Le ven. 7 déc. 2018 à 16:42, Steve Dower
As a slight aside, 8 out of 8 buildbot messages on the PR look like false positives, and none of the true positives sent a message. What happened there?
I don't know why the PR didn't get notifications about the regression. I got something more than 20 emails in my buildbot mail box :-) Let me have a look: https://github.com/python/cpython/pull/10245#issuecomment-445184041 There are multiple messages related to the 'custom' build: Pablo's created a custom build ("buildbot-custom" Git branch) which reverted your PR in master. It seems like Pablo's buildbot which sends notifications to PRs decided that his commit is attached to your PR. You can ignore all notifications from the "custom" build. The GitHub notifications are still experimental. If someone is interested, see how you can enhance these notifications ;-) s390x Debian 3.7 failed on "git clone". Well, that's a random network issue. We are working on trying to make this specific issue quiet: https://github.com/python/buildmaster-config/pull/69 Anyone is welcome to help to enhance the Python CIs, there are plenty of things to enhance ;-) Buildbots are complex beast, but the situation should be better nowadays than 2 years ago. The number of random failure is quite low. Maybe random failures are more visible today because notifications are sent on state change (basically when the color changes). Two years ago, all buildbots were always red (fail), so we never get any notification :-) I'm still fixing random failures every week: test_eintr, test_socket, etc. Victor