Hacker News new | ask | show | jobs
by Imustaskforhelp 153 days ago
Yes I believe the same too and I think we are on the same goal. I think that I can probably patch this code to install uv, let's say locally instead of globally if that's a major concern. I feel like its not that hard.
1 comments
quotemstr 153 days ago
It's easy enough to patch. It's the philosophy that bugs me. We already have a huge problem with routine workflows pulling things from the network (often, without even a semblance of hash-locking) and foregoing the traditional separation between environment setup and business logic. There's a lot of value into having discrete steps for downloading/installing stuff and doing development, because then you can pay special attention to the former, look for anything odd, read release notes, and so on. Between explicit, human-solicited upgrades, dev workflows should be using, ideally, vendored dependencies, or, if not that, then at least stuff that's hash-verified end-to-end.

Someday, someone is going to have a really big disaster that comes out of casual getting unauthenticated stuff from somebody else's computer.

link
Imustaskforhelp 153 days ago
I agree with you and you raise some good points

I think your reason of worrying is either that A) packages can update and contain malware or B) Uv's installation itself. might have malware if any of A) or B) get hacked

Regarding A) I feel like uv's dependencies can be pinned to a certain date to make them reproducible and this can come of help (https://docs.astral.sh/uv/guides/scripts/#improving-reproduc...)

Regarding B) I feel like they provide attestations via GitHub Artifact Attestations and the script could once again be modified to actually verify it via github attestations and they also provide ghcr artifacts (as such immutability) atleast of docker images and I looked further into it and it seems that you can use github artifacts to upload normal binary files as well so I will probably take a look into seeing if I can do something like this for uv's ghcr

Effectively after A) and B) the trust just ends up being reliant on Github's Microsoft infrastructure (usually) and perhaps python infrastructure which is on fastly

But I feel like this is for cases of extremely sensitive workflows But I feel like I might still take a look at it because security still feels very interesting to me and just because of this discussion, I can see at some pointers of following up on curiosity lol

Anyways would love to continue our discussion and probably update you on trying to make a script which could actually be completely pinned (atleast uv binary instead of just running a shell script from the astral servers in such case)

link