Justus Magin

The general requirement on reduce functions is that they behave like `numpy` ufuncs, e.g. `np.mean` (`xarray` does not support the other ways to call ufuncs, e.g. `np.mul.reduceat`). As far as...

this is a known issue, and one that we'd like to clean up (see #4825 for discussion). The short answer is that you should use `swap_dims` instead of `rename`: ```python...

wouldn't you be able to use `getattr` for that? `None` does not have a attribute named `data`, so this should work: ```python getattr(my_variable, "data", np.asarray(1.0)) ```

https://www.divio.com/blog/documentation/ might be a useful reference for this?

you can probably use the `minimum version policy` CI to check (we have a policy that slightly deviates from NEP29). I think we can drop `1.18`, but `1.19` should probably...

The cache will be cleared every day, or on changes to the contents of the environment file. I just reran the workflow which reported a cache miss so let's see...

with the implementation in #4972 you should already be able to specify a hypothesis strategy to create e.g. a random awkward array. Same with `dask` or other parallel computing frameworks:...

there's also the [experimental array api strategies](https://hypothesis.readthedocs.io/en/latest/numpy.html#array-api) built into hypothesis

there are still cases where it would be good to allow pushing to `main`, the most notable being the release bump / cleanup (but maybe we can temporarily allow pushes...

Yeah, deprecating return values is not fun. We somewhat recently did that for `Dataset.update` and found that we could only mention this in the documentation / release notes, which is...