Created attachment 34770 [details] [review]
Patch against TIP of GIMP 2.0.x branch

I've a feeling that this might apply pretty much directly to GIMP 2.2-pre. 
Maybe someone could try.  If not, I can try to forward-port it myself.

This would do as a changelog entry:
 Use a slower but much nicer technique for finding the
 two best colours to dither between when using fixed/positional
 dither methods.  Makes positional dither much less lame.

The patch applies cleanly to CVS HEAD with some minor offsets.

2004-12-12  Sven Neumann  <sven@gimp.org>

	Applied patch from Adam D. Moss (bug #161113):
	
	* app/core/gimpimage-convert.c: Use a slower but much nicer
	technique for finding the two best colours to dither between when
	using fixed/positional dither methods.  Makes positional dither
	much less lame.

Actually, mind backing it out again?  I mean I think it's okay, but after
hacking on the code a bit more I'm seeing some dither artifacts that I don't
trust, and I'm not sure right now whether it's something that was in that patch.
 Looking into it.

No, stupid me, the patch is fine.  I was wondering where it was getting magic
new colours from when dithering to a palette, and didn't realize that the
palette preview doesn't show all the palette colours.

Not that the new dither couldn't use some more tweaking and de-artifacting, but
it's still a big improvement I think.

Thanks, I'll post any further tweaks to a new bug.