Issue with numba 0.59.0

[israelmcmc]

David Palmore reported an issue installing cosipy related to numba version 0.59.0. It's not really an issue with cosipy, but I'll add it here to remember me to add it to the troubleshooting section in the documentation. David fixed this by downgrading to numba 0.57.0

[ckarwin]

Do you know what in cosipy is using numba? Is it a 3ML dependency?

[israelmcmc]

Here's the dependency tree in my environment. It seems numba is needed by astromodels and sparse.

pipdeptree.txt

[ckarwin]

Ah, I see, thanks.

[avalluvan]

To add, the API @numba.generated_jit has been deprecated in v0.59.0. "Use of @numba.extending.overload and the high-level extension API is recommended as a replacement."

[StevenBoggs]

Using the command:
% conda install numba=0.57.0
Seemed to get me past this hurdle for now....

[ckarwin]

Pierre Jean had the same issue with numba. Downgrading to version 0.57.0 also worked for him. Below I copy his initial error message, for reference:

The cosipy installation seemed to be fine (via conda) on my user on the server of Jürgen (linux) but I obtained the following error with the python command "from cosipy import BinnedData"

from cosipy import BinnedData
ImportError: cannot import name 'generated_jit' from 'numba' (/home/jean/.conda/envs/cosipy/lib/python3.10/site-packages/numba/__init__.py)

[israelmcmc]

I looked into this a little bit more. The traceback shows the issue propagates from the interpolation package (which call numba's generated_jit, and is called by astromodels). See this relevant issues:
EconForge/interpolation.py#103
EconForge/interpolation.py#110

I made a new release (v0.0.2a4) adding restricting numba<=0.58.0. It will throw a deprecation warning instead of an error.

I'll leave this issue open since we should remove the restriction on the numba version once this is fixed in the dependencies.