These changes prevent warnings when unreferencing the default input
source, locale and back rows which are never instanced by some locales
(and thus null).
The language input chooser is being leaked. The reason for the leak
is a "circular reference": the input chooser keeps a GtkBuilder
instance that gets destroyed when the chooser is, itself, destroyed.
However, the input chooser is never destroyed because the GtkBuilder
instance keeps a reference to it.
These changes dismiss the GtkBuilder once it is no longer needed in the
input chooser which fixes the leak.
When switching from a particular IPv4 or IPv6 "addresses" method to
another, make sure that "address", "DNS" and "routes" are applied or
ignored depending on that method's capabilities.
For example, when switching from manual to automatic IPv4 "addresses"
method, we need to make sure that the old IP address, gateway, etc.
aren't actually used in the new setting.
https://bugzilla.gnome.org/show_bug.cgi?id=734337
The first input source gets activated by default when users log in but
we don't currently allow users to set this (other than the awkward
remove all input sources and add them again in the right order).
Allowing user re-ordering of the input source list nicely allows users
to get the default input source they want.
https://bugzilla.gnome.org/show_bug.cgi?id=743400
Currently it looks a bit unpolished that the add/remove buttons on the
other side of the toolbar look linked while this pair doesn't. Note
that the input source settings button isn't always visible but in that
case the layout button will still show up correctly as non-linked.
https://bugzilla.gnome.org/show_bug.cgi?id=743400
Otherwise we could fail to validate the new configuration depending on
which output we're initializing mirror mode from.
gnome_rr_config_applicable() creates a temporary array of outputs for
validation and, in clone mode, all of them are assigned the same
geometry but, since it lacks context about which output the user is
currently on, it just uses the first one's geometry for all of
them. If this first output isn't the one the user is changing in the
UI then we never called _set_geometry() on it with a clone mode
geometry and thus the validation would fail because the geometry used
is that output's current mode which might not match the clone mode's.
https://bugzilla.gnome.org/show_bug.cgi?id=743816
