[00:11:37] This is really weird: https://gerrit.wikimedia.org/r/c/mediawiki/core/+/1205190 sqlite tests pass when you do "check sqlite" but they consistently fail when gate and submit runs [00:12:53] and they used to pass while selenium failed due to CF issues [00:38:57] > LoadBalancerTest::testNonZeroMasterLoad Failed asserting that 3 is identical to 0. [00:42:10] Amir1: it also passed on the first retry gate where selenium failed [00:42:23] like the whole thing doesn't make any sense [00:42:49] So probably a race condition or something else changed in master. [00:43:12] I feel like I'm one of these movies like flight plan that the main character is losing sense of what is reality and what is not [00:43:12] Although that'd have to be a very unlucky race.. [00:43:48] okay I'm going to try for the third time [00:46:21] none of these could have caused changes in sqlite: https://gerrit.wikimedia.org/r/q/project:mediawiki/core+status:merged+branch:master [02:37:49] It passed 🥴🥴 [02:37:58] Anyway [04:14:25] oh heck I remember looking at that patch (area of interest), sure don't see how you got a race cond out of that. some people have all the luck I guess [12:54:53] xSavitar: context for the less.php patch - https://github.com/wikimedia/less.php/issues/134#issuecomment-3551147774 [12:55:08] Looks like they found their way to Gerrit. [12:55:47] I tend to do it for them but it's nice they did it on their own. [12:57:38] Krinkle, thanks for the context. I had assigned the patch to JustHannah for review. Wanted to guide the author to fine it to a good state. [12:57:50] Do you prefer we import the GH issue to Phabricator so we have a task ID? [12:58:12] s/fine/refine [14:54:35] That second comment is a bit weird [14:54:43] Just almost a repeat of earlier, with a little expansion [15:02:32] https://php.watch/versions/8.5/fatal_error_backtraces could be useful [15:57:00] ori: left a post-merge CR at https://gerrit.wikimedia.org/r/c/mediawiki/libs/IPUtils/+/1205993 [16:03:11] https://doc.wikimedia.org/cover/mediawiki-libs-IPUtils/ coverage isn't far off being 100% [16:06:39] Was 97.98% 243 / 248 earlier [16:06:51] now 96.85% 246 / 254 [16:10:10] Yeah, this lack of coverage isn't represented in the coverage report because its a relation between two separate functions. and they are tested separately but not together. If the latter is changed to no longer call the former, it would break but not covered by tests. [16:10:36] Idem before this stack it was "covering" the array_map('trim') call but no inputs contained spaces [16:11:28] As you say, it's almost arbitrary ;) [16:22:04] good catch RE: isInRange though. [16:24:57] MW core does call it (among others), so it's not orphaned at least [17:53:53] https://integration.wikimedia.org/ci/job/quibble-for-mediawiki-core-vendor-sqlite-php81/1166/console : FAILURE in 10m 13s [17:53:57] 17:35:07 1) LoadBalancerTest::testNonZeroMasterLoad [17:53:57] 17:35:07 Failed asserting that 3 is identical to 0. [17:54:12] Amir1: flaky test seems to continue to be flaky :( [18:01:09] 😭