MultiIndex coordinates loc/sel not working in 0.19.0

[genric]

What happened:

TypeError: unhashable type: 'numpy.ndarray'

What you expected to happen:
No error

Minimal Complete Verifiable Example:
Take this for example:

import numpy as np
import pandas as pd
import xarray as xr
midx = pd.MultiIndex.from_arrays([["R", "R", "V", "V"], [0.1, 0.2, 0.7, 0.9]], names=("band", "wn"))
mda = xr.DataArray(np.random.rand(4), coords={"spec": midx}, dims="spec")
mda.loc[{'band': 'R', 'wn': 0.1}]

Environment:

xarray: 0.19.0, pandas: 1.3.1, numpy: 1.21.1,

[benbovy]

Thanks for the report @genric. Probably some regression introduced in #5322. I'm surprised this hasn't been caught in the tests. I'm also surprised that it seems to be fixed in #5692. I'll take a closer look.

[benbovy]

So the issue was that #3153 doesn't work well with multi-indexes that don't like numpy arrays nor even numpy scalars given as labels. Previously it worked because of multi-index virtual coordinates, i.e., the dtype of multi-index levels was ignored when internally re-organizing the indexers passed to .sel in xarray. However, this is not the case anymore as we are currently dropping virtual coordinates by real coordinates in the context of xarray's explicit indexes refactoring.

This should be fixed in #5692 (a551c7f), although #3153 still doesn't address the case of multi-index level coordinates with float32 values.