The xfail files record90-paint-alpha-clip-mask.xfail.png and record90-paint-alpha-clip.xfail.png override reference images record90-paint-alpha-clip-mask.*.ref.png and record90-paint-alpha-clip.*.ref.png, respectively. Yet the differences in these files is quite minor, even if we turn the threshold down to just a pixel: record90-paint-alpha-clip.base.argb32.ref.png record90-paint-alpha-clip.xfail.png PASS: Images are binary identical record90-paint-alpha-clip.base.rgb24.ref.png record90-paint-alpha-clip.xfail.png PASS: Images are binary identical record90-paint-alpha-clip.ref.png record90-paint-alpha-clip.xfail.png PASS: Images are perceptually indistinguishable record90-paint-alpha-clip-mask.base.argb32.ref.png record90-paint-alpha-clip-mask.xfail.png FAIL: Images are visibly different 4 pixels are different record90-paint-alpha-clip-mask.base.rgb24.ref.png record90-paint-alpha-clip-mask.xfail.png FAIL: Images are visibly different 4 pixels are different record90-paint-alpha-clip-mask.ref.png record90-paint-alpha-clip-mask.xfail.png PASS: Images are perceptually indistinguishable
Created attachment 81516 [details] [review] 0001-test-Drop-redundant-record90-paint-alpha-clip-xfail-.patch The two xfail.png's are unnecessary and can be safely dropped. Dropping one of them results in a handful of test cases PASSing, that had been XFAIL previously. The other was already PASSing, but only because the diff was within the threshold; with the XFAIL gone the test now gets an exact match.
You've underestimated the severity of any discrepancy in the record tests. (And I fudged the base ref pngs.)
(In reply to comment #2) > You've underestimated the severity of any discrepancy in the record tests. > (And I fudged the base ref pngs.) So does this mean that e.g. record90-paint-alpha-clip.base.argb32.ref.png is wrong and the identical file record90-paint-alpha-clip.xfail.png just works around this wrongness? Because in this cases there is no discrepancy at all (according to Bryce).
Yes, copying the base output as its ref image in this case was my mistake.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.