Comment 11 for bug 1418081

> > "Allowing the compositor to touch the scene with occluded() as it does is a bit messy."
>
> Totally agree.
>
> At some level the compositor "knows" which surfaces went into compositing. It ought to be able to determine
> whether these have unrendered surfaces without a mutating "touch".

This will happen when we get a smarter and richer scene(graph) that can represent whatever a shell may want display (in terms of client surfaces; think effects like expose, live previews etc) and which can give the DB compositors a final list of renderables to render.

That being said, the current mechanism (which queries on available surface buffers) doesn't fail because the compositors can "touch" the scene, but because the compositors are allowed to not render the surfaces we give them. Even if we had a better scene, we would need to guarantee that compositors actually use the surfaces we give them for the current mechanism to work.