Nix flakes are a really fantastic way to finely manage project dependencies. However, writing Nix expressions to specify dependencies is subtle and requires learning a whole new language. Can we do better?
Turns out yes. Demo:
The goal is to write an imperative CLI to automate the generation of the declarative dependencies specified in a flake.
Every time I start a project, I find myself copying a flake that works, and simply tweaking the name and list of buildInputs. The goal is to streamline that process and dodge writing any boilerplate code.
nix flake init
?This is meant to be more powerful. Templates generate boilerplate code, but without interaction. Interaction is key. The end user should be able to get their project up and running with minimal effort expended towards figuring out how to get nix to play nicely with their package and set of dependencies.
The idea is to have the user specify the type of dependencies,
flake inputs and outputs, and type of the package with skim
,
then to output a flake. This flake is then validated with rnix
.
If the user wants to modify an existing flake to add or remove
dependencies, this will also be possible. The flake shall be
parsed in with rnix
, and the user will be able to modify it.
As of now, basically none of the features exist. I've only got the proof of concept working: skim can be used for a cli and rnix can be used to modify the AST.
Further down the line, I'd like to make this even more interactive. This will involve querying github, crates.io, pypy, nixpkgs and more for packages, then piping them into skim for selection based on language.
The hope is to also provide automatic support for pre-existing nix expression generators such as node2nix, poetry2nix, cabal2nix, and naersk.
I'm using the rnix
parser to generate nix expressions,
and the skim
fuzzy finder for the cli.