I was writing this example of CSS filters: http://codepen.io/adobe/pen/KyEpe And I noticed that hue-rotate(0deg) changed the rendering of the image, which is incorrect. Due to a recent refactoring, one of the hue-rotate matrix entries has the wrong equation. I'll post a patch to fix it shortly. Our tests didn't catch this because the colors used in our tests (typically red / green) weren't affected by the incorrect entry. We need a test that touches the blue channel, too.
Created attachment 8488827 [details] [diff] [review] Patch "oneMinusLumB" should've been "lumB" in: aOutMatrix = lumB - oneMinusLumB * c - hueRotateB * s; To match the previous code (which changed to use variables in 1058756): aOutMatrix = 0.072f - 0.072f * c - 0.283f * s; For reference, lumB = 0.0722f. I haven't ran this patch though the Windows try bots yet; they might require a bit of fuzziness on the color results.
Created attachment 8489471 [details] [diff] [review] Patch [v2] Added a bit of fuzziness for D2D and changed the filter in the test from an SVG hue-rotate filter to a CSS hue-rotate filter. https://hg.mozilla.org/integration/mozilla-inbound/rev/390d478480f9 https://tbpl.mozilla.org/?tree=Try&rev=11849e0f6a68