(Message /usr/users/bradley/Mail/inbox:36)
Return-Path: drtr@ras.phy.cam.ac.uk
Received-Date: Fri, 7 Aug 92 14:06:17 EDT
Received: from HALIBUT.CIS.UPENN.EDU by noc2.dccs.upenn.edu
	id AA12422; Fri, 7 Aug 92 14:06:16 -0400
Return-Path: <drtr@ras.phy.cam.ac.uk>
Received: from ras.phy.cam.ac.uk by halibut.cis.upenn.edu
	id AA00260; Fri, 7 Aug 92 14:06:12 EDT
Posted-Date: Fri, 7 Aug 92 19:04 BST
Received: by ras.phy.cam.ac.uk (UK-Smail 3.1.25.1/3)
  id <m0mGYg8-0002UTC@ras.phy.cam.ac.uk>; Fri, 7 Aug 92 19:04 BST
Message-Id: <m0mGYg8-0002UTC@ras.phy.cam.ac.uk>
Date: Fri, 7 Aug 92 19:04 BST
From: drtr@ras.phy.cam.ac.uk (David Robinson)
To: bradley@halibut.cis.upenn.edu
Subject: xv fits patches
Cc: drtr1@ras.phy.cam.ac.uk

A while ago, (in June) I sent you some patches to xv-2.21 to allow it to read
and write FITS files. Have you incorporated these into the standard xv release?

I've just found a bug in some of the code I sent you. Specifically, if the
FITS file is a BITPIX=32 (=signed 4 byte int) file, and 
max-min > 2^31, then the program will give a garbage image. Here is a patch
to the file xvfits.c, which fixes the assumption that the difference between
two ints can be held in an int.

*** xvfits.c.orig	Fri Aug  7 16:50:10 1992
--- xvfits.c	Fri Aug  7 17:06:16 1992
***************
*** 567,581 ****
      {
  	register int *buffer=voidbuff;
  	register int max, min, maxmin_t;
! 	register float scale;
  
  	min = max = buffer[0];
  	for (i=1; i < n; i++, buffer++) maxmin(*buffer, max, min);
! 	scale = (max == min) ? 0. : 255./(float)(max-min);
  
  /* rescale and convert */
! 	for (i=0, buffer=voidbuff; i < n; i++)
! 	    cbuff[i] = (unsigned char)(scale*(float)(buffer[i]-min));
  
  /* convert float to unsigned char */
      } else if (fs->bitpix == -32)
--- 567,586 ----
      {
  	register int *buffer=voidbuff;
  	register int max, min, maxmin_t;
! 	register float scale, fmin;
  
  	min = max = buffer[0];
  	for (i=1; i < n; i++, buffer++) maxmin(*buffer, max, min);
! 	scale = (max == min) ? 1. : 255./((double)max-(double)min);
! 	fmin = (float)min;
  
  /* rescale and convert */
! 	if (scale < 255./2.1e9) /* is max-min too big for an int ? */
! 	    for (i=0, buffer=voidbuff; i < n; i++)
! 		cbuff[i] = (unsigned char)(scale*((float)buffer[i]-fmin));
! 	else /* use integer subtraction */
! 	    for (i=0, buffer=voidbuff; i < n; i++)
! 		cbuff[i] = (unsigned char)(scale*(float)(buffer[i]-min));
  
  /* convert float to unsigned char */
      } else if (fs->bitpix == -32)

--------------------------------- end of patch ----------------------------

David Robinson. (drtr@ras.phy.cam.ac.uk).