mariadb/sql/filesort.cc

/* Copyright (C) 2000 MySQL AB & MySQL Finland AB & TCX DataKonsult AB  

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; version 2 of the License.

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

   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */


/* Sorts a database */

#include "mysql_priv.h"
#ifdef HAVE_STDDEF_H
#include <stddef.h>			/* for macro offsetof */
#endif
#include <m_ctype.h>
#include "sql_sort.h"

#ifndef THREAD
#define SKIP_DBUG_IN_FILESORT
#endif

	/* How to write record_ref. */

#define WRITE_REF(file,from) \
if (my_b_write((file),(byte*) (from),param->ref_length)) \
  DBUG_RETURN(1);

	/* functions defined in this file */

static char **make_char_array(register uint fields, uint length, myf my_flag);
static BUFFPEK *read_buffpek_from_file(IO_CACHE *buffer_file, uint count);
static ha_rows find_all_keys(SORTPARAM *param,SQL_SELECT *select,
			     uchar * *sort_keys, IO_CACHE *buffer_file,
			     IO_CACHE *tempfile,IO_CACHE *indexfile);
static int write_keys(SORTPARAM *param,uchar * *sort_keys,
		      uint count, IO_CACHE *buffer_file, IO_CACHE *tempfile);
static void make_sortkey(SORTPARAM *param,uchar *to, byte *ref_pos);
static void register_used_fields(SORTPARAM *param);
static int merge_index(SORTPARAM *param,uchar *sort_buffer,
		       BUFFPEK *buffpek,
		       uint maxbuffer,IO_CACHE *tempfile,
		       IO_CACHE *outfile);
static bool save_index(SORTPARAM *param,uchar **sort_keys, uint count, 
                       FILESORT_INFO *table_sort);
static uint suffix_length(ulong string_length);
static uint sortlength(THD *thd, SORT_FIELD *sortorder, uint s_length,
		       bool *multi_byte_charset);
static SORT_ADDON_FIELD *get_addon_fields(THD *thd, Field **ptabfield,
                                          uint sortlength, uint *plength);
static void unpack_addon_fields(struct st_sort_addon_field *addon_field,
                                byte *buff);

/*
  Sort a table

  SYNOPSIS
    filesort()
    table		Table to sort
    sortorder		How to sort the table
    s_length		Number of elements in sortorder	
    select		Condition to apply to the rows
    ha_maxrows		Return only this many rows
    sort_positions	Set to 1 if we want to force sorting by position
			(Needed by UPDATE/INSERT or ALTER TABLE)
    examined_rows	Store number of examined rows here

  IMPLEMENTATION
    Creates a set of pointers that can be used to read the rows
    in sorted order. This should be done with the functions
    in records.cc
  
  REQUIREMENTS
    Before calling filesort, one must have done
    table->file->info(HA_STATUS_VARIABLE)

  NOTES
    If we sort by position (like if sort_positions is 1) filesort() will
    call table->prepare_for_position().

  RETURN
    HA_POS_ERROR	Error
    #			Number of rows

    examined_rows	will be set to number of examined rows

    The result set is stored in table->io_cache or
    table->record_pointers
*/

ha_rows filesort(THD *thd, TABLE *table, SORT_FIELD *sortorder, uint s_length,
		 SQL_SELECT *select, ha_rows max_rows,
                 bool sort_positions, ha_rows *examined_rows)
{
  int error;
  ulong memavl, min_sort_memory;
  uint maxbuffer;
  BUFFPEK *buffpek;
  ha_rows records= HA_POS_ERROR;
  uchar **sort_keys;
  IO_CACHE tempfile, buffpek_pointers, *selected_records_file, *outfile; 
  SORTPARAM param;
  bool multi_byte_charset;
  DBUG_ENTER("filesort");
  DBUG_EXECUTE("info",TEST_filesort(sortorder,s_length););
#ifdef SKIP_DBUG_IN_FILESORT
  DBUG_PUSH("");		/* No DBUG here */
#endif
  FILESORT_INFO table_sort;
  TABLE_LIST *tab= table->pos_in_table_list;
  Item_subselect *subselect= tab ? tab->containing_subselect() : 0;
  /* 
    Don't use table->sort in filesort as it is also used by 
    QUICK_INDEX_MERGE_SELECT. Work with a copy and put it back at the end 
    when index_merge select has finished with it.
  */
  memcpy(&table_sort, &table->sort, sizeof(FILESORT_INFO));
  table->sort.io_cache= NULL;
  
  outfile= table_sort.io_cache;
  my_b_clear(&tempfile);
  my_b_clear(&buffpek_pointers);
  buffpek=0;
  error= 1;
  bzero((char*) &param,sizeof(param));
  param.sort_length= sortlength(thd, sortorder, s_length, &multi_byte_charset);
  param.ref_length= table->file->ref_length;
  param.addon_field= 0;
  param.addon_length= 0;
  if (!(table->file->ha_table_flags() & HA_FAST_KEY_READ) &&
      !table->fulltext_searched && !sort_positions)
  {
    /* 
      Get the descriptors of all fields whose values are appended 
      to sorted fields and get its total length in param.spack_length.
    */
    param.addon_field= get_addon_fields(thd, table->field, 
                                        param.sort_length,
                                        &param.addon_length);
  }

  table_sort.addon_buf= 0;
  table_sort.addon_length= param.addon_length;
  table_sort.addon_field= param.addon_field;
  table_sort.unpack= unpack_addon_fields;
  if (param.addon_field)
  {
    param.res_length= param.addon_length;
    if (!(table_sort.addon_buf= (byte *) my_malloc(param.addon_length,
                                                    MYF(MY_WME))))
      goto err;
  }
  else
  {
    param.res_length= param.ref_length;
    /* 
      The reference to the record is considered 
      as an additional sorted field
    */
    param.sort_length+= param.ref_length;
  }
  param.rec_length= param.sort_length+param.addon_length;
  param.max_rows= max_rows;

  if (select && select->quick)
  {
    statistic_increment(thd->status_var.filesort_range_count, &LOCK_status);
  }
  else
  {
    statistic_increment(thd->status_var.filesort_scan_count, &LOCK_status);
  }
#ifdef CAN_TRUST_RANGE
  if (select && select->quick && select->quick->records > 0L)
  {
    records=min((ha_rows) (select->quick->records*2+EXTRA_RECORDS*2),
		table->file->stats.records)+EXTRA_RECORDS;
    selected_records_file=0;
  }
  else
#endif
  {
    records= table->file->estimate_rows_upper_bound();
    /*
      If number of records is not known, use as much of sort buffer 
      as possible. 
    */
    if (records == HA_POS_ERROR)
      records--;  // we use 'records+1' below.
    selected_records_file= 0;
  }

  if (multi_byte_charset &&
      !(param.tmp_buffer=my_malloc(param.sort_length,MYF(MY_WME))))
    goto err;

  memavl= thd->variables.sortbuff_size;
  min_sort_memory= max(MIN_SORT_MEMORY, param.sort_length*MERGEBUFF2);
  while (memavl >= min_sort_memory)
  {
    ulong old_memavl;
    ulong keys= memavl/(param.rec_length+sizeof(char*));
    param.keys=(uint) min(records+1, keys);
    if (table_sort.sort_keys ||
        (table_sort.sort_keys= (uchar **) make_char_array(param.keys, param.rec_length,
                                               MYF(0))))
      break;
    old_memavl=memavl;
    if ((memavl=memavl/4*3) < min_sort_memory && old_memavl > min_sort_memory)
      memavl= min_sort_memory;
  }
  sort_keys= table_sort.sort_keys;
  if (memavl < min_sort_memory)
  {
    my_error(ER_OUTOFMEMORY,MYF(ME_ERROR+ME_WAITTANG),
	     thd->variables.sortbuff_size);
    goto err;
  }
  if (open_cached_file(&buffpek_pointers,mysql_tmpdir,TEMP_PREFIX,
		       DISK_BUFFER_SIZE, MYF(MY_WME)))
    goto err;

  param.keys--;  			/* TODO: check why we do this */
  param.sort_form= table;
  param.end=(param.local_sortorder=sortorder)+s_length;
  if ((records=find_all_keys(&param,select,sort_keys, &buffpek_pointers,
			     &tempfile, selected_records_file)) ==
      HA_POS_ERROR)
    goto err;
  maxbuffer= (uint) (my_b_tell(&buffpek_pointers)/sizeof(*buffpek));

  if (maxbuffer == 0)			// The whole set is in memory
  {
    if (save_index(&param,sort_keys,(uint) records, &table_sort))
      goto err;
  }
  else
  {
    if (!table_sort.buffpek && table_sort.buffpek_len < maxbuffer &&
        !(table_sort.buffpek=
          (byte *) read_buffpek_from_file(&buffpek_pointers, maxbuffer)))
      goto err;
    buffpek= (BUFFPEK *) table_sort.buffpek;
    table_sort.buffpek_len= maxbuffer;
    close_cached_file(&buffpek_pointers);
	/* Open cached file if it isn't open */
    if (! my_b_inited(outfile) &&
	open_cached_file(outfile,mysql_tmpdir,TEMP_PREFIX,READ_RECORD_BUFFER,
			  MYF(MY_WME)))
      goto err;
    reinit_io_cache(outfile,WRITE_CACHE,0L,0,0);

    /*
      Use also the space previously used by string pointers in sort_buffer
      for temporary key storage.
    */
    param.keys=((param.keys*(param.rec_length+sizeof(char*))) /
		param.rec_length-1);
    maxbuffer--;				// Offset from 0
    if (merge_many_buff(&param,(uchar*) sort_keys,buffpek,&maxbuffer,
			&tempfile))
      goto err;
    if (flush_io_cache(&tempfile) ||
	reinit_io_cache(&tempfile,READ_CACHE,0L,0,0))
      goto err;
    if (merge_index(&param,(uchar*) sort_keys,buffpek,maxbuffer,&tempfile,
		    outfile))
      goto err;
  }
  if (records > param.max_rows)
    records=param.max_rows;
  error =0;

 err:
  if (param.tmp_buffer)
    x_free(param.tmp_buffer);
  if (!subselect || !subselect->is_uncacheable())
  {
    x_free((gptr) sort_keys);
    table_sort.sort_keys= 0;
    x_free((gptr) buffpek);
    table_sort.buffpek= 0;
    table_sort.buffpek_len= 0;
  }
  close_cached_file(&tempfile);
  close_cached_file(&buffpek_pointers);
  if (my_b_inited(outfile))
  {
    if (flush_io_cache(outfile))
      error=1;
    {
      my_off_t save_pos=outfile->pos_in_file;
      /* For following reads */
      if (reinit_io_cache(outfile,READ_CACHE,0L,0,0))
	error=1;
      outfile->end_of_file=save_pos;
    }
  }
  if (error)
    my_message(ER_FILSORT_ABORT, ER(ER_FILSORT_ABORT),
               MYF(ME_ERROR+ME_WAITTANG));
  else
    statistic_add(thd->status_var.filesort_rows,
		  (ulong) records, &LOCK_status);
  *examined_rows= param.examined_rows;
#ifdef SKIP_DBUG_IN_FILESORT
  DBUG_POP();			/* Ok to DBUG */
#endif
  memcpy(&table->sort, &table_sort, sizeof(FILESORT_INFO));
  DBUG_PRINT("exit",("records: %ld", (long) records));
  DBUG_RETURN(error ? HA_POS_ERROR : records);
} /* filesort */


void filesort_free_buffers(TABLE *table, bool full)
{
  if (table->sort.record_pointers)
  {
    my_free((gptr) table->sort.record_pointers,MYF(0));
    table->sort.record_pointers=0;
  }
  if (full)
  {
    if (table->sort.sort_keys )
    {
      x_free((gptr) table->sort.sort_keys);
      table->sort.sort_keys= 0;
    }
    if (table->sort.buffpek)
    {
      x_free((gptr) table->sort.buffpek);
      table->sort.buffpek= 0;
      table->sort.buffpek_len= 0;
    }
  }
  if (table->sort.addon_buf)
  {
    my_free((char *) table->sort.addon_buf, MYF(0));
    my_free((char *) table->sort.addon_field, MYF(MY_ALLOW_ZERO_PTR));
    table->sort.addon_buf=0;
    table->sort.addon_field=0;
  }
}

	/* Make a array of string pointers */

static char **make_char_array(register uint fields, uint length, myf my_flag)
{
  register char **pos;
  char **old_pos,*char_pos;
  DBUG_ENTER("make_char_array");

  if ((old_pos= (char**) my_malloc((uint) fields*(length+sizeof(char*)),
				    my_flag)))
  {
    pos=old_pos; char_pos=((char*) (pos+fields)) -length;
    while (fields--) *(pos++) = (char_pos+= length);
  }

  DBUG_RETURN(old_pos);
} /* make_char_array */


/* Read 'count' number of buffer pointers into memory */

static BUFFPEK *read_buffpek_from_file(IO_CACHE *buffpek_pointers, uint count)
{
  ulong length;
  BUFFPEK *tmp;
  DBUG_ENTER("read_buffpek_from_file");
  tmp=(BUFFPEK*) my_malloc(length=sizeof(BUFFPEK)*count, MYF(MY_WME));
  if (tmp)
  {
    if (reinit_io_cache(buffpek_pointers,READ_CACHE,0L,0,0) ||
	my_b_read(buffpek_pointers, (byte*) tmp, length))
    {
      my_free((char*) tmp, MYF(0));
      tmp=0;
    }
  }
  DBUG_RETURN(tmp);
}


/* 
  Search after sort_keys and write them into tempfile.
  SYNOPSIS
    find_all_keys()
      param             Sorting parameter
      select            Use this to get source data
      sort_keys         Array of pointers to sort key + addon buffers.
      buffpek_pointers  File to write BUFFPEKs describing sorted segments
                        in tempfile.
      tempfile          File to write sorted sequences of sortkeys to.
      indexfile         If !NULL, use it for source data (contains rowids)
  
  NOTE
    Basic idea:
      while (get_next_sortkey())
      {
        if (no free space in sort_keys buffers) 
        {
          sort sort_keys buffer;
          dump sorted sequence to 'tempfile';
          dump BUFFPEK describing sequence location into 'buffpek_pointers';
        }
        put sort key into 'sort_keys';
      }
      if (sort_keys has some elements && dumped at least once)
        sort-dump-dump as above;
      else
        don't sort, leave sort_keys array to be sorted by caller.
    
     All produced sequences are guaranteed to be non-empty.
  RETURN
    Number of records written on success.
    HA_POS_ERROR on error.
*/

static ha_rows find_all_keys(SORTPARAM *param, SQL_SELECT *select,
			     uchar **sort_keys,
			     IO_CACHE *buffpek_pointers,
			     IO_CACHE *tempfile, IO_CACHE *indexfile)
{
  int error,flag,quick_select;
  uint idx,indexpos,ref_length;
  byte *ref_pos,*next_pos,ref_buff[MAX_REFLENGTH];
  my_off_t record;
  TABLE *sort_form;
  volatile THD::killed_state *killed= &current_thd->killed;
  handler *file;
  MY_BITMAP *save_read_set, *save_write_set;
  DBUG_ENTER("find_all_keys");
  DBUG_PRINT("info",("using: %s",
                     (select ? select->quick ? "ranges" : "where":
                      "every row")));

  idx=indexpos=0;
  error=quick_select=0;
  sort_form=param->sort_form;
  file=sort_form->file;
  ref_length=param->ref_length;
  ref_pos= ref_buff;
  quick_select=select && select->quick;
  record=0;
  flag= ((!indexfile && file->ha_table_flags() & HA_REC_NOT_IN_SEQ)
	 || quick_select);
  if (indexfile || flag)
    ref_pos= &file->ref[0];
  next_pos=ref_pos;
  if (! indexfile && ! quick_select)
  {
    next_pos=(byte*) 0;			/* Find records in sequence */
    file->ha_rnd_init(1);
    file->extra_opt(HA_EXTRA_CACHE,
		    current_thd->variables.read_buff_size);
  }

  READ_RECORD read_record_info;
  if (quick_select)
  {
    if (select->quick->reset())
      DBUG_RETURN(HA_POS_ERROR);
    init_read_record(&read_record_info, current_thd, select->quick->head,
                     select, 1, 1);
  }

  /* Remember original bitmaps */
  save_read_set=  sort_form->read_set;
  save_write_set= sort_form->write_set;
  /* Set up temporary column read map for columns used by sort */
  bitmap_clear_all(&sort_form->tmp_set);
  /* Temporary set for register_used_fields and register_field_in_read_map */
  sort_form->read_set= &sort_form->tmp_set;
  register_used_fields(param);
  if (select && select->cond)
    select->cond->walk(&Item::register_field_in_read_map, 1,
                       (byte*) sort_form);
  sort_form->column_bitmaps_set(&sort_form->tmp_set, &sort_form->tmp_set);

  for (;;)
  {
    if (quick_select)
    {
      if ((error= read_record_info.read_record(&read_record_info)))
      {
        error= HA_ERR_END_OF_FILE;
        break;
      }
      file->position(sort_form->record[0]);
    }
    else					/* Not quick-select */
    {
      if (indexfile)
      {
	if (my_b_read(indexfile,(byte*) ref_pos,ref_length)) /* purecov: deadcode */
	{
	  error= my_errno ? my_errno : -1;		/* Abort */
	  break;
	}
	error=file->rnd_pos(sort_form->record[0],next_pos);
      }
      else
      {
	error=file->rnd_next(sort_form->record[0]);
	if (!flag)
	{
	  my_store_ptr(ref_pos,ref_length,record); // Position to row
	  record+= sort_form->s->db_record_offset;
	}
	else if (!error)
	  file->position(sort_form->record[0]);
      }
      if (error && error != HA_ERR_RECORD_DELETED)
	break;
    }

    if (*killed)
    {
      DBUG_PRINT("info",("Sort killed by user"));
      if (!indexfile && !quick_select)
      {
        (void) file->extra(HA_EXTRA_NO_CACHE);
        file->ha_rnd_end();
      }
      DBUG_RETURN(HA_POS_ERROR);		/* purecov: inspected */
    }
    if (error == 0)
      param->examined_rows++;
    if (error == 0 && (!select || select->skip_record() == 0))
    {
      if (idx == param->keys)
      {
	if (write_keys(param,sort_keys,idx,buffpek_pointers,tempfile))
	  DBUG_RETURN(HA_POS_ERROR);
	idx=0;
	indexpos++;
      }
      make_sortkey(param,sort_keys[idx++],ref_pos);
    }
    else
      file->unlock_row();
  }
  if (quick_select)
  {
    /*
      index_merge quick select uses table->sort when retrieving rows, so free
      resoures it has allocated.
    */
    end_read_record(&read_record_info);
  }
  else
  {
    (void) file->extra(HA_EXTRA_NO_CACHE);	/* End cacheing of records */
    if (!next_pos)
      file->ha_rnd_end();
  }

  /* Signal we should use orignal column read and write maps */
  sort_form->column_bitmaps_set(save_read_set, save_write_set);

  DBUG_PRINT("test",("error: %d  indexpos: %d",error,indexpos));
  if (error != HA_ERR_END_OF_FILE)
  {
    file->print_error(error,MYF(ME_ERROR | ME_WAITTANG)); /* purecov: inspected */
    DBUG_RETURN(HA_POS_ERROR);			/* purecov: inspected */
  }
  if (indexpos && idx &&
      write_keys(param,sort_keys,idx,buffpek_pointers,tempfile))
    DBUG_RETURN(HA_POS_ERROR);			/* purecov: inspected */
  DBUG_RETURN(my_b_inited(tempfile) ?
	      (ha_rows) (my_b_tell(tempfile)/param->rec_length) :
	      idx);
} /* find_all_keys */


/*
  Sort the buffer and write:
    1) the sorted sequence to tempfile
    2) a BUFFPEK describing the sorted sequence position to buffpek_pointers
  (was: Skriver en buffert med nycklar till filen)
  SYNOPSIS
    write_keys()
      param             Sort parameters
      sort_keys         Array of pointers to keys to sort
      count             Number of elements in sort_keys array 
      buffpek_pointers  One 'BUFFPEK' struct will be written into this file.
                        The BUFFPEK::{file_pos, count} will indicate where 
                        the sorted data was stored.
      tempfile          The sorted sequence will be written into this file.
    
  RETURN
    0 OK
    1 Error
*/

static int
write_keys(SORTPARAM *param, register uchar **sort_keys, uint count,
           IO_CACHE *buffpek_pointers, IO_CACHE *tempfile)
{
  uint sort_length, rec_length;
  uchar **end;
  BUFFPEK buffpek;
  DBUG_ENTER("write_keys");

  sort_length= param->sort_length;
  rec_length= param->rec_length;
#ifdef MC68000
  quicksort(sort_keys,count,sort_length);
#else
  my_string_ptr_sort((gptr) sort_keys, (uint) count, sort_length);
#endif
  if (!my_b_inited(tempfile) &&
      open_cached_file(tempfile, mysql_tmpdir, TEMP_PREFIX, DISK_BUFFER_SIZE,
                       MYF(MY_WME)))
    goto err;                                   /* purecov: inspected */
  buffpek.file_pos= my_b_tell(tempfile);
  if ((ha_rows) count > param->max_rows)
    count=(uint) param->max_rows;               /* purecov: inspected */
  buffpek.count=(ha_rows) count;
  for (end=sort_keys+count ; sort_keys != end ; sort_keys++)
    if (my_b_write(tempfile, (byte*) *sort_keys, (uint) rec_length))
      goto err;
  if (my_b_write(buffpek_pointers, (byte*) &buffpek, sizeof(buffpek)))
    goto err;
  DBUG_RETURN(0);

err:
  DBUG_RETURN(1);
} /* write_keys */


/*
  Store length as suffix in high-byte-first order
*/

static inline void store_length(uchar *to, uint length, uint pack_length)
{
  switch (pack_length) {
  case 1:
    *to= (uchar) length;
    break;
  case 2:
    mi_int2store(to, length);
    break;
  case 3:
    mi_int3store(to, length);
    break;
  default:
    mi_int4store(to, length);
    break;
  }
}


	/* makes a sort-key from record */

static void make_sortkey(register SORTPARAM *param,
			 register uchar *to, byte *ref_pos)
{
  reg3 Field *field;
  reg1 SORT_FIELD *sort_field;
  reg5 uint length;

  for (sort_field=param->local_sortorder ;
       sort_field != param->end ;
       sort_field++)
  {
    bool maybe_null=0;
    if ((field=sort_field->field))
    {						// Field
      if (field->maybe_null())
      {
	if (field->is_null())
	{
	  if (sort_field->reverse)
	    bfill(to,sort_field->length+1,(char) 255);
	  else
	    bzero((char*) to,sort_field->length+1);
	  to+= sort_field->length+1;
	  continue;
	}
	else
	  *to++=1;
      }
      field->sort_string((char*) to,sort_field->length);
    }
    else
    {						// Item
      Item *item=sort_field->item;
      maybe_null= item->maybe_null;
      switch (sort_field->result_type) {
      case STRING_RESULT:
      {
        CHARSET_INFO *cs=item->collation.collation;
        char fill_char= ((cs->state & MY_CS_BINSORT) ? (char) 0 : ' ');
        int diff;
        uint sort_field_length;

        if (maybe_null)
          *to++=1;
        /* All item->str() to use some extra byte for end null.. */
        String tmp((char*) to,sort_field->length+4,cs);
        String *res= item->str_result(&tmp);
        if (!res)
        {
          if (maybe_null)
            bzero((char*) to-1,sort_field->length+1);
          else
          {
            /* purecov: begin deadcode */
            /*
              This should only happen during extreme conditions if we run out
              of memory or have an item marked not null when it can be null.
              This code is here mainly to avoid a hard crash in this case.
            */
            DBUG_ASSERT(0);
            DBUG_PRINT("warning",
                       ("Got null on something that shouldn't be null"));
            bzero((char*) to,sort_field->length);	// Avoid crash
            /* purecov: end */
          }
          break;
        }
        length= res->length();
        sort_field_length= sort_field->length - sort_field->suffix_length;
        diff=(int) (sort_field_length - length);
        if (diff < 0)
        {
          diff=0;
          length= sort_field_length;
        }
        if (sort_field->suffix_length)
        {
          /* Store length last in result_string */
          store_length(to + sort_field_length, length,
                       sort_field->suffix_length);
        }
        if (sort_field->need_strxnfrm)
        {
          char *from=(char*) res->ptr();
          uint tmp_length;
          if ((uchar*) from == to)
          {
            set_if_smaller(length,sort_field->length);
            memcpy(param->tmp_buffer,from,length);
            from=param->tmp_buffer;
          }
          tmp_length= my_strnxfrm(cs,to,sort_field->length,
                                  (uchar*) from, length);
          DBUG_ASSERT(tmp_length == sort_field->length);
        }
        else
        {
          my_strnxfrm(cs,(uchar*)to,length,(const uchar*)res->ptr(),length);
          cs->cset->fill(cs, (char *)to+length,diff,fill_char);
        }
        break;
      }
      case INT_RESULT:
	{
          longlong value= item->val_int_result();
          if (maybe_null)
          {
	    *to++=1;				/* purecov: inspected */
            if (item->null_value)
            {
              if (maybe_null)
                bzero((char*) to-1,sort_field->length+1);
              else
              {
                DBUG_PRINT("warning",
                           ("Got null on something that shouldn't be null"));
                bzero((char*) to,sort_field->length);
              }
              break;
            }
          }
#if SIZEOF_LONG_LONG > 4
	  to[7]= (uchar) value;
	  to[6]= (uchar) (value >> 8);
	  to[5]= (uchar) (value >> 16);
	  to[4]= (uchar) (value >> 24);
	  to[3]= (uchar) (value >> 32);
	  to[2]= (uchar) (value >> 40);
	  to[1]= (uchar) (value >> 48);
          if (item->unsigned_flag)                    /* Fix sign */
            to[0]= (uchar) (value >> 56);
          else
            to[0]= (uchar) (value >> 56) ^ 128;	/* Reverse signbit */
#else
	  to[3]= (uchar) value;
	  to[2]= (uchar) (value >> 8);
	  to[1]= (uchar) (value >> 16);
          if (item->unsigned_flag)                    /* Fix sign */
            to[0]= (uchar) (value >> 24);
          else
            to[0]= (uchar) (value >> 24) ^ 128;	/* Reverse signbit */
#endif
	  break;
	}
      case DECIMAL_RESULT:
        {
          my_decimal dec_buf, *dec_val= item->val_decimal_result(&dec_buf);
          if (maybe_null)
          {
            if (item->null_value)
            { 
              bzero((char*)to, sort_field->length+1);
              to++;
              break;
            }
            *to++=1;
          }
          my_decimal2binary(E_DEC_FATAL_ERROR, dec_val, (char*)to,
                            item->max_length - (item->decimals ? 1:0),
                            item->decimals);
         break;
        }
      case REAL_RESULT:
	{
          double value= item->val_result();
	  if (maybe_null)
          {
            if (item->null_value)
            {
              bzero((char*) to,sort_field->length+1);
              to++;
              break;
            }
	    *to++=1;
          }
	  change_double_for_sort(value,(byte*) to);
	  break;
	}
      case ROW_RESULT:
      default: 
	// This case should never be choosen
	DBUG_ASSERT(0);
	break;
      }
    }
    if (sort_field->reverse)
    {							/* Revers key */
      if (maybe_null)
        to[-1]= ~to[-1];
      length=sort_field->length;
      while (length--)
      {
	*to = (uchar) (~ *to);
	to++;
      }
    }
    else
      to+= sort_field->length;
  }

  if (param->addon_field)
  {
    /* 
      Save field values appended to sorted fields.
      First null bit indicators are appended then field values follow.
      In this implementation we use fixed layout for field values -
      the same for all records.
    */
    SORT_ADDON_FIELD *addonf= param->addon_field;
    uchar *nulls= to;
    DBUG_ASSERT(addonf != 0);
    bzero((char *) nulls, addonf->offset);
    to+= addonf->offset;
    for ( ; (field= addonf->field) ; addonf++)
    {
      if (addonf->null_bit && field->is_null())
      {
        nulls[addonf->null_offset]|= addonf->null_bit;
#ifdef HAVE_purify
	bzero(to, addonf->length);
#endif
      }
      else
      {
        uchar *end= (uchar*) field->pack((char *) to, field->ptr);
#ifdef HAVE_purify
	uint length= (uint) ((to + addonf->length) - end);
	DBUG_ASSERT((int) length >= 0);
	if (length)
	  bzero(end, length);
#endif
      }
      to+= addonf->length;
    }
  }
  else
  {
    /* Save filepos last */
    memcpy((byte*) to, ref_pos, (size_s) param->ref_length);
  }
  return;
}


/*
  Register fields used by sorting in the sorted table's read set
*/

static void register_used_fields(SORTPARAM *param)
{
  reg1 SORT_FIELD *sort_field;
  TABLE *table=param->sort_form;
  MY_BITMAP *bitmap= table->read_set;

  for (sort_field= param->local_sortorder ;
       sort_field != param->end ;
       sort_field++)
  {
    Field *field;
    if ((field= sort_field->field))
    {
      if (field->table == table)
      bitmap_set_bit(bitmap, field->field_index);
    }
    else
    {						// Item
      sort_field->item->walk(&Item::register_field_in_read_map, 1,
                             (byte *) table);
    }
  }

  if (param->addon_field)
  {
    SORT_ADDON_FIELD *addonf= param->addon_field;
    Field *field;
    for ( ; (field= addonf->field) ; addonf++)
      bitmap_set_bit(bitmap, field->field_index);
  }
  else
  {
    /* Save filepos last */
    table->prepare_for_position();
  }
}


static bool save_index(SORTPARAM *param, uchar **sort_keys, uint count, 
                       FILESORT_INFO *table_sort)
{
  uint offset,res_length;
  byte *to;
  DBUG_ENTER("save_index");

  my_string_ptr_sort((gptr) sort_keys, (uint) count, param->sort_length);
  res_length= param->res_length;
  offset= param->rec_length-res_length;
  if ((ha_rows) count > param->max_rows)
    count=(uint) param->max_rows;
  if (!(to= table_sort->record_pointers= 
        (byte*) my_malloc(res_length*count, MYF(MY_WME))))
    DBUG_RETURN(1);                 /* purecov: inspected */
  for (uchar **end= sort_keys+count ; sort_keys != end ; sort_keys++)
  {
    memcpy(to, *sort_keys+offset, res_length);
    to+= res_length;
  }
  DBUG_RETURN(0);
}


	/* Merge buffers to make < MERGEBUFF2 buffers */

int merge_many_buff(SORTPARAM *param, uchar *sort_buffer,
		    BUFFPEK *buffpek, uint *maxbuffer, IO_CACHE *t_file)
{
  register int i;
  IO_CACHE t_file2,*from_file,*to_file,*temp;
  BUFFPEK *lastbuff;
  DBUG_ENTER("merge_many_buff");

  if (*maxbuffer < MERGEBUFF2)
    DBUG_RETURN(0);				/* purecov: inspected */
  if (flush_io_cache(t_file) ||
      open_cached_file(&t_file2,mysql_tmpdir,TEMP_PREFIX,DISK_BUFFER_SIZE,
			MYF(MY_WME)))
    DBUG_RETURN(1);				/* purecov: inspected */

  from_file= t_file ; to_file= &t_file2;
  while (*maxbuffer >= MERGEBUFF2)
  {
    reinit_io_cache(from_file,READ_CACHE,0L,0,0);
    reinit_io_cache(to_file,WRITE_CACHE,0L,0,0);
    lastbuff=buffpek;
    for (i=0 ; i <= (int) *maxbuffer-MERGEBUFF*3/2 ; i+=MERGEBUFF)
    {
      if (merge_buffers(param,from_file,to_file,sort_buffer,lastbuff++,
			buffpek+i,buffpek+i+MERGEBUFF-1,0))
	break;					/* purecov: inspected */
    }
    if (merge_buffers(param,from_file,to_file,sort_buffer,lastbuff++,
		      buffpek+i,buffpek+ *maxbuffer,0))
      break;					/* purecov: inspected */
    if (flush_io_cache(to_file))
      break;					/* purecov: inspected */
    temp=from_file; from_file=to_file; to_file=temp;
    setup_io_cache(from_file);
    setup_io_cache(to_file);
    *maxbuffer= (uint) (lastbuff-buffpek)-1;
  }
  close_cached_file(to_file);			// This holds old result
  if (to_file == t_file)
  {
    *t_file=t_file2;				// Copy result file
    setup_io_cache(t_file);
  }

  DBUG_RETURN(*maxbuffer >= MERGEBUFF2);	/* Return 1 if interrupted */
} /* merge_many_buff */


	/* Read data to buffer */
	/* This returns (uint) -1 if something goes wrong */

uint read_to_buffer(IO_CACHE *fromfile, BUFFPEK *buffpek,
		    uint rec_length)
{
  register uint count;
  uint length;

  if ((count=(uint) min((ha_rows) buffpek->max_keys,buffpek->count)))
  {
    if (my_pread(fromfile->file,(byte*) buffpek->base,
		 (length= rec_length*count),buffpek->file_pos,MYF_RW))
      return((uint) -1);			/* purecov: inspected */
    buffpek->key=buffpek->base;
    buffpek->file_pos+= length;			/* New filepos */
    buffpek->count-=	count;
    buffpek->mem_count= count;
  }
  return (count*rec_length);
} /* read_to_buffer */


/*
    Put all room used by freed buffer to use in adjacent buffer.  Note, that
    we can't simply distribute memory evenly between all buffers, because
    new areas must not overlap with old ones.
  SYNOPSIS
    reuse_freed_buff()
    queue      IN  list of non-empty buffers, without freed buffer
    reuse      IN  empty buffer
    key_length IN  key length
*/

void reuse_freed_buff(QUEUE *queue, BUFFPEK *reuse, uint key_length)
{
  uchar *reuse_end= reuse->base + reuse->max_keys * key_length;
  for (uint i= 0; i < queue->elements; ++i)
  {
    BUFFPEK *bp= (BUFFPEK *) queue_element(queue, i);
    if (bp->base + bp->max_keys * key_length == reuse->base)
    {
      bp->max_keys+= reuse->max_keys;
      return;
    }
    else if (bp->base == reuse_end)
    {
      bp->base= reuse->base;
      bp->max_keys+= reuse->max_keys;
      return;
    }
  }
  DBUG_ASSERT(0);
}


/* 
  Merge buffers to one buffer
  SYNOPSIS
    merge_buffers()
      param        Sort parameter
      from_file    File with source data (BUFFPEKs point to this file)
      to_file      File to write the sorted result data.
      sort_buffer  Buffer for data to store up to MERGEBUFF2 sort keys.
      lastbuff     OUT Store here BUFFPEK describing data written to to_file                   
      Fb           First element in source BUFFPEKs array
      Tb           Last element in source BUFFPEKs array
      flag

  RETURN
    0     - OK
    other - error
*/

int merge_buffers(SORTPARAM *param, IO_CACHE *from_file,
                  IO_CACHE *to_file, uchar *sort_buffer,
                  BUFFPEK *lastbuff, BUFFPEK *Fb, BUFFPEK *Tb,
                  int flag)
{
  int error;
  uint rec_length,sort_length,res_length,offset;
  ulong maxcount;
  ha_rows max_rows,org_max_rows;
  my_off_t to_start_filepos;
  uchar *strpos;
  BUFFPEK *buffpek;
  QUEUE queue;
  qsort2_cmp cmp;
  volatile THD::killed_state *killed= &current_thd->killed;
  THD::killed_state not_killable;
  DBUG_ENTER("merge_buffers");

  statistic_increment(current_thd->status_var.filesort_merge_passes,
		      &LOCK_status);
  if (param->not_killable)
  {
    killed= &not_killable;
    not_killable= THD::NOT_KILLED;
  }

  error=0;
  rec_length= param->rec_length;
  res_length= param->res_length;
  sort_length= param->sort_length;
  offset= rec_length-res_length;
  maxcount= (ulong) (param->keys/((uint) (Tb-Fb) +1));
  to_start_filepos= my_b_tell(to_file);
  strpos= (uchar*) sort_buffer;
  org_max_rows=max_rows= param->max_rows;

  /* The following will fire if there is not enough space in sort_buffer */
  DBUG_ASSERT(maxcount!=0);
  
  if (init_queue(&queue, (uint) (Tb-Fb)+1, offsetof(BUFFPEK,key), 0,
                 (queue_compare) (cmp= get_ptr_compare(sort_length)),
                 (void*) &sort_length))
    DBUG_RETURN(1);                                /* purecov: inspected */
  for (buffpek= Fb ; buffpek <= Tb ; buffpek++)
  {
    buffpek->base= strpos;
    buffpek->max_keys= maxcount;
    strpos+= (uint) (error= (int) read_to_buffer(from_file, buffpek,
                                                                         rec_length));
    if (error == -1)
      goto err;					/* purecov: inspected */
    buffpek->max_keys= buffpek->mem_count;	// If less data in buffers than expected
    queue_insert(&queue, (byte*) buffpek);
  }

  if (param->unique_buff)
  {
    /* 
       Called by Unique::get()
       Copy the first argument to param->unique_buff for unique removal.
       Store it also in 'to_file'.

       This is safe as we know that there is always more than one element
       in each block to merge (This is guaranteed by the Unique:: algorithm
    */
    buffpek= (BUFFPEK*) queue_top(&queue);
    memcpy(param->unique_buff, buffpek->key, rec_length);
    if (my_b_write(to_file, (byte*) buffpek->key, rec_length))
    {
      error=1; goto err;                        /* purecov: inspected */
    }
    buffpek->key+= rec_length;
    buffpek->mem_count--;
    if (!--max_rows)
    {
      error= 0;                                       /* purecov: inspected */
      goto end;                                       /* purecov: inspected */
    }
    queue_replaced(&queue);                        // Top element has been used
  }
  else
    cmp= 0;                                        // Not unique

  while (queue.elements > 1)
  {
    if (*killed)
    {
      error= 1; goto err;                        /* purecov: inspected */
    }
    for (;;)
    {
      buffpek= (BUFFPEK*) queue_top(&queue);
      if (cmp)                                        // Remove duplicates
      {
        if (!(*cmp)(&sort_length, &(param->unique_buff),
                    (uchar**) &buffpek->key))
              goto skip_duplicate;
            memcpy(param->unique_buff, (uchar*) buffpek->key, rec_length);
      }
      if (flag == 0)
      {
        if (my_b_write(to_file,(byte*) buffpek->key, rec_length))
        {
          error=1; goto err;                        /* purecov: inspected */
        }
      }
      else
      {
        if (my_b_write(to_file, (byte*) buffpek->key+offset, res_length))
        {
          error=1; goto err;                        /* purecov: inspected */
        }
      }
      if (!--max_rows)
      {
        error= 0;                               /* purecov: inspected */
        goto end;                               /* purecov: inspected */
      }

    skip_duplicate:
      buffpek->key+= rec_length;
      if (! --buffpek->mem_count)
      {
        if (!(error= (int) read_to_buffer(from_file,buffpek,
                                          rec_length)))
        {
          VOID(queue_remove(&queue,0));
          reuse_freed_buff(&queue, buffpek, rec_length);
          break;                        /* One buffer have been removed */
        }
        else if (error == -1)
          goto err;                        /* purecov: inspected */
      }
      queue_replaced(&queue);              /* Top element has been replaced */
    }
  }
  buffpek= (BUFFPEK*) queue_top(&queue);
  buffpek->base= sort_buffer;
  buffpek->max_keys= param->keys;

  /*
    As we know all entries in the buffer are unique, we only have to
    check if the first one is the same as the last one we wrote
  */
  if (cmp)
  {
    if (!(*cmp)(&sort_length, &(param->unique_buff), (uchar**) &buffpek->key))
    {
      buffpek->key+= rec_length;         // Remove duplicate
      --buffpek->mem_count;
    }
  }

  do
  {
    if ((ha_rows) buffpek->mem_count > max_rows)
    {                                        /* Don't write too many records */
      buffpek->mem_count= (uint) max_rows;
      buffpek->count= 0;                        /* Don't read more */
    }
    max_rows-= buffpek->mem_count;
    if (flag == 0)
    {
      if (my_b_write(to_file,(byte*) buffpek->key,
                     (rec_length*buffpek->mem_count)))
      {
        error= 1; goto err;                        /* purecov: inspected */
      }
    }
    else
    {
      register uchar *end;
      strpos= buffpek->key+offset;
      for (end= strpos+buffpek->mem_count*rec_length ;
           strpos != end ;
           strpos+= rec_length)
      {     
        if (my_b_write(to_file, (byte *) strpos, res_length))
        {
          error=1; goto err;                        
        }
      }
    }
  }
  while ((error=(int) read_to_buffer(from_file,buffpek, rec_length))
         != -1 && error != 0);

end:
  lastbuff->count= min(org_max_rows-max_rows, param->max_rows);
  lastbuff->file_pos= to_start_filepos;
err:
  delete_queue(&queue);
  DBUG_RETURN(error);
} /* merge_buffers */


	/* Do a merge to output-file (save only positions) */

static int merge_index(SORTPARAM *param, uchar *sort_buffer,
		       BUFFPEK *buffpek, uint maxbuffer,
		       IO_CACHE *tempfile, IO_CACHE *outfile)
{
  DBUG_ENTER("merge_index");
  if (merge_buffers(param,tempfile,outfile,sort_buffer,buffpek,buffpek,
		    buffpek+maxbuffer,1))
    DBUG_RETURN(1);				/* purecov: inspected */
  DBUG_RETURN(0);
} /* merge_index */


static uint suffix_length(ulong string_length)
{
  if (string_length < 256)
    return 1;
  if (string_length < 256L*256L)
    return 2;
  if (string_length < 256L*256L*256L)
    return 3;
  return 4;                                     // Can't sort longer than 4G
}



/*
  Calculate length of sort key

  SYNOPSIS
    sortlength()
    thd			Thread handler
    sortorder		Order of items to sort
    uint s_length	Number of items to sort
    multi_byte_charset  (out)
			Set to 1 if we are using multi-byte charset
			(In which case we have to use strxnfrm())

  NOTES
    sortorder->length is updated for each sort item
    sortorder->need_strxnfrm is set 1 if we have to use strxnfrm

  RETURN
    Total length of sort buffer in bytes
*/

static uint
sortlength(THD *thd, SORT_FIELD *sortorder, uint s_length,
           bool *multi_byte_charset)
{
  reg2 uint length;
  CHARSET_INFO *cs;
  *multi_byte_charset= 0;

  length=0;
  for (; s_length-- ; sortorder++)
  {
    sortorder->need_strxnfrm= 0;
    sortorder->suffix_length= 0;
    if (sortorder->field)
    {
      cs= sortorder->field->sort_charset();
      sortorder->length= sortorder->field->sort_length();

      if (use_strnxfrm((cs=sortorder->field->sort_charset())))
      {
        sortorder->need_strxnfrm= 1;
        *multi_byte_charset= 1;
        sortorder->length= cs->coll->strnxfrmlen(cs, sortorder->length);
      }
      if (sortorder->field->maybe_null())
	length++;				// Place for NULL marker
    }
    else
    {
      switch ((sortorder->result_type=sortorder->item->result_type())) {
      case STRING_RESULT:
	sortorder->length=sortorder->item->max_length;
        set_if_smaller(sortorder->length, thd->variables.max_sort_length);
	if (use_strnxfrm((cs=sortorder->item->collation.collation)))
	{ 
          sortorder->length= cs->coll->strnxfrmlen(cs, sortorder->length);
	  sortorder->need_strxnfrm= 1;
	  *multi_byte_charset= 1;
	}
        else if (cs == &my_charset_bin)
        {
          /* Store length last to be able to sort blob/varbinary */
          sortorder->suffix_length= suffix_length(sortorder->length);
          sortorder->length+= sortorder->suffix_length;
        }
	break;
      case INT_RESULT:
#if SIZEOF_LONG_LONG > 4
	sortorder->length=8;			// Size of intern longlong
#else
	sortorder->length=4;
#endif
	break;
      case DECIMAL_RESULT:
        sortorder->length=
          my_decimal_get_binary_size(sortorder->item->max_length - 
                                     (sortorder->item->decimals ? 1 : 0),
                                     sortorder->item->decimals);
        break;
      case REAL_RESULT:
	sortorder->length=sizeof(double);
	break;
      case ROW_RESULT:
      default: 
	// This case should never be choosen
	DBUG_ASSERT(0);
	break;
      }
      if (sortorder->item->maybe_null)
	length++;				// Place for NULL marker
    }
    set_if_smaller(sortorder->length, thd->variables.max_sort_length);
    length+=sortorder->length;
  }
  sortorder->field= (Field*) 0;			// end marker
  DBUG_PRINT("info",("sort_length: %d",length));
  return length;
}


/*
  Get descriptors of fields appended to sorted fields and
  calculate its total length

  SYNOPSIS
    get_addon_fields()
    thd                 Current thread
    ptabfields          Array of references to the table fields
    sortlength          Total length of sorted fields
    plength    out:     Total length of appended fields

  DESCRIPTION
    The function first finds out what fields are used in the result set.
    Then it calculates the length of the buffer to store the values of
    these fields together with the value of sort values. 
    If the calculated length is not greater than max_length_for_sort_data
    the function allocates memory for an array of descriptors containing
    layouts for the values of the non-sorted fields in the buffer and
    fills them.

  NOTES
    The null bits for the appended values are supposed to be put together
    and stored the buffer just ahead of the value of the first field.

  RETURN
    Pointer to the layout descriptors for the appended fields, if any
    NULL - if we do not store field values with sort data.
*/

static SORT_ADDON_FIELD *
get_addon_fields(THD *thd, Field **ptabfield, uint sortlength, uint *plength)
{
  Field **pfield;
  Field *field;
  SORT_ADDON_FIELD *addonf;
  uint length= 0;
  uint fields= 0;
  uint null_fields= 0;
  MY_BITMAP *read_set= (*ptabfield)->table->read_set;

  /*
    If there is a reference to a field in the query add it
    to the the set of appended fields.
    Note for future refinement:
    This this a too strong condition.
    Actually we need only the fields referred in the
    result set. And for some of them it makes sense to use 
    the values directly from sorted fields.
  */
  *plength= 0;

  for (pfield= ptabfield; (field= *pfield) ; pfield++)
  {
    if (!bitmap_is_set(read_set, field->field_index))
      continue;
    if (field->flags & BLOB_FLAG)
      return 0;
    length+= field->max_packed_col_length(field->pack_length());
    if (field->maybe_null())
      null_fields++;
    fields++;
  } 
  if (!fields)
    return 0;
  length+= (null_fields+7)/8;

  if (length+sortlength > thd->variables.max_length_for_sort_data ||
      !(addonf= (SORT_ADDON_FIELD *) my_malloc(sizeof(SORT_ADDON_FIELD)*
                                               (fields+1), MYF(MY_WME))))
    return 0;

  *plength= length;
  length= (null_fields+7)/8;
  null_fields= 0;
  for (pfield= ptabfield; (field= *pfield) ; pfield++)
  {
    if (!bitmap_is_set(read_set, field->field_index))
      continue;
    addonf->field= field;
    addonf->offset= length;
    if (field->maybe_null())
    {
      addonf->null_offset= null_fields/8;
      addonf->null_bit= 1<<(null_fields & 7);
      null_fields++;
    }
    else
    {
      addonf->null_offset= 0;
      addonf->null_bit= 0;
    }
    addonf->length= field->max_packed_col_length(field->pack_length());
    length+= addonf->length;
    addonf++;
  }
  addonf->field= 0;     // Put end marker
  
  DBUG_PRINT("info",("addon_length: %d",length));
  return (addonf-fields);
}


/*
  Copy (unpack) values appended to sorted fields from a buffer back to 
  their regular positions specified by the Field::ptr pointers.

  SYNOPSIS
    unpack_addon_fields()
    addon_field     Array of descriptors for appended fields
    buff            Buffer which to unpack the value from

  NOTES
    The function is supposed to be used only as a callback function
    when getting field values for the sorted result set.

  RETURN
    void.
*/

static void 
unpack_addon_fields(struct st_sort_addon_field *addon_field, byte *buff)
{
  Field *field;
  SORT_ADDON_FIELD *addonf= addon_field;

  for ( ; (field= addonf->field) ; addonf++)
  {
    if (addonf->null_bit && (addonf->null_bit & buff[addonf->null_offset]))
    {
      field->set_null();
      continue;
    }
    field->set_notnull();
    field->unpack(field->ptr, (char *) buff+addonf->offset);
  }
}

/*
** functions to change a double or float to a sortable string
** The following should work for IEEE
*/

#define DBL_EXP_DIG (sizeof(double)*8-DBL_MANT_DIG)

void change_double_for_sort(double nr,byte *to)
{
  uchar *tmp=(uchar*) to;
  if (nr == 0.0)
  {						/* Change to zero string */
    tmp[0]=(uchar) 128;
    bzero((char*) tmp+1,sizeof(nr)-1);
  }
  else
  {
#ifdef WORDS_BIGENDIAN
    memcpy_fixed(tmp,&nr,sizeof(nr));
#else
    {
      uchar *ptr= (uchar*) &nr;
#if defined(__FLOAT_WORD_ORDER) && (__FLOAT_WORD_ORDER == __BIG_ENDIAN)
      tmp[0]= ptr[3]; tmp[1]=ptr[2]; tmp[2]= ptr[1]; tmp[3]=ptr[0];
      tmp[4]= ptr[7]; tmp[5]=ptr[6]; tmp[6]= ptr[5]; tmp[7]=ptr[4];
#else
      tmp[0]= ptr[7]; tmp[1]=ptr[6]; tmp[2]= ptr[5]; tmp[3]=ptr[4];
      tmp[4]= ptr[3]; tmp[5]=ptr[2]; tmp[6]= ptr[1]; tmp[7]=ptr[0];
#endif
    }
#endif
    if (tmp[0] & 128)				/* Negative */
    {						/* make complement */
      uint i;
      for (i=0 ; i < sizeof(nr); i++)
	tmp[i]=tmp[i] ^ (uchar) 255;
    }
    else
    {					/* Set high and move exponent one up */
      ushort exp_part=(((ushort) tmp[0] << 8) | (ushort) tmp[1] |
		       (ushort) 32768);
      exp_part+= (ushort) 1 << (16-1-DBL_EXP_DIG);
      tmp[0]= (uchar) (exp_part >> 8);
      tmp[1]= (uchar) exp_part;
    }
  }
}