On Fri May 2 16:52:15 2025 +0000, Alex Henrie wrote:
I apologize for the stupid mistake. !7956 will fix it. Literally two minutes after I sent it, Bernhard sent an alternative fix in !7957 which would also be fine. I recognize that this happened primarily due to human error on my part. Nevertheless, another part of the problem is that the 32-bit tests always fail, and somehow this mistake did not cause the 64-bit tests to fail. That means that I need to spend some time helping with the intermittent test failures because they're making it too easy for me to overlook test failures that actually matter. Until the tests can be expected to pass, I will have to remember to read through the entire list of failures whenever I send a patch.
The test-linux-64 test suite only run a few tests that are otherwise skipped in a old/new wow64 environment like the one used for test-linux-32.
It's maybe a bit confusing but it otherwise would use another runner from >30min and it wasn't deemed useful to have a full test coverage run on every config.
The test-linux-32 sadly fails often, spuriously, although there have been up and downs I think it's getting better now and hopefully we'll get back to some occasional failures only.
Nevertheless, Gitlab now shows a short summary of failing tests (in the MRs "Test summary:" section, which can be expanded for a detailed list of failures) and I think we should be able to quickly check MR failures, and compare with other MRs or the testbot pattern page, to decide whether they are likely related or not.
Usually there's only a handful of spurious failures, with some well known returning failures, and a large number of/unusual failures should already still stand out, if we're paying attention.
Fwiw I'm not blaming you specifically here, it happens quite often (and I can be blamed for a fair amount the last few months). I'm mostly making sure it doesn't go unnoticed and try to raise awareness. At least as long as things get caught and fixed quickly it's probably fine. Fixing spurious failures would be nice too but that's much more difficult than it seems.
-
Rémi Bernon (＠rbernon)