[Bug 794909] ulpfecdec: output perfect seqnums

[GStreamer (GNOME Bugzilla)]

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

--- Comment #6 from Mikhail Fludkov <fludkov.me at gmail.com> ---
@(In reply to Mathieu Duponchelle from comment #4)
>I think it should either drop these lost events entirely
In that case, decoders will have no idea that they are dealing with a corrupt
stream. It is a too harsh approach. VP8 decoder is particularly dumb, not able
to detect that the stream is corrupted. 

> I do not think I agree: it should not be the job of the depayloader / decoder to sort out the good from the bad DISCONT flags, most of them are simply not equipped to do so.
You are right here. I guess the main reason why I said it brings more harm than
good that is because its hard for me to foresee potential issues with all the
elements based on basedepayload & video/audiodecoders. Hiding all gaps in
sequence numbers will backfire in some scenarios. 

Then let's modify the patch to reflect lost events we receive in new sequence
numbers and vice versa - rewrite sequence number in lost events. For example:
We receive packets: #0 #2 #3 lostevent#4 #5
We push out: #100 #101 #102 lostevent#103 #104
By doing that we hid that there was dropped FEC packet #1, but we didn't hide
gap in seqnums because of real packet loss. We still have lost events with new
sequence numbers. As you have said, FEC encoder is rewriting sequence numbers.
In that case, FEC decoder will behave symmetrically to the encoder. Which is
nice.

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