[FFmpeg-devel] FATE filter tests

[Timothy Gu]

On Mar 17, 2014 3:24 AM, "Clément Bœsch" <u at pkh.me> wrote:
>
> On Sun, Mar 16, 2014 at 01:25:47PM -0700, Timothy Gu wrote:
> > Hi all,
> >
> > Recently Iḿ trying to add some fate tests for video filters, and I can
> > see that there are two kinds of filter tests:
> >
> > * pixfmts tests
>
> As you said below, these are for testing the different pixfmts; it is
> useful for filters with multiple code path for each pixel formats. Also
> note that those tests are very slow to run.
>
> > * vsynth tests with framecrc, and
>
> This is the most straightforward way of adding tests.
>
> > * vsynth tests with video_filter.
> >
>
> I think this is mostly historical. Also note that video_filter is used by
> the pixfmts test. I suppose it was kept for having the test common between
> the two.
>
> Also note that the video_filter version only provides a single md5 for the
> whole output while framecrc gives more hints in case of mismatches (like
> which frame failed).
>
> Having video_filter output framecrc might be lead to too large reference
> files (there are tons of pixel formats for some filters). On the other
> hand, replacing framecrc tests with video_filter will loose some
> information.
>
> > It is obvious that pixfmts tests test all the pixfmts the filter
> > supports, but I can also see that with filters e.g. overlay you use
> > multiple vsynth framecrc's instead. What is the rationale behind that?
> >
>
> Overlay needs multiple input so I suppose it's a special case.
>
> > Also some tests e.g. crop use video_filter instead of framecrc.
> > framecrc has higher precision than video_filter, which uses md5. What
> > advantage does using video_filter have?
>
> Maybe faster, less size overhead, which will probably grow a lot in case
> of pixfmts tests.

OK, thanks for the explanation. But my other question is: is using
both pixfmts and framecrc preferred? Or is choosing one of them
enough?

Timothy