The following layout test is failing on Mac Release WK1 transforms/hittest-translated-content-off-to-infinity-and-back.html Probable cause: It looks like this test recently became flakey as of 7/27. The image failure here is failing to draw two green squares. I have not reproduced yet. The test is highly flakey Flakiness Dashboard: https://webkit-test-results.webkit.org/dashboards/flakiness_dashboard.html#showAllRuns=true&tests=transforms%2Fhittest-translated-content-off-to-infinity-and-back.html Image Diff: https://build.webkit.org/results/Apple%20High%20Sierra%20Release%20WK1%20(Tests)/r234545%20(7093)/transforms/hittest-translated-content-off-to-infinity-and-back-diffs.html
This is a curious one - looks like it became flaky on some bots on that day, but on others, it was flaky before, and became less so. Seems like an interesting case to reproduce. It's probably some problem that affects multiple tests randomly.
I was able to reproduce flakiness by running the tables/mozilla/other/ directory before transforms/hittest-translated-content-off-to-infinity-and-back.html, as these tend to run back to back in the test runner. An interaction between tests is happening here as running these two directories has a good chance of causing flakiness. running the tables/mozilla/other/ directory before transforms/hittest-translated-content-off-to-infinity-and-back.html by itself does not cause flakiness. and running the transforms/ directory by itself does not cause flakiness. Command used: run-webkit-tests --root testbuild-234649 tables/mozilla/other/ transforms/ -1 --iterations 15 --child-processes 1
Update: I bisected the tables/mozilla/other/ directory and found that tables/mozilla/other/slashlogo.html run before transforms/ yields the same flakiness if not more. also found the only running tables/mozilla/other/slashlogo.html and transforms/hittest-translated-content-off-to-infinity-and-back.html together does not cause flakiness. Command used: run-webkit-tests tables/mozilla/other/slashlogo.html transforms/ -1 --iterations 15 --child-processes 1 -v
Update: These four tests run in order can reproduce the flakiness: tables/mozilla/other/slashlogo.html transforms/hittest-translated-content-off-to-infinity-and-back.html transforms/2d/transform-2d.html transforms/2d/transform-2d-prefixed.html I noted that it never became %100 failing. it seemed to stay around a %50 of failure.
<rdar://problem/43941912>