What are our thoughts on still having tests of Python 2.7 in Pants? Motivation for this question is that I upgraded CI in a PR to use ubuntu-22.04 instead of the ubuntu-20.04 image (since I need a newer image for the tonic/hyper upgrade PR), but Python 2.7 is intentionally no longer provided in GitHub images. And the

actions/setup-python

action refuses to install 2.x versions. https://github.com/pantsbuild/pants/actions/runs/10492075946/job/29063241372#step:5:22. So if we want continued ability to test with Python 2.7, sort of on our own to supply the installation method.

I believe the PEX repo does something to download unsupported python versions like this. Maybe we can look at the CI setup there. I lean towards continuing to support 2.7.

The pex repo appears to build a Docker image with all of the Pythons in it. https://github.com/pex-tool/pex/blob/main/docker/base/Dockerfile

If I can ask a silly question. Do we know if we have users who are still using Python 2.7 while also upgrading Pants versions regularly?

I don’t know offhand. Frankly I’d rather we drop Python 2.7 support.

https://github.com/pantsbuild/pants/issues/21074 Not directly related, but relevant. When we drop support for older OS versions and the different mechanisms to do it. I’m generally fine with something that is a time or cost burden, that is not obviously used by a large chunk of the user base, to be gracefully sunset (when it, itself, is considered deprecated or not being updated)

I don't entirely agree with the flip side. If others maintain it or sponsor it, then great, but if continuing such support gets in the way of upgrading other parts of our stack, then I'm inclined to ditch it absent such maintenance or support.

Fair point

someone wrote a replacement for

actions/setup-python

which can install Python 2.7 via pyenv: https://github.com/LizardByte/setup-python-action

Maybe the user survey we discussed should check for any extant use of 2.7 to inform a decision

Maybe the user survey we discussed should check for any extant use of 2.7 to inform a decision

good point

And of 3.7 for that matter…

in the meantime, if I can confine the Python 2.7 tests to a single shard with a Docker container, then I can avoid the more general Docker-in-Docker problems if we had to run Docker backend tests in a Docker container

but the community cannot continue to support that sort of thing indefinitely

> but the community cannot continue to support that sort of thing indefinitely What would be a good way forward within the project to get consensus on ending support for EOL Python versions?

I changed my mind. I'm ok with dropping 2.7 and 3.7. If someone needs it, they can use an older version of pants or pay someone (as you said).

I can wait on my tonic/hyper PR until we branch

2.23.x

from

main

. Maybe then we deprecate 2.7 support in 2.23 and stop testing it in 2.24? Or should we give more notice?

If someone needs it, they can use an older version of pants or pay someone (as you said).

I mean, the other thing here is that, they can just stay on certain Pants versions - at least into 2.23ish timelines, and maybe even further. Some of my clients are still happily on 2.17 or whatever I left them at if they didn't feel like upgrading 🤷

True, if they are on an ancient version of Python, an ancient version of Pants is not likely to intimidate