Sub-pixel positioning

[kde at bartcoppens.be]

Quoting Sven Langkamp <longamp at reallygood.de>: 
> I think the problem would be that every conversion would cause a loss of  
> quality. If the error of the conversion algorithm is very small this 
wouldn't  
> be a problem. The error also gets bigger if there are many layers. 
> Did anybody try out how big the error is? 
>  
So if I get this right, does this mean that compositing of multiple layers 
will happen _only_ in the RGBA colour space? This could cause serious problems 
in the future. Imagine someone compositing 10 layers of CMYK(A) with 32bits 
for each channel. If combined in 8bit per channel RGBA, I can imagine that the 
resulting composition would be of questionable quality. 
 
On the other hand, if this means that layers are only rendered to RGBA at 
runtime, while compositing the layers in their native colourspace for 
exporting the image, this would be no real problem. It would still require 
investigating and implementing the various compositing functions for those 
colourspaces. 
 
Bart Coppens