If the user selects and deletes the current paragraph, the locusOfFocus
goes defunct and we stop presenting caret-moved events.

https://bugzilla.gnome.org/show_bug.cgi?id=745845

Gecko seems to like to kill certain accessible objects on us upon focus,
putting a new one in its place. This would completely mess up Orca's caret
navigation, so we added checks for these zombies and code to try to find
their replicants. That way, when the proverbial rug was snatched out from
under our feet, we could reposition ourselves on the replacement object,
and everything seemed fine. Now, however, the act of setting the caret in
the replicant seems to be sufficient to cause Gecko to kill that replicant.
This is why we can't nice things.

The change in this commit is to quietly update our locus of focus without
touching the zombie and its subsequent replicants in the hopes they'll
leave us alone so that we can actually navigate within the content.

Samuel Thibault authored Feb 17, 2015 and Joanmarie Diggs committed Feb 17, 2015

https://bugzilla.gnome.org/show_bug.cgi?id=744145

Attila Hammer authored Feb 17, 2015 and Joanmarie Diggs committed Feb 17, 2015

https://bugzilla.gnome.org/show_bug.cgi?id=738057

We shouldn't even be getting these events in the first place, but....

Orca used to automatically present the selected item in a newly-shown
switcher. It doesn't in GNOME 3.14. This is due to a couple of issues:

* The ordering of accessible events has changed, so first we get the
  accessible event for the selected item. Then we get the one for the
  window activation. As a result, Orca ignores the selected item because
  it doesn't yet know the user is in a GNOME Shell window. (bgo738701)

* The subsequent focused event for the containing panel, which does come
  after the window activation, should have be enough to cause Orca to
  automatically present the selected item even with the event ordering
  bug above. But this fails because Orca expects containers which contain
  selectable children to implement the accessible selection interface.
  GNOME Shell's switchers don't. (bgo738705)

This commit hacks around the second issue and is sufficient to cause
Orca to present the initially-selected item in a newly-shown switcher.

Aside from the fact that they're arguably not "real" tables, the
accessible hierarchy may be broken.

findReplicant() is a workaround for Gecko's killing the focused accessible
object in response to less-than-ideal authoring practices. If the root is
a table with a broken hierarchy, pyatspi's findDescendant() can hang. :(