[Bug 794909] ulpfecdec: output perfect seqnums

[GStreamer (GNOME Bugzilla)]

https://bugzilla.gnome.org/show_bug.cgi?id=794909

--- Comment #8 from Mathieu Duponchelle <mduponchelle1 at gmail.com> ---
I'll sum up the discussion Mikhail and I had:

We both agree that:

1) Seqnums should be rewritten, because rtpbasedepayload can't know whether
there is a jitterbuffer upstream which will take care of setting the DISCONT
flags

2) There needs to be some indication that *something* was lost sent downstream,
I initially assumed the vp8 depayloader for example would be capable of
determining this without relying on this information at all, thanks to the
picture-id and the M and S bits, but it isn't enough for the case where a
packet goes missing between these two markers.

The remaining bit on which we disagree is whether there should be gaps at all
in the output seqnums:

I think that there shouldn't be, because ulpfecdec has to cope with
rtpbasepayload's behaviour, and adding a property there to disable it would
just be messy and annoying to discover

Mikhail argues that it feels weird, I agree with that, but think we don't
really have a choice if we want things to work out of the box.

I propose the following:

Keep the patch as is, and simply remove the seqnum from the PacketLost events
we propagate. If need be, we can update downstream elements that would expect a
seqnum there, but for example the default implementation of handle_packet_loss
in rtpbasedepayload does not use that field at all.

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