bheztio

> You can't really optimize the trial and error feedback loop time.

Isn't this exactly what startups in this area are doing - automated experiments connected to AI. Of course at some stage things will "graduate" to human review but it still seems to have a lot of potential.

Last I tried numba, admittedly a long time ago, it was a pain due to extensive system requirements or alternatively requiring use of conda, which I dislike (for many reasons, but it's besides the point - I want to use standard python tools). Did they start to bundle their dependencies or..?

To be fair, at least when I tested it on a moderately sized codebase, ty was definitely faster than Pyrefly. Like twice as fast (0.5s vs 1s, give or take). However, it's not really critical to save that half a second for full checks. Both have gradual checking when used as a LSP. And on CI, there is overhead from other things.

However, it remains to be seen to what degree that speed difference is due to ty's smaller feature footprint. It just doesn't seem to check the same things. I found it too permissive, personally.

It's MIT licensed. If Astral ever truly dies, there'll be community versions of their software, because it's that "necessary".

You don't sound like a Python developer, that's probably the reason. When I search for "AppDirs", I only find references to what is e.g. XDG_BASE_DIRS on Linux (and others on other platforms)? Which I asuume is not what you mean. Most likely this is something particular to your environment, whatever that is.

In the Python ecosystem, uv is great for "project management" - lock files, project dependency environments ("venv"s) etc (similar to npm for js). In the Python ecosystem it's considered best practice to bundle binary dependencies with Python libraries, and then there'll be a binary distribution for each platform (os, libc).

Type annotations are generally great, if you've never had the displeasure of working with Python pre type-annotations, you'd know that tracking down "wtf type even is this" was a big part of just working in Python. The extra text/tokens spent on annotations are 100 % worth it, both for humans and LLM.

An issue you run into these days is that even if you have excellent type checking on your own code, some libraries don't. You of course want to avoid such libraries, or see if you can find type stubs. Gradual typing can suck, but well-typed Python is a bliss.

ty is too permissive when I've tried it. It got better with time, and is maybe even quite ok now, but still, the ideology seems to be to allow anything it doesn't quite understand. Which just doesn't align well with my idea of what a type checker is supposed to do. Pyrefly has much better checks IME (at least with strict mode). On top of that, they have specific support for things like Polars schema checking and array shape checks also (supposedly - haven't tried the latter). Which are just incredible niceties.

And pipenv, or just literally manually creating a venv. Uv didn't invent that, it's winning on speed mainly, and then capitalising on starting with a base that includes multiple general improvements in the Python ecosystem.