/* Extended regular expression matching and search library, version
   0.12.  (Implements POSIX draft P10003.2/D11.2, except for
   internationalization features.)

   Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998 Free Software Foundation, Inc.

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   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., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
   USA.	 */

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

#undef	_GNU_SOURCE
#define _GNU_SOURCE

#ifdef emacs
/* Converts the pointer to the char to BEG-based offset from the start.	 */
#define PTR_TO_OFFSET(d)						\
	POS_AS_IN_BUFFER (MATCHING_IN_FIRST_STRING			\
			  ? (d) - string1 : (d) - (string2 - size1))
#define POS_AS_IN_BUFFER(p) ((p) + (NILP (re_match_object) || BUFFERP (re_match_object)))
#else
#define PTR_TO_OFFSET(d) 0
#endif

#ifdef HAVE_CONFIG_H
#include <config.h>
#endif

/* We need this for `regex.h', and perhaps for the Emacs include files.	 */
#include <sys/types.h>

/* This is for other GNU distributions with internationalized messages.	 */
#if HAVE_LIBINTL_H || defined (_LIBC)
# include <libintl.h>
#else
# define gettext(msgid) (msgid)
#endif

#ifndef gettext_noop
/* This define is so xgettext can find the internationalizable
   strings.  */
#define gettext_noop(String) String
#endif

/* The `emacs' switch turns on certain matching commands
   that make sense only in Emacs. */
#ifdef emacs

#include "lisp.h"
#include "buffer.h"

/* Make syntax table lookup grant data in gl_state.  */
#define SYNTAX_ENTRY_VIA_PROPERTY

#include "syntax.h"
#include "charset.h"
#include "category.h"

#define malloc xmalloc
#define realloc xrealloc
#define free xfree

#else  /* not emacs */

/* If we are not linking with Emacs proper,
   we can't use the relocating allocator
   even if config.h says that we can.  */
#undef REL_ALLOC

#if defined (STDC_HEADERS) || defined (_LIBC)
#include <stdlib.h>
#else
char *malloc ();
char *realloc ();
#endif

/* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
   If nothing else has been done, use the method below.	 */
#ifdef INHIBIT_STRING_HEADER
#if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
#if !defined (bzero) && !defined (bcopy)
#undef INHIBIT_STRING_HEADER
#endif
#endif
#endif

/* This is the normal way of making sure we have a bcopy and a bzero.
   This is used in most programs--a few other programs avoid this
   by defining INHIBIT_STRING_HEADER.  */
#ifndef INHIBIT_STRING_HEADER
#if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
#include <string.h>
#ifndef bcmp
#define bcmp(s1, s2, n)	memcmp ((s1), (s2), (n))
#endif
#ifndef bcopy
#define bcopy(s, d, n)	memcpy ((d), (s), (n))
#endif
#ifndef bzero
#define bzero(s, n)	memset ((s), 0, (n))
#endif
#else
#include <strings.h>
#endif
#endif

/* 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 SWITCH_ENUM_BUG
#define SWITCH_ENUM_CAST(x) ((int)(x))
#else
#define SWITCH_ENUM_CAST(x) (x)
#endif

#ifdef SYNTAX_TABLE

extern char *re_syntax_table;

#else /* not SYNTAX_TABLE */

/* How many characters in the character set.  */
#define CHAR_SET_SIZE 256

static char re_syntax_table[CHAR_SET_SIZE];

static void
init_syntax_once ()
{
   register int c;
   static int done = 0;

   if (done)
     return;

   bzero (re_syntax_table, sizeof re_syntax_table);

   for (c = 'a'; c <= 'z'; c++)
     re_syntax_table[c] = Sword;

   for (c = 'A'; c <= 'Z'; c++)
     re_syntax_table[c] = Sword;

   for (c = '0'; c <= '9'; c++)
     re_syntax_table[c] = Sword;

   re_syntax_table['_'] = Sword;

   done = 1;
}

#endif /* not SYNTAX_TABLE */

#define SYNTAX(c) re_syntax_table[c]

/* Dummy macros for non-Emacs environments.  */
#define BASE_LEADING_CODE_P(c) (0)
#define WORD_BOUNDARY_P(c1, c2) (0)
#define CHAR_HEAD_P(p) (1)
#define SINGLE_BYTE_CHAR_P(c) (1)
#define SAME_CHARSET_P(c1, c2) (1)
#define MULTIBYTE_FORM_LENGTH(p, s) (1)
#define STRING_CHAR(p, s) (*(p))
#define STRING_CHAR_AND_LENGTH(p, s, actual_len) ((actual_len) = 1, *(p))
#define GET_CHAR_AFTER_2(c, p, str1, end1, str2, end2) \
  (c = ((p) == (end1) ? *(str2) : *(p)))
#define GET_CHAR_BEFORE_2(c, p, str1, end1, str2, end2) \
  (c = ((p) == (str2) ? *((end1) - 1) : *((p) - 1)))
#endif /* not emacs */

/* Get the interface, including the syntax bits.  */
#include "regex.h"

/* isalpha etc. are used for the character classes.  */
#include <ctype.h>

/* Jim Meyering writes:

   "... Some ctype macros are valid only for character codes that
   isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
   using /bin/cc or gcc but without giving an ansi option).  So, all
   ctype uses should be through macros like ISPRINT...	If
   STDC_HEADERS is defined, then autoconf has verified that the ctype
   macros don't need to be guarded with references to isascii. ...
   Defining isascii to 1 should let any compiler worth its salt
   eliminate the && through constant folding."	*/

#if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
#define ISASCII(c) 1
#else
#define ISASCII(c) isascii(c)
#endif

#ifdef isblank
#define ISBLANK(c) (ISASCII (c) && isblank (c))
#else
#define ISBLANK(c) ((c) == ' ' || (c) == '\t')
#endif
#ifdef isgraph
#define ISGRAPH(c) (ISASCII (c) && isgraph (c))
#else
#define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
#endif

#define ISPRINT(c) (ISASCII (c) && isprint (c))
#define ISDIGIT(c) (ISASCII (c) && isdigit (c))
#define ISALNUM(c) (ISASCII (c) && isalnum (c))
#define ISALPHA(c) (ISASCII (c) && isalpha (c))
#define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
#define ISLOWER(c) (ISASCII (c) && islower (c))
#define ISPUNCT(c) (ISASCII (c) && ispunct (c))
#define ISSPACE(c) (ISASCII (c) && isspace (c))
#define ISUPPER(c) (ISASCII (c) && isupper (c))
#define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))

#ifndef NULL
#define NULL (void *)0
#endif

/* We remove any previous definition of `SIGN_EXTEND_CHAR',
   since ours (we hope) works properly with all combinations of
   machines, compilers, `char' and `unsigned char' argument types.
   (Per Bothner suggested the basic approach.)	*/
#undef SIGN_EXTEND_CHAR
#if __STDC__
#define SIGN_EXTEND_CHAR(c) ((signed char) (c))
#else  /* not __STDC__ */
/* As in Harbison and Steele.  */
#define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
#endif

/* 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

#define REGEX_ALLOCATE malloc
#define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
#define REGEX_FREE free

#else /* not REGEX_MALLOC  */

/* Emacs already defines alloca, sometimes.  */
#ifndef alloca

/* Make alloca work the best possible way.  */
#ifdef __GNUC__
#define alloca __builtin_alloca
#else /* not __GNUC__ */
#if HAVE_ALLOCA_H
#include <alloca.h>
#else /* not __GNUC__ or HAVE_ALLOCA_H */
#if 0 /* It is a bad idea to declare alloca.  We always cast the result.  */
#ifndef _AIX /* Already did AIX, up at the top.	 */
char *alloca ();
#endif /* not _AIX */
#endif
#endif /* not HAVE_ALLOCA_H */
#endif /* not __GNUC__ */

#endif /* not alloca */

#define REGEX_ALLOCATE alloca

/* Assumes a `char *destination' variable.  */
#define REGEX_REALLOCATE(source, osize, nsize)				\
  (destination = (char *) alloca (nsize),				\
   bcopy (source, destination, osize),					\
   destination)

/* No need to do anything to free, after alloca.  */
#define REGEX_FREE(arg) ((void)0) /* Do nothing!  But inhibit gcc warning.  */

#endif /* not REGEX_MALLOC */

/* Define how to allocate the failure stack.  */

#if defined (REL_ALLOC) && defined (REGEX_MALLOC)

#define REGEX_ALLOCATE_STACK(size)				\
  r_alloc (&failure_stack_ptr, (size))
#define REGEX_REALLOCATE_STACK(source, osize, nsize)		\
  r_re_alloc (&failure_stack_ptr, (nsize))
#define REGEX_FREE_STACK(ptr)					\
  r_alloc_free (&failure_stack_ptr)

#else /* not using relocating allocator */

#ifdef REGEX_MALLOC

#define REGEX_ALLOCATE_STACK malloc
#define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
#define REGEX_FREE_STACK free

#else /* not REGEX_MALLOC */

#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)

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


/* True if `size1' is non-NULL and PTR is pointing anywhere inside
   `string1' or just past its end.  This works if PTR is NULL, which is
   a good thing.  */
#define FIRST_STRING_P(ptr)					\
  (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)

/* (Re)Allocate N items of type T using malloc, or fail.  */
#define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
#define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
#define RETALLOC_IF(addr, n, t) \
  if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
#define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))

#define BYTEWIDTH 8 /* In bits.	 */

#define STREQ(s1, s2) ((strcmp (s1, s2) == 0))

#undef MAX
#undef MIN
#define MAX(a, b) ((a) > (b) ? (a) : (b))
#define MIN(a, b) ((a) < (b) ? (a) : (b))

typedef char boolean;
#define false 0
#define true 1

static int re_match_2_internal ();

/* These are the command codes that appear in compiled regular
   expressions.	 Some opcodes are followed by argument bytes.  A
   command code can specify any interpretation whatsoever for its
   arguments.  Zero bytes may appear in the compiled regular expression.  */

typedef enum
{
  no_op = 0,

  /* Succeed right away--no more backtracking.	*/
  succeed,

	/* Followed by one byte giving n, then by n literal bytes.  */
  exactn,

	/* Matches any (more or less) character.  */
  anychar,

	/* Matches any one char belonging to specified set.  First
	   following byte is number of bitmap bytes.  Then come bytes
	   for a bitmap saying which chars are in.  Bits in each byte
	   are ordered low-bit-first.  A character is in the set if its
	   bit is 1.  A character too large to have a bit in the map is
	   automatically not in the set.  */
  charset,

	/* Same parameters as charset, but match any character that is
	   not one of those specified.	*/
  charset_not,

	/* Start remembering the text that is matched, for storing in a
	   register.  Followed by one byte with the register number, in
	   the range 0 to one less than the pattern buffer's re_nsub
	   field.  Then followed by one byte with the number of groups
	   inner to this one.  (This last has to be part of the
	   start_memory only because we need it in the on_failure_jump
	   of re_match_2.)  */
  start_memory,

	/* Stop remembering the text that is matched and store it in a
	   memory register.  Followed by one byte with the register
	   number, in the range 0 to one less than `re_nsub' in the
	   pattern buffer, and one byte with the number of inner groups,
	   just like `start_memory'.  (We need the number of inner
	   groups here because we don't have any easy way of finding the
	   corresponding start_memory when we're at a stop_memory.)  */
  stop_memory,

	/* Match a duplicate of something remembered. Followed by one
	   byte containing the register number.	 */
  duplicate,

	/* Fail unless at beginning of line.  */
  begline,

	/* Fail unless at end of line.	*/
  endline,

	/* Succeeds if at beginning of buffer (if emacs) or at beginning
	   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
	   match; otherwise change to jump.  This is used to jump
	   back to the beginning of a repeat.  If what follows this jump
	   clearly won't match what the repeat does, such that we can be
	   sure that there is no use backtracking out of repetitions
	   already matched, then we change it to a pop_failure_jump.
	   Followed by two-byte address.  */
  maybe_pop_jump,

	/* Jump to following two-byte address, and push a dummy failure
	   point. This failure point will be thrown away if an attempt
	   is made to use it for a failure.  A `+' construct makes this
	   before the first repeat.  Also used as an intermediary kind
	   of jump when compiling an alternative.  */
  dummy_failure_jump,

	/* Push a dummy failure point and continue.  Used at the end of
	   alternatives.  */
  push_dummy_failure,

	/* Followed by two-byte relative address and two-byte number n.
	   After matching N times, jump to the address upon failure.  */
  succeed_n,

	/* 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,

  wordchar,	/* Matches any word-constituent character.  */
  notwordchar,	/* Matches any char that is not a word-constituent.  */

  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.	*/

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

	/* Matches any character whose syntax is specified.  Followed by
	   a byte which contains a syntax code, e.g., Sword.  */
  syntaxspec,

	/* Matches any character whose syntax is not that specified.  */
  notsyntaxspec,

  /* 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).  */
  notcategoryspec
#endif /* emacs */
} re_opcode_t;

/* Common operations on the compiled pattern.  */

/* Store NUMBER in two contiguous bytes starting at DESTINATION.  */

#define STORE_NUMBER(destination, number)				\
  do {									\
    (destination)[0] = (number) & 0377;					\
    (destination)[1] = (number) >> 8;					\
  } while (0)

/* Same as STORE_NUMBER, except increment DESTINATION to
   the byte after where the number is stored.  Therefore, DESTINATION
   must be an lvalue.  */

#define STORE_NUMBER_AND_INCR(destination, number)			\
  do {									\
    STORE_NUMBER (destination, number);					\
    (destination) += 2;							\
  } while (0)

/* Put into DESTINATION a number stored in two contiguous bytes starting
   at SOURCE.  */

#define EXTRACT_NUMBER(destination, source)				\
  do {									\
    (destination) = *(source) & 0377;					\
    (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8;		\
  } while (0)

#ifdef DEBUG
static void
extract_number (dest, source)
    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.	*/
#undef EXTRACT_NUMBER
#define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
#endif /* not EXTRACT_MACROS */

#endif /* DEBUG */

/* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
   SOURCE must be an lvalue.  */

#define EXTRACT_NUMBER_AND_INCR(destination, source)			\
  do {									\
    EXTRACT_NUMBER (destination, source);				\
    (source) += 2;							\
  } while (0)

#ifdef DEBUG
static void
extract_number_and_incr (destination, source)
    int *destination;
    unsigned char **source;
{
  extract_number (destination, *source);
  *source += 2;
}

#ifndef EXTRACT_MACROS
#undef EXTRACT_NUMBER_AND_INCR
#define EXTRACT_NUMBER_AND_INCR(dest, src) \
  extract_number_and_incr (&dest, &src)
#endif /* not EXTRACT_MACROS */

#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)	\
		     | ((source)[2] << 16));	\
  } while (0)


/* Macros for charset. */

/* Size of bitmap of charset P in bytes.  P is a start of charset,
   i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not.  */
#define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)

/* Nonzero if charset P has range table.  */
#define CHARSET_RANGE_TABLE_EXISTS_P(p)	 ((p)[1] & 0x80)

/* Return the address of range table of charset P.  But not the start
   of table itself, but the before where the number of ranges is
   stored.  `2 +' means to skip re_opcode_t and size of bitmap.	 */
#define CHARSET_RANGE_TABLE(p) (&(p)[2 + CHARSET_BITMAP_SIZE (p)])

/* Test if C is listed in the bitmap of charset P.  */
#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									\
    {									\
      int range_start, range_end;					\
      unsigned char *p;							\
      unsigned char *range_table_end					\
	= CHARSET_RANGE_TABLE_END ((range_table), (count));		\
									\
      for (p = (range_table); p < range_table_end; p += 2 * 3)		\
	{								\
	  EXTRACT_CHARACTER (range_start, p);				\
	  EXTRACT_CHARACTER (range_end, p + 3);				\
									\
	  if (range_start <= (c) && (c) <= range_end)			\
	    {								\
	      (not) = !(not);						\
	      break;							\
	    }								\
	}								\
    }									\
  while (0)

/* Test if C is in range table of CHARSET.  The flag NOT is negated if
   C is listed in it.  */
#define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset)			\
  do									\
    {									\
      /* Number of ranges in range table. */				\
      int count;							\
      unsigned char *range_table = CHARSET_RANGE_TABLE (charset);	\
									\
      EXTRACT_NUMBER_AND_INCR (count, range_table);			\
      CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count);	\
    }									\
  while (0)

/* If DEBUG is defined, Regex prints many voluminous messages about what
   it is doing (if the variable `debug' is nonzero).  If linked with the
   main program in `iregex.c', you can enter patterns and strings
   interactively.  And if linked with the main program in `main.c' and
   the other test files, you can run the already-written tests.	 */

#ifdef DEBUG

/* We use standard I/O for debugging.  */
#include <stdio.h>

/* It is useful to test things that ``must'' be true when debugging.  */
#include <assert.h>

static int debug = 0;

#define DEBUG_STATEMENT(e) e
#define DEBUG_PRINT1(x) if (debug) printf (x)
#define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
#define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
#define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
#define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)				\
  if (debug) print_partial_compiled_pattern (s, e)
#define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)			\
  if (debug) print_double_string (w, s1, sz1, s2, sz2)


/* Print the fastmap in human-readable form.  */

void
print_fastmap (fastmap)
    char *fastmap;
{
  unsigned was_a_range = 0;
  unsigned i = 0;

  while (i < (1 << BYTEWIDTH))
    {
      if (fastmap[i++])
	{
	  was_a_range = 0;
	  putchar (i - 1);
	  while (i < (1 << BYTEWIDTH)  &&  fastmap[i])
	    {
	      was_a_range = 1;
	      i++;
	    }
	  if (was_a_range)
	    {
	      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.  */

void
print_partial_compiled_pattern (start, end)
    unsigned char *start;
    unsigned char *end;
{
  int mcnt, mcnt2;
  unsigned char *p = start;
  unsigned char *pend = end;

  if (start == NULL)
    {
      printf ("(null)\n");
      return;
    }

  /* Loop over pattern commands.  */
  while (p < pend)
    {
      printf ("%d:\t", p - start);

      switch ((re_opcode_t) *p++)
	{
	case no_op:
	  printf ("/no_op");
	  break;

	case exactn:
	  mcnt = *p++;
	  printf ("/exactn/%d", mcnt);
	  do
	    {
	      putchar ('/');
	      putchar (*p++);
	    }
	  while (--mcnt);
	  break;

	case start_memory:
	  mcnt = *p++;
	  printf ("/start_memory/%d/%d", mcnt, *p++);
	  break;

	case stop_memory:
	  mcnt = *p++;
	  printf ("/stop_memory/%d/%d", mcnt, *p++);
	  break;

	case duplicate:
	  printf ("/duplicate/%d", *p++);
	  break;

	case anychar:
	  printf ("/anychar");
	  break;

	case charset:
	case charset_not:
	  {
	    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 (c / 8 < *p
		  && (p[1 + (c/8)] & (1 << (c % 8))))
		{
		  /* Are we starting a range?  */
		  if (last + 1 == c && ! in_range)
		    {
		      putchar ('-');
		      in_range = 1;
		    }
		  /* Have we broken a range?  */
		  else if (last + 1 != c && in_range)
	      {
		      putchar (last);
		      in_range = 0;
		    }

		  if (! in_range)
		    putchar (c);

		  last = c;
	      }

	    if (in_range)
	      putchar (last);

	    putchar (']');

	    p += 1 + *p;
	  }
	  break;

	case begline:
	  printf ("/begline");
	  break;

	case endline:
	  printf ("/endline");
	  break;

	case on_failure_jump:
	  extract_number_and_incr (&mcnt, &p);
	  printf ("/on_failure_jump to %d", p + mcnt - start);
	  break;

	case on_failure_keep_string_jump:
	  extract_number_and_incr (&mcnt, &p);
	  printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
	  break;

	case dummy_failure_jump:
	  extract_number_and_incr (&mcnt, &p);
	  printf ("/dummy_failure_jump to %d", p + mcnt - start);
	  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:
	  printf ("/notwordbound");
	  break;

	case wordbeg:
	  printf ("/wordbeg");
	  break;

	case wordend:
	  printf ("/wordend");

#ifdef emacs
	case before_dot:
	  printf ("/before_dot");
	  break;

	case at_dot:
	  printf ("/at_dot");
	  break;

	case after_dot:
	  printf ("/after_dot");
	  break;

	case syntaxspec:
	  printf ("/syntaxspec");
	  mcnt = *p++;
	  printf ("/%d", mcnt);
	  break;

	case notsyntaxspec:
	  printf ("/notsyntaxspec");
	  mcnt = *p++;
	  printf ("/%d", mcnt);
	  break;
#endif /* emacs */

	case wordchar:
	  printf ("/wordchar");
	  break;

	case notwordchar:
	  printf ("/notwordchar");
	  break;

	case begbuf:
	  printf ("/begbuf");
	  break;

	case endbuf:
	  printf ("/endbuf");
	  break;

	default:
	  printf ("?%d", *(p-1));
	}

      putchar ('\n');
    }

  printf ("%d:\tend of pattern.\n", p - start);
}


void
print_compiled_pattern (bufp)
    struct re_pattern_buffer *bufp;
{
  unsigned char *buffer = bufp->buffer;

  print_partial_compiled_pattern (buffer, buffer + bufp->used);
  printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);

  if (bufp->fastmap_accurate && bufp->fastmap)
    {
      printf ("fastmap: ");
      print_fastmap (bufp->fastmap);
    }

  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: %d\n", bufp->syntax);
  /* Perhaps we should print the translate table?  */
}


void
print_double_string (where, string1, size1, string2, size2)
    const char *where;
    const char *string1;
    const char *string2;
    int size1;
    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]);
    }
}

#else /* not DEBUG */

#undef assert
#define assert(e)

#define DEBUG_STATEMENT(e)
#define DEBUG_PRINT1(x)
#define DEBUG_PRINT2(x1, x2)
#define DEBUG_PRINT3(x1, x2, x3)
#define DEBUG_PRINT4(x1, x2, x3, x4)
#define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
#define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)

#endif /* not DEBUG */

/* Set by `re_set_syntax' to the current regexp syntax to recognize.  Can
   also be assigned to arbitrarily: each pattern buffer stores its own
   syntax, so it can be changed between regex compilations.  */
/* This has no initializer because initialized variables in Emacs
   become read-only after dumping.  */
reg_syntax_t re_syntax_options;


/* Specify the precise syntax of regexps for compilation.  This provides
   for compatibility for various utilities which historically have
   different, incompatible syntaxes.

   The argument SYNTAX is a bit mask comprised of the various bits
   defined in regex.h.	We return the old syntax.  */

reg_syntax_t
re_set_syntax (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
   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?	 */

static const char *re_error_msgid[] =
  {
    gettext_noop ("Success"),	/* REG_NOERROR */
    gettext_noop ("No match"),	/* REG_NOMATCH */
    gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
    gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
    gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
    gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
    gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
    gettext_noop ("Unmatched [ or [^"),	/* REG_EBRACK */
    gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
    gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
    gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
    gettext_noop ("Invalid range end"),	/* REG_ERANGE */
    gettext_noop ("Memory exhausted"), /* REG_ESPACE */
    gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
    gettext_noop ("Premature end of regular expression"), /* REG_EEND */
    gettext_noop ("Regular expression too big"), /* REG_ESIZE */
    gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
  };

/* Avoiding alloca during matching, to placate r_alloc.	 */

/* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
   searching and matching functions should not call alloca.  On some
   systems, alloca is implemented in terms of malloc, and if we're
   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
   except the matching calls, which it leaves unprotected, in the
   faith that they will not malloc.  */

/* Normally, this is fine.  */
#define MATCH_MAY_ALLOCATE

/* When using GNU C, we are not REALLY using the C alloca, no matter
   what config.h may say.  So don't take precautions for it.  */
#ifdef __GNUC__
#undef C_ALLOCA
#endif

/* The match routines may not allocate if (1) they would do it with malloc
   and (2) it's not safe for them to use malloc.
   Note that if REL_ALLOC is defined, matching would not use malloc for the
   failure stack, but we would still use it for the register vectors;
   so REL_ALLOC should not affect this.	 */
#if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
#undef MATCH_MAY_ALLOCATE
#endif


/* 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.  */


/* Approximate 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
#define INIT_FAILURE_ALLOC 20
#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
{
  unsigned char *pointer;
  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.  */
} fail_stack_type;

#define FAIL_STACK_EMPTY()     (fail_stack.avail == 0)
#define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
#define FAIL_STACK_FULL()      (fail_stack.avail == fail_stack.size)


/* Define macros to initialize and free the failure stack.
   Do `return -2' if the alloc fails.  */

#ifdef MATCH_MAY_ALLOCATE
#define INIT_FAIL_STACK()						\
  do {									\
    fail_stack.stack = (fail_stack_elt_t *)				\
      REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * TYPICAL_FAILURE_SIZE	\
			    * sizeof (fail_stack_elt_t));		\
									\
    if (fail_stack.stack == NULL)					\
      return -2;							\
									\
    fail_stack.size = INIT_FAILURE_ALLOC;				\
    fail_stack.avail = 0;						\
  } while (0)

#define RESET_FAIL_STACK()  REGEX_FREE_STACK (fail_stack.stack)
#else
#define INIT_FAIL_STACK()						\
  do {									\
    fail_stack.avail = 0;						\
  } while (0)

#define RESET_FAIL_STACK()
#endif


/* Double the size of FAIL_STACK, up to a limit
   which allows 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.	*/

/* Factor to increase the failure stack size by
   when we increase it.
   This used to be 2, but 2 was too wasteful
   because the old discarded stacks added up to as much space
   were as ultimate, maximum-size stack.  */
#define FAIL_STACK_GROWTH_FACTOR 4

#define GROW_FAIL_STACK(fail_stack)					\
  (((fail_stack).size * sizeof (fail_stack_elt_t)			\
    >= re_max_failures * TYPICAL_FAILURE_SIZE)				\
   ? 0									\
   : ((fail_stack).stack						\
      = (fail_stack_elt_t *)						\
	REGEX_REALLOCATE_STACK ((fail_stack).stack,			\
	  (fail_stack).size * sizeof (fail_stack_elt_t),		\
	  MIN (re_max_failures * TYPICAL_FAILURE_SIZE,			\
	       ((fail_stack).size * sizeof (fail_stack_elt_t)		\
		* FAIL_STACK_GROWTH_FACTOR))),				\
									\
      (fail_stack).stack == NULL					\
      ? 0								\
      : ((fail_stack).size						\
	 = (MIN (re_max_failures * TYPICAL_FAILURE_SIZE,		\
		 ((fail_stack).size * sizeof (fail_stack_elt_t)		\
		  * FAIL_STACK_GROWTH_FACTOR))				\
	    / sizeof (fail_stack_elt_t)),				\
	 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 ()							\
    && !GROW_FAIL_STACK (FAIL_STACK))					\
   ? 0									\
   : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER,	\
      1))

/* Push a pointer value onto the failure stack.
   Assumes the variable `fail_stack'.  Probably should only
   be called from within `PUSH_FAILURE_POINT'.	*/
#define PUSH_FAILURE_POINTER(item)					\
  fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)

/* This pushes an integer-valued item onto the failure stack.
   Assumes the variable `fail_stack'.  Probably should only
   be called from within `PUSH_FAILURE_POINT'.	*/
#define PUSH_FAILURE_INT(item)					\
  fail_stack.stack[fail_stack.avail++].integer = (item)

/* Push a fail_stack_elt_t value onto the failure stack.
   Assumes the variable `fail_stack'.  Probably should only
   be called from within `PUSH_FAILURE_POINT'.	*/
#define PUSH_FAILURE_ELT(item)					\
  fail_stack.stack[fail_stack.avail++] =  (item)

/* These three POP... operations complement the three PUSH... operations.
   All assume that `fail_stack' is nonempty.  */
#define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
#define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
#define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]

/* Used to omit pushing failure point id's when we're not debugging.  */
#ifdef DEBUG
#define DEBUG_PUSH PUSH_FAILURE_INT
#define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
#else
#define DEBUG_PUSH(item)
#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.  GROW_FAIL_STACK requires `destination' be
   declared.

   Does `return FAILURE_CODE' if runs out of memory.  */

#define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code)	\
  do {									\
    char *destination;							\
    /* Must be int, so when we don't save any registers, the arithmetic	\
       of 0 + -1 isn't done as unsigned.  */				\
    int this_reg;							\
									\
    DEBUG_STATEMENT (failure_id++);					\
    DEBUG_STATEMENT (nfailure_points_pushed++);				\
    DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id);		\
    DEBUG_PRINT2 ("  Before push, next avail: %d\n", (fail_stack).avail);\
    DEBUG_PRINT2 ("			size: %d\n", (fail_stack).size);\
									\
    DEBUG_PRINT2 ("  slots needed: %d\n", NUM_FAILURE_ITEMS);		\
    DEBUG_PRINT2 ("	available: %d\n", REMAINING_AVAIL_SLOTS);	\
									\
    /* Ensure we have enough space allocated for what we will push.  */	\
    while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS)			\
      {									\
	if (!GROW_FAIL_STACK (fail_stack))				\
	  return failure_code;						\
									\
	DEBUG_PRINT2 ("\n  Doubled stack; size now: %d\n",		\
		       (fail_stack).size);				\
	DEBUG_PRINT2 ("	 slots available: %d\n", REMAINING_AVAIL_SLOTS);\
      }									\
									\
    /* Push the info, starting with the registers.  */			\
    DEBUG_PRINT1 ("\n");						\
									\
    if (1)								\
      for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
	   this_reg++)							\
	{								\
	  DEBUG_PRINT2 ("  Pushing reg: %d\n", this_reg);		\
	  DEBUG_STATEMENT (num_regs_pushed++);				\
									\
	  DEBUG_PRINT2 ("    start: 0x%x\n", regstart[this_reg]);	\
	  PUSH_FAILURE_POINTER (regstart[this_reg]);			\
									\
	  DEBUG_PRINT2 ("    end: 0x%x\n", regend[this_reg]);		\
	  PUSH_FAILURE_POINTER (regend[this_reg]);			\
									\
	  DEBUG_PRINT2 ("    info: 0x%x\n      ", reg_info[this_reg]);	\
	  DEBUG_PRINT2 (" match_null=%d",				\
			REG_MATCH_NULL_STRING_P (reg_info[this_reg]));	\
	  DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg]));	\
	  DEBUG_PRINT2 (" matched_something=%d",			\
			MATCHED_SOMETHING (reg_info[this_reg]));	\
	  DEBUG_PRINT2 (" ever_matched=%d",				\
			EVER_MATCHED_SOMETHING (reg_info[this_reg]));	\
	  DEBUG_PRINT1 ("\n");						\
	  PUSH_FAILURE_ELT (reg_info[this_reg].word);			\
	}								\
									\
    DEBUG_PRINT2 ("  Pushing  low active reg: %d\n", lowest_active_reg);\
    PUSH_FAILURE_INT (lowest_active_reg);				\
									\
    DEBUG_PRINT2 ("  Pushing high active reg: %d\n", highest_active_reg);\
    PUSH_FAILURE_INT (highest_active_reg);				\
									\
    DEBUG_PRINT2 ("  Pushing pattern 0x%x: ", pattern_place);		\
    DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend);		\
    PUSH_FAILURE_POINTER (pattern_place);				\
									\
    DEBUG_PRINT2 ("  Pushing string 0x%x: `", string_place);		\
    DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2,	\
				 size2);				\
    DEBUG_PRINT1 ("'\n");						\
    PUSH_FAILURE_POINTER (string_place);				\
									\
    DEBUG_PRINT2 ("  Pushing failure id: %u\n", failure_id);		\
    DEBUG_PUSH (failure_id);						\
  } while (0)

/* This is the number of items that are pushed and popped on the stack
   for each register.  */
#define NUM_REG_ITEMS  3

/* Individual items aside from the registers.  */
#ifdef DEBUG
#define NUM_NONREG_ITEMS 5 /* Includes failure point id.  */
#else
#define NUM_NONREG_ITEMS 4
#endif

/* Estimate the size of data pushed by a typical failure stack entry.
   An estimate is all we need, because all we use this for
   is to choose a limit for how big to make the failure stack.  */

#define TYPICAL_FAILURE_SIZE 20

/* This is how many items we actually use for a failure point.
   It depends on the regexp.  */
#define NUM_FAILURE_ITEMS				\
  (((0							\
     ? 0 : highest_active_reg - lowest_active_reg + 1)	\
    * NUM_REG_ITEMS)					\
   + NUM_NONREG_ITEMS)

/* How many items can still be added to the stack without overflowing it.  */
#define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)


/* Pops what PUSH_FAIL_STACK pushes.

   We restore into the parameters, all of which should be lvalues:
     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)\
{									\
  DEBUG_STATEMENT (fail_stack_elt_t failure_id;)			\
  int this_reg;								\
  const unsigned char *string_temp;					\
									\
  assert (!FAIL_STACK_EMPTY ());					\
									\
  /* Remove failure points and point to how many regs pushed.  */	\
  DEBUG_PRINT1 ("POP_FAILURE_POINT:\n");				\
  DEBUG_PRINT2 ("  Before pop, next avail: %d\n", fail_stack.avail);	\
  DEBUG_PRINT2 ("		     size: %d\n", fail_stack.size);	\
									\
  assert (fail_stack.avail >= NUM_NONREG_ITEMS);			\
									\
  DEBUG_POP (&failure_id);						\
  DEBUG_PRINT2 ("  Popping failure id: %u\n", failure_id);		\
									\
  /* If the saved string location is NULL, it came from an		\
     on_failure_keep_string_jump opcode, and we want to throw away the	\
     saved NULL, thus retaining our current position in the string.  */	\
  string_temp = POP_FAILURE_POINTER ();					\
  if (string_temp != NULL)						\
    str = (const char *) string_temp;					\
									\
  DEBUG_PRINT2 ("  Popping string 0x%x: `", str);			\
  DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2);	\
  DEBUG_PRINT1 ("'\n");							\
									\
  pat = (unsigned char *) POP_FAILURE_POINTER ();			\
  DEBUG_PRINT2 ("  Popping pattern 0x%x: ", pat);			\
  DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend);			\
									\
  /* Restore register info.  */						\
  high_reg = (unsigned) POP_FAILURE_INT ();				\
  DEBUG_PRINT2 ("  Popping high active reg: %d\n", high_reg);		\
									\
  low_reg = (unsigned) POP_FAILURE_INT ();				\
  DEBUG_PRINT2 ("  Popping  low active reg: %d\n", low_reg);		\
									\
  if (1)								\
    for (this_reg = high_reg; this_reg >= low_reg; this_reg--)		\
      {									\
	DEBUG_PRINT2 ("	   Popping reg: %d\n", this_reg);		\
									\
	reg_info[this_reg].word = POP_FAILURE_ELT ();			\
	DEBUG_PRINT2 ("	     info: 0x%x\n", reg_info[this_reg]);	\
									\
	regend[this_reg] = (const char *) POP_FAILURE_POINTER ();	\
	DEBUG_PRINT2 ("	     end: 0x%x\n", regend[this_reg]);		\
									\
	regstart[this_reg] = (const char *) POP_FAILURE_POINTER ();	\
	DEBUG_PRINT2 ("	     start: 0x%x\n", regstart[this_reg]);	\
      }									\
  else									\
    {									\
      for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
	{								\
	  reg_info[this_reg].word.integer = 0;				\
	  regend[this_reg] = 0;						\
	  regstart[this_reg] = 0;					\
	}								\
      highest_active_reg = high_reg;					\
    }									\
									\
  set_regs_matched_done = 0;						\
  DEBUG_STATEMENT (nfailure_points_popped++);				\
} /* POP_FAILURE_POINT */



/* 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.  */

typedef union
{
  fail_stack_elt_t word;
  struct
  {
      /* This field is one if this group can match the empty string,
	 zero if not.  If not yet determined,  `MATCH_NULL_UNSET_VALUE'.  */
#define MATCH_NULL_UNSET_VALUE 3
    unsigned match_null_string_p : 2;
    unsigned is_active : 1;
    unsigned matched_something : 1;
    unsigned ever_matched_something : 1;
  } bits;
} register_info_type;

#define REG_MATCH_NULL_STRING_P(R)  ((R).bits.match_null_string_p)
#define IS_ACTIVE(R)  ((R).bits.is_active)
#define MATCHED_SOMETHING(R)  ((R).bits.matched_something)
#define EVER_MATCHED_SOMETHING(R)  ((R).bits.ever_matched_something)


/* Call this when have matched a real character; it sets `matched' flags
   for the subexpressions which we are currently inside.  Also records
   that those subexprs have matched.  */
#define SET_REGS_MATCHED()						\
  do									\
    {									\
      if (!set_regs_matched_done)					\
	{								\
	  unsigned r;							\
	  set_regs_matched_done = 1;					\
	  for (r = lowest_active_reg; r <= highest_active_reg; r++)	\
	    {								\
	      MATCHED_SOMETHING (reg_info[r])				\
		= EVER_MATCHED_SOMETHING (reg_info[r])			\
		= 1;							\
	    }								\
	}								\
    }									\
  while (0)

/* Registers are set to a sentinel when they haven't yet matched.  */
static char reg_unset_dummy;
#define REG_UNSET_VALUE (&reg_unset_dummy)
#define REG_UNSET(e) ((e) == REG_UNSET_VALUE)

/* Subroutine declarations and macros for regex_compile.  */

static void store_op1 (), store_op2 ();
static void insert_op1 (), insert_op2 ();
static boolean at_begline_loc_p (), at_endline_loc_p ();
static boolean group_in_compile_stack ();

/* 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').  */
#ifndef PATFETCH
#define PATFETCH(c)							\
  do {if (p == pend) return REG_EEND;					\
    c = (unsigned char) *p++;						\
    if (RE_TRANSLATE_P (translate)) c = RE_TRANSLATE (translate, c);	\
  } while (0)
#endif

/* Fetch the next character in the uncompiled pattern, with no
   translation.	 */
#define PATFETCH_RAW(c)							\
  do {if (p == pend) return REG_EEND;					\
    c = (unsigned char) *p++;						\
  } while (0)

/* Go backwards one character in the pattern.  */
#define PATUNFETCH p--


/* If `translate' is non-null, return translate[D], else just D.  We
   cast the subscript to translate because some data is declared as
   `char *', to avoid warnings when a string constant is passed.  But
   when we use a character as a subscript we must make it unsigned.  */
#ifndef TRANSLATE
#define TRANSLATE(d) \
  (RE_TRANSLATE_P (translate) \
   ? (unsigned) RE_TRANSLATE (translate, (unsigned) (d)) : (d))
#endif


/* Macros for outputting the compiled pattern into `buffer'.  */

/* If the buffer isn't allocated when it comes in, use this.  */
#define INIT_BUF_SIZE  32

/* Make sure we have at least N more bytes of space in buffer.	*/
#define GET_BUFFER_SPACE(n)						\
    while (b - bufp->buffer + (n) > bufp->allocated)			\
      EXTEND_BUFFER ()

/* Make sure we have one more byte of buffer space and then add C to it.  */
#define BUF_PUSH(c)							\
  do {									\
    GET_BUFFER_SPACE (1);						\
    *b++ = (unsigned char) (c);						\
  } while (0)


/* Ensure we have two more bytes of buffer space and then append C1 and C2.  */
#define BUF_PUSH_2(c1, c2)						\
  do {									\
    GET_BUFFER_SPACE (2);						\
    *b++ = (unsigned char) (c1);					\
    *b++ = (unsigned char) (c2);					\
  } while (0)


/* As with BUF_PUSH_2, except for three bytes.	*/
#define BUF_PUSH_3(c1, c2, c3)						\
  do {									\
    GET_BUFFER_SPACE (3);						\
    *b++ = (unsigned char) (c1);					\
    *b++ = (unsigned char) (c2);					\
    *b++ = (unsigned char) (c3);					\
  } while (0)


/* Store a jump with opcode OP at LOC to location TO.  We store a
   relative address offset by the three bytes the jump itself occupies.	 */
#define STORE_JUMP(op, loc, to) \
  store_op1 (op, loc, (to) - (loc) - 3)

/* Likewise, for a two-argument jump.  */
#define STORE_JUMP2(op, loc, to, arg) \
  store_op2 (op, loc, (to) - (loc) - 3, arg)

/* Like `STORE_JUMP', but for inserting.  Assume `b' is the buffer end.	 */
#define INSERT_JUMP(op, loc, to) \
  insert_op1 (op, loc, (to) - (loc) - 3, b)

/* Like `STORE_JUMP2', but for inserting.  Assume `b' is the buffer end.  */
#define INSERT_JUMP2(op, loc, to, arg) \
  insert_op2 (op, loc, (to) - (loc) - 3, arg, b)


/* This is not an arbitrary limit: the arguments which represent offsets
   into the pattern are two bytes long.	 So if 2^16 bytes turns out to
   be too small, many things would have to change.  */
#define MAX_BUF_SIZE (1L << 16)


/* E…