[MPlayer-dev-eng] MPCF proposal

D Richard Felker III dalias at aerifal.cx
Thu Mar 13 17:49:31 CET 2003


On Thu, Mar 13, 2003 at 04:42:57PM +0100, Michael Niedermayer wrote:
> Hi
> 
> On Thursday 13 March 2003 00:16, D Richard Felker III wrote:
> > On Tue, Mar 11, 2003 at 01:50:11AM -0500, D Richard Felker III wrote:
> > > Somehow the MPCF discussion seems to have died off... So here's what I
> > > propose:
> > >
> > > 1) Anyone who has remaining complaints or issues that haven't yet been
> > > resolved with MPCF should post them now, and they should be discussed.
> >
> > One other issue I'd like to raise, if it's not already clear... As I
> > understand it, the index only stores keyframes. For video-only files
> > this is clear, and I assume for audio-only files, every packet is a
> > 'keyframe'. But what's the plan for audio+video muxed? Should the
> > index only point to audio packets that begin along with video
> > keyframes?
> hmm, its possible to have multiple video streams, so this doesnt seem like a 
> good idea, i allso would prefer to keep streams independant if possible, so 
> reencoding of the video stream is possible without reindexing the audio

IMHO multiple video streams is brain damaged, but I still agree that
it should be supported. Also this stupid hack I proposed would
complicate the muxer a great deal.

> > Or maybe audio shouldn't be indexed at all, and we should
> > just rely on the requirement that files be interleaved in order, so
> > that the player just starts with the first audio packet after the
> > video keyframe when seeking...
> hmm, sounds like this could cause problems perhaps, dunno

I don't see how it could be a problem as long as you have timestamps
for both.

> anyway indexes are small (~25kb per hour if we have 2 audio packets per sec) 
> so it doesnt seem worth to do weird tricks here

Ah, I forgot how much subpackets reduced the size impact on the index.
That's a good argument for keeping subpackets, too. So I agree, keep
index for all audio packets.

> btw, completly strict interleaving doesnt seem like a good idea either if we 
> allow multiple video streams, the performance loss would be too large IMHO 
> and even for many audio steams it is very likely bad
> so i suggest that interleaving is a MUST only for videos with 1 video stream & 
> <=2 audio streams
> any comments?

Interleaving should always be required. Otherwise playback from CDROM
is impossible. Even if it works it will destroy your drive in no time
seeking back and forth thousands of times... Besides, even a 1x drive
is a plenty fast to play most files single-video, single-audio files,
so I don't think io performance is an issue.

Rich



More information about the MPlayer-dev-eng mailing list