rolling keep_attrs & default True

[mathause]

• Closes ds.rolling() drops attributes and name  #4497
• Tests added
• Passes isort . && black . && mypy . && flake8
• User visible changes (including notable bug fixes) are documented in whats-new.rst

Only global attributes were retained for rolling operations. DataArrays would loose their attrs.

As per #3891 (comment) I also changed the default to True.

Note #4497 also mentions propagation of the name. This is not yet implemented.

[mathause]

Suggested change

	If True (default), the object's attributes (`attrs`) will be copied
	from the original object to the new one. If False, the new
	If None (default) or True, the object's attributes (`attrs`) will be
	copied from the original object to the new one. If False, the new

[keewis]

I'd replace the (default) and the , optional in the type spec with , default: None

[max-sixty]

Great, thanks!

Are the global configs still being honored in rolling? No strong view on whether they should be, though it affects whether we have defaults of None.

[mathause]

As mentioned in #4513 I should probably switch from ds.rolling(..., keep_attrs=True).mean() to ds.rolling(...).mean( keep_attrs=True). Let's see if there is feedback/ a consensus on #4513

[mathause]

This is ready for a review. It was a bit more complicated than anticipated - I had to thread keep_attrs through a lot of functions. So may not be trivial to check.

To summarize:

• Per where should keep_attrs be set in groupby, resample, weighted etc.? #4513 this switches from ds.rolling(..., keep_attrs=False).mean() to ds.rolling(...).mean(keep_attrs=False) & prints a deprecation warning.
• Per # Keep attrs by default? (keep_attrs) #3891 this now keeps attrs per default (it respects the global keep_attrs setting)
• Fixes ds.rolling() drops attributes and name  #4497

[mathause]

The warning used to be:

"and will raise an error in xarray 0.16.0"

I decided to continue warning and not raise an error - objections?

[max-sixty]

Wow, this was quite the herculean effort @mathause . Thanks. That solves any questions I had about whether the global configs are being honored...

To the extent we have to do more of these, we could consider whether we need to add keep_attrs kwargs, or we should recommend users removing attrs separately where needed. We'd still need to through the global defaults, though.

[max-sixty]

Any other thoughts before we merge?

[mathause]

I added a test to ensure da.name is conserved & the tests passed. So this is finished from my side. I guess this is not so easy to review as I change one or two lines per function in a lot of functions. So maybe someone could just review the tests? Or maybe you already did?

While working on this I thought rolling could profit from some refactoring but I kept it out of this PR. I might follow up with do this if I find time.

[max-sixty]

I think this looks great. I read through the tests carefully and browed the code. I would merge unless others have thoughts

[mathause]

Testing a global default and kwarg revealed some more issues. Should be fixed now.

I realize this does still not preserve the encoding. Not sure what our general take is on that.

[mathause]

I now do a deep copy here. I think this is safer. Else non-rolled DataArrays were connected to the original obj.

[max-sixty]

Thanks again @mathause !