There's very few chances that the universal-access will fit in the
minimum height of the shell for panels, so just nuke that.
This also fixes the panel taking 1px in height until we switch away from
it and back to it.
https://bugzilla.gnome.org/show_bug.cgi?id=761939
When using the small mode, and the scroll window's content
height is smaller than the screen, we'd end up with whitespace at the
bottom of the panel, as the panel's height is smaller than the window.
There's really no point showing a resolution that's smaller than the
fixed width of the shell, and isn't the preferred resolution for the
screen either. This stops the problematic 720x576 resolution showing on
an HD screen.
If the GCancellable is cancelled by the finalize() function, the
callback will still be called potentially with an instance of the
panel that's not valid anymore.
To avoid crashing in that case, only access the priv pointer
when we have no error.
https://bugzilla.gnome.org/show_bug.cgi?id=761846
It doesn't make sense to try to validate pages because widgets are
changing due to the whole dialog being destroyed and it causes a bunch
of warnings because some resources are cleared on each page dispose
method.
Avoid all that by disconnecting the page "changed" signal handler
before we start destroying widgets.
We don't need secrets for new connections and, in fact, trying to
retrieve secrets in that case will fail because we have a plain
NMConnection instead of a NMRemoteConnection.
The above mentioned error would result in the page never being
initialized.
The way we destroy the window during the handling of a clutter event may
cause the window to generate a focus event that gets processed immediately,
causing Clutter to deadlock on its global lock.
Bypass these issues in the places where we finalize out of Clutter events,
The Esc key presses are now handled at the GTK+ level, and we postpone
finalization (with either success or failure) to an idle.
Works around https://bugzilla.gnome.org/show_bug.cgi?id=747880
