backend: Fix Wayland backend idleCallbacks destruction

[PlasmaPower]

The old code was crashing for me on shutdown due to the fact that CWaylandBackend's reverse field order destruction destroys idleCallbacks before destroying outputs, meaning that when ~CWaylandOutput calls backend->idleCallbacks.clear(); it's clearing an already destroyed vector which crashes.

To fix this I just removed the FIXME hack and properly handled the potential sendFrameAndSetCallback UAF by using the weak pointer to check if the output was already destroyed or not.

[vaxerski]

why not just if (backend) then?

[PlasmaPower]

You mean checking that when idleCallbacks.clear() is called? I thought I might as well fix the FIXME while I was at it, though yeah checking !backend.expired() works (the expired check is necessary because this'll be called by the destructor of the backend so backend is still truthy). If you mean checking if (backend) in the callback, the concern there is that the output might be destroyed before the backend is I believe, and then the idle callback would be referencing an invalid output when called.

[vaxerski]

ah, I re-read it and noticed that github folded the diff weirdly. Now that makes sense, yes.

[vaxerski]

thanks!