Why Aren't We uv Yet?

40 points by aleyan


grym

The knee-jerk guess is that people don’t want a VC-backed tool [...]

This is not a knee jerk reaction; this is roughly half of why I intentionally stayed away from using astral products in load-bearing capabilities at work: this financing model guarantees a tool that will enshittify somehow, and I don't want to be holding the bag when it does and i've already drank the kool-aid. Acquisition by OpenAI was a worse outcome than expected, but I expected something like it.

The other half of why i avoided astral products is that I am broadly unimpressed with tools for language A that are written in language B by people outside A-language's in-group, particularly when they're presented as "One Big Tool That Solves All Your Five-Tool Problems ... But Faster"! Astral's products assuredly were and are presented in this way. As is usually the case, in the event they turn out to be 80/20 ports that are advertised as drop-in replacements but aren't, and that don't cover the hard 20% that I inevitably have to deal with. In the case of uv, that sales pitch backfired hard enough when the tool was new that it lead to the post-hoc creation of documents like this, which, while better than nothing, is an admission that the marketing bumf and the reality on the ground are farther apart than I'd prefer: I recall trying uv on my existing work project and it blew up immediately, and that certainly did not endear me to the tool.

The pip+setuptools+venv stack is community maintained, battle tested, covers all the weird nonsense i have to handle daily, and is slower than uv. 3 out of 4 ain't bad. Pipenv, poetry, and now uv are and were useful experiments that pushed the packaging ecosystem forward; none of them are, for me, worth moving /to/; they are useful experiments to learn from and potentially adopt features from.