KisIterator, quantum and pixels

[Adrian Page]

On Monday 05 July 2004 12:51 pm, Boudewijn Rempt wrote:
> > Yes. This is an implementation detail that should be hidden by the colour
> > strategy. If all interpretation of QUANTUMS is done by the strategies, it
> > makes it easier to implement things like
> > pre-multiplied/non-pre-multiplied alpha, since that is just a different
> > representation of the same pixel data.
>
> Maybe we should do an investigation of all those places and try to fix them
> soon; it might lead to some additions to the colour strategy api's.

Yes.

I think it might be an idea to make the compositing functions part of the 
strategy since they can be made most efficient by accessing the 
representation directly. We could perhaps have functions in the base class 
that can work using some abstract interface to work with channels and alpha, 
and override them to provide optimised versions where needed.

Adrian