When no feedbacks are available (either because the requested event does
not exist in the theme or because the system does not have required
capabilities, e.g. no vibration motor or LED) feedbackd will emit a
"feedback-ended" signal (with reason LFB_EVENT_STATE_REASON_NOT_FOUND)
on DBus _before_ completing the event triggering DBus call.
Since the DBus call was not completed yet, LfbEvent will not have
learned it's ID yet and therefore not emit it's own "feedback-ended"
GObject signal.
In this case the test will timeout since the main loop is never quit.
The ID is inserted into a hashtable below with GUINT_TO_POINTER
and is set using self->next_id++ with next_id also being a guint,
so change the type accordingly.
lfb_event_trigger_feedback_async() internally uses a GTask to handle
invocation of the user provided callback when g_task_return_*() is
called.
Currently GTask is completed first and only afterwards is the state of
the event being updated which can lead to suprising behaviour in code
like this:
static void
on_event_triggered (LfbEvent *event,
GAsyncResult *res,
CallsRinger *self)
{
g_autoptr (GError) err = NULL;
gboolean ok;
g_return_if_fail (LFB_IS_EVENT (event));
ok = lfb_event_trigger_feedback_finish (event, res, &err);
if (ok) {
LfbEventState state = lfb_event_get_state (event);
/* do something with state */
}
}
In this case `state` would return a stale value that's about to get
changed. Instead it would be better to *first* change the state and only
afterwards complete the GTask.
This also makes the sync and async variants consistent in terms of when
they change the event state.
Closes: #59
The old wording could be interpreted as the feedback being discarded if
it's noisier than the currently selected profile which is not the case.
Also take the new per application settings into account.
Remove the feedback and the associated callback data before invoking the
callback (which then triggers FbdFeebackBased:ended) since we otherwise
remove the data from the currently playing sound from the hash table
instead of removing the data from the sound that just finished playing.
Call flow for a repeating sound before:
on_sound_play_finished_callback()
data->callback()
fbd_feedback_get_ended()
on_fb_ended()
fbd_feedback_run()
g_hash_table_insert(..., feedback, data)
fbd_dev_sound_play()
gsound_context_play_full() [async]
g_hash_table_remove (..., data->feedback) <- removes wrong feedback
So (since the feedback is still the same but only the callback data
changes) we end up removing the data from the currently playing sound
from the hash table and any calls to fbd_dev_sound_stop() would fail
to find data and hence not cancel the playing sound.
Call flow for a repeating sound now:
on_sound_play_finished_callback()
g_hash_table_remove (..., data->feedback)
data->callback()
fbd_feedback_get_ended()
on_fb_ended()
fbd_feedback_run()
g_hash_table_insert(..., feedback, data)
fbd_dev_sound_play()
gsound_context_play_full() [async]
So we remove the data of the currently playing sound before adding
the feedback to the hash table again. No ownership is transfered to
the hash table so no cleanusp need to be changed.
Closes: https://source.puri.sm/Librem5/feedbackd/-/issues/46
Chris Vogel
Guido Gunther
Guido Günther
Evangelos Ribeiro Tzaras
Luca Weiss
Dylan Van Assche
Pablo Correa Gómez