pandas 1.3.x breaks test

[gdementen]

test_xlsx_xlwings_io fails with TypeError: Axis name should be None, int or str but is: nan (float64)

I tried many different dependencies combinations and the result is that our test suite is fully green with Pandas 1.2.x and fails that one test with Pandas 1.3.x (and least 1.3.0 and 1.3.1).

[gdementen]

The failure is about loading back an array with anonymous axes:

f = ndtest((Axis(3), Axis(2)), dtype=float)

The change in Pandas comes down to:

>>> DataFrame([[1, 2, 3]], columns=[None, 0, 1]).columns
>>> # or even simply:
>>> pd.Index([None, 0, 1])

which return with 1.2.x:

Index([None, 0, 1], dtype='object')

and with 1.3+:

Float64Index([nan, 0.0, 1.0], dtype='float64')

Unsure what to do from here. Probably need to handle NaN as column name somewhere

[gdementen]

In fact, the problem is even specific to the case of anonymous axis with only int (and float?) labels. For example, this still works:

>>> pd.Index([None, 'a', 'b'])
Index([None, 'a', 'b'], dtype='object')