[XviD-devel] pre0 milestone

[Christoph Lampert]

On Mon, 16 Dec 2002, Dirk Knop wrote:
> 
> But that's still undecided. I still vote for it, as it's fun to play 

> Hi,
> 
> Jim,
> are you proposing we just code for your needs?
> 
> *scratching head*
> 
> We discussed that stuff on IRC and those things which produce 
> non-standard-profile streams _could_ make their way into the codec - 
> with a flag indicating the status: XVID_NONSTANDARD or similar. In vfw 
> (which concerns you as a windows user) it would be an extra tab called 
> "NONSTANDARD COMPLIANT FEATURES".
> around with that.

I vote strongly against it! 

We should only include features which are likely to be decodable by
hardware players and all our "almost" MPEG-4 compatible competitors. 
(e.g. MPEG-quantization is not in Simple Profile, but I wouldn't consider
it too bad to use it, because it's in advanced simple anyway).

But real violations of MPEG-4, like changing Qpel flag at every frame are
something different! _If_ they make it into the codec which whatever extra
tab there is, people will use them. Simply look at our pre-alpha GMC! It's
not even decoded correctly, it doesn't give the slightest improvements,
but suddenly people use it (and complain)! Why??? Because Joe User has
not the faintest idea what all the features are, so if something _sounds_
cool, they use it. That's the same reason why so many people as for
"unstable" "bleeding edge" builds instead of using stable. Performance
isn't better, compatibility is not sure, but it wouldn't be "cool" to use
the stable release. 
If there is a tab for "better performance, but less compatability", people
will immediately use it and be one day be stuck with undecodable files
later. Or maybe not, because they only use XVID, nothing else. 
All this may not be our concern, but I still wouldn't like to have several
codecs, everyone with it's own features and for every file you have to
check which codec to activate. 
MPEG-4 Profile were invented for interchangability and lets stick to
that. Instead of adding features, we should make existent ones usable. 

gruel