DGdecode fetching incorrect frames

[foxyshadis]

I have a mpg that causes strange anomalies in dgdecode when playing backwards (and by extension, seeking). I have one file where if you start on 1080 and seek backwards, frame 1062 suddenly becomes frame 1082. (Or something past that, 1082-1137 are duplicate frames. This occurs in all versions of dgindex >= 1.40.

Doing anything to the file, even splitting it into an m1v, makes seeking work perfectly (though slower). I've scanned the mpeg for errors and haven't found any reported, although it might be otherwise invalid, I don't know.

Right at that point in dgindex, I notice, it won't seek backwards. I'll upload a chop of the first minute or so to your ftp.

[Guest]

Thanks for pointing this out and providing the stream. :goodpost:
It's actually a DGIndex problem. It's a very rare case in which the picture header start code has one of its bytes in the last four bytes before the end of pack. I understand it and at first sight it appears messy to fix. I will sleep on it and hope for inspiration.

[The lookahead that MPEG requires is a major PITA.]

[foxyshadis]

If it's too messy, at least there's a workaround. I'm just glad I found it and brought it to your attention.

[Guest]

If it's too messy, at least there's a workaround. I'm just glad I found it and brought it to your attention. It affects only random access; you'll get the right frames during linear play, as when encoding. But obviously, it needs to be fixed.

[Guest]

Just giving current status on this problem...

I have redesigned the core indexing implementation to avoid this problem. It's working well for me. I want to do some more regression testing and then fix the backward stepping by GOP before making a beta, probably tomorrow.

EDIT: OK, I'm putting out the new beta. Please see the main thread.