|
|
Log in / Subscribe / Register

PyTorch and the PyPI supply chain

PyTorch and the PyPI supply chain

Posted Jan 12, 2023 17:10 UTC (Thu) by mathstuf (subscriber, #69389)
In reply to: PyTorch and the PyPI supply chain by summentier
Parent article: PyTorch and the PyPI supply chain

> But trying to coerce setuptools to do what you want it to do is not fun. Its code is an undocumented mess, its abstractions are leaky and incoherent, and its architecture like a Jenga tower resting on top of a pile of Mikado sticks. Look at nontrivial setup scripts bundled (e.g., for project such as numpy or tensorflow): they resemble ancient incantations much more than actual code.

I think at one point, NumPy's additions to setuptools were on the same order of size as setuptools itself. SciPy probably didn't make things any easier.

I will agree about the undocumented mess 100% though. Figuring out what could go into some fields (globs, symlink traversal, etc.) involved tracing the value(s) through the code to where they hit some active API that actually used them. The duck typing helps with being able to get things done by abusing things like `../` traversal to grab things, but really hinders with making anyone aware of what is possible (and what of that is actually intended).


to post comments


Copyright © 2026, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds