Now that RefPtr is really a std::shared_ptr<>, we should use
std::weak_ref<> instead.

Note that a std::weak_ptr<> tells you nothing about whether
the underlying GObject is still alive, which Glib::RefPtr did.
It just tells you whether our std::shared_ptr<> still holds
a reference to it. That's why I removed one of the checks in
tests/giomm_listmodel/main.cc.

Instead of Glib::RefPtr<>::cast_dynamic().

Specifying a Deleter in make_refptr_for_instance().
And changing RefPtr::cast_dynamic<>() to std::dynamic_pointer_cast<>().

The glibmm_refptr test then becomes rather silly, and should maybe
just be removed.

Rename activate_action_vfunc() to activate_action_full_vfunc() ad
rename change_action_state_vfunc() to change_action_state_full_vfunc(),
matching the underlying C vfuncs.
Now that this class derives from ActionGroup, this was causing warnings
such as this, at least with clang++:

/opt/gnome/include/giomm-2.52/giomm/remoteactiongroup.h:180:18: error: 'Gio::RemoteActionGroup::activate_action_vfunc' hides overloaded virtual
      function [-Werror,-Woverloaded-virtual]
    virtual void activate_action_vfunc(const Glib::ustring& action_name, const Glib::VariantBase& parameter, const Glib::VariantBase& platform_data);
                 ^
/opt/gnome/include/giomm-2.52/giomm/actiongroup.h:574:18: note: hidden overloaded virtual function 'Gio::ActionGroup::activate_action_vfunc' declared
      here: different number of parameters (2 vs 3)
    virtual void activate_action_vfunc(const Glib::ustring& name, const Glib::VariantBase& parameter);

* glib/glibmm/dispatcher.cc: When a Windows HANDLE must be cast, cast it
to Glib::PollFD::fd_t instead of int. Bug 772074

* glib/glibmm/main.[h|cc]: file descriptor to poll can be either of gint,
  or gint64 type, depending on the platform. glibmm should follow this
  rule as well.

https://bugzilla.gnome.org/show_bug.cgi?id=772074

https://bugzilla.gnome.org/show_bug.cgi?id=774903

Bug 775210

* glib/glibmm/main.[cc|h]: Replace the std::map containing
ExtraSourceData with instance data in Source. The map was just a way
of avoiding an ABI break, but now we can break ABI. Bug 561885

Because GRemotActionGroup requires GActionGroup:
https://developer.gnome.org/gio/stable/GRemoteActionGroup.html#GRemoteActionGroup.prerequisites

Both are interfaces. This is like Gio::LoadableIcon, which
derives from Gio::Icon because GLoadableIcon requires GIcon.
FileIcon then derives and implements LoadableIcon, but doesn't
explicitly use _IMPLEMENTS_INTERFACE(Icon).

Gio::DBus::ActionGroup: Don't derive directly from ActionGroup
(instead just derive from, and implement Gio::RemoteActionGroup),
just as FileIcon doesn't derive directly from (or implement) Icon.

We can do this now that we can break ABI.

We can do this now that we can break ABI.

We can do this now that we can break ABI.

Because the requirement was removed from GActionMap in 2013:
https://git.gnome.org/browse/glib/commit/gio/gactionmap.c?id=390115f3851de49f93918cda9266ab857382cc3e

We can fix this now that we can break ABI.

We can fix this now that we can break ABI.

and add connect_impl_(bool notify, const sigc::slot_base& slot, bool after).
SignalProxyNormal then becomes more like SignalProxyDetailedBase.

Rename SignalProxyDetailed to SignalProxyDetailedBase, and
SignalProxyDetailedAnyType to SignalProxyDetailed.
Remove SignalProxyDetailed# aliases (# = 0..6).

Remove two_signal_methods from _WRAP_SIGNAL("changed"). It was added to avoid
breaking ABI.

The _WRAP_METHOD calls had the names typo'd as get_enum() + get_flags().

https://bugzilla.gnome.org/show_bug.cgi?id=774647

and let connect_property_changed() return a sigc::connection.

* gio/src/application.hg: Add dbus_register and dbus_unregister vfuncs.
* gio/src/gio_vfuncs.defs: Add dbus_register and dbus_unregister.
Remove shutdown. It's a signal.
Based on a patch by Ritesh Khadgaray <khadgaray@gmail.com>. Bug 762191

Remove the no_default_handler argument in _WRAP_SIGNAL() where it was added
only to avoid breaking ABI.

When the Thread classes were removed, the dispatcher examples were removed
from the list of check_PROGRAMS. I suppose that was a mistake. The dispatcher
examples were not removed. They still work. They use std::thread.

std::forward_list is ideally suited for storing pointers to the interfaces
of custom types. The list is often empty, never long. No need to use a pointer
to a container in order to save storage space (as I did in the previous
commit). An empty std::forward_list consists of nothing but a pointer.

* glib/glibmm/class.[cc|h]: Remove the clone_custom_type() overload without
an interface_class_vector_type argument.
* glib/glibmm/interface.cc:
* glib/glibmm/object.cc:
* glib/glibmm/objectbase.[cc|h]: Replace the std::map containing
ExtraObjectBaseData with instance data in ObjectBase. The map was just a way
of avoiding an ABI break, but now we can break ABI.
The new data is a std::unique_ptr<Class::interface_class_vector_type> rather
than a Class::interface_class_vector_type. It's a vector which is used only
during a short period during object construction, and only for custom objects.
With a pointer to the vector, it need not be created for the majority of
objects, and if it is created, it can be deleted when it's no longer needed.
* gio/src/application.ccg:
* glib/glibmm/main.cc: Add #include <mutex> that should have been there before,
but now became necessary, when it was removed from objectbase.h.

* glib/glibmm/objectbase.h: add convenient overload of get_property(),
  which gives a property value as a return value, not as an output
  parameter.

https://bugzilla.gnome.org/show_bug.cgi?id=771782

* tests/Makefile.am: Add giomm_stream_vfuncs.
* tests/giomm_stream_vfuncs/main.cc: New file.
An updated version of a patch by Krzysztof Kosiński <tweenk.pl@gmail.com>
Bug #572471

* gio/src/bufferedinputstream.hg: Add fill_vfunc().
* gio/src/inputstream.hg: Add read_vfunc(), skip_vfunc(), close_vfunc().
* gio/src/outputstream.hg: Add write_vfunc(), splice_vfunc(),
flush_vfunc(), close_vfunc().
An updated version of a patch by Krzysztof Kosiński <tweenk.pl@gmail.com>
Bug #572471

    * gio/src/settingsschemasource.hg: method get_default() returns
      transfer-none value, what means that we need to increase refcount
      before wrapping the pointer by RefPtr class.

https://bugzilla.gnome.org/show_bug.cgi?id=774593