[FFmpeg-devel] [PATCH] avcodec/mpegaudiodec_template: skip first frame after seeking

[wm4]

On Fri, 17 Apr 2015 19:47:50 +0200
Michael Niedermayer <michaelni at gmx.at> wrote:

> On Fri, Apr 17, 2015 at 07:21:23PM +0200, wm4 wrote:
> > On Fri, 17 Apr 2015 19:04:02 +0200
> > Michael Niedermayer <michaelni at gmx.at> wrote:
> > 
> > > This could be made optional if preferred
> > > ---
> > >  libavcodec/mpegaudiodec_template.c |    7 +++++++
> > >  1 file changed, 7 insertions(+)
> > > 
> > > diff --git a/libavcodec/mpegaudiodec_template.c b/libavcodec/mpegaudiodec_template.c
> > > index 2326a90..70c17a1 100644
> > > --- a/libavcodec/mpegaudiodec_template.c
> > > +++ b/libavcodec/mpegaudiodec_template.c
> > > @@ -87,6 +87,7 @@ typedef struct MPADecodeContext {
> > >      MPADSPContext mpadsp;
> > >      AVFloatDSPContext *fdsp;
> > >      AVFrame *frame;
> > > +    int flushed;
> > >  } MPADecodeContext;
> > >  
> > >  #define HEADER_SIZE 4
> > > @@ -1645,6 +1646,11 @@ static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
> > >          }
> > >      }
> > >  
> > > +    if (s->flushed) {
> > > +        s->avctx->internal->skip_samples = FFMAX(s->avctx->internal->skip_samples, s->frame->nb_samples);
> > > +        s->flushed = 0;
> > > +    }
> > > +
> > >      return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
> > >  }
> > >  
> > > @@ -1723,6 +1729,7 @@ static void mp_flush(MPADecodeContext *ctx)
> > >      memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf));
> > >      ctx->last_buf_size = 0;
> > >      ctx->dither_state = 0;
> > > +    ctx->flushed = 1;
> > >  }
> > >  
> > >  static void flush(AVCodecContext *avctx)
> > 
> > Would this not break gapless audio if you seek back to the first frame
> > after being done decoding? (Consider usecases like infinite looping.)
> 
> i did not test it but the skip value from AVPacket should override
> this if its set
> the first packet should contain skip sidedata or the mp3 would start
> with a "half" empty output frame
> 
> [...]

I wonder if it would be safer and generally less disruptive if the
demuxer added the skip value to the first AVPacket after every seek?