uv instead of pip?

[yajo]

It's known that uv is faster than pip, and it locks in parallel. That's by itself a good reason to experiment as an alternative to creating d2n locks.

However, they finally ship a multiplatform lock format! See astral-sh/uv#3347 (comment). There's a whole project ongoing to stabilize that effort and the format.

Do you think it'd be worth it to invest some time on trying to replace pip by uv, or at least adding a new backend for python projects?

[phaer]

Yes, I personally just lock resources but am following uv with great interest for the same reason. @purepani also showed interest into giving that a try iirc.

[phaer]

Thanks for highlighting the lock format btw, that wasn't there when I had last looked at uv.

https://github.com/astral-sh/packse/blob/main/uv.lock contains a nice example, all in all it locks a lot like the discussion in https://discuss.python.org/t/lock-files-again-but-this-time-w-sdists/46593.

I believe this is very interesting for the dream2nix use case as this locks sdists and wheels, meaning one could toggle between source builds and re-using wheels without the need to re-locking.

[purepani]

That's definitely something to add as a backend, but we should wait till the lock file format is stabilized: astral-sh/uv#4893

[phaer]

I agree that the lock-file stability wouldn't need to be a blocker for a first integration.

But looking at the latest release, it looks like the unstable lock flag has been removed from uv pip compile and is only supported by uv lock atm. The latter requires a pyproject.nix and doesn't allow arbitrary requirement strings. So the interface would probably resemble that of our pdm module.

[purepani]

In that case I might cook something after an exam I have in august.

[purepani]

As a side note, it looks like pdm is going to use the UV resolver soon, so we might not need to do much to get that speedup:
pdm-project/pdm#2693

[purepani]

Looks like they stabilized the lockfile(or at least, it is ready for stabilization), but it looks like the nixpkgs version for uv is out of date due to requiring rust 1.80, which is currently in staging, so I'm just going to wait till staging is merged to do this.

[purepani]

As a side note, it looks like pdm is going to use the UV resolver soon, so we might not need to do much to get that speedup: pdm-project/pdm#2693

Has been experimentally added! PR Here: #1067

[DavHau]

Does uv support multi system lock files by now?

[purepani]

It does(though I haven't tried it myself yet).

[DavHau]

PDM works pretty well I think and we already have a module for it, so why switch to uv?

[purepani]

It's significantly faster at locking(not a huge project but a difference of 1.5 seconds and ~10 seconds on one of my projects) l, and i would imagine that being able to support multiple package managers would allow for increased adoption

Though it's probably better if that RFC for lock files comes into existence at some point.

[phaer]

The proposed lock file also has one major advantage for source builds: It locks build-requires.

See the new PEP at https://peps.python.org/pep-0751/#package-build-requires

[yajo]

I discovered https://github.com/pyproject-nix/uv2nix. Would it make sense for both projects to join forces? cc @adisbladis