nix flake show: Support meta attribute for apps

[shivaraj-bh]

Motivation

Metadata information for flake apps will be useful while exploring a
flake using nix flake show

Context

The outputs that follow, uses the flake:

{
  inputs.nixpkgs.url = "github:nixos/nixpkgs/nixpkgs-unstable";
  outputs = { nixpkgs, ... }: {
    apps.aarch64-darwin.hello = {
      program = "${nixpkgs.lib.getExe nixpkgs.legacyPackages.aarch64-darwin.hello}";
      type = "app";
      meta.description = "I say hello!";
    };
  };
}

Output without the description:

❯ nix flake show
path:/Users/shivaraj/oss/tmp?lastModified=1723627908&narHash=sha256-D89aZ1hTqNCzvUIHLD6meJVNiApUoGvyMzEiowdf/dU%3D
└───apps
    └───aarch64-darwin
        └───hello: app

Output with the description:

❯ ../nix/outputs/out/bin/nix flake show
path:/Users/shivaraj/oss/tmp?lastModified=1723627908&narHash=sha256-D89aZ1hTqNCzvUIHLD6meJVNiApUoGvyMzEiowdf/dU%3D
└───apps
    └───aarch64-darwin
        └───hello: app: I say hello!

The description is also included in the json output of nix flake show:

{
  "apps": {
    "aarch64-darwin": {
      "hello": {
        "description": "I say hello!",
        "type": "app"
      }
    }
  }
}

Priorities and Process

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

[roberth]

It'd be lovely to have

• a characterization test for nix flake check
• warning for unknown attribute in app

but this is already a good change.

[shivaraj-bh]

• a characterization test for nix flake check
• warning for unknown attribute in app

I can add them. Do you suggest I do this here or in a different PR?

[roberth]

Awesome. Should be fine to do here in new commits.

• a characterization test for nix flake check

Grep diffAndAcceptInner for examples

• warning for unknown attribute in app

I think we already have some checks like that, so you could adapt it to work the same.

[kjeremy]

Is this a dupe of #10980

[roberth]

@kjeremy It's complementary, because this is about apps as opposed to packages and other "derivations".

Partially it kind of should be a duplicate because this PR doesn't solve the terminal layout issue. We could merge either one first and then apply your rendering solution to apps, but I'd slightly prefer to merge your PR first.

[roberth]

A question was raised on matrix whether this should be meta.description.
apps aren't derivations (or package attrsets), so it's not necessary to separate this into meta, but perhaps it'd be good for consistency nonetheless?

[shivaraj-bh]

nix flake check now throws an error when an unknown attribute is present:

❯ ../nix/outputs/out/bin/nix flake check
error:
       … while checking flake output 'apps'
         at /nix/store/n8v51gyfjrhjrfbabxwy0hh4q68q2v7v-source/flake.nix:5:6:
            4|      # formatter.aarch64-darwin = "foo";
            5|      apps.aarch64-darwin.hello = {
             |      ^
            6|       program = "${nixpkgs.lib.getExe nixpkgs.legacyPackages.aarch64-darwin.hello}";

       … while checking the app definition 'apps.aarch64-darwin.hello'
         at /nix/store/n8v51gyfjrhjrfbabxwy0hh4q68q2v7v-source/flake.nix:5:6:
            4|      # formatter.aarch64-darwin = "foo";
            5|      apps.aarch64-darwin.hello = {
             |      ^
            6|       program = "${nixpkgs.lib.getExe nixpkgs.legacyPackages.aarch64-darwin.hello}";

       error: app 'apps.aarch64-darwin.hello' has unsupported attribute 'unknown-attr'

It also throws error when type or program attributes are missing. For meta you get a warning:

❯ ../nix/outputs/out/bin/nix flake check
warning: app 'apps.aarch64-darwin.hello' lacks attribute 'meta'

[shivaraj-bh]

• nix run doesn’t yet report a warning when it detects an unknown attribute, should it? Also, should nix flake check throw this as an error or just a warning?
• nix flake check for apps doesn’t resolve the app, like it is done when nix fmt or nix run are called, thus we don’t check if the program is actually present in the store. (Side note: I did try doing sanity on apps by resolving them, but couldn’t find a way to do so, I will have to try again if that’s what is desired)

[edolstra]

I guess this should be meta.description?

However I'm not sure we need the meta prefix. It's necessary for derivations to distinguish derivation input attributes from metadata, but that's not the case for apps.

[edolstra]

Though I suppose meta is useful for distinguishing between attributes that should be an error if they're unknown, and attributes that should at most be a warning.

[roberth]

I think that's a good argument.
Also users are likely to write meta because they learned it for packages, and it's nice that x.meta.description or "" works for both, if that's going to be relevant.

[roberth]

If you could add the warning and sync the docs, this looks good to me 🚀

[shivaraj-bh]

If you could add the warning and sync the docs

Done!