This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/why-isnt-config-nixpkgs-config-nix-evaluated-for-nix-build/9579/4

Isn't this more about flake than unstable?

As for me it works using non-flake syntax, though fails when using flake syntax:

$ NIXPKGS_ALLOW_UNFREE=1 nix build -f '<nixpkgs>' steam
# builds successfully
$ NIXPKGS_ALLOW_UNFREE=1 nix build nixpkgs#steam
# does not
$ NIXPKGS_ALLOW_UNFREE=1 nix build --impure nixpkgs#steam
# does build again

I don't think nixpkgs should point flakes users to --impure as this breaks caching. It is convenient for sure, but it isn't a proper solution.

Instead you may consider defining a flake that is nixpkgs which is configured to allow unfree packages. This isn't ideal, but well, there is a good reason why nixUnstable is in fact unstable :).

{
  inputs.nixpkgs.url = "https://nixos.org/channels/nixos-20.09/nixexprs.tar.xz";
  outputs = { self, nixpkgs }: {
    defaultPackage.x86_64-linux = import nixpkgs {
      system = "x86_64-linux";
      config.allowUnfree = true;
    };
  };
}

It's also possible to avoid flakes entirely which avoids the issue.

{
  inputs.nixpkgs.url = "https://nixos.org/channels/nixos-20.09/nixexprs.tar.xz";
  outputs = { self, nixpkgs }: {
    defaultPackage.x86_64-linux = import nixpkgs {
      system = "x86_64-linux";
      config.allowUnfree = true;
    };
  };
}

This is indeed worth knowing, but I think it's too much information to put on a trace message like here. However, regarding:

I don't think nixpkgs should point flakes users to --impure as this breaks caching. It is convenient for sure, but it isn't a proper solution.

It depends on what you define as "proper". The use case I had in mind, (which was not specific enough indeed) is testing updates via e.g nix build .#zoom-us or nix profile install nixpkgs#zoom-us. I've adjusted the trace message according to your recommendation - it recommends not using --impure and it mentions it's an issue specific to # flakes syntax.

It depends on what you define as "proper". The use case I had in mind, (which was not specific enough indeed) is testing updates via e.g nix build .#zoom-us or nix profile install nixpkgs#zoom-us. I've adjusted the trace message according to your recommendation - it recommends not using --impure and it mentions it's an issue specific to # flakes syntax.

You can easily avoid flakes by using nix build -f . zoom-us instead. Flakes are experimental and I would say they should be avoided for now.

You can easily avoid flakes by using nix build -f . zoom-us instead. Flakes are experimental and I would say they should be avoided for now.

Hmm I see, but what about nix profile? It won't accept -f. so it seems.

Just because something is experimental doesn't mean we can't at least explain better a tiny detail to users of this experimental feature. We've had contributions before to nixos-rebuild which included flakes support, and this is what helps adopting such an experimental feature.

Besides, what is the worst thing that could happen? If at some point the flakes syntax would change, we'd only have to slightly update that message, and the price is 0 rebuilds, just as it is today. Please note @xfix that nothing in the new message's text recommends using flakes.

Providing instructions for flakes makes the error message more noisy for users who don't use flakes.

However, I think an improvement is still possible. It's possible to recognize whether an user is using flakes or not and provide a different message depending on that.

That being said, I think there should be a way to pass configuration options into Nix flakes, it's just that it wasn't implemented yet. Worth noting that this isn't --impure however, as --impure does way too much.

However, I think an improvement is still possible. It's possible to recognize whether an user is using flakes or not and provide a different message depending on that.

Indeed that would be ideal, though I'm not sure how would I do that - should it be a check of the current nix.package.version? Is that possible in the scope of that meta?

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/using-non-free-libraries-in-flakes/8632/15

@doronbehar sorry for bumping something from over a year ago but I managed to find my way here mostly because of your comments on discourse (in the above mentioned and others). Are there any updates on this? I know things might be a bit wild in your neck of the woods right about now.

@doronbehar sorry for bumping something from over a year ago but I managed to find my way here mostly because of your comments on discourse (in the above mentioned and others). Are there any updates on this?

Not really, the last suggestion by @xfix was to make that eval message different for nixUnstable and nixStable users, but (now) I think it's not possible to do it - nixpkgs on it's own doesn't know what nix version is used. Hence my opinion is that to give "unstable" nix more respect, this PR is one of the things we could improve.

I know things might be a bit wild in your neck of the woods right about now.

Thanks a lot for your concern ☺️. It's not such a big deal in the area where I live in.

flakes is about improving purity so recommending --impure loooks like the opposite of what we want. Reimporting nixpkgs is s uperior solution IMO. Not sure how to communicate that, point at the nixflake wiki page with this solution described ?

flakes is about improving purity so recommending --impure loooks like the opposite of what we want. Reimporting nixpkgs is s uperior solution IMO. Not sure how to communicate that, point at the nixflake wiki page with this solution described ?

Yea the problem is that it's impossible for that message to know whether it is being evaluated because someone imported Nixpkgs or because someone tried to evaluate a package from Nixpkgs' repo. In the first case I totally agree it'd be a bad advice. I'm not sure there's something we can do.. I'll close this.