Justus Magin

> if that's the current plan for handling numpy>=2.0. after writing this I've opened #9177 to deal with most of these issues. I didn't notice that RTD is still failing,...

you appear to have somehow broken `numpy` (or at least `numpy.__version__`: what the error is telling you is that `numpy.__version__ is None` evaluates to `True`). Can you check if `numpy`...

> which could be due to use of dicts internally perhaps? `dict` is deterministic since python 3.7, what you're looking for is `set`. Either way, the decision on whether or...

I think I figured out why: `create_test_data` creates a dataset that has three data variables, two of which do not have both indexed dims. Thus, if these variables are indexed...

you can also use a single-variable dataset, but yeah, that would eliminate the issue

> how is that meant to load values at all? `assert_identical` calls `load` to be able to compare values, which works _in-place_, i.e. afterwards `actual` is also in-memory

I think (1,2,3) sound good to me. I'm not sure if I would go with (4), since that would either introduce a new kwarg everywhere, or a new API that...

but doesn't that mean that we'd really want to make that two separate operations? One for shuffling the data such that groups are within a single (possibly shared) chunk, and...

I'm late to the discussion (thanks for sticking with us for so long, even though we've been really, really slow!), but to me `ffill` with `limit_direction="backwards"` seems to mean "take...

not sure whether this used to work (it could have), but the new string dtype in `numpy>=2` completely removes this kind of issue.