community Policy for minimum versions on dependencies

There's a discussion over in the Dask issue tracker around adopting a new policy for the minimum version of dependencies Dask should support. Currently the proposal is "we support versions which were released in the last 18 months".

This would mean, for example if we adopted this policy today, our minimum supported version of NumPy and pandas would be:

pandas >= 0.25.2 (our current minimum is 0.25.0)
numpy >= 1.17.2 (our current minimum is 1.15.1)

I'm bringing this up in the community issue tracker in case there are people here who might have thoughts on such a policy but don't watch the dask/dask issue tracker. If you have thoughts on this topic, in particular if adopting such a policy would negatively impact you, I encourage you to engage over in https://github.com/dask/dask/issues/7378.

Mar 15 '21 23:03 jrbourbeau

cc @dask/array @dask/dataframe

Mar 15 '21 23:03 jakirkham

As another data point, conda-forge is moving to NumPy 1.17 as a minimum ( https://github.com/conda-forge/conda-forge-pinning-feedstock/pull/1346 )

Mar 17 '21 01:03 jakirkham

I can say from the perspective of an Xarray developer, these changes are fine. Our current minimum supported versions are:

numpy=1.17
pandas=1.0

Somewhat related, I thought I'd mention that @crusaderky has setup a nice automated system for Xarray to keep its minimum version policy in sync: https://github.com/pydata/xarray/blob/master/ci/min_deps_check.py

Mar 17 '21 14:03 jhamman