Very stringent python dependencies

Hi, I just tried installing hail with pip but it has very stringent version dependencies. For example, it requires pandas>0.22,<0.24 which causes a pip install hail to downgrade my pandas version.

Would it be feasible to remove all those <0.x in the dependencies? Usually, new versions of e.g. pandas introduce less problems to the user than downgrades caused by pip packages (pip does not resolve dependencies in contrast to e.g. conda).

[johnc1231]

Updating pandas version in #7301 to (partially) address this issue. There’s still some larger questions about not pinning specific versions in the first place.

[tpoterba]

I’d propose to do an implicit dependency audit every time you push a new commit.

You can still pin versions on published packages, but use unpinned dependencies for CI testing.

I think our only option here would be to test twice – once with the major versions we explicitly depend on, and once with unbounded versions. I don’t really like this model, since it basically couples breaking changes in other tools to Hail commits, which isn’t how it actually works.

I’d much prefer a weekly cron job that tries to update dependencies, I think, but that thing isn’t trivial to build.