gistfile1.txt

After a grab (either a clutter grab or an X grab involving windows overlapping
the stage), clutter can be confused about what actor contains the pointer.

For https://bugzilla.gnome.org/show_bug.cgi?id=630842, when we suspect that
clutter might be out of sync, we recover by posting a fake CLUTTER_MOTION
event, which as a side effect will result in clutter eventually calling
_clutter_input_device_update(). But it would be nicer to have a guaranteed way
of doing this.

At first I thought we could just make _clutter_device_manager_update_devices()
public, but that's not quite enough, because that will used cached values for,
eg, the device coordinates, rather than looking up the current ones. (However,
note that nothing is currently using _clutter_device_manager_update_devices()
anyway, so its semantics could be changed to include this step as well.)