fix tests

[dcherian]

annoyingly some tests can fail depending on where they are in test_accessor.py

For example test_dicts fails if it is at the end of the file because airds.lon.attrs == {}.

Sometimes this is an xarray bug (pydata/xarray#4120).

It would be good to track this down.

[kmpaul]

See comment on pydata/xarray#4120. If you globally set keep_attrs=True (xr.set_options(keep_attrs=True)), the tests might work.

[dcherian]

Thanks for digging in to that.

If you globally set keep_attrs=True (xr.set_options(keep_attrs=True)), the tests might work.

Yes but it's nice that we're shaking out these bugs in xarray.

[kmpaul]

Yes but it's nice that we're shaking out these bugs in xarray.

Yes! Agreed. It does make me wonder, though, if cf-xarray should enforce keep_attrs=True globally, anyway. It seems like maintaining the attributes is necessary for reliable functionality.

[dcherian]

if cf-xarray should enforce keep_attrs=True globally, anyway. It seems like maintaining the attributes is necessary for reliable functionality.

I don't think we should set it but we could raise a warning at import time?

[kmpaul]

Wouldn't a user setting keep_attrs=False globally break CF-Xarray, since CF-Xarray depends on the attributes?

[jthielen]

Wouldn't a user setting keep_attrs=False globally break CF-Xarray, since CF-Xarray depends on the attributes?

Somewhat, if no operations are applied that strip attributes (such as a simple read-and-plot workflow), then everything should be fine.

[kmpaul]

Wouldn't a user setting keep_attrs=False globally break CF-Xarray, since CF-Xarray depends on the attributes?

Somewhat, if no operations are applied that strip attributes (such as a simple read-and-plot workflow), then everything should be fine.

Then I would think a warning would suffice. Unless we get lots of complaints from people saying manually setting keep_attrs all the time is annoying.