[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Python 3.9 & Numba



Hello,

I was trying to update dask & umap-learn both which led to a numba
dependency. Numba is currently broken in Debian unstable because we've
switched to Python 3.9, and the latest release of numba only supports
Python 3.8.
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=972246

However there's a pull request that was merged upstream that updates
numba to support Python 3.9, 
https://github.com/numba/numba/pull/6579
Unfortunately upstream doesn't feel like there's been enough testing
yet of it for an official release.

(Upstream's full comments
https://github.com/numba/numba/issues/6345#issuecomment-762413345 )

The pull request merged onto the last released version of numba 0.52.0
with just a bit of patch fuzz and a conflict in setup.py over the
version of llvmlite. (Debian has numba 0.51.2 in unstable)

I've run all of numba's test cases using the 3.9 compatibility patch,
all the tests cases for an updated version of python-sparse 0.11.2 [1],
and all the tests cases for dask & dask.distributed  without problem.

There's a bunch of packages that depend on numba and I'm assuming we'd
like to see them in bullseye.

I pushed the work I did to a fork of the numba repository.
https://salsa.debian.org/diane/numba

I'm wondering what we'd like to do next?

Does Mo Zhou want to review the commits to numba? Or should I push them
to the main numba packaging repository? (I'm in the python, science,
and med teams)

I was guessing the most likely path would be to make an experimental
release of numba 0.52.0 with the compatibility patch and then see how
pandas, astro team packages do with it.

But it's a complicated package capable of strange side effects and I
thought we should talk it over first.

Thank you,
Diane Trout

[1] As an aside we also have an out of date version of python-sparse as
we have an out of date upstream url. Discussion
https://github.com/dask/dask/issues/7078
and pypi does list the same location the dask developer told me.
https://github.com/pydata/sparse/
This might be relevant to xarray as they patched around the old version
of sparse 
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=960380


Reply to: