Add test for appending with cftime

[maxrjones]

No description provided.

[martindurant]

This is probably a bad choice of unit

"units": "seconds since 1970-01-01T00:00:00"

because that's also the standard UNIX definition, so having no units would have the same effect.

A nice concise error:

numpy.core._exceptions._UFuncNoLoopError: ufunc 'less' did not contain a loop with signature matching types (<class 'numpy.dtypes.Int64DType'>, <class 'numpy.dtypes.DateTime64DType'>) -> None

I'll look into the error, though, I would have expected this to work.

[martindurant]

OK, so of course xarray converts times automatically unless told otherwise.

[maxrjones]

Thanks for the fix @martindurant! Did you not like the drop_vars option in the xarray backend?

Also, a bit unrelated but do you mind sharing with me what ... passed to __getitem__ does for zarr (referencing the following)?

o = z[selector.split(":", 1)[1]][...]

[martindurant]

Did you not like the drop_vars option in the xarray backend?

It was not being applied after the dataset was already loaded. This does the right thing for a normal load, but if some array fails to load (as I found with some time conversion), the options needs to be in the first open_dataset call. So you can use the argument as before, but it should work better now :)

what ... passed to getitem does

It would be the same as [:] (i.e., grab the actual values as an array) but it also works for zero-dimension arrays.

[maxrjones]

Did you not like the drop_vars option in the xarray backend?

It was not being applied after the dataset was already loaded. This does the right thing for a normal load, but if some array fails to load (as I found with some time conversion), the options needs to be in the first open_dataset call. So you can use the argument as before, but it should work better now :)

what ... passed to getitem does

It would be the same as [:] (i.e., grab the actual values as an array) but it also works for zero-dimension arrays.

Thanks for the explanations and your work on the append feature!

[maxrjones]

The kerchunk tests take a long time to run in the CI, and it looks like most of the time is spent on the setup-micromamba step. Would you be open to enabling caching in setup-micromamba to speed up the workflows?

[martindurant]

Certainly would be happy to have caching. Actually, conda now uses the mamba solver, so it might be faster again - I don't mind, so long as it works!

[martindurant]

I used dict | syntax, which is py>=3.9; updated versions. Should probably update required python version generally.

[martindurant]

@maxrjones , env building on py39 seems stuck, any idea why?

[maxrjones]

@maxrjones , env building on py39 seems stuck, any idea why?

I pushed a couple fixes for the env errors