The following small example program draws a white image into a 3 byte BGR
image. The graphics used to draw has a Tx with a scale that is very close
to identity but is non-integral
As of the fix for 4916948 this goes through a different code path where
there's apparently rounding errors in the evaluation of the colour
components for the output pixels. the pixels in the source image are
all white but in the destination image they are all(?) 0xfdfefe

The problme is seen on windows (XP) and solaris 9 sparc.
One consequence of this is in printing code where such errors can occur
with fp arithmetic scaling between user, image and printer spaces.
The resulting printout is both not pure white and doesn't compress
using postscript run-length encoding. In one case this caused a spool
file to grow from 300K to 9.5Mb - and take much longer to generate and
print.

import java.awt.*;
import java.awt.image.*;

public class ImgCol {

    public static void main(String args[]) {
	
	int sz = 10;
	BufferedImage bi =
	    new BufferedImage(sz,sz, BufferedImage.TYPE_BYTE_GRAY);
	Graphics g = bi.getGraphics();
	g.setColor(Color.white);
	g.fillRect(0,0,sz,sz);
	System.out.println("pix before="+
			   Integer.toHexString(0xffffff&bi.getRGB(0,0)));
	BufferedImage bgrImg =
	    new BufferedImage(sz,sz, BufferedImage.TYPE_3BYTE_BGR);
	
	Graphics2D g2 = (Graphics2D)bgrImg.getGraphics();
	//g2.translate(0.0001,0.0001);
	g2.scale(1.0001,1.0001);
	g2.drawImage(bi, 0,0,sz,sz,0,0,sz,sz,Color.white,null);
	System.out.println("pix after="+
			   Integer.toHexString(0xffffff&bgrImg.getRGB(0,0)));
    }
}