[FFmpeg-devel] [PATCH] improvement on r21598 ("fix stream copy")

[Wolfram Gloger]

> From: Michael Niedermayer <michaelni at gmx.at>
> 
> well, what you write is a rather terse analysis ...
> because really the tests should match unless predicted and real timestamps
> differ and of course ignoring the current rounding bug reimar is working on

I think the rounding issue is orthogonal.

The test for pkt.pts which you added in r21598 is done for _every_
packet passing the main loop.

The test for opts_min OTOH can only hit if the predicted timestamps
for _all_ streams are bigger than recording time, i.e. if N packets
with dts+duration>recording_time have already been written.

So the tests do not match, one can get in a situation where only the
first test hits, even when all predicted timestamps equal "real"
timestamps.

> anyway, with your patch the opts_min test should be redundant and could
> be removed

That is true, "make test" does still pass for me after removing the
opts_min test.  Let me know if you want that incorporated in the patch
or separately.

Regards,
Wolfram.