[cairo] Operator optimization based on source or dest opacity

[Soeren Sandmann]

Antoine Azar <cairo at antoineazar.com> writes:

> >- pixman_image_is_opaque() needs to handle alpha-maps.
> 
> I'm not sure I understand what you mean. How do you check if an alpha 
> map is opaque without checking every pixel of it?

An alpha map is something you can set on an image using
pixman_image_set_alpha_map(). It is a way of using an external alpha
channel for the image.

Nobody uses them, but they are exposed through RENDER in the X server,
so we shouldn't break them. It's totally fine to just say

        if (image->common.alpha_map)
                return FALSE;

or something.

> >- In general, it looks to me like the code in pixman-pict.c that 
> >checks whether the optimization is applicable, is really checking 
> >whether the image is opaque. If so, it should be moved into 
> >pixman_image_is_opaque().
> 
> Yes and no. I originally didn't include the other checks in the 
> function because I wanted to keep it generic enough for other uses. 
> It basically just checks if a source contains alpha information or 
> not which I guess could be useful at other places. The other checks 
> take into account *how* the source is being used and if this 
> introduces any transparency. So I can see a valid point either way, 
> let me know if you would still rather have the checks included in 
> pixman_image_is_opaque.

I see the transformations and filters as integral to the images, not
as instructions on how to composite them. So a pixman image with a
transformation set is not "a normal image that will be used in a
transformed way", but rather "a set of pixels that happen to be the
result of a transformation applied to a different set of pixels".

Similary, when repeat is set, the image really is an infinite set of
pixels, not a finite set that happens to be repeated over and over.

Less philosophically, I don't see any way the information that an
image is opaque could ever be useful without also checking that the
transformation and filter don't reintroduce translucency.


Soren