Is this a known regression? Wasn't too successful at minimizing, maybe somebody could give it a shot:

nix eval "https://api.flakehub.com/f/pinned/HariAmoor-professional/nixos-config/0.1.15%2Brev-9663eb776b745a0eb637320c165729bf6acaac70/018a8b2d-121a-7d71-8db4-666a0ec6c497/source.tar.gz#nixosConfigurations.nixos.config.system.build.toplevel" --no-allow-import-from-derivation --no-eval-cache

Fails on 3ad1af8 with:

error: makeWrapper/makeShellWrapper must be in nativeBuildInputs

       note: trace involved the following derivations:
       derivation 'home-manager-generation'
       derivation 'home-manager-files'
       derivation 'home-manager-path'
       derivation 'bitwarden-x86_64-unknown-linux-gnu-2023.5.1'
       derivation 'npm-install-hook'
       derivation 'make-shell-wrapper-hook'

Works on 2.93. Sorry for the flakehub repro, don't have a lot of energy atm to keep minimizing this. I think a tell-tale sign is the fact that the derivation name is prefixed (x86_64-unknown-linux-gnu) as if nixpkgs is cross-compiled to gnu64, but it's not. raitobezarius this must have to do with the thunk equality changes?

2.93:

nix-repl> nixosConfigurations.nixos.config.nixpkgs.hostPlatform == nixosConfigurations.nixos.config.nixpkgs.buildPlatform
true

main:

nix-repl> nixosConfigurations.nixos.config.nixpkgs.hostPlatform == nixosConfigurations.nixos.config.nixpkgs.buildPlatform
false

So it indeed thinks there's cross now where there wasn't before.

Slightly reduced reproducer:

let
  inherit (import <nixpkgs> { }) lib;
in
(lib.evalModules {
  modules = [
    (
      { config, ... }:
      {
        options = {
          buildPlatform = lib.mkOption {
            type = lib.types.either lib.types.str lib.types.attrs;
            apply = lib.systems.elaborate;
            default = config.hostPlatform;
          };
          hostPlatform = lib.mkOption {
            type = lib.types.either lib.types.str lib.types.attrs;
            apply = lib.systems.elaborate;
            default = "x86_64-linux";
          };
          isCross = lib.mkOption { type = lib.types.bool; };
        };
        config = {
          isCross = config.buildPlatform == config.hostPlatform;
        };
      }
    )
  ];
}).config.isCross

Fully reduced reproducer:

with rec {
  a = {
    f = x: x;
    meow = true;
  };
  b = a // {
    meow = true;
  };
};
a == b

What's even more cursed is this:

nix::EvalState::eqValues (this=0x555555e8b770, v1=..., v2=..., pos=..., errorCtx="while testing two values for equality") at lix/libexpr/eval.cc:2613
2613        forceValue(v2, pos);
(gdb)
2616        if (v1.type() == nInt && v2.type() == nFloat) {
(gdb)
2619        if (v1.type() == nFloat && v2.type() == nInt) {
(gdb)
2624        if (v1.type() != v2.type()) return false;
(gdb)
2629        auto pointerEq = [&] { return v1.pointerEqProxy() == v2.pointerEqProxy(); };
(gdb)
2631        switch (v1.type()) {
(gdb)
2658                if (pointerEq()) return true;
(gdb)
2661                if (isDerivation(v1) && isDerivation(v2)) {
(gdb)
2669                if (v1.attrs()->size() != v2.attrs()->size()) return false;
(gdb) p v1.attrs()->size()
$1 = 2
(gdb) p v2.attrs()->size()
$2 = 3

It's not equal because the attrset size is incorrect? How???

In reply to @raitobezarius:matrix.org
i want to go and say that zomg the fact that we broke eval based on this behavior :')

In reply to @xokdvium:matrix.org
Updates never replace anything - they copy to a new attribute set though. It's puzzling

Yeah so doing this does help:

diff --git a/lix/libexpr/eval.cc b/lix/libexpr/eval.cc
index f1ba9e01c..5a5acd6aa 100644
--- a/lix/libexpr/eval.cc
+++ b/lix/libexpr/eval.cc
@@ -2682,6 +2682,7 @@ bool EvalState::eqValues(Value & v1, Value & v2, const PosIdx pos, std::string_v
 
         /* Functions are incomparable. */
         case nFunction:
+           if (pointerEq()) return true;
             return false;
 
         case nExternal:

But I'm not sure about all the consequences :)