http://tinderbox.mozilla.org/showlog.cgi?log=Cedar/1306773373.1306776029.13649.gz&fulltext=1 The test failure is not visually noticeable.
6 years ago
(In reply to comment #0) > The test failure is not visually noticeable. Elaborating on that slightly: At least in the latest instance, there's a single column of vertical pixels that mismatch in each of the far-right circles. The mismatching pixels are rgb(255,214,102) in the testcase vs. rgb(255,215,102) in the reference -- that's a difference of 1 in the green channel.
Per comment 40, it looks like we're just triggering an occasional visually-undetectable rounding error with a single-pixel-position gradients. While this might be a real bug, it's pretty minor & not at all what this test is meant to verify. There's actually no reason we need to use a gradient here, AFAIK -- we could just as easily be scaling a plain circle.
Created attachment 577186 [details] [diff] [review] fix: use smiley instead of gradient-fill This changes the test to use a smiley-face instead of a gradient, to hopefully avoid the randomorange, per previous comment. The smiley-face is essentially copypasted from skew-1.svg, except that I cut most of the values in half to get it to be the right size (r=20 instead of r=40). (one exception: I didn't cut the mouth-size in half, because I was too lazy to figure out what values to tweak and by how much -- and also, the double-size mouth actually looks fine. :) )
Comment on attachment 577186 [details] [diff] [review] fix: use smiley instead of gradient-fill Review of attachment 577186 [details] [diff] [review]: ----------------------------------------------------------------- Looks good to me. Thanks for fixing this Daniel!
No problem, thanks for the quick r+! Pushed to Try as a sanity-check before landing: https://tbpl.mozilla.org/?tree=Try&rev=7937d628cf63
https://hg.mozilla.org/integration/mozilla-inbound/rev/619852efb27e Assuming this really fixes the randomorange on m-c-based branches, we should land this on branches, too (since it's a test-only fix). I'll wait a week or two before doing that, to be sure m-c-based branches like this.