Closed lilyminium closed 2 months ago
Yeah I suspect that a pin is the way forward until the next release - sorry about that :(
Thanks @IAlibay! Just thinkign about this some more wrt the workshop and future maintainability of any MDAKits that might be created here, I was thinking that maybe the best way to do this would actually be in the install-mdanalysis
action -- just adding a 'numpy<2.0'
around here https://github.com/MDAnalysis/install-mdanalysis/blob/60ef7aab832dad506a65e28f4c927121b028191a/action.yaml#L107 . Any thoughts on that? That way if we do start supporting 2.0 then it'll automatically propagate through to kits that use the install-mdanalysis action without them needing to change anything.
Actually, I'm pretty sure I started pinning an upper version of numpy a while back - what version of MDA are you picking up?
Ok just checked, I started putting an upper pin for numpy in the package with release 2.6 and above.
As a minimum here (in the cookiecutter) I would suggest pinning the minimum MDA to >2.6, at least that'll ensure a proper failure when trying to combine np 2.0+ and MDA right now. Thoughts on this @lilyminium ?
Something does need doing about install-mdanalysis, will think about that now.
The failures only happen when installing from develop, not any particular release package. Happy to move the pin upwards but I'm not sure that's the issue here. The logs from the failed run indicate that MDA wasn't installed before the install-mdanalysis
action, so I don't think it could be the case where an older MDA is installed with numpy 2.0, then force-replaced with a newer one.
when installing develop
Ah ok, that's a key bit of information I didn't know. I have an idea, need to double check something.
Ok I think I got it, sorry there's just too many threads going on and I completely forgot about this https://github.com/MDAnalysis/mdanalysis/pull/4620
I'm going to aim to merge that PR and I think it'll clean things up, or at least it'll take us to a different state.
This should be fixed upstream - can you confirm that this is also resolved for you here @lilyminium ?
I re-ran the above-mentioned CI matrix and everything seems to run fine now: https://github.com/MDAnalysis/mdakit-cookie/actions/runs/9615305167 - probably means we're ok?
Thanks for putting in the fix @IAlibay! I'm happy that CI is all green now on https://github.com/MDAnalysis/mdakit-cookie. Closing now :-)
From mdakit-cookie CI -- https://github.com/MDAnalysis/mdakit-cookie/actions/runs/9615305167
Across Pythons 3.10-3.12 on MDAnalysis develop, we get a header size error:
Possibly needs pinning to numpy < 2.0?
Error not triggered by conda.