Update.

[thirdparty/glibc.git] / locale / programs / ld-ctype.c

/* Copyright (C) 1995-1999, 2000 Free Software Foundation, Inc.
   This file is part of the GNU C Library.
   Contributed by Ulrich Drepper <drepper@gnu.org>, 1995.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Library General Public License as
   published by the Free Software Foundation; either version 2 of the
   License, or (at your option) any later version.

   The GNU C Library 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
   Library General Public License for more details.

   You should have received a copy of the GNU Library General Public
   License along with the GNU C Library; see the file COPYING.LIB.  If not,
   write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
   Boston, MA 02111-1307, USA.  */

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

#include <alloca.h>
#include <byteswap.h>
#include <endian.h>
#include <errno.h>
#include <limits.h>
#include <obstack.h>
#include <stdlib.h>
#include <string.h>
#include <wchar.h>
#include <wctype.h>
#include <sys/uio.h>

#include "charmap.h"
#include "localeinfo.h"
#include "langinfo.h"
#include "linereader.h"
#include "locfile-token.h"
#include "locfile.h"
#include "localedef.h"

#include <assert.h>


#ifdef PREDEFINED_CLASSES
/* These are the extra bits not in wctype.h since these are not preallocated
   classes.  */
# define _ISwspecial1	(1 << 29)
# define _ISwspecial2	(1 << 30)
# define _ISwspecial3	(1 << 31)
#endif


/* The bit used for representing a special class.  */
#define BITPOS(class) ((class) - tok_upper)
#define BIT(class) (_ISbit (BITPOS (class)))
#define BITw(class) (_ISwbit (BITPOS (class)))

#define ELEM(ctype, collection, idx, value)				      \
  *find_idx (ctype, &ctype->collection idx, &ctype->collection##_max idx,     \
	     &ctype->collection##_act idx, value)


/* To be compatible with former implementations we for now restrict
   the number of bits for character classes to 16.  When compatibility
   is not necessary anymore increase the number to 32.  */
#define char_class_t uint16_t
#define char_class32_t uint32_t


/* Type to describe a transliteration action.  We have a possibly
   multiple character from-string and a set of multiple character
   to-strings.  All are 32bit values since this is what is used in
   the gconv functions.  */
struct translit_to_t
{
  uint32_t *str;

  struct translit_to_t *next;
};

struct translit_t
{
  uint32_t *from;

  const char *fname;
  size_t lineno;

  struct translit_to_t *to;

  struct translit_t *next;
};

struct translit_ignore_t
{
  uint32_t from;
  uint32_t to;
  uint32_t step;

  const char *fname;
  size_t lineno;

  struct translit_ignore_t *next;
};


/* The real definition of the struct for the LC_CTYPE locale.  */
struct locale_ctype_t
{
  uint32_t *charnames;
  size_t charnames_max;
  size_t charnames_act;

  struct repertoire_t *repertoire;

  /* We will allow up to 8 * sizeof (uint32_t) character classes.  */
#define MAX_NR_CHARCLASS (8 * sizeof (uint32_t))
  size_t nr_charclass;
  const char *classnames[MAX_NR_CHARCLASS];
  uint32_t last_class_char;
  uint32_t class256_collection[256];
  uint32_t *class_collection;
  size_t class_collection_max;
  size_t class_collection_act;
  uint32_t class_done;
  uint32_t class_offset;

  struct charseq **mbdigits;
  size_t mbdigits_act;
  size_t mbdigits_max;
  uint32_t *wcdigits;
  size_t wcdigits_act;
  size_t wcdigits_max;

  struct charseq *mboutdigits[10];
  uint32_t wcoutdigits[10];
  size_t outdigits_act;

  /* If the following number ever turns out to be too small simply
     increase it.  But I doubt it will.  --drepper@gnu */
#define MAX_NR_CHARMAP 16
  const char *mapnames[MAX_NR_CHARMAP];
  uint32_t *map_collection[MAX_NR_CHARMAP];
  uint32_t map256_collection[2][256];
  size_t map_collection_max[MAX_NR_CHARMAP];
  size_t map_collection_act[MAX_NR_CHARMAP];
  size_t map_collection_nr;
  size_t last_map_idx;
  int tomap_done[MAX_NR_CHARMAP];
  uint32_t map_offset;

  /* Transliteration information.  */
  const char *translit_copy_locale;
  const char *translit_copy_repertoire;
  struct translit_t *translit;
  struct translit_ignore_t *translit_ignore;
  uint32_t ntranslit_ignore;

  uint32_t *default_missing;
  const char *default_missing_file;
  size_t default_missing_lineno;

  /* The arrays for the binary representation.  */
  uint32_t plane_size;
  uint32_t plane_cnt;
  char_class_t *ctype_b;
  char_class32_t *ctype32_b;
  uint32_t *names;
  uint32_t **map;
  uint32_t **map32;
  struct iovec *class_3level;
  struct iovec *map_3level;
  uint32_t *class_name_ptr;
  uint32_t *map_name_ptr;
  unsigned char *width;
  struct iovec width_3level;
  uint32_t mb_cur_max;
  const char *codeset_name;
  uint32_t *translit_from_idx;
  uint32_t *translit_from_tbl;
  uint32_t *translit_to_idx;
  uint32_t *translit_to_tbl;
  uint32_t translit_idx_size;
  size_t translit_from_tbl_size;
  size_t translit_to_tbl_size;

  struct obstack mempool;
};


#define obstack_chunk_alloc xmalloc
#define obstack_chunk_free free


/* Prototypes for local functions.  */
static void ctype_startup (struct linereader *lr, struct localedef_t *locale,
			   struct charmap_t *charmap, int ignore_content);
static void ctype_class_new (struct linereader *lr,
			     struct locale_ctype_t *ctype, const char *name);
static void ctype_map_new (struct linereader *lr,
			   struct locale_ctype_t *ctype,
			   const char *name, struct charmap_t *charmap);
static uint32_t *find_idx (struct locale_ctype_t *ctype, uint32_t **table,
			   size_t *max, size_t *act, unsigned int idx);
static void set_class_defaults (struct locale_ctype_t *ctype,
				struct charmap_t *charmap,
				struct repertoire_t *repertoire);
static void allocate_arrays (struct locale_ctype_t *ctype,
			     struct charmap_t *charmap,
			     struct repertoire_t *repertoire);


static const char *longnames[] =
{
  "zero", "one", "two", "three", "four",
  "five", "six", "seven", "eight", "nine"
};
static const char *uninames[] =
{
  "U00000030", "U00000031", "U00000032", "U00000033", "U00000034",
  "U00000035", "U00000036", "U00000037", "U00000038", "U00000039"
};
static const unsigned char digits[] = "0123456789";


static void
ctype_startup (struct linereader *lr, struct localedef_t *locale,
	       struct charmap_t *charmap, int ignore_content)
{
  unsigned int cnt;
  struct locale_ctype_t *ctype;

  if (!ignore_content)
    {
      /* Allocate the needed room.  */
      locale->categories[LC_CTYPE].ctype = ctype =
	(struct locale_ctype_t *) xcalloc (1, sizeof (struct locale_ctype_t));

      /* We have seen no names yet.  */
      ctype->charnames_max = charmap->mb_cur_max == 1 ? 256 : 512;
      ctype->charnames =
	(unsigned int *) xmalloc (ctype->charnames_max
				  * sizeof (unsigned int));
      for (cnt = 0; cnt < 256; ++cnt)
	ctype->charnames[cnt] = cnt;
      ctype->charnames_act = 256;

      /* Fill character class information.  */
      ctype->last_class_char = ILLEGAL_CHAR_VALUE;
      /* The order of the following instructions determines the bit
	 positions!  */
      ctype_class_new (lr, ctype, "upper");
      ctype_class_new (lr, ctype, "lower");
      ctype_class_new (lr, ctype, "alpha");
      ctype_class_new (lr, ctype, "digit");
      ctype_class_new (lr, ctype, "xdigit");
      ctype_class_new (lr, ctype, "space");
      ctype_class_new (lr, ctype, "print");
      ctype_class_new (lr, ctype, "graph");
      ctype_class_new (lr, ctype, "blank");
      ctype_class_new (lr, ctype, "cntrl");
      ctype_class_new (lr, ctype, "punct");
      ctype_class_new (lr, ctype, "alnum");
#ifdef PREDEFINED_CLASSES
      /* The following are extensions from ISO 14652.  */
      ctype_class_new (lr, ctype, "left_to_right");
      ctype_class_new (lr, ctype, "right_to_left");
      ctype_class_new (lr, ctype, "num_terminator");
      ctype_class_new (lr, ctype, "num_separator");
      ctype_class_new (lr, ctype, "segment_separator");
      ctype_class_new (lr, ctype, "block_separator");
      ctype_class_new (lr, ctype, "direction_control");
      ctype_class_new (lr, ctype, "sym_swap_layout");
      ctype_class_new (lr, ctype, "char_shape_selector");
      ctype_class_new (lr, ctype, "num_shape_selector");
      ctype_class_new (lr, ctype, "non_spacing");
      ctype_class_new (lr, ctype, "non_spacing_level3");
      ctype_class_new (lr, ctype, "normal_connect");
      ctype_class_new (lr, ctype, "r_connect");
      ctype_class_new (lr, ctype, "no_connect");
      ctype_class_new (lr, ctype, "no_connect-space");
      ctype_class_new (lr, ctype, "vowel_connect");
#endif

      ctype->class_collection_max = charmap->mb_cur_max == 1 ? 256 : 512;
      ctype->class_collection
	= (uint32_t *) xcalloc (sizeof (unsigned long int),
				ctype->class_collection_max);
      ctype->class_collection_act = 256;

      /* Fill character map information.  */
      ctype->last_map_idx = MAX_NR_CHARMAP;
      ctype_map_new (lr, ctype, "toupper", charmap);
      ctype_map_new (lr, ctype, "tolower", charmap);
#ifdef PREDEFINED_CLASSES
      ctype_map_new (lr, ctype, "tosymmetric", charmap);
#endif

      /* Fill first 256 entries in `toXXX' arrays.  */
      for (cnt = 0; cnt < 256; ++cnt)
	{
	  ctype->map_collection[0][cnt] = cnt;
	  ctype->map_collection[1][cnt] = cnt;
#ifdef PREDEFINED_CLASSES
	  ctype->map_collection[2][cnt] = cnt;
#endif
	  ctype->map256_collection[0][cnt] = cnt;
	  ctype->map256_collection[1][cnt] = cnt;
	}

      obstack_init (&ctype->mempool);
    }
}


void
ctype_finish (struct localedef_t *locale, struct charmap_t *charmap)
{
  /* See POSIX.2, table 2-6 for the meaning of the following table.  */
#define NCLASS 12
  static const struct
  {
    const char *name;
    const char allow[NCLASS];
  }
  valid_table[NCLASS] =
  {
    /* The order is important.  See token.h for more information.
       M = Always, D = Default, - = Permitted, X = Mutually exclusive  */
    { "upper",  "--MX-XDDXXX-" },
    { "lower",  "--MX-XDDXXX-" },
    { "alpha",  "---X-XDDXXX-" },
    { "digit",  "XXX--XDDXXX-" },
    { "xdigit", "-----XDDXXX-" },
    { "space",  "XXXXX------X" },
    { "print",  "---------X--" },
    { "graph",  "---------X--" },
    { "blank",  "XXXXXM-----X" },
    { "cntrl",  "XXXXX-XX--XX" },
    { "punct",  "XXXXX-DD-X-X" },
    { "alnum",  "-----XDDXXX-" }
  };
  size_t cnt;
  int cls1, cls2;
  uint32_t space_value;
  struct charseq *space_seq;
  struct locale_ctype_t *ctype = locale->categories[LC_CTYPE].ctype;
  int warned;
  const void *key;
  size_t len;
  void *vdata;
  void *curs;

  /* Now resolve copying and also handle completely missing definitions.  */
  if (ctype == NULL)
    {
      const char *repertoire_name;

      /* First see whether we were supposed to copy.  If yes, find the
	 actual definition.  */
      if (locale->copy_name[LC_CTYPE] != NULL)
	{
	  /* Find the copying locale.  This has to happen transitively since
	     the locale we are copying from might also copying another one.  */
	  struct localedef_t *from = locale;

	  do
	    from = find_locale (LC_CTYPE, from->copy_name[LC_CTYPE],
				from->repertoire_name, charmap);
	  while (from->categories[LC_CTYPE].ctype == NULL
		 && from->copy_name[LC_CTYPE] != NULL);

	  ctype = locale->categories[LC_CTYPE].ctype
	    = from->categories[LC_CTYPE].ctype;
	}

      /* If there is still no definition issue an warning and create an
	 empty one.  */
      if (ctype == NULL)
	{
	  if (! be_quiet)
	    error (0, 0, _("No definition for %s category found"), "LC_CTYPE");
	  ctype_startup (NULL, locale, charmap, 0);
	  ctype = locale->categories[LC_CTYPE].ctype;
	}

      /* Get the repertoire we have to use.  */
      repertoire_name = locale->repertoire_name ?: repertoire_global;
      if (repertoire_name != NULL)
	ctype->repertoire = repertoire_read (repertoire_name);
    }

  /* We need the name of the currently used 8-bit character set to
     make correct conversion between this 8-bit representation and the
     ISO 10646 character set used internally for wide characters.  */
  ctype->codeset_name = charmap->code_set_name;
  if (ctype->codeset_name == NULL)
    {
      if (! be_quiet)
	error (0, 0, "no character set name specified in charmap");
      ctype->codeset_name = "//UNKNOWN//";
    }

  /* Set default value for classes not specified.  */
  set_class_defaults (ctype, charmap, ctype->repertoire);

  /* Check according to table.  */
  for (cnt = 0; cnt < ctype->class_collection_act; ++cnt)
    {
      uint32_t tmp = ctype->class_collection[cnt];

      if (tmp != 0)
	{
	  for (cls1 = 0; cls1 < NCLASS; ++cls1)
	    if ((tmp & _ISwbit (cls1)) != 0)
	      for (cls2 = 0; cls2 < NCLASS; ++cls2)
		if (valid_table[cls1].allow[cls2] != '-')
		  {
		    int eq = (tmp & _ISwbit (cls2)) != 0;
		    switch (valid_table[cls1].allow[cls2])
		      {
		      case 'M':
			if (!eq)
			  {
			    uint32_t value = ctype->charnames[cnt];

			    if (!be_quiet)
			      error (0, 0, _("\
character L'\\u%0*x' in class `%s' must be in class `%s'"),
				     value > 0xffff ? 8 : 4, value,
				     valid_table[cls1].name,
				     valid_table[cls2].name);
			  }
			break;

		      case 'X':
			if (eq)
			  {
			    uint32_t value = ctype->charnames[cnt];

			    if (!be_quiet)
			      error (0, 0, _("\
character L'\\u%0*x' in class `%s' must not be in class `%s'"),
				     value > 0xffff ? 8 : 4, value,
				     valid_table[cls1].name,
				     valid_table[cls2].name);
			  }
			break;

		      case 'D':
			ctype->class_collection[cnt] |= _ISwbit (cls2);
			break;

		      default:
			error (5, 0, _("internal error in %s, line %u"),
			       __FUNCTION__, __LINE__);
		      }
		  }
	}
    }

  for (cnt = 0; cnt < 256; ++cnt)
    {
      uint32_t tmp = ctype->class256_collection[cnt];

      if (tmp != 0)
	{
	  for (cls1 = 0; cls1 < NCLASS; ++cls1)
	    if ((tmp & _ISbit (cls1)) != 0)
	      for (cls2 = 0; cls2 < NCLASS; ++cls2)
		if (valid_table[cls1].allow[cls2] != '-')
		  {
		    int eq = (tmp & _ISbit (cls2)) != 0;
		    switch (valid_table[cls1].allow[cls2])
		      {
		      case 'M':
			if (!eq)
			  {
			    char buf[17];

			    snprintf (buf, sizeof buf, "\\%Zo", cnt);

			    if (!be_quiet)
			      error (0, 0, _("\
character '%s' in class `%s' must be in class `%s'"),
				     buf, valid_table[cls1].name,
				     valid_table[cls2].name);
			  }
			break;

		      case 'X':
			if (eq)
			  {
			    char buf[17];

			    snprintf (buf, sizeof buf, "\\%Zo", cnt);

			    if (!be_quiet)
			      error (0, 0, _("\
character '%s' in class `%s' must not be in class `%s'"),
				     buf, valid_table[cls1].name,
				     valid_table[cls2].name);
			  }
			break;

		      case 'D':
			ctype->class256_collection[cnt] |= _ISbit (cls2);
			break;

		      default:
			error (5, 0, _("internal error in %s, line %u"),
			       __FUNCTION__, __LINE__);
		      }
		  }
	}
    }

  /* ... and now test <SP> as a special case.  */
  space_value = 32;
  if (((cnt = BITPOS (tok_space),
	(ELEM (ctype, class_collection, , space_value)
	 & BITw (tok_space)) == 0)
       || (cnt = BITPOS (tok_blank),
	   (ELEM (ctype, class_collection, , space_value)
	    & BITw (tok_blank)) == 0)))
    {
      if (!be_quiet)
	error (0, 0, _("<SP> character not in class `%s'"),
	       valid_table[cnt].name);
    }
  else if (((cnt = BITPOS (tok_punct),
	     (ELEM (ctype, class_collection, , space_value)
	      & BITw (tok_punct)) != 0)
	    || (cnt = BITPOS (tok_graph),
		(ELEM (ctype, class_collection, , space_value)
		 & BITw (tok_graph))
		!= 0)))
    {
      if (!be_quiet)
	error (0, 0, _("<SP> character must not be in class `%s'"),
	       valid_table[cnt].name);
    }
  else
    ELEM (ctype, class_collection, , space_value) |= BITw (tok_print);

  space_seq = charmap_find_value (charmap, "SP", 2);
  if (space_seq == NULL)
    space_seq = charmap_find_value (charmap, "space", 5);
  if (space_seq == NULL)
    space_seq = charmap_find_value (charmap, "U00000020", 9);
  if (space_seq == NULL || space_seq->nbytes != 1)
    {
      if (!be_quiet)
	error (0, 0, _("character <SP> not defined in character map"));
    }
  else if (((cnt = BITPOS (tok_space),
	     (ctype->class256_collection[space_seq->bytes[0]]
	      & BIT (tok_space)) == 0)
	    || (cnt = BITPOS (tok_blank),
		(ctype->class256_collection[space_seq->bytes[0]]
		 & BIT (tok_blank)) == 0)))
    {
      if (!be_quiet)
	error (0, 0, _("<SP> character not in class `%s'"),
	       valid_table[cnt].name);
    }
  else if (((cnt = BITPOS (tok_punct),
	     (ctype->class256_collection[space_seq->bytes[0]]
	      & BIT (tok_punct)) != 0)
	    || (cnt = BITPOS (tok_graph),
		(ctype->class256_collection[space_seq->bytes[0]]
		 & BIT (tok_graph)) != 0)))
    {
      if (!be_quiet)
	error (0, 0, _("<SP> character must not be in class `%s'"),
	       valid_table[cnt].name);
    }
  else
    ctype->class256_collection[space_seq->bytes[0]] |= BIT (tok_print);

  /* Now that the tests are done make sure the name array contains all
     characters which are handled in the WIDTH section of the
     character set definition file.  */
  if (charmap->width_rules != NULL)
    for (cnt = 0; cnt < charmap->nwidth_rules; ++cnt)
      {
	unsigned char bytes[charmap->mb_cur_max];
	int nbytes = charmap->width_rules[cnt].from->nbytes;

	/* We have the range of character for which the width is
           specified described using byte sequences of the multibyte
           charset.  We have to convert this to UCS4 now.  And we
           cannot simply convert the beginning and the end of the
           sequence, we have to iterate over the byte sequence and
           convert it for every single character.  */
	memcpy (bytes, charmap->width_rules[cnt].from->bytes, nbytes);

	while (nbytes < charmap->width_rules[cnt].to->nbytes
	       || memcmp (bytes, charmap->width_rules[cnt].to->bytes,
			  nbytes) <= 0)
	  {
	    /* Find the UCS value for `bytes'.  */
	    int inner;
	    uint32_t wch;
	    struct charseq *seq = charmap_find_symbol (charmap, bytes, nbytes);

	    if (seq == NULL)
	      wch = ILLEGAL_CHAR_VALUE;
	    else if (seq->ucs4 != UNINITIALIZED_CHAR_VALUE)
	      wch = seq->ucs4;
	    else
	      wch = repertoire_find_value (ctype->repertoire, seq->name,
					   strlen (seq->name));

	    if (wch != ILLEGAL_CHAR_VALUE)
	      /* We are only interested in the side-effects of the
		 `find_idx' call.  It will add appropriate entries in
		 the name array if this is necessary.  */
	      (void) find_idx (ctype, NULL, NULL, NULL, wch);

	    /* "Increment" the bytes sequence.  */
	    inner = nbytes - 1;
	    while (inner >= 0 && bytes[inner] == 0xff)
	      --inner;

	    if (inner < 0)
	      {
		/* We have to extend the byte sequence.  */
		if (nbytes >= charmap->width_rules[cnt].to->nbytes)
		  break;

		bytes[0] = 1;
		memset (&bytes[1], 0, nbytes);
		++nbytes;
	      }
	    else
	      {
		++bytes[inner];
		while (++inner < nbytes)
		  bytes[inner] = 0;
	      }
	  }
      }

  /* Now set all the other characters of the character set to the
     default width.  */
  curs = NULL;
  while (iterate_table (&charmap->char_table, &curs, &key, &len, &vdata) == 0)
    {
      struct charseq *data = (struct charseq *) vdata;

      if (data->ucs4 == UNINITIALIZED_CHAR_VALUE)
	data->ucs4 = repertoire_find_value (ctype->repertoire,
					    data->name, len);

      if (data->ucs4 != ILLEGAL_CHAR_VALUE)
	(void) find_idx (ctype, NULL, NULL, NULL, data->ucs4);
    }

  /* There must be a multiple of 10 digits.  */
  if (ctype->mbdigits_act % 10 != 0)
    {
      assert (ctype->mbdigits_act == ctype->wcdigits_act);
      ctype->wcdigits_act -= ctype->mbdigits_act % 10;
      ctype->mbdigits_act -= ctype->mbdigits_act % 10;
      error (0, 0, _("`digit' category has not entries in groups of ten"));
    }

  /* Check the input digits.  There must be a multiple of ten available.
     In each group it could be that one or the other character is missing.
     In this case the whole group must be removed.  */
  cnt = 0;
  while (cnt < ctype->mbdigits_act)
    {
      size_t inner;
      for (inner = 0; inner < 10; ++inner)
	if (ctype->mbdigits[cnt + inner] == NULL)
	  break;

      if (inner == 10)
	cnt += 10;
      else
	{
	  /* Remove the group.  */
	  memmove (&ctype->mbdigits[cnt], &ctype->mbdigits[cnt + 10],
		   ((ctype->wcdigits_act - cnt - 10)
		    * sizeof (ctype->mbdigits[0])));
	  ctype->mbdigits_act -= 10;
	}
    }

  /* If no input digits are given use the default.  */
  if (ctype->mbdigits_act == 0)
    {
      if (ctype->mbdigits_max == 0)
	{
	  ctype->mbdigits = obstack_alloc (&charmap->mem_pool,
					   10 * sizeof (struct charseq *));
	  ctype->mbdigits_max = 10;
	}

      for (cnt = 0; cnt < 10; ++cnt)
	{
	  ctype->mbdigits[cnt] = charmap_find_symbol (charmap,
						      digits + cnt, 1);
	  if (ctype->mbdigits[cnt] == NULL)
	    {
	      ctype->mbdigits[cnt] = charmap_find_symbol (charmap,
							  longnames[cnt],
							  strlen (longnames[cnt]));
	      if (ctype->mbdigits[cnt] == NULL)
		{
		  /* Hum, this ain't good.  */
		  error (0, 0, _("\
no input digits defined and none of the standard names in the charmap"));

		  ctype->mbdigits[cnt] = obstack_alloc (&charmap->mem_pool,
							sizeof (struct charseq) + 1);

		  /* This is better than nothing.  */
		  ctype->mbdigits[cnt]->bytes[0] = digits[cnt];
		  ctype->mbdigits[cnt]->nbytes = 1;
		}
	    }
	}

      ctype->mbdigits_act = 10;
    }

  /* Check the wide character input digits.  There must be a multiple
     of ten available.  In each group it could be that one or the other
     character is missing.  In this case the whole group must be
     removed.  */
  cnt = 0;
  while (cnt < ctype->wcdigits_act)
    {
      size_t inner;
      for (inner = 0; inner < 10; ++inner)
	if (ctype->wcdigits[cnt + inner] == ILLEGAL_CHAR_VALUE)
	  break;

      if (inner == 10)
	cnt += 10;
      else
	{
	  /* Remove the group.  */
	  memmove (&ctype->wcdigits[cnt], &ctype->wcdigits[cnt + 10],
		   ((ctype->wcdigits_act - cnt - 10)
		    * sizeof (ctype->wcdigits[0])));
	  ctype->wcdigits_act -= 10;
	}
    }

  /* If no input digits are given use the default.  */
  if (ctype->wcdigits_act == 0)
    {
      if (ctype->wcdigits_max == 0)
	{
	  ctype->wcdigits = obstack_alloc (&charmap->mem_pool,
					   10 * sizeof (uint32_t));
	  ctype->wcdigits_max = 10;
	}

      for (cnt = 0; cnt < 10; ++cnt)
	ctype->wcdigits[cnt] = L'0' + cnt;

      ctype->mbdigits_act = 10;
    }

  /* Check the outdigits.  */
  warned = 0;
  for (cnt = 0; cnt < 10; ++cnt)
    if (ctype->mboutdigits[cnt] == NULL)
      {
	static struct charseq replace[2];

	if (!warned)
	  {
	    error (0, 0, _("\
not all characters used in `outdigit' are available in the charmap"));
	    warned = 1;
	  }

	replace[0].nbytes = 1;
	replace[0].bytes[0] = '?';
	replace[0].bytes[1] = '\0';
	ctype->mboutdigits[cnt] = &replace[0];
      }

  warned = 0;
  for (cnt = 0; cnt < 10; ++cnt)
    if (ctype->wcoutdigits[cnt] == 0)
      {
	if (!warned)
	  {
	    error (0, 0, _("\
not all characters used in `outdigit' are available in the repertoire"));
	    warned = 1;
	  }

	ctype->wcoutdigits[cnt] = L'?';
      }

  /* Sort the entries in the translit_ignore list.  */
  if (ctype->translit_ignore != NULL)
    {
      struct translit_ignore_t *firstp = ctype->translit_ignore;
      struct translit_ignore_t *runp;

      ctype->ntranslit_ignore = 1;

      for (runp = firstp->next; runp != NULL; runp = runp->next)
	{
	  struct translit_ignore_t *lastp = NULL;
	  struct translit_ignore_t *cmpp;

	  ++ctype->ntranslit_ignore;

	  for (cmpp = firstp; cmpp != NULL; lastp = cmpp, cmpp = cmpp->next)
	    if (runp->from < cmpp->from)
	      break;

	  runp->next = lastp;
	  if (lastp == NULL)
	    firstp = runp;
	}

      ctype->translit_ignore = firstp;
    }
}


void
ctype_output (struct localedef_t *locale, struct charmap_t *charmap,
	      const char *output_path)
{
  static const char nulbytes[4] = { 0, 0, 0, 0 };
  struct locale_ctype_t *ctype = locale->categories[LC_CTYPE].ctype;
  const size_t nelems = (_NL_ITEM_INDEX (_NL_NUM_LC_CTYPE)
			 + (oldstyle_tables
			    ? (ctype->map_collection_nr - 2)
			    : (ctype->nr_charclass + ctype->map_collection_nr)));
  struct iovec iov[2 + nelems + ctype->nr_charclass
		  + ctype->map_collection_nr + 2];
  struct locale_file data;
  uint32_t idx[nelems + 1];
  uint32_t default_missing_len;
  size_t elem, cnt, offset, total;
  char *cp;

  /* Now prepare the output: Find the sizes of the table we can use.  */
  allocate_arrays (ctype, charmap, ctype->repertoire);

  data.magic = LIMAGIC (LC_CTYPE);
  data.n = nelems;
  iov[0].iov_base = (void *) &data;
  iov[0].iov_len = sizeof (data);

  iov[1].iov_base = (void *) idx;
  iov[1].iov_len = nelems * sizeof (uint32_t);

  idx[0] = iov[0].iov_len + iov[1].iov_len;
  offset = 0;

  for (elem = 0; elem < nelems; ++elem)
    {
      if (elem < _NL_ITEM_INDEX (_NL_NUM_LC_CTYPE))
	switch (elem)
	  {
#define CTYPE_EMPTY(name) \
	  case name:							      \
	    iov[2 + elem + offset].iov_base = (void *) "";		      \
	    iov[2 + elem + offset].iov_len = 0;				      \
	    idx[elem + 1] = idx[elem];					      \
	    break

	  CTYPE_EMPTY(_NL_CTYPE_GAP1);
	  CTYPE_EMPTY(_NL_CTYPE_GAP2);
	  CTYPE_EMPTY(_NL_CTYPE_GAP3);

#define CTYPE_DATA(name, base, len)					      \
	  case _NL_ITEM_INDEX (name):					      \
	    iov[2 + elem + offset].iov_base = (base);			      \
	    iov[2 + elem + offset].iov_len = (len);			      \
	    idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;	      \
	    break

	  CTYPE_DATA (_NL_CTYPE_CLASS,
		      ctype->ctype_b,
		      (256 + 128) * sizeof (char_class_t));

	  CTYPE_DATA (_NL_CTYPE_TOUPPER,
		      ctype->map[0],
		      (256 + 128) * sizeof (uint32_t));
	  CTYPE_DATA (_NL_CTYPE_TOLOWER,
		      ctype->map[1],
		      (256 + 128) * sizeof (uint32_t));

	  CTYPE_DATA (_NL_CTYPE_TOUPPER32,
		      ctype->map32[0],
		      (oldstyle_tables ? ctype->plane_size * ctype->plane_cnt : 256)
		      * sizeof (uint32_t));
	  CTYPE_DATA (_NL_CTYPE_TOLOWER32,
		      ctype->map32[1],
		      (oldstyle_tables ? ctype->plane_size * ctype->plane_cnt : 256)
		      * sizeof (uint32_t));

	  CTYPE_DATA (_NL_CTYPE_CLASS32,
		      ctype->ctype32_b,
		      (oldstyle_tables ? ctype->plane_size * ctype->plane_cnt : 256)
		      * sizeof (char_class32_t));

	  CTYPE_DATA (_NL_CTYPE_NAMES,
		      ctype->names,
		      (oldstyle_tables ? ctype->plane_size * ctype->plane_cnt : 0)
		      * sizeof (uint32_t));

	  CTYPE_DATA (_NL_CTYPE_CLASS_OFFSET,
		      &ctype->class_offset, sizeof (uint32_t));

	  CTYPE_DATA (_NL_CTYPE_MAP_OFFSET,
		      &ctype->map_offset, sizeof (uint32_t));

	  CTYPE_DATA (_NL_CTYPE_TRANSLIT_TAB_SIZE,
		      &ctype->translit_idx_size, sizeof (uint32_t));

	  CTYPE_DATA (_NL_CTYPE_TRANSLIT_FROM_IDX,
		      ctype->translit_from_idx,
		      ctype->translit_idx_size * sizeof (uint32_t));

	  CTYPE_DATA (_NL_CTYPE_TRANSLIT_FROM_TBL,
		      ctype->translit_from_tbl,
		      ctype->translit_from_tbl_size);

	  CTYPE_DATA (_NL_CTYPE_TRANSLIT_TO_IDX,
		      ctype->translit_to_idx,
		      ctype->translit_idx_size * sizeof (uint32_t));

	  CTYPE_DATA (_NL_CTYPE_TRANSLIT_TO_TBL,
		      ctype->translit_to_tbl, ctype->translit_to_tbl_size);

	  CTYPE_DATA (_NL_CTYPE_HASH_SIZE,
		      &ctype->plane_size, sizeof (uint32_t));
	  CTYPE_DATA (_NL_CTYPE_HASH_LAYERS,
		      &ctype->plane_cnt, sizeof (uint32_t));

	  case _NL_ITEM_INDEX (_NL_CTYPE_CLASS_NAMES):
	    /* The class name array.  */
	    total = 0;
	    for (cnt = 0; cnt < ctype->nr_charclass; ++cnt, ++offset)
	      {
		iov[2 + elem + offset].iov_base
		  = (void *) ctype->classnames[cnt];
		iov[2 + elem + offset].iov_len
		  = strlen (ctype->classnames[cnt]) + 1;
		total += iov[2 + elem + offset].iov_len;
	      }
	    iov[2 + elem + offset].iov_base = (void *) nulbytes;
	    iov[2 + elem + offset].iov_len = 1 + (4 - ((total + 1) % 4));
	    total += 1 + (4 - ((total + 1) % 4));

	    idx[elem + 1] = idx[elem] + total;
	    break;

	  case _NL_ITEM_INDEX (_NL_CTYPE_MAP_NAMES):
	    /* The class name array.  */
	    total = 0;
	    for (cnt = 0; cnt < ctype->map_collection_nr; ++cnt, ++offset)
	      {
		iov[2 + elem + offset].iov_base
		  = (void *) ctype->mapnames[cnt];
		iov[2 + elem + offset].iov_len
		  = strlen (ctype->mapnames[cnt]) + 1;
		total += iov[2 + elem + offset].iov_len;
	      }
	    iov[2 + elem + offset].iov_base = (void *) nulbytes;
	    iov[2 + elem + offset].iov_len = 1 + (4 - ((total + 1) % 4));
	    total += 1 + (4 - ((total + 1) % 4));

	    idx[elem + 1] = idx[elem] + total;
	    break;

	  CTYPE_DATA (_NL_CTYPE_WIDTH,
		      (oldstyle_tables
		       ? ctype->width
		       : ctype->width_3level.iov_base),
		      (oldstyle_tables
		       ? (ctype->plane_size * ctype->plane_cnt + 3) & ~3ul
		       : ctype->width_3level.iov_len));

	  CTYPE_DATA (_NL_CTYPE_MB_CUR_MAX,
		      &ctype->mb_cur_max, sizeof (uint32_t));

	  case _NL_ITEM_INDEX (_NL_CTYPE_CODESET_NAME):
	    total = strlen (ctype->codeset_name) + 1;
	    if (total % 4 == 0)
	      iov[2 + elem + offset].iov_base = (char *) ctype->codeset_name;
	    else
	      {
		iov[2 + elem + offset].iov_base = alloca ((total + 3) & ~3);
		memset (mempcpy (iov[2 + elem + offset].iov_base,
				 ctype->codeset_name, total),
			'\0', 4 - (total & 3));
		total = (total + 3) & ~3;
	      }
	    iov[2 + elem + offset].iov_len = total;
	    idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    break;

	  case _NL_ITEM_INDEX (_NL_CTYPE_INDIGITS_MB_LEN):
	    iov[2 + elem + offset].iov_base = alloca (sizeof (uint32_t));
	    iov[2 + elem + offset].iov_len = sizeof (uint32_t);
	    *(uint32_t *) iov[2 + elem + offset].iov_base =
	      ctype->mbdigits_act / 10;
	    idx[elem + 1] = idx[elem] + sizeof (uint32_t);
	    break;

	  case _NL_ITEM_INDEX (_NL_CTYPE_INDIGITS_WC_LEN):
	    /* Align entries.  */
	    iov[2 + elem + offset].iov_base = (void *) nulbytes;
	    iov[2 + elem + offset].iov_len = (4 - idx[elem] % 4) % 4;
	    idx[elem] += iov[2 + elem + offset].iov_len;
	    ++offset;

	    iov[2 + elem + offset].iov_base = alloca (sizeof (uint32_t));
	    iov[2 + elem + offset].iov_len = sizeof (uint32_t);
	    *(uint32_t *) iov[2 + elem + offset].iov_base =
	      ctype->wcdigits_act / 10;
	    idx[elem + 1] = idx[elem] + sizeof (uint32_t);
	    break;

	  case _NL_ITEM_INDEX (_NL_CTYPE_INDIGITS0_MB) ... _NL_ITEM_INDEX (_NL_CTYPE_INDIGITS9_MB):
	    /* Compute the length of all possible characters.  For INDIGITS
	       there might be more than one.  We simply concatenate all of
	       them with a NUL byte following.  The NUL byte wouldn't be
	       necessary but it makes it easier for the user.  */
	    total = 0;

	    for (cnt = elem - _NL_ITEM_INDEX (_NL_CTYPE_INDIGITS0_MB);
		 cnt < ctype->mbdigits_act; cnt += 10)
	      total += ctype->mbdigits[cnt]->nbytes + 1;
	    iov[2 + elem + offset].iov_base = (char *) alloca (total);
	    iov[2 + elem + offset].iov_len = total;

	    cp = iov[2 + elem + offset].iov_base;
	    for (cnt = elem - _NL_ITEM_INDEX (_NL_CTYPE_INDIGITS0_MB);
		 cnt < ctype->mbdigits_act; cnt += 10)
	      {
		cp = mempcpy (cp, ctype->mbdigits[cnt]->bytes,
			      ctype->mbdigits[cnt]->nbytes);
		*cp++ = '\0';
	      }
	    idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    break;

	  case _NL_ITEM_INDEX (_NL_CTYPE_OUTDIGIT0_MB) ... _NL_ITEM_INDEX (_NL_CTYPE_OUTDIGIT9_MB):
	    /* Compute the length of all possible characters.  For INDIGITS
	       there might be more than one.  We simply concatenate all of
	       them with a NUL byte following.  The NUL byte wouldn't be
	       necessary but it makes it easier for the user.  */
	    cnt = elem - _NL_ITEM_INDEX (_NL_CTYPE_OUTDIGIT0_MB);
	    total = ctype->mboutdigits[cnt]->nbytes + 1;
	    iov[2 + elem + offset].iov_base = (char *) alloca (total);
	    iov[2 + elem + offset].iov_len = total;

	    *(char *) mempcpy (iov[2 + elem + offset].iov_base,
			       ctype->mboutdigits[cnt]->bytes,
			       ctype->mboutdigits[cnt]->nbytes) = '\0';
	    idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    break;

	  case _NL_ITEM_INDEX (_NL_CTYPE_INDIGITS0_WC) ... _NL_ITEM_INDEX (_NL_CTYPE_INDIGITS9_WC):
	    total = ctype->wcdigits_act / 10;

	    iov[2 + elem + offset].iov_base =
	      (uint32_t *) alloca (total * sizeof (uint32_t));
	    iov[2 + elem + offset].iov_len = total * sizeof (uint32_t);

	    for (cnt = elem - _NL_ITEM_INDEX (_NL_CTYPE_INDIGITS0_WC);
		 cnt < ctype->wcdigits_act; cnt += 10)
	      ((uint32_t *) iov[2 + elem + offset].iov_base)[cnt / 10]
		= ctype->wcdigits[cnt];
	    idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    break;

	  case _NL_ITEM_INDEX (_NL_CTYPE_OUTDIGIT0_WC):
	    /* Align entries.  */
	    iov[2 + elem + offset].iov_base = (void *) nulbytes;
	    iov[2 + elem + offset].iov_len = (4 - idx[elem] % 4) % 4;
	    idx[elem] += iov[2 + elem + offset].iov_len;
	    ++offset;
	    /* FALLTRHOUGH */

	  case _NL_ITEM_INDEX (_NL_CTYPE_OUTDIGIT1_WC) ... _NL_ITEM_INDEX (_NL_CTYPE_OUTDIGIT9_WC):
	    cnt = elem - _NL_ITEM_INDEX (_NL_CTYPE_OUTDIGIT0_WC);
	    iov[2 + elem + offset].iov_base = &ctype->wcoutdigits[cnt];
	    iov[2 + elem + offset].iov_len = sizeof (uint32_t);
	    idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    break;

	  case _NL_ITEM_INDEX(_NL_CTYPE_TRANSLIT_DEFAULT_MISSING_LEN):
	    default_missing_len = (ctype->default_missing
				   ? wcslen ((wchar_t *)ctype->default_missing)
				   : 0);
	    iov[2 + elem + offset].iov_base = &default_missing_len;
	    iov[2 + elem + offset].iov_len = sizeof (uint32_t);
	    idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    break;

	  case _NL_ITEM_INDEX(_NL_CTYPE_TRANSLIT_DEFAULT_MISSING):
	    iov[2 + elem + offset].iov_base =
	      ctype->default_missing ?: (uint32_t *) L"";
	    iov[2 + elem + offset].iov_len =
	      wcslen (iov[2 + elem + offset].iov_base);
	    idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    break;

	  case _NL_ITEM_INDEX(_NL_CTYPE_TRANSLIT_IGNORE_LEN):
	    iov[2 + elem + offset].iov_base = &ctype->ntranslit_ignore;
	    iov[2 + elem + offset].iov_len = sizeof (uint32_t);
	    idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    break;

	  case _NL_ITEM_INDEX(_NL_CTYPE_TRANSLIT_IGNORE):
	    {
	      uint32_t *ranges = (uint32_t *) alloca (ctype->ntranslit_ignore
						      * 3 * sizeof (uint32_t));
	      struct translit_ignore_t *runp;

	      iov[2 + elem + offset].iov_base = ranges;
	      iov[2 + elem + offset].iov_len = (ctype->ntranslit_ignore
						* 3 * sizeof (uint32_t));

	      for (runp = ctype->translit_ignore; runp != NULL;
		   runp = runp->next)
		{
		  *ranges++ = runp->from;
		  *ranges++ = runp->to;
		  *ranges++ = runp->step;
		}
	    }
	    /* Remove the following line in case a new entry is added
	       after _NL_CTYPE_TRANSLIT_DEFAULT_MISSING_LEN.  */
	    if (elem < nelems)
	      idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    break;

	  default:
	    assert (! "unknown CTYPE element");
	  }
      else
	{
	  /* Handle extra maps.  */
	  if (oldstyle_tables)
	    {
	      size_t nr = (elem - _NL_ITEM_INDEX (_NL_NUM_LC_CTYPE)) + 2;

	      iov[2 + elem + offset].iov_base = ctype->map32[nr];
	      iov[2 + elem + offset].iov_len = ((ctype->plane_size
						 * ctype->plane_cnt)
						* sizeof (uint32_t));

	      idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    }
	  else
	    {
	      size_t nr = elem - _NL_ITEM_INDEX (_NL_NUM_LC_CTYPE);
	      if (nr < ctype->nr_charclass)
		{
		  iov[2 + elem + offset] = ctype->class_3level[nr];
		}
	      else
		{
		  nr -= ctype->nr_charclass;
		  assert (nr < ctype->map_collection_nr);
		  iov[2 + elem + offset] = ctype->map_3level[nr];
		}
	      idx[elem + 1] = idx[elem] + iov[2 + elem + offset].iov_len;
	    }
	}
    }

  assert (2 + elem + offset == (nelems + ctype->nr_charclass
				+ ctype->map_collection_nr + 2 + 2));

  write_locale_data (output_path, "LC_CTYPE", 2 + elem + offset, iov);
}


/* Local functions.  */
static void
ctype_class_new (struct linereader *lr, struct locale_ctype_t *ctype,
		 const char *name)
{
  size_t cnt;

  for (cnt = 0; cnt < ctype->nr_charclass; ++cnt)
    if (strcmp (ctype->classnames[cnt], name) == 0)
      break;

  if (cnt < ctype->nr_charclass)
    {
      lr_error (lr, _("character class `%s' already defined"), name);
      return;
    }

  if (ctype->nr_charclass == MAX_NR_CHARCLASS)
    /* Exit code 2 is prescribed in P1003.2b.  */
    error (2, 0, _("\
implementation limit: no more than %Zd character classes allowed"),
	   MAX_NR_CHARCLASS);

  ctype->classnames[ctype->nr_charclass++] = name;
}


static void
ctype_map_new (struct linereader *lr, struct locale_ctype_t *ctype,
	       const char *name, struct charmap_t *charmap)
{
  size_t max_chars = 0;
  size_t cnt;

  for (cnt = 0; cnt < ctype->map_collection_nr; ++cnt)
    {
      if (strcmp (ctype->mapnames[cnt], name) == 0)
	break;

      if (max_chars < ctype->map_collection_max[cnt])
	max_chars = ctype->map_collection_max[cnt];
    }

  if (cnt < ctype->map_collection_nr)
    {
      lr_error (lr, _("character map `%s' already defined"), name);
      return;
    }

  if (ctype->map_collection_nr == MAX_NR_CHARMAP)
    /* Exit code 2 is prescribed in P1003.2b.  */
    error (2, 0, _("\
implementation limit: no more than %d character maps allowed"),
	   MAX_NR_CHARMAP);

  ctype->mapnames[cnt] = name;

  if (max_chars == 0)
    ctype->map_collection_max[cnt] = charmap->mb_cur_max == 1 ? 256 : 512;
  else
    ctype->map_collection_max[cnt] = max_chars;

  ctype->map_collection[cnt] = (uint32_t *)
    xcalloc (sizeof (uint32_t), ctype->map_collection_max[cnt]);
  ctype->map_collection_act[cnt] = 256;

  ++ctype->map_collection_nr;
}


/* We have to be prepared that TABLE, MAX, and ACT can be NULL.  This
   is possible if we only want to extend the name array.  */
static uint32_t *
find_idx (struct locale_ctype_t *ctype, uint32_t **table, size_t *max,
	  size_t *act, uint32_t idx)
{
  size_t cnt;

  if (idx < 256)
    return table == NULL ? NULL : &(*table)[idx];

  for (cnt = 256; cnt < ctype->charnames_act; ++cnt)
    if (ctype->charnames[cnt] == idx)
      break;

  /* We have to distinguish two cases: the name is found or not.  */
  if (cnt == ctype->charnames_act)
    {
      /* Extend the name array.  */
      if (ctype->charnames_act == ctype->charnames_max)
	{
	  ctype->charnames_max *= 2;
	  ctype->charnames = (uint32_t *)
	    xrealloc (ctype->charnames,
		      sizeof (uint32_t) * ctype->charnames_max);
	}
      ctype->charnames[ctype->charnames_act++] = idx;
    }

  if (table == NULL)
    /* We have done everything we are asked to do.  */
    return NULL;

  if (cnt >= *act)
    {
      if (cnt >= *max)
	{
	  size_t old_max = *max;
	  do
	    *max *= 2;
	  while (*max <= cnt);

	  *table =
	    (uint32_t *) xrealloc (*table, *max * sizeof (uint32_t));
	  memset (&(*table)[old_max], '\0',
		  (*max - old_max) * sizeof (uint32_t));
	}

      *act = cnt + 1;
    }

  return &(*table)[cnt];
}


static int
get_character (struct token *now, struct charmap_t *charmap,
	       struct repertoire_t *repertoire,
	       struct charseq **seqp, uint32_t *wchp)
{
  if (now->tok == tok_bsymbol)
    {
      /* This will hopefully be the normal case.  */
      *wchp = repertoire_find_value (repertoire, now->val.str.startmb,
				     now->val.str.lenmb);
      *seqp = charmap_find_value (charmap, now->val.str.startmb,
				  now->val.str.lenmb);
    }
  else if (now->tok == tok_ucs4)
    {
      char utmp[10];

      snprintf (utmp, sizeof (utmp), "U%08X", now->val.ucs4);
      *seqp = charmap_find_value (charmap, utmp, 9);

      if (*seqp == NULL)
	*seqp = repertoire_find_seq (repertoire, now->val.ucs4);

      if (*seqp == NULL)
	{
	  /* Compute the value in the charmap from the UCS value.  */
	  const char *symbol = repertoire_find_symbol (repertoire,
						       now->val.ucs4);

	  if (symbol == NULL)
	    *seqp = NULL;
	  else
	    *seqp = charmap_find_value (charmap, symbol, strlen (symbol));

	  if (*seqp == NULL)
	    {
	      if (repertoire != NULL)
		{
		  /* Insert a negative entry.  */
		  static const struct charseq negative
		    = { .ucs4 = ILLEGAL_CHAR_VALUE };
		  uint32_t *newp = obstack_alloc (&repertoire->mem_pool,
						  sizeof (uint32_t));
		  *newp = now->val.ucs4;

		  insert_entry (&repertoire->seq_table, newp,
				sizeof (uint32_t), (void *) &negative);
		}
	    }
	  else
	    (*seqp)->ucs4 = now->val.ucs4;
	}
      else if ((*seqp)->ucs4 != now->val.ucs4)
	*seqp = NULL;

      *wchp = now->val.ucs4;
    }
  else if (now->tok == tok_charcode)
    {
      /* We must map from the byte code to UCS4.  */
      *seqp = charmap_find_symbol (charmap, now->val.str.startmb,
				   now->val.str.lenmb);

      if (*seqp == NULL)
	*wchp = ILLEGAL_CHAR_VALUE;
      else
	{
	  if ((*seqp)->ucs4 == UNINITIALIZED_CHAR_VALUE)
	    (*seqp)->ucs4 = repertoire_find_value (repertoire, (*seqp)->name,
						   strlen ((*seqp)->name));
	  *wchp = (*seqp)->ucs4;
	}
    }
  else
    return 1;

  return 0;
}


/* Ellipsis like in `<foo123>..<foo12a>' or `<j1234>....<j1245>' and
   the .(2). counterparts.  */
static void
charclass_symbolic_ellipsis (struct linereader *ldfile,
			     struct locale_ctype_t *ctype,
			     struct charmap_t *charmap,
			     struct repertoire_t *repertoire,
			     struct token *now,
			     const char *last_str,
			     unsigned long int class256_bit,
			     unsigned long int class_bit, int base,
			     int ignore_content, int handle_digits, int step)
{
  const char *nowstr = now->val.str.startmb;
  char tmp[now->val.str.lenmb + 1];
  const char *cp;
  char *endp;
  unsigned long int from;
  unsigned long int to;

  /* We have to compute the ellipsis values using the symbolic names.  */
  assert (last_str != NULL);

  if (strlen (last_str) != now->val.str.lenmb)
    {
    invalid_range:
      lr_error (ldfile,
		_("`%s' and `%.*s' are no valid names for symbolic range"),
		last_str, (int) now->val.str.lenmb, nowstr);
      return;
    }

  if (memcmp (last_str, nowstr, now->val.str.lenmb) == 0)
    /* Nothing to do, the names are the same.  */
    return;

  for (cp = last_str; *cp == *(nowstr + (cp - last_str)); ++cp)
    ;

  errno = 0;
  from = strtoul (cp, &endp, base);
  if ((from == UINT_MAX && errno == ERANGE) || *endp != '\0')
    goto invalid_range;

  to = strtoul (nowstr + (cp - last_str), &endp, base);
  if ((to == UINT_MAX && errno == ERANGE)
      || (endp - nowstr) != now->val.str.lenmb || from >= to)
    goto invalid_range;

  /* OK, we have a range FROM - TO.  Now we can create the symbolic names.  */
  if (!ignore_content)
    {
      now->val.str.startmb = tmp;
      while ((from += step) <= to)
	{
	  struct charseq *seq;
	  uint32_t wch;

	  sprintf (tmp, (base == 10 ? "%.*s%0*d" : "%.*s%0*X"), cp - last_str,
		   last_str, now->val.str.lenmb - (cp - last_str), from);

	  get_character (now, charmap, repertoire, &seq, &wch);

	  if (seq != NULL && seq->nbytes == 1)
	    /* Yep, we can store information about this byte sequence.  */
	    ctype->class256_collection[seq->bytes[0]] |= class256_bit;

	  if (wch != ILLEGAL_CHAR_VALUE && class_bit != 0)
	    /* We have the UCS4 position.  */
	    *find_idx (ctype, &ctype->class_collection,
		       &ctype->class_collection_max,
		       &ctype->class_collection_act, wch) |= class_bit;

	  if (handle_digits == 1)
	    {
	      /* We must store the digit values.  */
	      if (ctype->mbdigits_act == ctype->mbdigits_max)
		{
		  ctype->mbdigits_max *= 2;
		  ctype->mbdigits = xrealloc (ctype->mbdigits,
					      (ctype->mbdigits_max
					       * sizeof (char *)));
		  ctype->wcdigits_max *= 2;
		  ctype->wcdigits = xrealloc (ctype->wcdigits,
					      (ctype->wcdigits_max
					       * sizeof (uint32_t)));
		}

	      ctype->mbdigits[ctype->mbdigits_act++] = seq;
	      ctype->wcdigits[ctype->wcdigits_act++] = wch;
	    }
	  else if (handle_digits == 2)
	    {
	      /* We must store the digit values.  */
	      if (ctype->outdigits_act >= 10)
		{
		  lr_error (ldfile, _("\
%s: field `%s' does not contain exactly ten entries"),
			    "LC_CTYPE", "outdigit");
		  return;
		}

	      ctype->mboutdigits[ctype->outdigits_act] = seq;
	      ctype->wcoutdigits[ctype->outdigits_act] = wch;
	      ++ctype->outdigits_act;
	    }
	}
    }
}


/* Ellipsis like in `<U1234>..<U2345>' or `<U1234>..(2)..<U2345>'.  */
static void
charclass_ucs4_ellipsis (struct linereader *ldfile,
			 struct locale_ctype_t *ctype,
			 struct charmap_t *charmap,
			 struct repertoire_t *repertoire,
			 struct token *now, uint32_t last_wch,
			 unsigned long int class256_bit,
			 unsigned long int class_bit, int ignore_content,
			 int handle_digits, int step)
{
  if (last_wch > now->val.ucs4)
    {
      lr_error (ldfile, _("\
to-value <U%0*X> of range is smaller than from-value <U%0*X>"),
		(now->val.ucs4 | last_wch) < 65536 ? 4 : 8, now->val.ucs4,
		(now->val.ucs4 | last_wch) < 65536 ? 4 : 8, last_wch);
      return;
    }

  if (!ignore_content)
    while ((last_wch += step) <= now->val.ucs4)
      {
	/* We have to find out whether there is a byte sequence corresponding
	   to this UCS4 value.  */
	struct charseq *seq;
	char utmp[10];

	snprintf (utmp, sizeof (utmp), "U%08X", last_wch);
	seq = charmap_find_value (charmap, utmp, 9);
	if (seq == NULL)
	  {
	    snprintf (utmp, sizeof (utmp), "U%04X", last_wch);
	    seq = charmap_find_value (charmap, utmp, 5);
	  }

	if (seq == NULL)
	  /* Try looking in the repertoire map.  */
	  seq = repertoire_find_seq (repertoire, last_wch);

	/* If this is the first time we look for this sequence create a new
	   entry.  */
	if (seq == NULL)
	  {
	    static const struct charseq negative
	      = { .ucs4 = ILLEGAL_CHAR_VALUE };

	    /* Find the symbolic name for this UCS4 value.  */
	    if (repertoire != NULL)
	      {
		const char *symbol = repertoire_find_symbol (repertoire,
							     last_wch);
		uint32_t *newp = obstack_alloc (&repertoire->mem_pool,
						sizeof (uint32_t));
		*newp = last_wch;

		if (symbol != NULL)
		  /* We have a name, now search the multibyte value.  */
		  seq = charmap_find_value (charmap, symbol, strlen (symbol));

		if (seq == NULL)
		  /* We have to create a fake entry.  */
		  seq = (struct charseq *) &negative;
		else
		  seq->ucs4 = last_wch;

		insert_entry (&repertoire->seq_table, newp, sizeof (uint32_t),
			      seq);
	      }
	    else
	      /* We have to create a fake entry.  */
	      seq = (struct charseq *) &negative;
	  }

	/* We have a name, now search the multibyte value.  */
	if (seq->ucs4 == last_wch && seq->nbytes == 1)
	  /* Yep, we can store information about this byte sequence.  */
	  ctype->class256_collection[(size_t) seq->bytes[0]]
	    |= class256_bit;

	/* And of course we have the UCS4 position.  */
	if (class_bit != 0)
	  *find_idx (ctype, &ctype->class_collection,
		     &ctype->class_collection_max,
		     &ctype->class_collection_act, last_wch) |= class_bit;

	if (handle_digits == 1)
	  {
	    /* We must store the digit values.  */
	    if (ctype->mbdigits_act == ctype->mbdigits_max)
	      {
		ctype->mbdigits_max *= 2;
		ctype->mbdigits = xrealloc (ctype->mbdigits,
					    (ctype->mbdigits_max
					     * sizeof (char *)));
		ctype->wcdigits_max *= 2;
		ctype->wcdigits = xrealloc (ctype->wcdigits,
					    (ctype->wcdigits_max
					     * sizeof (uint32_t)));
	      }

	    ctype->mbdigits[ctype->mbdigits_act++] = (seq->ucs4 == last_wch
						      ? seq : NULL);
	    ctype->wcdigits[ctype->wcdigits_act++] = last_wch;
	  }
	else if (handle_digits == 2)
	  {
	    /* We must store the digit values.  */
	    if (ctype->outdigits_act >= 10)
	      {
		lr_error (ldfile, _("\
%s: field `%s' does not contain exactly ten entries"),
			  "LC_CTYPE", "outdigit");
		return;
	      }

	    ctype->mboutdigits[ctype->outdigits_act] = (seq->ucs4 == last_wch
							? seq : NULL);
	    ctype->wcoutdigits[ctype->outdigits_act] = last_wch;
	    ++ctype->outdigits_act;
	  }
      }
}


/* Ellipsis as in `/xea/x12.../xea/x34'.  */
static void
charclass_charcode_ellipsis (struct linereader *ldfile,
			     struct locale_ctype_t *ctype,
			     struct charmap_t *charmap,
			     struct repertoire_t *repertoire,
			     struct token *now, char *last_charcode,
			     uint32_t last_charcode_len,
			     unsigned long int class256_bit,
			     unsigned long int class_bit, int ignore_content,
			     int handle_digits)
{
  /* First check whether the to-value is larger.  */
  if (now->val.charcode.nbytes != last_charcode_len)
    {
      lr_error (ldfile, _("\
start end end character sequence of range must have the same length"));
      return;
    }

  if (memcmp (last_charcode, now->val.charcode.bytes, last_charcode_len) > 0)
    {
      lr_error (ldfile, _("\
to-value character sequence is smaller than from-value sequence"));
      return;
    }

  if (!ignore_content)
    {
      do
	{
	  /* Increment the byte sequence value.  */
	  struct charseq *seq;
	  uint32_t wch;
	  int i;

	  for (i = last_charcode_len - 1; i >= 0; --i)
	    if (++last_charcode[i] != 0)
	      break;

	  if (last_charcode_len == 1)
	    /* Of course we have the charcode value.  */
	    ctype->class256_collection[(size_t) last_charcode[0]]
	      |= class256_bit;

	  /* Find the symbolic name.  */
	  seq = charmap_find_symbol (charmap, last_charcode,
				     last_charcode_len);
	  if (seq != NULL)
	    {
	      if (seq->ucs4 == UNINITIALIZED_CHAR_VALUE)
		seq->ucs4 = repertoire_find_value (repertoire, seq->name,
						   strlen (seq->name));
	      wch = seq == NULL ? ILLEGAL_CHAR_VALUE : seq->ucs4;

	      if (wch != ILLEGAL_CHAR_VALUE && class_bit != 0)
		*find_idx (ctype, &ctype->class_collection,
			   &ctype->class_collection_max,
			   &ctype->class_collection_act, wch) |= class_bit;
	    }
	  else
	    wch = ILLEGAL_CHAR_VALUE;

	  if (handle_digits == 1)
	    {
	      /* We must store the digit values.  */
	      if (ctype->mbdigits_act == ctype->mbdigits_max)
		{
		  ctype->mbdigits_max *= 2;
		  ctype->mbdigits = xrealloc (ctype->mbdigits,
					      (ctype->mbdigits_max
					       * sizeof (char *)));
		  ctype->wcdigits_max *= 2;
		  ctype->wcdigits = xrealloc (ctype->wcdigits,
					      (ctype->wcdigits_max
					       * sizeof (uint32_t)));
		}

	      seq = xmalloc (sizeof (struct charseq) + last_charcode_len);
	      memcpy ((char *) (seq + 1), last_charcode, last_charcode_len);
	      seq->nbytes = last_charcode_len;

	      ctype->mbdigits[ctype->mbdigits_act++] = seq;
	      ctype->wcdigits[ctype->wcdigits_act++] = wch;
	    }
	  else if (handle_digits == 2)
	    {
	      struct charseq *seq;
	      /* We must store the digit values.  */
	      if (ctype->outdigits_act >= 10)
		{
		  lr_error (ldfile, _("\
%s: field `%s' does not contain exactly ten entries"),
			    "LC_CTYPE", "outdigit");
		  return;
		}

	      seq = xmalloc (sizeof (struct charseq) + last_charcode_len);
	      memcpy ((char *) (seq + 1), last_charcode, last_charcode_len);
	      seq->nbytes = last_charcode_len;

	      ctype->mboutdigits[ctype->outdigits_act] = seq;
	      ctype->wcoutdigits[ctype->outdigits_act] = wch;
	      ++ctype->outdigits_act;
	    }
	}
      while (memcmp (last_charcode, now->val.charcode.bytes,
		     last_charcode_len) != 0);
    }
}


/* Read one transliteration entry.  */
static uint32_t *
read_widestring (struct linereader *ldfile, struct token *now,
		 struct charmap_t *charmap, struct repertoire_t *repertoire)
{
  uint32_t *wstr;

  if (now->tok == tok_default_missing)
    /* The special name "" will denote this case.  */
    wstr = ((uint32_t *) { 0 });
  else if (now->tok == tok_bsymbol)
    {
      /* Get the value from the repertoire.  */
      wstr = (uint32_t *) xmalloc (2 * sizeof (uint32_t));
      wstr[0] = repertoire_find_value (repertoire, now->val.str.startmb,
				       now->val.str.lenmb);
      if (wstr[0] == ILLEGAL_CHAR_VALUE)
	{
	  /* We cannot proceed, we don't know the UCS4 value.  */
	  free (wstr);
	  return NULL;
	}

      wstr[1] = 0;
    }
  else if (now->tok == tok_ucs4)
    {
      wstr = (uint32_t *) xmalloc (2 * sizeof (uint32_t));
      wstr[0] = now->val.ucs4;
      wstr[1] = 0;
    }
  else if (now->tok == tok_charcode)
    {
      /* Argh, we have to convert to the symbol name first and then to the
	 UCS4 value.  */
      struct charseq *seq = charmap_find_symbol (charmap,
						 now->val.str.startmb,
						 now->val.str.lenmb);
      if (seq == NULL)
	/* Cannot find the UCS4 value.  */
	return NULL;

      if (seq->ucs4 == UNINITIALIZED_CHAR_VALUE)
	seq->ucs4 = repertoire_find_value (repertoire, seq->name,
					   strlen (seq->name));
      if (seq->ucs4 == ILLEGAL_CHAR_VALUE)
	/* We cannot proceed, we don't know the UCS4 value.  */
	return NULL;

      wstr = (uint32_t *) xmalloc (2 * sizeof (uint32_t));
      wstr[0] = seq->ucs4;
      wstr[1] = 0;
    }
  else if (now->tok == tok_string)
    {
      wstr = now->val.str.startwc;
      if (wstr == NULL || wstr[0] == 0)
	return NULL;
    }
  else
    {
      if (now->tok != tok_eol && now->tok != tok_eof)
	lr_ignore_rest (ldfile, 0);
      SYNTAX_ERROR (_("%s: syntax error"), "LC_CTYPE");
      return (uint32_t *) -1l;
    }

  return wstr;
}


static void
read_translit_entry (struct linereader *ldfile, struct locale_ctype_t *ctype,
		     struct token *now, struct charmap_t *charmap,
		     struct repertoire_t *repertoire)
{
  uint32_t *from_wstr = read_widestring (ldfile, now, charmap, repertoire);
  struct translit_t *result;
  struct translit_to_t **top;
  struct obstack *ob = &ctype->mempool;
  int first;
  int ignore;

  if (from_wstr == NULL)
    /* There is no valid from string.  */
    return;

  result = (struct translit_t *) obstack_alloc (ob,
						sizeof (struct translit_t));
  result->from = from_wstr;
  result->fname = ldfile->fname;
  result->lineno = ldfile->lineno;
  result->next = NULL;
  result->to = NULL;
  top = &result->to;
  first = 1;
  ignore = 0;

  while (1)
    {
      uint32_t *to_wstr;

      /* Next we have one or more transliterations.  They are
	 separated by semicolons.  */
      now = lr_token (ldfile, charmap, repertoire);

      if (!first && (now->tok == tok_semicolon || now->tok == tok_eol))
	{
	  /* One string read.  */
	  const uint32_t zero = 0;

	  if (!ignore)
	    {
	      obstack_grow (ob, &zero, 4);
	      to_wstr = obstack_finish (ob);

	      *top = obstack_alloc (ob, sizeof (struct translit_to_t));
	      (*top)->str = to_wstr;
	      (*top)->next = NULL;
	    }

	  if (now->tok == tok_eol)
	    {
	      result->next = ctype->translit;
	      ctype->translit = result;
	      return;
	    }

	  if (!ignore)
	    top = &(*top)->next;
	  ignore = 0;
	}
      else
	{
	  to_wstr = read_widestring (ldfile, now, charmap, repertoire);
	  if (to_wstr == (uint32_t *) -1l)
	    {
	      /* An error occurred.  */
	      obstack_free (ob, result);
	      return;
	    }

	  if (to_wstr == NULL)
	    ignore = 1;
	  else
	    /* This value is usable.  */
	    obstack_grow (ob, to_wstr, wcslen ((wchar_t *) to_wstr) * 4);

	  first = 0;
	}
    }
}


static void
read_translit_ignore_entry (struct linereader *ldfile,
			    struct locale_ctype_t *ctype,
			    struct charmap_t *charmap,
			    struct repertoire_t *repertoire)
{
  /* We expect a semicolon-separated list of characters we ignore.  We are
     only interested in the wide character definitions.  These must be
     single characters, possibly defining a range when an ellipsis is used.  */
  while (1)
    {
      struct token *now = lr_token (ldfile, charmap, repertoire);
      struct translit_ignore_t *newp;
      uint32_t from;

      if (now->tok == tok_eol || now->tok == tok_eof)
	{
	  lr_error (ldfile,
		    _("premature end of `translit_ignore' definition"));
	  return;
	}

      if (now->tok != tok_bsymbol && now->tok != tok_ucs4)
	{
	  lr_error (ldfile, _("syntax error"));
	  lr_ignore_rest (ldfile, 0);
	  return;
	}

      if (now->tok == tok_ucs4)
	from = now->val.ucs4;
      else
	/* Try to get the value.  */
	from = repertoire_find_value (repertoire, now->val.str.startmb,
				      now->val.str.lenmb);

      if (from == ILLEGAL_CHAR_VALUE)
	{
	  lr_error (ldfile, "invalid character name");
	  newp = NULL;
	}
      else
	{
	  newp = (struct translit_ignore_t *)
	    obstack_alloc (&ctype->mempool, sizeof (struct translit_ignore_t));
	  newp->from = from;
	  newp->to = from;
	  newp->step = 1;

	  newp->next = ctype->translit_ignore;
	  ctype->translit_ignore = newp;
	}

      /* Now we expect either a semicolon, an ellipsis, or the end of the
	 line.  */
      now = lr_token (ldfile, charmap, repertoire);

      if (now->tok == tok_ellipsis2 || now->tok == tok_ellipsis2_2)
	{
	  /* XXX Should we bother implementing `....'?  `...' certainly
	     will not be implemented.  */
	  uint32_t to;
	  int step = now->tok == tok_ellipsis2_2 ? 2 : 1;

	  now = lr_token (ldfile, charmap, repertoire);

	  if (now->tok == tok_eol || now->tok == tok_eof)
	    {
	      lr_error (ldfile,
			_("premature end of `translit_ignore' definition"));
	      return;
	    }

	  if (now->tok != tok_bsymbol && now->tok != tok_ucs4)
	    {
	      lr_error (ldfile, _("syntax error"));
	      lr_ignore_rest (ldfile, 0);
	      return;
	    }

	  if (now->tok == tok_ucs4)
	    to = now->val.ucs4;
	  else
	    /* Try to get the value.  */
	    to = repertoire_find_value (repertoire, now->val.str.startmb,
					now->val.str.lenmb);

	  if (to == ILLEGAL_CHAR_VALUE)
	    lr_error (ldfile, "invalid character name");
	  else
	    {
	      /* Make sure the `to'-value is larger.  */
	      if (to >= from)
		{
		  newp->to = to;
		  newp->step = step;
		}
	      else
		lr_error (ldfile, _("\
to-value <U%0*X> of range is smaller than from-value <U%0*X>"),
			  (to | from) < 65536 ? 4 : 8, to,
			  (to | from) < 65536 ? 4 : 8, from);
	    }

	  /* And the next token.  */
	  now = lr_token (ldfile, charmap, repertoire);
	}

      if (now->tok == tok_eol || now->tok == tok_eof)
	/* We are done.  */
	return;

      if (now->tok == tok_semicolon)
	/* Next round.  */
	continue;

      /* If we come here something is wrong.  */
      lr_error (ldfile, _("syntax error"));
      lr_ignore_rest (ldfile, 0);
      return;
    }
}


/* The parser for the LC_CTYPE section of the locale definition.  */
void
ctype_read (struct linereader *ldfile, struct localedef_t *result,
	    struct charmap_t *charmap, const char *repertoire_name,
	    int ignore_content)
{
  struct repertoire_t *repertoire = NULL;
  struct locale_ctype_t *ctype;
  struct token *now;
  enum token_t nowtok;
  size_t cnt;
  struct charseq *last_seq;
  uint32_t last_wch = 0;
  enum token_t last_token;
  enum token_t ellipsis_token;
  int step;
  char last_charcode[16];
  size_t last_charcode_len = 0;
  const char *last_str = NULL;
  int mapidx;

  /* Get the repertoire we have to use.  */
  if (repertoire_name != NULL)
    repertoire = repertoire_read (repertoire_name);

  /* The rest of the line containing `LC_CTYPE' must be free.  */
  lr_ignore_rest (ldfile, 1);


  do
    {
      now = lr_token (ldfile, charmap, NULL);
      nowtok = now->tok;
    }
  while (nowtok == tok_eol);

  /* If we see `copy' now we are almost done.  */
  if (nowtok == tok_copy)
    {
      handle_copy (ldfile, charmap, repertoire_name, result, tok_lc_ctype,
		   LC_CTYPE, "LC_CTYPE", ignore_content);
      return;
    }

  /* Prepare the data structures.  */
  ctype_startup (ldfile, result, charmap, ignore_content);
  ctype = result->categories[LC_CTYPE].ctype;

  /* Remember the repertoire we use.  */
  if (!ignore_content)
    ctype->repertoire = repertoire;

  while (1)
    {
      unsigned long int class_bit = 0;
      unsigned long int class256_bit = 0;
      int handle_digits = 0;

      /* Of course we don't proceed beyond the end of file.  */
      if (nowtok == tok_eof)
	break;

      /* Ingore empty lines.  */
      if (nowtok == tok_eol)
	{
	  now = lr_token (ldfile, charmap, NULL);
	  nowtok = now->tok;
	  continue;
	}

      switch (nowtok)
	{
	case tok_charclass:
	  now = lr_token (ldfile, charmap, NULL);
	  while (now->tok == tok_ident || now->tok == tok_string)
	    {
	      ctype_class_new (ldfile, ctype, now->val.str.startmb);
	      now = lr_token (ldfile, charmap, NULL);
	      if (now->tok != tok_semicolon)
		break;
	      now = lr_token (ldfile, charmap, NULL);
	    }
	  if (now->tok != tok_eol)
	    SYNTAX_ERROR (_("\
%s: syntax error in definition of new character class"), "LC_CTYPE");
	  break;

	case tok_charconv:
	  now = lr_token (ldfile, charmap, NULL);
	  while (now->tok == tok_ident || now->tok == tok_string)
	    {
	      ctype_map_new (ldfile, ctype, now->val.str.startmb, charmap);
	      now = lr_token (ldfile, charmap, NULL);
	      if (now->tok != tok_semicolon)
		break;
	      now = lr_token (ldfile, charmap, NULL);
	    }
	  if (now->tok != tok_eol)
	    SYNTAX_ERROR (_("\
%s: syntax error in definition of new character map"), "LC_CTYPE");
	  break;

	case tok_class:
	  /* Ignore the rest of the line if we don't need the input of
	     this line.  */
	  if (ignore_content)
	    {
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }

	  /* We simply forget the `class' keyword and use the following
	     operand to determine the bit.  */
	  now = lr_token (ldfile, charmap, NULL);
	  if (now->tok == tok_ident || now->tok == tok_string)
	    {
	      /* Must can be one of the predefined class names.  */
	      for (cnt = 0; cnt < ctype->nr_charclass; ++cnt)
		if (strcmp (ctype->classnames[cnt], now->val.str.startmb) == 0)
		  break;
	      if (cnt >= ctype->nr_charclass)
		{
#ifdef PREDEFINED_CLASSES
		  if (now->val.str.lenmb == 8
		      && memcmp ("special1", now->val.str.startmb, 8) == 0)
		    class_bit = _ISwspecial1;
		  else if (now->val.str.lenmb == 8
		      && memcmp ("special2", now->val.str.startmb, 8) == 0)
		    class_bit = _ISwspecial2;
		  else if (now->val.str.lenmb == 8
		      && memcmp ("special3", now->val.str.startmb, 8) == 0)
		    class_bit = _ISwspecial3;
		  else
#endif
		    {
		      /* OK, it's a new class.  */
		      ctype_class_new (ldfile, ctype, now->val.str.startmb);

		      class_bit = _ISwbit (ctype->nr_charclass - 1);
		    }
		}
	      else
		{
		  class_bit = _ISwbit (cnt);

		  free (now->val.str.startmb);
		}
	    }
	  else if (now->tok == tok_digit)
	    goto handle_tok_digit;
	  else if (now->tok < tok_upper || now->tok > tok_blank)
	    goto err_label;
	  else
	    {
	      class_bit = BITw (now->tok);
	      class256_bit = BIT (now->tok);
	    }

	  /* The next character must be a semicolon.  */
	  now = lr_token (ldfile, charmap, NULL);
	  if (now->tok != tok_semicolon)
	    goto err_label;
	  goto read_charclass;

	case tok_upper:
	case tok_lower:
	case tok_alpha:
	case tok_alnum:
	case tok_space:
	case tok_cntrl:
	case tok_punct:
	case tok_graph:
	case tok_print:
	case tok_xdigit:
	case tok_blank:
	  /* Ignore the rest of the line if we don't need the input of
	     this line.  */
	  if (ignore_content)
	    {
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }

	  class_bit = BITw (now->tok);
	  class256_bit = BIT (now->tok);
	  handle_digits = 0;
	read_charclass:
	  ctype->class_done |= class_bit;
	  last_token = tok_none;
	  ellipsis_token = tok_none;
	  step = 1;
	  now = lr_token (ldfile, charmap, NULL);
	  while (now->tok != tok_eol && now->tok != tok_eof)
	    {
	      uint32_t wch;
	      struct charseq *seq;

	      if (ellipsis_token == tok_none)
		{
		  if (get_character (now, charmap, repertoire, &seq, &wch))
		    goto err_label;

		  if (!ignore_content && seq != NULL && seq->nbytes == 1)
		    /* Yep, we can store information about this byte
		       sequence.  */
		    ctype->class256_collection[seq->bytes[0]] |= class256_bit;

		  if (!ignore_content && wch != ILLEGAL_CHAR_VALUE
		      && class_bit != 0)
		    /* We have the UCS4 position.  */
		    *find_idx (ctype, &ctype->class_collection,
			       &ctype->class_collection_max,
			       &ctype->class_collection_act, wch) |= class_bit;

		  last_token = now->tok;
		  /* Terminate the string.  */
		  if (last_token == tok_bsymbol)
		    {
		      now->val.str.startmb[now->val.str.lenmb] = '\0';
		      last_str = now->val.str.startmb;
		    }
		  else
		    last_str = NULL;
		  last_seq = seq;
		  last_wch = wch;
		  memcpy (last_charcode, now->val.charcode.bytes, 16);
		  last_charcode_len = now->val.charcode.nbytes;

		  if (!ignore_content && handle_digits == 1)
		    {
		      /* We must store the digit values.  */
		      if (ctype->mbdigits_act == ctype->mbdigits_max)
			{
			  ctype->mbdigits_max += 10;
			  ctype->mbdigits = xrealloc (ctype->mbdigits,
						      (ctype->mbdigits_max
						       * sizeof (char *)));
			  ctype->wcdigits_max += 10;
			  ctype->wcdigits = xrealloc (ctype->wcdigits,
						      (ctype->wcdigits_max
						       * sizeof (uint32_t)));
			}

		      ctype->mbdigits[ctype->mbdigits_act++] = seq;
		      ctype->wcdigits[ctype->wcdigits_act++] = wch;
		    }
		  else if (!ignore_content && handle_digits == 2)
		    {
		      /* We must store the digit values.  */
		      if (ctype->outdigits_act >= 10)
			{
			  lr_error (ldfile, _("\
%s: field `%s' does not contain exactly ten entries"),
			    "LC_CTYPE", "outdigit");
			  goto err_label;
			}

		      ctype->mboutdigits[ctype->outdigits_act] = seq;
		      ctype->wcoutdigits[ctype->outdigits_act] = wch;
		      ++ctype->outdigits_act;
		    }
		}
	      else
		{
		  /* Now it gets complicated.  We have to resolve the
		     ellipsis problem.  First we must distinguish between
		     the different kind of ellipsis and this must match the
		     tokens we have seen.  */
		  assert (last_token != tok_none);

		  if (last_token != now->tok)
		    {
		      lr_error (ldfile, _("\
ellipsis range must be marked by two operands of same type"));
		      lr_ignore_rest (ldfile, 0);
		      break;
		    }

		  if (last_token == tok_bsymbol)
		    {
		      if (ellipsis_token == tok_ellipsis3)
			lr_error (ldfile, _("with symbolic name range values \
the absolute ellipsis `...' must not be used"));

		      charclass_symbolic_ellipsis (ldfile, ctype, charmap,
						   repertoire, now, last_str,
						   class256_bit, class_bit,
						   (ellipsis_token
						    == tok_ellipsis4
						    ? 10 : 16),
						   ignore_content,
						   handle_digits, step);
		    }
		  else if (last_token == tok_ucs4)
		    {
		      if (ellipsis_token != tok_ellipsis2)
			lr_error (ldfile, _("\
with UCS range values one must use the hexadecimal symbolic ellipsis `..'"));

		      charclass_ucs4_ellipsis (ldfile, ctype, charmap,
					       repertoire, now, last_wch,
					       class256_bit, class_bit,
					       ignore_content, handle_digits,
					       step);
		    }
		  else
		    {
		      assert (last_token == tok_charcode);

		      if (ellipsis_token != tok_ellipsis3)
			lr_error (ldfile, _("\
with character code range values one must use the absolute ellipsis `...'"));

		      charclass_charcode_ellipsis (ldfile, ctype, charmap,
						   repertoire, now,
						   last_charcode,
						   last_charcode_len,
						   class256_bit, class_bit,
						   ignore_content,
						   handle_digits);
		    }

		  /* Now we have used the last value.  */
		  last_token = tok_none;
		}

	      /* Next we expect a semicolon or the end of the line.  */
	      now = lr_token (ldfile, charmap, NULL);
	      if (now->tok == tok_eol || now->tok == tok_eof)
		break;

	      if (last_token != tok_none
		  && now->tok >= tok_ellipsis2 && now->tok <= tok_ellipsis4_2)
		{
		  if (now->tok == tok_ellipsis2_2)
		    {
		      now->tok = tok_ellipsis2;
		      step = 2;
		    }
		  else if (now->tok == tok_ellipsis4_2)
		    {
		      now->tok = tok_ellipsis4;
		      step = 2;
		    }

		  ellipsis_token = now->tok;

		  now = lr_token (ldfile, charmap, NULL);
		  continue;
		}

	      if (now->tok != tok_semicolon)
		goto err_label;

	      /* And get the next character.  */
	      now = lr_token (ldfile, charmap, NULL);

	      ellipsis_token = tok_none;
	      step = 1;
	    }
	  break;

	case tok_digit:
	  /* Ignore the rest of the line if we don't need the input of
	     this line.  */
	  if (ignore_content)
	    {
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }

	handle_tok_digit:
	  class_bit = _ISwdigit;
	  class256_bit = _ISdigit;
	  handle_digits = 1;
	  goto read_charclass;

	case tok_outdigit:
	  /* Ignore the rest of the line if we don't need the input of
	     this line.  */
	  if (ignore_content)
	    {
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }

	  if (ctype->outdigits_act != 0)
	    lr_error (ldfile, _("\
%s: field `%s' declared more than once"),
		      "LC_CTYPE", "outdigit");
	  class_bit = 0;
	  class256_bit = 0;
	  handle_digits = 2;
	  goto read_charclass;

	case tok_toupper:
	  /* Ignore the rest of the line if we don't need the input of
	     this line.  */
	  if (ignore_content)
	    {
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }

	  mapidx = 0;
	  goto read_mapping;

	case tok_tolower:
	  /* Ignore the rest of the line if we don't need the input of
	     this line.  */
	  if (ignore_content)
	    {
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }

	  mapidx = 1;
	  goto read_mapping;

	case tok_map:
	  /* Ignore the rest of the line if we don't need the input of
	     this line.  */
	  if (ignore_content)
	    {
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }

	  /* We simply forget the `map' keyword and use the following
	     operand to determine the mapping.  */
	  now = lr_token (ldfile, charmap, NULL);
	  if (now->tok == tok_ident || now->tok == tok_string)
	    {
	      size_t cnt;

	      for (cnt = 2; cnt < ctype->map_collection_nr; ++cnt)
		if (strcmp (now->val.str.startmb, ctype->mapnames[cnt]) == 0)
		  break;

	      if (cnt < ctype->map_collection_nr)
		free (now->val.str.startmb);
	      else
		/* OK, it's a new map.  */
		ctype_map_new (ldfile, ctype, now->val.str.startmb, charmap);

	      mapidx = cnt;
	    }
	  else if (now->tok < tok_toupper || now->tok > tok_tolower)
	    goto err_label;
	  else
	    mapidx = now->tok - tok_toupper;

	  now = lr_token (ldfile, charmap, NULL);
	  /* This better should be a semicolon.  */
	  if (now->tok != tok_semicolon)
	    goto err_label;

	read_mapping:
	  /* Test whether this mapping was already defined.  */
	  if (ctype->tomap_done[mapidx])
	    {
	      lr_error (ldfile, _("duplicated definition for mapping `%s'"),
			ctype->mapnames[mapidx]);
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }
	  ctype->tomap_done[mapidx] = 1;

	  now = lr_token (ldfile, charmap, NULL);
	  while (now->tok != tok_eol && now->tok != tok_eof)
	    {
	      struct charseq *from_seq;
	      uint32_t from_wch;
	      struct charseq *to_seq;
	      uint32_t to_wch;

	      /* Every pair starts with an opening brace.  */
	      if (now->tok != tok_open_brace)
		goto err_label;

	      /* Next comes the from-value.  */
	      now = lr_token (ldfile, charmap, NULL);
	      if (get_character (now, charmap, repertoire, &from_seq,
				 &from_wch) != 0)
		goto err_label;

	      /* The next is a comma.  */
	      now = lr_token (ldfile, charmap, NULL);
	      if (now->tok != tok_comma)
		goto err_label;

	      /* And the other value.  */
	      now = lr_token (ldfile, charmap, NULL);
	      if (get_character (now, charmap, repertoire, &to_seq,
				 &to_wch) != 0)
		goto err_label;

	      /* And the last thing is the closing brace.  */
	      now = lr_token (ldfile, charmap, NULL);
	      if (now->tok != tok_close_brace)
		goto err_label;

	      if (!ignore_content)
		{
		  if (mapidx < 2 && from_seq != NULL && to_seq != NULL
		      && from_seq->nbytes == 1 && to_seq->nbytes == 1)
		    /* We can use this value.  */
		    ctype->map256_collection[mapidx][from_seq->bytes[0]]
		      = to_seq->bytes[0];

		  if (from_wch != ILLEGAL_CHAR_VALUE
		      && to_wch != ILLEGAL_CHAR_VALUE)
		    /* Both correct values.  */
		    *find_idx (ctype, &ctype->map_collection[mapidx],
			       &ctype->map_collection_max[mapidx],
			       &ctype->map_collection_act[mapidx],
			       from_wch) = to_wch;
		}

	      /* Now comes a semicolon or the end of the line/file.  */
	      now = lr_token (ldfile, charmap, NULL);
	      if (now->tok == tok_semicolon)
		now = lr_token (ldfile, charmap, NULL);
	    }
	  break;

	case tok_translit_start:
	  /* Ignore the rest of the line if we don't need the input of
	     this line.  */
	  if (ignore_content)
	    {
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }

	  /* The rest of the line better should be empty.  */
	  lr_ignore_rest (ldfile, 1);

	  /* We count here the number of allocated entries in the `translit'
	     array.  */
	  cnt = 0;

	  /* We proceed until we see the `translit_end' token.  */
	  while (now = lr_token (ldfile, charmap, repertoire),
		 now->tok != tok_translit_end && now->tok != tok_eof)
	    {
	      if (now->tok == tok_eol)
		/* Ignore empty lines.  */
		continue;

	      if (now->tok == tok_translit_end)
		{
		  lr_ignore_rest (ldfile, 0);
		  break;
		}

	      if (now->tok == tok_include)
		{
		  /* We have to include locale.  */
		  const char *locale_name;
		  const char *repertoire_name;

		  now = lr_token (ldfile, charmap, NULL);
		  /* This should be a string or an identifier.  In any
		     case something to name a locale.  */
		  if (now->tok != tok_string && now->tok != tok_ident)
		    {
		    translit_syntax:
		      lr_error (ldfile, _("%s: syntax error"), "LC_CTYPE");
		      lr_ignore_rest (ldfile, 0);
		      continue;
		    }
		  locale_name = now->val.str.startmb;

		  /* Next should be a semicolon.  */
		  now = lr_token (ldfile, charmap, NULL);
		  if (now->tok != tok_semicolon)
		    goto translit_syntax;

		  /* Now the repertoire name.  */
		  now = lr_token (ldfile, charmap, NULL);
		  if ((now->tok != tok_string && now->tok != tok_ident)
		      || now->val.str.startmb == NULL)
		    goto translit_syntax;
		  repertoire_name = now->val.str.startmb;

		  /* We must not have more than one `include'.  */
		  if (ctype->translit_copy_locale != NULL)
		    {
		      lr_error (ldfile, _("\
%s: only one `include' instruction allowed"), "LC_CTYPE");
		      lr_ignore_rest (ldfile, 0);
		      continue;
		    }

		  ctype->translit_copy_locale = locale_name;
		  ctype->translit_copy_repertoire = repertoire_name;

		  /* The rest of the line must be empty.  */
		  lr_ignore_rest (ldfile, 1);

		  /* Make sure the locale is read.  */
		  add_to_readlist (LC_CTYPE, ctype->translit_copy_locale,
				   repertoire_name, 1, NULL);
		  continue;
		}
	      else if (now->tok == tok_default_missing)
		{
		  uint32_t *wstr;

		  /* We expect a single character or string as the
		     argument.  */
		  now = lr_token (ldfile, charmap, NULL);
		  wstr = read_widestring (ldfile, now, charmap, repertoire);

		  if (wstr != NULL)
		    {
		      if (ctype->default_missing != NULL)
			{
			  lr_error (ldfile, _("\
%s: duplicate `default_missing' definition"), "LC_CTYPE");
			  error_at_line (0, 0, ctype->default_missing_file,
					 ctype->default_missing_lineno,
					 _("previous definition was here"));
			}
		      else
			{
			  ctype->default_missing = wstr;
			  ctype->default_missing_file = ldfile->fname;
			  ctype->default_missing_lineno = ldfile->lineno;
			}
		    }
		  lr_ignore_rest (ldfile, 1);
		  continue;
		}
	      else if (now->tok == tok_translit_ignore)
		{
		  read_translit_ignore_entry (ldfile, ctype, charmap,
					      repertoire);
		  continue;
		}

	      read_translit_entry (ldfile, ctype, now, charmap, repertoire);
	    }
	  break;

	case tok_ident:
	  /* Ignore the rest of the line if we don't need the input of
	     this line.  */
	  if (ignore_content)
	    {
	      lr_ignore_rest (ldfile, 0);
	      break;
	    }

	  /* This could mean one of several things.  First test whether
	     it's a character class name.  */
	  for (cnt = 0; cnt < ctype->nr_charclass; ++cnt)
	    if (strcmp (now->val.str.startmb, ctype->classnames[cnt]) == 0)
	      break;
	  if (cnt < ctype->nr_charclass)
	    {
	      class_bit = _ISwbit (cnt);
	      class256_bit = cnt <= 11 ? _ISbit (cnt) : 0;
	      free (now->val.str.startmb);
	      goto read_charclass;
	    }
	  for (cnt = 0; cnt < ctype->map_collection_nr; ++cnt)
	    if (strcmp (now->val.str.startmb, ctype->mapnames[cnt]) == 0)
	      break;
	  if (cnt < ctype->map_collection_nr)
	    {
	      mapidx = cnt;
	      free (now->val.str.startmb);
	      goto read_mapping;
            }
#ifdef PREDEFINED_CLASSES
	  if (strcmp (now->val.str.startmb, "special1") == 0)
	    {
	      class_bit = _ISwspecial1;
	      free (now->val.str.startmb);
	      goto read_charclass;
	    }
	  if (strcmp (now->val.str.startmb, "special2") == 0)
	    {
	      class_bit = _ISwspecial2;
	      free (now->val.str.startmb);
	      goto read_charclass;
	    }
	  if (strcmp (now->val.str.startmb, "special3") == 0)
	    {
	      class_bit = _ISwspecial3;
	      free (now->val.str.startmb);
	      goto read_charclass;
	    }
	  if (strcmp (now->val.str.startmb, "tosymmetric") == 0)
	    {
	      mapidx = 2;
	      goto read_mapping;
	    }
#endif
	  break;

	case tok_end:
	  /* Next we assume `LC_CTYPE'.  */
	  now = lr_token (ldfile, charmap, NULL);
	  if (now->tok == tok_eof)
	    break;
	  if (now->tok == tok_eol)
	    lr_error (ldfile, _("%s: incomplete `END' line"),
		      "LC_CTYPE");
	  else if (now->tok != tok_lc_ctype)
	    lr_error (ldfile, _("\
%1$s: definition does not end with `END %1$s'"), "LC_CTYPE");
	  lr_ignore_rest (ldfile, now->tok == tok_lc_ctype);
	  return;

	default:
	err_label:
	  if (now->tok != tok_eof)
	    SYNTAX_ERROR (_("%s: syntax error"), "LC_CTYPE");
	}

      /* Prepare for the next round.  */
      now = lr_token (ldfile, charmap, NULL);
      nowtok = now->tok;
    }

  /* When we come here we reached the end of the file.  */
  lr_error (ldfile, _("%s: premature end of file"), "LC_CTYPE");
}


static void
set_class_defaults (struct locale_ctype_t *ctype, struct charmap_t *charmap,
		    struct repertoire_t *repertoire)
{
  size_t cnt;

  /* These function defines the default values for the classes and conversions
     according to POSIX.2 2.5.2.1.
     It may seem that the order of these if-blocks is arbitrary but it is NOT.
     Don't move them unless you know what you do!  */

  void set_default (int bitpos, int from, int to)
    {
      char tmp[2];
      int ch;
      int bit = _ISbit (bitpos);
      int bitw = _ISwbit (bitpos);
      /* Define string.  */
      strcpy (tmp, "?");

      for (ch = from; ch <= to; ++ch)
	{
	  struct charseq *seq;
	  tmp[0] = ch;

	  seq = charmap_find_value (charmap, tmp, 1);
	  if (seq == NULL)
	    {
	      if (!be_quiet)
		error (0, 0, _("\
%s: character `%s' not defined in charmap while needed as default value"),
		       "LC_CTYPE", tmp);
	    }
	  else if (seq->nbytes != 1)
	    error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
		   "LC_CTYPE", tmp);
	  else
	    ctype->class256_collection[seq->bytes[0]] |= bit;

	  /* No need to search here, the ASCII value is also the Unicode
	     value.  */
	  ELEM (ctype, class_collection, , ch) |= bitw;
	}
    }

  /* Set default values if keyword was not present.  */
  if ((ctype->class_done & BITw (tok_upper)) == 0)
    /* "If this keyword [lower] is not specified, the lowercase letters
        `A' through `Z', ..., shall automatically belong to this class,
	with implementation defined character values."  [P1003.2, 2.5.2.1]  */
    set_default (BITPOS (tok_upper), 'A', 'Z');

  if ((ctype->class_done & BITw (tok_lower)) == 0)
    /* "If this keyword [lower] is not specified, the lowercase letters
        `a' through `z', ..., shall automatically belong to this class,
	with implementation defined character values."  [P1003.2, 2.5.2.1]  */
    set_default (BITPOS (tok_lower), 'a', 'z');

  if ((ctype->class_done & BITw (tok_alpha)) == 0)
    {
      /* Table 2-6 in P1003.2 says that characters in class `upper' or
	 class `lower' *must* be in class `alpha'.  */
      unsigned long int mask = BIT (tok_upper) | BIT (tok_lower);
      unsigned long int maskw = BITw (tok_upper) | BITw (tok_lower);

      for (cnt = 0; cnt < 256; ++cnt)
	if ((ctype->class256_collection[cnt] & mask) != 0)
	  ctype->class256_collection[cnt] |= BIT (tok_alpha);

      for (cnt = 0; cnt < ctype->class_collection_act; ++cnt)
	if ((ctype->class_collection[cnt] & maskw) != 0)
	  ctype->class_collection[cnt] |= BITw (tok_alpha);
    }

  if ((ctype->class_done & BITw (tok_digit)) == 0)
    /* "If this keyword [digit] is not specified, the digits `0' through
        `9', ..., shall automatically belong to this class, with
	implementation-defined character values."  [P1003.2, 2.5.2.1]  */
    set_default (BITPOS (tok_digit), '0', '9');

  /* "Only characters specified for the `alpha' and `digit' keyword
     shall be specified.  Characters specified for the keyword `alpha'
     and `digit' are automatically included in this class.  */
  {
    unsigned long int mask = BIT (tok_alpha) | BIT (tok_digit);
    unsigned long int maskw = BITw (tok_alpha) | BITw (tok_digit);

    for (cnt = 0; cnt < 256; ++cnt)
      if ((ctype->class256_collection[cnt] & mask) != 0)
	ctype->class256_collection[cnt] |= BIT (tok_alnum);

    for (cnt = 0; cnt < ctype->class_collection_act; ++cnt)
      if ((ctype->class_collection[cnt] & maskw) != 0)
	ctype->class_collection[cnt] |= BITw (tok_alnum);
  }

  if ((ctype->class_done & BITw (tok_space)) == 0)
    /* "If this keyword [space] is not specified, the characters <space>,
        <form-feed>, <newline>, <carriage-return>, <tab>, and
	<vertical-tab>, ..., shall automatically belong to this class,
	with implementation-defined character values."  [P1003.2, 2.5.2.1]  */
    {
      struct charseq *seq;

      seq = charmap_find_value (charmap, "space", 5);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "SP", 2);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "U00000020", 9);
      if (seq == NULL)
	{
	  if (!be_quiet)
	    error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
		   "LC_CTYPE", "<space>");
	}
      else if (seq->nbytes != 1)
	error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
	       "LC_CTYPE", "<space>");
      else
	ctype->class256_collection[seq->bytes[0]] |= BIT (tok_space);

      /* No need to search.  */
      ELEM (ctype, class_collection, , L' ') |= BITw (tok_space);

      seq = charmap_find_value (charmap, "form-feed", 9);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "U0000000C", 9);
      if (seq == NULL)
	{
	  if (!be_quiet)
	    error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
		   "LC_CTYPE", "<form-feed>");
	}
      else if (seq->nbytes != 1)
	error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
	       "LC_CTYPE", "<form-feed>");
      else
	ctype->class256_collection[seq->bytes[0]] |= BIT (tok_space);

      /* No need to search.  */
      ELEM (ctype, class_collection, , L'\f') |= BITw (tok_space);


      seq = charmap_find_value (charmap, "newline", 7);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "U0000000A", 9);
      if (seq == NULL)
	{
	  if (!be_quiet)
	    error (0, 0, _("\
character `%s' not defined while needed as default value"),
		   "<newline>");
	}
      else if (seq->nbytes != 1)
	error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
	       "LC_CTYPE", "<newline>");
      else
	ctype->class256_collection[seq->bytes[0]] |= BIT (tok_space);

      /* No need to search.  */
      ELEM (ctype, class_collection, , L'\n') |= BITw (tok_space);


      seq = charmap_find_value (charmap, "carriage-return", 15);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "U0000000D", 9);
      if (seq == NULL)
	{
	  if (!be_quiet)
	    error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
		   "LC_CTYPE", "<carriage-return>");
	}
      else if (seq->nbytes != 1)
	error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
	       "LC_CTYPE", "<carriage-return>");
      else
	ctype->class256_collection[seq->bytes[0]] |= BIT (tok_space);

      /* No need to search.  */
      ELEM (ctype, class_collection, , L'\r') |= BITw (tok_space);


      seq = charmap_find_value (charmap, "tab", 3);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "U00000009", 9);
      if (seq == NULL)
	{
	  if (!be_quiet)
	    error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
		   "LC_CTYPE", "<tab>");
	}
      else if (seq->nbytes != 1)
	error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
	       "LC_CTYPE", "<tab>");
      else
	ctype->class256_collection[seq->bytes[0]] |= BIT (tok_space);

      /* No need to search.  */
      ELEM (ctype, class_collection, , L'\t') |= BITw (tok_space);


      seq = charmap_find_value (charmap, "vertical-tab", 12);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "U0000000B", 9);
      if (seq == NULL)
	{
	  if (!be_quiet)
	    error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
		   "LC_CTYPE", "<vertical-tab>");
	}
      else if (seq->nbytes != 1)
	error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
	       "LC_CTYPE", "<vertical-tab>");
      else
	ctype->class256_collection[seq->bytes[0]] |= BIT (tok_space);

      /* No need to search.  */
      ELEM (ctype, class_collection, , L'\v') |= BITw (tok_space);
    }

  if ((ctype->class_done & BITw (tok_xdigit)) == 0)
    /* "If this keyword is not specified, the digits `0' to `9', the
        uppercase letters `A' through `F', and the lowercase letters `a'
	through `f', ..., shell automatically belong to this class, with
	implementation defined character values."  [P1003.2, 2.5.2.1]  */
    {
      set_default (BITPOS (tok_xdigit), '0', '9');
      set_default (BITPOS (tok_xdigit), 'A', 'F');
      set_default (BITPOS (tok_xdigit), 'a', 'f');
    }

  if ((ctype->class_done & BITw (tok_blank)) == 0)
    /* "If this keyword [blank] is unspecified, the characters <space> and
       <tab> shall belong to this character class."  [P1003.2, 2.5.2.1]  */
   {
      struct charseq *seq;

      seq = charmap_find_value (charmap, "space", 5);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "SP", 2);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "U00000020", 9);
      if (seq == NULL)
	{
	  if (!be_quiet)
	    error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
		   "LC_CTYPE", "<space>");
	}
      else if (seq->nbytes != 1)
	error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
	       "LC_CTYPE", "<space>");
      else
	ctype->class256_collection[seq->bytes[0]] |= BIT (tok_blank);

      /* No need to search.  */
      ELEM (ctype, class_collection, , L' ') |= BITw (tok_blank);


      seq = charmap_find_value (charmap, "tab", 3);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "U00000009", 9);
      if (seq == NULL)
	{
	  if (!be_quiet)
	    error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
		   "LC_CTYPE", "<tab>");
	}
      else if (seq->nbytes != 1)
	error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
	       "LC_CTYPE", "<tab>");
      else
	ctype->class256_collection[seq->bytes[0]] |= BIT (tok_blank);

      /* No need to search.  */
      ELEM (ctype, class_collection, , L'\t') |= BITw (tok_blank);
    }

  if ((ctype->class_done & BITw (tok_graph)) == 0)
    /* "If this keyword [graph] is not specified, characters specified for
        the keywords `upper', `lower', `alpha', `digit', `xdigit' and `punct',
	shall belong to this character class."  [P1003.2, 2.5.2.1]  */
    {
      unsigned long int mask = BIT (tok_upper) | BIT (tok_lower) |
	BIT (tok_alpha) | BIT (tok_digit) | BIT (tok_xdigit) | BIT (tok_punct);
      unsigned long int maskw = BITw (tok_upper) | BITw (tok_lower) |
	BITw (tok_alpha) | BITw (tok_digit) | BITw (tok_xdigit) |
	BITw (tok_punct);
      size_t cnt;

      for (cnt = 0; cnt < ctype->class_collection_act; ++cnt)
	if ((ctype->class_collection[cnt] & maskw) != 0)
	  ctype->class_collection[cnt] |= BITw (tok_graph);

      for (cnt = 0; cnt < 256; ++cnt)
	if ((ctype->class256_collection[cnt] & mask) != 0)
	  ctype->class256_collection[cnt] |= BIT (tok_graph);
    }

  if ((ctype->class_done & BITw (tok_print)) == 0)
    /* "If this keyword [print] is not provided, characters specified for
        the keywords `upper', `lower', `alpha', `digit', `xdigit', `punct',
	and the <space> character shall belong to this character class."
	[P1003.2, 2.5.2.1]  */
    {
      unsigned long int mask = BIT (tok_upper) | BIT (tok_lower) |
	BIT (tok_alpha) | BIT (tok_digit) | BIT (tok_xdigit) | BIT (tok_punct);
      unsigned long int maskw = BITw (tok_upper) | BITw (tok_lower) |
	BITw (tok_alpha) | BITw (tok_digit) | BITw (tok_xdigit) |
	BITw (tok_punct);
      size_t cnt;
      struct charseq *seq;

      for (cnt = 0; cnt < ctype->class_collection_act; ++cnt)
	if ((ctype->class_collection[cnt] & maskw) != 0)
	  ctype->class_collection[cnt] |= BITw (tok_print);

      for (cnt = 0; cnt < 256; ++cnt)
	if ((ctype->class256_collection[cnt] & mask) != 0)
	  ctype->class256_collection[cnt] |= BIT (tok_print);


      seq = charmap_find_value (charmap, "space", 5);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "SP", 2);
      if (seq == NULL)
	seq = charmap_find_value (charmap, "U00000020", 9);
      if (seq == NULL)
	{
	  if (!be_quiet)
	    error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
		   "LC_CTYPE", "<space>");
	}
      else if (seq->nbytes != 1)
	error (0, 0, _("\
%s: character `%s' in charmap not representable with one byte"),
	       "LC_CTYPE", "<space>");
      else
	ctype->class256_collection[seq->bytes[0]] |= BIT (tok_print);

      /* No need to search.  */
      ELEM (ctype, class_collection, , L' ') |= BITw (tok_print);
    }

  if (ctype->tomap_done[0] == 0)
    /* "If this keyword [toupper] is not specified, the lowercase letters
        `a' through `z', and their corresponding uppercase letters `A' to
	`Z', ..., shall automatically be included, with implementation-
	defined character values."  [P1003.2, 2.5.2.1]  */
    {
      char tmp[4];
      int ch;

      strcpy (tmp, "<?>");

      for (ch = 'a'; ch <= 'z'; ++ch)
	{
	  struct charseq *seq_from, *seq_to;

	  tmp[1] = (char) ch;

	  seq_from = charmap_find_value (charmap, &tmp[1], 1);
	  if (seq_from == NULL)
	    {
	      if (!be_quiet)
		error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
		       "LC_CTYPE", tmp);
	    }
	  else if (seq_from->nbytes != 1)
	    {
	      if (!be_quiet)
		error (0, 0, _("\
%s: character `%s' needed as default value not representable with one byte"),
		       "LC_CTYPE", tmp);
	    }
	  else
	    {
	      /* This conversion is implementation defined.  */
	      tmp[1] = (char) (ch + ('A' - 'a'));
	      seq_to = charmap_find_value (charmap, &tmp[1], 1);
	      if (seq_to == NULL)
		{
		  if (!be_quiet)
		    error (0, 0, _("\
%s: character `%s' not defined while needed as default value"),
			   "LC_CTYPE", tmp);
		}
	      else if (seq_to->nbytes != 1)
		{
		  if (!be_quiet)
		    error (0, 0, _("\
%s: character `%s' needed as default value not representable with one byte"),
			   "LC_CTYPE", tmp);
		}
	      else
		/* The index [0] is determined by the order of the
		   `ctype_map_newP' calls in `ctype_startup'.  */
		ctype->map256_collection[0][seq_from->bytes[0]]
		  = seq_to->bytes[0];
	    }

	  /* No need to search.  */
	  ELEM (ctype, map_collection, [0], ch) = ch + ('A' - 'a');
	}
    }

  if (ctype->tomap_done[1] == 0)
    /* "If this keyword [tolower] is not specified, the mapping shall be
       the reverse mapping of the one specified to `toupper'."  [P1003.2]  */
    {
      for (cnt = 0; cnt < ctype->map_collection_act[0]; ++cnt)
	if (ctype->map_collection[0][cnt] != 0)
	  ELEM (ctype, map_collection, [1],
		ctype->map_collection[0][cnt])
	    = ctype->charnames[cnt];

      for (cnt = 0; cnt < 256; ++cnt)
	if (ctype->map256_collection[0][cnt] != 0)
	  ctype->map256_collection[1][ctype->map256_collection[0][cnt]] = cnt;
    }

  if (ctype->outdigits_act == 0)
    {
      for (cnt = 0; cnt < 10; ++cnt)
	{
	  ctype->mboutdigits[cnt] = charmap_find_symbol (charmap,
							 digits + cnt, 1);

	  if (ctype->mboutdigits[cnt] == NULL)
	    ctype->mboutdigits[cnt] = charmap_find_symbol (charmap,
							   longnames[cnt],
							   strlen (longnames[cnt]));

	  if (ctype->mboutdigits[cnt] == NULL)
	    ctype->mboutdigits[cnt] = charmap_find_symbol (charmap,
							   uninames[cnt], 9);

	  if (ctype->mboutdigits[cnt] == NULL)
	    {
	      /* Provide a replacement.  */
	      error (0, 0, _("\
no output digits defined and none of the standard names in the charmap"));

	      ctype->mboutdigits[cnt] = obstack_alloc (&charmap->mem_pool,
						       sizeof (struct charseq)
						       + 1);

	      /* This is better than nothing.  */
	      ctype->mboutdigits[cnt]->bytes[0] = digits[cnt];
	      ctype->mboutdigits[cnt]->nbytes = 1;
	    }

	  ctype->wcoutdigits[cnt] = L'0' + cnt;
	}

      ctype->outdigits_act = 10;
    }
}


/* Construction of sparse 3-level tables.
   See wchar-lookup.h for their structure and the meaning of p and q.  */

struct wctype_table
{
  /* Parameters.  */
  unsigned int p;
  unsigned int q;
  /* Working representation.  */
  size_t level1_alloc;
  size_t level1_size;
  uint32_t *level1;
  size_t level2_alloc;
  size_t level2_size;
  uint32_t *level2;
  size_t level3_alloc;
  size_t level3_size;
  uint32_t *level3;
  /* Compressed representation.  */
  size_t result_size;
  char *result;
};

/* Initialize.  Assumes t->p and t->q have already been set.  */
static inline void
wctype_table_init (struct wctype_table *t)
{
  t->level1_alloc = t->level1_size = 0;
  t->level2_alloc = t->level2_size = 0;
  t->level3_alloc = t->level3_size = 0;
}

/* Retrieve an entry.  */
static inline int
wctype_table_get (struct wctype_table *t, uint32_t wc)
{
  uint32_t index1 = wc >> (t->q + t->p + 5);
  if (index1 < t->level1_size)
    {
      uint32_t lookup1 = t->level1[index1];
      if (lookup1 != ~((uint32_t) 0))
	{
	  uint32_t index2 = ((wc >> (t->p + 5)) & ((1 << t->q) - 1))
			    + (lookup1 << t->q);
	  uint32_t lookup2 = t->level2[index2];
	  if (lookup2 != ~((uint32_t) 0))
	    {
	      uint32_t index3 = ((wc >> 5) & ((1 << t->p) - 1))
				+ (lookup2 << t->p);
	      uint32_t lookup3 = t->level3[index3];
	      uint32_t index4 = wc & 0x1f;

	      return (lookup3 >> index4) & 1;
	    }
	}
    }
  return 0;
}

/* Add one entry.  */
static void
wctype_table_add (struct wctype_table *t, uint32_t wc)
{
  uint32_t index1 = wc >> (t->q + t->p + 5);
  uint32_t index2 = (wc >> (t->p + 5)) & ((1 << t->q) - 1);
  uint32_t index3 = (wc >> 5) & ((1 << t->p) - 1);
  uint32_t index4 = wc & 0x1f;
  size_t i, i1, i2;

  if (index1 >= t->level1_size)
    {
      if (index1 >= t->level1_alloc)
	{
	  size_t alloc = 2 * t->level1_alloc;
	  if (alloc <= index1)
	    alloc = index1 + 1;
	  t->level1 = (t->level1_alloc > 0
		       ? (uint32_t *) xrealloc ((char *) t->level1,
						alloc * sizeof (uint32_t))
		       : (uint32_t *) xmalloc (alloc * sizeof (uint32_t)));
	  t->level1_alloc = alloc;
	}
      while (index1 >= t->level1_size)
	t->level1[t->level1_size++] = ~((uint32_t) 0);
    }

  if (t->level1[index1] == ~((uint32_t) 0))
    {
      if (t->level2_size == t->level2_alloc)
	{
	  size_t alloc = 2 * t->level2_alloc + 1;
	  t->level2 = (t->level2_alloc > 0
		       ? (uint32_t *) xrealloc ((char *) t->level2,
						(alloc << t->q) * sizeof (uint32_t))
		       : (uint32_t *) xmalloc ((alloc << t->q) * sizeof (uint32_t)));
	  t->level2_alloc = alloc;
	}
      i1 = t->level2_size << t->q;
      i2 = (t->level2_size + 1) << t->q;
      for (i = i1; i < i2; i++)
	t->level2[i] = ~((uint32_t) 0);
      t->level1[index1] = t->level2_size++;
    }

  index2 += t->level1[index1] << t->q;

  if (t->level2[index2] == ~((uint32_t) 0))
    {
      if (t->level3_size == t->level3_alloc)
	{
	  size_t alloc = 2 * t->level3_alloc + 1;
	  t->level3 = (t->level3_alloc > 0
		       ? (uint32_t *) xrealloc ((char *) t->level3,
						(alloc << t->p) * sizeof (uint32_t))
		       : (uint32_t *) xmalloc ((alloc << t->p) * sizeof (uint32_t)));
	  t->level3_alloc = alloc;
	}
      i1 = t->level3_size << t->p;
      i2 = (t->level3_size + 1) << t->p;
      for (i = i1; i < i2; i++)
	t->level3[i] = 0;
      t->level2[index2] = t->level3_size++;
    }

  index3 += t->level2[index2] << t->p;

  t->level3[index3] |= (uint32_t)1 << index4;
}

/* Finalize and shrink.  */
static void
wctype_table_finalize (struct wctype_table *t)
{
  size_t i, j, k;
  uint32_t reorder3[t->level3_size];
  uint32_t reorder2[t->level2_size];
  uint32_t level1_offset, level2_offset, level3_offset;

  /* Uniquify level3 blocks.  */
  k = 0;
  for (j = 0; j < t->level3_size; j++)
    {
      for (i = 0; i < k; i++)
	if (memcmp (&t->level3[i << t->p], &t->level3[j << t->p],
		    (1 << t->p) * sizeof (uint32_t)) == 0)
	  break;
      /* Relocate block j to block i.  */
      reorder3[j] = i;
      if (i == k)
	{
	  if (i != j)
	    memcpy (&t->level3[i << t->p], &t->level3[j << t->p],
		    (1 << t->p) * sizeof (uint32_t));
	  k++;
	}
    }
  t->level3_size = k;

  for (i = 0; i < (t->level2_size << t->q); i++)
    if (t->level2[i] != ~((uint32_t) 0))
      t->level2[i] = reorder3[t->level2[i]];

  /* Uniquify level2 blocks.  */
  k = 0;
  for (j = 0; j < t->level2_size; j++)
    {
      for (i = 0; i < k; i++)
	if (memcmp (&t->level2[i << t->q], &t->level2[j << t->q],
		    (1 << t->q) * sizeof (uint32_t)) == 0)
	  break;
      /* Relocate block j to block i.  */
      reorder2[j] = i;
      if (i == k)
	{
	  if (i != j)
	    memcpy (&t->level2[i << t->q], &t->level2[j << t->q],
		    (1 << t->q) * sizeof (uint32_t));
	  k++;
	}
    }
  t->level2_size = k;

  for (i = 0; i < t->level1_size; i++)
    if (t->level1[i] != ~((uint32_t) 0))
      t->level1[i] = reorder2[t->level1[i]];

  /* Create and fill the resulting compressed representation.  */
  t->result_size =
    5 * sizeof (uint32_t)
    + t->level1_size * sizeof (uint32_t)
    + (t->level2_size << t->q) * sizeof (uint32_t)
    + (t->level3_size << t->p) * sizeof (uint32_t);
  t->result = (char *) xmalloc (t->result_size);

  level1_offset =
    5 * sizeof (uint32_t);
  level2_offset =
    5 * sizeof (uint32_t)
    + t->level1_size * sizeof (uint32_t);
  level3_offset =
    5 * sizeof (uint32_t)
    + t->level1_size * sizeof (uint32_t)
    + (t->level2_size << t->q) * sizeof (uint32_t);

  ((uint32_t *) t->result)[0] = t->q + t->p + 5;
  ((uint32_t *) t->result)[1] = t->level1_size;
  ((uint32_t *) t->result)[2] = t->p + 5;
  ((uint32_t *) t->result)[3] = (1 << t->q) - 1;
  ((uint32_t *) t->result)[4] = (1 << t->p) - 1;

  for (i = 0; i < t->level1_size; i++)
    ((uint32_t *) (t->result + level1_offset))[i] =
      (t->level1[i] == ~((uint32_t) 0)
       ? 0
       : (t->level1[i] << t->q) * sizeof (uint32_t) + level2_offset);

  for (i = 0; i < (t->level2_size << t->q); i++)
    ((uint32_t *) (t->result + level2_offset))[i] =
      (t->level2[i] == ~((uint32_t) 0)
       ? 0
       : (t->level2[i] << t->p) * sizeof (uint32_t) + level3_offset);

  for (i = 0; i < (t->level3_size << t->p); i++)
    ((uint32_t *) (t->result + level3_offset))[i] = t->level3[i];

  if (t->level1_alloc > 0)
    free (t->level1);
  if (t->level2_alloc > 0)
    free (t->level2);
  if (t->level3_alloc > 0)
    free (t->level3);
}

struct wcwidth_table
{
  /* Parameters.  */
  unsigned int p;
  unsigned int q;
  /* Working representation.  */
  size_t level1_alloc;
  size_t level1_size;
  uint32_t *level1;
  size_t level2_alloc;
  size_t level2_size;
  uint32_t *level2;
  size_t level3_alloc;
  size_t level3_size;
  uint8_t *level3;
  /* Compressed representation.  */
  size_t result_size;
  char *result;
};

/* Initialize.  Assumes t->p and t->q have already been set.  */
static inline void
wcwidth_table_init (struct wcwidth_table *t)
{
  t->level1_alloc = t->level1_size = 0;
  t->level2_alloc = t->level2_size = 0;
  t->level3_alloc = t->level3_size = 0;
}

/* Retrieve an entry.  */
static inline uint8_t
wcwidth_table_get (struct wcwidth_table *t, uint32_t wc)
{
  uint32_t index1 = wc >> (t->q + t->p);
  if (index1 < t->level1_size)
    {
      uint32_t lookup1 = t->level1[index1];
      if (lookup1 != ~((uint32_t) 0))
	{
	  uint32_t index2 = ((wc >> t->p) & ((1 << t->q) - 1))
			    + (lookup1 << t->q);
	  uint32_t lookup2 = t->level2[index2];
	  if (lookup2 != ~((uint32_t) 0))
	    {
	      uint32_t index3 = (wc & ((1 << t->p) - 1))
				+ (lookup2 << t->p);
	      uint8_t lookup3 = t->level3[index3];

	      return lookup3;
	    }
	}
    }
  return 0xff;
}

/* Add one entry.  */
static void
wcwidth_table_add (struct wcwidth_table *t, uint32_t wc, uint8_t width)
{
  uint32_t index1 = wc >> (t->q + t->p);
  uint32_t index2 = (wc >> t->p) & ((1 << t->q) - 1);
  uint32_t index3 = wc & ((1 << t->p) - 1);
  size_t i, i1, i2;

  if (width == wcwidth_table_get (t, wc))
    return;

  if (index1 >= t->level1_size)
    {
      if (index1 >= t->level1_alloc)
	{
	  size_t alloc = 2 * t->level1_alloc;
	  if (alloc <= index1)
	    alloc = index1 + 1;
	  t->level1 = (t->level1_alloc > 0
		       ? (uint32_t *) xrealloc ((char *) t->level1,
						alloc * sizeof (uint32_t))
		       : (uint32_t *) xmalloc (alloc * sizeof (uint32_t)));
	  t->level1_alloc = alloc;
	}
      while (index1 >= t->level1_size)
	t->level1[t->level1_size++] = ~((uint32_t) 0);
    }

  if (t->level1[index1] == ~((uint32_t) 0))
    {
      if (t->level2_size == t->level2_alloc)
	{
	  size_t alloc = 2 * t->level2_alloc + 1;
	  t->level2 = (t->level2_alloc > 0
		       ? (uint32_t *) xrealloc ((char *) t->level2,
						(alloc << t->q) * sizeof (uint32_t))
		       : (uint32_t *) xmalloc ((alloc << t->q) * sizeof (uint32_t)));
	  t->level2_alloc = alloc;
	}
      i1 = t->level2_size << t->q;
      i2 = (t->level2_size + 1) << t->q;
      for (i = i1; i < i2; i++)
	t->level2[i] = ~((uint32_t) 0);
      t->level1[index1] = t->level2_size++;
    }

  index2 += t->level1[index1] << t->q;

  if (t->level2[index2] == ~((uint32_t) 0))
    {
      if (t->level3_size == t->level3_alloc)
	{
	  size_t alloc = 2 * t->level3_alloc + 1;
	  t->level3 = (t->level3_alloc > 0
		       ? (uint8_t *) xrealloc ((char *) t->level3,
					       (alloc << t->p) * sizeof (uint8_t))
		       : (uint8_t *) xmalloc ((alloc << t->p) * sizeof (uint8_t)));
	  t->level3_alloc = alloc;
	}
      i1 = t->level3_size << t->p;
      i2 = (t->level3_size + 1) << t->p;
      for (i = i1; i < i2; i++)
	t->level3[i] = 0xff;
      t->level2[index2] = t->level3_size++;
    }

  index3 += t->level2[index2] << t->p;

  t->level3[index3] = width;
}

/* Finalize and shrink.  */
static void
wcwidth_table_finalize (struct wcwidth_table *t)
{
  size_t i, j, k;
  uint32_t reorder3[t->level3_size];
  uint32_t reorder2[t->level2_size];
  uint32_t level1_offset, level2_offset, level3_offset, last_offset;

  /* Uniquify level3 blocks.  */
  k = 0;
  for (j = 0; j < t->level3_size; j++)
    {
      for (i = 0; i < k; i++)
	if (memcmp (&t->level3[i << t->p], &t->level3[j << t->p],
		    (1 << t->p) * sizeof (uint8_t)) == 0)
	  break;
      /* Relocate block j to block i.  */
      reorder3[j] = i;
      if (i == k)
	{
	  if (i != j)
	    memcpy (&t->level3[i << t->p], &t->level3[j << t->p],
		    (1 << t->p) * sizeof (uint8_t));
	  k++;
	}
    }
  t->level3_size = k;

  for (i = 0; i < (t->level2_size << t->q); i++)
    if (t->level2[i] != ~((uint32_t) 0))
      t->level2[i] = reorder3[t->level2[i]];

  /* Uniquify level2 blocks.  */
  k = 0;
  for (j = 0; j < t->level2_size; j++)
    {
      for (i = 0; i < k; i++)
	if (memcmp (&t->level2[i << t->q], &t->level2[j << t->q],
		    (1 << t->q) * sizeof (uint32_t)) == 0)
	  break;
      /* Relocate block j to block i.  */
      reorder2[j] = i;
      if (i == k)
	{
	  if (i != j)
	    memcpy (&t->level2[i << t->q], &t->level2[j << t->q],
		    (1 << t->q) * sizeof (uint32_t));
	  k++;
	}
    }
  t->level2_size = k;

  for (i = 0; i < t->level1_size; i++)
    if (t->level1[i] != ~((uint32_t) 0))
      t->level1[i] = reorder2[t->level1[i]];

  /* Create and fill the resulting compressed representation.  */
  last_offset =
    5 * sizeof (uint32_t)
    + t->level1_size * sizeof (uint32_t)
    + (t->level2_size << t->q) * sizeof (uint32_t)
    + (t->level3_size << t->p) * sizeof (uint8_t);
  t->result_size = (last_offset + 3) & ~3ul;
  t->result = (char *) xmalloc (t->result_size);

  level1_offset =
    5 * sizeof (uint32_t);
  level2_offset =
    5 * sizeof (uint32_t)
    + t->level1_size * sizeof (uint32_t);
  level3_offset =
    5 * sizeof (uint32_t)
    + t->level1_size * sizeof (uint32_t)
    + (t->level2_size << t->q) * sizeof (uint32_t);

  ((uint32_t *) t->result)[0] = t->q + t->p;
  ((uint32_t *) t->result)[1] = t->level1_size;
  ((uint32_t *) t->result)[2] = t->p;
  ((uint32_t *) t->result)[3] = (1 << t->q) - 1;
  ((uint32_t *) t->result)[4] = (1 << t->p) - 1;

  for (i = 0; i < t->level1_size; i++)
    ((uint32_t *) (t->result + level1_offset))[i] =
      (t->level1[i] == ~((uint32_t) 0)
       ? 0
       : (t->level1[i] << t->q) * sizeof (uint32_t) + level2_offset);

  for (i = 0; i < (t->level2_size << t->q); i++)
    ((uint32_t *) (t->result + level2_offset))[i] =
      (t->level2[i] == ~((uint32_t) 0)
       ? 0
       : (t->level2[i] << t->p) * sizeof (uint8_t) + level3_offset);

  for (i = 0; i < (t->level3_size << t->p); i++)
    ((uint8_t *) (t->result + level3_offset))[i] = t->level3[i];

  if (last_offset < t->result_size)
    memset (t->result + last_offset, 0, t->result_size - last_offset);

  if (t->level1_alloc > 0)
    free (t->level1);
  if (t->level2_alloc > 0)
    free (t->level2);
  if (t->level3_alloc > 0)
    free (t->level3);
}

struct wctrans_table
{
  /* Parameters.  */
  unsigned int p;
  unsigned int q;
  /* Working representation.  */
  size_t level1_alloc;
  size_t level1_size;
  uint32_t *level1;
  size_t level2_alloc;
  size_t level2_size;
  uint32_t *level2;
  size_t level3_alloc;
  size_t level3_size;
  int32_t *level3;
  /* Compressed representation.  */
  size_t result_size;
  char *result;
};

/* Initialize.  Assumes t->p and t->q have already been set.  */
static inline void
wctrans_table_init (struct wctrans_table *t)
{
  t->level1_alloc = t->level1_size = 0;
  t->level2_alloc = t->level2_size = 0;
  t->level3_alloc = t->level3_size = 0;
}

/* Retrieve an entry.  */
static inline uint32_t
wctrans_table_get (struct wctrans_table *t, uint32_t wc)
{
  uint32_t index1 = wc >> (t->q + t->p);
  if (index1 < t->level1_size)
    {
      uint32_t lookup1 = t->level1[index1];
      if (lookup1 != ~((uint32_t) 0))
	{
	  uint32_t index2 = ((wc >> t->p) & ((1 << t->q) - 1))
			    + (lookup1 << t->q);
	  uint32_t lookup2 = t->level2[index2];
	  if (lookup2 != ~((uint32_t) 0))
	    {
	      uint32_t index3 = (wc & ((1 << t->p) - 1))
				+ (lookup2 << t->p);
	      int32_t lookup3 = t->level3[index3];

	      return wc + lookup3;
	    }
	}
    }
  return wc;
}

/* Add one entry.  */
static void
wctrans_table_add (struct wctrans_table *t, uint32_t wc, uint32_t mapped_wc)
{
  uint32_t index1 = wc >> (t->q + t->p);
  uint32_t index2 = (wc >> t->p) & ((1 << t->q) - 1);
  uint32_t index3 = wc & ((1 << t->p) - 1);
  int32_t value;
  size_t i, i1, i2;

  if (mapped_wc == wctrans_table_get (t, wc))
    return;

  value = (int32_t) mapped_wc - (int32_t) wc;

  if (index1 >= t->level1_size)
    {
      if (index1 >= t->level1_alloc)
	{
	  size_t alloc = 2 * t->level1_alloc;
	  if (alloc <= index1)
	    alloc = index1 + 1;
	  t->level1 = (t->level1_alloc > 0
		       ? (uint32_t *) xrealloc ((char *) t->level1,
						alloc * sizeof (uint32_t))
		       : (uint32_t *) xmalloc (alloc * sizeof (uint32_t)));
	  t->level1_alloc = alloc;
	}
      while (index1 >= t->level1_size)
	t->level1[t->level1_size++] = ~((uint32_t) 0);
    }

  if (t->level1[index1] == ~((uint32_t) 0))
    {
      if (t->level2_size == t->level2_alloc)
	{
	  size_t alloc = 2 * t->level2_alloc + 1;
	  t->level2 = (t->level2_alloc > 0
		       ? (uint32_t *) xrealloc ((char *) t->level2,
						(alloc << t->q) * sizeof (uint32_t))
		       : (uint32_t *) xmalloc ((alloc << t->q) * sizeof (uint32_t)));
	  t->level2_alloc = alloc;
	}
      i1 = t->level2_size << t->q;
      i2 = (t->level2_size + 1) << t->q;
      for (i = i1; i < i2; i++)
	t->level2[i] = ~((uint32_t) 0);
      t->level1[index1] = t->level2_size++;
    }

  index2 += t->level1[index1] << t->q;

  if (t->level2[index2] == ~((uint32_t) 0))
    {
      if (t->level3_size == t->level3_alloc)
	{
	  size_t alloc = 2 * t->level3_alloc + 1;
	  t->level3 = (t->level3_alloc > 0
		       ? (int32_t *) xrealloc ((char *) t->level3,
					       (alloc << t->p) * sizeof (int32_t))
		       : (int32_t *) xmalloc ((alloc << t->p) * sizeof (int32_t)));
	  t->level3_alloc = alloc;
	}
      i1 = t->level3_size << t->p;
      i2 = (t->level3_size + 1) << t->p;
      for (i = i1; i < i2; i++)
	t->level3[i] = 0;
      t->level2[index2] = t->level3_size++;
    }

  index3 += t->level2[index2] << t->p;

  t->level3[index3] = value;
}

/* Finalize and shrink.  */
static void
wctrans_table_finalize (struct wctrans_table *t)
{
  size_t i, j, k;
  uint32_t reorder3[t->level3_size];
  uint32_t reorder2[t->level2_size];
  uint32_t level1_offset, level2_offset, level3_offset;

  /* Uniquify level3 blocks.  */
  k = 0;
  for (j = 0; j < t->level3_size; j++)
    {
      for (i = 0; i < k; i++)
	if (memcmp (&t->level3[i << t->p], &t->level3[j << t->p],
		    (1 << t->p) * sizeof (int32_t)) == 0)
	  break;
      /* Relocate block j to block i.  */
      reorder3[j] = i;
      if (i == k)
	{
	  if (i != j)
	    memcpy (&t->level3[i << t->p], &t->level3[j << t->p],
		    (1 << t->p) * sizeof (int32_t));
	  k++;
	}
    }
  t->level3_size = k;

  for (i = 0; i < (t->level2_size << t->q); i++)
    if (t->level2[i] != ~((uint32_t) 0))
      t->level2[i] = reorder3[t->level2[i]];

  /* Uniquify level2 blocks.  */
  k = 0;
  for (j = 0; j < t->level2_size; j++)
    {
      for (i = 0; i < k; i++)
	if (memcmp (&t->level2[i << t->q], &t->level2[j << t->q],
		    (1 << t->q) * sizeof (uint32_t)) == 0)
	  break;
      /* Relocate block j to block i.  */
      reorder2[j] = i;
      if (i == k)
	{
	  if (i != j)
	    memcpy (&t->level2[i << t->q], &t->level2[j << t->q],
		    (1 << t->q) * sizeof (uint32_t));
	  k++;
	}
    }
  t->level2_size = k;

  for (i = 0; i < t->level1_size; i++)
    if (t->level1[i] != ~((uint32_t) 0))
      t->level1[i] = reorder2[t->level1[i]];

  /* Create and fill the resulting compressed representation.  */
  t->result_size =
    5 * sizeof (uint32_t)
    + t->level1_size * sizeof (uint32_t)
    + (t->level2_size << t->q) * sizeof (uint32_t)
    + (t->level3_size << t->p) * sizeof (int32_t);
  t->result = (char *) xmalloc (t->result_size);

  level1_offset =
    5 * sizeof (uint32_t);
  level2_offset =
    5 * sizeof (uint32_t)
    + t->level1_size * sizeof (uint32_t);
  level3_offset =
    5 * sizeof (uint32_t)
    + t->level1_size * sizeof (uint32_t)
    + (t->level2_size << t->q) * sizeof (uint32_t);

  ((uint32_t *) t->result)[0] = t->q + t->p;
  ((uint32_t *) t->result)[1] = t->level1_size;
  ((uint32_t *) t->result)[2] = t->p;
  ((uint32_t *) t->result)[3] = (1 << t->q) - 1;
  ((uint32_t *) t->result)[4] = (1 << t->p) - 1;

  for (i = 0; i < t->level1_size; i++)
    ((uint32_t *) (t->result + level1_offset))[i] =
      (t->level1[i] == ~((uint32_t) 0)
       ? 0
       : (t->level1[i] << t->q) * sizeof (uint32_t) + level2_offset);

  for (i = 0; i < (t->level2_size << t->q); i++)
    ((uint32_t *) (t->result + level2_offset))[i] =
      (t->level2[i] == ~((uint32_t) 0)
       ? 0
       : (t->level2[i] << t->p) * sizeof (int32_t) + level3_offset);

  for (i = 0; i < (t->level3_size << t->p); i++)
    ((int32_t *) (t->result + level3_offset))[i] = t->level3[i];

  if (t->level1_alloc > 0)
    free (t->level1);
  if (t->level2_alloc > 0)
    free (t->level2);
  if (t->level3_alloc > 0)
    free (t->level3);
}


static void
allocate_arrays (struct locale_ctype_t *ctype, struct charmap_t *charmap,
		 struct repertoire_t *repertoire)
{
  size_t idx;
  size_t width_table_size;
  const void *key;
  size_t len;
  void *vdata;
  void *curs;

  /* First we have to decide how we organize the arrays.  It is easy
     for a one-byte character set.  But multi-byte character set
     cannot be stored flat because the chars might be sparsely used.
     So we determine an optimal hashing function for the used
     characters.

     We use a very trivial hashing function to store the sparse
     table.  CH % TABSIZE is used as an index.  To solve multiple hits
     we have N planes.  This guarantees a fixed search time for a
     character [N / 2].  In the following code we determine the minimum
     value for TABSIZE * N, where TABSIZE >= 256.

     Some people complained that this algorithm takes too long.  Well,
     go on, improve it.  But changing the step size is *not* an
     option.  Some people changed this to use only sizes of prime
     numbers.  Think again, do some math.  We are looking for the
     optimal solution, not something which works in general.  Unless
     somebody can provide a dynamic programming solution I think this
     implementation is as good as it can get.  */
  size_t min_total = UINT_MAX;
  size_t act_size = 256;

  if (oldstyle_tables)
    {
      if (!be_quiet && ctype->charnames_act > 512)
	fputs (_("\
Computing table size for character classes might take a while..."),
	       stderr);

      /* While we want to have a small total size we are willing to use a
	 little bit larger table if this reduces the number of layers.
	 Therefore we add a little penalty to the number of planes.
	 Maybe this constant has to be adjusted a bit.  */
#define PENALTY 128
      do
	{
	  size_t cnt[act_size];
	  size_t act_planes = 1;

	  memset (cnt, '\0', sizeof cnt);

	  for (idx = 0; idx < 256; ++idx)
	    cnt[idx] = 1;

	  for (idx = 0; idx < ctype->charnames_act; ++idx)
	    if (ctype->charnames[idx] >= 256)
	      {
		size_t nr = ctype->charnames[idx] % act_size;

		if (++cnt[nr] > act_planes)
		  {
		    act_planes = cnt[nr];
		    if ((act_size + PENALTY) * act_planes >= min_total)
		      break;
		  }
	      }

	  if ((act_size + PENALTY) * act_planes < min_total)
	    {
	      min_total = (act_size + PENALTY) * act_planes;
	      ctype->plane_size = act_size;
	      ctype->plane_cnt = act_planes;
	    }

	  ++act_size;
	}
      while (act_size < min_total);

      if (!be_quiet && ctype->charnames_act > 512)
	fputs (_(" done\n"), stderr);


      ctype->names = (uint32_t *) xcalloc (ctype->plane_size
					   * ctype->plane_cnt,
					   sizeof (uint32_t));

      for (idx = 1; idx < 256; ++idx)
	ctype->names[idx] = idx;

      /* Trick: change the 0th entry's name to 1 to mark the cell occupied.  */
      ctype->names[0] = 1;

      for (idx = 256; idx < ctype->charnames_act; ++idx)
	{
	  size_t nr = (ctype->charnames[idx] % ctype->plane_size);
	  size_t depth = 0;

	  while (ctype->names[nr + depth * ctype->plane_size])
	    ++depth;
	  assert (depth < ctype->plane_cnt);

	  ctype->names[nr + depth * ctype->plane_size] = ctype->charnames[idx];

	  /* Now for faster access remember the index in the NAMES_B array.  */
	  ctype->charnames[idx] = nr + depth * ctype->plane_size;
	}
      ctype->names[0] = 0;
    }
  else
    {
      ctype->plane_size = 0;
      ctype->plane_cnt = 0;
      ctype->names = NULL;
    }

  /* You wonder about this amount of memory?  This is only because some
     users do not manage to address the array with unsigned values or
     data types with range >= 256.  '\200' would result in the array
     index -128.  To help these poor people we duplicate the entries for
     128 up to 255 below the entry for \0.  */
  ctype->ctype_b = (char_class_t *) xcalloc (256 + 128,
					     sizeof (char_class_t));
  ctype->ctype32_b = (char_class32_t *)
    xcalloc ((oldstyle_tables ? ctype->plane_size * ctype->plane_cnt : 256),
	     sizeof (char_class32_t));
  if (!oldstyle_tables)
    ctype->class_3level = (struct iovec *)
      xmalloc (ctype->nr_charclass * sizeof (struct iovec));

  /* This is the array accessed using the multibyte string elements.  */
  for (idx = 0; idx < 256; ++idx)
    ctype->ctype_b[128 + idx] = ctype->class256_collection[idx];

  /* Mirror first 127 entries.  We must take care that entry -1 is not
     mirrored because EOF == -1.  */
  for (idx = 0; idx < 127; ++idx)
    ctype->ctype_b[idx] = ctype->ctype_b[256 + idx];

  if (oldstyle_tables)
    {
      /* The 32 bit array contains all characters.  */
      for (idx = 0; idx < ctype->class_collection_act; ++idx)
	ctype->ctype32_b[ctype->charnames[idx]] = ctype->class_collection[idx];
    }
  else
    {
      /* The 32 bit array contains all characters < 0x100.  */
      for (idx = 0; idx < ctype->class_collection_act; ++idx)
	if (ctype->charnames[idx] < 0x100)
	  ctype->ctype32_b[ctype->charnames[idx]] = ctype->class_collection[idx];
    }

  if (!oldstyle_tables)
    {
      size_t nr;

      for (nr = 0; nr < ctype->nr_charclass; nr++)
	{
	  struct wctype_table t;

	  t.p = 4; /* or: 5 */
	  t.q = 7; /* or: 6 */
	  wctype_table_init (&t);

	  for (idx = 0; idx < ctype->class_collection_act; ++idx)
	    if (ctype->class_collection[idx] & _ISwbit (nr))
	      wctype_table_add (&t, ctype->charnames[idx]);

	  wctype_table_finalize (&t);

	  if (verbose)
	    fprintf (stderr, _("%s: table for class \"%s\": %lu bytes\n"),
		     "LC_CTYPE", ctype->classnames[nr],
		     (unsigned long int) t.result_size);

	  ctype->class_3level[nr].iov_base = t.result;
	  ctype->class_3level[nr].iov_len = t.result_size;
	}
    }

  /* Room for table of mappings.  */
  ctype->map = (uint32_t **) xmalloc (2 * sizeof (uint32_t *));
  ctype->map32 = (uint32_t **) xmalloc (ctype->map_collection_nr
					* sizeof (uint32_t *));
  if (!oldstyle_tables)
    ctype->map_3level = (struct iovec *)
      xmalloc (ctype->map_collection_nr * sizeof (struct iovec));

  /* Fill in all mappings.  */
  for (idx = 0; idx < 2; ++idx)
    {
      unsigned int idx2;

      /* Allocate table.  */
      ctype->map[idx] = (uint32_t *) xmalloc ((256 + 128) * sizeof (uint32_t));

      /* Copy values from collection.  */
      for (idx2 = 0; idx2 < 256; ++idx2)
	ctype->map[idx][128 + idx2] = ctype->map256_collection[idx][idx2];

      /* Mirror first 127 entries.  We must take care not to map entry
	 -1 because EOF == -1.  */
      for (idx2 = 0; idx2 < 127; ++idx2)
	ctype->map[idx][idx2] = ctype->map[idx][256 + idx2];

      /* EOF must map to EOF.  */
      ctype->map[idx][127] = EOF;
    }

  for (idx = 0; idx < ctype->map_collection_nr; ++idx)
    {
      unsigned int idx2;

      /* Allocate table.  */
      ctype->map32[idx] = (uint32_t *)
	xmalloc ((oldstyle_tables ? ctype->plane_size * ctype->plane_cnt : 256)
		 * sizeof (uint32_t));

      /* Copy default value (identity mapping).  */
      if (oldstyle_tables)
	memcpy (ctype->map32[idx], ctype->names,
		ctype->plane_size * ctype->plane_cnt * sizeof (uint32_t));
      else
	for (idx2 = 0; idx2 < 256; ++idx2)
	  ctype->map32[idx][idx2] = idx2;

      /* Copy values from collection.  */
      for (idx2 = 0; idx2 < 256; ++idx2)
	if (ctype->map_collection[idx][idx2] != 0)
	  ctype->map32[idx][idx2] = ctype->map_collection[idx][idx2];

      if (oldstyle_tables)
	while (idx2 < ctype->map_collection_act[idx])
	  {
	    if (ctype->map_collection[idx][idx2] != 0)
	      ctype->map32[idx][ctype->charnames[idx2]] =
		ctype->map_collection[idx][idx2];
	    ++idx2;
	  }
    }

  if (!oldstyle_tables)
    {
      size_t nr;

      for (nr = 0; nr < ctype->map_collection_nr; nr++)
	{
	  struct wctrans_table t;

	  t.p = 7;
	  t.q = 9;
	  wctrans_table_init (&t);

	  for (idx = 0; idx < ctype->map_collection_act[nr]; ++idx)
	    if (ctype->map_collection[nr][idx] != 0)
	      wctrans_table_add (&t, ctype->charnames[idx],
				 ctype->map_collection[nr][idx]);

	  wctrans_table_finalize (&t);

	  if (verbose)
	    fprintf (stderr, _("%s: table for map \"%s\": %lu bytes\n"),
		     "LC_CTYPE", ctype->mapnames[nr],
		     (unsigned long int) t.result_size);

	  ctype->map_3level[nr].iov_base = t.result;
	  ctype->map_3level[nr].iov_len = t.result_size;
	}
    }

  /* Extra array for class and map names.  */
  ctype->class_name_ptr = (uint32_t *) xmalloc (ctype->nr_charclass
						* sizeof (uint32_t));
  ctype->map_name_ptr = (uint32_t *) xmalloc (ctype->map_collection_nr
					      * sizeof (uint32_t));

  if (oldstyle_tables)
    {
      ctype->class_offset = 0; /* not really used */
      ctype->map_offset = 0; /* not really used */
    }
  else
    {
      ctype->class_offset = _NL_ITEM_INDEX (_NL_NUM_LC_CTYPE);
      ctype->map_offset = ctype->class_offset + ctype->nr_charclass;
    }

  /* Array for width information.  Because the expected width are very
     small we use only one single byte.  This saves space.  */
  if (oldstyle_tables)
    {
      width_table_size = (ctype->plane_size * ctype->plane_cnt + 3) & ~3ul;
      ctype->width = (unsigned char *) xmalloc (width_table_size);

      /* Initialize with -1.  */
      memset (ctype->width, '\xff', width_table_size);
      if (charmap->width_rules != NULL)
	{
	  size_t cnt;

	  for (cnt = 0; cnt < charmap->nwidth_rules; ++cnt)
	    {
	      unsigned char bytes[charmap->mb_cur_max];
	      int nbytes = charmap->width_rules[cnt].from->nbytes;

	      /* We have the range of character for which the width is
		 specified described using byte sequences of the multibyte
		 charset.  We have to convert this to UCS4 now.  And we
		 cannot simply convert the beginning and the end of the
		 sequence, we have to iterate over the byte sequence and
		 convert it for every single character.  */
	      memcpy (bytes, charmap->width_rules[cnt].from->bytes, nbytes);

	      while (nbytes < charmap->width_rules[cnt].to->nbytes
		     || memcmp (bytes, charmap->width_rules[cnt].to->bytes,
				nbytes) <= 0)
		{
		  /* Find the UCS value for `bytes'.  */
		  int inner;
		  uint32_t wch;
		  struct charseq *seq =
		    charmap_find_symbol (charmap, bytes, nbytes);

		  if (seq == NULL)
		    wch = ILLEGAL_CHAR_VALUE;
		  else if (seq->ucs4 != UNINITIALIZED_CHAR_VALUE)
		    wch = seq->ucs4;
		  else
		    wch = repertoire_find_value (ctype->repertoire, seq->name,
						 strlen (seq->name));

		  if (wch != ILLEGAL_CHAR_VALUE)
		    {
		      /* Store the value.  */
		      size_t nr = wch % ctype->plane_size;
		      size_t depth = 0;

		      while (ctype->names[nr + depth * ctype->plane_size] != wch)
			{
			  ++depth;
			  assert (depth < ctype->plane_cnt);
			}

		      ctype->width[nr + depth * ctype->plane_size]
			= charmap->width_rules[cnt].width;
		    }

		  /* "Increment" the bytes sequence.  */
		  inner = nbytes - 1;
		  while (inner >= 0 && bytes[inner] == 0xff)
		    --inner;

		  if (inner < 0)
		    {
		      /* We have to extend the byte sequence.  */
		      if (nbytes >= charmap->width_rules[cnt].to->nbytes)
			break;

		      bytes[0] = 1;
		      memset (&bytes[1], 0, nbytes);
		      ++nbytes;
		    }
		  else
		    {
		      ++bytes[inner];
		      while (++inner < nbytes)
			bytes[inner] = 0;
		    }
		}
	    }
	}

      /* Now set all the other characters of the character set to the
	 default width.  */
      curs = NULL;
      while (iterate_table (&charmap->char_table, &curs, &key, &len, &vdata) == 0)
	{
	  struct charseq *data = (struct charseq *) vdata;
	  size_t nr;
	  size_t depth;

	  if (data->ucs4 == UNINITIALIZED_CHAR_VALUE)
	    data->ucs4 = repertoire_find_value (ctype->repertoire,
						data->name, len);

	  if (data->ucs4 != ILLEGAL_CHAR_VALUE)
	    {
	      nr = data->ucs4 % ctype->plane_size;
	      depth = 0;

	      while (ctype->names[nr + depth * ctype->plane_size] != data->ucs4)
		{
		  ++depth;
		  assert (depth < ctype->plane_cnt);
		}

	      if (ctype->width[nr + depth * ctype->plane_size]
		  == (unsigned char) '\xff')
		ctype->width[nr + depth * ctype->plane_size] =
		  charmap->width_default;
	    }
	}
    }
  else
    {
      struct wcwidth_table t;

      t.p = 7;
      t.q = 9;
      wcwidth_table_init (&t);

      /* First set all the characters of the character set to the default width.  */
      curs = NULL;
      while (iterate_table (&charmap->char_table, &curs, &key, &len, &vdata) == 0)
	{
	  struct charseq *data = (struct charseq *) vdata;

	  if (data->ucs4 == UNINITIALIZED_CHAR_VALUE)
	    data->ucs4 = repertoire_find_value (ctype->repertoire,
						data->name, len);

	  if (data->ucs4 != ILLEGAL_CHAR_VALUE)
	    wcwidth_table_add (&t, data->ucs4, charmap->width_default);
	}

      /* Now add the explicitly specified widths.  */
      if (charmap->width_rules != NULL)
	{
	  size_t cnt;

	  for (cnt = 0; cnt < charmap->nwidth_rules; ++cnt)
	    {
	      unsigned char bytes[charmap->mb_cur_max];
	      int nbytes = charmap->width_rules[cnt].from->nbytes;

	      /* We have the range of character for which the width is
		 specified described using byte sequences of the multibyte
		 charset.  We have to convert this to UCS4 now.  And we
		 cannot simply convert the beginning and the end of the
		 sequence, we have to iterate over the byte sequence and
		 convert it for every single character.  */
	      memcpy (bytes, charmap->width_rules[cnt].from->bytes, nbytes);

	      while (nbytes < charmap->width_rules[cnt].to->nbytes
		     || memcmp (bytes, charmap->width_rules[cnt].to->bytes,
				nbytes) <= 0)
		{
		  /* Find the UCS value for `bytes'.  */
		  int inner;
		  uint32_t wch;
		  struct charseq *seq =
		    charmap_find_symbol (charmap, bytes, nbytes);

		  if (seq == NULL)
		    wch = ILLEGAL_CHAR_VALUE;
		  else if (seq->ucs4 != UNINITIALIZED_CHAR_VALUE)
		    wch = seq->ucs4;
		  else
		    wch = repertoire_find_value (ctype->repertoire, seq->name,
						 strlen (seq->name));

		  if (wch != ILLEGAL_CHAR_VALUE)
		    /* Store the value.  */
		    wcwidth_table_add (&t, wch, charmap->width_rules[cnt].width);

		  /* "Increment" the bytes sequence.  */
		  inner = nbytes - 1;
		  while (inner >= 0 && bytes[inner] == 0xff)
		    --inner;

		  if (inner < 0)
		    {
		      /* We have to extend the byte sequence.  */
		      if (nbytes >= charmap->width_rules[cnt].to->nbytes)
			break;

		      bytes[0] = 1;
		      memset (&bytes[1], 0, nbytes);
		      ++nbytes;
		    }
		  else
		    {
		      ++bytes[inner];
		      while (++inner < nbytes)
			bytes[inner] = 0;
		    }
		}
	    }
	}

      wcwidth_table_finalize (&t);

      if (verbose)
	fprintf (stderr, _("%s: table for width: %lu bytes\n"),
		 "LC_CTYPE", (unsigned long int) t.result_size);

      ctype->width_3level.iov_base = t.result;
      ctype->width_3level.iov_len = t.result_size;
    }

  /* Set MB_CUR_MAX.  */
  ctype->mb_cur_max = charmap->mb_cur_max;

  /* Now determine the table for the transliteration information.

     XXX It is not yet clear to me whether it is worth implementing a
     complicated algorithm which uses a hash table to locate the entries.
     For now I'll use a simple array which can be searching using binary
     search.  */
  if (ctype->translit_copy_locale != NULL)
    {
      /* Fold in the transliteration information from the locale mentioned
	 in the `include' statement.  */
      struct locale_ctype_t *here = ctype;

      do
	{
	  struct localedef_t *other = find_locale (LC_CTYPE,
						   here->translit_copy_locale,
						   repertoire->name, charmap);

	  if (other == NULL)
	    {
	      error (0, 0, _("\
%s: transliteration data from locale `%s' not available"),
		     "LC_CTYPE", here->translit_copy_locale);
	      break;
	    }

	  here = other->categories[LC_CTYPE].ctype;

	  /* Enqueue the information if necessary.  */
	  if (here->translit != NULL)
	    {
	      struct translit_t *endp = here->translit;
	      while (endp->next != NULL)
		endp = endp->next;

	      endp->next = ctype->translit;
	      ctype->translit = here->translit;
	    }
	}
      while (here->translit_copy_locale != NULL);
    }

  if (ctype->translit != NULL)
    {
      /* First count how many entries we have.  This is the upper limit
	 since some entries from the included files might be overwritten.  */
      size_t number = 0;
      size_t cnt;
      struct translit_t *runp = ctype->translit;
      struct translit_t **sorted;
      size_t from_len, to_len;

      while (runp != NULL)
	{
	  ++number;
	  runp = runp->next;
	}

      /* Next we allocate an array large enough and fill in the values.  */
      sorted = (struct translit_t **) alloca (number
					      * sizeof (struct translit_t **));
      runp = ctype->translit;
      number = 0;
      do
	{
	  /* Search for the place where to insert this string.
	     XXX Better use a real sorting algorithm later.  */
	  size_t idx = 0;
	  int replace = 0;

	  while (idx < number)
	    {
	      int res = wcscmp ((const wchar_t *) sorted[idx]->from,
				(const wchar_t *) runp->from);
	      if (res == 0)
		{
		  replace = 1;
		  break;
		}
	      if (res > 0)
		break;
	      ++idx;
	    }

	  if (replace)
	    sorted[idx] = runp;
	  else
	    {
	      memmove (&sorted[idx + 1], &sorted[idx],
		       (number - idx) * sizeof (struct translit_t *));
	      sorted[idx] = runp;
	      ++number;
	    }

	  runp = runp->next;
	}
      while (runp != NULL);

      /* The next step is putting all the possible transliteration
	 strings in one memory block so that we can write it out.
	 We need several different blocks:
	 - index to the from-string array
	 - from-string array
	 - index to the to-string array
	 - to-string array.
      */
      from_len = to_len = 0;
      for (cnt = 0; cnt < number; ++cnt)
	{
	  struct translit_to_t *srunp;
	  from_len += wcslen ((const wchar_t *) sorted[cnt]->from) + 1;
	  srunp = sorted[cnt]->to;
	  while (srunp != NULL)
	    {
	      to_len += wcslen ((const wchar_t *) srunp->str) + 1;
	      srunp = srunp->next;
	    }
	  /* Plus one for the extra NUL character marking the end of
	     the list for the current entry.  */
	  ++to_len;
	}

      /* We can allocate the arrays for the results.  */
      ctype->translit_from_idx = xmalloc (number * sizeof (uint32_t));
      ctype->translit_from_tbl = xmalloc (from_len * sizeof (uint32_t));
      ctype->translit_to_idx = xmalloc (number * sizeof (uint32_t));
      ctype->translit_to_tbl = xmalloc (to_len * sizeof (uint32_t));

      from_len = 0;
      to_len = 0;
      for (cnt = 0; cnt < number; ++cnt)
	{
	  size_t len;
	  struct translit_to_t *srunp;

	  ctype->translit_from_idx[cnt] = from_len;
	  ctype->translit_to_idx[cnt] = to_len;

	  len = wcslen ((const wchar_t *) sorted[cnt]->from) + 1;
	  wmemcpy ((wchar_t *) &ctype->translit_from_tbl[from_len],
		   (const wchar_t *) sorted[cnt]->from, len);
	  from_len += len;

	  ctype->translit_to_idx[cnt] = to_len;
	  srunp = sorted[cnt]->to;
	  while (srunp != NULL)
	    {
	      len = wcslen ((const wchar_t *) srunp->str) + 1;
	      wmemcpy ((wchar_t *) &ctype->translit_to_tbl[to_len],
		       (const wchar_t *) srunp->str, len);
	      to_len += len;
	      srunp = srunp->next;
	    }
	  ctype->translit_to_tbl[to_len++] = L'\0';
	}

      /* Store the information about the length.  */
      ctype->translit_idx_size = number;
      ctype->translit_from_tbl_size = from_len * sizeof (uint32_t);
      ctype->translit_to_tbl_size = to_len * sizeof (uint32_t);
    }
  else
    {
      /* Provide some dummy pointers since we have nothing to write out.  */
      static uint32_t no_str = { 0 };

      ctype->translit_from_idx = &no_str;
      ctype->translit_from_tbl = &no_str;
      ctype->translit_to_tbl = &no_str;
      ctype->translit_idx_size = 0;
      ctype->translit_from_tbl_size = 0;
      ctype->translit_to_tbl_size = 0;
    }
}