Is there ongoing work or plan for integrating

uv

as an optional dependency resolver in Pants? I know there are feature requests for it, but was curious as to whether anybody has looked to pick it up. Faster resolution of environments would be a great improvement to have.

I am looking at this in the context of a new Python backend, it is basically the first feature that will be implemented

What's the status of the new Python backend? This seems relevant since there has been alot of pockets of thrash and it seems no one pocket is aware of the other. I know of Pim's work, some other convert uv lockfile -> Pex lockfile banditing and then there's all the Pex work that supports Pim (

--pylock

) and a whole other approach via

--venv-repository

. This question seems to be asked every month or so for the last >1 year and there is never a coherent response afaict.

I am performing a similar conversion of pex.lock -> uv.lock to add to the reference, but unhappy with that for the obvious reasons. IMO the slow dependency resolution that occurs with medium-and-up sized repos is the biggest hurdle by far currently in the Pants build system, but that’s just my view on it.

@hundreds-carpet-28072 out of curiosity, are you using https://github.com/pex-tool/pex/releases/tag/v2.64.0 yet? Although Pants really should be moving in some direction on faster resolves, I've been chipping away at what I can in the background to make Pex resolves faster. Pex >= 2.64 + Pip 25.3 yield some good results: https://github.com/pex-tool/pex/pull/2968#issuecomment-3451000127 This is not

uv

good, but it may make the wait for Pants action on this front more bearable.

I did a PoC on top of https://github.com/pantsbuild/pants/pull/22518 using the approach that John outlined in https://github.com/pantsbuild/pants/issues/20679#issuecomment-3462720148 Not sure if/when I'll be able to come back to it, so might as well share it. The code is mostly llm-authored, is missing tests (outside of the shell script), etc... but it does generate lockfiles and can be used to build and run pexes. a lockfile with pytorch generates in ~2 seconds as expected. https://github.com/pantsbuild/pants/compare/main...jcrobak:pants:uv-experiment?expand=1 One remaining item is cross-platform/environment support. IIUC, uv+pex can build (binary-only) cross platform pex's, so it might easier than running uv via a docker container/environment (which doesn't work in this branch). It probably warrants a design doc/written proposal.

@hundreds-carpet-28072 out of curiosity, are you using https://github.com/pex-tool/pex/releases/tag/v2.64.0 yet? Although Pants really should be moving in some direction on faster resolves, I’ve been chipping away at what I can in the background to make Pex resolves faster. Pex >= 2.64 + Pip 25.3 yield some good results: https://github.com/pex-tool/pex/pull/2968#issuecomment-3451000127 This is not

uv

good, but it may make the wait for Pants action on this front more bearable.

@brief-scientist-13682 I was aware of this and am planning to integrate this, but unfortunately we’re on quite an old version of Pants which doesn’t have support for that new a version of Pex being used. Our updates to Pants were delayed due to the new-to-us lockfile generation slowing down user workflows too much - for which we now have a hacky uv conversion solution, but now need to play catch up with newer versions.

Pex has full backwards compatibility.

It would seem I’m mistaken! I could’ve sworn there was a stricter range of versions for the version of Pants we’re at, but I’m able to override it, actually. I’ll happily do some tests now with pex==2.66.0 and see what improvements I can get, thanks!!

You should just use latest Pex. And also Pip 25.3 is important.