by making sure the call-added signal is emitted early enough
so that all consumers (display, ringer, etc) have a chance of getting
notified when the call state changes from f.e. DIALING to DISCONNECTED
similar to how its done in 03d960ccaf
for the dummy provider.
* removed nua_callstate_authenticating as it was never hit
* move debug statement further up, because we might not reach it if no
corresponding call is found
* handle 401 and 407 the same way
note: we should record which realm we're authenticating against during
REGISTER so we can prompt the user for additional credentials when
challenged for a different realm - still happens when calling from
a sip.linphone.org account to a jmp.bwapp.bwsip.io account.
IPv6 should work, but sofia's outbound engine keep printing
errors involving the outbound engine. Working theory:
Failing ICMPv6 (pings) can make sofia think we don't have connectivity.
Note that we also don't get any answers from the SIP servers we tried so far.
And adapt to changes
It's easier to track changes if we are using GListModel than
a GList. Also, eventually we shall have multiple providers,
and we can use GtkFlattenListModel to merge multiple origins
without manually managing pointers ourself.
Since several widgets has APIs to handle GListModels, it will
also help us create widgets from GListModels easily.
The 'property-changed' and 'disconnect-reason' signals weren't disconnected on the
voice_call object cleanup, and callbacks would thus still be fired once a second
call happens, resulting in a crash.
Signed-off-by: Eugenio Paolantonio (g7) <me@medesimo.eu>
This fixes this two issues:
- Emit the events origin-removed, origin-added when required.
- Remove call immediately when disconnected and emit events:
By using `g_idle_add ()` we can't be sure when the call is actually
removed and could stay potentionaly for evere, as long the main loop is
busy. This can lead to unexpected behavior. If the calls shouldn't be
freed then it's required to increase the ref count. This is similar how
the `remove` event works in a `GtkContainer`.
This is part of a larger refactoring effort, with the goal to replace
all the get_*() member functions in provider abstraction interfaces with
GObject properties. See also: https://source.puri.sm/Librem5/calls/issues/6
How `g_type_class_peek ()` was used it didn't return the correct parent
class in most cases.
G_DEFINE_TYPE macro creates a pointer we can use to get the parent
class `n_p_parent_class`.
Because we didn't use the correct parent class the object initialisation
wasn't fully completed for some GtkWidgets.
See https://developer.gnome.org/gobject/stable/chapter-gobject.html#gobject-instantiation
for more information.
This commit makes use of the `n_p_parent_class pointer` created for this
specific use case where ever possible.
Fixes: https://source.puri.sm/Librem5/calls/issues/118
