Comment 35 for bug 867424

Created attachment 8737263
Stop using libgnome and libgnomeui on Linux v3

Updated this patch again, and tested on master and on top of 45.0.1. It works
with both gtk2 and gtk3 builds but I'd still like feedback about whether I'm
doing the right thing with all the gtk plumbing.

Solved Issues
-------------

"Doesn't solve bug #557601 - DisconnectFromSM is called at the right time when
quitting firefox, but is too early on shutdown. I don't understand why yet."

This was just PEBKAC. Turns out that if you run your test firefox build from a
terminal it will reliably be killed when the terminal quits - who knew? Running
with nohup or from e.g. krunner works and firefox survives until it's finished
cleaning up. Firefox started by session restore also behaves as expected.

Changes from v2
---------------

* Rebase to master
  - Unbitrot to match "Bug 1207245 - part 6 - rename nsRefPtr<T> to RefPtr<T>"
  - Unbitrot: add missing nsThreadUtils.h include
* Add keyword for SetClientState logging
* Add symbol to mozgtk so that gtk3 builds link
* Written a new commit message

Questions
---------

* Is the #define gtk_function gtk_function_, #undef gtk_function dance in the
  gtk compat defines the right approach?
* Is the mozgtk stub symbol definition sufficient for gtk3 builds?
* are gio watch callbacks always guaranteed to be on the main thread?
* what should I use to log instead of printf? What log facilities are still
 available in late shutdown?

Remaining issues
----------------

* The interaction call/response with the session manager may need some work
  - in fact it needs to be tested at all.
* Doesn't set a couple of SM properties that gnome-client did
* Doesn't use grabs to prevent interaction at the wrong time like gnome-client
  did. But the libegg SM implementation doesn't do anything to prevent
  interaction, as far as I can tell. So I'm not sure what the correct behaviour
  is.