cc @glemaitre

Nop. It does not seem to work. But I don't understand why only one build is failing while we have several builds with NumPy 1.21.2?

I did not look at it yet, but I found this entry in the release note: https://numpy.org/doc/stable/release/1.21.0-notes.html#new-min-digits-argument-for-printing-float-values

And I can't reproduce it on a conda environment with the same versions. Hmmm

@rgommers anywhere in particular you could point us at?

That looks unfamiliar to me. It may have been an unintended side effect of numpy/numpy#18629. @ahaldane, @seberg any thoughts on this one?

I don't really understand it, or how it would have to do anything with this. Would it be possible to get the print options to see if it is possible/how to reproduce it? Because I cannot locally.

@seberg here's the print options right where the error happens:

{'edgeitems': 3, 'threshold': 1000, 'floatmode': 'maxprec', 'precision': 8, 'suppress': False, 'linewidth': 75, 'nanstr': 'nan', 'infstr': 'inf', 'sign': '-', 'formatter': None, 'legacy': False}

It's very confusing, because locally I have the same numpy version, the same print options, but this is what I get:

>>> numpy.float64(9.987631416202845e-01)
0.9987631416202845
>>> numpy.array([9.987631416202845e-01])
array([0.99876314])

which is what I'd expect, but it's not what we get in the CI.

Confusing, is it possible that the runner has custom, non-default, printing for float objects? Maybe try printing repr/string for the normal Python float?

So it seems like some patch on python or other libs on the conda main channel. Using conda-forge fixes the issue (I think): #21337

I think we can close this debug PR because this specific problem has gone away.