[Ffmpeg-devel] Re: Matroska Patch

[Rich Felker]

On Wed, Mar 22, 2006 at 12:27:28AM +0300, Mike Matsnev wrote:
> M?ns Rullg?rd wrote:
> >The cluster time is unsigned.  You can't treat it as signed.
> Yes, cluster time is unsigned, but block time relative to cluster is signed 
> 16 bit int.
> This is an annoying artifact in the spec, but it's too late to change it.
> In practice you don't see negative pts, and there aren't huge pts requiring 
> all 64 bits
> too, so it's unclear what is a better choice. I'd allow negative timestamps 
> that can
> be used as preroll by some muxers.

IMO negative timestamps are a very bad idea. They're invalid in some
formats and may be used for in-band "no timestamp available"
signalling in certain APIs.

Rich