[cairo] [PATCH 2/3 v2] test: Test a variety of scales when downscaling

[Chris Wilson]

On Tue, Sep 10, 2013 at 10:47:26AM +0200, Uli Schlachter wrote:
> On 10.09.2013 00:59, Bryce W. Harrington wrote:
> > Downscaling from 96 to 24 is easy since it's an even multiple, so try
> > scaling by -1 pixel too.
> > 
> > This adds a 1:1 scaling test case as well, which should pass through the
> > image unchanged.
> > 
> > Signed-off-by: Bryce Harrington <b.harrington at samsung.com>
> > ---
> >  test/pixman-downscale.c                    |   25 +++++++++++++++++++------
> >  test/reference/pixman-downscale-24.ref.png |  Bin 0 -> 191 bytes
> >  test/reference/pixman-downscale-95.ref.png |  Bin 0 -> 285 bytes
> >  test/reference/pixman-downscale-96.ref.png |  Bin 0 -> 301 bytes
> >  test/reference/pixman-downscale.ref.png    |  Bin 191 -> 0 bytes
> >  5 files changed, 19 insertions(+), 6 deletions(-)
> >  create mode 100644 test/reference/pixman-downscale-24.ref.png
> >  create mode 100644 test/reference/pixman-downscale-95.ref.png
> >  create mode 100644 test/reference/pixman-downscale-96.ref.png
> >  delete mode 100644 test/reference/pixman-downscale.ref.png
> [...]
> 
> Test results:
> 
> pixman-downscale-95 and pixman-downscale-24 always fail (as expected) and
> pixman-downscale-96 mostly succeeds, but fails for similar surfaces (run
> cairo-test-suite with -a). No idea why similar surfaces should influence
> anything here.
> 
> Anyone has an idea?
> 
> Uli
> 
> P.S.: As you see, no real complaints against this patch series. Some of the
> failures look weird, but I'd rather have Chris decide if this is fine or not,
> since he has much more of a clue than I do.

I'd rather we had failing tests than to stop test suite enhancements
because we have no idea why they fail. I'm even happy with the reference
images causing known failures for the other backends - once we have a
second backend that offers comparable quality in downscaling, we can
debate their relative merits. Until then everything else is a miserable
failure.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre