How to change coordinate values to a MultiIndex?

[AlexAndorra]

I'm running into an issue and I can't figure out if it's something the dev team wanted, or if it's a bug. In both cases, I don't know how to solve this yet, so thanks a lot for any pointers!

Context

I have an xarray object coming from a Bayesian model (I'm using PyMC and ArviZ). One of the dimensions in the model is called obs_idx, with coordinates np.arange(1989) (that's the amount of data I have).

What I wanna do

For analysis and plot, I wanna switch obs_idx's coordinates to the Pandas MultiIndex that uniquely identifies the observations. So I'm doing:

obs_idx_detailed = census_data.set_index(demographics_order).index

# make ArviZ aware of the strata we're gonna select by:
idata_multinomial.posterior_predictive.coords.update({"obs_idx": obs_idx_detailed})

This used to work really well with xarray 2022.9.0, as you can see in the screenshot below:

What I cannot do for now

As of xarray 2023.2.0, this doesn't work anymore, and I can't understand why for now:

As you can see, the MultiIndex is passed to obs_idx, but the new coordinates are not added, so afterwards I cannot select on the xarray object by doing, for example:

idata_multinomial.posterior_predictive["answers"].sel(dict(nationality="Other", unit="Lasnamäe"))

I logically get:

KeyError: "'nationality' is not a valid dimension or coordinate"

Note that:

idata_multinomial.posterior_predictive.assign_coords(obs_idx=obs_idx_detailed)

doesn't work either -- it's giving me the exact same xarray object as in the screenshot above.

Question

Does someone have an idea of what's going on, and how I can solve this?

Thanks a lot in advance, and of course thank you for all your OSS work!

PS: this behavior appeared with 2022.12.0