Make all mdakits imported & used in the core library optional and don't ship them on the conda-forge recipe with v2.9

[IAlibay]

I want to do a py3.13 release that works on conda-forge, but the way we are currently dealing with deprecated analysis modules that are now mdakits is blocking me.

Long story short, there's a cyclic packaging issue that is resolvable on the medium term (make base packages etc..), but we don't have time to deal with this on the short term and I would love to unblock a lot of py3.13 compatibility issues downstream.

I realise this effectively means we break the API early but I don't think it's worth blocking a major python release for analyses that are deprecated and we can just say "install this package and things will work as you expect them for now".

[IAlibay]

Could I get the @MDAnalysis/coredevs to weigh in on a one week timeline please?

Ideally I would like to have a decision made by next weekend so we can move ahead with a new release.

[orbeckst]

I am ok with the proposed solution. Py 3.13 is important.

Given that we're sort-of breaking semantic versioning (at least at the installation level, not at the code level) we should

• be clear in our CHANGELOG and release blog post about what's happening
• have clear error messages when the optional import fails

[orbeckst]

(sorry didn't intend to close the issue, pressed wrong button)

Quick question: How will this work with the PyPi install?

[IAlibay]

(sorry didn't intend to close the issue, pressed wrong button)

Quick question: How will this work with the PyPi install?

Same thing as conda-forge unfortunately - folks will need to manually install the additional kits.

We can make the warning messages very clear about how they can go about doing so though (i.e. put the pip & conda install options in the warning).

[RMeli]

I think with a clear warning, this should only be a minor annoyance to the users. Or at least I hope so. I'd be in favour of this change to move things forward.

[yuxuanzhuang]

I am also in favor of this change.

[tylerjereddy]

Sounds "ok," especially if each such import has already been deprecated. In which release did the deprecation happen? I think there's a convention in the community to "execute" deprecations after warning for two releases. If you're cutting to just 1 release where you warned that might be a bit tighter than desired, but for getting 3.13 support ready on limited bandwidth I'd agree with the others that it seems worth it even if not ideal.

[IAlibay]

In which release did the deprecation happen?

I think at least one was in the last release unfortunately.

[IAlibay]

Thanks all, it's been over a week so I'm going to assume that this is now approved. I'll go ahead with this change next weekend.