[Ffmpeg-devel] invalid new backstep: mpegaudio decoding either faulty or seeking is broken

[Cyrus A]

Cyrus A wrote:
> Thanos Kyritsis wrote:
>> Good day,
>>
>> I was using a ffmpeg CVS snapshot from before you guys switched to 
>> SVN. These days I decided to update my snapshot to the latest SVN and 
>> I've been noticing some things:
>>
>> There is a lot of:
>>
>> backstep:457, lastbuf:332
>> invalid new backstep 300
>> backstep:455, lastbuf:332
>> invalid new backstep 300
>>
>> output when seeking and decoding mpeg audio from avi streams.
>> I'm both using own code and the ffmpeg tool for these tasks.
>>
>> So far, I've came up to three occasions:
>>
>> 1) My own code and ffmpeg tool sometimes fail when seeking and cutting:
>>
>> ffmpeg -ss 00:00:40.332 -t 2 -i movie.avi -acodec pcm_s16le clip.wav
>>
>> This fails. clip.wav does not contain the decoded audio from within 
>> movie's 00:00:40.332 - 00:00:42.332
>>
>> 2) There is another avi movie for which both my own code and ffmpeg 
>> tool fail to decode the 1st 20 seconds, no seeking here:
>>
>> ffmpeg -i movie.avi -acodec pcm_s16le clip.wav
>>
>> after this full decoding, movie.avi is Y duration and clip.wav is 
>> Y-20 seconds duration. Those 1st 20 seconds produced a lot of 
>> "invalid new backstep" output.
>>
>> ffplay does the same: the video/audio sync gets screwed, it plays the 
>> video from the beginning and the audio with a +20 seconds offset. The 
>> old code (just before you change to SVN) works like a charm (both 
>> ffplay and ffmpeg).
>>
>> 3) Sometimes my own code succeeds to seek and decode, while at the 
>> same time the ffmpeg tool fails.
>>
>> I'm using a code structure similar (but much more minimal) to the 
>> ffmpeg tool structure in my own project to call av_seek_frame and do 
>> the av_decode_audio loops, but there are some differences on the way 
>> I skip some frames in order to get from the I frame to an exact P 
>> frame (after calling av_seek_frame()).
>>
>>
>> However, the common factor to all these is the "invalid new backstep" 
>> output. That wasn't happening some months before.
>>
>> Do you feel there are currently known issues with mpeg audio decoding 
>> or seeking ? Or would you like me to prepare some media clips and 
>> upload them for you to test ?
>>
>>
>>   
> I'm seeing this exact same behavior trying to re-encode MPEG-2 files 
> from MPEG-4 avis. I would assume there is definitely a bug here.
>
> Try using an older version of ffmpeg. It worked for me. (How old, I'm 
> not sure. Mine was REALLY old.)
>
> Cyrus
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at mplayerhq.hu
> http://lists.mplayerhq.hu/mailman/listinfo/ffmpeg-devel
Does anyone have any ideas to help us out?

BTW, the problem is with the original MPEG4/mp3 encoding, in my case. 
FFmpeg will reencode other MPEG-4/mp3s created on my old 32-bit computer 
setups without problems. That is, the SVN version of ffmpeg on 64-bit 
can take an mpeg4/mp3 avi recorded on a 32-bit machine and produce a 
solid, universally viewable (linux and windows) mpeg file.

Any help here would be greatly appreciated.

Cyrus