Reimar Döffinger <[email protected]> added the comment:

On Mon, Feb 14, 2011 at 01:23:32PM +0000, Karl Blomster wrote:
> I'm not qualified to argue about whether the flushing is technically wrong or
> not, but I'd certainly agree it's suboptimal since with other decoders it
> generally IS possible to implement decently working random access.

Decently working should be possible, but not "bit-exact to decoding from the 
start".
You could try fixing it yourself, in libavcodec/mpegaudiodec.c there is a flush 
function.
It memsets synth_buf to 0.
Try to find the mimimum number of other things you also have to set to 0 to fix 
it,
the possibly relevant things should be
last_buf + last_buf_size
synth_buf_offset
sb_samples
mdct_buf
granules

________________________________________________
FFmpeg issue tracker <[email protected]>
<https://roundup.ffmpeg.org/issue2598>
________________________________________________

Reply via email to