[vlc-devel] [RFC PATCH 0/9] audio decoder fallback: the packetizer way

[Thomas Guillem]

On Tue, Jul 12, 2016, at 12:09, Rémi Denis-Courmont wrote:
> Le 2016-07-11 18:02, Thomas Guillem a écrit :
> > On Sun, Jul 10, 2016, at 15:54, Rémi Denis-Courmont wrote:
> >> On lauantaina 9. heinäkuuta 2016 12.20.08 EEST Thomas Guillem wrote:
> >> > This is an other proposal to handle audio decoder fallback, the 
> >> goal is to
> >> > have passthrough decoder that can fallback to PCM decoders and the 
> >> other
> >> > way.
> >>
> >> AFAIK, we already have decoder fallback for audio. We don´t have
> > Do we ? Where ?
> >
> >> glitch-free
> >> fallback because it´s essentially impossible.
> >>
> >> And I don´t see why S/PDIF payloaders and PCM decoders could not use 
> >> the
> >> same
> >> packetizer. In fact, I don´t see why they would use different 
> >> packetizers
> >> at
> >
> > S/PDIF and PCM can use the same packetizer.
> > My main problem is that some demuxers say that the format is 
> > packetized
> > when it's not really. In that case, there is no packetizer at all. I
> > don't really what to do in that case.
> 
> I am not buying any of that.
> 
> Each codecs has certain semantics on what the ES format parameters mean 
> (notably extra data), how the unpacketized and/or packetized blocks are 
> sequenced, what the block timestamps and flags mean. This applies to 
> demuxers, muxers, decoders, encoders and packetizers. If there is a 
> disagreement, it's clearly a bug in one or both plugins.

I think some demuxers shouldn't indicate that their format is
packetized.

Note that fmt->b_packetized is set to true by default from
es_format_Init(). A lot of demuxers don't override it and may assume
that it's false by default.

> 
> > The decoder_RequestPacketizer() function doesn't really need the
> > packetizer_name argument. This function could just force the usage of
> > any packetizer (that matches the format).
> > One other solution could be to create an extra packetizer even if the
> > format from the demuxer is packetized.
> 
> That makes no sense to me.
> 
> -- 
> Rémi Denis-Courmont
> http://www.remlab.net/
> _______________________________________________
> vlc-devel mailing list
> To unsubscribe or modify your subscription options:
> https://mailman.videolan.org/listinfo/vlc-devel