Following the recent additions to the browser tests:
This is probably caused by Chrome being picky with z-index
Following the recent additions to the browser tests:
This is probably caused by Chrome being picky with z-index
| Subject | Repo | Branch | Lines +/- | |
|---|---|---|---|---|
| Fix browser tests in Chrome | mediawiki/extensions/UploadWizard | master | +72 -104 |
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | • Gilles | T91651 Significantly increase UploadWizard test coverage | |||
| Resolved | • Gilles | T86770 Browser tests no longer work on Chrome |
@Cmcmahon has noted that this is an issue everyone is seeing. We can definitely fix it by putting "sleep 1" in front of every click action, but that's not exactly the best way.
I'm going to do a bisect on a few commits to see what I can find.
Hm, I am now of the opinion that this isn't a problem in UploadWizard, because the tests fail in Chrome all the way back to October...
It's certainly a matter of waiting for the right things, or short of that being possible, adding sleeps.
Change 191645 had a related patch set uploaded (by Gilles):
Fix browser tests in Chrome
The first run on the integration machine has a bunch of failures: https://integration.wikimedia.org/ci/job/browsertests-UploadWizard-commons.wikimedia.beta.wmflabs.org-linux-chrome-sauce/500/#showFailuresLink moving this back to development and bumping the points count.
I guess the differences are probably due to this running on linux while my machine is OS X. Running the test on saucelabs should answer that question.
The good news is that it all seems to be the same failure being repeated.
The test is doing what it's supposed to, I looked for hacks but couldn't find anything that works. It seems to be a limitation of the linux chrome driver in relationship to the upload button hack done in UploadWizard. I think that the sanest thing to do is to test UploadWizard using OS X/Windows Chrome, since we know that at least the OS X driver doesn't suffer from that issue.
I'll go ahead and change the configuration on the integration server.
Turns out that it's just a Chrome version issue. 28 fails, 31 (chosen randomly) works fine. Bumping the version to 40 on the integration server solves the issue: