It looks like since 0.55.1, numba removed the upper bound bin on numpy (previously was 1.22). I was wondering if it would be possible to cut a new release with those changes so that I am able to use numba with a later version of numpy!
I have the same problem. today I did a fresh install of python, and I got numpy â1.22.3â and when I install numpy 0.55.1, I get âImportError: Numba needs NumPy 1.21 or lessâ
is there a temporary workaround for this?
for now I just commented out L140: raise ImportError(âNumba needs NumPy 1.21 or lessâ) in site-packages/numba/init.py
Thank you for asking about this. We are currently working on NumPy 1.22 support here:
We are currently (today) still missing the 1.22 packages on defaults which would allow us to cut a clean build.
We are planning to release a 0.55.2 soon (next few days, maybe weeks) which will contain only the 1.22 patches â this should hopefully be compatible.
As for the NumPy version pin, for Numba, when we release, we pin the latest known working and tested version of NumPy on the release branch. So, the main branch doesnât have any NumPy maximum, which enables smoother testing of new releases, but the released versions always will. We do this defensively because all the recent NumPy releases were incompatible with Numba for some reason and required additional patches to work.