Allow for consistenty ordered coords

[max-sixty]

• Tests added
• Passes isort . && black . && mypy . && flake8
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

I took an initial stab at allowing for consistently ordered coords, which came up recently, by using a list rather than a set. It's fairly involved, and the currently proposed code fails quite a few tests.

Another option would be to use OrderedSet. Or to leave it as a set.

[dcherian]

Another option would be to use OrderedSet.

Would this be a smaller change overall (aside from typing)?

[max-sixty]

Would this be a smaller change overall (aside from typing)?

I think slightly smaller, yes.

[max-sixty]

As discussed in dev meeting:

• one case to look at is that of pulling a variable out of a dataset
• consider using an OrderedSet, which will perform better in worst case

[dcherian]

one case to look at is that of pulling a variable out of a dataset

@max-sixty see #4744

[keewis]

I forgot to submit these suggestions before the meeting, so they might be out-of-date (but possibly still helpful).

[keewis]

sorted always returns a list:

Suggested change

	return list(sorted(self._data._coords))
	return sorted(self._data._coords)

[keewis]

not sure if we can come up with something more efficient (or if being efficient is really necessary), but this seems a bit wasteful (we iterate over set(vars_to_replace) - set(coord_names) twice). Maybe something like this:

Suggested change

	new_coord_names = coord_names + [x for x in vars_to_replace if x not in coord_names]
	new_coord_names = [x for x in new_coord_names if x not in vars_to_remove]
	new_coord_names = [x for x in coord_names if x not in vars_to_remove]
	new_coord_names += [
	x
	for x in vars_to_replace
	if x not in coord_names and x not in vars_to_remove
	]

?

[keewis]

same here:

Suggested change

	new_coord_names = coord_names + [x for x in vars_to_create if x not in coord_names]
	new_coord_names = [x for x in new_coord_names if x not in vars_to_remove]
	new_coord_names = [x for x in coord_names if x not in vars_to_remove]
	new_coord_names += [
	x
	for x in vars_to_create
	if x not in coord_names and x not in vars_to_remove
	]

[keewis]

I'm not sure if user-defined is a good idea for Dataset, see #4649. sorted would have the advantage of always being predictable.

[max-sixty]

Going to close this in favor of using an OrderedSet — while I won't do this now, I can in the future, or someone is welcome to take this on. I should have realized that would be a better design in the first place!