 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3528 /* Otherwise match next pattern command. */
3529 #ifdef SWITCH_ENUM_BUG
3530 switch ((int) ((re_opcode_t) *p++))
3532 switch ((re_opcode_t) *p++)
3535 /* Ignore these. Used to ignore the n of succeed_n's which
3536 currently have n == 0. */
3538 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3542 /* Match the next n pattern characters exactly. The following
3543 byte in the pattern defines n, and the n bytes after that
3544 are the characters to match. */
3547 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3549 /* This is written out as an if-else so we don't waste time
3550 testing `translate' inside the loop. */
3556 if (translate[(unsigned char) *d++] != (char) *p++)
3566 if (*d++ != (char) *p++) goto fail;
3570 SET_REGS_MATCHED ();
3574 /* Match any character except possibly a newline or a null. */
3576 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3580 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3581 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3584 SET_REGS_MATCHED ();
3585 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3593 register unsigned char c;
3594 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3596 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3599 c = TRANSLATE (*d); /* The character to match. */
3601 /* Cast to `unsigned' instead of `unsigned char' in case the
3602 bit list is a full 32 bytes long. */
3603 if (c < (unsigned) (*p * BYTEWIDTH)
3604 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3609 if (!not) goto fail;
3611 SET_REGS_MATCHED ();
3617 /* The beginning of a group is represented by start_memory.
3618 The arguments are the register number in the next byte, and the
3619 number of groups inner to this one in the next. The text
3620 matched within the group is recorded (in the internal
3621 registers data structure) under the register number. */
3623 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3625 /* Find out if this group can match the empty string. */
3626 p1 = p; /* To send to group_match_null_string_p. */
3628 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3629 REG_MATCH_NULL_STRING_P (reg_info[*p])
3630 = group_match_null_string_p (&p1, pend, reg_info);
3632 /* Save the position in the string where we were the last time
3633 we were at this open-group operator in case the group is
3634 operated upon by a repetition operator, e.g., with `(a*)*b'
3635 against `ab'; then we want to ignore where we are now in
3636 the string in case this attempt to match fails. */
3637 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3638 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3640 DEBUG_PRINT2 (" old_regstart: %d\n",
3641 POINTER_TO_OFFSET (old_regstart[*p]));
3644 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
3646 IS_ACTIVE (reg_info[*p]) = 1;
3647 MATCHED_SOMETHING (reg_info[*p]) = 0;
3649 /* This is the new highest active register. */
3650 highest_active_reg = *p;
3652 /* If nothing was active before, this is the new lowest active
3654 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3655 lowest_active_reg = *p;
3657 /* Move past the register number and inner group count. */
3662 /* The stop_memory opcode represents the end of a group. Its
3663 arguments are the same as start_memory's: the register
3664 number, and the number of inner groups. */
3666 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
3668 /* We need to save the string position the last time we were at
3669 this close-group operator in case the group is operated
3670 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
3671 against `aba'; then we want to ignore where we are now in
3672 the string in case this attempt to match fails. */
3673 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3674 ? REG_UNSET (regend[*p]) ? d : regend[*p]
3676 DEBUG_PRINT2 (" old_regend: %d\n",
3677 POINTER_TO_OFFSET (old_regend[*p]));
3680 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
3682 /* This register isn't active anymore. */
3683 IS_ACTIVE (reg_info[*p]) = 0;
3685 /* If this was the only register active, nothing is active
3687 if (lowest_active_reg == highest_active_reg)
3689 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3690 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3693 { /* We must scan for the new highest active register, since
3694 it isn't necessarily one less than now: consider
3695 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
3696 new highest active register is 1. */
3697 unsigned char r = *p - 1;
3698 while (r > 0 && !IS_ACTIVE (reg_info[r]))
3701 /* If we end up at register zero, that means that we saved
3702 the registers as the result of an `on_failure_jump', not
3703 a `start_memory', and we jumped to past the innermost
3704 `stop_memory'. For example, in ((.)*) we save
3705 registers 1 and 2 as a result of the *, but when we pop
3706 back to the second ), we are at the stop_memory 1.
3707 Thus, nothing is active. */
3710 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3711 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3714 highest_active_reg = r;
3717 /* If just failed to match something this time around with a
3718 group that's operated on by a repetition operator, try to
3719 force exit from the ``loop'', and restore the register
3720 information for this group that we had before trying this
3722 if ((!MATCHED_SOMETHING (reg_info[*p])
3723 || (re_opcode_t) p[-3] == start_memory)
3726 boolean is_a_jump_n = false;
3730 switch ((re_opcode_t) *p1++)
3734 case pop_failure_jump:
3735 case maybe_pop_jump:
3737 case dummy_failure_jump:
3738 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3748 /* If the next operation is a jump backwards in the pattern
3749 to an on_failure_jump right before the start_memory
3750 corresponding to this stop_memory, exit from the loop
3751 by forcing a failure after pushing on the stack the
3752 on_failure_jump's jump in the pattern, and d. */
3753 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
3754 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
3756 /* If this group ever matched anything, then restore
3757 what its registers were before trying this last
3758 failed match, e.g., with `(a*)*b' against `ab' for
3759 regstart[1], and, e.g., with `((a*)*(b*)*)*'
3760 against `aba' for regend[3].
3762 Also restore the registers for inner groups for,
3763 e.g., `((a*)(b*))*' against `aba' (register 3 would
3764 otherwise get trashed). */
3766 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
3770 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
3772 /* Restore this and inner groups' (if any) registers. */
3773 for (r = *p; r < *p + *(p + 1); r++)
3775 regstart[r] = old_regstart[r];
3777 /* xx why this test? */
3778 if ((int) old_regend[r] >= (int) regstart[r])
3779 regend[r] = old_regend[r];
3783 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
3784 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
3790 /* Move past the register number and the inner group count. */
3795 /* \<digit> has been turned into a `duplicate' command which is
3796 followed by the numeric value of <digit> as the register number. */
3799 register const char *d2, *dend2;
3800 int regno = *p++; /* Get which register to match against. */
3801 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
3803 /* Can't back reference a group which we've never matched. */
3804 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
3807 /* Where in input to try to start matching. */
3808 d2 = regstart[regno];
3810 /* Where to stop matching; if both the place to start and
3811 the place to stop matching are in the same string, then
3812 set to the place to stop, otherwise, for now have to use
3813 the end of the first string. */
3815 dend2 = ((FIRST_STRING_P (regstart[regno])
3816 == FIRST_STRING_P (regend[regno]))
3817 ? regend[regno] : end_match_1);
3820 /* If necessary, advance to next segment in register
3824 if (dend2 == end_match_2) break;
3825 if (dend2 == regend[regno]) break;
3827 /* End of string1 => advance to string2. */
3829 dend2 = regend[regno];
3831 /* At end of register contents => success */
3832 if (d2 == dend2) break;
3834 /* If necessary, advance to next segment in data. */
3837 /* How many characters left in this segment to match. */
3840 /* Want how many consecutive characters we can match in
3841 one shot, so, if necessary, adjust the count. */
3842 if (mcnt > dend2 - d2)
3845 /* Compare that many; failure if mismatch, else move
3848 ? bcmp_translate (d, d2, mcnt, translate)
3849 : bcmp (d, d2, mcnt))
3851 d += mcnt, d2 += mcnt;
3857 /* begline matches the empty string at the beginning of the string
3858 (unless `not_bol' is set in `bufp'), and, if
3859 `newline_anchor' is set, after newlines. */
3861 DEBUG_PRINT1 ("EXECUTING begline.\n");
3863 if (AT_STRINGS_BEG (d))
3865 if (!bufp->not_bol) break;
3867 else if (d[-1] == '\n' && bufp->newline_anchor)
3871 /* In all other cases, we fail. */
3875 /* endline is the dual of begline. */
3877 DEBUG_PRINT1 ("EXECUTING endline.\n");
3879 if (AT_STRINGS_END (d))
3881 if (!bufp->not_eol) break;
3884 /* We have to ``prefetch'' the next character. */
3885 else if ((d == end1 ? *string2 : *d) == '\n'
3886 && bufp->newline_anchor)
3893 /* Match at the very beginning of the data. */
3895 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
3896 if (AT_STRINGS_BEG (d))
3901 /* Match at the very end of the data. */
3903 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
3904 if (AT_STRINGS_END (d))
3909 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
3910 pushes NULL as the value for the string on the stack. Then
3911 `pop_failure_point' will keep the current value for the
3912 string, instead of restoring it. To see why, consider
3913 matching `foo\nbar' against `.*\n'. The .* matches the foo;
3914 then the . fails against the \n. But the next thing we want
3915 to do is match the \n against the \n; if we restored the
3916 string value, we would be back at the foo.
3918 Because this is used only in specific cases, we don't need to
3919 check all the things that `on_failure_jump' does, to make
3920 sure the right things get saved on the stack. Hence we don't
3921 share its code. The only reason to push anything on the
3922 stack at all is that otherwise we would have to change
3923 `anychar's code to do something besides goto fail in this
3924 case; that seems worse than this. */
3925 case on_failure_keep_string_jump:
3926 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
3928 EXTRACT_NUMBER_AND_INCR (mcnt, p);
3929 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
3931 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
3935 /* Uses of on_failure_jump:
3937 Each alternative starts with an on_failure_jump that points
3938 to the beginning of the next alternative. Each alternative
3939 except the last ends with a jump that in effect jumps past
3940 the rest of the alternatives. (They really jump to the
3941 ending jump of the following alternative, because tensioning
3942 these jumps is a hassle.)
3944 Repeats start with an on_failure_jump that points past both
3945 the repetition text and either the following jump or
3946 pop_failure_jump back to this on_failure_jump. */
3947 case on_failure_jump:
3949 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
3951 EXTRACT_NUMBER_AND_INCR (mcnt, p);
3952 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
3954 /* If this on_failure_jump comes right before a group (i.e.,
3955 the original * applied to a group), save the information
3956 for that group and all inner ones, so that if we fail back
3957 to this point, the group's information will be correct.
3958 For example, in \(a*\)*\1, we need the preceding group,
3959 and in \(\(a*\)b*\)\2, we need the inner group. */
3961 /* We can't use `p' to check ahead because we push
3962 a failure point to `p + mcnt' after we do this. */
3965 /* We need to skip no_op's before we look for the
3966 start_memory in case this on_failure_jump is happening as
3967 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
3969 while (p1 < pend && (re_opcode_t) *p1 == no_op)
3972 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
3974 /* We have a new highest active register now. This will
3975 get reset at the start_memory we are about to get to,
3976 but we will have saved all the registers relevant to
3977 this repetition op, as described above. */
3978 highest_active_reg = *(p1 + 1) + *(p1 + 2);
3979 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
3980 lowest_active_reg = *(p1 + 1);
3983 DEBUG_PRINT1 (":\n");
3984 PUSH_FAILURE_POINT (p + mcnt, d, -2);
3988 /* A smart repeat ends with `maybe_pop_jump'.
3989 We change it to either `pop_failure_jump' or `jump'. */
3990 case maybe_pop_jump:
3991 EXTRACT_NUMBER_AND_INCR (mcnt, p);
3992 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
3994 register unsigned char *p2 = p;
3996 /* Compare the beginning of the repeat with what in the
3997 pattern follows its end. If we can establish that there
3998 is nothing that they would both match, i.e., that we
3999 would have to backtrack because of (as in, e.g., `a*a')
4000 then we can change to pop_failure_jump, because we'll
4001 never have to backtrack.
4003 This is not true in the case of alternatives: in
4004 `(a|ab)*' we do need to backtrack to the `ab' alternative
4005 (e.g., if the string was `ab'). But instead of trying to
4006 detect that here, the alternative has put on a dummy
4007 failure point which is what we will end up popping. */
4009 /* Skip over open/close-group commands. */
4010 while (p2 + 2 < pend
4011 && ((re_opcode_t) *p2 == stop_memory
4012 || (re_opcode_t) *p2 == start_memory))
4013 p2 += 3; /* Skip over args, too. */
4015 /* If we're at the end of the pattern, we can change. */
4018 /* Consider what happens when matching ":\(.*\)"
4019 against ":/". I don't really understand this code
4021 p[-3] = (unsigned char) pop_failure_jump;
4023 (" End of pattern: change to `pop_failure_jump'.\n");
4026 else if ((re_opcode_t) *p2 == exactn
4027 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4029 register unsigned char c
4030 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4033 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4034 to the `maybe_finalize_jump' of this case. Examine what
4036 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4038 p[-3] = (unsigned char) pop_failure_jump;
4039 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4043 else if ((re_opcode_t) p1[3] == charset
4044 || (re_opcode_t) p1[3] == charset_not)
4046 int not = (re_opcode_t) p1[3] == charset_not;
4048 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4049 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4052 /* `not' is equal to 1 if c would match, which means
4053 that we can't change to pop_failure_jump. */
4056 p[-3] = (unsigned char) pop_failure_jump;
4057 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4062 p -= 2; /* Point at relative address again. */
4063 if ((re_opcode_t) p[-1] != pop_failure_jump)
4065 p[-1] = (unsigned char) jump;
4066 DEBUG_PRINT1 (" Match => jump.\n");
4067 goto unconditional_jump;
4069 /* Note fall through. */
4072 /* The end of a simple repeat has a pop_failure_jump back to
4073 its matching on_failure_jump, where the latter will push a
4074 failure point. The pop_failure_jump takes off failure
4075 points put on by this pop_failure_jump's matching
4076 on_failure_jump; we got through the pattern to here from the
4077 matching on_failure_jump, so didn't fail. */
4078 case pop_failure_jump:
4080 /* We need to pass separate storage for the lowest and
4081 highest registers, even though we don't care about the
4082 actual values. Otherwise, we will restore only one
4083 register from the stack, since lowest will == highest in
4084 `pop_failure_point'. */
4085 unsigned dummy_low_reg, dummy_high_reg;
4086 unsigned char *pdummy;
4089 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4090 POP_FAILURE_POINT (sdummy, pdummy,
4091 dummy_low_reg, dummy_high_reg,
4092 reg_dummy, reg_dummy, reg_info_dummy);
4094 /* Note fall through. */
4097 /* Unconditionally jump (without popping any failure points). */
4100 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4101 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4102 p += mcnt; /* Do the jump. */
4103 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4107 /* We need this opcode so we can detect where alternatives end
4108 in `group_match_null_string_p' et al. */
4110 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4111 goto unconditional_jump;
4114 /* Normally, the on_failure_jump pushes a failure point, which
4115 then gets popped at pop_failure_jump. We will end up at
4116 pop_failure_jump, also, and with a pattern of, say, `a+', we
4117 are skipping over the on_failure_jump, so we have to push
4118 something meaningless for pop_failure_jump to pop. */
4119 case dummy_failure_jump:
4120 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4121 /* It doesn't matter what we push for the string here. What
4122 the code at `fail' tests is the value for the pattern. */
4123 PUSH_FAILURE_POINT (0, 0, -2);
4124 goto unconditional_jump;
4127 /* At the end of an alternative, we need to push a dummy failure
4128 point in case we are followed by a `pop_failure_jump', because
4129 we don't want the failure point for the alternative to be
4130 popped. For example, matching `(a|ab)*' against `aab'
4131 requires that we match the `ab' alternative. */
4132 case push_dummy_failure:
4133 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4134 /* See comments just above at `dummy_failure_jump' about the
4136 PUSH_FAILURE_POINT (0, 0, -2);
4139 /* Have to succeed matching what follows at least n times.
4140 After that, handle like `on_failure_jump'. */
4142 EXTRACT_NUMBER (mcnt, p + 2);
4143 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4146 /* Originally, this is how many times we HAVE to succeed. */
4151 STORE_NUMBER_AND_INCR (p, mcnt);
4152 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4156 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4157 p[2] = (unsigned char) no_op;
4158 p[3] = (unsigned char) no_op;
4164 EXTRACT_NUMBER (mcnt, p + 2);
4165 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4167 /* Originally, this is how many times we CAN jump. */
4171 STORE_NUMBER (p + 2, mcnt);
4172 goto unconditional_jump;
4174 /* If don't have to jump any more, skip over the rest of command. */
4181 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4183 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4185 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4186 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4187 STORE_NUMBER (p1, mcnt);
4192 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4193 if (AT_WORD_BOUNDARY (d))
4198 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4199 if (AT_WORD_BOUNDARY (d))
4204 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4205 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4210 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4211 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4212 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4219 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4220 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4225 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4226 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4231 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4232 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4235 #else /* not emacs19 */
4237 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4238 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4241 #endif /* not emacs19 */
4244 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4249 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4253 if (SYNTAX (*d++) != (enum syntaxcode) mcnt)
4255 SET_REGS_MATCHED ();
4259 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4261 goto matchnotsyntax;
4264 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4268 if (SYNTAX (*d++) == (enum syntaxcode) mcnt)
4270 SET_REGS_MATCHED ();
4273 #else /* not emacs */
4275 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4277 if (!WORDCHAR_P (d))
4279 SET_REGS_MATCHED ();
4284 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4288 SET_REGS_MATCHED ();
4291 #endif /* not emacs */
4296 continue; /* Successfully executed one pattern command; keep going. */
4299 /* We goto here if a matching operation fails. */
4301 if (!FAIL_STACK_EMPTY ())
4302 { /* A restart point is known. Restore to that state. */
4303 DEBUG_PRINT1 ("\nFAIL:\n");
4304 POP_FAILURE_POINT (d, p,
4305 lowest_active_reg, highest_active_reg,
4306 regstart, regend, reg_info);
4308 /* If this failure point is a dummy, try the next one. */
4312 /* If we failed to the end of the pattern, don't examine *p. */
4316 boolean is_a_jump_n = false;
4318 /* If failed to a backwards jump that's part of a repetition
4319 loop, need to pop this failure point and use the next one. */
4320 switch ((re_opcode_t) *p)
4324 case maybe_pop_jump:
4325 case pop_failure_jump:
4328 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4331 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4333 && (re_opcode_t) *p1 == on_failure_jump))
4341 if (d >= string1 && d <= end1)
4345 break; /* Matching at this starting point really fails. */
4349 goto restore_best_regs;
4353 return -1; /* Failure to match. */
4356 /* Subroutine definitions for re_match_2. */
4359 /* We are passed P pointing to a register number after a start_memory.
4361 Return true if the pattern up to the corresponding stop_memory can
4362 match the empty string, and false otherwise.
4364 If we find the matching stop_memory, sets P to point to one past its number.
4365 Otherwise, sets P to an undefined byte less than or equal to END.
4367 We don't handle duplicates properly (yet). */
4370 group_match_null_string_p (p, end, reg_info)
4371 unsigned char **p, *end;
4372 register_info_type *reg_info;
4375 /* Point to after the args to the start_memory. */
4376 unsigned char *p1 = *p + 2;
4380 /* Skip over opcodes that can match nothing, and return true or
4381 false, as appropriate, when we get to one that can't, or to the
4382 matching stop_memory. */
4384 switch ((re_opcode_t) *p1)
4386 /* Could be either a loop or a series of alternatives. */
4387 case on_failure_jump:
4389 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4391 /* If the next operation is not a jump backwards in the
4396 /* Go through the on_failure_jumps of the alternatives,
4397 seeing if any of the alternatives cannot match nothing.
4398 The last alternative starts with only a jump,
4399 whereas the rest start with on_failure_jump and end
4400 with a jump, e.g., here is the pattern for `a|b|c':
4402 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4403 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4406 So, we have to first go through the first (n-1)
4407 alternatives and then deal with the last one separately. */
4410 /* Deal with the first (n-1) alternatives, which start
4411 with an on_failure_jump (see above) that jumps to right
4412 past a jump_past_alt. */
4414 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4416 /* `mcnt' holds how many bytes long the alternative
4417 is, including the ending `jump_past_alt' and
4420 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4424 /* Move to right after this alternative, including the
4428 /* Break if it's the beginning of an n-th alternative
4429 that doesn't begin with an on_failure_jump. */
4430 if ((re_opcode_t) *p1 != on_failure_jump)
4433 /* Still have to check that it's not an n-th
4434 alternative that starts with an on_failure_jump. */
4436 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4437 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4439 /* Get to the beginning of the n-th alternative. */
4445 /* Deal with the last alternative: go back and get number
4446 of the `jump_past_alt' just before it. `mcnt' contains
4447 the length of the alternative. */
4448 EXTRACT_NUMBER (mcnt, p1 - 2);
4450 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4453 p1 += mcnt; /* Get past the n-th alternative. */
4459 assert (p1[1] == **p);
4465 if (!common_op_match_null_string_p (&p1, end, reg_info))
4468 } /* while p1 < end */
4471 } /* group_match_null_string_p */
4474 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4475 It expects P to be the first byte of a single alternative and END one
4476 byte past the last. The alternative can contain groups. */
4479 alt_match_null_string_p (p, end, reg_info)
4480 unsigned char *p, *end;
4481 register_info_type *reg_info;
4484 unsigned char *p1 = p;
4488 /* Skip over opcodes that can match nothing, and break when we get
4489 to one that can't. */
4491 switch ((re_opcode_t) *p1)
4494 case on_failure_jump:
4496 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4501 if (!common_op_match_null_string_p (&p1, end, reg_info))
4504 } /* while p1 < end */
4507 } /* alt_match_null_string_p */
4510 /* Deals with the ops common to group_match_null_string_p and
4511 alt_match_null_string_p.
4513 Sets P to one after the op and its arguments, if any. */
4516 common_op_match_null_string_p (p, end, reg_info)
4517 unsigned char **p, *end;
4518 register_info_type *reg_info;
4523 unsigned char *p1 = *p;
4525 switch ((re_opcode_t) *p1++)
4545 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4546 ret = group_match_null_string_p (&p1, end, reg_info);
4548 /* Have to set this here in case we're checking a group which
4549 contains a group and a back reference to it. */
4551 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4552 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4558 /* If this is an optimized succeed_n for zero times, make the jump. */
4560 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4568 /* Get to the number of times to succeed. */
4570 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4575 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4583 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
4591 /* All other opcodes mean we cannot match the empty string. */
4597 } /* common_op_match_null_string_p */
4600 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
4601 bytes; nonzero otherwise. */
4604 bcmp_translate (s1, s2, len, translate)
4605 unsigned char *s1, *s2;
4609 register unsigned char *p1 = s1, *p2 = s2;
4612 if (translate[*p1++] != translate[*p2++]) return 1;
4618 /* Entry points for GNU code. */
4620 /* re_compile_pattern is the GNU regular expression compiler: it
4621 compiles PATTERN (of length SIZE) and puts the result in BUFP.
4622 Returns 0 if the pattern was valid, otherwise an error string.
4624 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
4625 are set in BUFP on entry.
4627 We call regex_compile to do the actual compilation. */
4630 re_compile_pattern (pattern, length, bufp)
4631 const char *pattern;
4633 struct re_pattern_buffer *bufp;
4637 /* GNU code is written to assume at least RE_NREGS registers will be set
4638 (and at least one extra will be -1). */
4639 bufp->regs_allocated = REGS_UNALLOCATED;
4641 /* And GNU code determines whether or not to get register information
4642 by passing null for the REGS argument to re_match, etc., not by
4646 /* Match anchors at newline. */
4647 bufp->newline_anchor = 1;
4649 ret = regex_compile (pattern, length, re_syntax_options, bufp);
4651 return re_error_msg[(int) ret];
4654 /* Entry points compatible with 4.2 BSD regex library. We don't define
4655 them if this is an Emacs or POSIX compilation. */
4657 #if !defined (emacs) && !defined (_POSIX_SOURCE)
4659 /* BSD has one and only one pattern buffer. */
4660 static struct re_pattern_buffer re_comp_buf;
4670 if (!re_comp_buf.buffer)
4671 return "No previous regular expression";
4675 if (!re_comp_buf.buffer)
4677 re_comp_buf.buffer = (unsigned char *) malloc (200);
4678 if (re_comp_buf.buffer == NULL)
4679 return "Memory exhausted";
4680 re_comp_buf.allocated = 200;
4682 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
4683 if (re_comp_buf.fastmap == NULL)
4684 return "Memory exhausted";
4687 /* Since `re_exec' always passes NULL for the `regs' argument, we
4688 don't need to initialize the pattern buffer fields which affect it. */
4690 /* Match anchors at newlines. */
4691 re_comp_buf.newline_anchor = 1;
4693 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
4695 /* Yes, we're discarding `const' here. */
4696 return (char *) re_error_msg[(int) ret];
4704 const int len = strlen (s);
4706 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
4708 #endif /* not emacs and not _POSIX_SOURCE */
4710 /* POSIX.2 functions. Don't define these for Emacs. */
4714 /* regcomp takes a regular expression as a string and compiles it.
4716 PREG is a regex_t *. We do not expect any fields to be initialized,
4717 since POSIX says we shouldn't. Thus, we set
4719 `buffer' to the compiled pattern;
4720 `used' to the length of the compiled pattern;
4721 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
4722 REG_EXTENDED bit in CFLAGS is set; otherwise, to
4723 RE_SYNTAX_POSIX_BASIC;
4724 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
4725 `fastmap' and `fastmap_accurate' to zero;
4726 `re_nsub' to the number of subexpressions in PATTERN.
4728 PATTERN is the address of the pattern string.
4730 CFLAGS is a series of bits which affect compilation.
4732 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
4733 use POSIX basic syntax.
4735 If REG_NEWLINE is set, then . and [^...] don't match newline.
4736 Also, regexec will try a match beginning after every newline.
4738 If REG_ICASE is set, then we considers upper- and lowercase
4739 versions of letters to be equivalent when matching.
4741 If REG_NOSUB is set, then when PREG is passed to regexec, that
4742 routine will report only success or failure, and nothing about the
4745 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
4746 the return codes and their meanings.) */
4749 regcomp (preg, pattern, cflags)
4751 const char *pattern;
4756 = (cflags & REG_EXTENDED) ?
4757 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
4759 /* regex_compile will allocate the space for the compiled pattern. */
4761 preg->allocated = 0;
4763 /* Don't bother to use a fastmap when searching. This simplifies the
4764 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
4765 characters after newlines into the fastmap. This way, we just try
4769 if (cflags & REG_ICASE)
4773 preg->translate = (char *) malloc (CHAR_SET_SIZE);
4774 if (preg->translate == NULL)
4775 return (int) REG_ESPACE;
4777 /* Map uppercase characters to corresponding lowercase ones. */
4778 for (i = 0; i < CHAR_SET_SIZE; i++)
4779 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
4782 preg->translate = NULL;
4784 /* If REG_NEWLINE is set, newlines are treated differently. */
4785 if (cflags & REG_NEWLINE)
4786 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
4787 syntax &= ~RE_DOT_NEWLINE;
4788 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
4789 /* It also changes the matching behavior. */
4790 preg->newline_anchor = 1;
4793 preg->newline_anchor = 0;
4795 preg->no_sub = !!(cflags & REG_NOSUB);
4797 /* POSIX says a null character in the pattern terminates it, so we
4798 can use strlen here in compiling the pattern. */
4799 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
4801 /* POSIX doesn't distinguish between an unmatched open-group and an
4802 unmatched close-group: both are REG_EPAREN. */
4803 if (ret == REG_ERPAREN) ret = REG_EPAREN;
4809 /* regexec searches for a given pattern, specified by PREG, in the
4812 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
4813 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
4814 least NMATCH elements, and we set them to the offsets of the
4815 corresponding matched substrings.
4817 EFLAGS specifies `execution flags' which affect matching: if
4818 REG_NOTBOL is set, then ^ does not match at the beginning of the
4819 string; if REG_NOTEOL is set, then $ does not match at the end.
4821 We return 0 if we find a match and REG_NOMATCH if not. */
4824 regexec (preg, string, nmatch, pmatch, eflags)
4825 const regex_t *preg;
4828 regmatch_t pmatch[];
4832 struct re_registers regs;
4833 regex_t private_preg;
4834 int len = strlen (string);
4835 boolean want_reg_info = !preg->no_sub && nmatch > 0;
4837 private_preg = *preg;
4839 private_preg.not_bol = !!(eflags & REG_NOTBOL);
4840 private_preg.not_eol = !!(eflags & REG_NOTEOL);
4842 /* The user has told us exactly how many registers to return
4843 information about, via `nmatch'. We have to pass that on to the
4844 matching routines. */
4845 private_preg.regs_allocated = REGS_FIXED;
4849 regs.num_regs = nmatch;
4850 regs.start = TALLOC (nmatch, regoff_t);
4851 regs.end = TALLOC (nmatch, regoff_t);
4852 if (regs.start == NULL || regs.end == NULL)
4853 return (int) REG_NOMATCH;
4856 /* Perform the searching operation. */
4857 ret = re_search (&private_preg, string, len,
4858 /* start: */ 0, /* range: */ len,
4859 want_reg_info ? ®s : (struct re_registers *) 0);
4861 /* Copy the register information to the POSIX structure. */
4868 for (r = 0; r < nmatch; r++)
4870 pmatch[r].rm_so = regs.start[r];
4871 pmatch[r].rm_eo = regs.end[r];
4875 /* If we needed the temporary register info, free the space now. */
4880 /* We want zero return to mean success, unlike `re_search'. */
4881 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
4885 /* Returns a message corresponding to an error code, ERRCODE, returned
4886 from either regcomp or regexec. We don't use PREG here. */
4889 regerror (errcode, preg, errbuf, errbuf_size)
4891 const regex_t *preg;
4899 || errcode >= (sizeof (re_error_msg) / sizeof (re_error_msg[0])))
4900 /* Only error codes returned by the rest of the code should be passed
4901 to this routine. If we are given anything else, or if other regex
4902 code generates an invalid error code, then the program has a bug.
4903 Dump core so we can fix it. */
4906 msg = re_error_msg[errcode];
4908 /* POSIX doesn't require that we do anything in this case, but why
4913 msg_size = strlen (msg) + 1; /* Includes the null. */
4915 if (errbuf_size != 0)
4917 if (msg_size > errbuf_size)
4919 strncpy (errbuf, msg, errbuf_size - 1);
4920 errbuf[errbuf_size - 1] = 0;
4923 strcpy (errbuf, msg);
4930 /* Free dynamically allocated space used by PREG. */
4936 if (preg->buffer != NULL)
4937 free (preg->buffer);
4938 preg->buffer = NULL;
4940 preg->allocated = 0;
4943 if (preg->fastmap != NULL)
4944 free (preg->fastmap);
4945 preg->fastmap = NULL;
4946 preg->fastmap_accurate = 0;
4948 if (preg->translate != NULL)
4949 free (preg->translate);
4950 preg->translate = NULL;
4953 #endif /* not emacs */
4957 make-backup-files: t
4959 trim-versions-without-asking: nil