> The thing I dislike about NixOS the most is ... nix. But I guess that is hard to change now. Every good idea to be ruined via horrible jokes of an underperforming programming language ...

I don’t get the hate for Nix, honestly. (I don’t get the complaints that it’s difficult, either, but I’m guessing you’re not making one here. I do get the complaint that the standard library is a joke, but you’re not making that one either that I can see.) The derivation and flake stuff excepted, Nix is essentially the minimal way to add lazy functions to JSON, plus a couple of syntax tweaks. The only performance-related thing you could vary here is the laziness, and it’s essential to the design of Nixpkgs and especially NixOS (the only config generator I know that doesn’t suck).

I’ll grant that the application of Nix to Nixpkgs is not in any reasonable sense fast, but it looks like a large part of that is fairly inherent to the problem: you’ve got a humongous blob of code that you’re going to (lazily and in part) evaluate once. That’s not really something typical dynamic-language optimization techniques excels at, whatever the language.

There’s still probably at least an order of magnitude to be had compared to mainline Nix the implementation, like in every codebase that hasn’t undergone a concerted effort to not lose performance for stupid reasons, but there isn’t much I can find to blame Nix the language for.