[10:24:24] https://twitter.com/__jakub_g/status/1333711930004025346 [10:59:07] I was able to try the automatic IE to Edge redirect in a Windows 7 VM and it's really not that bad. I'll note however that Edge wasn't installed by any Windows update. But as soon as you open IE, Microsoft advertises it everywhere, including the default homepage [11:42:32] ah, turns out that Edge isn't installed automatically because the VM is Enterprise Edition [11:42:51] for consumer Windows it does get installed through automatic windows updates: https://support.microsoft.com/en-us/help/4567409/update-for-the-new-microsoft-edge [11:44:04] now I wonder if this get advertised in the IE11 UA string, which would let us know how many people would be affected by us being added to the list [11:45:30] hmmm no, unfortunately [22:54:17] Krinkle: added a comment to https://gerrit.wikimedia.org/r/c/mediawiki/core/+/589465/10 [23:02:37] AaronSchulz: I spent some time today looking at begin/commitMasterChanges with Nikki/CPT and the way these are invovled in running deferred updates and jobs. Definitely needs a lot of improvements to the documetnations surrounding that :D lot of rabbit holes we went down but a couple of things stood out that I'll dump here for reflectionL [23:03:57] - we very precisely fake __METHOD__ on behalf of a DU/Job object so that it matches what __METHOD__ would we inside the run/doUpdate method. This isn't obcious since most jobs and updates never make use of that. As a result it's pretty had to figure out what's wrong with something like "transaction ActivityUpdateJob::run" since nowhere in that class does it do anything with LBFactory. Finding yout way from that class to to e.g. [23:03:57] JobRunner's begin/commit call is non-trivial. [23:04:56] - We use commitMasterChanges in DU, in JobRunner and in MediaWIki.php several times for committing changes with a bogus/useless __METHOD__ parameter that is ignored since it is doing an implicit transaction. This leads to a goose chase for a balanced begin* call that doesn't exist. [23:06:07] - We use commitMasterChanges sometimes when the code already knows there was not begin, and there were no other changes, but we do it anyway just for the purpose of asserting that there is no transaction round, kin dof like a sanity check / checkpoint. It might make sense in those cases to factor that out into a safer ans simpler method that just performs that assertion, unrelated to committing. [23:07:08] - Duplication of JobRunner in EventBus was another time-sink. Probably high up on the list of thigns to get rid of in favour of re-using core with added options as needed. [23:07:34] It turns out that 1 line difference between the two is the likely cause of T269021 [23:07:35] T269021: MediaWiki::restInPeace: transaction round 'ActivityUpdateJob::run' still running - https://phabricator.wikimedia.org/T269021