w3c test imports sometimes also require rebaselining non-wpt tests |
|||
Issue descriptionIn the latest w3c-test-autoroller CLs (eg https://codereview.chromium.org/2464743003/), there are some non-imported tests which appear to fail consistently. foolip@ also noticed this in a recent change. My hypothesis for this is that when importing, we update third_party/WebKit/LayoutTests/resources/testharness.js, and this affects tests outside of imported/wpt. Possible change 1: Stop updating files in LayoutTests/resources/, require humans to manually update those files. Possible change 2: When there are changes in LayoutTests/resources/, update baselines and expectations for any newly-failing tests, even if they're not in imported/wpt. I think change 2 sounds easier to manage, even though it means that "updating wpt" implies "updating testharness.js and potentially anything that depends on it".
,
Nov 2 2016
So now I've made it so that when update-w3c-test-expectations invokes webkit-patch rebaseline-cl, it doesn't use --only-changed-tests, so it will try to download new baselines for any tests which seem to be newly failing. However, when there are flaky tests that fail with the patch but not without, then new baselines may (wrongly) be downloaded for those tests, which will cause the import to fail. I noticed this when trying to manually update wpt in https://codereview.chromium.org/2468053002. In the best case, this flakiness would be an occasional but not common problem, and would just result in imports occasionally failing.
,
Nov 15 2016
,
Nov 17 2016
Issue 659180 has been merged into this issue. |
|||
►
Sign in to add a comment |
|||
Comment 1 by bugdroid1@chromium.org
, Oct 31 2016