`pkgs.callPackage` is used to parameterize the construction of Nix Derivation. To understand its purpose, let's first consider how we would define a Nix package (also known as a Derivation) without using `pkgs.callPackage`.
While the definition of this Derivation is quite concise, most Derivations in nixpkgs are much more complex. In previous sections, we introduced and extensively used the `import xxx.nix` method to import Nix expressions from other Nix files, which can enhance code maintainability.
1. To enhance maintainability, you can store the definition of the Derivation in a separate file, e.g., `hello.nix`.
1. However, the context within `hello.nix` itself doesn't include the `pkgs` variable, so you'll need to modify its content to pass `pkgs` as a parameter to `hello.nix`.
2. In places where you need to use this Derivation, you can import it using `import ./hello.nix pkgs` and use `pkgs` as a parameter to execute the function defined within.
In the previous example without `pkgs.callPackage`, we directly passed `pkgs` as a parameter to `hello.nix`. However, this approach has some drawbacks:
1. All other dependencies of the `hello` Derivation are tightly coupled with `pkgs`.
1. If we need custom dependencies, we have to modify either `pkgs` or the content of `hello.nix`, which can be cumbersome.
2. In cases where `hello.nix` becomes complex, it's challenging to determine which Derivations from `pkgs` it relies on, making it difficult to analyze the dependencies between Derivations.
`pkgs.callPackage`, as a tool for parameterizing the construction of Derivations, addresses these issues. Let's take a look at its source code and comments [nixpkgs/lib/customisation.nix#L101-L121](https://github.com/NixOS/nixpkgs/blob/fe138d3/lib/customisation.nix#L101-L121):
In essence, `pkgs.callPackage` is used as `pkgs.callPackage fn args`, where `fn` is a Nix file or function, and `args` is an attribute set. Here's how it works:
1.`pkgs.callPackge fn args` first checks if `fn` is a function or a file. If it's a file, it imports the function defined within.
1. After this step, you have a function, typically with parameters like `lib`, `stdenv`, `fetchurl`, and possibly some custom parameters.
2. Next, `pkgs.callPackge fn args` merges `args` with the `pkgs` attribute set. If there are conflicts, the parameters in `args` will override those in `pkgs`.
3. Then, `pkgs.callPackge fn args` extracts the parameters of the `fn` function from the merged attribute set and uses them to execute the function.
4. The result of the function execution is a Derivation, which is a Nix package.
What can a Nix file or function, used as an argument to `pkgs.callPackge`, look like? You can examine examples we've used before in [Nixpkgs's Advanced Usage - Introduction](./intro.md): `hello.nix`, `fcitx5-rime.nix`, `vscode/with-extensions.nix`, and `firefox/common.nix`. All of them can be imported using `pkgs.callPackage`.
For instance, if you've defined a custom NixOS kernel configuration in `kernel.nix` and made the development branch name and kernel source code configurable:
As shown above, by using `pkgs.callPackage`, you can pass different `src` and `boardName` values to the `kernel.nix` function to generate different kernel packages. This allows you to adapt the same `kernel.nix` to different kernel source code and development boards.
The advantages of `pkgs.callPackage` are:
1. Derivation definitions are parameterized, and all dependencies of the Derivation are the function parameters in its definition. This makes it easy to analyze dependencies between Derivations.
2. All dependencies and other custom parameters of the Derivation can be easily replaced by using the second parameter of `pkgs.callPackage`, greatly enhancing Derivation reusability.
3. While achieving the above two functionalities, it does not increase code complexity, as all dependencies in `pkgs` can be automatically injected.
So it's always recommended to use `pkgs.callPackage` to define Derivations.