wegel / 2nex

2nex

Full source-bootstrappable

I'm doing a (free) reproducible build system/distro (just a hobby, won't be big and professional like Guix or Nix) for Linux. This has been brewing for a few months, and is starting to get ready.

Jokes aside, I love the idea of an immutable, reproducible OS, but I find that I don't like how the build scripts are implemented in "strange" programming languages in Nix (the Nix Language) and Guix (Scheme, which is Turing-complete; not something you necessarily want for a build script language), and I don't like the complexity of the configuration languages they are using, although they are incredibly powerful (too much?). Note that I LOVE functional programming languages, but I don't think build scripts and configuration are where they're most needed.

I also don't understand why exactly Guix and Nix are searching for the Holy Full Source Bootstrap, because we already have it. Maybe they don't have it, but it's there. Guix says:

If you run guix pull today, you get a package graph of more than 22,000 nodes rooted in a 357-byte program; something that had never been achieved, to our knowledge, since the birth of Unix.

The 22,000 nodes is very impressive (and the >80,000 packages of Nix too), but the 357-byte root program is, to me, quite useless. They also need a ~25MB binary package to bootstrap the whole thing. Yes, the 357-byte program thing is VERY cool. But it's not necessary. Remember, the goal is to achieve reproducibility, and ultimately, you still have to build the "normal" GNU toolchain to be able to compile most software. They even say it in the "Full-Source Bootstrap" doc:

hex0 (the 357-byte root program) builds hex1 and then on to M0, hex2, M1, mescc-tools and finally M2-Planet. Then, using mescc-tools, M2-Planet we build Mes. From here on starts the more traditional C-based bootstrap of the GNU System.

Why would we take this particular route to achieve reproducibility exactly? What we need, it seems to me, is a reproducible GNU toolchain, and once we have that, we only have to make sure that everything else that's built from there is also reproducible. Sounds simple enough to me, especially if the build system actually enforces reproducibility.

So I present 2nex. Currently, it's only a proof-of-concept sandbox to make a fully reproducible, fully source-bootstrappable, no fixed binary build system (eg, any x64 system with a gcc toolchain should work), but eventually I'd like to build on that to make a full distro for multiple architectures, and I also have some ideas for the configuration part of the OS (e.g., what's handled by the Nix language in NixOS).

How it works

Packages are described in a yaml Manifest (see manifests). The Manifest lists the sources, dependencies (in the form of other packages), and build instructions. The Manifest includes the checksum of the result of the build; we thus can know that the result of the build is what we expect, and is reproducible. The builder is used to build the Manifest, and the bundles are then commited to an ostree repository.

The builder

The builder is a currently a simple (and pretty badly written) rust program that reads the Manifest, builds and verify it, and then commits the result to an ostree repository. It makes sure that the build environment itself is controlled and reproducible, and that the build is reproducible.

Status

This is still at the very early proof of concept stage. The bootstrap toolchain build is reproducible on my machine and on a VM; I'm sure there are many corner cases that need to be addressed to make it full reproducible in all cases.