eachDefaultSystem docs are sorely missing a real example

[colemickens]

I'm very unconvinced about using a flake import to override the systems that "eachDefaultSystem" is going to build for. I love flakes but seeing inputs used for this seems confusing to me.

Further, I've been looking at/for examples, etc, and have no idea how to get jj (https://github.com/martinvonz/jj) to support riscv64-linux.

[colemickens]

This pattern is throwing up red flags for me. Least of all, overriding a transitive dependency with a path doesn't seem to work.

I have:

{
 systems = { url = "path:./flake.systems.nix"; flake = false; };
 jj = { url = "github:martinvonz/jj"; inputs."flake-utils".inputs."systems".follows = "system"; };
}

and I can see from reading the lock file that the override is not working.

[colemickens]

I guess I can override flake-utils, with it's system overridden, and then pass that to jj as an override, but ... this is going to get hairy fast. So this sorta works:

{
    systems = { url = "path:./flake.systems.nix"; flake = false; };
    flake-utils = { url = "github:numtide/flake-utils"; inputs."systems".follows = "systems"; };
    jj = { url = "github:martinvonz/jj"; inputs."flake-utils".follows = "flake-utils"; };
}

Now I have multiple (enough, really) flake-utils through my flake.lock, but now I also have this version that is slightly different. Now I have to decide if I override every potential flake input that consume flake-utils.

Also, a question -- should I have been able to override that dependency transitively?

[zimbatm]

I think your last example is the right way to do it, given the current design constraints of flakes.

The best is to flatten the inputs as much as possible so there is only one version of each dependency.

[Moise225]

latest.json

[malik-n]

I am also looking for another example. Trying to simplify my flake.nix with eachDefaultSystem but turning it unusable.

Is there any example flake.nix that defines nixosConfigurations and formatter? Using treefmt maybe

[Buttars]

It's not clear at all why the former doesn't produce the latter.

formatter = inputs.flake-utils.lib.eachDefaultSystem (
  system: nixpkgs.legacyPackages.${system}.nixfmt-rfc-style
);

formatter = {
  x86_64-linux = nixpkgs.legacyPackages.x86_64-linux.nixfmt-rfc-style;
  ...
};

UPDATE: I find myself back here after a couple months. The implementation of these utility functions requires wrapping the assertion and values to get the desired output. This is not great for readability. I'd like to think everyone would prefer to set the property equal to eachDefaultSystem or similar and construct the value from within the function. I fully expected eachDefaultSystem to return a set of systems as root attributes and the computed values as the children like demonstrated in my comment above.

If I wanted to get the outcome I expected I'd have to do this:

inputs.flake-utils.lib.eachDefaultSystem (
  system: { formatter = nixpkgs.legacyPackages.${system}.nixfmt-rfc-style }
);

I wouldn't mind opening a PR to add another method but I'm not sure this project is receiving updates any longer.

Working example of expected output using defaultSystems and genAttrs without using eachDefaultSystem.

 formatter = nixpkgs.lib.genAttrs inputs.flake-utils.lib.defaultSystems (system:
   nixpkgs.legacyPackages.${system}.nixfmt-rfc-style
 );