Closed dhomeier closed 1 year ago
I think I would rather skip one test if possible as you suggest?
Done. It's still tested in the py38 and py311-lts envs now.
Any thoughts how to address the actual GlueSerializer
failure?
Something like https://github.com/numpy/numpy/pull/23034 (and if so, would this have to be done upstream)?
Description
The
GlueSerializer
failure/hang first reported in #2353 has now made it into Numpy 1.25, so to keep jobs from just timing out (and burning CI hours), this pins all tests to 1.24.x as a temporary workaround. Alternatively one could skiptest_save_aggregate_slice
conditionally on Numpy >= 1.25, but that would leave that test to be run in only one or two tests – but perhaps better, than lagging by an entire Numpy release with everything else? Might reactivate anlts
job withnumpy==1.24.*
...Workaround for #2428