Bugzilla

(In reply to Daniel Holbert [:dholbert] from comment #3)
> On WPT.fyi dashboard, the differing pixels are all `RGB(0, 127, 0)` in the testcase vs. `RGB(0, 128, 0)` in the reference case.  So it's not exactly that red pixels are showing through, but rather that we're picking up a slight lack-of-color when compositing together this antialiased fringe with the (green) pixels below it.

(Or maybe this is a pixel from the outer edge of the red diamond's antialiasing fringe, where we're like *almost* entirely green, and the teensy amount of redness gets rounded down to 0 in the red color channel, but has enough of an impact to also round down the green color channel from 128 to 127.)

(In reply to Daniel Holbert [:dholbert] from comment #3)
> On WPT.fyi dashboard, the differing pixels are all `RGB(0, 127, 0)` in the testcase vs. `RGB(0, 128, 0)` in the reference case.  So it's not exactly that red pixels are showing through, but rather that we're picking up a slight lack-of-color when compositing together this antialiased fringe with the (green) pixels below it.

(Or maybe this is a pixel from the outer edge of the red diamond's antialiasing fringe that doesn't entirely get covered up, where we're like *almost* entirely green, and the teensy amount of redness gets rounded down to 0 in the red color channel, but has enough of an impact to also round down the green color channel from 128 to 127.)