uv: Python packaging in Rust
uv: Python packaging in Rust
uv is an extremely fast Python package installer and resolver, designed as a drop-in alternative to pip and pip-tools.
You're viewing a single thread.
Is pip the real bottle neck in software development? How many dependencies are you managing that require this speed?
The longest running tool in our toolchain is
pytest, followed somewhat closely bypoetry. I work on a project with several teams, so I end up needing to reinstall deps pretty frequently.It's not something I'm actively looking for, but I wouldn't say no to some improvements. But I need the additional features of
poetry, so I probably won't bother until it gets a bit more sophisticated.The slowness of poetry is mostly due to network bottlenecks. While a Rust based tool, might be faster, it won't be a dramatic improvement, unlike the one we have seen with Ruff.
True, but poetry still has a long startup time (probably reading the lock file?) before it gets to the network parts. Also, a lot of the time spent in our projects is compiling/installing dependencies, not downloading them, and at least the install part could probably be sped up (I'm guessing async Rust is faster than whatever Poetry is doing).
But yeah, I'm not expecting huge wins here, but if it saves me a minute or so when rebuilding my docker images (currently takes >5 min in the poetry part alone), that's worth taking the time to switch.
I'm a lot more excited about projects like Ruff though, but any part of the Python tooling ecosystem is interesting.
Damn. Poetry alone takes 5 minutes in Docker rebuilds? That's good to know.
It does for us, ymmv. pip wasn't that much faster, so the version resolution isn't our main bottleneck, we just have a lot of packages (like 30 direct deps, and they have their own deps; seems like ~140 packages in our biggest microservice).
Damn
Well, I don't need to reinstall deps very often, like every 2 weeks at most. My regular workflow is like a 5s restart of the container, and that's if something breaks with the auto-reload.
But I still wouldn't say no to some speed improvements. It's way better than our FE codebase, which takes way longer.