aarch64-darwin segfault trying to build system

[n8henrie]

aarch64-darwin segfault trying to build system

[yorickvP]

IIRC: it's hard to detect stack overflows on darwin, the error message for that doesn't always work, so this might be an ordinary stack overflow instead of memory corruption.

[nixos-discourse]

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

https://discourse.nixos.org/t/rust-in-nix-discussion-thread/5092/21

[n8henrie]

No luck with GC_DONT_GC=1 this time

[n8henrie]

Thanks for your input, I think you're probably correct.

This shows the basics idea of my flake setup, which has been working fairly well for the last year or so:

{
  inputs = {
    nixpkgs.url = "github:nixos/nixpkgs";
    nixpkgs-stable.url = "github:nixos/nixpkgs/release-23.11";
  };

  outputs = {
    self,
    nixpkgs,
    ...
  } @ inputs: let
    inherit (nixpkgs) lib;

    hosts = {
      mymachine.system = "aarch64-darwin";
      othermachine = {
        system = "aarch64-darwin";
        nixpkgs = inputs.nixpkgs-stable;
      };
    };
  in
    lib.foldlAttrs (acc: hostname: conf: let
      nixpkgs = conf.nixpkgs or inputs.nixpkgs;
    in
      lib.recursiveUpdate acc {
        packages.${conf.system}.default = nixpkgs.legacyPackages.${conf.system}.hello;
      }) {}
    hosts;
}

(This minimal example builds, whereas mine segfaults.)

In my actually crashing flake, if I change nixpkgs = conf.nixpkgs or inputs.nixpkgs; to nixpkgs = conf.nixpkgs or self.inputs.nixpkgs;, I go from a segfault to an infinite recursion error, even though AFAICT inputs.nixpkgs and self.inputs.nixpkgs should be the same thing.

I'm also not sure why it gives me segfault or infinite recursion whereas this works. Maybe I'm barking up the wrong tree.

[n8henrie]

I noted that 2.20 seems to have some changes relevant to stack overflows; this seems to confirm the issue (with a much better error message):

$ nix run github:nixos/nix/2.20-maintenance -- build .#darwinConfigurations.natepro.config.system.build.toplevel
warning: updating lock file '/Users/n8henrie/git/nixos/flake.lock':
• Updated input 'modules':
    'path:/nix/store/4d3ygplv20ymmsx0az7yh136asmm1vr2-source/modules?lastModified=1&narHash=sha256-zPNgdnLrKUd2c0wMzH3wyivqivC5JQmAHMCEVz00ZG4%3D' (1970-01-01)
  → 'path:/nix/store/my2zyiknkk118jjblmf3gd3arcwf62hn-source/modules?lastModified=1&narHash=sha256-zPNgdnLrKUd2c0wMzH3wyivqivC5JQmAHMCEVz00ZG4%3D' (1970-01-01)
error:
       … while evaluating the attribute 'config'
         at /nix/store/8gsd9h76snq5sgcnwjip3c3p9wxbrjzm-source/lib/modules.nix:322:9:
          321|         options = checked options;
          322|         config = checked (removeAttrs config [ "_module" ]);
             |         ^
          323|         _module = checked (config._module);

       … while calling the 'seq' builtin
         at /nix/store/8gsd9h76snq5sgcnwjip3c3p9wxbrjzm-source/lib/modules.nix:322:18:
          321|         options = checked options;
          322|         config = checked (removeAttrs config [ "_module" ]);
             |                  ^
          323|         _module = checked (config._module);

       (stack trace truncated; use '--show-trace' to show the full trace)

       error: stack overflow; max-call-depth exceeded
       at /nix/store/8gsd9h76snq5sgcnwjip3c3p9wxbrjzm-source/lib/trivial.nix:438:10:
          437|     then f.__functionArgs or (functionArgs (f.__functor f))
          438|     else builtins.functionArgs f;
             |          ^
          439|

I suspect this is already covered in #8281

Thanks for pointing me in the right direction @yorickvP!