Open jeromekelleher opened 1 month ago
On reflection, the reasonable thing to do here is to catch the pickle-time error and raise a helpful error message. We don't need to support mixed numpy version environments, as it's an intermediate format after all.
We get a backtrace like:
I'm not sure there's an awful lot we can do about this, unfortunately. Numpy 2.x and 1.x are going to coexist for some time, so it's not clear what we can do here.
The easy solution would be to pin to numpy > 2 and bump the ICF metadata version. However, pinning to numpy > 2 isn't helpful in the short term, I think.