comparison src/regex.c @ 31312:e6b19a60e035

* regex.h (RE_NO_NEWLINE_ANCHOR): New syntax flag. (struct re_pattern_buffer): Remove newline_anchor. * regex.c: Keep namespace clean for GNU libc by renaming <fun> to __<fun> and using `weak_alias (__<fun>, <fun>)'. (re_max_failures, fail_stack): Use size_t rather than unsigned. (regex_compile): For ^ and $, choose between buffer and line (beg|end) depending on the new RE_NO_NEWLINE_ANCHOR syntax flag. (print_compiled_pattern, re_search_2, mutually_exclusive_p) (re_match_2_internal, re_compile_pattern, re_comp, regcomp): Get rid of references to newline_anchor. (regcomp): Allocate and precompute a fastmap.

31311:a04fdd7a9207

   - structure the opcode space into opcode+flag.
   - merge with glibc's regex.[ch].
   - replace succeed_n + jump_n with a combined operation so that the counter
     can simply be decremented when popping the failure_point without having
     to stack up failure_count entries.
   - get rid of `newline_anchor'.
 */

/* AIX requires this to be the first thing in the file. */
#if defined _AIX && !defined REGEX_MALLOC
  #pragma alloca

#if defined STDC_HEADERS && !defined emacs
# include <stddef.h>
#else
/* We need this for `regex.h', and perhaps for the Emacs include files.  */
# include <sys/types.h>
#endif

/* This is for other GNU distributions with internationalized messages.  */
#if HAVE_LIBINTL_H || defined _LIBC
# include <libintl.h>

    }

  printf ("re_nsub: %d\t", bufp->re_nsub);
  printf ("regs_alloc: %d\t", bufp->regs_allocated);
  printf ("can_be_null: %d\t", bufp->can_be_null);
  printf ("newline_anchor: %d\n", bufp->newline_anchor);
  printf ("no_sub: %d\t", bufp->no_sub);
  printf ("not_bol: %d\t", bufp->not_bol);
  printf ("not_eol: %d\t", bufp->not_eol);
  printf ("syntax: %lx\n", bufp->syntax);
  fflush (stdout);

  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
   in regex.h.  Obviously the order here has to be same as there.
   POSIX doesn't require that we do anything for REG_NOERROR,
   but why not be nice?  */

#endif

/* Roughly the maximum number of failure points on the stack.  Would be
   exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
   This is a variable only so users of regex can assign to it; we never
   change it ourselves.	 */
#if defined MATCH_MAY_ALLOCATE
/* Note that 4400 is enough to cause a crash on Alpha OSF/1,
   whose default stack limit is 2mb.  In order for a larger
   value to work reliably, you have to try to make it accord
   with the process stack limit.  */
int re_max_failures = 40000;
#else
int re_max_failures = 4000;
#endif

union fail_stack_elt
{
   const unsigned char *pointer;
  unsigned int integer;
};

typedef union fail_stack_elt fail_stack_elt_t;

typedef struct
{
  fail_stack_elt_t *stack;
  unsigned size;
  unsigned avail;		/* Offset of next open position.  */
  unsigned frame;		/* Offset of the cur constructed frame.  */
} fail_stack_type;

#define PATTERN_STACK_EMPTY()     (fail_stack.avail == 0)
#define FAIL_STACK_EMPTY()     (fail_stack.frame == 0)
#define FAIL_STACK_FULL()      (fail_stack.avail == fail_stack.size)

     `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.  */

/* Insert the `jump' from the end of last alternative to "here".
   The space for the jump has already been allocated. */
#define FIXUP_ALT_JUMP()						\
do {									\

		   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 (begline);
	    else
	      goto normal_char;
	  }
	  break;

		   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);
	     else
	       goto normal_char;
	   }
	   break;

       worklist.
     We guarantee termination by ignoring backward jumps (more or less),
     so that `p' is monotonically increasing.  More to the point, we
     never set `p' (or push) anything `<= p1'.  */

  /* If can_be_null is set, then the fastmap will not be used anyway.  */
  while (1)
    {
      /* `p1' is used as a marker of how far back a `on_failure_jump'
	 can go without being ignored.  It is normally equal to `p'
	 (which prevents any backward `on_failure_jump') except right

  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));
  if (analysis < -1)
    return analysis;
  bufp->can_be_null = (analysis != 0);
  return 0;
} /* re_compile_fastmap */

/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
   ENDS.  Subsequent matches using PATTERN_BUFFER and REGS will use

      bufp->regs_allocated = REGS_UNALLOCATED;
      regs->num_regs = 0;
      regs->start = regs->end = (regoff_t *) 0;
    }
}

/* Searching routines.	*/

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

     struct re_registers *regs;
{
  return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
		      regs, size);
}

/* End address of virtual concatenation of string.  */
#define STOP_ADDR_VSTRING(P)				\
  (((P) >= size1 ? string2 + size2 : string1 + size1))

  re_char *string2 = (re_char*) str2;
  register char *fastmap = bufp->fastmap;
  register RE_TRANSLATE_TYPE translate = bufp->translate;
  int total_size = size1 + size2;
  int endpos = startpos + range;
  int anchored_start = 0;

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

  /* Check for out-of-range STARTPOS.  */

  if (fastmap && !bufp->fastmap_accurate)
    if (re_compile_fastmap (bufp) == -2)
      return -2;

  /* See whether the pattern is anchored.  */
  if (bufp->buffer[0] == begline)
    anchored_start = 1;

#ifdef emacs
  gl_state.object = re_match_object;
  {
    int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));

	 skip quickly past places we cannot match.
	 We don't bother to treat startpos == 0 specially
	 because that case doesn't repeat.  */
      if (anchored_start && startpos > 0)
	{
	  if (! (bufp->newline_anchor
		 && ((startpos <= size1 ? string1[startpos - 1]
		      : string2[startpos - size1 - 1])
		     == '\n')))
	    goto advance;
	}

      /* If a fastmap is supplied, skip quickly over characters that
	 cannot be the start of a match.  If the pattern can match the

	    }
	}
    }
  return -1;
} /* re_search_2 */

/* Declarations and macros for re_match_2.  */

static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
				    register int len,

	  return 1;
	}
      break;
      
    case endline:
      if (!bufp->newline_anchor)
	break;
      /* Fallthrough */
    case exactn:
      {
	register unsigned int c
	  = (re_opcode_t) *p2 == endline ? '\n'
	  : RE_STRING_CHAR(p2 + 2, pend - p2 - 2);

# if defined C_ALLOCA && !defined REGEX_MALLOC
  alloca (0);
# endif
  return result;
}
#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.	*/

#if defined C_ALLOCA && !defined REGEX_MALLOC
  alloca (0);
#endif
  return result;
}

/* 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)

	  }
	  break;


	/* 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))
	    {

	    }
	  else
	    {
	      unsigned char c;
	      GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
	      if (c == '\n' && bufp->newline_anchor)
		break;
	    }
	  /* In all other cases, we fail.  */
	  goto fail;

	      if (!bufp->not_eol) break;
	    }
	  else
	    {
	      PREFETCH_NOLIMIT ();
	      if (*d == '\n' && bufp->newline_anchor)
		break;
	    }
	  goto fail;

  /* 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 ((re_char*) 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.  */

#if defined _REGEX_RE_COMP || defined _LIBC

	return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
    }

  /* Since `re_exec' always passes NULL for the `regs' argument, we
     don't need to initialize the pattern buffer fields which affect it.  */

  /* 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;

     `buffer' to the compiled pattern;
     `used' to the length of the compiled pattern;
     `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
       REG_EXTENDED bit in CFLAGS is set; otherwise, to
       RE_SYNTAX_POSIX_BASIC;
     `newline_anchor' to REG_NEWLINE being set in CFLAGS;
     `fastmap' and `fastmap_accurate' to zero;
     `re_nsub' to the number of subexpressions in PATTERN.

   PATTERN is the address of the pattern string.

   CFLAGS is a series of bits which affect compilation.

  /* 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;

  /* If REG_NEWLINE is set, newlines are treated differently.  */
  if (cflags & REG_NEWLINE)
    { /* REG_NEWLINE implies neither . nor [^...] match newline.  */
      syntax &= ~RE_DOT_NEWLINE;
      syntax |= RE_HAT_LISTS_NOT_NEWLINE;
      /* It also changes the matching behavior.  */
      preg->newline_anchor = 1;
    }
  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 ((re_char*) 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.

{
  int ret;
  struct re_registers regs;
  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);

      if (regs.start == NULL)
	return (int) REG_NOMATCH;
      regs.end = regs.start + nmatch;
    }

  /* Perform the searching operation.  */
  ret = re_search (&private_preg, string, len,
		   /* start: */ 0, /* range: */ len,
		   want_reg_info ? &regs : (struct re_registers *) 0);

    }

  /* We want zero return to mean success, unlike `re_search'.  */
  return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
}


/* Returns a message corresponding to an error code, ERRCODE, returned
   from either regcomp or regexec.   We don't use PREG here.  */

	strcpy (errbuf, msg);
    }

  return msg_size;
}


/* Free dynamically allocated space used by PREG.  */

void

  if (preg->translate != NULL)
    free (preg->translate);
  preg->translate = NULL;
}

#endif /* not emacs  */

31312:e6b19a60e035

   - structure the opcode space into opcode+flag.
   - merge with glibc's regex.[ch].
   - replace succeed_n + jump_n with a combined operation so that the counter
     can simply be decremented when popping the failure_point without having
     to stack up failure_count entries.
 */

/* AIX requires this to be the first thing in the file. */
#if defined _AIX && !defined REGEX_MALLOC
  #pragma alloca

#if defined STDC_HEADERS && !defined emacs
# include <stddef.h>
#else
/* We need this for `regex.h', and perhaps for the Emacs include files.  */
# include <sys/types.h>
#endif

#ifdef _LIBC
/* We have to keep the namespace clean.  */
# define regfree(preg) __regfree (preg)
# define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
# define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
# define regerror(errcode, preg, errbuf, errbuf_size) \
	__regerror(errcode, preg, errbuf, errbuf_size)
# define re_set_registers(bu, re, nu, st, en) \
	__re_set_registers (bu, re, nu, st, en)
# define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
	__re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
# define re_match(bufp, string, size, pos, regs) \
	__re_match (bufp, string, size, pos, regs)
# define re_search(bufp, string, size, startpos, range, regs) \
	__re_search (bufp, string, size, startpos, range, regs)
# define re_compile_pattern(pattern, length, bufp) \
	__re_compile_pattern (pattern, length, bufp)
# define re_set_syntax(syntax) __re_set_syntax (syntax)
# define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
	__re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
# define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)

# define WEAK_ALIAS(a,b) weak_alias (a, b)

/* We are also using some library internals.  */
# include <locale/localeinfo.h>
# include <locale/elem-hash.h>
# include <langinfo.h>
#else
# define WEAK_ALIAS(a,b)
#endif

/* This is for other GNU distributions with internationalized messages.  */
#if HAVE_LIBINTL_H || defined _LIBC
# include <libintl.h>

    }

  printf ("re_nsub: %d\t", bufp->re_nsub);
  printf ("regs_alloc: %d\t", bufp->regs_allocated);
  printf ("can_be_null: %d\t", bufp->can_be_null);
  printf ("no_sub: %d\t", bufp->no_sub);
  printf ("not_bol: %d\t", bufp->not_bol);
  printf ("not_eol: %d\t", bufp->not_eol);
  printf ("syntax: %lx\n", bufp->syntax);
  fflush (stdout);

  reg_syntax_t ret = re_syntax_options;

  re_syntax_options = syntax;
  return ret;
}
WEAK_ALIAS (__re_set_syntax, re_set_syntax)

/* This table gives an error message for each of the error codes listed
   in regex.h.  Obviously the order here has to be same as there.
   POSIX doesn't require that we do anything for REG_NOERROR,
   but why not be nice?  */

#endif

/* Roughly the maximum number of failure points on the stack.  Would be
   exactly that if always used TYPICAL_FAILURE_SIZE items each time we failed.
   This is a variable only so users of regex can assign to it; we never
   change it ourselves.  */
# if defined MATCH_MAY_ALLOCATE
/* Note that 4400 was enough to cause a crash on Alpha OSF/1,
   whose default stack limit is 2mb.  In order for a larger
   value to work reliably, you have to try to make it accord
   with the process stack limit.  */
size_t re_max_failures = 40000;
# else
size_t re_max_failures = 4000;
# endif

union fail_stack_elt
{
  const unsigned char *pointer;
  /* This should be the biggest `int' that's no bigger than a pointer.  */
  long integer;
};

typedef union fail_stack_elt fail_stack_elt_t;

typedef struct
{
  fail_stack_elt_t *stack;
  size_t size;
  size_t avail;	/* Offset of next open position.  */
  size_t frame;	/* Offset of the cur constructed frame.  */
} fail_stack_type;

#define PATTERN_STACK_EMPTY()     (fail_stack.avail == 0)
#define FAIL_STACK_EMPTY()     (fail_stack.frame == 0)
#define FAIL_STACK_FULL()      (fail_stack.avail == fail_stack.size)

     `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' field is neither examined nor set.  */

/* Insert the `jump' from the end of last alternative to "here".
   The space for the jump has already been allocated. */
#define FIXUP_ALT_JUMP()						\
do {									\

		   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;
	  }
	  break;

		   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
	       goto normal_char;
	   }
	   break;

       worklist.
     We guarantee termination by ignoring backward jumps (more or less),
     so that `p' is monotonically increasing.  More to the point, we
     never set `p' (or push) anything `<= p1'.  */

  while (1)
    {
      /* `p1' is used as a marker of how far back a `on_failure_jump'
	 can go without being ignored.  It is normally equal to `p'
	 (which prevents any backward `on_failure_jump') except right

  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);
  if (analysis < -1)
    return analysis;
  return 0;
} /* re_compile_fastmap */

/* Set REGS to hold NUM_REGS registers, storing them in STARTS and
   ENDS.  Subsequent matches using PATTERN_BUFFER and REGS will use

      bufp->regs_allocated = REGS_UNALLOCATED;
      regs->num_regs = 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. */

     struct re_registers *regs;
{
  return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
		      regs, size);
}
WEAK_ALIAS (__re_search, re_search)

/* End address of virtual concatenation of string.  */
#define STOP_ADDR_VSTRING(P)				\
  (((P) >= size1 ? string2 + size2 : string1 + size1))

  re_char *string2 = (re_char*) str2;
  register char *fastmap = bufp->fastmap;
  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 (fastmap && !bufp->fastmap_accurate)
    if (re_compile_fastmap (bufp) == -2)
      return -2;

  /* See whether the pattern is anchored.  */
  anchored_start = (bufp->buffer[0] == begline);

#ifdef emacs
  gl_state.object = re_match_object;
  {
    int charpos = SYNTAX_TABLE_BYTE_TO_CHAR (POS_AS_IN_BUFFER (startpos));

	 skip quickly past places we cannot match.
	 We don't bother to treat startpos == 0 specially
	 because that case doesn't repeat.  */
      if (anchored_start && startpos > 0)
	{
	  if (! ((startpos <= size1 ? string1[startpos - 1]
		  : string2[startpos - size1 - 1])
		 == '\n'))
	    goto advance;
	}

      /* If a fastmap is supplied, skip quickly over characters that
	 cannot be the start of a match.  If the pattern can match the

	    }
	}
    }
  return -1;
} /* re_search_2 */
WEAK_ALIAS (__re_search_2, re_search_2)

/* Declarations and macros for re_match_2.  */

static int bcmp_translate _RE_ARGS((re_char *s1, re_char *s2,
				    register int len,

	  return 1;
	}
      break;
      
    case endline:
    case exactn:
      {
	register unsigned int c
	  = (re_opcode_t) *p2 == endline ? '\n'
	  : RE_STRING_CHAR(p2 + 2, pend - p2 - 2);

# if defined C_ALLOCA && !defined REGEX_MALLOC
  alloca (0);
# endif
  return result;
}
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.	*/

#if defined C_ALLOCA && !defined REGEX_MALLOC
  alloca (0);
#endif
  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)

	  }
	  break;


	/* begline matches the empty string at the beginning of the string
	   (unless `not_bol' is set in `bufp'), and after newlines.  */
	case begline:
	  DEBUG_PRINT1 ("EXECUTING begline.\n");

	  if (AT_STRINGS_BEG (d))
	    {

	    }
	  else
	    {
	      unsigned char c;
	      GET_CHAR_BEFORE_2 (c, d, string1, end1, string2, end2);
	      if (c == '\n')
		break;
	    }
	  /* In all other cases, we fail.  */
	  goto fail;

	      if (!bufp->not_eol) break;
	    }
	  else
	    {
	      PREFETCH_NOLIMIT ();
	      if (*d == '\n')
		break;
	    }
	  goto fail;

  /* 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;

  ret = regex_compile ((re_char*) pattern, length, re_syntax_options, bufp);

  if (!ret)
    return NULL;
  return gettext (re_error_msgid[(int) ret]);
}
WEAK_ALIAS (__re_compile_pattern, re_compile_pattern)

/* Entry points compatible with 4.2 BSD regex library.  We don't define
   them unless specifically requested.  */

#if defined _REGEX_RE_COMP || defined _LIBC

	return (char *) gettext (re_error_msgid[(int) REG_ESPACE]);
    }

  /* Since `re_exec' always passes NULL for the `regs' argument, we
     don't need to initialize the pattern buffer fields which affect it.  */

  ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);

  if (!ret)
    return NULL;

     `buffer' to the compiled pattern;
     `used' to the length of the compiled pattern;
     `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
       REG_EXTENDED bit in CFLAGS is set; otherwise, to
       RE_SYNTAX_POSIX_BASIC;
     `fastmap' to an allocated space for the fastmap;
     `fastmap_accurate' to zero;
     `re_nsub' to the number of subexpressions in PATTERN.

   PATTERN is the address of the pattern string.

   CFLAGS is a series of bits which affect compilation.

  /* regex_compile will allocate the space for the compiled pattern.  */
  preg->buffer = 0;
  preg->allocated = 0;
  preg->used = 0;

  /* Try to allocate space for the fastmap.  */
  preg->fastmap = (char *) malloc (1 << BYTEWIDTH);

  if (cflags & REG_ICASE)
    {
      unsigned i;

  /* If REG_NEWLINE is set, newlines are treated differently.  */
  if (cflags & REG_NEWLINE)
    { /* REG_NEWLINE implies neither . nor [^...] match newline.  */
      syntax &= ~RE_DOT_NEWLINE;
      syntax |= RE_HAT_LISTS_NOT_NEWLINE;
    }
  else
    syntax |= RE_NO_NEWLINE_ANCHOR;

  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 ((re_char*) 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;

  if (ret == REG_NOERROR && preg->fastmap)
    { /* Compute the fastmap now, since regexec cannot modify the pattern
	 buffer.  */
      re_compile_fastmap (preg);
      if (preg->can_be_null)
	{ /* The fastmap can't be used anyway.  */
	  free (preg->fastmap);
	  preg->fastmap = NULL;
	}
    }
  return (int) ret;
}
WEAK_ALIAS (__regcomp, regcomp)


/* regexec searches for a given pattern, specified by PREG, in the
   string STRING.

{
  int ret;
  struct re_registers regs;
  regex_t private_preg;
  int len = strlen (string);
  boolean want_reg_info = !preg->no_sub && nmatch > 0 && pmatch;

  private_preg = *preg;

  private_preg.not_bol = !!(eflags & REG_NOTBOL);
  private_preg.not_eol = !!(eflags & REG_NOTEOL);

      if (regs.start == NULL)
	return (int) REG_NOMATCH;
      regs.end = regs.start + nmatch;
    }

  /* Instead of using not_eol to implement REG_NOTEOL, we could simply
     pass (&private_preg, string, len + 1, 0, len, ...) pretending the string
     was a little bit longer but still only matching the real part.
     This works because the `endline' will check for a '\n' and will find a
     '\0', correctly deciding that this is not the end of a line.
     But it doesn't work out so nicely for REG_NOTBOL, since we don't have
     a convenient '\0' there.  For all we know, the string could be preceded
     by '\n' which would throw things off.  */

  /* Perform the searching operation.  */
  ret = re_search (&private_preg, string, len,
		   /* start: */ 0, /* range: */ len,
		   want_reg_info ? &regs : (struct re_registers *) 0);

    }

  /* We want zero return to mean success, unlike `re_search'.  */
  return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
}
WEAK_ALIAS (__regexec, regexec)


/* Returns a message corresponding to an error code, ERRCODE, returned
   from either regcomp or regexec.   We don't use PREG here.  */

	strcpy (errbuf, msg);
    }

  return msg_size;
}
WEAK_ALIAS (__regerror, regerror)


/* Free dynamically allocated space used by PREG.  */

void

  if (preg->translate != NULL)
    free (preg->translate);
  preg->translate = NULL;
}
WEAK_ALIAS (__regfree, regfree)

#endif /* not emacs  */