Comparison with deptry?

[maresb]

Hey @zbowling, I'm curious what you're up to here. Looks very interesting!

Currently my go-to tool for this type of stuff is https://github.com/fpgmaas/deptry. Have you seen it yet?

[zbowling]

Oh neat! Never knew about that other project. Looks like it's similar but some distinct differences in approach. I found https://tweag.github.io/FawltyDeps/ before too but wanted it faster and have much better diagnostic info.

Like I want to tell when a dependency is used in certain patterns that allow it to be optional safely with some degree of confidence (it's an inexact science but certain patterns are common and I can largely detect them like checking if a dependency is imported inside "try: ... import somedep ... except ImportEror:" block or where a whole module is not guarded like that. Idea is cases where can't tell what you are doing you can add to an exclude/include list in a toml (like a pyproject.toml or separate one).

I'm super focused hardcore on speed so it can run quickly on your different envs your code runs in. Like some folks have many env setups around their code for dev, test, prod, etc but also the implicit env that gets created at install time if they are making a package with their dep (and exponential combinations of extras). I want to support requirements.txt, conda yamls, pyproject, and pixi and maybe imperative and not declarative classic setup.py without actually installing. Also nothing really supports conda well I found.

I am experimenting with having a synthetic fast pass here too make that happen because that is a lot of envs a user's code can run in and if I setup each one that code could run just to detect modules would be slow to setup to test in for complex projects. I want it to be so fast it can be run in precommit/presubmit test. This synthetic is going to be hardest backend to get working right but if I can pull it off and hammer out the big bugs it will be wicked fast.

But still want a more confident fallback where it actually fetches transitive deps into a cached isolated venv to validate modules exist that might match. Because there is the pypi and conda ecosystem it means likely driving venv/uv and conda/pixi with backends. I also want to support just interrogating inside any already setup venv if invoked within one too instead of building one.

One mode is I want to be able to run is pointing at an already packaged sdist, wheel or conda package too, more than just a flat source tree.

All these environment setups are different but the plumbing and algo to check in them (besides synthetic) is largely all the same. Just a bunch of target python code that needs to be scanned against a set of search paths to look for modules.

I'm still very early and things aren't working yet end to end but I'll post up a page on the approach once I flush out more of it and the minimal viable use case is done. When it's ready I'll post up my feature backlog and try to grind them down.