comparison src/regex.c @ 83548:c71725faff1a

Merged from [e-mail]. Last-minute emacsclient rewrites be damned! Patches applied: * [e-mail]/emacs--devo--0--patch-490 Update from CVS * [e-mail]/emacs--devo--0--patch-491 Update from CVS * [e-mail]/emacs--devo--0--patch-492 Update from CVS * [e-mail]/emacs--devo--0--patch-493 Update from CVS * [e-mail]/emacs--devo--0--patch-494 Update from CVS * [e-mail]/emacs--devo--0--patch-495 Merge from gnus--rel--5.10 * [e-mail]/emacs--devo--0--patch-496 Update from CVS * [e-mail]/emacs--devo--0--patch-497 Update from CVS * [e-mail]/emacs--devo--0--patch-498 Update from CVS * [e-mail]/emacs--devo--0--patch-499 Update from CVS * [e-mail]/emacs--devo--0--patch-500 Update from CVS * [e-mail]/emacs--devo--0--patch-501 Update from CVS * [e-mail]/emacs--devo--0--patch-502 Update from CVS * [e-mail]/emacs--devo--0--patch-503 Update from CVS * [e-mail]/emacs--devo--0--patch-504 Merge from gnus--rel--5.10 * [e-mail]/emacs--devo--0--patch-505 Update from CVS * [e-mail]/emacs--devo--0--patch-506 Update from CVS * [e-mail]/emacs--devo--0--patch-507 Update from CVS * [e-mail]/emacs--devo--0--patch-508 Update from CVS * [e-mail]/emacs--devo--0--patch-509 Update from CVS * [e-mail]/emacs--devo--0--patch-510 Update from CVS * [e-mail]/emacs--devo--0--patch-511 Update from CVS * [e-mail]/emacs--devo--0--patch-512 Update from CVS * [e-mail]/emacs--devo--0--patch-513 Update from CVS * [e-mail]/emacs--devo--0--patch-514 Update from CVS * [e-mail]/emacs--devo--0--patch-515 Merge from gnus--rel--5.10 * [e-mail]/emacs--devo--0--patch-516 Update from CVS * [e-mail]/emacs--devo--0--patch-517 Update from CVS * [e-mail]/emacs--devo--0--patch-518 Merge from gnus--rel--5.10 * [e-mail]/emacs--devo--0--patch-519 Update from CVS: etc/TUTORIAL.cn: Updated. * [e-mail]/emacs--devo--0--patch-520 Merge from erc--emacs--22 * [e-mail]/emacs--devo--0--patch-521 Update from CVS * [e-mail]/emacs--devo--0--patch-522 Update from CVS * [e-mail]/emacs--devo--0--patch-523 Update from CVS * [e-mail]/emacs--devo--0--patch-524 Update from CVS * [e-mail]/emacs--devo--0--patch-525 Update from CVS * [e-mail]/emacs--devo--0--patch-526 Update from CVS * [e-mail]/emacs--devo--0--patch-527 Merge from gnus--rel--5.10 * [e-mail]/emacs--devo--0--patch-528 Update from CVS * [e-mail]/emacs--devo--0--patch-529 Update from CVS * [e-mail]/emacs--devo--0--patch-530 Update from CVS * [e-mail]/emacs--devo--0--patch-531 Update from CVS * [e-mail]/emacs--devo--0--patch-532 Update from CVS * [e-mail]/emacs--devo--0--patch-533 Update from CVS * [e-mail]/emacs--devo--0--patch-534 Merge from gnus--rel--5.10 * [e-mail]/emacs--devo--0--patch-535 Update from CVS * [e-mail]/gnus--rel--5.10--patch-161 Update from CVS * [e-mail]/gnus--rel--5.10--patch-162 Update from CVS * [e-mail]/gnus--rel--5.10--patch-163 Merge from emacs--devo--0 * [e-mail]/gnus--rel--5.10--patch-164 Update from CVS * [e-mail]/gnus--rel--5.10--patch-165 Update from CVS * [e-mail]/gnus--rel--5.10--patch-166 Update from CVS * [e-mail]/gnus--rel--5.10--patch-167 Update from CVS * [e-mail]/gnus--rel--5.10--patch-168 Update from CVS * [e-mail]/gnus--rel--5.10--patch-169 Merge from emacs--devo--0 * [e-mail]/gnus--rel--5.10--patch-170 Update from CVS git-archimport-id: [e-mail]--2004/emacs--multi-tty--0--patch-588

83547:0912b745fc75

   the Free Software Foundation; either version 2, or (at your option)
   any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.	 See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
   USA.	 */

/* TODO:
   - structure the opcode space into opcode+flag.
   - merge with glibc's regex.[ch].
   - replace (succeed_n + jump_n + set_number_at) with something that doesn't

# ifdef free
#  undef free
# endif
# define free xfree

/* Converts the pointer to the char to BEG-based offset from the start.	 */
# define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
# define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))

# define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
# define RE_STRING_CHAR(p, s) \

#  define REGEX_ALLOCATE_STACK alloca

#  define REGEX_REALLOCATE_STACK(source, osize, nsize)			\
   REGEX_REALLOCATE (source, osize, nsize)
/* No need to explicitly free anything.	 */
#  define REGEX_FREE_STACK(arg) ((void)0)

# endif /* not REGEX_MALLOC */
#endif /* not using relocating allocator */

	/* Followed by two byte relative address to which to jump.  */
  jump,

	/* 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,

	/* Followed by two-byte relative address, and two-byte number n.
	   Jump to the address N times, then fail.  */
  jump_n,

	/* Set the following two-byte relative address to the
	   subsequent two-byte number.	The address *includes* the two
	   bytes of number.  */
  set_number_at,

  wordbeg,	/* Succeeds if at word beginning.  */
  wordend,	/* Succeeds if at word end.  */

  wordbound,	/* Succeeds if at a word boundary.  */
  notwordbound,	/* Succeeds if not at a word boundary.	*/

  symbeg,       /* Succeeds if at symbol beginning.  */
  symend,       /* Succeeds if at symbol end.  */

	/* Matches any character whose syntax is specified.  Followed by

  ,before_dot,	/* Succeeds if before point.  */
  at_dot,	/* Succeeds if at point.  */
  after_dot,	/* Succeeds if after point.  */

  /* Matches any character whose category-set contains the specified
     category.	The operator is followed by a byte which contains a
     category code (mnemonic ASCII character).	*/
  categoryspec,

  /* Matches any character whose category-set does not contain the
     specified category.  The operator is followed by a byte which
     contains the category code (mnemonic ASCII character).  */

#endif /* DEBUG */

/* Store a multibyte character in three contiguous bytes starting
   DESTINATION, and increment DESTINATION to the byte after where the
   character is stored.	 Therefore, DESTINATION must be an lvalue.  */

#define STORE_CHARACTER_AND_INCR(destination, character)	\
  do {								\
    (destination)[0] = (character) & 0377;			\
    (destination)[1] = ((character) >> 8) & 0377;		\
    (destination)[2] = (character) >> 16;			\
    (destination) += 3;						\
  } while (0)

/* Put into DESTINATION a character stored in three contiguous bytes
   starting at SOURCE.	*/

#define EXTRACT_CHARACTER(destination, source)	\
  do {						\
    (destination) = ((source)[0]		\
		     | ((source)[1] << 8)	\

#define CHARSET_LOOKUP_BITMAP(p, c)				\
  ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH			\
   && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))

/* Return the address of end of RANGE_TABLE.  COUNT is number of
   ranges (which is a pair of (start, end)) in the RANGE_TABLE.	 `* 2'
   is start of range and end of range.	`* 3' is size of each start
   and end.  */
#define CHARSET_RANGE_TABLE_END(range_table, count)	\
  ((range_table) + (count) * 2 * 3)

/* Test if C is in RANGE_TABLE.	 A flag NOT is negated if C is in.
   COUNT is number of ranges in RANGE_TABLE.  */
#define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count)	\
  do									\
    {									\
      re_wchar_t range_start, range_end;				\

     STR -- the saved data position.
     PAT -- the saved pattern position.
     REGSTART, REGEND -- arrays of string positions.

   Also assumes the variables `fail_stack' and (if debugging), `bufp',
   `pend', `string1', `size1', `string2', and `size2'.	*/

#define POP_FAILURE_POINT(str, pat)                                     \
do {									\
  assert (!FAIL_STACK_EMPTY ());					\
									\

/* Structure to manage work area for range table.  */
struct range_table_work_area
{
  int *table;			/* actual work area.  */
  int allocated;		/* allocated size for work area in bytes.  */
  int used;			/* actually used size in words.	 */
  int bits;			/* flag to record character classes */
};

/* Make sure that WORK_AREA can hold more N multibyte characters.
   This is used only in set_image_of_range and set_image_of_range_1.

    tem = set_image_of_range (&work_area, start, end, translate);	\
    if (tem > 0)							\
      FREE_STACK_RETURN (tem);						\
  } while (0)

/* Free allocated memory for WORK_AREA.	 */
#define FREE_RANGE_TABLE_WORK_AREA(work_area)	\
  do {						\
    if ((work_area).table)			\
      free ((work_area).table);			\
  } while (0)

  /* Place in the uncompiled pattern (i.e., the {) to
     which to go back if the interval is invalid.  */
  re_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;

  /* Counts open-groups as they are encountered.  Remembered for the
     matching close-group on the compile stack, so the same register

  if (bufp->allocated == 0)
    {
      if (bufp->buffer)
	{ /* If zero allocated, but buffer is non-null, try to realloc
	     enough space.  This loses if buffer's address is bogus, but
	     that is the user's responsibility.	 */
	  RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
	}
      else
	{ /* Caller did not allocate a buffer.	Do it for them.	 */
	  bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
	}
      if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);

      bufp->allocated = INIT_BUF_SIZE;

	    main_pend = pend;
	    main_pattern = pattern;
	    p = pattern = whitespace_regexp;
	    pend = p + strlen (p);
	    break;
	  }    

	case '^':
	  {
	    if (   /* If at start of pattern, it's an operator.	 */
		   p == pattern + 1
		   /* If context independent, it's an operator.	 */
		|| syntax & RE_CONTEXT_INDEP_ANCHORS
		   /* Otherwise, depends on what's come before.	 */
		|| at_begline_loc_p (pattern, p, syntax))
	      BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
	    else
	      goto normal_char;
	  }

	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 ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
	     else

	      else if (!(syntax & RE_CONTEXT_INDEP_OPS))
		goto normal_char;
	    }

	  {
	    /* 1 means zero (many) matches is allowed.	*/
	    boolean zero_times_ok = 0, many_times_ok = 0;
	    boolean greedy = 1;

	    /* If there is a sequence of repetition chars, collapse it
	       down to just one (the right one).  We can't combine
	       interval operators with these because of, e.g., `a{2}*',
	       which should only match an even number of `a's.	*/

	    for (;;)
	      {
		if ((syntax & RE_FRUGAL)
		    && c == '?' && (zero_times_ok || many_times_ok))

	       to an empty pattern.  */
	    if (!laststart || laststart == b)
	      break;

	    /* Now we know whether or not zero matches is allowed
	       and also whether or not two or more matches is allowed.	*/
	    if (greedy)
	      {
		if (many_times_ok)
		  {
		    boolean simple = skip_one_char (laststart) == b;

	    GET_BUFFER_SPACE (34);

	    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;

	    /* Push the number of bytes in the bitmap.	*/
	    BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);

	    /* Clear the whole map.  */
	    bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);

	    /* charset_not matches newline according to a syntax bit.  */
	    if ((re_opcode_t) b[-2] == charset_not
		&& (syntax & RE_HAT_LISTS_NOT_NEWLINE))
	      SET_LIST_BIT ('\n');

	    /* Read in characters and ranges, setting map bits.	 */
	    for (;;)
	      {
		boolean escaped_char = false;
		const unsigned char *p2 = p;

		    PATFETCH (c);
		    escaped_char = true;
		  }
		else
		  {
		    /* Could be the end of the bracket expression.	If it's
		       not (i.e., when the bracket expression is `[]' so
		       far), the ']' character bit gets set way below.  */
		    if (c == ']' && p2 != p1)
		      break;
		  }

		   class.  */

		if (!escaped_char &&
		    syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
		  {
		    /* Leave room for the null.	 */
		    unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
		    const unsigned char *class_beg;

		    PATFETCH (c);
		    c1 = 0;

							 re_wctype_to_bit (cc));

                        for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
			  {
			    int translated = TRANSLATE (ch);
			    if (re_iswctype (btowc (ch), cc))
			      SET_LIST_BIT (translated);
			  }

			/* In most cases the matching rule for char classes
			   only uses the syntax table for multibyte chars,

			p = class_beg;
			SET_LIST_BIT ('[');

			/* Because the `:' may starts the range, we
			   can't simply set bit and repeat the loop.
			   Instead, just set it to C and handle below.	*/
			c = ':';
		      }
		  }

		if (p < pend && p[0] == '-' && p[1] != ']')

		      }
		    else
		      {
			for (this_char = range_start; this_char <= range_end;
			     this_char++)
			  SET_LIST_BIT (TRANSLATE (this_char));
		      }
		  }
		else
		  /* ... into range table.  */
		  SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1);
	      }

	    /* 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];

	    /* Build real range table from work area.  */

		    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;

		 ``can't happen''.  */
	      assert (compile_stack.avail != 0);
	      {
		/* 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

		   won't actually generate any code, so we'll have to
		   clear pending_exact explicitly.  */
		pending_exact = 0;

		/* We're at the end of the group, so now we know how many
		   groups were inside this one.	 */
		if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0)
		  BUF_PUSH_2 (stop_memory, this_group_regnum);
	      }
	      break;

	    handle_alt:
	      if (syntax & RE_LIMITED_OPS)
		goto normal_char;

	      /* Insert before the previous alternative a jump which
		 jumps to this alternative if the former fails.	 */
	      GET_BUFFER_SPACE (3);
	      INSERT_JUMP (on_failure_jump, begalt, b + 6);
	      pending_exact = 0;
	      b += 3;

				      b + 5 + nbytes,
				      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;
			startoffset += 5;

	       else
		 goto normal_char;

#ifdef emacs
	    /* There is no way to specify the before_dot and after_dot
	       operators.  rms says this is ok.	 --karl	 */
	    case '=':
	      BUF_PUSH (at_dot);
	      break;

	    case 's':

	      || pending_exact + *pending_exact + 1 != b

	      /* We have only one byte following the exactn for the count.  */
	      || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH

	      /* If followed by a repetition operator.	*/
	      || (p != pend && (*p == '*' || *p == '^'))
	      || ((syntax & RE_BK_PLUS_QM)
		  ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
		  : p != pend && (*p == '+' || *p == '?'))
	      || ((syntax & RE_INTERVALS)

  FREE_STACK_RETURN (REG_NOERROR);
} /* regex_compile */

/* Subroutines for `regex_compile'.  */

/* Store OP at LOC followed by two-byte integer parameter ARG.	*/

static void
store_op1 (op, loc, arg)
    re_opcode_t op;
    unsigned char *loc;

{
  int j, k;
  boolean not;

  /* If all elements for base leading-codes in fastmap is set, this
     flag is set true.	*/
  boolean match_any_multibyte_characters = false;

  assert (p);

  /* The loop below works as follows:

	  p++;
	  continue;


      /* Following are the cases which match a character.  These end
	 with `break'.	*/

	case exactn:
	  if (fastmap)
	    {
	      int c = RE_STRING_CHAR (p + 1, pend - p);

	  abort (); /* We have listed all the cases.  */
	} /* switch *p++ */

      /* Getting here means we have found the possible starting
	 characters for one path of the pattern -- and that the empty
	 string does not match.	 We need not follow this path further.  */
      return 0;
    } /* while p */

  /* We reached the end without matching anything.  */
  return 1;

  char *fastmap = bufp->fastmap;
  int analysis;

  assert (fastmap && bufp->buffer);

  bzero (fastmap, 1 << BYTEWIDTH);  /* Assume nothing's valid.	*/
  bufp->fastmap_accurate = 1;	    /* It will be when we're done.  */

  analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
			    fastmap, RE_MULTIBYTE_P (bufp));
  bufp->can_be_null = (analysis != 0);

      regs->start = regs->end = (regoff_t *) 0;
    }
}
WEAK_ALIAS (__re_set_registers, re_set_registers)

/* Searching routines.	*/

/* Like re_search_2, below, but only one string is specified, and
   doesn't let you say where to stop matching. */

int

  register RE_TRANSLATE_TYPE translate = bufp->translate;
  int total_size = size1 + size2;
  int endpos = startpos + range;
  boolean anchored_start;

  /* Nonzero if we have to concern multibyte character.	 */
  const boolean multibyte = RE_MULTIBYTE_P (bufp);

  /* Check for out-of-range STARTPOS.  */
  if (startpos < 0 || startpos > total_size)
    return -1;

	}

      /* If a fastmap is supplied, skip quickly over characters that
	 cannot be the start of a match.  If the pattern can match the
	 null string, however, we don't need to skip characters; we want
	 the first null string.	 */
      if (fastmap && startpos < total_size && !bufp->can_be_null)
	{
	  register re_char *d;
	  register re_wchar_t buf_ch;

	  d = POS_ADDR_VSTRING (startpos);

	  if (range > 0)	/* Searching forwards.	*/
	    {
	      register int lim = 0;
	      int irange = range;

	      if (startpos < size1 && startpos + range >= size1)

		      break;
		  } while (1);

	      startpos += irange - range;
	    }
	  else				/* Searching backwards.	 */
	    {
	      int room = (startpos >= size1
			  ? size2 + size1 - startpos
			  : size1 - startpos);
	      buf_ch = RE_STRING_CHAR (d, room);

       d = string2;							\
       dend = end_match_2;						\
     }									\

/* 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)


/* Test if D points to a character which is word-constituent.  We have

/* The comment at case wordbound is following one, but we don't use
   AT_WORD_BOUNDARY anymore to support multibyte form.

   The DEC Alpha C compiler 3.x generates incorrect code for the
   test	 WORDCHAR_P (d - 1) != WORDCHAR_P (d)  in the expansion of
   AT_WORD_BOUNDARY, so this code is disabled.	Expanding the
   macro and introducing temporary variables works around the bug.  */

#if 0
/* Test if the character before D and the one at D differ with respect
   to being word-constituent.  */

	 such case.  */
      else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
	{
	  /* Now, we are sure that P2 has no range table.
	     So, for the size of bitmap in P2, `p2[1]' is
	     enough.	But P1 may have range table, so the
	     size of bitmap table of P1 is extracted by
	     using macro `CHARSET_BITMAP_SIZE'.

	     Since we know that all the character listed in
	     P2 is ASCII, it is enough to test only bitmap

	    }
	  else if ((re_opcode_t) *p1 == charset_not)
	    {
	      int idx;
	      /* We win if the charset_not inside the loop lists
		 every character listed in the charset after.	 */
	      for (idx = 0; idx < (int) p2[1]; idx++)
		if (! (p2[2 + idx] == 0
		       || (idx < CHARSET_BITMAP_SIZE (p1)
			   && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
		  break;

WEAK_ALIAS (__re_match, re_match)
#endif /* not emacs */

#ifdef emacs
/* In Emacs, this is the string or buffer in which we
   are matching.  It is used for looking up syntax properties.	*/
Lisp_Object re_match_object;
#endif

/* re_match_2 matches the compiled pattern in BUFP against the
   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
   failure stack overflowing).	Otherwise, we return the length of the
   matched substring.  */

int
re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
     struct re_pattern_buffer *bufp;

  return result;
}
WEAK_ALIAS (__re_match_2, re_match_2)

/* This is a separate function so that we can force an alloca cleanup
   afterwards.	*/
static int
re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
     struct re_pattern_buffer *bufp;
     re_char *string1, *string2;
     int size1, size2;

  /* Just past the end of the corresponding string.  */
  re_char *end1, *end2;

  /* Pointers into string1 and string2, just past the last characters in
     each to consider matching.	 */
  re_char *end_match_1, *end_match_2;

  /* Where we are in the data, and the end of the current string.  */
  re_char *d, *dend;

  /* Where we are in the pattern, and the end of the pattern.  */
  re_char *p = bufp->buffer;
  re_char *pend = p + bufp->used;

  /* We use this to map every character in the string.	*/
  RE_TRANSLATE_TYPE translate = bufp->translate;

  /* Nonzero if we have to concern multibyte character.	 */
  const boolean multibyte = RE_MULTIBYTE_P (bufp);

  /* Failure point stack.  Each place that can handle a failure further
     down the line pushes a failure point on this stack.  It consists of
     regstart, and regend for all registers corresponding to
     the subexpressions we're currently inside, plus the number of such
     registers, and, finally, two char *'s.  The first char * is where
     to resume scanning the pattern; the second one is where to resume
     scanning the strings.	*/
#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global.	 */
  fail_stack_type fail_stack;
#endif
#ifdef DEBUG
  unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
#endif

     it is allocated relocatably.  */
  fail_stack_elt_t *failure_stack_ptr;
#endif

  /* We fill all the registers internally, independent of what we
     return, for use in backreferences.	 The number here includes
     an element for register zero.  */
  size_t num_regs = bufp->re_nsub + 1;

  /* Information on the contents of registers. These are pointers into
     the input strings; they record just what was matched (on this

  re_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
     the end of the best match so far in a separate variable.  We
     initialize this to NULL so that when we backtrack the first time
     and need to test it, it's not garbage.  */

     register information struct.  */
  for (reg = 1; reg < num_regs; reg++)
    regstart[reg] = regend[reg] = NULL;

  /* We move `string1' into `string2' if the latter's empty -- but not if
     `string1' is null.	 */
  if (size2 == 0 && string1 != NULL)
    {
      string2 = string1;
      size2 = size1;
      string1 = 0;

  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 ("\n%p: ", p);

				    == FIRST_STRING_P (d));
	      /* 1 if this match is the best seen so far.  */
	      boolean best_match_p;

	      /* AIX compiler got confused when this was combined
		 with the previous declaration.	 */
	      if (same_str_p)
		best_match_p = d > match_end;
	      else
		best_match_p = !FIRST_STRING_P (d);

		restore_best_regs:
		  /* Restore best match.  It may happen that `dend ==
		     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);

	  /* If caller wants register contents data back, do it.  */
	  if (regs && !bufp->no_sub)
	    {
	      /* Have the register data arrays been allocated?	*/
	      if (bufp->regs_allocated == REGS_UNALLOCATED)
		{ /* No.  So allocate them with malloc.	 We need one
		     extra element beyond `num_regs' for the `-1' marker
		     GNU code uses.  */
		  regs->num_regs = MAX (RE_NREGS, num_regs + 1);
		  regs->start = TALLOC (regs->num_regs, regoff_t);
		  regs->end = TALLOC (regs->num_regs, regoff_t);

		    }
		}
	      else
		{
		  /* These braces fend off a "empty body in an else-statement"
		     warning under GCC when assert expands to nothing.	*/
		  assert (bufp->regs_allocated == REGS_FIXED);
		}

	      /* Convert the pointer data in `regstart' and `regend' to
		 indices.  Register zero has to be set differently,

	      /* 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.	 */
	      for (reg = num_regs; reg < regs->num_regs; reg++)
		regs->start[reg] = regs->end[reg] = -1;
	    } /* regs && !bufp->no_sub */

	  DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",

	  FREE_VARIABLES ();
	  return mcnt;
	}

      /* Otherwise match next pattern command.	*/
      switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
	{
	/* Ignore these.  Used to ignore the n of succeed_n's which
	   currently have n == 0.  */
	case no_op:

	case succeed:
	  DEBUG_PRINT1 ("EXECUTING succeed.\n");
	  goto succeed_label;

	/* Match the next n pattern characters exactly.	 The following
	   byte in the pattern defines n, and the n bytes after that
	   are the characters to match.	 */
	case exactn:
	  mcnt = *p++;
	  DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);

	  /* Remember the start point to rollback upon failure.  */

	/* The beginning of a group is represented by start_memory.
	   The argument is the register number.  The text
	   matched within the group is recorded (in the internal
	   registers data structure) under the register number.	 */
	case start_memory:
	  DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);

	  /* In case we need to undo this operation (via backtracking).  */
	  PUSH_FAILURE_REG ((unsigned int)*p);

	/* \<digit> has been turned into a `duplicate' command which is
	   followed by the numeric value of <digit> as the register number.  */
	case duplicate:
	  {
	    register re_char *d2, *dend2;
	    int regno = *p++;	/* Get which register to match against.	 */
	    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];

	    /* Remember the start point to rollback upon failure.  */
	    dfail = d;

	/* on_failure_keep_string_jump is used to optimize `.*\n'.  It
	   pushes NULL as the value for the string on the stack.  Then
	   `POP_FAILURE_POINT' will keep the current value for the
	   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

	case jump:
	unconditional_jump:
	  IMMEDIATE_QUIT_CHECK;
	  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 %p).\n", p);
	  break;


	/* Have to succeed matching what follows at least n times.

	      s2 = SYNTAX (c2);

	      if (/* Case 2: Only one of S1 and S2 is Sword.  */
		  ((s1 == Sword) != (s2 == Sword))
		  /* Case 3: Both of S1 and S2 are Sword, and macro
		     WORD_BOUNDARY_P (C1, C2) returns nonzero.	*/
		  || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
		not = !not;
	    }
	  if (not)
	    break;

	case wordbeg:
	  DEBUG_PRINT1 ("EXECUTING wordbeg.\n");

	  /* We FAIL in one of the following cases: */

	  /* Case 1: D is at the end of string.	 */
	  if (AT_STRINGS_END (d))
	    goto fail;
	  else
	    {
	      /* C1 is the character before D, S1 is the syntax of C1, C2

		  UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
#endif
		  s1 = SYNTAX (c1);

		  /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
		     returns 0.	 */
		  if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
		    goto fail;
		}
	    }
	  break;

		  UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
#endif
		  s2 = SYNTAX (c2);

		  /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
		     returns 0.	 */
		  if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
	  goto fail;
		}
	    }
	  break;

	case symbeg:
	  DEBUG_PRINT1 ("EXECUTING symbeg.\n");

	  /* We FAIL in one of the following cases: */

	  /* Case 1: D is at the end of string.	 */
	  if (AT_STRINGS_END (d))
	    goto fail;
	  else
	    {
	      /* C1 is the character before D, S1 is the syntax of C1, C2

	      UPDATE_SYNTAX_TABLE (charpos);
#endif
	      PREFETCH ();
	      c2 = RE_STRING_CHAR (d, dend - d);
	      s2 = SYNTAX (c2);
	
	      /* Case 2: S2 is neither Sword nor Ssymbol. */
	      if (s2 != Sword && s2 != Ssymbol)
		goto fail;

	      /* Case 3: D is not at the beginning of string ... */

83548:c71725faff1a

   the Free Software Foundation; either version 2, or (at your option)
   any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
   USA.  */

/* TODO:
   - structure the opcode space into opcode+flag.
   - merge with glibc's regex.[ch].
   - replace (succeed_n + jump_n + set_number_at) with something that doesn't

# ifdef free
#  undef free
# endif
# define free xfree

/* Converts the pointer to the char to BEG-based offset from the start.  */
# define PTR_TO_OFFSET(d) POS_AS_IN_BUFFER (POINTER_TO_OFFSET (d))
# define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))

# define RE_MULTIBYTE_P(bufp) ((bufp)->multibyte)
# define RE_STRING_CHAR(p, s) \

#  define REGEX_ALLOCATE_STACK alloca

#  define REGEX_REALLOCATE_STACK(source, osize, nsize)			\
   REGEX_REALLOCATE (source, osize, nsize)
/* No need to explicitly free anything.  */
#  define REGEX_FREE_STACK(arg) ((void)0)

# endif /* not REGEX_MALLOC */
#endif /* not using relocating allocator */

	/* Followed by two byte relative address to which to jump.  */
  jump,

	/* 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,

	/* Followed by two-byte relative address, and two-byte number n.
	   Jump to the address N times, then fail.  */
  jump_n,

	/* Set the following two-byte relative address to the
	   subsequent two-byte number.  The address *includes* the two
	   bytes of number.  */
  set_number_at,

  wordbeg,	/* Succeeds if at word beginning.  */
  wordend,	/* Succeeds if at word end.  */

  wordbound,	/* Succeeds if at a word boundary.  */
  notwordbound,	/* Succeeds if not at a word boundary.  */

  symbeg,       /* Succeeds if at symbol beginning.  */
  symend,       /* Succeeds if at symbol end.  */

	/* Matches any character whose syntax is specified.  Followed by

  ,before_dot,	/* Succeeds if before point.  */
  at_dot,	/* Succeeds if at point.  */
  after_dot,	/* Succeeds if after point.  */

  /* Matches any character whose category-set contains the specified
     category.  The operator is followed by a byte which contains a
     category code (mnemonic ASCII character).  */
  categoryspec,

  /* Matches any character whose category-set does not contain the
     specified category.  The operator is followed by a byte which
     contains the category code (mnemonic ASCII character).  */

#endif /* DEBUG */

/* Store a multibyte character in three contiguous bytes starting
   DESTINATION, and increment DESTINATION to the byte after where the
   character is stored.  Therefore, DESTINATION must be an lvalue.  */

#define STORE_CHARACTER_AND_INCR(destination, character)	\
  do {								\
    (destination)[0] = (character) & 0377;			\
    (destination)[1] = ((character) >> 8) & 0377;		\
    (destination)[2] = (character) >> 16;			\
    (destination) += 3;						\
  } while (0)

/* Put into DESTINATION a character stored in three contiguous bytes
   starting at SOURCE.  */

#define EXTRACT_CHARACTER(destination, source)	\
  do {						\
    (destination) = ((source)[0]		\
		     | ((source)[1] << 8)	\

#define CHARSET_LOOKUP_BITMAP(p, c)				\
  ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH			\
   && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))

/* Return the address of end of RANGE_TABLE.  COUNT is number of
   ranges (which is a pair of (start, end)) in the RANGE_TABLE.  `* 2'
   is start of range and end of range.  `* 3' is size of each start
   and end.  */
#define CHARSET_RANGE_TABLE_END(range_table, count)	\
  ((range_table) + (count) * 2 * 3)

/* Test if C is in RANGE_TABLE.  A flag NOT is negated if C is in.
   COUNT is number of ranges in RANGE_TABLE.  */
#define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count)	\
  do									\
    {									\
      re_wchar_t range_start, range_end;				\

     STR -- the saved data position.
     PAT -- the saved pattern position.
     REGSTART, REGEND -- arrays of string positions.

   Also assumes the variables `fail_stack' and (if debugging), `bufp',
   `pend', `string1', `size1', `string2', and `size2'.  */

#define POP_FAILURE_POINT(str, pat)                                     \
do {									\
  assert (!FAIL_STACK_EMPTY ());					\
									\

/* Structure to manage work area for range table.  */
struct range_table_work_area
{
  int *table;			/* actual work area.  */
  int allocated;		/* allocated size for work area in bytes.  */
  int used;			/* actually used size in words.  */
  int bits;			/* flag to record character classes */
};

/* Make sure that WORK_AREA can hold more N multibyte characters.
   This is used only in set_image_of_range and set_image_of_range_1.

    tem = set_image_of_range (&work_area, start, end, translate);	\
    if (tem > 0)							\
      FREE_STACK_RETURN (tem);						\
  } while (0)

/* Free allocated memory for WORK_AREA.  */
#define FREE_RANGE_TABLE_WORK_AREA(work_area)	\
  do {						\
    if ((work_area).table)			\
      free ((work_area).table);			\
  } while (0)

  /* Place in the uncompiled pattern (i.e., the {) to
     which to go back if the interval is invalid.  */
  re_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;

  /* Counts open-groups as they are encountered.  Remembered for the
     matching close-group on the compile stack, so the same register

  if (bufp->allocated == 0)
    {
      if (bufp->buffer)
	{ /* If zero allocated, but buffer is non-null, try to realloc
	     enough space.  This loses if buffer's address is bogus, but
	     that is the user's responsibility.  */
	  RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
	}
      else
	{ /* Caller did not allocate a buffer.  Do it for them.  */
	  bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
	}
      if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);

      bufp->allocated = INIT_BUF_SIZE;

	    main_pend = pend;
	    main_pattern = pattern;
	    p = pattern = whitespace_regexp;
	    pend = p + strlen (p);
	    break;
	  }

	case '^':
	  {
	    if (   /* If at start of pattern, it's an operator.  */
		   p == pattern + 1
		   /* If context independent, it's an operator.  */
		|| syntax & RE_CONTEXT_INDEP_ANCHORS
		   /* Otherwise, depends on what's come before.  */
		|| at_begline_loc_p (pattern, p, syntax))
	      BUF_PUSH ((syntax & RE_NO_NEWLINE_ANCHOR) ? begbuf : begline);
	    else
	      goto normal_char;
	  }

	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 ((syntax & RE_NO_NEWLINE_ANCHOR) ? endbuf : endline);
	     else

	      else if (!(syntax & RE_CONTEXT_INDEP_OPS))
		goto normal_char;
	    }

	  {
	    /* 1 means zero (many) matches is allowed.  */
	    boolean zero_times_ok = 0, many_times_ok = 0;
	    boolean greedy = 1;

	    /* If there is a sequence of repetition chars, collapse it
	       down to just one (the right one).  We can't combine
	       interval operators with these because of, e.g., `a{2}*',
	       which should only match an even number of `a's.  */

	    for (;;)
	      {
		if ((syntax & RE_FRUGAL)
		    && c == '?' && (zero_times_ok || many_times_ok))

	       to an empty pattern.  */
	    if (!laststart || laststart == b)
	      break;

	    /* Now we know whether or not zero matches is allowed
	       and also whether or not two or more matches is allowed.  */
	    if (greedy)
	      {
		if (many_times_ok)
		  {
		    boolean simple = skip_one_char (laststart) == b;

	    GET_BUFFER_SPACE (34);

	    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;

	    /* Push the number of bytes in the bitmap.  */
	    BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);

	    /* Clear the whole map.  */
	    bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);

	    /* charset_not matches newline according to a syntax bit.  */
	    if ((re_opcode_t) b[-2] == charset_not
		&& (syntax & RE_HAT_LISTS_NOT_NEWLINE))
	      SET_LIST_BIT ('\n');

	    /* Read in characters and ranges, setting map bits.  */
	    for (;;)
	      {
		boolean escaped_char = false;
		const unsigned char *p2 = p;

		    PATFETCH (c);
		    escaped_char = true;
		  }
		else
		  {
		    /* Could be the end of the bracket expression.  If it's
		       not (i.e., when the bracket expression is `[]' so
		       far), the ']' character bit gets set way below.  */
		    if (c == ']' && p2 != p1)
		      break;
		  }

		   class.  */

		if (!escaped_char &&
		    syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
		  {
		    /* Leave room for the null.  */
		    unsigned char str[CHAR_CLASS_MAX_LENGTH + 1];
		    const unsigned char *class_beg;

		    PATFETCH (c);
		    c1 = 0;

							 re_wctype_to_bit (cc));

                        for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
			  {
			    int translated = TRANSLATE (ch);
			    if (translated < (1 << BYTEWIDTH)
				&& re_iswctype (btowc (ch), cc))
			      SET_LIST_BIT (translated);
			  }

			/* In most cases the matching rule for char classes
			   only uses the syntax table for multibyte chars,

			p = class_beg;
			SET_LIST_BIT ('[');

			/* Because the `:' may starts the range, we
			   can't simply set bit and repeat the loop.
			   Instead, just set it to C and handle below.  */
			c = ':';
		      }
		  }

		if (p < pend && p[0] == '-' && p[1] != ']')

		      }
		    else
		      {
			for (this_char = range_start; this_char <= range_end;
			     this_char++)
			  {
			    int translated = TRANSLATE (this_char);
			    if (translated < (1 << BYTEWIDTH))
			      SET_LIST_BIT (translated);
			    else
			      SET_RANGE_TABLE_WORK_AREA
				(range_table_work, translated, translated);
			  }
		      }
		  }
		else
		  /* ... into range table.  */
		  SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1);
	      }

	    /* 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];

	    /* Build real range table from work area.  */

		    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;

		 ``can't happen''.  */
	      assert (compile_stack.avail != 0);
	      {
		/* 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

		   won't actually generate any code, so we'll have to
		   clear pending_exact explicitly.  */
		pending_exact = 0;

		/* We're at the end of the group, so now we know how many
		   groups were inside this one.  */
		if (this_group_regnum <= MAX_REGNUM && this_group_regnum > 0)
		  BUF_PUSH_2 (stop_memory, this_group_regnum);
	      }
	      break;

	    handle_alt:
	      if (syntax & RE_LIMITED_OPS)
		goto normal_char;

	      /* Insert before the previous alternative a jump which
		 jumps to this alternative if the former fails.  */
	      GET_BUFFER_SPACE (3);
	      INSERT_JUMP (on_failure_jump, begalt, b + 6);
	      pending_exact = 0;
	      b += 3;

				      b + 5 + nbytes,
				      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;
			startoffset += 5;

	       else
		 goto normal_char;

#ifdef emacs
	    /* There is no way to specify the before_dot and after_dot
	       operators.  rms says this is ok.  --karl  */
	    case '=':
	      BUF_PUSH (at_dot);
	      break;

	    case 's':

	      || pending_exact + *pending_exact + 1 != b

	      /* We have only one byte following the exactn for the count.  */
	      || *pending_exact >= (1 << BYTEWIDTH) - MAX_MULTIBYTE_LENGTH

	      /* If followed by a repetition operator.  */
	      || (p != pend && (*p == '*' || *p == '^'))
	      || ((syntax & RE_BK_PLUS_QM)
		  ? p + 1 < pend && *p == '\\' && (p[1] == '+' || p[1] == '?')
		  : p != pend && (*p == '+' || *p == '?'))
	      || ((syntax & RE_INTERVALS)

  FREE_STACK_RETURN (REG_NOERROR);
} /* regex_compile */

/* Subroutines for `regex_compile'.  */

/* Store OP at LOC followed by two-byte integer parameter ARG.  */

static void
store_op1 (op, loc, arg)
    re_opcode_t op;
    unsigned char *loc;

{
  int j, k;
  boolean not;

  /* If all elements for base leading-codes in fastmap is set, this
     flag is set true.  */
  boolean match_any_multibyte_characters = false;

  assert (p);

  /* The loop below works as follows:

	  p++;
	  continue;


      /* Following are the cases which match a character.  These end
	 with `break'.  */

	case exactn:
	  if (fastmap)
	    {
	      int c = RE_STRING_CHAR (p + 1, pend - p);

	  abort (); /* We have listed all the cases.  */
	} /* switch *p++ */

      /* Getting here means we have found the possible starting
	 characters for one path of the pattern -- and that the empty
	 string does not match.  We need not follow this path further.  */
      return 0;
    } /* while p */

  /* We reached the end without matching anything.  */
  return 1;

  char *fastmap = bufp->fastmap;
  int analysis;

  assert (fastmap && bufp->buffer);

  bzero (fastmap, 1 << BYTEWIDTH);  /* Assume nothing's valid.  */
  bufp->fastmap_accurate = 1;	    /* It will be when we're done.  */

  analysis = analyse_first (bufp->buffer, bufp->buffer + bufp->used,
			    fastmap, RE_MULTIBYTE_P (bufp));
  bufp->can_be_null = (analysis != 0);

      regs->start = regs->end = (regoff_t *) 0;
    }
}
WEAK_ALIAS (__re_set_registers, re_set_registers)

/* Searching routines.  */

/* Like re_search_2, below, but only one string is specified, and
   doesn't let you say where to stop matching. */

int

  register RE_TRANSLATE_TYPE translate = bufp->translate;
  int total_size = size1 + size2;
  int endpos = startpos + range;
  boolean anchored_start;

  /* Nonzero if we have to concern multibyte character.  */
  const boolean multibyte = RE_MULTIBYTE_P (bufp);

  /* Check for out-of-range STARTPOS.  */
  if (startpos < 0 || startpos > total_size)
    return -1;

	}

      /* If a fastmap is supplied, skip quickly over characters that
	 cannot be the start of a match.  If the pattern can match the
	 null string, however, we don't need to skip characters; we want
	 the first null string.  */
      if (fastmap && startpos < total_size && !bufp->can_be_null)
	{
	  register re_char *d;
	  register re_wchar_t buf_ch;

	  d = POS_ADDR_VSTRING (startpos);

	  if (range > 0)	/* Searching forwards.  */
	    {
	      register int lim = 0;
	      int irange = range;

	      if (startpos < size1 && startpos + range >= size1)

		      break;
		  } while (1);

	      startpos += irange - range;
	    }
	  else				/* Searching backwards.  */
	    {
	      int room = (startpos >= size1
			  ? size2 + size1 - startpos
			  : size1 - startpos);
	      buf_ch = RE_STRING_CHAR (d, room);

       d = string2;							\
       dend = end_match_2;						\
     }									\

/* 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)


/* Test if D points to a character which is word-constituent.  We have

/* The comment at case wordbound is following one, but we don't use
   AT_WORD_BOUNDARY anymore to support multibyte form.

   The DEC Alpha C compiler 3.x generates incorrect code for the
   test	 WORDCHAR_P (d - 1) != WORDCHAR_P (d)  in the expansion of
   AT_WORD_BOUNDARY, so this code is disabled.  Expanding the
   macro and introducing temporary variables works around the bug.  */

#if 0
/* Test if the character before D and the one at D differ with respect
   to being word-constituent.  */

	 such case.  */
      else if (!multibyte || !CHARSET_RANGE_TABLE_EXISTS_P (p2))
	{
	  /* Now, we are sure that P2 has no range table.
	     So, for the size of bitmap in P2, `p2[1]' is
	     enough.  But P1 may have range table, so the
	     size of bitmap table of P1 is extracted by
	     using macro `CHARSET_BITMAP_SIZE'.

	     Since we know that all the character listed in
	     P2 is ASCII, it is enough to test only bitmap

	    }
	  else if ((re_opcode_t) *p1 == charset_not)
	    {
	      int idx;
	      /* We win if the charset_not inside the loop lists
		 every character listed in the charset after.  */
	      for (idx = 0; idx < (int) p2[1]; idx++)
		if (! (p2[2 + idx] == 0
		       || (idx < CHARSET_BITMAP_SIZE (p1)
			   && ((p2[2 + idx] & ~ p1[2 + idx]) == 0))))
		  break;

WEAK_ALIAS (__re_match, re_match)
#endif /* not emacs */

#ifdef emacs
/* In Emacs, this is the string or buffer in which we
   are matching.  It is used for looking up syntax properties.  */
Lisp_Object re_match_object;
#endif

/* re_match_2 matches the compiled pattern in BUFP against the
   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
   failure stack overflowing).  Otherwise, we return the length of the
   matched substring.  */

int
re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
     struct re_pattern_buffer *bufp;

  return result;
}
WEAK_ALIAS (__re_match_2, re_match_2)

/* This is a separate function so that we can force an alloca cleanup
   afterwards.  */
static int
re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
     struct re_pattern_buffer *bufp;
     re_char *string1, *string2;
     int size1, size2;

  /* Just past the end of the corresponding string.  */
  re_char *end1, *end2;

  /* Pointers into string1 and string2, just past the last characters in
     each to consider matching.  */
  re_char *end_match_1, *end_match_2;

  /* Where we are in the data, and the end of the current string.  */
  re_char *d, *dend;

  /* Where we are in the pattern, and the end of the pattern.  */
  re_char *p = bufp->buffer;
  re_char *pend = p + bufp->used;

  /* We use this to map every character in the string.  */
  RE_TRANSLATE_TYPE translate = bufp->translate;

  /* Nonzero if we have to concern multibyte character.  */
  const boolean multibyte = RE_MULTIBYTE_P (bufp);

  /* Failure point stack.  Each place that can handle a failure further
     down the line pushes a failure point on this stack.  It consists of
     regstart, and regend for all registers corresponding to
     the subexpressions we're currently inside, plus the number of such
     registers, and, finally, two char *'s.  The first char * is where
     to resume scanning the pattern; the second one is where to resume
     scanning the strings.  */
#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global.  */
  fail_stack_type fail_stack;
#endif
#ifdef DEBUG
  unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
#endif

     it is allocated relocatably.  */
  fail_stack_elt_t *failure_stack_ptr;
#endif

  /* We fill all the registers internally, independent of what we
     return, for use in backreferences.  The number here includes
     an element for register zero.  */
  size_t num_regs = bufp->re_nsub + 1;

  /* Information on the contents of registers. These are pointers into
     the input strings; they record just what was matched (on this

  re_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
     the end of the best match so far in a separate variable.  We
     initialize this to NULL so that when we backtrack the first time
     and need to test it, it's not garbage.  */

     register information struct.  */
  for (reg = 1; reg < num_regs; reg++)
    regstart[reg] = regend[reg] = NULL;

  /* We move `string1' into `string2' if the latter's empty -- but not if
     `string1' is null.  */
  if (size2 == 0 && string1 != NULL)
    {
      string2 = string1;
      size2 = size1;
      string1 = 0;

  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 ("\n%p: ", p);

				    == FIRST_STRING_P (d));
	      /* 1 if this match is the best seen so far.  */
	      boolean best_match_p;

	      /* AIX compiler got confused when this was combined
		 with the previous declaration.  */
	      if (same_str_p)
		best_match_p = d > match_end;
	      else
		best_match_p = !FIRST_STRING_P (d);

		restore_best_regs:
		  /* Restore best match.  It may happen that `dend ==
		     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);

	  /* If caller wants register contents data back, do it.  */
	  if (regs && !bufp->no_sub)
	    {
	      /* Have the register data arrays been allocated?	*/
	      if (bufp->regs_allocated == REGS_UNALLOCATED)
		{ /* No.  So allocate them with malloc.  We need one
		     extra element beyond `num_regs' for the `-1' marker
		     GNU code uses.  */
		  regs->num_regs = MAX (RE_NREGS, num_regs + 1);
		  regs->start = TALLOC (regs->num_regs, regoff_t);
		  regs->end = TALLOC (regs->num_regs, regoff_t);

		    }
		}
	      else
		{
		  /* These braces fend off a "empty body in an else-statement"
		     warning under GCC when assert expands to nothing.  */
		  assert (bufp->regs_allocated == REGS_FIXED);
		}

	      /* Convert the pointer data in `regstart' and `regend' to
		 indices.  Register zero has to be set differently,

	      /* 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.  */
	      for (reg = num_regs; reg < regs->num_regs; reg++)
		regs->start[reg] = regs->end[reg] = -1;
	    } /* regs && !bufp->no_sub */

	  DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",

	  FREE_VARIABLES ();
	  return mcnt;
	}

      /* Otherwise match next pattern command.  */
      switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
	{
	/* Ignore these.  Used to ignore the n of succeed_n's which
	   currently have n == 0.  */
	case no_op:

	case succeed:
	  DEBUG_PRINT1 ("EXECUTING succeed.\n");
	  goto succeed_label;

	/* Match the next n pattern characters exactly.  The following
	   byte in the pattern defines n, and the n bytes after that
	   are the characters to match.  */
	case exactn:
	  mcnt = *p++;
	  DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);

	  /* Remember the start point to rollback upon failure.  */

	/* The beginning of a group is represented by start_memory.
	   The argument is the register number.  The text
	   matched within the group is recorded (in the internal
	   registers data structure) under the register number.  */
	case start_memory:
	  DEBUG_PRINT2 ("EXECUTING start_memory %d:\n", *p);

	  /* In case we need to undo this operation (via backtracking).  */
	  PUSH_FAILURE_REG ((unsigned int)*p);

	/* \<digit> has been turned into a `duplicate' command which is
	   followed by the numeric value of <digit> as the register number.  */
	case duplicate:
	  {
	    register re_char *d2, *dend2;
	    int regno = *p++;	/* Get which register to match against.  */
	    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];

	    /* Remember the start point to rollback upon failure.  */
	    dfail = d;

	/* on_failure_keep_string_jump is used to optimize `.*\n'.  It
	   pushes NULL as the value for the string on the stack.  Then
	   `POP_FAILURE_POINT' will keep the current value for the
	   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

	case jump:
	unconditional_jump:
	  IMMEDIATE_QUIT_CHECK;
	  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 %p).\n", p);
	  break;


	/* Have to succeed matching what follows at least n times.

	      s2 = SYNTAX (c2);

	      if (/* Case 2: Only one of S1 and S2 is Sword.  */
		  ((s1 == Sword) != (s2 == Sword))
		  /* Case 3: Both of S1 and S2 are Sword, and macro
		     WORD_BOUNDARY_P (C1, C2) returns nonzero.  */
		  || ((s1 == Sword) && WORD_BOUNDARY_P (c1, c2)))
		not = !not;
	    }
	  if (not)
	    break;

	case wordbeg:
	  DEBUG_PRINT1 ("EXECUTING wordbeg.\n");

	  /* We FAIL in one of the following cases: */

	  /* Case 1: D is at the end of string.  */
	  if (AT_STRINGS_END (d))
	    goto fail;
	  else
	    {
	      /* C1 is the character before D, S1 is the syntax of C1, C2

		  UPDATE_SYNTAX_TABLE_BACKWARD (charpos - 1);
#endif
		  s1 = SYNTAX (c1);

		  /* ... and S1 is Sword, and WORD_BOUNDARY_P (C1, C2)
		     returns 0.  */
		  if ((s1 == Sword) && !WORD_BOUNDARY_P (c1, c2))
		    goto fail;
		}
	    }
	  break;

		  UPDATE_SYNTAX_TABLE_FORWARD (charpos + 1);
#endif
		  s2 = SYNTAX (c2);

		  /* ... and S2 is Sword, and WORD_BOUNDARY_P (C1, C2)
		     returns 0.  */
		  if ((s2 == Sword) && !WORD_BOUNDARY_P (c1, c2))
	  goto fail;
		}
	    }
	  break;

	case symbeg:
	  DEBUG_PRINT1 ("EXECUTING symbeg.\n");

	  /* We FAIL in one of the following cases: */

	  /* Case 1: D is at the end of string.  */
	  if (AT_STRINGS_END (d))
	    goto fail;
	  else
	    {
	      /* C1 is the character before D, S1 is the syntax of C1, C2

	      UPDATE_SYNTAX_TABLE (charpos);
#endif
	      PREFETCH ();
	      c2 = RE_STRING_CHAR (d, dend - d);
	      s2 = SYNTAX (c2);

	      /* Case 2: S2 is neither Sword nor Ssymbol. */
	      if (s2 != Sword && s2 != Ssymbol)
		goto fail;

	      /* Case 3: D is not at the beginning of string ... */