search

NixOS / ofborg

@ofborg tooling automation https://monitoring.ofborg.org/dashboard/db/ofborg
https://ofborg.org
MIT License
232 stars 166 forks source link

Allow unfree redistributable builds? #687

Open roberth opened 1 month ago

roberth commented 1 month ago

Currently software such as vault is not built by ofborg, and I don't really understand why. bsl11 may not be open source, but it is source-available and redistributable = true;.

I could see an ideological argument here, but I don't think that should an excuse for being unhelpful to Nixpkgs maintainers. So I'd like to ask, is this limitation intentional?

SomeoneSerge commented 3 weeks ago

Preceding conversation:

A relevant conversation is https://github.com/NixOS/nixpkgs/issues/83884, in particular the suggestion to introduce a "curated [white]list" of licenses instead of unfreeRedistributable (also an example of implementation from Gentoo)

As to my opinion, I think the solution is hierarchical:

  1. We should evaluate everything (unfree, insecure, cross) in order to compensate for the lack of static typing. IMO this we should start doing now, maybe in a concurrent job. The only excuse not to do this is Nix's performance, and it's a bad excuse because developers' time is more expensive than the compute
  2. We should build the tests for whitelisted unfree licenses, but we should abstain from distributing the closures to the "user"
  3. For a subset of white-listed licenses, we should "build" and cache the FODs (e.g. the singular contention point about the CUDA EULA is whether "patchelf"-ing substitutes a "modification"; there's seemingly no reason, however, not to keep the original tarballs from disappearing)

A tangential point is that we seem to lack tools to plug a third party Ofborg-like CI in and out of the Nixpkgs forge...