[Libav-user] Segfault on corrupted video

[Michael Bradshaw]

On Wed, Jun 27, 2012 at 5:58 PM, Tim Pitman <tapitman11 at gmail.com> wrote:
> Alrighty,
>
> I was relying on the return value from avcodec_decode_video2.
> Basically I assumed that if it was > 0 then the picture was valid.
> However, I was curious about what got_picture was for so if changed:
>
> if (ret > 0)
>
> to
>
> if (ret > 0 && got_picture)
>
> It no longer crashes, but instead seems to freeze for a few seconds,
> which is definitely a big step in the right direction.
>
> What is the difference between the return value and got_picture?

Some codecs require you to decode several frames before you can start
getting usable pictures. For example, if you look at
http://en.wikipedia.org/wiki/Video_compression_picture_types you will
see that in the sample picture the frame sequence goes I-P-B-I. You
won't get a usable picture from the B-frame until you decode the last
I-frame, so in this example that B-frame would ret > 0 (because it was
successful) but got_picture == 0 (because it's buffering the picture
and has to wait for the next frame before you can get a usable frame
out).

This is also why you have to flush some codecs. Look at what the docs
say about avcodec_decode_video2 for codecs that have CODEC_CAP_DELAY
capability set.

That "freeze" for a few seconds is your app just decoding the first
several frames until it finally has enough information decoded that
you start getting usable frames out.

--Michael