1765640 - Avoid subpixel antialiasing differences in reftest select-3.html

Status: RESOLVED FIXED

(Core :: Layout: Positioned, task)

Description

[Daniel Holbert [:dholbert]]

reftest select-3.html has an extremely complex fuzzy-failure annotation right now -- and in fact it's not-complex enough, because the test still triggers an unexpected failure with a different amount of fuzziness on my local machine.

Looking at the test failure in reftest-analyzer, we're failing because the testcase has subpixel (colorful) antialiasing, whereas the reference case has grayscale antialiasing (presumably because it uses position:fixed).

Per bug 1763411 comment 8, this is not-entirely-surprising; subpixel AA gets enabled/disabled depending on whether WebRender can be absolutely sure there's an opaque background behind the text (and the heuristics that it uses to do this aren't perfect).

Anyway; we should fix the test to avoid tripping this unimportant-to-the-thing-actually-being-tested rendering difference between the testcase and reference case, so that we don't have to maintain a bespoke platform-specific fuzzy failure annotation.

Comment 1

[Daniel Holbert [:dholbert]]

I'm going to make two changes here:
(1) create a copy of the test that uses transparent text (so that it won't trigger subpixel antialiasing). To the extent that this test is screening for positioning differences between position:absolute vs. position:fixed select widgets, this new test will continue to screen for that.

(2) adjust select-3-ref.html to just be position:absolute. This lets us continue to test how the testcase looks (text-and-all), vs. a reference case that is still different/simpler-in-structure. (The testcase has an extra position:relative div around the abspos thing, which I think is the real important difference between the two in terms of testing the original bug.)

Comment 2

[Daniel Holbert [:dholbert]]

Attachment: Bug 1765640 part 1: Create a copy of reftest "select-3.html" with transparent text, to avoid antialiasing-related fuzzy failures. r?#layout-reviewers

Comment 3

[Daniel Holbert [:dholbert]]

Attachment: Bug 1765640 part 2: Change reference case select-3-ref.html to use position:absolute instead of fixed, to avoid fuzzy failures due to antialiasing differences. r?#layout-reviewers

This should still serve as a good reference case for select-3.html, since it's
still simpler-in-structure than that testcase.

To the extent that the testcase really does want to test positioning of select
elements with position:absolute vs. position:fixed (I'm not sure if it does,
but just in case it does): we still have coverage for that, via the new copy of
this reftest called "select-4.html" added in the previous patch in this series.

Depends on D144225

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/1bda7c512831
part 1: Create a copy of reftest "select-3.html" with transparent text, to avoid antialiasing-related fuzzy failures. r=hiro
https://hg.mozilla.org/integration/autoland/rev/58432e7f0940
part 2: Change reference case select-3-ref.html to use position:absolute instead of fixed, to avoid fuzzy failures due to antialiasing differences. r=hiro

Comment 5

[Norisz Fay [:noriszfay]]

https://hg.mozilla.org/mozilla-central/rev/1bda7c512831
https://hg.mozilla.org/mozilla-central/rev/58432e7f0940