written on Thursday, February 15, 2024
Two weeks ago I asked the question again about What Rye should be. There has been one thing that I have not publicly shared before and that is that ever since Rye exists I have also been talking to Charlie Marsh about Python packaging and Python tooling. It turns out that we had some shared ideas of what an ideal Python tooling landscape would look like. That has lead to some very interesting back and forths. To make a potentially very long story short: Together with Astral's release of uv they will take stewardship of Rye. For the details read on.
For me Rye is an exciting test bed of what Python tooling can be. I have been using this test bed to run a of experiments over the last year. I learned a lot about what is missing in the ecosystem by building it and where the challenges are. What I enjoyed the most of working on it so far has been the feedback from various people on it. I wanted to explore what a “cargo for Python” is like and it's becoming ever more evident what that might look like. At the same time from the very start I was very clear in questioning its existence.
Since we were talking I was able to run an experiment which has been to put in Astral's uv as replacement for pip-tools. If you are not familiar with it yet: uv today is a drop-in replacement for pip-tools and venv. The why is pretty clear: it's much faster than pip-tools. Instead of taking 5 seconds to sync a virtualenv, it's almost instant. It's hard to overstate how impactful this is in terms of developer experience.
For entirely unrelated reasons Rye today already picks some of Astral's tools to power other functionality. If you invoke rye fmt and rye check it behind the scenes uses Astral's ruff to do so. They are fast, sufficiently oppinonated and they do not require installing them into the virtualenv of the project. They are quickly becoming the obvious choice if you are used to excellent tooling from other ecosystems. So as it stands, three things that Rye does are either already picking Astral tools, or will soon default to doing so.
This led to a few conversations if it would make sense for Astral to continue the work on Rye and build it out into that “cargo for Python”. I'm very much convinced that there should be such a tool and that is something Charlie from Astral shares. Where we landed is a plan that looks like the following:
Rye will continue to be a test bed for what Python tooling can be. We will move the project under Astral's stewardship with the desire to use it to further explore what a good UX can be and we will be quite liberal in trying different things. For instance now that the package installation process is blazing fast, I really want to see if we can remove the need of calling sync manually. There are also a lot of questions remaining like how to make the most of the indygreg builds or what lock files should look like in a Python world. I also want to go deep on exploring a multi-version Python import system.
Rye will turn into a blessed breeding ground of different things. As the user experience becomes more obvious uv itself will turn from what it is today — low level plumbing — into that higher level tool with a clear migration path of folks using rye to that new uv.
To try Rye on top of uv today install or update to the latest version and enable the experimental uv support:
$ rye config --set-bool behavior.use-uv=true
To learn more about uv and rye head over to GitHub:
You might have some more questions about this so I compiled a basic FAQ:
Thanks to AI there is.
This entry was tagged announcement, python and rye