1 /* Extended regular expression matching and search library,
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
6 Copyright (C) 1993, 1994, 1995 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)
34 /* We need this for `regex.h', and perhaps for the Emacs include files. */
35 #include <sys/types.h>
37 /* This is for other GNU distributions with internationalized messages. */
38 #if HAVE_LIBINTL_H || defined (_LIBC)
41 # define gettext(msgid) (msgid)
44 /* The `emacs' switch turns on certain matching commands
45 that make sense only in Emacs. */
54 /* If we are not linking with Emacs proper,
55 we can't use the relocating allocator
56 even if config.h says that we can. */
59 #if defined (STDC_HEADERS) || defined (_LIBC)
66 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
67 If nothing else has been done, use the method below. */
68 #ifdef INHIBIT_STRING_HEADER
69 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
70 #if !defined (bzero) && !defined (bcopy)
71 #undef INHIBIT_STRING_HEADER
76 /* This is the normal way of making sure we have a bcopy and a bzero.
77 This is used in most programs--a few other programs avoid this
78 by defining INHIBIT_STRING_HEADER. */
79 #ifndef INHIBIT_STRING_HEADER
80 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
83 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
86 #define bcopy(s, d, n) memcpy ((d), (s), (n))
89 #define bzero(s, n) memset ((s), 0, (n))
96 /* Define the syntax stuff for \<, \>, etc. */
98 /* This must be nonzero for the wordchar and notwordchar pattern
99 commands in re_match_2. */
104 #ifdef SWITCH_ENUM_BUG
105 #define SWITCH_ENUM_CAST(x) ((int)(x))
107 #define SWITCH_ENUM_CAST(x) (x)
112 extern char *re_syntax_table;
114 #else /* not SYNTAX_TABLE */
116 /* How many characters in the character set. */
117 #define CHAR_SET_SIZE 256
119 static char re_syntax_table[CHAR_SET_SIZE];
130 bzero (re_syntax_table, sizeof re_syntax_table);
132 for (c = 'a'; c <= 'z'; c++)
133 re_syntax_table[c] = Sword;
135 for (c = 'A'; c <= 'Z'; c++)
136 re_syntax_table[c] = Sword;
138 for (c = '0'; c <= '9'; c++)
139 re_syntax_table[c] = Sword;
141 re_syntax_table['_'] = Sword;
146 #endif /* not SYNTAX_TABLE */
148 #define SYNTAX(c) re_syntax_table[c]
150 #endif /* not emacs */
152 /* Get the interface, including the syntax bits. */
155 /* isalpha etc. are used for the character classes. */
158 /* Jim Meyering writes:
160 "... Some ctype macros are valid only for character codes that
161 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
162 using /bin/cc or gcc but without giving an ansi option). So, all
163 ctype uses should be through macros like ISPRINT... If
164 STDC_HEADERS is defined, then autoconf has verified that the ctype
165 macros don't need to be guarded with references to isascii. ...
166 Defining isascii to 1 should let any compiler worth its salt
167 eliminate the && through constant folding." */
169 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
172 #define ISASCII(c) isascii(c)
176 #define ISBLANK(c) (ISASCII (c) && isblank (c))
178 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
181 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
183 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
186 #define ISPRINT(c) (ISASCII (c) && isprint (c))
187 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
188 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
189 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
190 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
191 #define ISLOWER(c) (ISASCII (c) && islower (c))
192 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
193 #define ISSPACE(c) (ISASCII (c) && isspace (c))
194 #define ISUPPER(c) (ISASCII (c) && isupper (c))
195 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
198 #define NULL (void *)0
201 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
202 since ours (we hope) works properly with all combinations of
203 machines, compilers, `char' and `unsigned char' argument types.
204 (Per Bothner suggested the basic approach.) */
205 #undef SIGN_EXTEND_CHAR
207 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
208 #else /* not __STDC__ */
209 /* As in Harbison and Steele. */
210 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
213 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
214 use `alloca' instead of `malloc'. This is because using malloc in
215 re_search* or re_match* could cause memory leaks when C-g is used in
216 Emacs; also, malloc is slower and causes storage fragmentation. On
217 the other hand, malloc is more portable, and easier to debug.
219 Because we sometimes use alloca, some routines have to be macros,
220 not functions -- `alloca'-allocated space disappears at the end of the
221 function it is called in. */
225 #define REGEX_ALLOCATE malloc
226 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
227 #define REGEX_FREE free
229 #else /* not REGEX_MALLOC */
231 /* Emacs already defines alloca, sometimes. */
234 /* Make alloca work the best possible way. */
236 #define alloca __builtin_alloca
237 #else /* not __GNUC__ */
240 #else /* not __GNUC__ or HAVE_ALLOCA_H */
241 #if 0 /* It is a bad idea to declare alloca. We always cast the result. */
242 #ifndef _AIX /* Already did AIX, up at the top. */
244 #endif /* not _AIX */
246 #endif /* not HAVE_ALLOCA_H */
247 #endif /* not __GNUC__ */
249 #endif /* not alloca */
251 #define REGEX_ALLOCATE alloca
253 /* Assumes a `char *destination' variable. */
254 #define REGEX_REALLOCATE(source, osize, nsize) \
255 (destination = (char *) alloca (nsize), \
256 bcopy (source, destination, osize), \
259 /* No need to do anything to free, after alloca. */
260 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
262 #endif /* not REGEX_MALLOC */
264 /* Define how to allocate the failure stack. */
266 #if defined (REL_ALLOC) && defined (REGEX_MALLOC)
268 #define REGEX_ALLOCATE_STACK(size) \
269 r_alloc (&failure_stack_ptr, (size))
270 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
271 r_re_alloc (&failure_stack_ptr, (nsize))
272 #define REGEX_FREE_STACK(ptr) \
273 r_alloc_free (&failure_stack_ptr)
275 #else /* not using relocating allocator */
279 #define REGEX_ALLOCATE_STACK malloc
280 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
281 #define REGEX_FREE_STACK free
283 #else /* not REGEX_MALLOC */
285 #define REGEX_ALLOCATE_STACK alloca
287 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
288 REGEX_REALLOCATE (source, osize, nsize)
289 /* No need to explicitly free anything. */
290 #define REGEX_FREE_STACK(arg)
292 #endif /* not REGEX_MALLOC */
293 #endif /* not using relocating allocator */
296 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
297 `string1' or just past its end. This works if PTR is NULL, which is
299 #define FIRST_STRING_P(ptr) \
300 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
302 /* (Re)Allocate N items of type T using malloc, or fail. */
303 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
304 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
305 #define RETALLOC_IF(addr, n, t) \
306 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
307 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
309 #define BYTEWIDTH 8 /* In bits. */
311 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
315 #define MAX(a, b) ((a) > (b) ? (a) : (b))
316 #define MIN(a, b) ((a) < (b) ? (a) : (b))
318 typedef char boolean;
322 static int re_match_2_internal ();
324 /* These are the command codes that appear in compiled regular
325 expressions. Some opcodes are followed by argument bytes. A
326 command code can specify any interpretation whatsoever for its
327 arguments. Zero bytes may appear in the compiled regular expression. */
333 /* Succeed right away--no more backtracking. */
336 /* Followed by one byte giving n, then by n literal bytes. */
339 /* Matches any (more or less) character. */
342 /* Matches any one char belonging to specified set. First
343 following byte is number of bitmap bytes. Then come bytes
344 for a bitmap saying which chars are in. Bits in each byte
345 are ordered low-bit-first. A character is in the set if its
346 bit is 1. A character too large to have a bit in the map is
347 automatically not in the set. */
350 /* Same parameters as charset, but match any character that is
351 not one of those specified. */
354 /* Start remembering the text that is matched, for storing in a
355 register. Followed by one byte with the register number, in
356 the range 0 to one less than the pattern buffer's re_nsub
357 field. Then followed by one byte with the number of groups
358 inner to this one. (This last has to be part of the
359 start_memory only because we need it in the on_failure_jump
363 /* Stop remembering the text that is matched and store it in a
364 memory register. Followed by one byte with the register
365 number, in the range 0 to one less than `re_nsub' in the
366 pattern buffer, and one byte with the number of inner groups,
367 just like `start_memory'. (We need the number of inner
368 groups here because we don't have any easy way of finding the
369 corresponding start_memory when we're at a stop_memory.) */
372 /* Match a duplicate of something remembered. Followed by one
373 byte containing the register number. */
376 /* Fail unless at beginning of line. */
379 /* Fail unless at end of line. */
382 /* Succeeds if at beginning of buffer (if emacs) or at beginning
383 of string to be matched (if not). */
386 /* Analogously, for end of buffer/string. */
389 /* Followed by two byte relative address to which to jump. */
392 /* Same as jump, but marks the end of an alternative. */
395 /* Followed by two-byte relative address of place to resume at
396 in case of failure. */
399 /* Like on_failure_jump, but pushes a placeholder instead of the
400 current string position when executed. */
401 on_failure_keep_string_jump,
403 /* Throw away latest failure point and then jump to following
404 two-byte relative address. */
407 /* Change to pop_failure_jump if know won't have to backtrack to
408 match; otherwise change to jump. This is used to jump
409 back to the beginning of a repeat. If what follows this jump
410 clearly won't match what the repeat does, such that we can be
411 sure that there is no use backtracking out of repetitions
412 already matched, then we change it to a pop_failure_jump.
413 Followed by two-byte address. */
416 /* Jump to following two-byte address, and push a dummy failure
417 point. This failure point will be thrown away if an attempt
418 is made to use it for a failure. A `+' construct makes this
419 before the first repeat. Also used as an intermediary kind
420 of jump when compiling an alternative. */
423 /* Push a dummy failure point and continue. Used at the end of
427 /* Followed by two-byte relative address and two-byte number n.
428 After matching N times, jump to the address upon failure. */
431 /* Followed by two-byte relative address, and two-byte number n.
432 Jump to the address N times, then fail. */
435 /* Set the following two-byte relative address to the
436 subsequent two-byte number. The address *includes* the two
440 wordchar, /* Matches any word-constituent character. */
441 notwordchar, /* Matches any char that is not a word-constituent. */
443 wordbeg, /* Succeeds if at word beginning. */
444 wordend, /* Succeeds if at word end. */
446 wordbound, /* Succeeds if at a word boundary. */
447 notwordbound /* Succeeds if not at a word boundary. */
450 ,before_dot, /* Succeeds if before point. */
451 at_dot, /* Succeeds if at point. */
452 after_dot, /* Succeeds if after point. */
454 /* Matches any character whose syntax is specified. Followed by
455 a byte which contains a syntax code, e.g., Sword. */
458 /* Matches any character whose syntax is not that specified. */
463 /* Common operations on the compiled pattern. */
465 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
467 #define STORE_NUMBER(destination, number) \
469 (destination)[0] = (number) & 0377; \
470 (destination)[1] = (number) >> 8; \
473 /* Same as STORE_NUMBER, except increment DESTINATION to
474 the byte after where the number is stored. Therefore, DESTINATION
475 must be an lvalue. */
477 #define STORE_NUMBER_AND_INCR(destination, number) \
479 STORE_NUMBER (destination, number); \
480 (destination) += 2; \
483 /* Put into DESTINATION a number stored in two contiguous bytes starting
486 #define EXTRACT_NUMBER(destination, source) \
488 (destination) = *(source) & 0377; \
489 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
494 extract_number (dest, source)
496 unsigned char *source;
498 int temp = SIGN_EXTEND_CHAR (*(source + 1));
499 *dest = *source & 0377;
503 #ifndef EXTRACT_MACROS /* To debug the macros. */
504 #undef EXTRACT_NUMBER
505 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
506 #endif /* not EXTRACT_MACROS */
510 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
511 SOURCE must be an lvalue. */
513 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
515 EXTRACT_NUMBER (destination, source); \
521 extract_number_and_incr (destination, source)
523 unsigned char **source;
525 extract_number (destination, *source);
529 #ifndef EXTRACT_MACROS
530 #undef EXTRACT_NUMBER_AND_INCR
531 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
532 extract_number_and_incr (&dest, &src)
533 #endif /* not EXTRACT_MACROS */
537 /* If DEBUG is defined, Regex prints many voluminous messages about what
538 it is doing (if the variable `debug' is nonzero). If linked with the
539 main program in `iregex.c', you can enter patterns and strings
540 interactively. And if linked with the main program in `main.c' and
541 the other test files, you can run the already-written tests. */
545 /* We use standard I/O for debugging. */
548 /* It is useful to test things that ``must'' be true when debugging. */
551 static int debug = 0;
553 #define DEBUG_STATEMENT(e) e
554 #define DEBUG_PRINT1(x) if (debug) printf (x)
555 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
556 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
557 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
558 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
559 if (debug) print_partial_compiled_pattern (s, e)
560 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
561 if (debug) print_double_string (w, s1, sz1, s2, sz2)
564 /* Print the fastmap in human-readable form. */
567 print_fastmap (fastmap)
570 unsigned was_a_range = 0;
573 while (i < (1 << BYTEWIDTH))
579 while (i < (1 << BYTEWIDTH) && fastmap[i])
595 /* Print a compiled pattern string in human-readable form, starting at
596 the START pointer into it and ending just before the pointer END. */
599 print_partial_compiled_pattern (start, end)
600 unsigned char *start;
604 unsigned char *p = start;
605 unsigned char *pend = end;
613 /* Loop over pattern commands. */
616 printf ("%d:\t", p - start);
618 switch ((re_opcode_t) *p++)
626 printf ("/exactn/%d", mcnt);
637 printf ("/start_memory/%d/%d", mcnt, *p++);
642 printf ("/stop_memory/%d/%d", mcnt, *p++);
646 printf ("/duplicate/%d", *p++);
656 register int c, last = -100;
657 register int in_range = 0;
659 printf ("/charset [%s",
660 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
662 assert (p + *p < pend);
664 for (c = 0; c < 256; c++)
666 && (p[1 + (c/8)] & (1 << (c % 8))))
668 /* Are we starting a range? */
669 if (last + 1 == c && ! in_range)
674 /* Have we broken a range? */
675 else if (last + 1 != c && in_range)
704 case on_failure_jump:
705 extract_number_and_incr (&mcnt, &p);
706 printf ("/on_failure_jump to %d", p + mcnt - start);
709 case on_failure_keep_string_jump:
710 extract_number_and_incr (&mcnt, &p);
711 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
714 case dummy_failure_jump:
715 extract_number_and_incr (&mcnt, &p);
716 printf ("/dummy_failure_jump to %d", p + mcnt - start);
719 case push_dummy_failure:
720 printf ("/push_dummy_failure");
724 extract_number_and_incr (&mcnt, &p);
725 printf ("/maybe_pop_jump to %d", p + mcnt - start);
728 case pop_failure_jump:
729 extract_number_and_incr (&mcnt, &p);
730 printf ("/pop_failure_jump to %d", p + mcnt - start);
734 extract_number_and_incr (&mcnt, &p);
735 printf ("/jump_past_alt to %d", p + mcnt - start);
739 extract_number_and_incr (&mcnt, &p);
740 printf ("/jump to %d", p + mcnt - start);
744 extract_number_and_incr (&mcnt, &p);
745 extract_number_and_incr (&mcnt2, &p);
746 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
750 extract_number_and_incr (&mcnt, &p);
751 extract_number_and_incr (&mcnt2, &p);
752 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
756 extract_number_and_incr (&mcnt, &p);
757 extract_number_and_incr (&mcnt2, &p);
758 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
762 printf ("/wordbound");
766 printf ("/notwordbound");
778 printf ("/before_dot");
786 printf ("/after_dot");
790 printf ("/syntaxspec");
792 printf ("/%d", mcnt);
796 printf ("/notsyntaxspec");
798 printf ("/%d", mcnt);
803 printf ("/wordchar");
807 printf ("/notwordchar");
819 printf ("?%d", *(p-1));
825 printf ("%d:\tend of pattern.\n", p - start);
830 print_compiled_pattern (bufp)
831 struct re_pattern_buffer *bufp;
833 unsigned char *buffer = bufp->buffer;
835 print_partial_compiled_pattern (buffer, buffer + bufp->used);
836 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
838 if (bufp->fastmap_accurate && bufp->fastmap)
840 printf ("fastmap: ");
841 print_fastmap (bufp->fastmap);
844 printf ("re_nsub: %d\t", bufp->re_nsub);
845 printf ("regs_alloc: %d\t", bufp->regs_allocated);
846 printf ("can_be_null: %d\t", bufp->can_be_null);
847 printf ("newline_anchor: %d\n", bufp->newline_anchor);
848 printf ("no_sub: %d\t", bufp->no_sub);
849 printf ("not_bol: %d\t", bufp->not_bol);
850 printf ("not_eol: %d\t", bufp->not_eol);
851 printf ("syntax: %d\n", bufp->syntax);
852 /* Perhaps we should print the translate table? */
857 print_double_string (where, string1, size1, string2, size2)
870 if (FIRST_STRING_P (where))
872 for (this_char = where - string1; this_char < size1; this_char++)
873 putchar (string1[this_char]);
878 for (this_char = where - string2; this_char < size2; this_char++)
879 putchar (string2[this_char]);
883 #else /* not DEBUG */
888 #define DEBUG_STATEMENT(e)
889 #define DEBUG_PRINT1(x)
890 #define DEBUG_PRINT2(x1, x2)
891 #define DEBUG_PRINT3(x1, x2, x3)
892 #define DEBUG_PRINT4(x1, x2, x3, x4)
893 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
894 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
896 #endif /* not DEBUG */
898 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
899 also be assigned to arbitrarily: each pattern buffer stores its own
900 syntax, so it can be changed between regex compilations. */
901 /* This has no initializer because initialized variables in Emacs
902 become read-only after dumping. */
903 reg_syntax_t re_syntax_options;
906 /* Specify the precise syntax of regexps for compilation. This provides
907 for compatibility for various utilities which historically have
908 different, incompatible syntaxes.
910 The argument SYNTAX is a bit mask comprised of the various bits
911 defined in regex.h. We return the old syntax. */
914 re_set_syntax (syntax)
917 reg_syntax_t ret = re_syntax_options;
919 re_syntax_options = syntax;
923 /* This table gives an error message for each of the error codes listed
924 in regex.h. Obviously the order here has to be same as there.
925 POSIX doesn't require that we do anything for REG_NOERROR,
926 but why not be nice? */
928 static const char *re_error_msgid[] =
929 { "Success", /* REG_NOERROR */
930 "No match", /* REG_NOMATCH */
931 "Invalid regular expression", /* REG_BADPAT */
932 "Invalid collation character", /* REG_ECOLLATE */
933 "Invalid character class name", /* REG_ECTYPE */
934 "Trailing backslash", /* REG_EESCAPE */
935 "Invalid back reference", /* REG_ESUBREG */
936 "Unmatched [ or [^", /* REG_EBRACK */
937 "Unmatched ( or \\(", /* REG_EPAREN */
938 "Unmatched \\{", /* REG_EBRACE */
939 "Invalid content of \\{\\}", /* REG_BADBR */
940 "Invalid range end", /* REG_ERANGE */
941 "Memory exhausted", /* REG_ESPACE */
942 "Invalid preceding regular expression", /* REG_BADRPT */
943 "Premature end of regular expression", /* REG_EEND */
944 "Regular expression too big", /* REG_ESIZE */
945 "Unmatched ) or \\)", /* REG_ERPAREN */
948 /* Avoiding alloca during matching, to placate r_alloc. */
950 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
951 searching and matching functions should not call alloca. On some
952 systems, alloca is implemented in terms of malloc, and if we're
953 using the relocating allocator routines, then malloc could cause a
954 relocation, which might (if the strings being searched are in the
955 ralloc heap) shift the data out from underneath the regexp
958 Here's another reason to avoid allocation: Emacs
959 processes input from X in a signal handler; processing X input may
960 call malloc; if input arrives while a matching routine is calling
961 malloc, then we're scrod. But Emacs can't just block input while
962 calling matching routines; then we don't notice interrupts when
963 they come in. So, Emacs blocks input around all regexp calls
964 except the matching calls, which it leaves unprotected, in the
965 faith that they will not malloc. */
967 /* Normally, this is fine. */
968 #define MATCH_MAY_ALLOCATE
970 /* When using GNU C, we are not REALLY using the C alloca, no matter
971 what config.h may say. So don't take precautions for it. */
976 /* The match routines may not allocate if (1) they would do it with malloc
977 and (2) it's not safe for them to use malloc.
978 Note that if REL_ALLOC is defined, matching would not use malloc for the
979 failure stack, but we would still use it for the register vectors;
980 so REL_ALLOC should not affect this. */
981 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
982 #undef MATCH_MAY_ALLOCATE
986 /* Failure stack declarations and macros; both re_compile_fastmap and
987 re_match_2 use a failure stack. These have to be macros because of
988 REGEX_ALLOCATE_STACK. */
991 /* Number of failure points for which to initially allocate space
992 when matching. If this number is exceeded, we allocate more
993 space, so it is not a hard limit. */
994 #ifndef INIT_FAILURE_ALLOC
995 #define INIT_FAILURE_ALLOC 5
998 /* Roughly the maximum number of failure points on the stack. Would be
999 exactly that if always used MAX_FAILURE_SPACE each time we failed.
1000 This is a variable only so users of regex can assign to it; we never
1001 change it ourselves. */
1002 #if defined (MATCH_MAY_ALLOCATE)
1003 int re_max_failures = 20000;
1005 int re_max_failures = 2000;
1008 union fail_stack_elt
1010 unsigned char *pointer;
1014 typedef union fail_stack_elt fail_stack_elt_t;
1018 fail_stack_elt_t *stack;
1020 unsigned avail; /* Offset of next open position. */
1023 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1024 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1025 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1028 /* Define macros to initialize and free the failure stack.
1029 Do `return -2' if the alloc fails. */
1031 #ifdef MATCH_MAY_ALLOCATE
1032 #define INIT_FAIL_STACK() \
1034 fail_stack.stack = (fail_stack_elt_t *) \
1035 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1037 if (fail_stack.stack == NULL) \
1040 fail_stack.size = INIT_FAILURE_ALLOC; \
1041 fail_stack.avail = 0; \
1044 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1046 #define INIT_FAIL_STACK() \
1048 fail_stack.avail = 0; \
1051 #define RESET_FAIL_STACK()
1055 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1057 Return 1 if succeeds, and 0 if either ran out of memory
1058 allocating space for it or it was already too large.
1060 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1062 #define DOUBLE_FAIL_STACK(fail_stack) \
1063 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1065 : ((fail_stack).stack = (fail_stack_elt_t *) \
1066 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1067 (fail_stack).size * sizeof (fail_stack_elt_t), \
1068 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1070 (fail_stack).stack == NULL \
1072 : ((fail_stack).size <<= 1, \
1076 /* Push pointer POINTER on FAIL_STACK.
1077 Return 1 if was able to do so and 0 if ran out of memory allocating
1079 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1080 ((FAIL_STACK_FULL () \
1081 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1083 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1086 /* Push a pointer value onto the failure stack.
1087 Assumes the variable `fail_stack'. Probably should only
1088 be called from within `PUSH_FAILURE_POINT'. */
1089 #define PUSH_FAILURE_POINTER(item) \
1090 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1092 /* This pushes an integer-valued item onto the failure stack.
1093 Assumes the variable `fail_stack'. Probably should only
1094 be called from within `PUSH_FAILURE_POINT'. */
1095 #define PUSH_FAILURE_INT(item) \
1096 fail_stack.stack[fail_stack.avail++].integer = (item)
1098 /* Push a fail_stack_elt_t value onto the failure stack.
1099 Assumes the variable `fail_stack'. Probably should only
1100 be called from within `PUSH_FAILURE_POINT'. */
1101 #define PUSH_FAILURE_ELT(item) \
1102 fail_stack.stack[fail_stack.avail++] = (item)
1104 /* These three POP... operations complement the three PUSH... operations.
1105 All assume that `fail_stack' is nonempty. */
1106 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1107 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1108 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1110 /* Used to omit pushing failure point id's when we're not debugging. */
1112 #define DEBUG_PUSH PUSH_FAILURE_INT
1113 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1115 #define DEBUG_PUSH(item)
1116 #define DEBUG_POP(item_addr)
1120 /* Push the information about the state we will need
1121 if we ever fail back to it.
1123 Requires variables fail_stack, regstart, regend, reg_info, and
1124 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1127 Does `return FAILURE_CODE' if runs out of memory. */
1129 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1131 char *destination; \
1132 /* Must be int, so when we don't save any registers, the arithmetic \
1133 of 0 + -1 isn't done as unsigned. */ \
1136 DEBUG_STATEMENT (failure_id++); \
1137 DEBUG_STATEMENT (nfailure_points_pushed++); \
1138 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1139 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1140 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1142 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1143 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1145 /* Ensure we have enough space allocated for what we will push. */ \
1146 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1148 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1149 return failure_code; \
1151 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1152 (fail_stack).size); \
1153 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1156 /* Push the info, starting with the registers. */ \
1157 DEBUG_PRINT1 ("\n"); \
1160 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1163 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1164 DEBUG_STATEMENT (num_regs_pushed++); \
1166 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1167 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1169 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1170 PUSH_FAILURE_POINTER (regend[this_reg]); \
1172 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1173 DEBUG_PRINT2 (" match_null=%d", \
1174 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1175 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1176 DEBUG_PRINT2 (" matched_something=%d", \
1177 MATCHED_SOMETHING (reg_info[this_reg])); \
1178 DEBUG_PRINT2 (" ever_matched=%d", \
1179 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1180 DEBUG_PRINT1 ("\n"); \
1181 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1184 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1185 PUSH_FAILURE_INT (lowest_active_reg); \
1187 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1188 PUSH_FAILURE_INT (highest_active_reg); \
1190 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1191 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1192 PUSH_FAILURE_POINTER (pattern_place); \
1194 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1195 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1197 DEBUG_PRINT1 ("'\n"); \
1198 PUSH_FAILURE_POINTER (string_place); \
1200 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1201 DEBUG_PUSH (failure_id); \
1204 /* This is the number of items that are pushed and popped on the stack
1205 for each register. */
1206 #define NUM_REG_ITEMS 3
1208 /* Individual items aside from the registers. */
1210 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1212 #define NUM_NONREG_ITEMS 4
1215 /* We push at most this many items on the stack. */
1216 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1218 /* We actually push this many items. */
1219 #define NUM_FAILURE_ITEMS \
1221 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1225 /* How many items can still be added to the stack without overflowing it. */
1226 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1229 /* Pops what PUSH_FAIL_STACK pushes.
1231 We restore into the parameters, all of which should be lvalues:
1232 STR -- the saved data position.
1233 PAT -- the saved pattern position.
1234 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1235 REGSTART, REGEND -- arrays of string positions.
1236 REG_INFO -- array of information about each subexpression.
1238 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1239 `pend', `string1', `size1', `string2', and `size2'. */
1241 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1243 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1245 const unsigned char *string_temp; \
1247 assert (!FAIL_STACK_EMPTY ()); \
1249 /* Remove failure points and point to how many regs pushed. */ \
1250 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1251 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1252 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1254 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1256 DEBUG_POP (&failure_id); \
1257 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1259 /* If the saved string location is NULL, it came from an \
1260 on_failure_keep_string_jump opcode, and we want to throw away the \
1261 saved NULL, thus retaining our current position in the string. */ \
1262 string_temp = POP_FAILURE_POINTER (); \
1263 if (string_temp != NULL) \
1264 str = (const char *) string_temp; \
1266 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1267 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1268 DEBUG_PRINT1 ("'\n"); \
1270 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1271 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1272 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1274 /* Restore register info. */ \
1275 high_reg = (unsigned) POP_FAILURE_INT (); \
1276 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1278 low_reg = (unsigned) POP_FAILURE_INT (); \
1279 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1282 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1284 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1286 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1287 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1289 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1290 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1292 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1293 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1297 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1299 reg_info[this_reg].word.integer = 0; \
1300 regend[this_reg] = 0; \
1301 regstart[this_reg] = 0; \
1303 highest_active_reg = high_reg; \
1306 set_regs_matched_done = 0; \
1307 DEBUG_STATEMENT (nfailure_points_popped++); \
1308 } /* POP_FAILURE_POINT */
1312 /* Structure for per-register (a.k.a. per-group) information.
1313 Other register information, such as the
1314 starting and ending positions (which are addresses), and the list of
1315 inner groups (which is a bits list) are maintained in separate
1318 We are making a (strictly speaking) nonportable assumption here: that
1319 the compiler will pack our bit fields into something that fits into
1320 the type of `word', i.e., is something that fits into one item on the
1325 fail_stack_elt_t word;
1328 /* This field is one if this group can match the empty string,
1329 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1330 #define MATCH_NULL_UNSET_VALUE 3
1331 unsigned match_null_string_p : 2;
1332 unsigned is_active : 1;
1333 unsigned matched_something : 1;
1334 unsigned ever_matched_something : 1;
1336 } register_info_type;
1338 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1339 #define IS_ACTIVE(R) ((R).bits.is_active)
1340 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1341 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1344 /* Call this when have matched a real character; it sets `matched' flags
1345 for the subexpressions which we are currently inside. Also records
1346 that those subexprs have matched. */
1347 #define SET_REGS_MATCHED() \
1350 if (!set_regs_matched_done) \
1353 set_regs_matched_done = 1; \
1354 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1356 MATCHED_SOMETHING (reg_info[r]) \
1357 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1364 /* Registers are set to a sentinel when they haven't yet matched. */
1365 static char reg_unset_dummy;
1366 #define REG_UNSET_VALUE (®_unset_dummy)
1367 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1369 /* Subroutine declarations and macros for regex_compile. */
1371 static void store_op1 (), store_op2 ();
1372 static void insert_op1 (), insert_op2 ();
1373 static boolean at_begline_loc_p (), at_endline_loc_p ();
1374 static boolean group_in_compile_stack ();
1375 static reg_errcode_t compile_range ();
1377 /* Fetch the next character in the uncompiled pattern---translating it
1378 if necessary. Also cast from a signed character in the constant
1379 string passed to us by the user to an unsigned char that we can use
1380 as an array index (in, e.g., `translate'). */
1382 #define PATFETCH(c) \
1383 do {if (p == pend) return REG_EEND; \
1384 c = (unsigned char) *p++; \
1385 if (translate) c = (unsigned char) translate[c]; \
1389 /* Fetch the next character in the uncompiled pattern, with no
1391 #define PATFETCH_RAW(c) \
1392 do {if (p == pend) return REG_EEND; \
1393 c = (unsigned char) *p++; \
1396 /* Go backwards one character in the pattern. */
1397 #define PATUNFETCH p--
1400 /* If `translate' is non-null, return translate[D], else just D. We
1401 cast the subscript to translate because some data is declared as
1402 `char *', to avoid warnings when a string constant is passed. But
1403 when we use a character as a subscript we must make it unsigned. */
1405 #define TRANSLATE(d) \
1406 (translate ? (char) translate[(unsigned char) (d)] : (d))
1410 /* Macros for outputting the compiled pattern into `buffer'. */
1412 /* If the buffer isn't allocated when it comes in, use this. */
1413 #define INIT_BUF_SIZE 32
1415 /* Make sure we have at least N more bytes of space in buffer. */
1416 #define GET_BUFFER_SPACE(n) \
1417 while (b - bufp->buffer + (n) > bufp->allocated) \
1420 /* Make sure we have one more byte of buffer space and then add C to it. */
1421 #define BUF_PUSH(c) \
1423 GET_BUFFER_SPACE (1); \
1424 *b++ = (unsigned char) (c); \
1428 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1429 #define BUF_PUSH_2(c1, c2) \
1431 GET_BUFFER_SPACE (2); \
1432 *b++ = (unsigned char) (c1); \
1433 *b++ = (unsigned char) (c2); \
1437 /* As with BUF_PUSH_2, except for three bytes. */
1438 #define BUF_PUSH_3(c1, c2, c3) \
1440 GET_BUFFER_SPACE (3); \
1441 *b++ = (unsigned char) (c1); \
1442 *b++ = (unsigned char) (c2); \
1443 *b++ = (unsigned char) (c3); \
1447 /* Store a jump with opcode OP at LOC to location TO. We store a
1448 relative address offset by the three bytes the jump itself occupies. */
1449 #define STORE_JUMP(op, loc, to) \
1450 store_op1 (op, loc, (to) - (loc) - 3)
1452 /* Likewise, for a two-argument jump. */
1453 #define STORE_JUMP2(op, loc, to, arg) \
1454 store_op2 (op, loc, (to) - (loc) - 3, arg)
1456 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1457 #define INSERT_JUMP(op, loc, to) \
1458 insert_op1 (op, loc, (to) - (loc) - 3, b)
1460 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1461 #define INSERT_JUMP2(op, loc, to, arg) \
1462 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1465 /* This is not an arbitrary limit: the arguments which represent offsets
1466 into the pattern are two bytes long. So if 2^16 bytes turns out to
1467 be too small, many things would have to change. */
1468 #define MAX_BUF_SIZE (1L << 16)
1471 /* Extend the buffer by twice its current size via realloc and
1472 reset the pointers that pointed into the old block to point to the
1473 correct places in the new one. If extending the buffer results in it
1474 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1475 #define EXTEND_BUFFER() \
1477 unsigned char *old_buffer = bufp->buffer; \
1478 if (bufp->allocated == MAX_BUF_SIZE) \
1480 bufp->allocated <<= 1; \
1481 if (bufp->allocated > MAX_BUF_SIZE) \
1482 bufp->allocated = MAX_BUF_SIZE; \
1483 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1484 if (bufp->buffer == NULL) \
1485 return REG_ESPACE; \
1486 /* If the buffer moved, move all the pointers into it. */ \
1487 if (old_buffer != bufp->buffer) \
1489 b = (b - old_buffer) + bufp->buffer; \
1490 begalt = (begalt - old_buffer) + bufp->buffer; \
1491 if (fixup_alt_jump) \
1492 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1494 laststart = (laststart - old_buffer) + bufp->buffer; \
1495 if (pending_exact) \
1496 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1501 /* Since we have one byte reserved for the register number argument to
1502 {start,stop}_memory, the maximum number of groups we can report
1503 things about is what fits in that byte. */
1504 #define MAX_REGNUM 255
1506 /* But patterns can have more than `MAX_REGNUM' registers. We just
1507 ignore the excess. */
1508 typedef unsigned regnum_t;
1511 /* Macros for the compile stack. */
1513 /* Since offsets can go either forwards or backwards, this type needs to
1514 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1515 typedef int pattern_offset_t;
1519 pattern_offset_t begalt_offset;
1520 pattern_offset_t fixup_alt_jump;
1521 pattern_offset_t inner_group_offset;
1522 pattern_offset_t laststart_offset;
1524 } compile_stack_elt_t;
1529 compile_stack_elt_t *stack;
1531 unsigned avail; /* Offset of next open position. */
1532 } compile_stack_type;
1535 #define INIT_COMPILE_STACK_SIZE 32
1537 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1538 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1540 /* The next available element. */
1541 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1544 /* Set the bit for character C in a list. */
1545 #define SET_LIST_BIT(c) \
1546 (b[((unsigned char) (c)) / BYTEWIDTH] \
1547 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1550 /* Get the next unsigned number in the uncompiled pattern. */
1551 #define GET_UNSIGNED_NUMBER(num) \
1555 while (ISDIGIT (c)) \
1559 num = num * 10 + c - '0'; \
1567 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1569 #define IS_CHAR_CLASS(string) \
1570 (STREQ (string, "alpha") || STREQ (string, "upper") \
1571 || STREQ (string, "lower") || STREQ (string, "digit") \
1572 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1573 || STREQ (string, "space") || STREQ (string, "print") \
1574 || STREQ (string, "punct") || STREQ (string, "graph") \
1575 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1577 #ifndef MATCH_MAY_ALLOCATE
1579 /* If we cannot allocate large objects within re_match_2_internal,
1580 we make the fail stack and register vectors global.
1581 The fail stack, we grow to the maximum size when a regexp
1583 The register vectors, we adjust in size each time we
1584 compile a regexp, according to the number of registers it needs. */
1586 static fail_stack_type fail_stack;
1588 /* Size with which the following vectors are currently allocated.
1589 That is so we can make them bigger as needed,
1590 but never make them smaller. */
1591 static int regs_allocated_size;
1593 static const char ** regstart, ** regend;
1594 static const char ** old_regstart, ** old_regend;
1595 static const char **best_regstart, **best_regend;
1596 static register_info_type *reg_info;
1597 static const char **reg_dummy;
1598 static register_info_type *reg_info_dummy;
1600 /* Make the register vectors big enough for NUM_REGS registers,
1601 but don't make them smaller. */
1604 regex_grow_registers (num_regs)
1607 if (num_regs > regs_allocated_size)
1609 RETALLOC_IF (regstart, num_regs, const char *);
1610 RETALLOC_IF (regend, num_regs, const char *);
1611 RETALLOC_IF (old_regstart, num_regs, const char *);
1612 RETALLOC_IF (old_regend, num_regs, const char *);
1613 RETALLOC_IF (best_regstart, num_regs, const char *);
1614 RETALLOC_IF (best_regend, num_regs, const char *);
1615 RETALLOC_IF (reg_info, num_regs, register_info_type);
1616 RETALLOC_IF (reg_dummy, num_regs, const char *);
1617 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1619 regs_allocated_size = num_regs;
1623 #endif /* not MATCH_MAY_ALLOCATE */
1625 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1626 Returns one of error codes defined in `regex.h', or zero for success.
1628 Assumes the `allocated' (and perhaps `buffer') and `translate'
1629 fields are set in BUFP on entry.
1631 If it succeeds, results are put in BUFP (if it returns an error, the
1632 contents of BUFP are undefined):
1633 `buffer' is the compiled pattern;
1634 `syntax' is set to SYNTAX;
1635 `used' is set to the length of the compiled pattern;
1636 `fastmap_accurate' is zero;
1637 `re_nsub' is the number of subexpressions in PATTERN;
1638 `not_bol' and `not_eol' are zero;
1640 The `fastmap' and `newline_anchor' fields are neither
1641 examined nor set. */
1643 /* Return, freeing storage we allocated. */
1644 #define FREE_STACK_RETURN(value) \
1645 return (free (compile_stack.stack), value)
1647 static reg_errcode_t
1648 regex_compile (pattern, size, syntax, bufp)
1649 const char *pattern;
1651 reg_syntax_t syntax;
1652 struct re_pattern_buffer *bufp;
1654 /* We fetch characters from PATTERN here. Even though PATTERN is
1655 `char *' (i.e., signed), we declare these variables as unsigned, so
1656 they can be reliably used as array indices. */
1657 register unsigned char c, c1;
1659 /* A random temporary spot in PATTERN. */
1662 /* Points to the end of the buffer, where we should append. */
1663 register unsigned char *b;
1665 /* Keeps track of unclosed groups. */
1666 compile_stack_type compile_stack;
1668 /* Points to the current (ending) position in the pattern. */
1669 const char *p = pattern;
1670 const char *pend = pattern + size;
1672 /* How to translate the characters in the pattern. */
1673 RE_TRANSLATE_TYPE translate = bufp->translate;
1675 /* Address of the count-byte of the most recently inserted `exactn'
1676 command. This makes it possible to tell if a new exact-match
1677 character can be added to that command or if the character requires
1678 a new `exactn' command. */
1679 unsigned char *pending_exact = 0;
1681 /* Address of start of the most recently finished expression.
1682 This tells, e.g., postfix * where to find the start of its
1683 operand. Reset at the beginning of groups and alternatives. */
1684 unsigned char *laststart = 0;
1686 /* Address of beginning of regexp, or inside of last group. */
1687 unsigned char *begalt;
1689 /* Place in the uncompiled pattern (i.e., the {) to
1690 which to go back if the interval is invalid. */
1691 const char *beg_interval;
1693 /* Address of the place where a forward jump should go to the end of
1694 the containing expression. Each alternative of an `or' -- except the
1695 last -- ends with a forward jump of this sort. */
1696 unsigned char *fixup_alt_jump = 0;
1698 /* Counts open-groups as they are encountered. Remembered for the
1699 matching close-group on the compile stack, so the same register
1700 number is put in the stop_memory as the start_memory. */
1701 regnum_t regnum = 0;
1704 DEBUG_PRINT1 ("\nCompiling pattern: ");
1707 unsigned debug_count;
1709 for (debug_count = 0; debug_count < size; debug_count++)
1710 putchar (pattern[debug_count]);
1715 /* Initialize the compile stack. */
1716 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1717 if (compile_stack.stack == NULL)
1720 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1721 compile_stack.avail = 0;
1723 /* Initialize the pattern buffer. */
1724 bufp->syntax = syntax;
1725 bufp->fastmap_accurate = 0;
1726 bufp->not_bol = bufp->not_eol = 0;
1728 /* Set `used' to zero, so that if we return an error, the pattern
1729 printer (for debugging) will think there's no pattern. We reset it
1733 /* Always count groups, whether or not bufp->no_sub is set. */
1736 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1737 /* Initialize the syntax table. */
1738 init_syntax_once ();
1741 if (bufp->allocated == 0)
1744 { /* If zero allocated, but buffer is non-null, try to realloc
1745 enough space. This loses if buffer's address is bogus, but
1746 that is the user's responsibility. */
1747 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1750 { /* Caller did not allocate a buffer. Do it for them. */
1751 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1753 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1755 bufp->allocated = INIT_BUF_SIZE;
1758 begalt = b = bufp->buffer;
1760 /* Loop through the uncompiled pattern until we're at the end. */
1769 if ( /* If at start of pattern, it's an operator. */
1771 /* If context independent, it's an operator. */
1772 || syntax & RE_CONTEXT_INDEP_ANCHORS
1773 /* Otherwise, depends on what's come before. */
1774 || at_begline_loc_p (pattern, p, syntax))
1784 if ( /* If at end of pattern, it's an operator. */
1786 /* If context independent, it's an operator. */
1787 || syntax & RE_CONTEXT_INDEP_ANCHORS
1788 /* Otherwise, depends on what's next. */
1789 || at_endline_loc_p (p, pend, syntax))
1799 if ((syntax & RE_BK_PLUS_QM)
1800 || (syntax & RE_LIMITED_OPS))
1804 /* If there is no previous pattern... */
1807 if (syntax & RE_CONTEXT_INVALID_OPS)
1808 FREE_STACK_RETURN (REG_BADRPT);
1809 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1814 /* Are we optimizing this jump? */
1815 boolean keep_string_p = false;
1817 /* 1 means zero (many) matches is allowed. */
1818 char zero_times_ok = 0, many_times_ok = 0;
1820 /* If there is a sequence of repetition chars, collapse it
1821 down to just one (the right one). We can't combine
1822 interval operators with these because of, e.g., `a{2}*',
1823 which should only match an even number of `a's. */
1827 zero_times_ok |= c != '+';
1828 many_times_ok |= c != '?';
1836 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1839 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1841 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1844 if (!(c1 == '+' || c1 == '?'))
1859 /* If we get here, we found another repeat character. */
1862 /* Star, etc. applied to an empty pattern is equivalent
1863 to an empty pattern. */
1867 /* Now we know whether or not zero matches is allowed
1868 and also whether or not two or more matches is allowed. */
1870 { /* More than one repetition is allowed, so put in at the
1871 end a backward relative jump from `b' to before the next
1872 jump we're going to put in below (which jumps from
1873 laststart to after this jump).
1875 But if we are at the `*' in the exact sequence `.*\n',
1876 insert an unconditional jump backwards to the .,
1877 instead of the beginning of the loop. This way we only
1878 push a failure point once, instead of every time
1879 through the loop. */
1880 assert (p - 1 > pattern);
1882 /* Allocate the space for the jump. */
1883 GET_BUFFER_SPACE (3);
1885 /* We know we are not at the first character of the pattern,
1886 because laststart was nonzero. And we've already
1887 incremented `p', by the way, to be the character after
1888 the `*'. Do we have to do something analogous here
1889 for null bytes, because of RE_DOT_NOT_NULL? */
1890 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1892 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1893 && !(syntax & RE_DOT_NEWLINE))
1894 { /* We have .*\n. */
1895 STORE_JUMP (jump, b, laststart);
1896 keep_string_p = true;
1899 /* Anything else. */
1900 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1902 /* We've added more stuff to the buffer. */
1906 /* On failure, jump from laststart to b + 3, which will be the
1907 end of the buffer after this jump is inserted. */
1908 GET_BUFFER_SPACE (3);
1909 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1917 /* At least one repetition is required, so insert a
1918 `dummy_failure_jump' before the initial
1919 `on_failure_jump' instruction of the loop. This
1920 effects a skip over that instruction the first time
1921 we hit that loop. */
1922 GET_BUFFER_SPACE (3);
1923 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1938 boolean had_char_class = false;
1940 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1942 /* Ensure that we have enough space to push a charset: the
1943 opcode, the length count, and the bitset; 34 bytes in all. */
1944 GET_BUFFER_SPACE (34);
1948 /* We test `*p == '^' twice, instead of using an if
1949 statement, so we only need one BUF_PUSH. */
1950 BUF_PUSH (*p == '^' ? charset_not : charset);
1954 /* Remember the first position in the bracket expression. */
1957 /* Push the number of bytes in the bitmap. */
1958 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1960 /* Clear the whole map. */
1961 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1963 /* charset_not matches newline according to a syntax bit. */
1964 if ((re_opcode_t) b[-2] == charset_not
1965 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1966 SET_LIST_BIT ('\n');
1968 /* Read in characters and ranges, setting map bits. */
1971 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1975 /* \ might escape characters inside [...] and [^...]. */
1976 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1978 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1985 /* Could be the end of the bracket expression. If it's
1986 not (i.e., when the bracket expression is `[]' so
1987 far), the ']' character bit gets set way below. */
1988 if (c == ']' && p != p1 + 1)
1991 /* Look ahead to see if it's a range when the last thing
1992 was a character class. */
1993 if (had_char_class && c == '-' && *p != ']')
1994 FREE_STACK_RETURN (REG_ERANGE);
1996 /* Look ahead to see if it's a range when the last thing
1997 was a character: if this is a hyphen not at the
1998 beginning or the end of a list, then it's the range
2001 && !(p - 2 >= pattern && p[-2] == '[')
2002 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2006 = compile_range (&p, pend, translate, syntax, b);
2007 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2010 else if (p[0] == '-' && p[1] != ']')
2011 { /* This handles ranges made up of characters only. */
2014 /* Move past the `-'. */
2017 ret = compile_range (&p, pend, translate, syntax, b);
2018 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2021 /* See if we're at the beginning of a possible character
2024 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2025 { /* Leave room for the null. */
2026 char str[CHAR_CLASS_MAX_LENGTH + 1];
2031 /* If pattern is `[[:'. */
2032 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2037 if (c == ':' || c == ']' || p == pend
2038 || c1 == CHAR_CLASS_MAX_LENGTH)
2044 /* If isn't a word bracketed by `[:' and:`]':
2045 undo the ending character, the letters, and leave
2046 the leading `:' and `[' (but set bits for them). */
2047 if (c == ':' && *p == ']')
2050 boolean is_alnum = STREQ (str, "alnum");
2051 boolean is_alpha = STREQ (str, "alpha");
2052 boolean is_blank = STREQ (str, "blank");
2053 boolean is_cntrl = STREQ (str, "cntrl");
2054 boolean is_digit = STREQ (str, "digit");
2055 boolean is_graph = STREQ (str, "graph");
2056 boolean is_lower = STREQ (str, "lower");
2057 boolean is_print = STREQ (str, "print");
2058 boolean is_punct = STREQ (str, "punct");
2059 boolean is_space = STREQ (str, "space");
2060 boolean is_upper = STREQ (str, "upper");
2061 boolean is_xdigit = STREQ (str, "xdigit");
2063 if (!IS_CHAR_CLASS (str))
2064 FREE_STACK_RETURN (REG_ECTYPE);
2066 /* Throw away the ] at the end of the character
2070 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2072 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2074 /* This was split into 3 if's to
2075 avoid an arbitrary limit in some compiler. */
2076 if ( (is_alnum && ISALNUM (ch))
2077 || (is_alpha && ISALPHA (ch))
2078 || (is_blank && ISBLANK (ch))
2079 || (is_cntrl && ISCNTRL (ch)))
2081 if ( (is_digit && ISDIGIT (ch))
2082 || (is_graph && ISGRAPH (ch))
2083 || (is_lower && ISLOWER (ch))
2084 || (is_print && ISPRINT (ch)))
2086 if ( (is_punct && ISPUNCT (ch))
2087 || (is_space && ISSPACE (ch))
2088 || (is_upper && ISUPPER (ch))
2089 || (is_xdigit && ISXDIGIT (ch)))
2092 had_char_class = true;
2101 had_char_class = false;
2106 had_char_class = false;
2111 /* Discard any (non)matching list bytes that are all 0 at the
2112 end of the map. Decrease the map-length byte too. */
2113 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2121 if (syntax & RE_NO_BK_PARENS)
2128 if (syntax & RE_NO_BK_PARENS)
2135 if (syntax & RE_NEWLINE_ALT)
2142 if (syntax & RE_NO_BK_VBAR)
2149 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2150 goto handle_interval;
2156 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2158 /* Do not translate the character after the \, so that we can
2159 distinguish, e.g., \B from \b, even if we normally would
2160 translate, e.g., B to b. */
2166 if (syntax & RE_NO_BK_PARENS)
2167 goto normal_backslash;
2173 if (COMPILE_STACK_FULL)
2175 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2176 compile_stack_elt_t);
2177 if (compile_stack.stack == NULL) return REG_ESPACE;
2179 compile_stack.size <<= 1;
2182 /* These are the values to restore when we hit end of this
2183 group. They are all relative offsets, so that if the
2184 whole pattern moves because of realloc, they will still
2186 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2187 COMPILE_STACK_TOP.fixup_alt_jump
2188 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2189 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2190 COMPILE_STACK_TOP.regnum = regnum;
2192 /* We will eventually replace the 0 with the number of
2193 groups inner to this one. But do not push a
2194 start_memory for groups beyond the last one we can
2195 represent in the compiled pattern. */
2196 if (regnum <= MAX_REGNUM)
2198 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2199 BUF_PUSH_3 (start_memory, regnum, 0);
2202 compile_stack.avail++;
2207 /* If we've reached MAX_REGNUM groups, then this open
2208 won't actually generate any code, so we'll have to
2209 clear pending_exact explicitly. */
2215 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2217 if (COMPILE_STACK_EMPTY)
2218 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2219 goto normal_backslash;
2221 FREE_STACK_RETURN (REG_ERPAREN);
2225 { /* Push a dummy failure point at the end of the
2226 alternative for a possible future
2227 `pop_failure_jump' to pop. See comments at
2228 `push_dummy_failure' in `re_match_2'. */
2229 BUF_PUSH (push_dummy_failure);
2231 /* We allocated space for this jump when we assigned
2232 to `fixup_alt_jump', in the `handle_alt' case below. */
2233 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2236 /* See similar code for backslashed left paren above. */
2237 if (COMPILE_STACK_EMPTY)
2238 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2241 FREE_STACK_RETURN (REG_ERPAREN);
2243 /* Since we just checked for an empty stack above, this
2244 ``can't happen''. */
2245 assert (compile_stack.avail != 0);
2247 /* We don't just want to restore into `regnum', because
2248 later groups should continue to be numbered higher,
2249 as in `(ab)c(de)' -- the second group is #2. */
2250 regnum_t this_group_regnum;
2252 compile_stack.avail--;
2253 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2255 = COMPILE_STACK_TOP.fixup_alt_jump
2256 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2258 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2259 this_group_regnum = COMPILE_STACK_TOP.regnum;
2260 /* If we've reached MAX_REGNUM groups, then this open
2261 won't actually generate any code, so we'll have to
2262 clear pending_exact explicitly. */
2265 /* We're at the end of the group, so now we know how many
2266 groups were inside this one. */
2267 if (this_group_regnum <= MAX_REGNUM)
2269 unsigned char *inner_group_loc
2270 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2272 *inner_group_loc = regnum - this_group_regnum;
2273 BUF_PUSH_3 (stop_memory, this_group_regnum,
2274 regnum - this_group_regnum);
2280 case '|': /* `\|'. */
2281 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2282 goto normal_backslash;
2284 if (syntax & RE_LIMITED_OPS)
2287 /* Insert before the previous alternative a jump which
2288 jumps to this alternative if the former fails. */
2289 GET_BUFFER_SPACE (3);
2290 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2294 /* The alternative before this one has a jump after it
2295 which gets executed if it gets matched. Adjust that
2296 jump so it will jump to this alternative's analogous
2297 jump (put in below, which in turn will jump to the next
2298 (if any) alternative's such jump, etc.). The last such
2299 jump jumps to the correct final destination. A picture:
2305 If we are at `b', then fixup_alt_jump right now points to a
2306 three-byte space after `a'. We'll put in the jump, set
2307 fixup_alt_jump to right after `b', and leave behind three
2308 bytes which we'll fill in when we get to after `c'. */
2311 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2313 /* Mark and leave space for a jump after this alternative,
2314 to be filled in later either by next alternative or
2315 when know we're at the end of a series of alternatives. */
2317 GET_BUFFER_SPACE (3);
2326 /* If \{ is a literal. */
2327 if (!(syntax & RE_INTERVALS)
2328 /* If we're at `\{' and it's not the open-interval
2330 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2331 || (p - 2 == pattern && p == pend))
2332 goto normal_backslash;
2336 /* If got here, then the syntax allows intervals. */
2338 /* At least (most) this many matches must be made. */
2339 int lower_bound = -1, upper_bound = -1;
2341 beg_interval = p - 1;
2345 if (syntax & RE_NO_BK_BRACES)
2346 goto unfetch_interval;
2348 FREE_STACK_RETURN (REG_EBRACE);
2351 GET_UNSIGNED_NUMBER (lower_bound);
2355 GET_UNSIGNED_NUMBER (upper_bound);
2356 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2359 /* Interval such as `{1}' => match exactly once. */
2360 upper_bound = lower_bound;
2362 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2363 || lower_bound > upper_bound)
2365 if (syntax & RE_NO_BK_BRACES)
2366 goto unfetch_interval;
2368 FREE_STACK_RETURN (REG_BADBR);
2371 if (!(syntax & RE_NO_BK_BRACES))
2373 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2380 if (syntax & RE_NO_BK_BRACES)
2381 goto unfetch_interval;
2383 FREE_STACK_RETURN (REG_BADBR);
2386 /* We just parsed a valid interval. */
2388 /* If it's invalid to have no preceding re. */
2391 if (syntax & RE_CONTEXT_INVALID_OPS)
2392 FREE_STACK_RETURN (REG_BADRPT);
2393 else if (syntax & RE_CONTEXT_INDEP_OPS)
2396 goto unfetch_interval;
2399 /* If the upper bound is zero, don't want to succeed at
2400 all; jump from `laststart' to `b + 3', which will be
2401 the end of the buffer after we insert the jump. */
2402 if (upper_bound == 0)
2404 GET_BUFFER_SPACE (3);
2405 INSERT_JUMP (jump, laststart, b + 3);
2409 /* Otherwise, we have a nontrivial interval. When
2410 we're all done, the pattern will look like:
2411 set_number_at <jump count> <upper bound>
2412 set_number_at <succeed_n count> <lower bound>
2413 succeed_n <after jump addr> <succeed_n count>
2415 jump_n <succeed_n addr> <jump count>
2416 (The upper bound and `jump_n' are omitted if
2417 `upper_bound' is 1, though.) */
2419 { /* If the upper bound is > 1, we need to insert
2420 more at the end of the loop. */
2421 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2423 GET_BUFFER_SPACE (nbytes);
2425 /* Initialize lower bound of the `succeed_n', even
2426 though it will be set during matching by its
2427 attendant `set_number_at' (inserted next),
2428 because `re_compile_fastmap' needs to know.
2429 Jump to the `jump_n' we might insert below. */
2430 INSERT_JUMP2 (succeed_n, laststart,
2431 b + 5 + (upper_bound > 1) * 5,
2435 /* Code to initialize the lower bound. Insert
2436 before the `succeed_n'. The `5' is the last two
2437 bytes of this `set_number_at', plus 3 bytes of
2438 the following `succeed_n'. */
2439 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2442 if (upper_bound > 1)
2443 { /* More than one repetition is allowed, so
2444 append a backward jump to the `succeed_n'
2445 that starts this interval.
2447 When we've reached this during matching,
2448 we'll have matched the interval once, so
2449 jump back only `upper_bound - 1' times. */
2450 STORE_JUMP2 (jump_n, b, laststart + 5,
2454 /* The location we want to set is the second
2455 parameter of the `jump_n'; that is `b-2' as
2456 an absolute address. `laststart' will be
2457 the `set_number_at' we're about to insert;
2458 `laststart+3' the number to set, the source
2459 for the relative address. But we are
2460 inserting into the middle of the pattern --
2461 so everything is getting moved up by 5.
2462 Conclusion: (b - 2) - (laststart + 3) + 5,
2463 i.e., b - laststart.
2465 We insert this at the beginning of the loop
2466 so that if we fail during matching, we'll
2467 reinitialize the bounds. */
2468 insert_op2 (set_number_at, laststart, b - laststart,
2469 upper_bound - 1, b);
2474 beg_interval = NULL;
2479 /* If an invalid interval, match the characters as literals. */
2480 assert (beg_interval);
2482 beg_interval = NULL;
2484 /* normal_char and normal_backslash need `c'. */
2487 if (!(syntax & RE_NO_BK_BRACES))
2489 if (p > pattern && p[-1] == '\\')
2490 goto normal_backslash;
2495 /* There is no way to specify the before_dot and after_dot
2496 operators. rms says this is ok. --karl */
2504 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2510 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2517 BUF_PUSH (wordchar);
2523 BUF_PUSH (notwordchar);
2536 BUF_PUSH (wordbound);
2540 BUF_PUSH (notwordbound);
2551 case '1': case '2': case '3': case '4': case '5':
2552 case '6': case '7': case '8': case '9':
2553 if (syntax & RE_NO_BK_REFS)
2559 FREE_STACK_RETURN (REG_ESUBREG);
2561 /* Can't back reference to a subexpression if inside of it. */
2562 if (group_in_compile_stack (compile_stack, c1))
2566 BUF_PUSH_2 (duplicate, c1);
2572 if (syntax & RE_BK_PLUS_QM)
2575 goto normal_backslash;
2579 /* You might think it would be useful for \ to mean
2580 not to translate; but if we don't translate it
2581 it will never match anything. */
2589 /* Expects the character in `c'. */
2591 /* If no exactn currently being built. */
2594 /* If last exactn not at current position. */
2595 || pending_exact + *pending_exact + 1 != b
2597 /* We have only one byte following the exactn for the count. */
2598 || *pending_exact == (1 << BYTEWIDTH) - 1
2600 /* If followed by a repetition operator. */
2601 || *p == '*' || *p == '^'
2602 || ((syntax & RE_BK_PLUS_QM)
2603 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2604 : (*p == '+' || *p == '?'))
2605 || ((syntax & RE_INTERVALS)
2606 && ((syntax & RE_NO_BK_BRACES)
2608 : (p[0] == '\\' && p[1] == '{'))))
2610 /* Start building a new exactn. */
2614 BUF_PUSH_2 (exactn, 0);
2615 pending_exact = b - 1;
2622 } /* while p != pend */
2625 /* Through the pattern now. */
2628 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2630 if (!COMPILE_STACK_EMPTY)
2631 FREE_STACK_RETURN (REG_EPAREN);
2633 /* If we don't want backtracking, force success
2634 the first time we reach the end of the compiled pattern. */
2635 if (syntax & RE_NO_POSIX_BACKTRACKING)
2638 free (compile_stack.stack);
2640 /* We have succeeded; set the length of the buffer. */
2641 bufp->used = b - bufp->buffer;
2646 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2647 print_compiled_pattern (bufp);
2651 #ifndef MATCH_MAY_ALLOCATE
2652 /* Initialize the failure stack to the largest possible stack. This
2653 isn't necessary unless we're trying to avoid calling alloca in
2654 the search and match routines. */
2656 int num_regs = bufp->re_nsub + 1;
2658 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2659 is strictly greater than re_max_failures, the largest possible stack
2660 is 2 * re_max_failures failure points. */
2661 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2663 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2666 if (! fail_stack.stack)
2668 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2669 * sizeof (fail_stack_elt_t));
2672 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2674 * sizeof (fail_stack_elt_t)));
2675 #else /* not emacs */
2676 if (! fail_stack.stack)
2678 = (fail_stack_elt_t *) malloc (fail_stack.size
2679 * sizeof (fail_stack_elt_t));
2682 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2684 * sizeof (fail_stack_elt_t)));
2685 #endif /* not emacs */
2688 regex_grow_registers (num_regs);
2690 #endif /* not MATCH_MAY_ALLOCATE */
2693 } /* regex_compile */
2695 /* Subroutines for `regex_compile'. */
2697 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2700 store_op1 (op, loc, arg)
2705 *loc = (unsigned char) op;
2706 STORE_NUMBER (loc + 1, arg);
2710 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2713 store_op2 (op, loc, arg1, arg2)
2718 *loc = (unsigned char) op;
2719 STORE_NUMBER (loc + 1, arg1);
2720 STORE_NUMBER (loc + 3, arg2);
2724 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2725 for OP followed by two-byte integer parameter ARG. */
2728 insert_op1 (op, loc, arg, end)
2734 register unsigned char *pfrom = end;
2735 register unsigned char *pto = end + 3;
2737 while (pfrom != loc)
2740 store_op1 (op, loc, arg);
2744 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2747 insert_op2 (op, loc, arg1, arg2, end)
2753 register unsigned char *pfrom = end;
2754 register unsigned char *pto = end + 5;
2756 while (pfrom != loc)
2759 store_op2 (op, loc, arg1, arg2);
2763 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2764 after an alternative or a begin-subexpression. We assume there is at
2765 least one character before the ^. */
2768 at_begline_loc_p (pattern, p, syntax)
2769 const char *pattern, *p;
2770 reg_syntax_t syntax;
2772 const char *prev = p - 2;
2773 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2776 /* After a subexpression? */
2777 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2778 /* After an alternative? */
2779 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2783 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2784 at least one character after the $, i.e., `P < PEND'. */
2787 at_endline_loc_p (p, pend, syntax)
2788 const char *p, *pend;
2791 const char *next = p;
2792 boolean next_backslash = *next == '\\';
2793 const char *next_next = p + 1 < pend ? p + 1 : 0;
2796 /* Before a subexpression? */
2797 (syntax & RE_NO_BK_PARENS ? *next == ')'
2798 : next_backslash && next_next && *next_next == ')')
2799 /* Before an alternative? */
2800 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2801 : next_backslash && next_next && *next_next == '|');
2805 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2806 false if it's not. */
2809 group_in_compile_stack (compile_stack, regnum)
2810 compile_stack_type compile_stack;
2815 for (this_element = compile_stack.avail - 1;
2818 if (compile_stack.stack[this_element].regnum == regnum)
2825 /* Read the ending character of a range (in a bracket expression) from the
2826 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2827 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2828 Then we set the translation of all bits between the starting and
2829 ending characters (inclusive) in the compiled pattern B.
2831 Return an error code.
2833 We use these short variable names so we can use the same macros as
2834 `regex_compile' itself. */
2836 static reg_errcode_t
2837 compile_range (p_ptr, pend, translate, syntax, b)
2838 const char **p_ptr, *pend;
2839 RE_TRANSLATE_TYPE translate;
2840 reg_syntax_t syntax;
2845 const char *p = *p_ptr;
2846 int range_start, range_end;
2851 /* Even though the pattern is a signed `char *', we need to fetch
2852 with unsigned char *'s; if the high bit of the pattern character
2853 is set, the range endpoints will be negative if we fetch using a
2856 We also want to fetch the endpoints without translating them; the
2857 appropriate translation is done in the bit-setting loop below. */
2858 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2859 range_start = ((const unsigned char *) p)[-2];
2860 range_end = ((const unsigned char *) p)[0];
2862 /* Have to increment the pointer into the pattern string, so the
2863 caller isn't still at the ending character. */
2866 /* If the start is after the end, the range is empty. */
2867 if (range_start > range_end)
2868 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2870 /* Here we see why `this_char' has to be larger than an `unsigned
2871 char' -- the range is inclusive, so if `range_end' == 0xff
2872 (assuming 8-bit characters), we would otherwise go into an infinite
2873 loop, since all characters <= 0xff. */
2874 for (this_char = range_start; this_char <= range_end; this_char++)
2876 SET_LIST_BIT (TRANSLATE (this_char));
2882 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2883 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2884 characters can start a string that matches the pattern. This fastmap
2885 is used by re_search to skip quickly over impossible starting points.
2887 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2888 area as BUFP->fastmap.
2890 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2893 Returns 0 if we succeed, -2 if an internal error. */
2896 re_compile_fastmap (bufp)
2897 struct re_pattern_buffer *bufp;
2900 #ifdef MATCH_MAY_ALLOCATE
2901 fail_stack_type fail_stack;
2903 #ifndef REGEX_MALLOC
2906 /* We don't push any register information onto the failure stack. */
2907 unsigned num_regs = 0;
2909 register char *fastmap = bufp->fastmap;
2910 unsigned char *pattern = bufp->buffer;
2911 unsigned long size = bufp->used;
2912 unsigned char *p = pattern;
2913 register unsigned char *pend = pattern + size;
2915 /* This holds the pointer to the failure stack, when
2916 it is allocated relocatably. */
2917 fail_stack_elt_t *failure_stack_ptr;
2919 /* Assume that each path through the pattern can be null until
2920 proven otherwise. We set this false at the bottom of switch
2921 statement, to which we get only if a particular path doesn't
2922 match the empty string. */
2923 boolean path_can_be_null = true;
2925 /* We aren't doing a `succeed_n' to begin with. */
2926 boolean succeed_n_p = false;
2928 assert (fastmap != NULL && p != NULL);
2931 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2932 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2933 bufp->can_be_null = 0;
2937 if (p == pend || *p == succeed)
2939 /* We have reached the (effective) end of pattern. */
2940 if (!FAIL_STACK_EMPTY ())
2942 bufp->can_be_null |= path_can_be_null;
2944 /* Reset for next path. */
2945 path_can_be_null = true;
2947 p = fail_stack.stack[--fail_stack.avail].pointer;
2955 /* We should never be about to go beyond the end of the pattern. */
2958 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2961 /* I guess the idea here is to simply not bother with a fastmap
2962 if a backreference is used, since it's too hard to figure out
2963 the fastmap for the corresponding group. Setting
2964 `can_be_null' stops `re_search_2' from using the fastmap, so
2965 that is all we do. */
2967 bufp->can_be_null = 1;
2971 /* Following are the cases which match a character. These end
2980 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2981 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2987 /* Chars beyond end of map must be allowed. */
2988 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2991 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2992 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2998 for (j = 0; j < (1 << BYTEWIDTH); j++)
2999 if (SYNTAX (j) == Sword)
3005 for (j = 0; j < (1 << BYTEWIDTH); j++)
3006 if (SYNTAX (j) != Sword)
3013 int fastmap_newline = fastmap['\n'];
3015 /* `.' matches anything ... */
3016 for (j = 0; j < (1 << BYTEWIDTH); j++)
3019 /* ... except perhaps newline. */
3020 if (!(bufp->syntax & RE_DOT_NEWLINE))
3021 fastmap['\n'] = fastmap_newline;
3023 /* Return if we have already set `can_be_null'; if we have,
3024 then the fastmap is irrelevant. Something's wrong here. */
3025 else if (bufp->can_be_null)
3028 /* Otherwise, have to check alternative paths. */
3035 for (j = 0; j < (1 << BYTEWIDTH); j++)
3036 if (SYNTAX (j) == (enum syntaxcode) k)
3043 for (j = 0; j < (1 << BYTEWIDTH); j++)
3044 if (SYNTAX (j) != (enum syntaxcode) k)
3049 /* All cases after this match the empty string. These end with
3069 case push_dummy_failure:
3074 case pop_failure_jump:
3075 case maybe_pop_jump:
3078 case dummy_failure_jump:
3079 EXTRACT_NUMBER_AND_INCR (j, p);
3084 /* Jump backward implies we just went through the body of a
3085 loop and matched nothing. Opcode jumped to should be
3086 `on_failure_jump' or `succeed_n'. Just treat it like an
3087 ordinary jump. For a * loop, it has pushed its failure
3088 point already; if so, discard that as redundant. */
3089 if ((re_opcode_t) *p != on_failure_jump
3090 && (re_opcode_t) *p != succeed_n)
3094 EXTRACT_NUMBER_AND_INCR (j, p);
3097 /* If what's on the stack is where we are now, pop it. */
3098 if (!FAIL_STACK_EMPTY ()
3099 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3105 case on_failure_jump:
3106 case on_failure_keep_string_jump:
3107 handle_on_failure_jump:
3108 EXTRACT_NUMBER_AND_INCR (j, p);
3110 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3111 end of the pattern. We don't want to push such a point,
3112 since when we restore it above, entering the switch will
3113 increment `p' past the end of the pattern. We don't need
3114 to push such a point since we obviously won't find any more
3115 fastmap entries beyond `pend'. Such a pattern can match
3116 the null string, though. */
3119 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3121 RESET_FAIL_STACK ();
3126 bufp->can_be_null = 1;
3130 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3131 succeed_n_p = false;
3138 /* Get to the number of times to succeed. */
3141 /* Increment p past the n for when k != 0. */
3142 EXTRACT_NUMBER_AND_INCR (k, p);
3146 succeed_n_p = true; /* Spaghetti code alert. */
3147 goto handle_on_failure_jump;
3164 abort (); /* We have listed all the cases. */
3167 /* Getting here means we have found the possible starting
3168 characters for one path of the pattern -- and that the empty
3169 string does not match. We need not follow this path further.
3170 Instead, look at the next alternative (remembered on the
3171 stack), or quit if no more. The test at the top of the loop
3172 does these things. */
3173 path_can_be_null = false;
3177 /* Set `can_be_null' for the last path (also the first path, if the
3178 pattern is empty). */
3179 bufp->can_be_null |= path_can_be_null;
3182 RESET_FAIL_STACK ();
3184 } /* re_compile_fastmap */
3186 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3187 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3188 this memory for recording register information. STARTS and ENDS
3189 must be allocated using the malloc library routine, and must each
3190 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3192 If NUM_REGS == 0, then subsequent matches should allocate their own
3195 Unless this function is called, the first search or match using
3196 PATTERN_BUFFER will allocate its own register data, without
3197 freeing the old data. */
3200 re_set_registers (bufp, regs, num_regs, starts, ends)
3201 struct re_pattern_buffer *bufp;
3202 struct re_registers *regs;
3204 regoff_t *starts, *ends;
3208 bufp->regs_allocated = REGS_REALLOCATE;
3209 regs->num_regs = num_regs;
3210 regs->start = starts;
3215 bufp->regs_allocated = REGS_UNALLOCATED;
3217 regs->start = regs->end = (regoff_t *) 0;
3221 /* Searching routines. */
3223 /* Like re_search_2, below, but only one string is specified, and
3224 doesn't let you say where to stop matching. */
3227 re_search (bufp, string, size, startpos, range, regs)
3228 struct re_pattern_buffer *bufp;
3230 int size, startpos, range;
3231 struct re_registers *regs;
3233 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3238 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3239 virtual concatenation of STRING1 and STRING2, starting first at index
3240 STARTPOS, then at STARTPOS + 1, and so on.
3242 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3244 RANGE is how far to scan while trying to match. RANGE = 0 means try
3245 only at STARTPOS; in general, the last start tried is STARTPOS +
3248 In REGS, return the indices of the virtual concatenation of STRING1
3249 and STRING2 that matched the entire BUFP->buffer and its contained
3252 Do not consider matching one past the index STOP in the virtual
3253 concatenation of STRING1 and STRING2.
3255 We return either the position in the strings at which the match was
3256 found, -1 if no match, or -2 if error (such as failure
3260 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3261 struct re_pattern_buffer *bufp;
3262 const char *string1, *string2;
3266 struct re_registers *regs;
3270 register char *fastmap = bufp->fastmap;
3271 register RE_TRANSLATE_TYPE translate = bufp->translate;
3272 int total_size = size1 + size2;
3273 int endpos = startpos + range;
3275 /* Check for out-of-range STARTPOS. */
3276 if (startpos < 0 || startpos > total_size)
3279 /* Fix up RANGE if it might eventually take us outside
3280 the virtual concatenation of STRING1 and STRING2.
3281 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
3283 range = 0 - startpos;
3284 else if (endpos > total_size)
3285 range = total_size - startpos;
3287 /* If the search isn't to be a backwards one, don't waste time in a
3288 search for a pattern that must be anchored. */
3289 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3298 /* In a forward search for something that starts with \=.
3299 don't keep searching past point. */
3300 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
3302 range = PT - startpos;
3308 /* Update the fastmap now if not correct already. */
3309 if (fastmap && !bufp->fastmap_accurate)
3310 if (re_compile_fastmap (bufp) == -2)
3313 /* Loop through the string, looking for a place to start matching. */
3316 /* If a fastmap is supplied, skip quickly over characters that
3317 cannot be the start of a match. If the pattern can match the
3318 null string, however, we don't need to skip characters; we want
3319 the first null string. */
3320 if (fastmap && startpos < total_size && !bufp->can_be_null)
3322 if (range > 0) /* Searching forwards. */
3324 register const char *d;
3325 register int lim = 0;
3328 if (startpos < size1 && startpos + range >= size1)
3329 lim = range - (size1 - startpos);
3331 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3333 /* Written out as an if-else to avoid testing `translate'
3337 && !fastmap[(unsigned char)
3338 translate[(unsigned char) *d++]])
3341 while (range > lim && !fastmap[(unsigned char) *d++])
3344 startpos += irange - range;
3346 else /* Searching backwards. */
3348 register char c = (size1 == 0 || startpos >= size1
3349 ? string2[startpos - size1]
3350 : string1[startpos]);
3352 if (!fastmap[(unsigned char) TRANSLATE (c)])
3357 /* If can't match the null string, and that's all we have left, fail. */
3358 if (range >= 0 && startpos == total_size && fastmap
3359 && !bufp->can_be_null)
3362 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3363 startpos, regs, stop);
3364 #ifndef REGEX_MALLOC
3393 /* Declarations and macros for re_match_2. */
3395 static int bcmp_translate ();
3396 static boolean alt_match_null_string_p (),
3397 common_op_match_null_string_p (),
3398 group_match_null_string_p ();
3400 /* This converts PTR, a pointer into one of the search strings `string1'
3401 and `string2' into an offset from the beginning of that string. */
3402 #define POINTER_TO_OFFSET(ptr) \
3403 (FIRST_STRING_P (ptr) \
3404 ? ((regoff_t) ((ptr) - string1)) \
3405 : ((regoff_t) ((ptr) - string2 + size1)))
3407 /* Macros for dealing with the split strings in re_match_2. */
3409 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3411 /* Call before fetching a character with *d. This switches over to
3412 string2 if necessary. */
3413 #define PREFETCH() \
3416 /* End of string2 => fail. */ \
3417 if (dend == end_match_2) \
3419 /* End of string1 => advance to string2. */ \
3421 dend = end_match_2; \
3425 /* Test if at very beginning or at very end of the virtual concatenation
3426 of `string1' and `string2'. If only one string, it's `string2'. */
3427 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3428 #define AT_STRINGS_END(d) ((d) == end2)
3431 /* Test if D points to a character which is word-constituent. We have
3432 two special cases to check for: if past the end of string1, look at
3433 the first character in string2; and if before the beginning of
3434 string2, look at the last character in string1. */
3435 #define WORDCHAR_P(d) \
3436 (SYNTAX ((d) == end1 ? *string2 \
3437 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3440 /* Test if the character before D and the one at D differ with respect
3441 to being word-constituent. */
3442 #define AT_WORD_BOUNDARY(d) \
3443 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3444 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3447 /* Free everything we malloc. */
3448 #ifdef MATCH_MAY_ALLOCATE
3449 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3450 #define FREE_VARIABLES() \
3452 REGEX_FREE_STACK (fail_stack.stack); \
3453 FREE_VAR (regstart); \
3454 FREE_VAR (regend); \
3455 FREE_VAR (old_regstart); \
3456 FREE_VAR (old_regend); \
3457 FREE_VAR (best_regstart); \
3458 FREE_VAR (best_regend); \
3459 FREE_VAR (reg_info); \
3460 FREE_VAR (reg_dummy); \
3461 FREE_VAR (reg_info_dummy); \
3464 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3465 #endif /* not MATCH_MAY_ALLOCATE */
3467 /* These values must meet several constraints. They must not be valid
3468 register values; since we have a limit of 255 registers (because
3469 we use only one byte in the pattern for the register number), we can
3470 use numbers larger than 255. They must differ by 1, because of
3471 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3472 be larger than the value for the highest register, so we do not try
3473 to actually save any registers when none are active. */
3474 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3475 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3477 /* Matching routines. */
3479 #ifndef emacs /* Emacs never uses this. */
3480 /* re_match is like re_match_2 except it takes only a single string. */
3483 re_match (bufp, string, size, pos, regs)
3484 struct re_pattern_buffer *bufp;
3487 struct re_registers *regs;
3489 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3494 #endif /* not emacs */
3497 /* re_match_2 matches the compiled pattern in BUFP against the
3498 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3499 and SIZE2, respectively). We start matching at POS, and stop
3502 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3503 store offsets for the substring each group matched in REGS. See the
3504 documentation for exactly how many groups we fill.
3506 We return -1 if no match, -2 if an internal error (such as the
3507 failure stack overflowing). Otherwise, we return the length of the
3508 matched substring. */
3511 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3512 struct re_pattern_buffer *bufp;
3513 const char *string1, *string2;
3516 struct re_registers *regs;
3519 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3525 /* This is a separate function so that we can force an alloca cleanup
3528 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3529 struct re_pattern_buffer *bufp;
3530 const char *string1, *string2;
3533 struct re_registers *regs;
3536 /* General temporaries. */
3540 /* Just past the end of the corresponding string. */
3541 const char *end1, *end2;
3543 /* Pointers into string1 and string2, just past the last characters in
3544 each to consider matching. */
3545 const char *end_match_1, *end_match_2;
3547 /* Where we are in the data, and the end of the current string. */
3548 const char *d, *dend;
3550 /* Where we are in the pattern, and the end of the pattern. */
3551 unsigned char *p = bufp->buffer;
3552 register unsigned char *pend = p + bufp->used;
3554 /* Mark the opcode just after a start_memory, so we can test for an
3555 empty subpattern when we get to the stop_memory. */
3556 unsigned char *just_past_start_mem = 0;
3558 /* We use this to map every character in the string. */
3559 RE_TRANSLATE_TYPE translate = bufp->translate;
3561 /* Failure point stack. Each place that can handle a failure further
3562 down the line pushes a failure point on this stack. It consists of
3563 restart, regend, and reg_info for all registers corresponding to
3564 the subexpressions we're currently inside, plus the number of such
3565 registers, and, finally, two char *'s. The first char * is where
3566 to resume scanning the pattern; the second one is where to resume
3567 scanning the strings. If the latter is zero, the failure point is
3568 a ``dummy''; if a failure happens and the failure point is a dummy,
3569 it gets discarded and the next next one is tried. */
3570 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3571 fail_stack_type fail_stack;
3574 static unsigned failure_id = 0;
3575 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3578 /* This holds the pointer to the failure stack, when
3579 it is allocated relocatably. */
3580 fail_stack_elt_t *failure_stack_ptr;
3582 /* We fill all the registers internally, independent of what we
3583 return, for use in backreferences. The number here includes
3584 an element for register zero. */
3585 unsigned num_regs = bufp->re_nsub + 1;
3587 /* The currently active registers. */
3588 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3589 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3591 /* Information on the contents of registers. These are pointers into
3592 the input strings; they record just what was matched (on this
3593 attempt) by a subexpression part of the pattern, that is, the
3594 regnum-th regstart pointer points to where in the pattern we began
3595 matching and the regnum-th regend points to right after where we
3596 stopped matching the regnum-th subexpression. (The zeroth register
3597 keeps track of what the whole pattern matches.) */
3598 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3599 const char **regstart, **regend;
3602 /* If a group that's operated upon by a repetition operator fails to
3603 match anything, then the register for its start will need to be
3604 restored because it will have been set to wherever in the string we
3605 are when we last see its open-group operator. Similarly for a
3607 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3608 const char **old_regstart, **old_regend;
3611 /* The is_active field of reg_info helps us keep track of which (possibly
3612 nested) subexpressions we are currently in. The matched_something
3613 field of reg_info[reg_num] helps us tell whether or not we have
3614 matched any of the pattern so far this time through the reg_num-th
3615 subexpression. These two fields get reset each time through any
3616 loop their register is in. */
3617 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3618 register_info_type *reg_info;
3621 /* The following record the register info as found in the above
3622 variables when we find a match better than any we've seen before.
3623 This happens as we backtrack through the failure points, which in
3624 turn happens only if we have not yet matched the entire string. */
3625 unsigned best_regs_set = false;
3626 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3627 const char **best_regstart, **best_regend;
3630 /* Logically, this is `best_regend[0]'. But we don't want to have to
3631 allocate space for that if we're not allocating space for anything
3632 else (see below). Also, we never need info about register 0 for
3633 any of the other register vectors, and it seems rather a kludge to
3634 treat `best_regend' differently than the rest. So we keep track of
3635 the end of the best match so far in a separate variable. We
3636 initialize this to NULL so that when we backtrack the first time
3637 and need to test it, it's not garbage. */
3638 const char *match_end = NULL;
3640 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3641 int set_regs_matched_done = 0;
3643 /* Used when we pop values we don't care about. */
3644 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3645 const char **reg_dummy;
3646 register_info_type *reg_info_dummy;
3650 /* Counts the total number of registers pushed. */
3651 unsigned num_regs_pushed = 0;
3654 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3658 #ifdef MATCH_MAY_ALLOCATE
3659 /* Do not bother to initialize all the register variables if there are
3660 no groups in the pattern, as it takes a fair amount of time. If
3661 there are groups, we include space for register 0 (the whole
3662 pattern), even though we never use it, since it simplifies the
3663 array indexing. We should fix this. */
3666 regstart = REGEX_TALLOC (num_regs, const char *);
3667 regend = REGEX_TALLOC (num_regs, const char *);
3668 old_regstart = REGEX_TALLOC (num_regs, const char *);
3669 old_regend = REGEX_TALLOC (num_regs, const char *);
3670 best_regstart = REGEX_TALLOC (num_regs, const char *);
3671 best_regend = REGEX_TALLOC (num_regs, const char *);
3672 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3673 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3674 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3676 if (!(regstart && regend && old_regstart && old_regend && reg_info
3677 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3685 /* We must initialize all our variables to NULL, so that
3686 `FREE_VARIABLES' doesn't try to free them. */
3687 regstart = regend = old_regstart = old_regend = best_regstart
3688 = best_regend = reg_dummy = NULL;
3689 reg_info = reg_info_dummy = (register_info_type *) NULL;
3691 #endif /* MATCH_MAY_ALLOCATE */
3693 /* The starting position is bogus. */
3694 if (pos < 0 || pos > size1 + size2)
3700 /* Initialize subexpression text positions to -1 to mark ones that no
3701 start_memory/stop_memory has been seen for. Also initialize the
3702 register information struct. */
3703 for (mcnt = 1; mcnt < num_regs; mcnt++)
3705 regstart[mcnt] = regend[mcnt]
3706 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3708 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3709 IS_ACTIVE (reg_info[mcnt]) = 0;
3710 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3711 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3714 /* We move `string1' into `string2' if the latter's empty -- but not if
3715 `string1' is null. */
3716 if (size2 == 0 && string1 != NULL)
3723 end1 = string1 + size1;
3724 end2 = string2 + size2;
3726 /* Compute where to stop matching, within the two strings. */
3729 end_match_1 = string1 + stop;
3730 end_match_2 = string2;
3735 end_match_2 = string2 + stop - size1;
3738 /* `p' scans through the pattern as `d' scans through the data.
3739 `dend' is the end of the input string that `d' points within. `d'
3740 is advanced into the following input string whenever necessary, but
3741 this happens before fetching; therefore, at the beginning of the
3742 loop, `d' can be pointing at the end of a string, but it cannot
3744 if (size1 > 0 && pos <= size1)
3751 d = string2 + pos - size1;
3755 DEBUG_PRINT1 ("The compiled pattern is: ");
3756 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3757 DEBUG_PRINT1 ("The string to match is: `");
3758 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3759 DEBUG_PRINT1 ("'\n");
3761 /* This loops over pattern commands. It exits by returning from the
3762 function if the match is complete, or it drops through if the match
3763 fails at this starting point in the input data. */
3766 DEBUG_PRINT2 ("\n0x%x: ", p);
3769 { /* End of pattern means we might have succeeded. */
3770 DEBUG_PRINT1 ("end of pattern ... ");
3772 /* If we haven't matched the entire string, and we want the
3773 longest match, try backtracking. */
3774 if (d != end_match_2)
3776 /* 1 if this match ends in the same string (string1 or string2)
3777 as the best previous match. */
3778 boolean same_str_p = (FIRST_STRING_P (match_end)
3779 == MATCHING_IN_FIRST_STRING);
3780 /* 1 if this match is the best seen so far. */
3781 boolean best_match_p;
3783 /* AIX compiler got confused when this was combined
3784 with the previous declaration. */
3786 best_match_p = d > match_end;
3788 best_match_p = !MATCHING_IN_FIRST_STRING;
3790 DEBUG_PRINT1 ("backtracking.\n");
3792 if (!FAIL_STACK_EMPTY ())
3793 { /* More failure points to try. */
3795 /* If exceeds best match so far, save it. */
3796 if (!best_regs_set || best_match_p)
3798 best_regs_set = true;
3801 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3803 for (mcnt = 1; mcnt < num_regs; mcnt++)
3805 best_regstart[mcnt] = regstart[mcnt];
3806 best_regend[mcnt] = regend[mcnt];
3812 /* If no failure points, don't restore garbage. And if
3813 last match is real best match, don't restore second
3815 else if (best_regs_set && !best_match_p)
3818 /* Restore best match. It may happen that `dend ==
3819 end_match_1' while the restored d is in string2.
3820 For example, the pattern `x.*y.*z' against the
3821 strings `x-' and `y-z-', if the two strings are
3822 not consecutive in memory. */
3823 DEBUG_PRINT1 ("Restoring best registers.\n");
3826 dend = ((d >= string1 && d <= end1)
3827 ? end_match_1 : end_match_2);
3829 for (mcnt = 1; mcnt < num_regs; mcnt++)
3831 regstart[mcnt] = best_regstart[mcnt];
3832 regend[mcnt] = best_regend[mcnt];
3835 } /* d != end_match_2 */
3838 DEBUG_PRINT1 ("Accepting match.\n");
3840 /* If caller wants register contents data back, do it. */
3841 if (regs && !bufp->no_sub)
3843 /* Have the register data arrays been allocated? */
3844 if (bufp->regs_allocated == REGS_UNALLOCATED)
3845 { /* No. So allocate them with malloc. We need one
3846 extra element beyond `num_regs' for the `-1' marker
3848 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3849 regs->start = TALLOC (regs->num_regs, regoff_t);
3850 regs->end = TALLOC (regs->num_regs, regoff_t);
3851 if (regs->start == NULL || regs->end == NULL)
3856 bufp->regs_allocated = REGS_REALLOCATE;
3858 else if (bufp->regs_allocated == REGS_REALLOCATE)
3859 { /* Yes. If we need more elements than were already
3860 allocated, reallocate them. If we need fewer, just
3862 if (regs->num_regs < num_regs + 1)
3864 regs->num_regs = num_regs + 1;
3865 RETALLOC (regs->start, regs->num_regs, regoff_t);
3866 RETALLOC (regs->end, regs->num_regs, regoff_t);
3867 if (regs->start == NULL || regs->end == NULL)
3876 /* These braces fend off a "empty body in an else-statement"
3877 warning under GCC when assert expands to nothing. */
3878 assert (bufp->regs_allocated == REGS_FIXED);
3881 /* Convert the pointer data in `regstart' and `regend' to
3882 indices. Register zero has to be set differently,
3883 since we haven't kept track of any info for it. */
3884 if (regs->num_regs > 0)
3886 regs->start[0] = pos;
3887 regs->end[0] = (MATCHING_IN_FIRST_STRING
3888 ? ((regoff_t) (d - string1))
3889 : ((regoff_t) (d - string2 + size1)));
3892 /* Go through the first `min (num_regs, regs->num_regs)'
3893 registers, since that is all we initialized. */
3894 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3896 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3897 regs->start[mcnt] = regs->end[mcnt] = -1;
3901 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3903 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3907 /* If the regs structure we return has more elements than
3908 were in the pattern, set the extra elements to -1. If
3909 we (re)allocated the registers, this is the case,
3910 because we always allocate enough to have at least one
3912 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3913 regs->start[mcnt] = regs->end[mcnt] = -1;
3914 } /* regs && !bufp->no_sub */
3916 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3917 nfailure_points_pushed, nfailure_points_popped,
3918 nfailure_points_pushed - nfailure_points_popped);
3919 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3921 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3925 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3931 /* Otherwise match next pattern command. */
3932 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3934 /* Ignore these. Used to ignore the n of succeed_n's which
3935 currently have n == 0. */
3937 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3941 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3944 /* Match the next n pattern characters exactly. The following
3945 byte in the pattern defines n, and the n bytes after that
3946 are the characters to match. */
3949 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3951 /* This is written out as an if-else so we don't waste time
3952 testing `translate' inside the loop. */
3958 if ((unsigned char) translate[(unsigned char) *d++]
3959 != (unsigned char) *p++)
3969 if (*d++ != (char) *p++) goto fail;
3973 SET_REGS_MATCHED ();
3977 /* Match any character except possibly a newline or a null. */
3979 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3983 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3984 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3987 SET_REGS_MATCHED ();
3988 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3996 register unsigned char c;
3997 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3999 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
4002 c = TRANSLATE (*d); /* The character to match. */
4004 /* Cast to `unsigned' instead of `unsigned char' in case the
4005 bit list is a full 32 bytes long. */
4006 if (c < (unsigned) (*p * BYTEWIDTH)
4007 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4012 if (!not) goto fail;
4014 SET_REGS_MATCHED ();
4020 /* The beginning of a group is represented by start_memory.
4021 The arguments are the register number in the next byte, and the
4022 number of groups inner to this one in the next. The text
4023 matched within the group is recorded (in the internal
4024 registers data structure) under the register number. */
4026 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4028 /* Find out if this group can match the empty string. */
4029 p1 = p; /* To send to group_match_null_string_p. */
4031 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4032 REG_MATCH_NULL_STRING_P (reg_info[*p])
4033 = group_match_null_string_p (&p1, pend, reg_info);
4035 /* Save the position in the string where we were the last time
4036 we were at this open-group operator in case the group is
4037 operated upon by a repetition operator, e.g., with `(a*)*b'
4038 against `ab'; then we want to ignore where we are now in
4039 the string in case this attempt to match fails. */
4040 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4041 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4043 DEBUG_PRINT2 (" old_regstart: %d\n",
4044 POINTER_TO_OFFSET (old_regstart[*p]));
4047 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4049 IS_ACTIVE (reg_info[*p]) = 1;
4050 MATCHED_SOMETHING (reg_info[*p]) = 0;
4052 /* Clear this whenever we change the register activity status. */
4053 set_regs_matched_done = 0;
4055 /* This is the new highest active register. */
4056 highest_active_reg = *p;
4058 /* If nothing was active before, this is the new lowest active
4060 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4061 lowest_active_reg = *p;
4063 /* Move past the register number and inner group count. */
4065 just_past_start_mem = p;
4070 /* The stop_memory opcode represents the end of a group. Its
4071 arguments are the same as start_memory's: the register
4072 number, and the number of inner groups. */
4074 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4076 /* We need to save the string position the last time we were at
4077 this close-group operator in case the group is operated
4078 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4079 against `aba'; then we want to ignore where we are now in
4080 the string in case this attempt to match fails. */
4081 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4082 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4084 DEBUG_PRINT2 (" old_regend: %d\n",
4085 POINTER_TO_OFFSET (old_regend[*p]));
4088 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4090 /* This register isn't active anymore. */
4091 IS_ACTIVE (reg_info[*p]) = 0;
4093 /* Clear this whenever we change the register activity status. */
4094 set_regs_matched_done = 0;
4096 /* If this was the only register active, nothing is active
4098 if (lowest_active_reg == highest_active_reg)
4100 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4101 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4104 { /* We must scan for the new highest active register, since
4105 it isn't necessarily one less than now: consider
4106 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4107 new highest active register is 1. */
4108 unsigned char r = *p - 1;
4109 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4112 /* If we end up at register zero, that means that we saved
4113 the registers as the result of an `on_failure_jump', not
4114 a `start_memory', and we jumped to past the innermost
4115 `stop_memory'. For example, in ((.)*) we save
4116 registers 1 and 2 as a result of the *, but when we pop
4117 back to the second ), we are at the stop_memory 1.
4118 Thus, nothing is active. */
4121 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4122 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4125 highest_active_reg = r;
4128 /* If just failed to match something this time around with a
4129 group that's operated on by a repetition operator, try to
4130 force exit from the ``loop'', and restore the register
4131 information for this group that we had before trying this
4133 if ((!MATCHED_SOMETHING (reg_info[*p])
4134 || just_past_start_mem == p - 1)
4137 boolean is_a_jump_n = false;
4141 switch ((re_opcode_t) *p1++)
4145 case pop_failure_jump:
4146 case maybe_pop_jump:
4148 case dummy_failure_jump:
4149 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4159 /* If the next operation is a jump backwards in the pattern
4160 to an on_failure_jump right before the start_memory
4161 corresponding to this stop_memory, exit from the loop
4162 by forcing a failure after pushing on the stack the
4163 on_failure_jump's jump in the pattern, and d. */
4164 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4165 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4167 /* If this group ever matched anything, then restore
4168 what its registers were before trying this last
4169 failed match, e.g., with `(a*)*b' against `ab' for
4170 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4171 against `aba' for regend[3].
4173 Also restore the registers for inner groups for,
4174 e.g., `((a*)(b*))*' against `aba' (register 3 would
4175 otherwise get trashed). */
4177 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4181 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4183 /* Restore this and inner groups' (if any) registers. */
4184 for (r = *p; r < *p + *(p + 1); r++)
4186 regstart[r] = old_regstart[r];
4188 /* xx why this test? */
4189 if (old_regend[r] >= regstart[r])
4190 regend[r] = old_regend[r];
4194 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4195 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4201 /* Move past the register number and the inner group count. */
4206 /* \<digit> has been turned into a `duplicate' command which is
4207 followed by the numeric value of <digit> as the register number. */
4210 register const char *d2, *dend2;
4211 int regno = *p++; /* Get which register to match against. */
4212 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4214 /* Can't back reference a group which we've never matched. */
4215 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4218 /* Where in input to try to start matching. */
4219 d2 = regstart[regno];
4221 /* Where to stop matching; if both the place to start and
4222 the place to stop matching are in the same string, then
4223 set to the place to stop, otherwise, for now have to use
4224 the end of the first string. */
4226 dend2 = ((FIRST_STRING_P (regstart[regno])
4227 == FIRST_STRING_P (regend[regno]))
4228 ? regend[regno] : end_match_1);
4231 /* If necessary, advance to next segment in register
4235 if (dend2 == end_match_2) break;
4236 if (dend2 == regend[regno]) break;
4238 /* End of string1 => advance to string2. */
4240 dend2 = regend[regno];
4242 /* At end of register contents => success */
4243 if (d2 == dend2) break;
4245 /* If necessary, advance to next segment in data. */
4248 /* How many characters left in this segment to match. */
4251 /* Want how many consecutive characters we can match in
4252 one shot, so, if necessary, adjust the count. */
4253 if (mcnt > dend2 - d2)
4256 /* Compare that many; failure if mismatch, else move
4259 ? bcmp_translate (d, d2, mcnt, translate)
4260 : bcmp (d, d2, mcnt))
4262 d += mcnt, d2 += mcnt;
4264 /* Do this because we've match some characters. */
4265 SET_REGS_MATCHED ();
4271 /* begline matches the empty string at the beginning of the string
4272 (unless `not_bol' is set in `bufp'), and, if
4273 `newline_anchor' is set, after newlines. */
4275 DEBUG_PRINT1 ("EXECUTING begline.\n");
4277 if (AT_STRINGS_BEG (d))
4279 if (!bufp->not_bol) break;
4281 else if (d[-1] == '\n' && bufp->newline_anchor)
4285 /* In all other cases, we fail. */
4289 /* endline is the dual of begline. */
4291 DEBUG_PRINT1 ("EXECUTING endline.\n");
4293 if (AT_STRINGS_END (d))
4295 if (!bufp->not_eol) break;
4298 /* We have to ``prefetch'' the next character. */
4299 else if ((d == end1 ? *string2 : *d) == '\n'
4300 && bufp->newline_anchor)
4307 /* Match at the very beginning of the data. */
4309 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4310 if (AT_STRINGS_BEG (d))
4315 /* Match at the very end of the data. */
4317 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4318 if (AT_STRINGS_END (d))
4323 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4324 pushes NULL as the value for the string on the stack. Then
4325 `pop_failure_point' will keep the current value for the
4326 string, instead of restoring it. To see why, consider
4327 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4328 then the . fails against the \n. But the next thing we want
4329 to do is match the \n against the \n; if we restored the
4330 string value, we would be back at the foo.
4332 Because this is used only in specific cases, we don't need to
4333 check all the things that `on_failure_jump' does, to make
4334 sure the right things get saved on the stack. Hence we don't
4335 share its code. The only reason to push anything on the
4336 stack at all is that otherwise we would have to change
4337 `anychar's code to do something besides goto fail in this
4338 case; that seems worse than this. */
4339 case on_failure_keep_string_jump:
4340 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4342 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4343 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4345 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4349 /* Uses of on_failure_jump:
4351 Each alternative starts with an on_failure_jump that points
4352 to the beginning of the next alternative. Each alternative
4353 except the last ends with a jump that in effect jumps past
4354 the rest of the alternatives. (They really jump to the
4355 ending jump of the following alternative, because tensioning
4356 these jumps is a hassle.)
4358 Repeats start with an on_failure_jump that points past both
4359 the repetition text and either the following jump or
4360 pop_failure_jump back to this on_failure_jump. */
4361 case on_failure_jump:
4363 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4365 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4366 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4368 /* If this on_failure_jump comes right before a group (i.e.,
4369 the original * applied to a group), save the information
4370 for that group and all inner ones, so that if we fail back
4371 to this point, the group's information will be correct.
4372 For example, in \(a*\)*\1, we need the preceding group,
4373 and in \(zz\(a*\)b*\)\2, we need the inner group. */
4375 /* We can't use `p' to check ahead because we push
4376 a failure point to `p + mcnt' after we do this. */
4379 /* We need to skip no_op's before we look for the
4380 start_memory in case this on_failure_jump is happening as
4381 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4383 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4386 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4388 /* We have a new highest active register now. This will
4389 get reset at the start_memory we are about to get to,
4390 but we will have saved all the registers relevant to
4391 this repetition op, as described above. */
4392 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4393 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4394 lowest_active_reg = *(p1 + 1);
4397 DEBUG_PRINT1 (":\n");
4398 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4402 /* A smart repeat ends with `maybe_pop_jump'.
4403 We change it to either `pop_failure_jump' or `jump'. */
4404 case maybe_pop_jump:
4405 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4406 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4408 register unsigned char *p2 = p;
4410 /* Compare the beginning of the repeat with what in the
4411 pattern follows its end. If we can establish that there
4412 is nothing that they would both match, i.e., that we
4413 would have to backtrack because of (as in, e.g., `a*a')
4414 then we can change to pop_failure_jump, because we'll
4415 never have to backtrack.
4417 This is not true in the case of alternatives: in
4418 `(a|ab)*' we do need to backtrack to the `ab' alternative
4419 (e.g., if the string was `ab'). But instead of trying to
4420 detect that here, the alternative has put on a dummy
4421 failure point which is what we will end up popping. */
4423 /* Skip over open/close-group commands.
4424 If what follows this loop is a ...+ construct,
4425 look at what begins its body, since we will have to
4426 match at least one of that. */
4430 && ((re_opcode_t) *p2 == stop_memory
4431 || (re_opcode_t) *p2 == start_memory))
4433 else if (p2 + 6 < pend
4434 && (re_opcode_t) *p2 == dummy_failure_jump)
4441 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4442 to the `maybe_finalize_jump' of this case. Examine what
4445 /* If we're at the end of the pattern, we can change. */
4448 /* Consider what happens when matching ":\(.*\)"
4449 against ":/". I don't really understand this code
4451 p[-3] = (unsigned char) pop_failure_jump;
4453 (" End of pattern: change to `pop_failure_jump'.\n");
4456 else if ((re_opcode_t) *p2 == exactn
4457 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4459 register unsigned char c
4460 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4462 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4464 p[-3] = (unsigned char) pop_failure_jump;
4465 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4469 else if ((re_opcode_t) p1[3] == charset
4470 || (re_opcode_t) p1[3] == charset_not)
4472 int not = (re_opcode_t) p1[3] == charset_not;
4474 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4475 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4478 /* `not' is equal to 1 if c would match, which means
4479 that we can't change to pop_failure_jump. */
4482 p[-3] = (unsigned char) pop_failure_jump;
4483 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4487 else if ((re_opcode_t) *p2 == charset)
4490 register unsigned char c
4491 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4494 if ((re_opcode_t) p1[3] == exactn
4495 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4496 && (p2[1 + p1[4] / BYTEWIDTH]
4497 & (1 << (p1[4] % BYTEWIDTH)))))
4499 p[-3] = (unsigned char) pop_failure_jump;
4500 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4504 else if ((re_opcode_t) p1[3] == charset_not)
4507 /* We win if the charset_not inside the loop
4508 lists every character listed in the charset after. */
4509 for (idx = 0; idx < (int) p2[1]; idx++)
4510 if (! (p2[2 + idx] == 0
4511 || (idx < (int) p1[4]
4512 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4517 p[-3] = (unsigned char) pop_failure_jump;
4518 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4521 else if ((re_opcode_t) p1[3] == charset)
4524 /* We win if the charset inside the loop
4525 has no overlap with the one after the loop. */
4527 idx < (int) p2[1] && idx < (int) p1[4];
4529 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4532 if (idx == p2[1] || idx == p1[4])
4534 p[-3] = (unsigned char) pop_failure_jump;
4535 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4540 p -= 2; /* Point at relative address again. */
4541 if ((re_opcode_t) p[-1] != pop_failure_jump)
4543 p[-1] = (unsigned char) jump;
4544 DEBUG_PRINT1 (" Match => jump.\n");
4545 goto unconditional_jump;
4547 /* Note fall through. */
4550 /* The end of a simple repeat has a pop_failure_jump back to
4551 its matching on_failure_jump, where the latter will push a
4552 failure point. The pop_failure_jump takes off failure
4553 points put on by this pop_failure_jump's matching
4554 on_failure_jump; we got through the pattern to here from the
4555 matching on_failure_jump, so didn't fail. */
4556 case pop_failure_jump:
4558 /* We need to pass separate storage for the lowest and
4559 highest registers, even though we don't care about the
4560 actual values. Otherwise, we will restore only one
4561 register from the stack, since lowest will == highest in
4562 `pop_failure_point'. */
4563 unsigned dummy_low_reg, dummy_high_reg;
4564 unsigned char *pdummy;
4567 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4568 POP_FAILURE_POINT (sdummy, pdummy,
4569 dummy_low_reg, dummy_high_reg,
4570 reg_dummy, reg_dummy, reg_info_dummy);
4572 /* Note fall through. */
4575 /* Unconditionally jump (without popping any failure points). */
4578 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4579 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4580 p += mcnt; /* Do the jump. */
4581 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4585 /* We need this opcode so we can detect where alternatives end
4586 in `group_match_null_string_p' et al. */
4588 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4589 goto unconditional_jump;
4592 /* Normally, the on_failure_jump pushes a failure point, which
4593 then gets popped at pop_failure_jump. We will end up at
4594 pop_failure_jump, also, and with a pattern of, say, `a+', we
4595 are skipping over the on_failure_jump, so we have to push
4596 something meaningless for pop_failure_jump to pop. */
4597 case dummy_failure_jump:
4598 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4599 /* It doesn't matter what we push for the string here. What
4600 the code at `fail' tests is the value for the pattern. */
4601 PUSH_FAILURE_POINT (0, 0, -2);
4602 goto unconditional_jump;
4605 /* At the end of an alternative, we need to push a dummy failure
4606 point in case we are followed by a `pop_failure_jump', because
4607 we don't want the failure point for the alternative to be
4608 popped. For example, matching `(a|ab)*' against `aab'
4609 requires that we match the `ab' alternative. */
4610 case push_dummy_failure:
4611 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4612 /* See comments just above at `dummy_failure_jump' about the
4614 PUSH_FAILURE_POINT (0, 0, -2);
4617 /* Have to succeed matching what follows at least n times.
4618 After that, handle like `on_failure_jump'. */
4620 EXTRACT_NUMBER (mcnt, p + 2);
4621 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4624 /* Originally, this is how many times we HAVE to succeed. */
4629 STORE_NUMBER_AND_INCR (p, mcnt);
4630 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4634 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4635 p[2] = (unsigned char) no_op;
4636 p[3] = (unsigned char) no_op;
4642 EXTRACT_NUMBER (mcnt, p + 2);
4643 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4645 /* Originally, this is how many times we CAN jump. */
4649 STORE_NUMBER (p + 2, mcnt);
4650 goto unconditional_jump;
4652 /* If don't have to jump any more, skip over the rest of command. */
4659 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4661 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4663 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4664 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4665 STORE_NUMBER (p1, mcnt);
4670 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4671 if (AT_WORD_BOUNDARY (d))
4676 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4677 if (AT_WORD_BOUNDARY (d))
4682 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4683 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4688 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4689 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4690 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4696 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4697 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4702 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4703 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4708 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4709 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4714 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4719 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4723 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4725 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4727 SET_REGS_MATCHED ();
4731 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4733 goto matchnotsyntax;
4736 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4740 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4742 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4744 SET_REGS_MATCHED ();
4747 #else /* not emacs */
4749 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4751 if (!WORDCHAR_P (d))
4753 SET_REGS_MATCHED ();
4758 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4762 SET_REGS_MATCHED ();
4765 #endif /* not emacs */
4770 continue; /* Successfully executed one pattern command; keep going. */
4773 /* We goto here if a matching operation fails. */
4775 if (!FAIL_STACK_EMPTY ())
4776 { /* A restart point is known. Restore to that state. */
4777 DEBUG_PRINT1 ("\nFAIL:\n");
4778 POP_FAILURE_POINT (d, p,
4779 lowest_active_reg, highest_active_reg,
4780 regstart, regend, reg_info);
4782 /* If this failure point is a dummy, try the next one. */
4786 /* If we failed to the end of the pattern, don't examine *p. */
4790 boolean is_a_jump_n = false;
4792 /* If failed to a backwards jump that's part of a repetition
4793 loop, need to pop this failure point and use the next one. */
4794 switch ((re_opcode_t) *p)
4798 case maybe_pop_jump:
4799 case pop_failure_jump:
4802 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4805 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4807 && (re_opcode_t) *p1 == on_failure_jump))
4815 if (d >= string1 && d <= end1)
4819 break; /* Matching at this starting point really fails. */
4823 goto restore_best_regs;
4827 return -1; /* Failure to match. */
4830 /* Subroutine definitions for re_match_2. */
4833 /* We are passed P pointing to a register number after a start_memory.
4835 Return true if the pattern up to the corresponding stop_memory can
4836 match the empty string, and false otherwise.
4838 If we find the matching stop_memory, sets P to point to one past its number.
4839 Otherwise, sets P to an undefined byte less than or equal to END.
4841 We don't handle duplicates properly (yet). */
4844 group_match_null_string_p (p, end, reg_info)
4845 unsigned char **p, *end;
4846 register_info_type *reg_info;
4849 /* Point to after the args to the start_memory. */
4850 unsigned char *p1 = *p + 2;
4854 /* Skip over opcodes that can match nothing, and return true or
4855 false, as appropriate, when we get to one that can't, or to the
4856 matching stop_memory. */
4858 switch ((re_opcode_t) *p1)
4860 /* Could be either a loop or a series of alternatives. */
4861 case on_failure_jump:
4863 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4865 /* If the next operation is not a jump backwards in the
4870 /* Go through the on_failure_jumps of the alternatives,
4871 seeing if any of the alternatives cannot match nothing.
4872 The last alternative starts with only a jump,
4873 whereas the rest start with on_failure_jump and end
4874 with a jump, e.g., here is the pattern for `a|b|c':
4876 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4877 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4880 So, we have to first go through the first (n-1)
4881 alternatives and then deal with the last one separately. */
4884 /* Deal with the first (n-1) alternatives, which start
4885 with an on_failure_jump (see above) that jumps to right
4886 past a jump_past_alt. */
4888 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4890 /* `mcnt' holds how many bytes long the alternative
4891 is, including the ending `jump_past_alt' and
4894 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4898 /* Move to right after this alternative, including the
4902 /* Break if it's the beginning of an n-th alternative
4903 that doesn't begin with an on_failure_jump. */
4904 if ((re_opcode_t) *p1 != on_failure_jump)
4907 /* Still have to check that it's not an n-th
4908 alternative that starts with an on_failure_jump. */
4910 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4911 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4913 /* Get to the beginning of the n-th alternative. */
4919 /* Deal with the last alternative: go back and get number
4920 of the `jump_past_alt' just before it. `mcnt' contains
4921 the length of the alternative. */
4922 EXTRACT_NUMBER (mcnt, p1 - 2);
4924 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4927 p1 += mcnt; /* Get past the n-th alternative. */
4933 assert (p1[1] == **p);
4939 if (!common_op_match_null_string_p (&p1, end, reg_info))
4942 } /* while p1 < end */
4945 } /* group_match_null_string_p */
4948 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4949 It expects P to be the first byte of a single alternative and END one
4950 byte past the last. The alternative can contain groups. */
4953 alt_match_null_string_p (p, end, reg_info)
4954 unsigned char *p, *end;
4955 register_info_type *reg_info;
4958 unsigned char *p1 = p;
4962 /* Skip over opcodes that can match nothing, and break when we get
4963 to one that can't. */
4965 switch ((re_opcode_t) *p1)
4968 case on_failure_jump:
4970 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4975 if (!common_op_match_null_string_p (&p1, end, reg_info))
4978 } /* while p1 < end */
4981 } /* alt_match_null_string_p */
4984 /* Deals with the ops common to group_match_null_string_p and
4985 alt_match_null_string_p.
4987 Sets P to one after the op and its arguments, if any. */
4990 common_op_match_null_string_p (p, end, reg_info)
4991 unsigned char **p, *end;
4992 register_info_type *reg_info;
4997 unsigned char *p1 = *p;
4999 switch ((re_opcode_t) *p1++)
5019 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5020 ret = group_match_null_string_p (&p1, end, reg_info);
5022 /* Have to set this here in case we're checking a group which
5023 contains a group and a back reference to it. */
5025 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5026 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5032 /* If this is an optimized succeed_n for zero times, make the jump. */
5034 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5042 /* Get to the number of times to succeed. */
5044 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5049 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5057 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5065 /* All other opcodes mean we cannot match the empty string. */
5071 } /* common_op_match_null_string_p */
5074 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5075 bytes; nonzero otherwise. */
5078 bcmp_translate (s1, s2, len, translate)
5079 unsigned char *s1, *s2;
5081 RE_TRANSLATE_TYPE translate;
5083 register unsigned char *p1 = s1, *p2 = s2;
5086 if (translate[*p1++] != translate[*p2++]) return 1;
5092 /* Entry points for GNU code. */
5094 /* re_compile_pattern is the GNU regular expression compiler: it
5095 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5096 Returns 0 if the pattern was valid, otherwise an error string.
5098 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5099 are set in BUFP on entry.
5101 We call regex_compile to do the actual compilation. */
5104 re_compile_pattern (pattern, length, bufp)
5105 const char *pattern;
5107 struct re_pattern_buffer *bufp;
5111 /* GNU code is written to assume at least RE_NREGS registers will be set
5112 (and at least one extra will be -1). */
5113 bufp->regs_allocated = REGS_UNALLOCATED;
5115 /* And GNU code determines whether or not to get register information
5116 by passing null for the REGS argument to re_match, etc., not by
5120 /* Match anchors at newline. */
5121 bufp->newline_anchor = 1;
5123 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5127 return gettext (re_error_msgid[(int) ret]);
5130 /* Entry points compatible with 4.2 BSD regex library. We don't define
5131 them unless specifically requested. */
5133 #ifdef _REGEX_RE_COMP
5135 /* BSD has one and only one pattern buffer. */
5136 static struct re_pattern_buffer re_comp_buf;
5146 if (!re_comp_buf.buffer)
5147 return gettext ("No previous regular expression");
5151 if (!re_comp_buf.buffer)
5153 re_comp_buf.buffer = (unsigned char *) malloc (200);
5154 if (re_comp_buf.buffer == NULL)
5155 return gettext (re_error_msgid[(int) REG_ESPACE]);
5156 re_comp_buf.allocated = 200;
5158 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5159 if (re_comp_buf.fastmap == NULL)
5160 return gettext (re_error_msgid[(int) REG_ESPACE]);
5163 /* Since `re_exec' always passes NULL for the `regs' argument, we
5164 don't need to initialize the pattern buffer fields which affect it. */
5166 /* Match anchors at newlines. */
5167 re_comp_buf.newline_anchor = 1;
5169 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5174 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5175 return (char *) gettext (re_error_msgid[(int) ret]);
5183 const int len = strlen (s);
5185 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5187 #endif /* _REGEX_RE_COMP */
5189 /* POSIX.2 functions. Don't define these for Emacs. */
5193 /* regcomp takes a regular expression as a string and compiles it.
5195 PREG is a regex_t *. We do not expect any fields to be initialized,
5196 since POSIX says we shouldn't. Thus, we set
5198 `buffer' to the compiled pattern;
5199 `used' to the length of the compiled pattern;
5200 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5201 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5202 RE_SYNTAX_POSIX_BASIC;
5203 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5204 `fastmap' and `fastmap_accurate' to zero;
5205 `re_nsub' to the number of subexpressions in PATTERN.
5207 PATTERN is the address of the pattern string.
5209 CFLAGS is a series of bits which affect compilation.
5211 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5212 use POSIX basic syntax.
5214 If REG_NEWLINE is set, then . and [^...] don't match newline.
5215 Also, regexec will try a match beginning after every newline.
5217 If REG_ICASE is set, then we considers upper- and lowercase
5218 versions of letters to be equivalent when matching.
5220 If REG_NOSUB is set, then when PREG is passed to regexec, that
5221 routine will report only success or failure, and nothing about the
5224 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5225 the return codes and their meanings.) */
5228 regcomp (preg, pattern, cflags)
5230 const char *pattern;
5235 = (cflags & REG_EXTENDED) ?
5236 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5238 /* regex_compile will allocate the space for the compiled pattern. */
5240 preg->allocated = 0;
5243 /* Don't bother to use a fastmap when searching. This simplifies the
5244 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5245 characters after newlines into the fastmap. This way, we just try
5249 if (cflags & REG_ICASE)
5254 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
5255 * sizeof (*(RE_TRANSLATE_TYPE)0));
5256 if (preg->translate == NULL)
5257 return (int) REG_ESPACE;
5259 /* Map uppercase characters to corresponding lowercase ones. */
5260 for (i = 0; i < CHAR_SET_SIZE; i++)
5261 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5264 preg->translate = NULL;
5266 /* If REG_NEWLINE is set, newlines are treated differently. */
5267 if (cflags & REG_NEWLINE)
5268 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5269 syntax &= ~RE_DOT_NEWLINE;
5270 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5271 /* It also changes the matching behavior. */
5272 preg->newline_anchor = 1;
5275 preg->newline_anchor = 0;
5277 preg->no_sub = !!(cflags & REG_NOSUB);
5279 /* POSIX says a null character in the pattern terminates it, so we
5280 can use strlen here in compiling the pattern. */
5281 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5283 /* POSIX doesn't distinguish between an unmatched open-group and an
5284 unmatched close-group: both are REG_EPAREN. */
5285 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5291 /* regexec searches for a given pattern, specified by PREG, in the
5294 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5295 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5296 least NMATCH elements, and we set them to the offsets of the
5297 corresponding matched substrings.
5299 EFLAGS specifies `execution flags' which affect matching: if
5300 REG_NOTBOL is set, then ^ does not match at the beginning of the
5301 string; if REG_NOTEOL is set, then $ does not match at the end.
5303 We return 0 if we find a match and REG_NOMATCH if not. */
5306 regexec (preg, string, nmatch, pmatch, eflags)
5307 const regex_t *preg;
5310 regmatch_t pmatch[];
5314 struct re_registers regs;
5315 regex_t private_preg;
5316 int len = strlen (string);
5317 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5319 private_preg = *preg;
5321 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5322 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5324 /* The user has told us exactly how many registers to return
5325 information about, via `nmatch'. We have to pass that on to the
5326 matching routines. */
5327 private_preg.regs_allocated = REGS_FIXED;
5331 regs.num_regs = nmatch;
5332 regs.start = TALLOC (nmatch, regoff_t);
5333 regs.end = TALLOC (nmatch, regoff_t);
5334 if (regs.start == NULL || regs.end == NULL)
5335 return (int) REG_NOMATCH;
5338 /* Perform the searching operation. */
5339 ret = re_search (&private_preg, string, len,
5340 /* start: */ 0, /* range: */ len,
5341 want_reg_info ? ®s : (struct re_registers *) 0);
5343 /* Copy the register information to the POSIX structure. */
5350 for (r = 0; r < nmatch; r++)
5352 pmatch[r].rm_so = regs.start[r];
5353 pmatch[r].rm_eo = regs.end[r];
5357 /* If we needed the temporary register info, free the space now. */
5362 /* We want zero return to mean success, unlike `re_search'. */
5363 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5367 /* Returns a message corresponding to an error code, ERRCODE, returned
5368 from either regcomp or regexec. We don't use PREG here. */
5371 regerror (errcode, preg, errbuf, errbuf_size)
5373 const regex_t *preg;
5381 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5382 /* Only error codes returned by the rest of the code should be passed
5383 to this routine. If we are given anything else, or if other regex
5384 code generates an invalid error code, then the program has a bug.
5385 Dump core so we can fix it. */
5388 msg = gettext (re_error_msgid[errcode]);
5390 msg_size = strlen (msg) + 1; /* Includes the null. */
5392 if (errbuf_size != 0)
5394 if (msg_size > errbuf_size)
5396 strncpy (errbuf, msg, errbuf_size - 1);
5397 errbuf[errbuf_size - 1] = 0;
5400 strcpy (errbuf, msg);
5407 /* Free dynamically allocated space used by PREG. */
5413 if (preg->buffer != NULL)
5414 free (preg->buffer);
5415 preg->buffer = NULL;
5417 preg->allocated = 0;
5420 if (preg->fastmap != NULL)
5421 free (preg->fastmap);
5422 preg->fastmap = NULL;
5423 preg->fastmap_accurate = 0;
5425 if (preg->translate != NULL)
5426 free (preg->translate);
5427 preg->translate = NULL;
5430 #endif /* not emacs */
5434 make-backup-files: t
5436 trim-versions-without-asking: nil