Increase built-in fuzzy threshold in WPT test transform3d-sorting-006.html
Categories
(Core :: Layout, task)
Tracking
()
Tracking | Status | |
---|---|---|
firefox95 | --- | fixed |
People
(Reporter: dholbert, Assigned: dholbert)
References
(Blocks 1 open bug)
Details
Attachments
(1 file)
The WPT test transform3d-sorting-006.html gained a fairly-generous fuzzy annotation <meta name=fuzzy content="0-100;0-500">
a few months ago, to make it pass in Chromium, here (merged to our tree in bug 1715894):
https://github.com/web-platform-tests/wpt/commit/47390a2d7a51364a5124abfff92f2dd7a6bd8a6f
As it turns out, Firefox also has some fuzzy mismatches on this test, slightly more than Chromium does -- we have 544 fuzzy pixels (rather than 500) with a maxDifference of 162 (rather than 100), at least on my Linux machine. However, all of the fuzzy pixels are in the antialiased fringes of transformed shapes -- the overall stacking order is correct, which is what the test seems to be getting at.
Given that the test already has a built-in fuzzy annotation which is sufficient for Chromium but not-quite-sufficient for Firefox, it seems like we should adjust that annotation to make it accommodate Firefox's level of antialiasing fuzziness.
This test is included in the compat2021 list, BTW; hence, I'm marking this as blocking that metabug.
Assignee | ||
Comment 1•3 years ago
|
||
This test already contains a pretty-generous fuzzy annotation, which was added
to make it pass in Chromium. It turns out that Firefox requires a slightly
more-generous annotation in order to pass. This commit just adjusts the
existing annotation to account for Firefox's observed level of fuzziness.
Note that the mismatching pixels are all part of antialiased fringes at the
edges of transformed shapes. If you disregard these fringes, nothing in the
test appears to be meaningfully failing. So, this commit isn't papering over
any "real" failure, aside from these fringes that the test is already
papering-over via its preexisting annotation.
Updated•3 years ago
|
Updated•3 years ago
|
Assignee | ||
Comment 2•3 years ago
•
|
||
Note: the fuzzy mismatching pixels here are just an indication that antialiasing happens a bit differently for transformed pixels in preserve-3d
contexts vs. in 2D transforms.
I filed bug 1737904 on that problem.
(Chromium presumably had that same behavior at some point, which would explain why this test has a fuzzy annotation baked into it, per comment 0 here. Though I'm not seeing that behavior in Chrome on my current machine, so they may have improved things in recent months.)
Comment 5•3 years ago
|
||
bugherder |
Description
•