[mythtv] [mythtv-commits] Ticket #4621: dropped frames in MPEG4AVC (H.264) streams due to non monotone dts

[Janne Grunau]

On Tuesday 12 February 2008 00:39:20 Mark Spieth wrote:
> >> does this also cater for mpeg2 pts/dts wrapping at 26.5h?
> >> otherwise it could cause problems for mpeg1/2.
> >
> > no, thanks for reminding me. I'll fix it asap.
>
> Im not sure, but I suspect in
>
> if  (temppts <= lastvpts && temppts + 10000 > lastvpts)
>
> temppts cant be both smaller and larger than lastvpts simultaneously.
> or will also not do things appropriately.

?
nobody claims that it should be simultaneously larger and smaller. It 
checks if the pts is at most 10 seconds after lastvpts.

> also need to consider pts effects when skipping forwards and
> backwards. also skipping across the wrap boundary.

lastvpts is reseted after skipping.

> wrap detection can be eliminated by setting the start pts/dts to 0 in
> lavf and translating all subsequent timestamps.
> This is why I went down this path when I investigated the mpeg wrap
> problem ages ago.
> currently I dont have the lavf/utils.c installed in my codebase but I
> can dig it out again if you like.

It shouldn't be necessary. This is only a workaround for our libav* 
generating some invalid DTSs.
I could only see the current code to create AV sync issues if the 
calculation is used all the time. But that is fortunately not the case. 

Janne