[FFmpeg-cvslog] r25066 - trunk/libavcodec/imgconvert.c

[Reimar Döffinger]

On Tue, Sep 21, 2010 at 03:50:50PM +0100, M?ns Rullg?rd wrote:
> Reimar D?ffinger <Reimar.Doeffinger at gmx.de> writes:
> 
> > On Tue, Sep 21, 2010 at 03:05:08PM +0100, M?ns Rullg?rd wrote:
> >> Stefano Sabatini <stefano.sabatini-lala at poste.it> writes:
> >> > and I can barely remember the reason of the gray8=PAL thing.
> >> 
> >> It seems that if anyone is able to remember or understand this bizarre
> >> state of affairs, they are unwilling to explain it to the rest of us.
> >
> > I did my best, if anyone did not understand it it's beyond my abilities
> > to explain.
> 
> I got lost at the part about how doing a table lookup per pixel would
> somehow be easier than not doing it.

You have to do it for PAL8 code. Thus it allows the PAL8 code to be used
for all formats (e.g. for conversion to BGR32), otherwise you'd have
to write a separate conversion function for each format.
(Yes, for libswscale we could instead generate the "palette" in libswscale,
but that would not be nice for anyone doing something similar but not
using libswscale).

> > This "bug" can only appear in codecs that do not use get_buffer,
> > which limits the issue quite a lot IMO.
> 
> Since you apparently have some notion of how it should be fixed, could
> you _please_ try to make a patch?

I do think the patch I sent with the email from
Fri, 10 Sep 2010 20:06:52 +0200
should be quite close, it just is a bit too ugly.
But I'll try to find some time for it if nobody does
(e.g. I'd really like to change rawdec not to allocate
a buffer when it will never be used, and also add a
palette to the context so we do not need to malloc
for it, and after that it would be nicer to implement
- assuming these changes are possible without breaking
anything).