[FFmpeg-devel] fate: Do not report side data size

Vittorio Giovara vittorio.giovara at gmail.com
Wed Mar 8 16:45:59 EET 2017


On Wed, Mar 8, 2017 at 6:51 AM, Michael Niedermayer
<michael at niedermayer.cc> wrote:
>> Removing the side_data_size from output should be fine, as its a
>> implementation detail and as seen here can even vary between
>> architecture or possibly even compiler.
>> Maybe someone that uses that ffprobe output more often can comment?
>
> I use all kinds of stuff
> if something is removed from ffprobes output it wont be tested anymore.
> We should test more not less.

We should test better, not more.

>> An alternative for fixing fate would be to use fixed size fields in
>> the new sidedata, although the possibility of it breaking similarly
>> again in the future then remains.
>
> I strongly prefer fixed size to be used in side data over platform
> dependant fields. Not only does size become testable but theres also
> a platform specific difference less in the interface which should
> help bug reproducability between platforms

I won't comment on this statement which I find wrong in many ways, but
I'm just going to say that for the case at hand
sizeof(AVSphericalMapping) is not part of the ABI, and so it is
allowed to have different sizes on different architectures.
Same for many other side data objects, this one just happens to use
size_t, which is the right type for expressing a size.

Printing the size of a struct adds absolutely no value to a test, and
changing a type because of a test is a hacky workaround I have no
intention to pursue.
-- 
Vittorio


More information about the ffmpeg-devel mailing list