I've tried, but I can't reproduce this test failure elsewhere, either during the build of the package nor for the standalone source code, so I'm lowering the severity.
One thing that makes me suspicious is that the failures are when running the tests with python 3.8, while the new 3.9 is run earlier and ends with a success: this would make me think about a failure to run the tests twice in a row, but even that is running correctly when I try it locally. If anybody else is able to reproduce this and can explain how I'd be happy to investigate more. -- Elena ``of Valhalla''
