Implement BaseCoordinateFrame.frame property

[eerovaher]

Description

It is not uncommon to have code that extracts (or could extract) the coordinate data from a SkyCoord as a BaseCoordinateFrame by accessing the frame property of the SkyCoord. Such code can now handle both SkyCoord and BaseCoordinateFrame instances in a uniform manner. Making use of this opportunity in astropy addresses Ruff rule PLR0915 (too-many-statements) in astropy.coordinates.sky_coordinate_parsers._get_frame_without_data() and a UP038 (non-pep604-isinstance) violation in astropy.vizualisation.wcsaxes.core._transform_plot_args().

BaseCoordinateFrame.frame could also be useful for implementing a SupportsFrame protocol, but I'm not planning to do that yet.
EDIT: I've implemented the protocol, lets see which way the discussion goes.

Most of the changes in this pull request are in coordinates, but it would be good if the visualization maintainers could review the one change I made there.

• By checking this box, the PR author has requested that maintainers do NOT use the "Squash and Merge" button. Maintainers should respect this when possible; however, the final decision is at the discretion of the maintainer that merges the PR.

[github-actions]

Thank you for your contribution to Astropy! 🌌 This checklist is meant to remind the package maintainers who will review this pull request of some common things to look for.

• Do the proposed changes actually accomplish desired goals?
• Do the proposed changes follow the Astropy coding guidelines?
• Are tests added/updated as required? If so, do they follow the Astropy testing guidelines?
• Are docs added/updated as required? If so, do they follow the Astropy documentation guidelines?
• Is rebase and/or squash necessary? If so, please provide the author with appropriate instructions. Also see instructions for rebase and squash.
• Did the CI pass? If no, are the failures related? If you need to run daily and weekly cron jobs as part of the PR, please apply the "Extra CI" label. Codestyle issues can be fixed by the bot.
• Is a change log needed? If yes, did the change log check pass? If no, add the "no-changelog-entry-needed" label. If this is a manual backport, use the "skip-changelog-checks" label unless special changelog handling is necessary.
• Is this a big PR that makes a "What's new?" entry worthwhile and if so, is (1) a "what's new" entry included in this PR and (2) the "whatsnew-needed" label applied?
• At the time of adding the milestone, if the milestone set requires a backport to release branch(es), apply the appropriate "backport-X.Y.x" label(s) before merge.

[mhvk]

I'm a bit of two minds on this. On the one hand, I like the simplifications in the code, but on the other hand it would seem like those mostly could be reached just as well by judicious use of getattr(other, "frame", other) followed by an isinstance check. I also , I don't really like properties that return self and am a little unsure about the implicit API change, that anything that has a .frame attribute will now work a bit like a SkyCoord - though my general inclination towards ducktyping makes this less of an issue.

Given that I'm on the fence, maybe best to wait and see what others think.

[mhvk]

I'd put a comment here that this is for compatibility with SkyCoord.

[neutrinoceros]

If I understand correctly, what matters is what these classes have in common, so I would suggest to formally define this common base as a typing.Protocol and use it here instead of an Union.

[mhvk]

I actually like that @eerovaher has kept to the simple & readable for now; that seemed to be the conclusion too from the long discussion we had on typing, that we use it first mostly as documentation.

[neutrinoceros]

that's what I'm proposing too. A protocol would document exactly which parts of these classes are needed here.

[mhvk]

But it might well be less readable - even a simple SkyCoordLike is less clear than just listing two options.

Anyway, it is no big deal, I just want to avoid overzealous typing when the benefits are unclear (and it seems obvious that the whole ecosystem is still rapidly evolving).

[neutrinoceros]

The description got me on the fence too because I don't think PLR0915 and UP038 are particularly useful rules to enforce and I was worried that was the whole motivation for this PR. However, reading through the diff I appreciate the idea of making BaseCoordinateFrame more like SkyCoord as it visibly helps with reducing coupling in a couple places that can now stay blissfully unaware what exact type they are being passed. As far as type annotations go I think defining an actual Protocol would be useful documentation for what exactly are the important common interfaces between these two classes, but other than that LGTM.

[Cadair]

UP038 are particularly useful rules to enforce

I personally have a the exact opposite opinion to this, I think it's entirely pointless (a tuple is just as clear as a union operator), it dosen't play nicely with models (as they override the | operator) and according to ruff it's slower! "Note that this results in slower code.". So UP038 is one of the few rules that live in my ignore config permanently.

[Cadair]

I am also on the fence about this, I don't like it from an API clarity point of view, it's really weird. I also don't know how many people in the wild would actually find this useful (I suspect it will be handy in sunpy in the same way it's handy internally for astropy) as we really don't recommend using coordinate frame objects over SkyCoord anywhere?

[neutrinoceros]

I personally have a the exact opposite opinion to this

@Cadair did you miss the negation in my sentence ? I think we actually agree !

[Cadair]

Haha yes. Don't read GH until after coffee lol

[eerovaher]

I've expanded this pull request by implementing the SupportsFrame protocol.

@mhvk commented

On the one hand, I like the simplifications in the code, but on the other hand it would seem like those mostly could be reached just as well by judicious use of getattr(other, "frame", other)

The code can indeed be simplified with or without implementing BaseCoordinateFrame.frame, but at least mypy seems to have some trouble figuring out what the return type of getattr(other, "frame", other) is, so for type checking the ability to simply access other.frame is much better.

...am a little unsure about the implicit API change, that anything that has a .frame attribute will now work a bit like a SkyCoord

We can get better duck typing both by implementing BaseCoordinateFrame.frame or by using getattr(other, "frame", other), but it will be much simpler to specify what is required for duck-typing by implementing a protocol, and then there is little reason why BaseCoordinateFrame shouldn't implement that protocol.

@neutrinoceros wrote

As far as type annotations go I think defining an actual Protocol would be useful documentation

The reason I didn't implement the protocol when I opened the pull request was that I wasn't sure how it should be documented. We are currently not documenting protocols in the user documentation (we don't really have any to document) and the utility of the protocol in type annotations is limited by the fact that the fraction of annotated code in astropy is very small.

@Cadair commented

I also don't know how many people in the wild would actually find this useful (I suspect it will be handy in sunpy in the same way it's handy internally for astropy) as we really don't recommend using coordinate frame objects over SkyCoord anywhere?

In astropy the classes that implement the SupportsFrame protocol are SkyCoord and BaseCoordinateFrame, but perhaps some users would like to have some custom class that does not inherit from either, and now with the protocol in place it starts being possible.

[eerovaher]

You can have a look at how SupportsFrame is documented.