Call package vs import in Nix

Note

Today i'm going to briefly talk about two ways to call nix functions that return derivations (which is the usual case in default.nix files).

Let's say we have the following package in some-package.nix (some random haskell package):

{pkgs ? import <nixpkgs> {}, ...}:

let
  somePackage = pkgs.haskellPackages.callCabal2nix "some-package" ./. { };
in
pkgs.stdenv.mkDerivation {
  name = "some-package";
  src = ./.;
  installPhase = '''
    mkdir -p $out/bin

    makeWrapper ${somePackage}/bin/some-package $out/bin/some-package \
      --prefix PATH : ${pkgs.lib.makeBinPath [ pkgs.nix ]}
  ''';
}

And then we want to import this package somewhere else. We also want to make sure that it is built with a version of nixpkgs that we've defined, so we can have maximum reproducibility.

Here's two ways to do it:

pkgs.callPackage ./some-package.nix { };

And another

import ./some-package.nix { inherit pkgs; } # our custom pkgs

So callPackage will automatically pass in the pkgs it's called with to some-package, while import will use either the default value in some-package.nix (or error out if there is no default value) or the nixpkgs that is passed explicitly. This is useful when you want to explicitly control the version, while calling callPackage with another version of nixpkgs for example.

callPackage is the more popular option, and it's pretty handy as it passes arguments to derivation functions automatically, but it can be a source of confusion as to which nixpkgs is passed to the function. When you need more explicitness, use import.

logo