swap_dims drops coordinate units

[simonkeys]

Using swap_dims on a DataArray with quantified units drops the units on the new dimension. Dequantifying and then quantifying preserves the units:

import xarray as xr
import pint_xarray

da = xr.DataArray(
    [1, 2, 3], 
    dims=['foo'], 
    coords={
        "foo": ('foo', [1,2,3], {"units": "s"}), 
        "bar": ('foo', [4,5,6], {"units": "kg"})
    },
    attrs={"units": 'm'}
).pint.quantify()

[
    da.bar.pint.units,
    da.swap_dims(foo='bar').bar.pint.units,
    da.pint.dequantify().swap_dims(foo='bar').pint.quantify().bar.pint.units
]

Returns:

[<Unit('kilogram')>, None, <Unit('kilogram')>]

xarray 2026.1.0, pint-xarray 0.6.0

[keewis]

this is pretty much to be expected (although maybe not well documented): swap_dims drops the index on "foo" and creates a new default index along "bar", which strips the units of the variable. To avoid that, you can create the index yourself (da.set_xindex("bar", pint_xarray.PintIndex), or something similar), after which swap_dims would do exactly the same as rename_dims.

We might want to add an accessor method that makes creating indexes easier.