⚠️ Nightly upstream-dev CI failed ⚠️

[github-actions]

Workflow Run URL

Python 3.10 Test Summary

[keewis]

the interpolate tests fail because of pandas-dev/pandas#53962

[keewis]

we now get a lot of TypeErrors related to the addition of a datetime.timedelta to a cftime.datetime object. Interestingly enough, cftime itself doesn't appear to have been changed recently, so we might need more debugging to figure out which other dependency changed. Edit: this appears to have been pandas

cc @spencerkclark

[keewis]

the numpy representation issues seem related to NEP 51 (the actual change was merged in numpy/numpy#22449)

[spencerkclark]

we now get a lot of TypeErrors related to the addition of a datetime.timedelta to a cftime.datetime object. Interestingly enough, cftime itself doesn't appear to have been changed recently, so we might need more debugging to figure out which other dependency changed. Edit: this appears to have been pandas

Thanks for the ping. I'll try and look into this soon.

[spencerkclark]

After some investigation, the cftime-related failures appear to be an upstream issue related to the interaction of the official release of Cython 3.0.0 a couple weeks ago, and arithmetic operations in cftime. See discussion in Unidata/cftime#271; I think we did not see this before, because Cython 3.0.0 was in beta.

[spencerkclark]

The cftime-related failures should hopefully be fixed in the next run now that Unidata/cftime#305 has been merged.

[keewis]

The ComplexWarning import errors are from scipy but caused by a change in numpy. scipy has already adapted to that but will only upload a new build this weekend, so this should go away soon.

There's also a few AttributeErrors for iris.cube, but I don't really understand those... we have the same version of iris as for previous runs that didn't fail, so maybe this is related to the numpy / scipy issue?

Edit: both of them went away, so either the iris was intermittent, or it was related to the numpy rename