Comment

Comment on Bonfire & Guix, a love story -- fishinthecalculator

<- View Parent

fishinthecalculator@lemmy.ml ⁨1⁩ ⁨week⁩ ago

I find Guix far better on almost every remark, in no particular order:

as you said some part of the Nix community is made of techbros (even if Guix attracts some fossbros as well)
the way governance is structured in the Nix community is brittle, just see the drama from which all the new Nix forks spawned
better documentation. The doc for Nix is scattered, the Guix manual, albeit not perfect, is much more complete
the Guile language is far clearer than Nix, also you don’t have to use it only for package recipes, you can build full applications with it
the Guix story around trustability of binaries is far better (checkout how Guix boostrapps everything), entires classes of vulnerabilities are prevented by design
the Guix UX is far better designed imho, the command line is intuitive and well documented and features are easily composable
the community is not diverging, as is the case for Nix flakes

source

Sort:hotnew top

jim3692@discuss.online ⁨1⁩ ⁨week⁩ ago
I don’t have any experience with guix, so I will not express any opinions towards that.

However, regarding NixOS:

Yes, as a person with experience in the Nix language, I can confirm it’s awful

The documentation of NixOS is a known issue, and there are currently efforts to improve it

Talking about the trustability of binaries, by doing a quick search, I read that Guix builds are reproducible. This is true for NixOS as well. All upstreamed packages must have their version and the hash of the code (or artifact), to allow validation.

The community of NixOS is opting to maintaining flakes, because:

Some applications can simply not be built following the Nix guidelines. Examples are some electron apps (like Falkor) and apps that have weird toolchains (like bubblejail)

The reviewing process takes way too long, and PRs for upstreaming are often ignored. This forces a lot of people to just PR a flake.nix to the application, or maintain their own overlays (overlays are like overriding the available packages, while flakes are more like distributing Nix code in general)
source
- fishinthecalculator@lemmy.ml ⁨1⁩ ⁨week⁩ ago
  My point on binaries was not really about reproducibility as nowadays most distro have reproducible builds: Arch, Debian, RHEL, SUSE and probably more. My point is that packages in Guix are bootstrapped from a very small binary seed, something like 357 bytes, which highly mitigates the risk of Trusting Trust attacks
  
  source
  - jim3692@discuss.online ⁨1⁩ ⁨week⁩ ago
    It’s the first time I see the concept of bootstrappability in the context of security.
    
    Is it really worth the effort?
    
    There are multiple ways to run a supply chain attack. With bootstrappability, one can be sure that the compiler is trusted, but what about the code that the compiler compiles? There was this recent attack to XZ utils, which shows that more attention is needed on the code being merged and compiled.
    
    I think that this just creates a false sense of security.
    
    Contrary to that, I had read about a BSD team (I think FreeBSD) that reviews all the code before each release. This way they have achieved ~5 RCE exploits throughout their entire history.
    
    source
    fishinthecalculator@lemmy.ml ⁨1⁩ ⁨week⁩ ago
    I think it’s worth the effort since it prevents numerous risks at the root, for sure it’s not enough. I agree that bootstrapping wouldn’t necessarily solve the XZ attack, but I think that should be solved by big tech paying FOSS maintainers enough or at all to prevent them from burning out.
    
    About the BSD experience that looks like a big amount of work but definitely worth it, I’m sure they didn’t ship many packages as Guix ships but I guess the projects have different goals and requirements.
    
    source
    unhrpetby@sh.itjust.works ⁨1⁩ ⁨week⁩ ago
    
    There was this recent attack to XZ utils, which shows that more attention is needed on the code being merged and compiled.
    
    XZ was made possible largely because there was unaudited binary data. One part as test data in the repo, and the other part within the pre-built releases. Bootstrapping everything from source would have required that these binaries had an auditable source, thus allowing public eyes to review the code and likely stopping the attack.
    
    Pulled from here:
    
    Every unauditable binary also leaves us vulnerable to compiler backdoors as described by Ken Thompson in the 1984 paper Reflections on Trusting Trust and beautifully explained by Carl Dong in his Bitcoin Build System Security talk.
    
    It is therefore equally important that we continue towards our final goal: A Full Source bootstrap; removing all unauditable binary seeds.
    
    source