clarify lazy behaviour and eager loading chunks=None in open_*-functions

[kmuehlbauer]

• Closes WritableCFDataStore realizes variable data when loading with object stored Zarr store #10612
• User visible changes (including notable bug fixes) are documented in whats-new.rst

[valeriupredoi]

looks brilliant to me, cheers very much @kmuehlbauer 🍻

[kmuehlbauer]

@keewis It would be great if you could have a look on the wording here, too before I get this in. Thanks!

[valeriupredoi]

Thanks very much for this @kmuehlbauer and cheers for reviewing @dcherian 🍻

I was talking to @pp-mo about this in pp-mo/ncdata#145 where we are trying to pop in a user warning if any of the Zarr-loaded Xarray variables have realized data ie Numpy ndarrays, and it's actually rather difficult for me to have a robust warning that doesn't emit false alarms or sits cushy on true positives, so I thought I'd ask you if you'd be willing to pop such a warning in Xarray. I can most definitely open a new issue on this, but I thought I'd first touch base with you - something a la:

import warnings


def _raise_warning(var):
    """Raise a warnings.warning if variable data not lazy."""
    warn_msg = (
        f"Variable {var.name}{var.dims} has fully realized "
        "data, if you need lazy data, then add "
        "chunks={} or chunks="auto" as argument to Xarray open_dataset."
    )
    warnings.warn(warn_msg, UserWarning, stacklevel=2)


in func: or class:
            if isinstance(var.data, np.ndarray):

[dcherian]

if any of the Zarr-loaded Xarray variables have realized data ie Numpy ndarrays

This only happens when we create a Pandas index, I don't think a warning is needed. Certainly it would be good to make our test quite better about checking these things

[valeriupredoi]

if any of the Zarr-loaded Xarray variables have realized data ie Numpy ndarrays

This only happens when we create a Pandas index, I don't think a warning is needed. Certainly it would be good to make our test quite better about checking these things

well, it happens with chunks=None, regardless of the input data (well, Zarr, in my used case), and if some, like myself above, miss that, they'll end up with a whole lotta Numpys 😀 Hence me wondering if you folks think a warning would be useful.

Hopefully with this in now, no more numpties like I was 🍺