fix inputDerivation in case of __structuredAttrs

[Radvendii]

package.inputDerivation is broken when package sets __structuredAttrs = true;. You can test this with

$ nix-build --expr 'with import <nixpkgs> {}; (pkgs.hello.overrideAttrs { __structuredAttrs = true; }).inputDerivation'
this derivation will be built:
  /nix/store/5cg7cp1ym7zs9lzxvnqryj2s058m5gld-hello-2.12.1.drv
these 2 paths will be fetched (0.99 MiB download, 0.99 MiB unpacked):
  /nix/store/pa10z4ngm0g83kx9mssrqzz30s84vq7k-hello-2.12.1.tar.gz
  /nix/store/b8w73v17699k1zdnd31lvzzcp5f0rmni-version-check-hook
copying path '/nix/store/pa10z4ngm0g83kx9mssrqzz30s84vq7k-hello-2.12.1.tar.gz' from 'https://cache.nixos.org'...
copying path '/nix/store/b8w73v17699k1zdnd31lvzzcp5f0rmni-version-check-hook' from 'https://cache.nixos.org'...
building '/nix/store/5cg7cp1ym7zs9lzxvnqryj2s058m5gld-hello-2.12.1.drv'...
error: output '/nix/store/30kpdjhl0iy97z3lama80q93b6jgxh5y-hello-2.12.1' is not allowed to refer to the following paths:
         /nix/store/30kpdjhl0iy97z3lama80q93b6jgxh5y-hello-2.12.1
         /nix/store/b8w73v17699k1zdnd31lvzzcp5f0rmni-version-check-hook
         /nix/store/l9k32vj2aczxw62134j1x0dsh569jz2l-bash-5.2p37
         /nix/store/lfwfj17y9fpjb73nsj2m35rmkh587a0x-stdenv-linux
         /nix/store/pa10z4ngm0g83kx9mssrqzz30s84vq7k-hello-2.12.1.tar.gz
         /nix/store/shkw4qm9qcw5sc5n1k5jznc83ny02r39-default-builder.sh
         /nix/store/vj1c3wf9c11a0qs6p3ymfvrnsdgsdcbq-source-stdenv.sh

There is an attempt to account for this in the implementation of inputDerivation, but it's broken. The issue is that it assumes setting allowedRefrences and allowedRequisites to null will be like not setting them at all, which is false.

It does seem to work as top-level attributes (though I'm honestly not sure why this works, and I couldn't get it to work in my own derivations¹), but in outputChecks.out.allowedReferences, it definitely doesn't work.

In any case, Rather than setting allowedReferences, disallowedReferences, allowedRequisites and disallowedRequisites to values that we hope mimic the default behaviour, we can remove those attributes to get the default behaviour.

Related issues:

• flake: Bump nixpkgs to 25.11 nix#14688
• Parse allowed* = null properly in structuredAttrs  nix#14750

Things done

• Built on platform:
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• Tested, as applicable:
  • NixOS tests in nixos/tests.
  • Package tests at passthru.tests.
  • Tests in lib/tests or pkgs/test for functions and "core" functionality.
• Ran nixpkgs-review on this PR. See nixpkgs-review usage.
• Tested basic functionality of all binary files, usually in ./result/bin/.
• Nixpkgs Release Notes
  • Package update: when the change is major or breaking.
• NixOS Release Notes
  • Module addition: when adding a new NixOS module.
  • Module update: when the change is significant.
• Fits CONTRIBUTING.md, pkgs/README.md, maintainers/README.md and other READMEs.

---

Add a 👍 reaction to pull requests you find important.

Footnotes

1. It works currently because __ignoreNulls = true; is added. However, this only seems to work for top-level attributes, not outputChecks.out.foo. I still think getting rid of the attributes entirely is the right way forward. ↩

[Radvendii]

It would also be nice if this could get back-ported to 25.11.

Since 25.11, separateDebugInfo = true; and setting {dis,}allowed{References,Requisites} requires __structuredAttrs = true;.

Which makes this bug much more likely to happen. (or at least, that's why we ran into it in nix)

[wolfgangwalther]

Dupliate of #463882 ?

[Radvendii]

Ah good call. I prefer my solution, but I would take the rename and test from the other.

@infinisil What do you think?

[Ericson2314]

Let's get the tests from that other PR (and add more of them), too.

[Radvendii]

@Ericson2314 This time I intentionally reintroduced the typo temporarily to make sure the tests I added would catch it.

[Radvendii]

Is it normal for CI to be stuck in queue for a whole day? I don't mind necessarily, I just want to make sure there's nothing I should be doing to push this forward.

[Radvendii]

Okay, I tested this locally using nix-build -A tests.stdenv, and there were indeed problems with the tests. So I fixed those problems, and tested again. These tests catch it if we misspell or remove one of the outputCheckAttrs.

And they pass with this PR as it currently is.

[infinisil]

@infinisil What do you think?

I don't mind as long as it works, thanks for the additional tests, I'll close mine!

[nixpkgs-ci]

Successfully created backport PR for release-25.11:

• [Backport release-25.11] fix inputDerivation in case of __structuredAttrs #470487