Open Bug 1962907 Opened 25 days ago Updated 17 days ago

[SHIP on Android] Failure in /css/motion/offset-path-shape-polygon-001.html

Categories

(Core :: CSS Transitions and Animations, defect)

defect

Tracking

()

People

(Reporter: kaya, Unassigned)

References

Details

We are enabling SHIP by default on Android. W-swr-nofis variant is failing. (nofis variants will have SHIP enabled along with the patch in the blocked ticket.) Worth noting: the failure happens when the patch for the blocked ticket (which enables SHIP by default) is applied. I could not spot any .ini config file for the failing test.
Here's the try run: https://treeherder.mozilla.org/jobs?repo=try&selectedTaskRun=DCKoA9-SS9GGE1-w7BFXzA.0&revision=45a68cfbc37a6b5b0802fde7aaf6e859fc18e5bf

Blocks: 1951246

This is a fuzzy failure on a single pixel.

https://hg-edge.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://firefox-ci-tc.services.mozilla.com/api/queue/v1/task/DCKoA9-SS9GGE1-w7BFXzA/runs/0/artifacts/public/logs/live_backing.log&only_show_unexpected=1

The failure is:
"Maximum difference per channel 153, 1 pixels differ"
The rotated-square's leftmost-corner pixel is an antialiased-light-green color in the reference case, vs. white in the testcase.

This log also shows fuzzy-failures for offset-rotate-005.html, though in that case it's "Maximum difference per channel 1, 10 pixels differ" (with the 10 pixels being along the antialiased diagonal top-edge of a rotated rectangle.

These are probably inconsequential painting inconsistencies that are fiddly with respect to timing; and SHIP impacts the timing here, which is how it ends up "causing" these failures. Given that these are imperceptible-to-users, we should just add .ini annotations to allow these fuzzy failures (as part of the SHIP patch stack) and not worry about it too much. You can do that by generating an .ini file with a fuzzy annotation like this one

Specifically, you'll want to create a file called testing/web-platform/meta/css/motion/offset-path-shape-polygon-001.html.ini with these contents:

[offset-path-shape-polygon-001.html]
  fuzzy:
    if os == "android": maxDifference=153;totalPixels=0-1

And similar for offset-rotate-005.html, with maxDifference=1;totalPixels=0-10

Hopefully you're able to do that as part of the enabling-SHIP work (it's probably easiest to do in the context of your existing patch-stack, to confirm that the test-failure stops being reported), but let me know if you need help.

Severity: -- → S4
Component: DOM: CSS Object Model → CSS Transitions and Animations
Flags: needinfo?(kkaya)

Thanks for the response Daniel.
That sounds good, I'll adjust the test annotation accordingly along with the SHIP-enabling work.

Flags: needinfo?(kkaya)
You need to log in before you can comment on or make changes to this bug.