[MPlayer-cvslog] CVS: main/DOCS/tech mpcf.txt,1.67,1.68

D Richard Felker III dalias at aerifal.cx
Fri Mar 4 21:39:59 CET 2005


On Fri, Mar 04, 2005 at 08:26:20PM +0100, Michael Niedermayer wrote:
> so with tar.gz or similar general propose containers theres little use for 
> recovering from errors, as the user notices it immedeately when he tries to 
> do anything with the file but for a multimedia container format its likely 
> that the user will not know about the damage before he plays the file which 
> could be long after he downloaded it

IMO it's not quite that bad, because the user should check the md5sum,
or if using p2p the p2p network should automatically check hashes..
But still I agree it's bad.
Actually I think more corruption causes from windows systems crashing
with buggy filesystem drivers, not bad downloads..
Scratched CDs are a big cause too.

> so we should IMHO
> * require players to warn the user about detected damages (players not doing 
> this would not be compliant to the nut spec)

"Warning the user" is fairly vague; hard to determine if the player is
compliant or not. What if the player is an embedded device with no
message display? Or what if the user specifies -quiet/-silent? Do
these violate the spec? :)

> * provide a small and easy to use tool (under bsd license) to check if a nut 
> file is undamaged as far as this is quickly checkable

Strongly agree!
Actually I'd like to have a simple demuxer (outside the lavf
structure) and muxer (not necessarily optimal, but usable) both under
BSD license as well, to encourage people to adopt nut.

> maybe we should just drop the index mess and return to a single index at the 
> end, it would be so much simpler for the muxer and demuxer and damage to the 
> index could be detected very quickly while still allowing (slow) seeking in 
> damaged files
> rich whats your oppinion about that?

I like it a lot. But I still think we need a good way for dealing with
excessive numbers of keyframes, e.g. audio, while still seeking to the
nearest keyframe (including all keyframes). This is orthogonal from
what we've been talking about but I'd still like to know if you have a
clean solution for it.

Rich




More information about the MPlayer-cvslog mailing list