El 21/3/25 a las 19:08, Jeremy Bícha escribió:
I uploaded gcr 4 4.4.0.1-1 today with an upstream change that disabled parallel running for some of the gck tests. Could you check if the build failure bug appears to be fixed in this version?
Ok, I've built the package 200 times in my setup. On AWS machines of type m7a.medium and r7a.medium, the failure rate is 100%, i.e. the build always fails. Incidentally, those machines have 1 CPU, but I'm unable to tell if that's what triggers the build failure or maybe the overall speed. In either case, build-essential has never implied multi-core, so RC or not, this is still a violation of Debian Policy. On AWS machines of type r7a.large, m7a.large or c7a.large, the failure rate is around 8%. Those machines have 2 CPUs. This is a lot better than before, but still flaky. I've put all the failed build logs I got here for you to see: https://people.debian.org/~sanvila/build-logs/gcr4/ If the tests fail in some cases but they work in other cases, it means that a test failure does not mean the program is wrong, i.e. the tests are flaky, in which case it makes absolutely no sense to punish QA testers with build failures that "have to be ignored". The right way to ignore a test failure is by disabling it to begin with. Please disable the flaky tests, they do more harm than good. Thanks.
