[Bug 750761] inputselector: Handle different duration track selection

[GStreamer (GNOME Bugzilla)]

https://bugzilla.gnome.org/show_bug.cgi?id=750761

--- Comment #46 from kevin <kevinbing.song at gmail.com> ---
(In reply to Sebastian Dröge (slomo) from comment #45)
> (In reply to kevin from comment #44)
> > (In reply to Sebastian Dröge (slomo) from comment #43)
> > > Review of attachment 305894 [details] [review] [review] [review]:
> > > 
> > > ::: plugins/elements/gstinputselector.c
> > > @@ +459,3 @@
> > > +static gboolean
> > > +gst_input_selector_eos_wait (GstInputSelector * self, GstSelectorPad * pad)
> > > +{
> > > 
> > > So this basically blocks EOS until the pad becomes the active pad, right?
> > > How does that make things work correctly, and how is it different to before?
> > > 
> > No, eos_wait () will block until flush or state change. So can switch track
> > between EOS track and non-EOS track on the fly. So can switch track:
> > EOS->non-EOS->EOS->non-EOS. Old code can't support this.
> 
> Ah right, but this will break gapless playback. It should also unblock if
> all streams are EOS.
Ok, I will check it.
> 
> 
> > > What would now still happen is that the active EOS is forwarded, and all
> > > other streams are blocked on EOS until they become active. But if the active
> > > stream is EOS, all buffers of the inactive streams are just dropped until
> > > they themselves go EOS. So while we now can switch from an EOS to a non-EOS
> > > stream, we will most likely still have dropped a lot of data instead of
> > > waiting at the current time.
> > > 
> > All inactive track will blocked when active reach EOS. All inactive track is
> > waiting active track advancing, but active no any data.
> 
> Where are they waiting in the code? I didn't see any waiting in the chain
> function related to this
gst_input_selector_wait_running_time () will be blocked.
> 
> > > @@ +473,3 @@
> > > +      }
> > > +
> > > +      gst_pad_push_event (self->srcpad, gst_event_new_eos ());
> > > 
> > > You should ideally forward exactly the EOS event that arrived here instead
> > > of a new one :)
> > > 
> > No, eos_wait () may send EOS event many times. it will send eos event every
> > time when switch track to EOS track.
> 
> You can send the same EOS event multiple times by getting a new reference
> before gst_pad_push_event()
Ok, will update later.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.