distinguish between defined & used schemas

[sellout]

As @edolstra says here

flake-schemas/flake.nix

Line 214 in 7649320

# FIXME: distinguish between available and active schemas?

it would be great to be able to define schemas distinct from the set of schemas used by the flake. Partially because you might not want to use all the schemas you define, but more significantly, you don’t want to re-export the upstream schemas you’re using, as then schemas = nix-darwin.schemas // home-manager.schemas; would incidentally provide you with the “base” schemas, but not necessarily the versions of them that you expected. As schemas get more widely adopted, you’re forcing the user to topo-sort the inputs that provide schemas. E.g., schemas = nix-darwin.schemas // home-manager.schemas // flake-schemas.schemas; does the right thing, but schemas = flake-schemas.schemas // nix-darwin.schemas // home-manager.schemas; does not.

My proposal would be to have the existing outputs.schemas be the schema definitions – these are schemas that are output by the flake. Then a top-level attribute, schemas would be the schemas used by the flake.

The complication of this approach is that you need to apply schemas not just to the outputs, but to the top-level attrset. The current flake schemas become “output schemas”, and the schemas proposed here would be able to affect elements outside of outputs.

Another use case for this generalization is options as in @lheckemann’s NixCon talk¹.

Here’s an example flake trying to show how I think this might look/work.

{
  description = "schema example";

  inputs.flake-schemas.url = "github:DeterminateSystems/flake-schemas";

  # This takes `inputs` (like `outputs` does), but it also takes `outputs` to be able to find locally-defined schemas.
  schemas = {flake-schemas, ...}: outputs: {
    # get the schemas for the top-level attributes we use
    inherit (flake-schemas.schemas) description inputs schemas;
    outputs = {
      # get the schemas for the `outputs` attributes we use
      inherit (flake-schemas.schemas.outputs) overlays packages schemas;
      inherit (outputs.schemas.outputs) myNewOutputs;
      myPrivateOutputs = {
        version = 1;
        doc = "An output whose schema isn’t available elsewhere.";
        inventory = flake-schemas.lib.derivationsInventory "myPrivateOutputs" false;
      };
    };
  };

  outputs = {flake-schemas, ...}: {
    overlays.default = final: prev: {…};
    packages = {…};
    schemas.outputs.myNewOutputs = {
      version = 1;
      doc = "An output whose schema is published by this flake.";
      inventory = flake-schemas.lib.derivationsInventory "myNewOutputs" false;
    };
  };
}

The top-level schemas attribute in this case could be simplified to something like the following, if you don’t want to restrict to exactly the schemas you’re using.

  schemas = flake-schemas.schemas // {
    outputs = flake-schemas.schemas.outputs // outputs.schemas.outputs // {
      myPrivateOutputs = {
        version = 1;
        doc = "An output whose schema isn’t available elsewhere.";
        inventory = flake-schemas.lib.derivationsInventory "myPrivateOutputs" false;
      };
    };
  };

Footnotes

1. I would suggest that @lheckemann’s proposal be split similarly to this one – an outputs.options to define options and a top-level options for ones available to the flake. This would also eliminate the need for the manual namespacing (like "org.nixos.nixpgs.config"), since normal flake references would serve the same purpose. ↩