Because things, on aurora the nightly builds have the app.update.channel "aurora" but on-push builds are "default". Rather than using http://mxr.mozilla.org/mozilla-central/source/browser/modules/test/browser_ProcessHangNotifications.js#92 the test needs to be using whatever it is that the feature uses to decide on the number of buttons, because as https://treeherder.mozilla.org/#/jobs?repo=try&revision=e283c520abef&selectedJob=14559021 shows, the feature does know that an on-push build of (simulated) aurora is aurora despite not having that as the update channel, and the notification does have three buttons.

Commented out the checks for the number of buttons in https://hg.mozilla.org/releases/mozilla-aurora/rev/1c4b9f6f5dfa

Disabling on trunk too, so green on the next merge doesn't depend on me remembering to land this again. Sorry for the needinfo when I don't need any info, but I've been told that some people ignore all bugmail except for review and needinfo requests, so that a cc is either plenty or the equivalent of not ccing and filing in Core::General, and that there is no way I can tell whether or not someone just discards their bugmail unread. It's an odd world.

thanks for the heads up.

The leave-open keyword is there and there is no activity for 6 months. :Dolske, maybe it's time to close this bug?

Declaring new-year needinfo bankruptcy.

Amusing. On-push DevEdition builds no longer exist. Even though aurora no longer exists, the only DevEdition builds are nighlies, which set the update channel "aurora". So although I won't, because it would be the wrong thing to do, because the test should be testing the thing which it uses to add the button, rather than another thing which is only sometimes the same, that commenting-out could just be reverted and nothing in automation would fail. Of course, every developer's own build of devedition would fail, but we have a long history of not caring about whether our tests run anywhere but in automation. Personally, all this bugspam from beancounters about bugs where something was disabled is starting to make me think that if we can't just explain to the beancounters to STFU about how many bugs are open, we should just start removing entire tests from the tree rather than disabling them, since we have absolutely no will whatsoever to ever fix any test failure.

Pushed by mconley@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/f8191f73b797 Use AppConstants.MOZ_DEV_EDITION to detect DevEdition rather than using UpdateChannel. r=mossop