xemacs-beta: src/regex.c comparison

comparison src/regex.c @ 183:e121b013d1f0 r20-3b18

Import from CVS: tag r20-3b18

author	cvs
date	Mon, 13 Aug 2007 09:54:23 +0200
parents	9ad43877534d
children	3d6bfa290dbd

comparison

equal deleted inserted replaced

-:f07455f06202
+:e121b013d1f0
 /* Define the syntax stuff for \<, \>, etc.  */
 /* This must be nonzero for the wordchar and notwordchar pattern
 commands in re_match_2.  */
 #ifndef Sword
 #define Sword 1
 #endif
 #ifdef SYNTAX_TABLE
 /* Should we use malloc or alloca?  If REGEX_MALLOC is not defined, we
 use `alloca' instead of `malloc'.  This is because using malloc in
 re_search* or re_match* could cause memory leaks when C-g is used in
 Emacs; also, malloc is slower and causes storage fragmentation.  On
 the other hand, malloc is more portable, and easier to debug.
 Because we sometimes use alloca, some routines have to be macros,
 not functions -- `alloca'-allocated space disappears at the end of the
 function it is called in.  */
 #ifdef REGEX_MALLOC
 #include <alloca.h>
 #else /* not __GNUC__ or HAVE_ALLOCA_H */
 #ifndef _AIX /* Already did AIX, up at the top.  */
 char *alloca ();
 #endif /* not _AIX */
 #endif /* not HAVE_ALLOCA_H */
 #endif /* not __GNUC__ */
 #endif /* not alloca */
 #define REGEX_ALLOCATE alloca
 of string to be matched (if not).  */
 begbuf,
 /* Analogously, for end of buffer/string.  */
 endbuf,
 /* Followed by two byte relative address to which to jump.  */
 jump,
 	/* Same as jump, but marks the end of an alternative.  */
 jump_past_alt,
 /* Followed by two-byte relative address of place to resume at
 in case of failure.  */
 on_failure_jump,
 /* Like on_failure_jump, but pushes a placeholder instead of the
 current string position when executed.  */
 on_failure_keep_string_jump,
 /* Throw away latest failure point and then jump to following
 two-byte relative address.  */
 pop_failure_jump,
 /* Change to pop_failure_jump if know won't have to backtrack to
 /* 97/2/17 jhod: The following two were merged back in from the Mule
 2.3 code to enable some language specific processing */
 ,categoryspec,     /* Matches entries in the character category tables */
 notcategoryspec    /* The opposite of the above */
 #endif
 } re_opcode_t;
 /* Common operations on the compiled pattern.  */
 /* Store NUMBER in two contiguous bytes starting at DESTINATION.  */
 #ifdef DEBUG
 static void
 extract_number (int *dest, unsigned char *source)
 {
 int temp = SIGN_EXTEND_CHAR (*(source + 1));
 *dest = *source & 0377;
 *dest += temp << 8;
 }
 #ifndef EXTRACT_MACROS /* To debug the macros.  */
 } while (0)
 #ifdef DEBUG
 static void
 extract_number_and_incr (int *destination, unsigned char **source)
 {
 extract_number (destination, *source);
 *source += 2;
 }
 #ifndef EXTRACT_MACROS
 static void
 print_fastmap (char *fastmap)
 {
 unsigned was_a_range = 0;
 unsigned i = 0;
 while (i < (1 << BYTEWIDTH))
 {
 if (fastmap[i++])
 	{
 	  was_a_range = 0;
 printf ("-");
 putchar (i - 1);
 }
 }
 }
 putchar ('\n');
 }
 /* Print a compiled pattern string in human-readable form, starting at
 the START pointer into it and ending just before the pointer END.  */
 if (start == NULL)
 {
 printf ("(null)\n");
 return;
 }
 /* Loop over pattern commands.  */
 while (p < pend)
 {
 printf ("%d:\t", p - start);
 register int c, last = -100;
 	    register int in_range = 0;
 	    printf ("/charset [%s",
 	            (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
 assert (p + *p < pend);
 for (c = 0; c < 256; c++)
 	      if (((unsigned char) (c / 8) < *p)
 		  && (p[1 + (c/8)] & (1 << (c % 8))))
 		  else if (last + 1 != c && in_range)
 		    {
 		      putchar (last);
 		      in_range = 0;
 		    }
 		  if (! in_range)
 		    putchar (c);
 		  last = c;
 }
 break;
 	case push_dummy_failure:
 printf ("/push_dummy_failure");
 break;
 case maybe_pop_jump:
 extract_number_and_incr (&mcnt, &p);
 	  printf ("/maybe_pop_jump to %d", p + mcnt - start);
 	  break;
 case pop_failure_jump:
 	  extract_number_and_incr (&mcnt, &p);
 	  printf ("/pop_failure_jump to %d", p + mcnt - start);
 	  break;
 case jump_past_alt:
 	  extract_number_and_incr (&mcnt, &p);
 	  printf ("/jump_past_alt to %d", p + mcnt - start);
 	  break;
 case jump:
 	  extract_number_and_incr (&mcnt, &p);
 	  printf ("/jump to %d", p + mcnt - start);
 	  break;
 case succeed_n:
 extract_number_and_incr (&mcnt, &p);
 extract_number_and_incr (&mcnt2, &p);
 	  printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
 break;
 case jump_n:
 extract_number_and_incr (&mcnt, &p);
 extract_number_and_incr (&mcnt2, &p);
 	  printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
 break;
 case set_number_at:
 extract_number_and_incr (&mcnt, &p);
 extract_number_and_incr (&mcnt2, &p);
 	  printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
 break;
 case wordbound:
 	  printf ("/wordbound");
 	  break;
 	case notwordbound:
 break;
 	case wordbeg:
 	  printf ("/wordbeg");
 	  break;
 	case wordend:
 	  printf ("/wordend");
 #ifdef emacs
 	case before_dot:
 	  printf ("/before_dot");
 break;
 	case syntaxspec:
 printf ("/syntaxspec");
 	  mcnt = *p++;
 	  printf ("/%d", mcnt);
 break;
 	case notsyntaxspec:
 printf ("/notsyntaxspec");
 	  mcnt = *p++;
 	  printf ("/%d", mcnt);
 	  break;
 #ifdef MULE
 /* 97/2/17 jhod Mule category patch */
 	case categoryspec:
 	  printf ("/categoryspec");
 	  mcnt = *p++;
 	  printf ("/%d", mcnt);
 	  break;
 	case notcategoryspec:
 	  printf ("/notcategoryspec");
 	  mcnt = *p++;
 	  printf ("/%d", mcnt);
 	  break;
 /* end of category patch */
 #endif /* MULE */
 #endif /* emacs */
 	case wordchar:
 	  printf ("/wordchar");
 break;
 	case notwordchar:
 	  printf ("/notwordchar");
 break;
 	case begbuf:
 static void
 print_double_string (CONST char *where, CONST char *string1, int size1,
 		     CONST char *string2, int size2)
 {
 unsigned this_char;
 if (where == NULL)
 printf ("(null)");
 else
 {
 if (FIRST_STRING_P (where))
 {
 for (this_char = where - string1; this_char < size1; this_char++)
 putchar (string1[this_char]);
 where = string2;
 }
 for (this_char = where - string2; this_char < size2; this_char++)
 putchar (string2[this_char]);
 }
 reg_syntax_t
 re_set_syntax (reg_syntax_t syntax)
 {
 reg_syntax_t ret = re_syntax_options;
 re_syntax_options = syntax;
 return ret;
 }
 /* This table gives an error message for each of the error codes listed
 using the relocating allocator routines, then malloc could cause a
 relocation, which might (if the strings being searched are in the
 ralloc heap) shift the data out from underneath the regexp
 routines.
 Here's another reason to avoid allocation: Emacs
 processes input from X in a signal handler; processing X input may
 call malloc; if input arrives while a matching routine is calling
 malloc, then we're scrod.  But Emacs can't just block input while
 calling matching routines; then we don't notice interrupts when
 they come in.  So, Emacs blocks input around all regexp calls
 /* Failure stack declarations and macros; both re_compile_fastmap and
 re_match_2 use a failure stack.  These have to be macros because of
 REGEX_ALLOCATE_STACK.  */
 /* Number of failure points for which to initially allocate space
 when matching.  If this number is exceeded, we allocate more
 space, so it is not a hard limit.  */
 #ifndef INIT_FAILURE_ALLOC
 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
 Return 1 if succeeds, and 0 if either ran out of memory
 allocating space for it or it was already too large.
 REGEX_REALLOCATE_STACK requires `destination' be declared.   */
 #define DOUBLE_FAIL_STACK(fail_stack)					\
 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS		\
 ? 0									\
 ? 0								\
 : ((fail_stack).size <<= 1, 					\
 1)))
 /* Push pointer POINTER on FAIL_STACK.
 Return 1 if was able to do so and 0 if ran out of memory allocating
 space to do so.  */
 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK)				\
 ((FAIL_STACK_FULL ()							\
 && !DOUBLE_FAIL_STACK (FAIL_STACK))					\
 #define DEBUG_POP(item_addr)
 #endif
 /* Push the information about the state we will need
 if we ever fail back to it.
 Requires variables fail_stack, regstart, regend, reg_info, and
 num_regs be declared.  DOUBLE_FAIL_STACK requires `destination' be
 declared.
 Does `return FAILURE_CODE' if runs out of memory.  */
 #if !defined (REGEX_MALLOC) && !defined (REL_ALLOC)
 #define DECLARE_DESTINATION char *destination;
 #else
 STR -- the saved data position.
 PAT -- the saved pattern position.
 LOW_REG, HIGH_REG -- the highest and lowest active registers.
 REGSTART, REGEND -- arrays of string positions.
 REG_INFO -- array of information about each subexpression.
 Also assumes the variables `fail_stack' and (if debugging), `bufp',
 `pend', `string1', `size1', `string2', and `size2'.  */
 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
 {									\
 /* Structure for per-register (a.k.a. per-group) information.
 Other register information, such as the
 starting and ending positions (which are addresses), and the list of
 inner groups (which is a bits list) are maintained in separate
 variables.
 We are making a (strictly speaking) nonportable assumption here: that
 the compiler will pack our bit fields into something that fits into
 the type of `word', i.e., is something that fits into one item on the
 failure stack.  */
 #define REG_UNSET_VALUE (&reg_unset_dummy)
 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
 /* Subroutine declarations and macros for regex_compile.  */
 /* Fetch the next character in the uncompiled pattern---translating it
 if necessary.  Also cast from a signed character in the constant
 string passed to us by the user to an unsigned char that we can use
 as an array index (in, e.g., `translate').  */
 #define PATFETCH(c)							\
 do {if (p == pend) return REG_EEND;					\
 typedef struct
 {
 pattern_offset_t begalt_offset;
 pattern_offset_t fixup_alt_jump;
 pattern_offset_t inner_group_offset;
 pattern_offset_t laststart_offset;
 regnum_t regnum;
 } compile_stack_elt_t;
 typedef struct
 if (p == pend) 						\
 break; 							\
 PATFETCH (c);						\
 } 								\
 } 								\
 }
 #define CHAR_CLASS_MAX_LENGTH  6 /* Namely, `xdigit'.  */
 #define IS_CHAR_CLASS(string)						\
 (STREQ (string, "alpha") || STREQ (string, "upper")			\
 static int regs_allocated_size;
 static CONST char **     regstart, **     regend;
 static CONST char ** old_regstart, ** old_regend;
 static CONST char **best_regstart, **best_regend;
 static register_info_type *reg_info;
 static CONST char **reg_dummy;
 static register_info_type *reg_info_dummy;
 /* Make the register vectors big enough for NUM_REGS registers,
 but don't make them smaller.  */
 `syntax' is set to SYNTAX;
 `used' is set to the length of the compiled pattern;
 `fastmap_accurate' is zero;
 `re_nsub' is the number of subexpressions in PATTERN;
 `not_bol' and `not_eol' are zero;
 The `fastmap' and `newline_anchor' fields are neither
 examined nor set.  */
 /* Return, freeing storage we allocated.  */
 #define FREE_STACK_RETURN(value)		\
 array indices.  The macros that fetch a character from the pattern
 make sure to coerce to unsigned char before assigning, so we won't
 get bitten by negative numbers here. */
 /* XEmacs change: used to be unsigned char. */
 register EMACS_INT c, c1;
 /* A random temporary spot in PATTERN.  */
 CONST char *p1;
 /* Points to the end of the buffer, where we should append.  */
 register unsigned char *b;
 /* Keeps track of unclosed groups.  */
 compile_stack_type compile_stack;
 /* Points to the current (ending) position in the pattern.  */
 CONST char *p = pattern;
 CONST char *pend = pattern + size;
 /* How to translate the characters in the pattern.  */
 char *translate = bufp->translate;
 /* Address of the count-byte of the most recently inserted `exactn'
 command.  This makes it possible to tell if a new exact-match
 unsigned char *begalt;
 /* Place in the uncompiled pattern (i.e., the {) to
 which to go back if the interval is invalid.  */
 CONST char *beg_interval;
 /* Address of the place where a forward jump should go to the end of
 the containing expression.  Each alternative of an `or' -- except the
 last -- ends with a forward jump of this sort.  */
 unsigned char *fixup_alt_jump = 0;
 #ifdef DEBUG
 DEBUG_PRINT1 ("\nCompiling pattern: ");
 if (debug)
 {
 unsigned debug_count;
 for (debug_count = 0; debug_count < size; debug_count++)
 putchar (pattern[debug_count]);
 putchar ('\n');
 }
 #endif /* DEBUG */
 /* Set `used' to zero, so that if we return an error, the pattern
 printer (for debugging) will think there's no pattern.  We reset it
 at the end.  */
 bufp->used = 0;
 /* Always count groups, whether or not bufp->no_sub is set.  */
 bufp->re_nsub = 0;
 #if !defined (emacs) && !defined (SYNTAX_TABLE)
 /* Initialize the syntax table.  */
 init_syntax_once ();
 #endif
 case '$':
 {
 if (   /* If at end of pattern, it's an operator.  */
 p == pend
 /* If context independent, it's an operator.  */
 || syntax & RE_CONTEXT_INDEP_ANCHORS
 /* Otherwise, depends on what's next.  */
 || at_endline_loc_p (p, pend, syntax))
 BUF_PUSH (endline);
 }
 {
 /* Are we optimizing this jump?  */
 boolean keep_string_p = false;
 /* 1 means zero (many) matches is allowed.  */
 char zero_times_ok = 0, many_times_ok = 0;
 /* If there is a sequence of repetition chars, collapse it
 down to just one (the right one).  We can't combine
 /* If we get here, we found another repeat character.  */
 }
 /* Star, etc. applied to an empty pattern is equivalent
 to an empty pattern.  */
 if (!laststart)
 break;
 /* Now we know whether or not zero matches is allowed
 and also whether or not two or more matches is allowed.  */
 if (many_times_ok)
 { /* More than one repetition is allowed, so put in at the
 end a backward relative jump from `b' to before the next
 jump we're going to put in below (which jumps from
 laststart to after this jump).
 But if we are at the `*' in the exact sequence `.*\n',
 insert an unconditional jump backwards to the .,
 instead of the beginning of the loop.  This way we only
 push a failure point once, instead of every time
 laststart = b;
 /* We test `*p == '^' twice, instead of using an if
 statement, so we only need one BUF_PUSH.  */
 BUF_PUSH (*p == '^' ? charset_not : charset);
 if (*p == '^')
 p++;
 /* Remember the first position in the bracket expression.  */
 p1 = p;
 /* Look ahead to see if it's a range when the last thing
 was a character: if this is a hyphen not at the
 beginning or the end of a list, then it's the range
 operator.  */
 if (c == '-'
 && !(p - 2 >= pattern && p[-2] == '[')
 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
 && *p != ']')
 {
 reg_errcode_t ret;
 str[c1++] = c;
 }
 str[c1] = '\0';
 /* If isn't a word bracketed by `[:' and:`]':
 undo the ending character, the letters, and leave
 the leading `:' and `[' (but set bits for them).  */
 if (c == ':' && *p == ']')
 {
 int ch;
 boolean is_alnum = STREQ (str, "alnum");
 boolean is_print = STREQ (str, "print");
 boolean is_punct = STREQ (str, "punct");
 boolean is_space = STREQ (str, "space");
 boolean is_upper = STREQ (str, "upper");
 boolean is_xdigit = STREQ (str, "xdigit");
 if (!IS_CHAR_CLASS (str))
 			  FREE_STACK_RETURN (REG_ECTYPE);
 /* Throw away the ] at the end of the character
 class.  */
 PATFETCH (c);
 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
 {
 had_char_class = true;
 }
 else
 {
 c1++;
 while (c1--)
 PATUNFETCH;
 SET_EITHER_BIT ('[');
 SET_EITHER_BIT (':');
 had_char_class = false;
 }
 		break;
 	      }
 #endif /* MULE */
 /* Discard any (non)matching list bytes that are all 0 at the
 end of the map.  Decrease the map-length byte too.  */
 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
 b[-1]--;
 b += b[-1];
 	  }
 	  break;
 handle_open:
 bufp->re_nsub++;
 regnum++;
 if (COMPILE_STACK_FULL)
 {
 RETALLOC (compile_stack.stack, compile_stack.size << 1,
 compile_stack_elt_t);
 if (compile_stack.stack == NULL) return REG_ESPACE;
 compile_stack.size <<= 1;
 /* These are the values to restore when we hit end of this
 group.  They are all relative offsets, so that if the
 whole pattern moves because of realloc, they will still
 be valid.  */
 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
 COMPILE_STACK_TOP.fixup_alt_jump
 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
 COMPILE_STACK_TOP.regnum = regnum;
 /* We will eventually replace the 0 with the number of
 if (regnum <= MAX_REGNUM)
 {
 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
 BUF_PUSH_3 (start_memory, regnum, 0);
 }
 compile_stack.avail++;
 fixup_alt_jump = 0;
 laststart = 0;
 begalt = b;
 { /* Push a dummy failure point at the end of the
 alternative for a possible future
 `pop_failure_jump' to pop.  See comments at
 `push_dummy_failure' in `re_match_2'.  */
 BUF_PUSH (push_dummy_failure);
 /* We allocated space for this jump when we assigned
 to `fixup_alt_jump', in the `handle_alt' case below.  */
 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
 }
 /* We don't just want to restore into `regnum', because
 later groups should continue to be numbered higher,
 as in `(ab)c(de)' -- the second group is #2.  */
 regnum_t this_group_regnum;
 compile_stack.avail--;
 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
 fixup_alt_jump
 = COMPILE_STACK_TOP.fixup_alt_jump
 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
 : 0;
 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
 this_group_regnum = COMPILE_STACK_TOP.regnum;
 		/* If we've reached MAX_REGNUM groups, then this open
 		   won't actually generate any code, so we'll have to
 groups were inside this one.  */
 if (this_group_regnum <= MAX_REGNUM)
 {
 unsigned char *inner_group_loc
 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
 *inner_group_loc = regnum - this_group_regnum;
 BUF_PUSH_3 (stop_memory, this_group_regnum,
 regnum - this_group_regnum);
 }
 }
 which gets executed if it gets matched.  Adjust that
 jump so it will jump to this alternative's analogous
 jump (put in below, which in turn will jump to the next
 (if any) alternative's such jump, etc.).  The last such
 jump jumps to the correct final destination.  A picture:
 _____ _____
 |   | |   |
 |   v |   v
 a | b   | c
 If we are at `b', then fixup_alt_jump right now points to a
 three-byte space after `a'.  We'll put in the jump, set
 fixup_alt_jump to right after `b', and leave behind three
 bytes which we'll fill in when we get to after `c'.  */
 laststart = 0;
 begalt = b;
 break;
 case '{':
 /* If \{ is a literal.  */
 if (!(syntax & RE_INTERVALS)
 /* If we're at `\{' and it's not the open-interval
 operator.  */
 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
 || (p - 2 == pattern  &&  p == pend))
 goto normal_backslash;
 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
 || lower_bound > upper_bound)
 {
 if (syntax & RE_NO_BK_BRACES)
 goto unfetch_interval;
 else
 FREE_STACK_RETURN (REG_BADBR);
 }
 if (!(syntax & RE_NO_BK_BRACES))
 {
 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
 PATFETCH (c);
 }
 if (c != '}')
 {
 if (syntax & RE_NO_BK_BRACES)
 goto unfetch_interval;
 else
 FREE_STACK_RETURN (REG_BADBR);
 }
 /* We just parsed a valid interval.  */
 succeed_n <after jump addr> <succeed_n count>
 <body of loop>
 jump_n <succeed_n addr> <jump count>
 (The upper bound and `jump_n' are omitted if
 `upper_bound' is 1, though.)  */
 else
 { /* If the upper bound is > 1, we need to insert
 more at the end of the loop.  */
 unsigned nbytes = 10 + (upper_bound > 1) * 10;
 GET_BUFFER_SPACE (nbytes);
 INSERT_JUMP2 (succeed_n, laststart,
 b + 5 + (upper_bound > 1) * 5,
 lower_bound);
 b += 5;
 /* Code to initialize the lower bound.  Insert
 before the `succeed_n'.  The `5' is the last two
 bytes of this `set_number_at', plus 3 bytes of
 the following `succeed_n'.  */
 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
 b += 5;
 if (upper_bound > 1)
 { /* More than one repetition is allowed, so
 append a backward jump to the `succeed_n'
 that starts this interval.
 When we've reached this during matching,
 we'll have matched the interval once, so
 jump back only `upper_bound - 1' times.  */
 STORE_JUMP2 (jump_n, b, laststart + 5,
 upper_bound - 1);
 for the relative address.  But we are
 inserting into the middle of the pattern --
 so everything is getting moved up by 5.
 Conclusion: (b - 2) - (laststart + 3) + 5,
 i.e., b - laststart.
 We insert this at the beginning of the loop
 so that if we fail during matching, we'll
 reinitialize the bounds.  */
 insert_op2 (set_number_at, laststart, b - laststart,
 upper_bound - 1, b);
 assert (beg_interval);
 p = beg_interval;
 beg_interval = NULL;
 /* normal_char and normal_backslash need `c'.  */
 PATFETCH (c);
 if (!(syntax & RE_NO_BK_BRACES))
 {
 if (p > pattern  &&  p[-1] == '\\')
 goto normal_backslash;
 operators.  rms says this is ok.  --karl  */
 case '=':
 BUF_PUSH (at_dot);
 break;
 case 's':
 laststart = b;
 PATFETCH (c);
 	      /* XEmacs addition */
 	      if (c >= 0x80 || syntax_spec_code[c] == 0377)
 		FREE_STACK_RETURN (REG_ESYNTAX);
 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
 break;
 #ifdef MULE
 /* 97.2.17 jhod merged in to XEmacs from mule-2.3 */
 	    case 'c':
 	      laststart = b;
 	      PATFETCH_RAW (c);
 	      if (c < 32 || c > 127)
 		FREE_STACK_RETURN (REG_ECATEGORY);
 	      BUF_PUSH_2 (categoryspec, c);
 	    /* `q' points to the beginning of the next char. */
 	    CONST char *q = p - 1;
 	    INC_CHARPTR (q);
 	    /* If no exactn currently being built.  */
 	    if (!pending_exact
 		/* If last exactn not at current position.  */
 		|| pending_exact + *pending_exact + 1 != b
 		/* We have only one byte following the exactn for the count. */
 		|| ((unsigned int) (*pending_exact + (q - p)) >=
 		    ((unsigned int) (1 << BYTEWIDTH) - 1))
 		/* If followed by a repetition operator.  */
 		    && ((syntax & RE_NO_BK_BRACES)
 			? *q == '{'
 			: (q[0] == '\\' && q[1] == '{'))))
 	      {
 		/* Start building a new exactn.  */
 		laststart = b;
 		BUF_PUSH_2 (exactn, 0);
 		pending_exact = b - 1;
 	      }
 	    BUF_PUSH (c);
 	    (*pending_exact)++;
 	    while (p < q)
 	      {
 	    break;
 	  }
 } /* switch (c) */
 } /* while p != pend */
 /* Through the pattern now.  */
 if (fixup_alt_jump)
 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
 if (!COMPILE_STACK_EMPTY)
 FREE_STACK_RETURN (REG_EPAREN);
 /* If we don't want backtracking, force success
 the first time we reach the end of the compiled pattern.  */
 if (syntax & RE_NO_POSIX_BACKTRACKING)
 	fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
 #ifdef emacs
 	if (! fail_stack.stack)
 	  fail_stack.stack
 	    = (fail_stack_elt_t *) xmalloc (fail_stack.size
 					    * sizeof (fail_stack_elt_t));
 	else
 	  fail_stack.stack
 	    = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
 					     (fail_stack.size
 					      * sizeof (fail_stack_elt_t)));
 #else /* not emacs */
 	if (! fail_stack.stack)
 	  fail_stack.stack
 	    = (fail_stack_elt_t *) malloc (fail_stack.size
 					   * sizeof (fail_stack_elt_t));
 	else
 	  fail_stack.stack
 	    = (fail_stack_elt_t *) realloc (fail_stack.stack,
 					    (fail_stack.size
 register unsigned char *pfrom = end;
 register unsigned char *pto = end + 3;
 while (pfrom != loc)
 *--pto = *--pfrom;
 store_op1 (op, loc, arg);
 }
 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2.  */
 static void
 insert_op2 (re_opcode_t op, unsigned char *loc, int arg1, int arg2,
 	    unsigned char *end)
 {
 register unsigned char *pfrom = end;
 register unsigned char *pto = end + 5;
 while (pfrom != loc)
 *--pto = *--pfrom;
 store_op2 (op, loc, arg1, arg2);
 }
 /* P points to just after a ^ in PATTERN.  Return true if that ^ comes
 static boolean
 at_begline_loc_p (CONST char *pattern, CONST char *p, reg_syntax_t syntax)
 {
 CONST char *prev = p - 2;
 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
 return
 /* After a subexpression?  */
 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
 /* After an alternative?  */
 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
 at_endline_loc_p (CONST char *p, CONST char *pend, int syntax)
 {
 CONST char *next = p;
 boolean next_backslash = *next == '\\';
 CONST char *next_next = p + 1 < pend ? p + 1 : 0;
 return
 /* Before a subexpression?  */
 (syntax & RE_NO_BK_PARENS ? *next == ')'
 : next_backslash && next_next && *next_next == ')')
 /* Before an alternative?  */
 || (syntax & RE_NO_BK_VBAR ? *next == '|'
 : next_backslash && next_next && *next_next == '|');
 }
 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
 false if it's not.  */
 static boolean
 group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum)
 {
 int this_element;
 for (this_element = compile_stack.avail - 1;
 this_element >= 0;
 this_element--)
 if (compile_stack.stack[this_element].regnum == regnum)
 return true;
 return false;
 /* Read the ending character of a range (in a bracket expression) from the
 uncompiled pattern *P_PTR (which ends at PEND).  We assume the
 starting character is in `P[-2]'.  (`P[-1]' is the character `-'.)
 Then we set the translation of all bits between the starting and
 ending characters (inclusive) in the compiled pattern B.
 Return an error code.
 We use these short variable names so we can use the same macros as
 `regex_compile' itself.  */
 static reg_errcode_t
 compile_range (CONST char **p_ptr, CONST char *pend, char *translate,
 {
 unsigned this_char;
 CONST char *p = *p_ptr;
 int range_start, range_end;
 if (p == pend)
 return REG_ERANGE;
 /* Even though the pattern is a signed `char *', we need to fetch
 with unsigned char *'s; if the high bit of the pattern character
 is set, the range endpoints will be negative if we fetch using a
 signed char *.
 We also want to fetch the endpoints without translating them; the
 appropriate translation is done in the bit-setting loop below.  */
 /* The SVR4 compiler on the 3B2 had trouble with unsigned CONST char *.  */
 range_start = ((CONST unsigned char *) p)[-2];
 range_end   = ((CONST unsigned char *) p)[0];
 loop, since all characters <= 0xff.  */
 for (this_char = range_start; this_char <= range_end; this_char++)
 {
 SET_LIST_BIT (TRANSLATE (this_char));
 }
 return REG_NOERROR;
 }
 #ifdef MULE
 compile_extended_range (CONST char **p_ptr, CONST char *pend, char *translate,
 			reg_syntax_t syntax, Lisp_Object rtab)
 {
 Emchar this_char, range_start, range_end;
 CONST Bufbyte *p;
 if (*p_ptr == pend)
 return REG_ERANGE;
 p = (CONST Bufbyte *) *p_ptr;
 range_end = charptr_emchar (p);
 p--; /* back to '-' */
 DEC_CHARPTR (p); /* back to start of range */
 /* We also want to fetch the endpoints without translating them; the
 appropriate translation is done in the bit-setting loop below.  */
 range_start = charptr_emchar (p);
 INC_CHARPTR (*p_ptr);
 /* If the start is after the end, the range is empty.  */
 if (range_start > range_end)
 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
 /* Can't have ranges spanning different charsets, except maybe for
 ranges entirely witin the first 256 chars. */
 if ((range_start >= 0x100 || range_end >= 0x100)
 && CHAR_LEADING_BYTE (range_start) !=
 CHAR_LEADING_BYTE (range_end))
 return REG_ERANGESPAN;
 SET_RANGETAB_BIT (TRANSLATE (this_char));
 }
 if (this_char <= range_end)
 put_range_table (rtab, this_char, range_end, Qt);
 return REG_NOERROR;
 }
 #endif /* MULE */
 characters can start a string that matches the pattern.  This fastmap
 is used by re_search to skip quickly over impossible starting points.
 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
 area as BUFP->fastmap.
 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
 the pattern buffer.
 Returns 0 if we succeed, -2 if an internal error.   */
 #ifdef MATCH_MAY_ALLOCATE
 fail_stack_type fail_stack;
 #endif
 DECLARE_DESTINATION
 /* We don't push any register information onto the failure stack.  */
 register char *fastmap = bufp->fastmap;
 unsigned char *pattern = bufp->buffer;
 unsigned long size = bufp->used;
 unsigned char *p = pattern;
 register unsigned char *pend = pattern + size;
 /* We aren't doing a `succeed_n' to begin with.  */
 boolean succeed_n_p = false;
 assert (fastmap != NULL && p != NULL);
 INIT_FAIL_STACK ();
 bzero (fastmap, 1 << BYTEWIDTH);  /* Assume nothing's valid.  */
 bufp->fastmap_accurate = 1;	    /* It will be when we're done.  */
 bufp->can_be_null = 0;
 while (1)
 {
 if (p == pend || *p == succeed)
 	{
 	  /* We have reached the (effective) end of pattern.  */
 	    break;
 	}
 /* We should never be about to go beyond the end of the pattern.  */
 assert (p < pend);
 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
 	{
 /* I guess the idea here is to simply not bother with a fastmap
 if a backreference is used, since it's too hard to figure out
 		 (regex_emacs_buffer->mirror_syntax_table), j) ==
 		(enum syntaxcode) k)
 	      fastmap[j] = 1;
 	  for (j = 0x80; j < 0xA0; j++)
 	    {
-	      if (j == PRE_LEADING_BYTE_PRIVATE_1
+	      if (LEADING_BYTE_PREFIX_P(j))
-		  || j == PRE_LEADING_BYTE_PRIVATE_2)
 		/* too complicated to calculate this right */
 		fastmap[j] = 1;
 	      else
 		{
 		  int multi_p;
 		 (regex_emacs_buffer->mirror_syntax_table), j) !=
 		(enum syntaxcode) k)
 	      fastmap[j] = 1;
 	  for (j = 0x80; j < 0xA0; j++)
 	    {
-	      if (j == PRE_LEADING_BYTE_PRIVATE_1
+	      if (LEADING_BYTE_PREFIX_P(j))
-		  || j == PRE_LEADING_BYTE_PRIVATE_2)
 		/* too complicated to calculate this right */
 		fastmap[j] = 1;
 	      else
 		{
 		  int multi_p;
 	case maybe_pop_jump:
 	case jump:
 case jump_past_alt:
 	case dummy_failure_jump:
 EXTRACT_NUMBER_AND_INCR (j, p);
 	  p += j;
 	  if (j > 0)
 	    continue;
 /* Jump backward implies we just went through the body of a
 loop and matched nothing.  Opcode jumped to should be
 `on_failure_jump' or `succeed_n'.  Just treat it like an
 ordinary jump.  For a * loop, it has pushed its failure
 point already; if so, discard that as redundant.  */
 	      && (re_opcode_t) *p != succeed_n)
 	    continue;
 p++;
 EXTRACT_NUMBER_AND_INCR (j, p);
 p += j;
 /* If what's on the stack is where we are now, pop it.  */
 if (!FAIL_STACK_EMPTY ()
 	      && fail_stack.stack[fail_stack.avail - 1].pointer == p)
 fail_stack.avail--;
 continue;
 continue;
 	case succeed_n:
 /* Get to the number of times to succeed.  */
 p += 2;
 /* Increment p past the n for when k != 0.  */
 EXTRACT_NUMBER_AND_INCR (k, p);
 if (k == 0)
 	    {
 int
 re_search (struct re_pattern_buffer *bufp, CONST char *string, int size,
 	   int startpos, int range, struct re_registers *regs)
 {
 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
 		      regs, size);
 }
 #ifndef emacs
 /* Snarfed from src/lisp.h, needed for compiling [ce]tags. */
 STARTPOS, then at STARTPOS + 1, and so on.
 With MULE, STARTPOS is a byte position, not a char position.  And the
 search will increment STARTPOS by the width of the current leading
 character.
 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
 RANGE is how far to scan while trying to match.  RANGE = 0 means try
 only at STARTPOS; in general, the last start tried is STARTPOS +
 RANGE.
 With MULE, RANGE is a byte position, not a char position.  The last
 start tried is the character starting <= STARTPOS + RANGE.
 In REGS, return the indices of the virtual concatenation of STRING1
 and STRING2 that matched the entire BUFP->buffer and its contained
 subexpressions.
 Do not consider matching one past the index STOP in the virtual
 concatenation of STRING1 and STRING2.
 We return either the position in the strings at which the match was
 found, -1 if no match, or -2 if error (such as failure
 Charcount d_size;
 /* Check for out-of-range STARTPOS.  */
 if (startpos < 0 || startpos > total_size)
 return -1;
 /* Fix up RANGE if it might eventually take us outside
 the virtual concatenation of STRING1 and STRING2.  */
 if (endpos < 0)
 range = 0 - startpos;
 else if (endpos > total_size)
 /* Update the fastmap now if not correct already.  */
 if (fastmap && !bufp->fastmap_accurate)
 if (re_compile_fastmap (bufp) == -2)
 return -2;
 #ifdef REGEX_BEGLINE_CHECK
 {
 int i = 0;
 while (i < bufp->used)
 }
 #endif
 /* Loop through the string, looking for a place to start matching.  */
 for (;;)
 {
 #ifdef REGEX_BEGLINE_CHECK
 /* If the regex is anchored at the beginning of a line (i.e. with a ^),
 	 then we can speed things up by skipping to the next beginning-of-
 	 line. */
 if (anchored_at_begline && startpos > 0 && startpos != size1 &&
 	      startpos += irange - range;
 	    }
 	  else				/* Searching backwards.  */
 	    {
 	      unsigned char c = (size1 == 0 || startpos >= size1
 				 ? string2[startpos - size1]
 				 : string1[startpos]);
 #ifdef MULE
 	      if (c < 0x80 && !fastmap[(unsigned char) TRANSLATE (c)])
 #else
 	      if (!fastmap[(unsigned char) TRANSLATE (c)])
 #endif
 #endif
 if (val >= 0)
 	return startpos;
 if (val == -2)
 	return -2;
 advance:
 if (!range)
 	break;
 else if (range > 0)
 	{
 	  d = ((CONST unsigned char *)
 	       (startpos >= size1 ? string2 - size1 : string1) + startpos);
 	  d_size = charcount_to_bytecount (d, 1);
 	  range -= d_size;
 /* Test if at very beginning or at very end of the virtual concatenation
 of `string1' and `string2'.  If only one string, it's `string2'.  */
 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
 #define AT_STRINGS_END(d) ((d) == end2)
 /* XEmacs change:
 If the given position straddles the string gap, return the equivalent
 position that is before or after the gap, respectively; otherwise,
 return the same position. */
 /* re_match_2 matches the compiled pattern in BUFP against the
 (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 and
 SIZE2, respectively).  We start matching at POS, and stop matching
 at STOP.
 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
 store offsets for the substring each group matched in REGS.  See the
 documentation for exactly how many groups we fill.
 We return -1 if no match, -2 if an internal error (such as the
 each to consider matching.  */
 CONST char *end_match_1, *end_match_2;
 /* Where we are in the data, and the end of the current string.  */
 CONST char *d, *dend;
 /* Where we are in the pattern, and the end of the pattern.  */
 unsigned char *p = bufp->buffer;
 register unsigned char *pend = p + bufp->used;
 /* Mark the opcode just after a start_memory, so we can test for an
 /* We fill all the registers internally, independent of what we
 return, for use in backreferences.  The number here includes
 an element for register zero.  */
 unsigned num_regs = bufp->re_nsub + 1;
 /* The currently active registers.  */
 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
 /* Information on the contents of registers. These are pointers into
 field of reg_info[reg_num] helps us tell whether or not we have
 matched any of the pattern so far this time through the reg_num-th
 subexpression.  These two fields get reset each time through any
 loop their register is in.  */
 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global.  */
 register_info_type *reg_info;
 #endif
 /* The following record the register info as found in the above
 variables when we find a match better than any we've seen before.
 This happens as we backtrack through the failure points, which in
 turn happens only if we have not yet matched the entire string. */
 unsigned best_regs_set = false;
 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global.  */
 CONST char **best_regstart, **best_regend;
 #endif
 /* Logically, this is `best_regend[0]'.  But we don't want to have to
 allocate space for that if we're not allocating space for anything
 else (see below).  Also, we never need info about register 0 for
 any of the other register vectors, and it seems rather a kludge to
 treat `best_regend' differently than the rest.  So we keep track of
 register_info_type *reg_info_dummy;
 #endif
 #ifdef DEBUG
 /* Counts the total number of registers pushed.  */
 unsigned num_regs_pushed = 0;
 #endif
 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
 INIT_FAIL_STACK ();
 #ifdef MATCH_MAY_ALLOCATE
 /* Do not bother to initialize all the register variables if there are
 no groups in the pattern, as it takes a fair amount of time.  If
 there are groups, we include space for register 0 (the whole
 pattern), even though we never use it, since it simplifies the
 best_regend = REGEX_TALLOC (num_regs, CONST char *);
 reg_info = REGEX_TALLOC (num_regs, register_info_type);
 reg_dummy = REGEX_TALLOC (num_regs, CONST char *);
 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
 if (!(regstart && regend && old_regstart && old_regend && reg_info
 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
 {
 FREE_VARIABLES ();
 return -2;
 }
 }
 if (pos < 0 || pos > size1 + size2)
 {
 FREE_VARIABLES ();
 return -1;
 }
 /* Initialize subexpression text positions to -1 to mark ones that no
 start_memory/stop_memory has been seen for. Also initialize the
 register information struct.  */
 for (mcnt = 1; mcnt < num_regs; mcnt++)
 {
 regstart[mcnt] = regend[mcnt]
 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
 IS_ACTIVE (reg_info[mcnt]) = 0;
 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
 }
 /* We move `string1' into `string2' if the latter's empty -- but not if
 `string1' is null.  */
 if (size2 == 0 && string1 != NULL)
 {
 string2 = string1;
 {
 end_match_1 = end1;
 end_match_2 = string2 + stop - size1;
 }
 /* `p' scans through the pattern as `d' scans through the data.
 `dend' is the end of the input string that `d' points within.  `d'
 is advanced into the following input string whenever necessary, but
 this happens before fetching; therefore, at the beginning of the
 loop, `d' can be pointing at the end of a string, but it cannot
 equal `string2'.  */
 DEBUG_PRINT1 ("The compiled pattern is: ");
 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
 DEBUG_PRINT1 ("The string to match is: `");
 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
 DEBUG_PRINT1 ("'\n");
 /* This loops over pattern commands.  It exits by returning from the
 function if the match is complete, or it drops through if the match
 fails at this starting point in the input data.  */
 for (;;)
 {
 DEBUG_PRINT2 ("\n0x%p: ", p);
 if (p == pend)
 	{ /* End of pattern means we might have succeeded.  */
 DEBUG_PRINT1 ("end of pattern ... ");
 	  /* If we haven't matched the entire string, and we want the
 longest match, try backtracking.  */
 if (d != end_match_2)
 	    {
 	      /* 1 if this match ends in the same string (string1 or string2)
 		 as the best previous match.  */
 	      boolean same_str_p = (FIRST_STRING_P (match_end)
 				    == MATCHING_IN_FIRST_STRING);
 	      /* 1 if this match is the best seen so far.  */
 	      boolean best_match_p;
 	      /* AIX compiler got confused when this was combined
 		best_match_p = d > match_end;
 	      else
 		best_match_p = !MATCHING_IN_FIRST_STRING;
 DEBUG_PRINT1 ("backtracking.\n");
 if (!FAIL_STACK_EMPTY ())
 { /* More failure points to try.  */
 /* If exceeds best match so far, save it.  */
 if (!best_regs_set || best_match_p)
 {
 best_regs_set = true;
 match_end = d;
 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
 for (mcnt = 1; mcnt < num_regs; mcnt++)
 {
 best_regstart[mcnt] = regstart[mcnt];
 best_regend[mcnt] = regend[mcnt];
 }
 }
 goto fail;
 }
 /* If no failure points, don't restore garbage.  And if
 last match is real best match, don't restore second
 best one. */
 end_match_1' while the restored d is in string2.
 For example, the pattern `x.*y.*z' against the
 strings `x-' and `y-z-', if the two strings are
 not consecutive in memory.  */
 DEBUG_PRINT1 ("Restoring best registers.\n");
 d = match_end;
 dend = ((d >= string1 && d <= end1)
 		           ? end_match_1 : end_match_2);
 		  for (mcnt = 1; mcnt < num_regs; mcnt++)
 regs->start[0] = pos;
 regs->end[0] = (MATCHING_IN_FIRST_STRING
 				  ? ((regoff_t) (d - string1))
 			          : ((regoff_t) (d - string2 + size1)));
 }
 /* Go through the first `min (num_regs, regs->num_regs)'
 registers, since that is all we initialized.  */
 	      for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
 		{
 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
 			= (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
 regs->end[mcnt]
 			= (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
 }
 		}
 /* If the regs structure we return has more elements than
 were in the pattern, set the extra elements to -1.  If
 we (re)allocated the registers, this is the case,
 because we always allocate enough to have at least one
 -1 at the end.  */
 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
 nfailure_points_pushed, nfailure_points_popped,
 nfailure_points_pushed - nfailure_points_popped);
 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
 			    ? string1
 			    : string2 - size1);
 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
 FREE_VARIABLES ();
 	      not = !not;
 	    p += 1 + *p;
 	    if (!not) goto fail;
 	    SET_REGS_MATCHED ();
 INC_CHARPTR (d); /* XEmacs change */
 	    break;
 	  }
 	    if (EQ (Qt, unified_range_table_lookup (p, c, Qnil)))
 	      not = !not;
 	    p += unified_range_table_bytes_used (p);
 	    if (!not) goto fail;
 	    SET_REGS_MATCHED ();
 	    INC_CHARPTR (d);
 	    break;
 	  }
 #endif
 case start_memory:
 	  DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
 /* Find out if this group can match the empty string.  */
 	  p1 = p;		/* To send to group_match_null_string_p.  */
 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
 REG_MATCH_NULL_STRING_P (reg_info[*p])
 = group_match_null_string_p (&p1, pend, reg_info);
 /* Save the position in the string where we were the last time
 we were at this open-group operator in case the group is
 operated upon by a repetition operator, e.g., with `(a*)*b'
 against `ab'; then we want to ignore where we are now in
 the string in case this attempt to match fails.  */
 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
 : regstart[*p];
 	  DEBUG_PRINT2 ("  old_regstart: %d\n",
 			 POINTER_TO_OFFSET (old_regstart[*p]));
 regstart[*p] = d;
 	  DEBUG_PRINT2 ("  regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
 IS_ACTIVE (reg_info[*p]) = 1;
 MATCHED_SOMETHING (reg_info[*p]) = 0;
 	  /* Clear this whenever we change the register activity status.  */
 	  set_regs_matched_done = 0;
 /* This is the new highest active register.  */
 highest_active_reg = *p;
 /* If nothing was active before, this is the new lowest active
 register.  */
 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
 lowest_active_reg = *p;
 /* The stop_memory opcode represents the end of a group.  Its
 arguments are the same as start_memory's: the register
 number, and the number of inner groups.  */
 	case stop_memory:
 	  DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
 /* We need to save the string position the last time we were at
 this close-group operator in case the group is operated
 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
 against `aba'; then we want to ignore where we are now in
 the string in case this attempt to match fails.  */
 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
 ? REG_UNSET (regend[*p]) ? d : regend[*p]
 			   : regend[*p];
 	  DEBUG_PRINT2 ("      old_regend: %d\n",
 			 POINTER_TO_OFFSET (old_regend[*p]));
 regend[*p] = d;
 	  DEBUG_PRINT2 ("      regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
 (a(b)c(d(e)f)g).  When group 3 ends, after the f), the
 new highest active register is 1.  */
 unsigned char r = *p - 1;
 while (r > 0 && !IS_ACTIVE (reg_info[r]))
 r--;
 /* If we end up at register zero, that means that we saved
 the registers as the result of an `on_failure_jump', not
 a `start_memory', and we jumped to past the innermost
 `stop_memory'.  For example, in ((.)*) we save
 registers 1 and 2 as a result of the *, but when we pop
 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
 }
 else
 highest_active_reg = r;
 }
 /* If just failed to match something this time around with a
 group that's operated on by a repetition operator, try to
 force exit from the ``loop'', and restore the register
 information for this group that we had before trying this
 last match.  */
 if ((!MATCHED_SOMETHING (reg_info[*p])
 || just_past_start_mem == p - 1)
 	      && (p + 2) < pend)
 {
 boolean is_a_jump_n = false;
 p1 = p + 2;
 mcnt = 0;
 switch ((re_opcode_t) *p1++)
 {
 case jump_n:
 		  case dummy_failure_jump:
 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
 		    if (is_a_jump_n)
 		      p1 += 2;
 break;
 default:
 /* do nothing */ ;
 }
 	      p1 += mcnt;
 /* If the next operation is a jump backwards in the pattern
 	         to an on_failure_jump right before the start_memory
 corresponding to this stop_memory, exit from the loop
 by forcing a failure after pushing on the stack the
 on_failure_jump's jump in the pattern, and d.  */
 /* If this group ever matched anything, then restore
 what its registers were before trying this last
 failed match, e.g., with `(a*)*b' against `ab' for
 regstart[1], and, e.g., with `((a*)*(b*)*)*'
 against `aba' for regend[3].
 Also restore the registers for inner groups for,
 e.g., `((a*)(b*))*' against `aba' (register 3 would
 otherwise get trashed).  */
 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
 		    {
 		      unsigned r;
 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
 		      /* Restore this and inner groups' (if any) registers.  */
 for (r = *p; r < *p + *(p + 1); r++)
 {
 regstart[r] = old_regstart[r];
 /* xx why this test?  */
 if (old_regend[r] >= regstart[r])
 regend[r] = old_regend[r];
 }
 }
 		  p1++;
 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
 goto fail;
 }
 }
 /* Move past the register number and the inner group count.  */
 p += 2;
 break;
 	    DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
 	    /* Can't back reference a group which we've never matched.  */
 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
 goto fail;
 /* Where in input to try to start matching.  */
 d2 = regstart[regno];
 /* Where to stop matching; if both the place to start and
 the place to stop matching are in the same string, then
 set to the place to stop, otherwise, for now have to use
 the end of the first string.  */
 dend2 = ((FIRST_STRING_P (regstart[regno])
 		      == FIRST_STRING_P (regend[regno]))
 		     ? regend[regno] : end_match_1);
 	    for (;;)
 	      {
 		/* If necessary, advance to next segment in register
 		/* If necessary, advance to next segment in data.  */
 		PREFETCH ();
 		/* How many characters left in this segment to match.  */
 		mcnt = dend - d;
 		/* Want how many consecutive characters we can match in
 one shot, so, if necessary, adjust the count.  */
 if (mcnt > dend2 - d2)
 		  mcnt = dend2 - d2;
 		/* Compare that many; failure if mismatch, else move
 past them.  */
 		if (translate
 ? bcmp_translate ((unsigned char *) d,
 				      (unsigned char *) d2, mcnt, translate)
 : memcmp (d, d2, mcnt))
 		  goto fail;
 		d += mcnt, d2 += mcnt;
 		/* Do this because we've match some characters.  */
 /* begline matches the empty string at the beginning of the string
 (unless `not_bol' is set in `bufp'), and, if
 `newline_anchor' is set, after newlines.  */
 	case begline:
 DEBUG_PRINT1 ("EXECUTING begline.\n");
 if (AT_STRINGS_BEG (d))
 {
 if (!bufp->not_bol) break;
 }
 else if (d[-1] == '\n' && bufp->newline_anchor)
 if (AT_STRINGS_END (d))
 {
 if (!bufp->not_eol) break;
 }
 /* We have to ``prefetch'' the next character.  */
 else if ((d == end1 ? *string2 : *d) == '\n'
 && bufp->newline_anchor)
 {
 break;
 string, instead of restoring it.  To see why, consider
 matching `foo\nbar' against `.*\n'.  The .* matches the foo;
 then the . fails against the \n.  But the next thing we want
 to do is match the \n against the \n; if we restored the
 string value, we would be back at the foo.
 Because this is used only in specific cases, we don't need to
 check all the things that `on_failure_jump' does, to make
 sure the right things get saved on the stack.  Hence we don't
 share its code.  The only reason to push anything on the
 stack at all is that otherwise we would have to change
 `anychar's code to do something besides goto fail in this
 case; that seems worse than this.  */
 case on_failure_keep_string_jump:
 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
 EXTRACT_NUMBER_AND_INCR (mcnt, p);
 DEBUG_PRINT3 (" %d (to 0x%p):\n", mcnt, p + mcnt);
 PUSH_FAILURE_POINT (p + mcnt, (void *) 0, -2);
 break;
 	/* Uses of on_failure_jump:
 Each alternative starts with an on_failure_jump that points
 to the beginning of the next alternative.  Each alternative
 except the last ends with a jump that in effect jumps past
 the rest of the alternatives.  (They really jump to the
 ending jump of the following alternative, because tensioning
 pattern follows its end. If we can establish that there
 is nothing that they would both match, i.e., that we
 would have to backtrack because of (as in, e.g., `a*a')
 then we can change to pop_failure_jump, because we'll
 never have to backtrack.
 This is not true in the case of alternatives: in
 `(a|ab)*' we do need to backtrack to the `ab' alternative
 (e.g., if the string was `ab').  But instead of trying to
 detect that here, the alternative has put on a dummy
 failure point which is what we will end up popping.  */
 		  break;
 	      }
 	    p1 = p + mcnt;
 	    /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
 	       to the `maybe_finalize_jump' of this case.  Examine what
 	       follows.  */
 /* If we're at the end of the pattern, we can change.  */
 if (p2 == pend)
 	      {
 {
 		    p[-3] = (unsigned char) pop_failure_jump;
 DEBUG_PRINT3 ("  %c != %c => pop_failure_jump.\n",
 c, p1[5]);
 }
 		else if ((re_opcode_t) p1[3] == charset
 			 || (re_opcode_t) p1[3] == charset_not)
 		  {
 		    int not = (re_opcode_t) p1[3] == charset_not;
 		    if (c < (unsigned char) (p1[4] * BYTEWIDTH)
 			&& p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
 		      not = !not;
 /* `not' is equal to 1 if c would match, which means
 {
 		    p[-3] = (unsigned char) pop_failure_jump;
 DEBUG_PRINT3 ("  %c != %c => pop_failure_jump.\n",
 c, p1[5]);
 }
 		else if ((re_opcode_t) p1[3] == charset_not)
 		  {
 		    int idx;
 		    /* We win if the charset_not inside the loop
 		       lists every character listed in the charset after.  */
 dummy_low_reg, dummy_high_reg,
 reg_dummy, reg_dummy, reg_info_dummy);
 }
 /* Note fall through.  */
 /* Unconditionally jump (without popping any failure points).  */
 case jump:
 	unconditional_jump:
 	  EXTRACT_NUMBER_AND_INCR (mcnt, p);	/* Get the amount to jump.  */
 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
 	  p += mcnt;				/* Do the jump.  */
 DEBUG_PRINT2 ("(to 0x%p).\n", p);
 	  break;
 /* We need this opcode so we can detect where alternatives end
 in `group_match_null_string_p' et al.  */
 case jump_past_alt:
 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
 goto unconditional_jump;
 PUSH_FAILURE_POINT ((void *) 0, (void *) 0, -2);
 break;
 /* Have to succeed matching what follows at least n times.
 After that, handle like `on_failure_jump'.  */
 case succeed_n:
 EXTRACT_NUMBER (mcnt, p + 2);
 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
 assert (mcnt >= 0);
 /* Originally, this is how many times we HAVE to succeed.  */
 	      p[2] = (unsigned char) no_op;
 p[3] = (unsigned char) no_op;
 goto on_failure;
 }
 break;
 case jump_n:
 EXTRACT_NUMBER (mcnt, p + 2);
 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
 /* Originally, this is how many times we CAN jump.  */
 if (mcnt)
 {
 mcnt--;
 STORE_NUMBER (p + 2, mcnt);
 	       goto unconditional_jump;
 }
 /* If don't have to jump any more, skip over the rest of command.  */
 	  else
 	    p += 4;
 break;
 	case set_number_at:
 	  {
 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
 EXTRACT_NUMBER_AND_INCR (mcnt, p);
 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
 	  if (BUF_PTR_BYTE_POS (regex_emacs_buffer, (unsigned char *) d) >=
 	      BUF_PT (regex_emacs_buffer))
 	    goto fail;
 	  break;
 	case at_dot:
 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
 	  if (BUF_PTR_BYTE_POS (regex_emacs_buffer, (unsigned char *) d)
 	      != BUF_PT (regex_emacs_buffer))
 	    goto fail;
 	  break;
 	case after_dot:
 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
 if (BUF_PTR_BYTE_POS (regex_emacs_buffer, (unsigned char *) d)
 	      <= BUF_PT (regex_emacs_buffer))
 	    goto fail;
 				    mcnt, should_succeed))
 	      goto fail;
 	    SET_REGS_MATCHED ();
 	  }
 	  break;
 	case notcategoryspec:
 	  should_succeed = 0;
 	  goto matchornotcategory;
 /* end of category patch */
 #endif /* MULE */
 if (!WORDCHAR_P_UNSAFE ((int) (*d)))
 goto fail;
 	  SET_REGS_MATCHED ();
 d++;
 	  break;
 	case notwordchar:
 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
 	  PREFETCH ();
 if (!WORDCHAR_P_UNSAFE ((int) (*d)))
 goto fail;
 SET_REGS_MATCHED ();
 d++;
 	  break;
 #endif /* not emacs */
 default:
 abort ();
 	}
 continue;  /* Successfully executed one pattern command; keep going.  */
 /* If we failed to the end of the pattern, don't examine *p.  */
 	  assert (p <= pend);
 if (p < pend)
 {
 boolean is_a_jump_n = false;
 /* If failed to a backwards jump that's part of a repetition
 loop, need to pop this failure point and use the next one.  */
 switch ((re_opcode_t) *p)
 {
 case jump_n:
 case maybe_pop_jump:
 case pop_failure_jump:
 case jump:
 p1 = p + 1;
 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
 p1 += mcnt;
 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
 || (!is_a_jump_n
 && (re_opcode_t) *p1 == on_failure_jump))
 goto fail;
 /* Subroutine definitions for re_match_2.  */
 /* We are passed P pointing to a register number after a start_memory.
 Return true if the pattern up to the corresponding stop_memory can
 match the empty string, and false otherwise.
 If we find the matching stop_memory, sets P to point to one past its number.
 Otherwise, sets P to an undefined byte less than or equal to END.
 We don't handle duplicates properly (yet).  */
 			   register_info_type *reg_info)
 {
 int mcnt;
 /* Point to after the args to the start_memory.  */
 unsigned char *p1 = *p + 2;
 while (p1 < end)
 {
 /* Skip over opcodes that can match nothing, and return true or
 	 false, as appropriate, when we get to one that can't, or to the
 matching stop_memory.  */
 switch ((re_opcode_t) *p1)
 {
 /* Could be either a loop or a series of alternatives.  */
 case on_failure_jump:
 p1++;
 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
 /* If the next operation is not a jump backwards in the
 	     pattern.  */
 	  if (mcnt >= 0)
 	    {
 whereas the rest start with on_failure_jump and end
 with a jump, e.g., here is the pattern for `a|b|c':
 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
 /exactn/1/c
 So, we have to first go through the first (n-1)
 alternatives and then deal with the last one separately.  */
 {
 /* `mcnt' holds how many bytes long the alternative
 is, including the ending `jump_past_alt' and
 its number.  */
 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
 				                      reg_info))
 return false;
 /* Move to right after this alternative, including the
 		     jump_past_alt.  */
 p1 += mcnt;
 /* Break if it's the beginning of an n-th alternative
 that doesn't begin with an on_failure_jump.  */
 if ((re_opcode_t) *p1 != on_failure_jump)
 break;
 		  /* Still have to check that it's not an n-th
 		     alternative that starts with an on_failure_jump.  */
 		  p1++;
 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
 p1 += mcnt;	/* Get past the n-th alternative.  */
 } /* if mcnt > 0 */
 break;
 case stop_memory:
 	  assert (p1[1] == **p);
 *p = p1 + 2;
 return true;
 default:
 if (!common_op_match_null_string_p (&p1, end, reg_info))
 return false;
 }
 } /* while p1 < end */
 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
 It expects P to be the first byte of a single alternative and END one
 byte past the last. The alternative can contain groups.  */
 static boolean
 alt_match_null_string_p (unsigned char *p, unsigned char *end,
 			 register_info_type *reg_info)
 {
 int mcnt;
 unsigned char *p1 = p;
 while (p1 < end)
 {
 /* Skip over opcodes that can match nothing, and break when we get
 to one that can't.  */
 switch ((re_opcode_t) *p1)
 {
 	/* It's a loop.  */
 case on_failure_jump:
 p1++;
 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
 p1 += mcnt;
 break;
 	default:
 if (!common_op_match_null_string_p (&p1, end, reg_info))
 return false;
 }
 }  /* while p1 < end */
 return true;
 } /* alt_match_null_string_p */
 /* Deals with the ops common to group_match_null_string_p and
 alt_match_null_string_p.
 Sets P to one after the op and its arguments, if any.  */
 static boolean
 common_op_match_null_string_p (unsigned char **p, unsigned char *end,
 			       register_info_type *reg_info)
 case start_memory:
 reg_no = *p1;
 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
 ret = group_match_null_string_p (&p1, end, reg_info);
 /* Have to set this here in case we're checking a group which
 contains a group and a back reference to it.  */
 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
 if (!ret)
 return false;
 break;
 /* If this is an optimized succeed_n for zero times, make the jump.  */
 case jump:
 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
 if (mcnt >= 0)
 p1 += mcnt;
 return false;
 break;
 case succeed_n:
 /* Get to the number of times to succeed.  */
 p1 += 2;
 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
 if (mcnt == 0)
 {
 p1 -= 4;
 }
 else
 return false;
 break;
 case duplicate:
 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
 return false;
 break;
 case set_number_at:
 } /* common_op_match_null_string_p */
 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
 bytes; nonzero otherwise.  */
 static int
 bcmp_translate (CONST unsigned char *s1, CONST unsigned char *s2,
 		register int len, char *translate)
 {
 register CONST unsigned char *p1 = s1, *p2 = s2;
 /* Entry points for GNU code.  */
 /* re_compile_pattern is the GNU regular expression compiler: it
 compiles PATTERN (of length SIZE) and puts the result in BUFP.
 Returns 0 if the pattern was valid, otherwise an error string.
 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
 are set in BUFP on entry.
 We call regex_compile to do the actual compilation.  */
 CONST char *
 re_compile_pattern (CONST char *pattern, int length,
 		    struct re_pattern_buffer *bufp)
 {
 reg_errcode_t ret;
 /* GNU code is written to assume at least RE_NREGS registers will be set
 (and at least one extra will be -1).  */
 bufp->regs_allocated = REGS_UNALLOCATED;
 /* And GNU code determines whether or not to get register information
 by passing null for the REGS argument to re_match, etc., not by
 setting no_sub.  */
 bufp->no_sub = 0;
 /* Match anchors at newline.  */
 bufp->newline_anchor = 1;
 ret = regex_compile (pattern, length, re_syntax_options, bufp);
 if (!ret)
 return NULL;
 return gettext (re_error_msgid[(int) ret]);
 }
 /* Entry points compatible with 4.2 BSD regex library.  We don't define
 them unless specifically requested.  */
 #ifdef _REGEX_RE_COMP
 char *
 re_comp (CONST char *s)
 {
 reg_errcode_t ret;
 if (!s)
 {
 if (!re_comp_buf.buffer)
 	return gettext ("No previous regular expression");
 return 0;
 /* Match anchors at newlines.  */
 re_comp_buf.newline_anchor = 1;
 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
 if (!ret)
 return NULL;
 /* Yes, we're discarding `CONST' here if !HAVE_LIBINTL.  */
 return (char *) gettext (re_error_msgid[(int) ret]);
 /* regex_compile will allocate the space for the compiled pattern.  */
 preg->buffer = 0;
 preg->allocated = 0;
 preg->used = 0;
 /* Don't bother to use a fastmap when searching.  This simplifies the
 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
 characters after newlines into the fastmap.  This way, we just try
 every character.  */
 preg->fastmap = 0;
 if (cflags & REG_ICASE)
 {
 unsigned i;
 preg->translate = (char *) malloc (CHAR_SET_SIZE);
 if (preg->translate == NULL)
 return (int) REG_ESPACE;
 /* Map uppercase characters to corresponding lowercase ones.  */
 else
 preg->newline_anchor = 0;
 preg->no_sub = !!(cflags & REG_NOSUB);
 /* POSIX says a null character in the pattern terminates it, so we
 can use strlen here in compiling the pattern.  */
 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
 /* POSIX doesn't distinguish between an unmatched open-group and an
 unmatched close-group: both are REG_EPAREN.  */
 if (ret == REG_ERPAREN) ret = REG_EPAREN;
 return (int) ret;
 }
 /* regexec searches for a given pattern, specified by PREG, in the
 string STRING.
 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
 `regcomp', we ignore PMATCH.  Otherwise, we assume PMATCH has at
 least NMATCH elements, and we set them to the offsets of the
 corresponding matched substrings.
 EFLAGS specifies `execution flags' which affect matching: if
 REG_NOTBOL is set, then ^ does not match at the beginning of the
 string; if REG_NOTEOL is set, then $ does not match at the end.
 We return 0 if we find a match and REG_NOMATCH if not.  */
 int
 regexec (CONST regex_t *preg, CONST char *string, size_t nmatch,
 	 regmatch_t pmatch[], int eflags)
 regex_t private_preg;
 int len = strlen (string);
 boolean want_reg_info = !preg->no_sub && nmatch > 0;
 private_preg = *preg;
 private_preg.not_bol = !!(eflags & REG_NOTBOL);
 private_preg.not_eol = !!(eflags & REG_NOTEOL);
 /* The user has told us exactly how many registers to return
 information about, via `nmatch'.  We have to pass that on to the
 matching routines.  */
 private_preg.regs_allocated = REGS_FIXED;
 if (want_reg_info)
 {
 regs.num_regs = nmatch;
 regs.start = TALLOC (nmatch, regoff_t);
 regs.end = TALLOC (nmatch, regoff_t);
 /* Perform the searching operation.  */
 ret = re_search (&private_preg, string, len,
 /* start: */ 0, /* range: */ len,
 want_reg_info ? &regs : (struct re_registers *) 0);
 /* Copy the register information to the POSIX structure.  */
 if (want_reg_info)
 {
 if (ret >= 0)
 {
 CONST char *msg;
 size_t msg_size;
 if (errcode < 0
 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
 /* Only error codes returned by the rest of the code should be passed
 to this routine.  If we are given anything else, or if other regex
 code generates an invalid error code, then the program has a bug.
 Dump core so we can fix it.  */
 abort ();
 msg = gettext (re_error_msgid[errcode]);
 msg_size = strlen (msg) + 1; /* Includes the null.  */
 if (errbuf_size != 0)
 {
 if (msg_size > errbuf_size)
 {
 strncpy (errbuf, msg, errbuf_size - 1);
 regfree (regex_t *preg)
 {
 if (preg->buffer != NULL)
 free (preg->buffer);
 preg->buffer = NULL;
 preg->allocated = 0;
 preg->used = 0;
 if (preg->fastmap != NULL)
 free (preg->fastmap);

Mercurial > hg > xemacs-beta

comparison src/regex.c @ 183:e121b013d1f0 r20-3b18