(Maybe) start running test_smoothness.html in CI
Categories
(Core :: Panning and Zooming, task, P3)
Tracking
()
People
(Reporter: kats, Unassigned)
References
Details
test_smoothness.html has never run in automation, per bug 1251612 comment 0. That bug ports it from being a chrome mochitest to a plain mochitest, but keeps it disabled. We may want enable it in CI if we can make it robust enough and guarantee that it only runs in hardware-vsync configurations. And we should probably hook up the WR side of the frame uniformity data collection if we do this.
|
||
Comment 1•1 year ago
|
||
test_smoothness.html does;
- send a native wheel event in a rAF callback
- repeat 1 on 100 frames
in the meantime in APZ
- collect each transform (i.e. each scroll delta by the wheel events)
then
- calculate the standard deviation of the collected transforms and see if it's widely spread or not. And if it's not widely spread, then the test gets passed.
It looks to me the test case ultimately tests every Vsync tick happens constantly at the same pace (i.e. Vsync duration), so I don't think we need the test case now.
Note that the test case might catch regressions something like, there's a very heavy task in APZ that the task isn't finished within a Vsync tick. That's said for such kind of regressions we should write a new talos test (or something similar) doing like ASAP mode on the main-thread because Vsync duration is depending on machines so that even if the test failed at a revision on our CI, it's possible that it was significantly regressed at an older revision.
I am going to close this bug as WONTFIX and will file a new bug to remove test_smoothness.html and relevant stuff.
|
|
||
Comment 2•1 year ago
|
||
(In reply to Hiroyuki Ikezoe (:hiro) from comment #1)
I am going to close this bug as WONTFIX and will file a new bug to remove test_smoothness.html and relevant stuff.
Filed bug 1894596.
Description
•