aggregation functions treat duck arrays differently depending on dtype

[keewis]

While working on #3238, I tried replacing np.arange with np.linspace to create test arrays:

>>> ureg = pint.UnitRegistry()
>>> # with int values
>>> array = np.arange(10).astype(int) * ureg.m
>>> np.max(array)
<Quantity(9, 'meter')>
>>> np.max(xr.DataArray(data=array))  # works as expected
<xarray.DataArray ()>
<Quantity(9, 'meter')>
>>> # now with floats
>>> array = np.arange(10).astype(float) * ureg.m
>>> np.max(array)
<Quantity(9.0, 'meter')>
>>> np.max(xr.DataArray(data=array))  # unit information is lost
<xarray.DataArray ()>
array(9.)

Judging by the build logs of #3238, this seems to be the case for all aggregation functions except from np.median and of course those that return booleans or indices.

[keewis]

This seams to be an issue with pint and is worked on in hgrecco/pint#764: using that PR instead of the version available in conda-forge makes all functions fail with a TypeError regardless of dtype.

So I guess this can be closed?

[keewis]

now that I hit this issue using the example from #3238 again, this seems to be a bug in xarray. For reference, this is the mentioned example that fails even with a pint version with __array_function__:

>>> xr.DataArray(data=np.arange(10).astype(int) * ureg.m).median()
<xarray.DataArray ()>
<Quantity(4.5, 'meter')>
>>> xr.DataArray(data=np.arange(10).astype(float) * ureg.m).median()
<xarray.DataArray ()>
array(4.5)

[shoyer]

@keewis could you clarify that example? Both those examples appear to be the same code, with different results!

[keewis]

that's true. I edited it, the float version should fail. The new example is actually the same as the first one, but using DataArray.median() instead of np.max()

[shoyer]

Can you try merging in the latest version of xarray master into your branch? I think this issue was fixed just recently by #3254. When I test this myself, both versions seem to do the right thing:

In [5]: xr.DataArray(data=np.arange(10).astype(float) * ureg.m).median()
Out[5]:
<xarray.DataArray ()>
<Quantity(4.5, 'meter')>

In [6]: xr.DataArray(data=np.arange(10).astype(int) * ureg.m).median()
Out[6]:
<xarray.DataArray ()>
<Quantity(4.5, 'meter')>

[keewis]

you're right, after merging the issue is gone for me, too.