[Ffmpeg-devel] swscale and regression tests

[Michael Niedermayer]

Hi

On Mon, Aug 14, 2006 at 11:43:37AM +0200, Luca Abeni wrote:
> Hi all,
> 
> as someone already noticed, if ffmpeg is configured with
> "--enable-swscaler --enable-gpl", then "make test" fails. I investigated
> this problem, and I am trying to find a solution.
> 
> A first problem is that some ffmpeg's pixel formats are not supported by
> swscale. This causes, for example, failures on the jpeg, mjpeg, and
> similar tests.
> 
> When I manually disable the tests involving unsupported formats, some
> tests still fail. I checked them, and it turned out that all the failing
> tests involve rescaling (for example the dv tests) or pixel format
> conversions.
> Double thinking about it, I arrived to the conclusion that these
> failures have to be expected, because swscale uses different algorithms
> respect to libavcodec's scaling and format conversion routines... So,
> the results can be different. Is this understanding correct? (I mean:
> video rescaling and format conversion are approximated operations,
> right?)

yes


> 
> 
> So, how to fix "make test" with swscale? I can think about 2 possible
> solutions, but there are probably more:
> 1) leave the libavocodec scaling and format conversion routines in
> place, and always use them in the tests (even if swscale is enabled)
> 2) use different *.ref files whe swscale is enabled

could we make old and sws scaling runtime switchable? if thats not too
diffcult then we should do that and have regression tests for both

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

In the past you could go to a library and read, borrow or copy any book
Today you'd get arrested for mere telling someone where the library is