[MPlayer-users] 10l hacks in vd_ffmpeg.c and colorspace issues

[D Richard Felker III]

Hmm, I just noticed that vd_ffmpeg.c does some dirty hacks to
'convert' 422P to YV12 when the output device doesn't support 422P.
With -vo mga, this is resulting in chroma planes flickering on and off
every other frame, but only when triple buffering is used. I.e.
default mode looks bad, -dr looks fine, but -dr -double also looks
bad. I glanced at the code but I don't see what's wrong right off.
Commenting out the line for the hack:

//          if(avctx->pix_fmt==PIX_FMT_YUV422P) return CONTROL_TRUE;// half stride

fixes the problem of course, but ideally the hack should be fixed so
that it works.

On a related issue, if I just insert -vop scale to correct the
problem, things are ok, but the scale filter selects to output YV12.
Shouldn't it select YUY2 (if possible) instead when the input material
is 422P so as not to lose samples? I can do this manually with -vop
format=yuy2,scale, but it seems like the default behavior should be
not to do conversions that drop samples if we don't have to...

Rich