[AGG] Re: SpanGenerators

[Maxim S. <mc...@an...>]

> I am experimenting with 2.4 release and modified SpanGenerator design. The
> new functions are a bit easier to use and the concept of a 'renderer' now
> seems to be redundant.

Interesting! What are the changes?

> I encountered a problem with 'sbool_combine_shapes_aa' function which 
> still
> accepts the 'Renderer' as the last parameter:
>
> agg::sbool_combine_shapes_aa(agg::sbool_and, sh, ras, sl1, sl, result, 
> ren);
>
> Shouldn't the last parameter be 'render_base' in the new design?

First of all, I'd like to keep the renderer concept. It's convenient when 
you use a function template like that:

template<class Ren> void render(Ren& ren)
{
   . . .
}
Here "ren" can be solid, gradient, etc. That is, I'd like to preserve 
compile time polymorphism.

This is why "sbool_combine_shapes_aa" requires it. It can't take 
renderer_base, because it doesn't have to draw anything. It can be a just a 
next step in boolean operations. You can use "scanline_storage_aa" as "ren" 
to accumulate the new set of scanlines. Maybe it's just not the best name, 
but a scanline renderer is an abstract that takes a scanline and does what 
it wants. I used the renderer interface for the scanline boolean because it 
perfectly fitted all needs. So, it can draw directly, it also can accumulate 
scanlines in a separate storage for future use.
So that, I'd like to keep the concept of the scanline renderer as a class 
(along with functions "render_*").

McSeem