[FFmpeg-devel] [RFC] Specifying KEYINT (-g) has no effect in libx264

[Stefano Sabatini]

On date Wednesday 2011-05-25 16:57:48 -0700, Baptiste Coudurier encoded:
> Hi Etienne,
> 
> First thanks a lot for the patch.
> 
> On 05/25/2011 04:45 PM, Etienne Buira wrote:
> > On Mon, May 23, 2011 at 01:32:15AM -0700, Baptiste Coudurier wrote:
> >> On 5/22/11 8:33 PM, Michael Niedermayer wrote:
> >>> On Sun, May 22, 2011 at 10:40:07PM +0200, Stefano Sabatini wrote:
> >>>> Hi,
> >>>>
> >>>> see issue #157.
> >>>>
> >>>> The problem was possibly introduced in:
> >>>>
> >>>> commit 0140d3f0921e5cbb6ea8706acb0307f7ff57a133
> >>>> Author: Baptiste Coudurier <baptiste.coudurier at gmail.com>
> >>>> Date:   Sat Apr 16 16:50:50 2011 -0700
> >>>>
> >>>>     In libx264 wrapper, add -preset and -tune options
> >>>>
> >>>> and depends on the fact that x264_param_default_preset() which is
> >>>> called in x264_init(), calls x264_param_default() which defaults the
> >>>> already set options.
> >>>>
> >>>> Indeed order of options matters, with the current mechanism options
> >>>> are set in the relevant contexts, and globally processed in the object
> >>>> initialization, so the order specified on the commandline is basically
> >>>> ignored.
> >>>>
> >>>> In the case of libx264 we may process profile/preset *before* the
> >>>> other options set in the context, so that specific settings will
> >>>> override profile/preset information.
> >>
> >> No, order doesn't matter, as long as a preset is specified, it will
> >> override other options.
> >>
> >>>> This has the problem that most of the times user-set options are
> >>>> changed to medium profile or libx264 will complain&exit, see:
> >>>>
> >>>> thus making fine-tuned user options pointless (since they are changed
> >>>> to medium preset and/or will be rejected by libx264).
> >>
> >> No, you can still specify fine-tuned options without a preset.
> >>
> >>> I see 3 obvious solutions
> >>>
> >>> A. is mplayers to just use -x264opts which ive added
> >>>
> >>> B. keep track of which options have been set by the user and which are
> >>>    default.
> >>
> >> 'B' is how it should be done. It's trivial, just remap "g" in the
> >> libx264.c the way weightp was mapped.
> > 
> > Hi all.
> > 
> > Patch attached.
> > 
> > But now, I think it would be better to keep track of what options have
> > been set at libavutil level. I don't know if ffmpeg would benefit a lot
> > from that, but certainly libavutil.
> > 
> > 
> > ffmpeg_move_relevant_options_to_libx264.diff

I'm not sure this approach is a good idea (and honestly looks an hell
to maintain). So please wait and leave me some time to comment.
-- 
FFmpeg = Frenzy and Free Mastodontic Perennial Elected Goblin