use uv as package manager

[miloth]

Hi @jorenham , this aims at solving #49. The main changes are:

• Package manager moved to uv.
• Command runner changed to just.
• tox and pre-commit use uv as venv backend.

It is a proof of concept, untested on GHA, hence the "draft" status. I just wanted to check if changing the command runner was an acceptable trade-off before continuing the work.

Resolves #49

[jorenham]

Thanks a lot for this!
I wanted to see what the CI thought of it, so removed the draft status for a bit.

[jorenham]

What are the advantages for using just over poe?
FWIW; I believe that poe (unlike the name suggests) doesn't require poetry to work

[jorenham]

FYI, I also tried migrating this to uv a while back:
https://github.com/jorenham/scipy-stubs/blob/uv/pyproject.toml

But for some reason I wasn't able to get it to work 🤔. My guess is that it has to do with either hatch or uv itself not supporting stub-only packages. And although I haven't verified this, I suspect this has to do with the - in the scipy-stubs package name, which for "normal" packages must be a valid python name.

[miloth]

Yeah, I see the problem. I think it's related to how uv installs the packages in the venv. I came up with a workaround in the latest commit, basically forcing the install as a built wheel. Now the stubtest CI passes.

Going back to the choice between poe and just. I believe using Python as a command runner is a bit unnecessary. You are adding the complexity of an additional layer of interpreter for no added benefit. I prefer just, as it is a compiled tool, so much more self-contained. In the end though, the commands specified here are quite simple, so the choice will not have a massive impact. In the end, the uv team are considering their own command runner, so it could all end up under one tool anyways.

[jorenham]

Ok that makes sense 👌🏻. I'm not familiar with just, but judging from its popularity, it's pretty solid.

BTW, don't forget update the CONTRIBUTING.md, which explicitly mentions poe now.

[miloth]

The contribution docs should be sorted now. The things I think are worth considering in this PR are:

• How the command runner and tox work together. The commands here are quite simple, so I think it could be feasible to get rid of them and put all the config in tox. I'll create a child PR into this one to showcase how. Leveraging the tox-uv plugin, it should work fairly nicely the CIs.
• I'd like to investigate a bit better why the stubs are not recognized when installed with uv sync...
• How do you feel about dividing the pyproject.toml in smaller self-contained sections? It's 300+ lines, so it could be a bit difficult finding specific pieces. I was thinking about taking out ruff.toml and tox.toml. I can do this in a separate PR, to have one per topic.

[jorenham]

• How do you feel about dividing the pyproject.toml in smaller self-contained sections? It's 300+ lines, so it could be a bit difficult finding specific pieces. I was thinking about taking out ruff.toml and tox.toml. I can do this in a separate PR, to have one per topic.

Hmm, I see what you mean, but I usually try to minimize the amount of (config) files in the repo root. Having a central place for configuration can also be helpful in my experience, which can be especially practical when we change a config option that affects multiple tools (e.g. mypy and pyright).

Perhaps it could help if we'd add a comment between sections, so that they're easier to visually distinguish 🤔

[jorenham]

Hmm, I see that just doesn't automatically run in the uv environment. So becauseuv run just stubtest kinda defeats the purpose; maybe we could prefix the just commands with a uv run in the justfile? That should also help simplify the ci.yml workflow.

Oh and speaking of GHA, the just docs mention the extractions/setup-just@v2 and taiki-e/install-action@just actions. So maybe one of those could help as well.

---

In the meantime I upgraded ruff and fixed the new errors, which is what's causing the merge conflicts, as this PR also included some fixes for those ruff errors. Anyway; there's no need to worry about those ruff errors now; and you can just reset those .pyi files to get rid of the conflicts.

---

• How the command runner and tox work together. The commands here are quite simple, so I think it could be feasible to get rid of them and put all the config in tox. I'll create a child PR into this one to showcase how. Leveraging the tox-uv plugin, it should work fairly nicely the CIs.

That sounds interesting; looking forward to it :)

---

• I'd like to investigate a bit better why the stubs are not recognized when installed with uv sync...

My gut feeling tells me it has something to do with hatch (and I based that on totally nothing 👌🏻). So perhaps switching to another build tool could help with this?

[lucascolley]

uv run just stubtest

just to throw another hat into the ring 😅 with https://pixi.sh this would look like pixi run stubtest or pixi r stubtest.

Probably good to migrate to uv first though, then only consider pixi if there would be further benefits. I assume some of the dependencies here are PyPI-only anyway, so maybe the benefits of access to the Conda ecosystem won't be substantial.

[jorenham]

There's also spin of course, which is what numpy and scipy use

[lucascolley]

scipy almost uses spin 😅 scipy/scipy#21674 (and that is for just tasks rather than dependency management as well)

[jorenham]

The contribution docs should be sorted now. The things I think are worth considering in this PR are:

• How the command runner and tox work together. The commands here are quite simple, so I think it could be feasible to get rid of them and put all the config in tox. I'll create a child PR into this one to showcase how. Leveraging the tox-uv plugin, it should work fairly nicely the CIs.
• I'd like to investigate a bit better why the stubs are not recognized when installed with uv sync...
• How do you feel about dividing the pyproject.toml in smaller self-contained sections? It's 300+ lines, so it could be a bit difficult finding specific pieces. I was thinking about taking out ruff.toml and tox.toml. I can do this in a separate PR, to have one per topic.

@miloth I also don't mind merging the UV part of this PR as it currently is. We could then reconsider replacing poe in a follow-up PR.

[miloth]

Hi @jorenham , I've just come back from a long work trip, no real time to do some fun coding. I'll clean it up next week if you want. Or feel free to do as you suggested in the meantime!

[jorenham]

I've just come back from a long work trip, no real time to do some fun coding. I'll clean it up next week if you want.

No worries; I'm just happy to see that you're still interested :)

In the meantime some merge conflicts have accumulated, but those shouldn't be difficult to deal with.

Or feel free to do as you suggested in the meantime!

What suggestion of mine are you referring to here?

[jorenham]

Thanks @miloth!