Datatree support

[JoelLucaAdams]

• Add support for xarray.DataTree. Thankfully this is quite a simple addition that takes an existing dataset, converts the names back to the original ones in the raw SDF files and then builds a datatree using the from_dict method.
• When the user generates a xarray.DataTree an additional attribute called flat_structure_name is added that points to the name used to reference the variable when opening as a regular xarray.Dataset.
• Add tests and documentation.

Resolves #23

The motivation behind this addition is so that we are able to interface with a package like xr-tui. This makes interfacing with and playing around with datasets on HPCs significantly easier.

Requires samueljackson92/xr-tui#4 to work with open_mfdatatree.

[JoelLucaAdams]

Well, it's unfortunate to see that this doesn't work with Python 3.10. The last supported version of xarray is 2025.6.1 and the version where the issues with failing xarray.DataTrees is resolved is in 2025.9.1.

I think we should drop support for 3.10 for several reasons:

• Its EOL is approximately October 2026 PEP 619
• xarray has dropped support for this since June 2025
• xr-tui doesn't support 3.10
• We already support 3.11, 3.12, 3.13 and 3.14.

[LiamPattinson]

When it comes to supporting Python versions, there are two places I always reference:

• The official Python version status page
• SPEC 0, which specifies minimum version support for the scientific python ecosystem.

Python 3.10 itself will still be supported until the end of this year, but SPEC 0 recommends keeping to the cutting-edge. If you want to keep pace with the likes of NumPy and Xarray, the recommendation is actually to drop 3.10 and 3.11 at this stage.

With the rise of tools like uv, it's less important than it used to be to support older Python versions, but it is likely to frustrate a lot of users who are just using the version that ships with their OS or runs on their cluster.

It's your call if you want to drop 3.10. I'm probably overly conservative about this because I recall being stuck on 3.8 for ages in the days before uv, but dropping 3.10 and maintaining 3.11 might strike a nice middle ground here.

[JoelLucaAdams]

Thanks for the advice Liam! I think to stay in-line with SPEC 0 but still allow for people who may have slightly older versions of Python installed I shall follow your advice and drop 3.10 but maintain 3.11

[LiamPattinson]

I've rebased on main, apologies that GitHub is now giving me credit for all of your hard work! The failing readthedocs build seems to be because my commit was immediately followed by an auto-commit that cancelled the previous workflow.

I tried to resolve a ruff format issue without realising that the repo contains a conflict between black and ruff. I would personally recommend switching to ruff to simplify the number of dev dependencies and replacing the auto-committing black action with one that simply raises an error for unformatted code, as in my experience this tends to cause fewer surprises for contributors.

[LiamPattinson]

Rebased again to remove the little dance with the conflicting formatting tools.

[JoelLucaAdams]

Thanks for fixing the conflict!

Regarding the ruff vs black formatter I agree we should just fully switch to ruff. Could you either fix this in a separate branch or raise a PR about it and I'll get round to it when I have more time

[LiamPattinson]

Thanks for fixing the conflict!

Regarding the ruff vs black formatter I agree we should just fully switch to ruff. Could you either fix this in a separate branch or raise a PR about it and I'll get round to it when I have more time

No worries! I'll raise an issue/PR for the workflow change later.

I've read through the code and all looks good to me, but I haven't yet had a chance to play around with the DataTree type and confirm it behaves sensibly. I've complete my review sometime this afternoon.

[LiamPattinson]

All looks good! I've tested it myself and the DataTrees are a lovely addition to the library. I can think of a few previous projects I've worked on where they would have been really helpful.

[LiamPattinson]

I must admit, I'm quite impressed that this works. I would have expected much more manual intervention to be necessary!

[JoelLucaAdams]

I was just as shocked that it worked when I first tried it, such an elegant system!