unfree changes between v0.15.0 and v0.16.0?

[blemouzy]

Hello,

After upgrading den from v0.15.0 to v0.16.0, unfree battery seems to not work anymore:

       error: Package ‘drawio-29.0.3’ in /nix/store/m5dr0w2wb9lca77ar3idab31szd6mix7-source/pkgs/by-name/dr/drawio/package.nix:132 has an unfree license (‘asl20 unfreeRedistributable’), refusing to evaluate.

       a) To temporarily allow unfree packages, you can use an environment variable
          for a single invocation of the nix tools.

            $ export NIXPKGS_ALLOW_UNFREE=1

          Note: When using `nix shell`, `nix build`, `nix develop`, etc with a flake,
                then pass `--impure` in order to allow use of environment variables.

       b) For `nixos-rebuild` you can set
         { nixpkgs.config.allowUnfree = true; }
       in configuration.nix to override this.

       Alternatively you can configure a predicate to allow specific packages:
         { nixpkgs.config.allowUnfreePredicate = pkg: builtins.elem (lib.getName pkg) [
             "drawio"
           ];
         }

       c) For `nix-env`, `nix-build`, `nix-shell` or any other Nix command you can add
         { allowUnfree = true; }
       to ~/.config/nixpkgs/config.nix.

with the following aspects:

{ den, inputs, ... }:
{
  den.aspects.apps._.diagram = {
    includes = [
      (den._.unfree [
        "drawio"
      ])
    ];

    homeManager =
      { pkgs, ... }:
      {
        home.packages = with pkgs; [
          drawio
        ];
      };
  };
}

---

# User aspect

{ __findFile, den, ... }:
{
  den.aspects.benjamin = {
    includes = [ <apps/diagram> ];
  };
}

I don't see any API change in the documentation.

It is a regression?

[sini]

After sleeping on this I think there might be a real issue here that was both intentional, and unintentional. I apologize for the regression in behavior. I haven't tested yet to verify, but what I suspect is happening is this:

We updated aspects to give them proper identity based on their declaration path and name.

De-duplication triggers on this improved identity. De-duplication is aware of perHost/perUser parametric aspects.

This didn't account for parametrized aspects which aren't fed by the context pipeline, as is the example with the unfree battery.

Tests likely didn't catch this because they only test one unfree evocation. I tried to replicate your report verbatim when it got linked to me, and my test passed. That's why I asked if your flake was public on matrix. :)

So... I think that's probably what's happening and if it is, it might be complicated to fix -- not just a one liner -- but I promise to fix it ASAP.

[sini]

I created an issue to track this: #460

[blemouzy]

I apologize for the regression in behavior.

Don't worry, I just have to stay on den v0.15.0.

We updated aspects to give them proper identity based on their declaration path and name.
De-duplication triggers on this improved identity. De-duplication is aware of perHost/perUser parametric aspects.
This didn't account for parametrized aspects which aren't fed by the context pipeline, as is the example with the unfree battery.

I don't really understand :(
Is there something I can try to confirm this hypothesis?

That's why I asked if your flake was public on matrix.

No, sorry, I still have to integrate sops-nix to hide my secrets.
I will try to do a minimal example with the issue.

[sini]

If it's what I think it is, I can reproduce myself.

Can you confirm you're using den._.unfree from multiple aspects?

[sini]

I'll try to re-explain my understanding of the issue more clearly.

For the tree of all aspects expanded via their includes, we're de-duplicating them.

So if 'desktop' includes 'zsh' and 'develop' includes 'zsh', zsh doesn't get pulled in twice.

So let's say each aspect is a function. F.

We're de-duplicating based on F. Now, for den._.unfree [ "drawio" ] -- that's not just F(), that's F(X). So if you elsewhere say den._.unfree [ "steam" ] we would incorrectly identify F, even though that's clearly a different F(Y).

[sini]

At least, that's my theory.

[blemouzy]

Can you confirm you're using den._.unfree from multiple aspects?

Yes, I have many <apps/xxx> aspects with unfree.

I will try to do a minimal example with the issue.

Please find a minimal extract of my configuration with the issue in https://github.com/blemouzy/bugs-den:

git clone https://github.com/blemouzy/bugs-den -b issue-460
nix build .#nixos-vm-test

[sini]

Thanks. When I have a fix I'll test this to verify it covers you. I /might/ have something tomorrow.

[sini]

I confirmed this is fixed on the latest commit on main.

[blemouzy]

Thanks, that indeed fixes the unfree issue!

Now I'm trying to build my configuration again and I have another regression; I will start a new discussion for it (#469).