1734477 - WPT test "rotate-transform-equivalent.html" fails in Firefox (with WebRender)

Status: RESOLVED WORKSFORME

(Core :: Graphics: WebRender, defect, P3)

Description

[Daniel Holbert [:dholbert]]

Attachment: test log, suitable for pasting into reftest-analyzer.xhtml

WPT test rotate-transform-equivalent.html fails in Firefox, with a very slight positioning difference.

https://wpt.fyi/results/css/css-transforms/animation/rotate-transform-equivalent.html

We show a slight "halo" (an antialiasing artifact?) around the green shape (a few pixels away from its edge), and the halo is a slightly different color in the testcase vs. the reference case.

The "halo" seems to be a WebRender-specific artifact -- it's not present at all in Firefox-with-WebRender-disabled, and it's not present in Chrome (both of which pass this WPT test).

Comment 1

[Daniel Holbert [:dholbert]]

Note, this test is included in the "compat2021" test set:
https://github.com/Ecosystem-Infra/wpt-results-analysis/blob/master/compat-2021/css-transforms-tests.txt

If possible, it would be nice to fix the bug (or fix the test to avoid the bug, if we believe that this fringe is an acceptable artifact & shouldn't be something that WPT tests assert against).

Comment 2

[Daniel Holbert [:dholbert]]

It looks like the "halo" artifact here is from the element's background "bleeding" beyond its border, via antialiasing on pixels that are only partly covered up, or something like that.

This artifact happens in Chrome as well, FWIW, in 2D transforms like the first example below -- but it's not present if there's any 3D component to the transform as in the second example below:

data:text/html,<div style="rotate: 0 0 1 45deg; border: 2px solid white; background: red; height: 50px; width: 50px">
data:text/html,<div style="rotate: 0 0.01 1 45deg; border: 2px solid white; background: red; height: 50px; width: 50px">

Comment 3

[Daniel Holbert [:dholbert]]

BTW: the fuzzy threshold seems to be pretty non-intermittent here, on a given platform -- but the fuzzy threshold may vary per-platform.

• On my local linux machine, I have 348 pixels different, maximum difference per channel 35
• On try, Linux64 and Windows and android-em-7-0-x86_64-lite-qr seem to match that except with maximum difference per channel 34
• ...though Android 7.0 x86-64 WebRender instead seems to reliably report Found 277 pixels different, maximum difference per channel 46 (possibly different due to rendering at a different resolution?)

Comment 4

[BugBot [:suhaib / :marco/ :calixte]]

The severity field is not set for this bug.
:jimm, could you have a look please?

For more information, please visit auto_nag documentation.

Comment 5

[BugBot [:suhaib / :marco/ :calixte]]

The severity field is not set for this bug.
:jimm, could you have a look please?

For more information, please visit auto_nag documentation.

Comment 6

[Glenn Watson [:gw]]

Passes now.