Code review comment for lp:~vanvugt/mir/unocclude

> This pushes me to think we have to shift occlusion detection to SurfaceStack,
> as it seems the common place this could be done. The compositors sure have to
> have a role in submitting what sort of occlusion they should do, but I don't
> want the compositors slowed down by having to send IPC events.
>
> Now of course, this is tricky, because we could have multiple compositors, but
> I think can be done as the SurfaceStack gets a bit smarter about what
> compositors are viewing the SurfaceStack.

+1
scene just needs to gain knowledge about the existing outputs. Their position within
the scene - the camera parameters. Those information could also be used to limit compositor
thread wakeups. Additionally that information is necessary to properly dispatch user input
within 3D shells.

We can solve the issue with decoration increasing the size of surfaces in a different way.
Treat decorations and window shadows as separate scene elements with a separate position
and size, that the shell adds to the scene..
Or maybe simpler: Extend the scenes occlusion logic with a filter-filter that filters out
false positives.
  as a graphical effect that extend the size of surfaces, which