Cross Compiling: Difference between revisions

From NixOS Wiki
imported>Symphorien
show how to use pkgsCross to get a shell, and remove mentions of pre-18.09
Frontear (talk | contribs)
m more clear edits with correct references to platforms
 
(9 intermediate revisions by 4 users not shown)
Line 1: Line 1:
{{Expansion}}
{{Expansion}}
For building arm software check out the Article [[NixOS on ARM]]<br/>
[[Nixpkgs]] provides excellent support in configuring it for cross-platform compiling tasks since 18.09<sup>[citation needed]</sup>.
If you are looking for building 32bit software, check out [[Packaging/32bit Applications]]<br/>
Quick example to cross compile a package: [[Cheatsheet#Cross-compile_packages]].


== Cross-Compiling a package in nixpkgs ==
In order to prepare Nixpkgs for a cross-compiling environment, it needs to be aware of both the platform that performs the build-step, and the platform that will execute the resulting binaries. The former is referred to as the <code>buildPlatform</code>, while the latter is <code>hostPlatform</code>.<blockquote>If you were compiling a program from your system for a Raspberry PI, you would be the <code>buildPlatform</code> whereas the Raspberry PI would be the <code>hostPlatform</code>.</blockquote>Furthermore, in order to provide more granular control to declaring dependencies in these environments, Nixpkgs derivations expose an exhaustive set of attributes that can explicitly define when are where dependencies are required. A full reference to these can be found in the [https://nixos.org/manual/nixpkgs/unstable/#ssec-stdenv-dependencies-propagated Nixpkgs manual].
Cross-compilation is well supported in nixpkgs since 18.09.
The basic idea is to use <code>pkgsCross.platform</code> instead of <code>pkgs</code>:
<syntaxHighlight lang=bash>
nix-build '<nixpkgs>' -A pkgsCross.raspberryPi.openssl
</syntaxHighlight>


== How to obtain a shell with a cross compiler ==
== Getting Started ==
Create a file <code>crossShell.nix</code> as follows:
Nixpkgs exposes two configuration attributes that map internally to the expected behaviors of the build/host platforms as described above. These attributes can be set when importing Nixpkgs as a Nix expression:<syntaxhighlight lang="nix">
<syntaxhighlight lang="nix">
let
with import <nixpkgs> {
  pkgs = import <nixpkgs> {
  crossSystem = {
    localSystem = "x86_64-linux"; # buildPlatform
     config = "aarch64-unknown-linux-gnu";
    crossSystem = "aarch64-linux"; # hostPlatform
  };
in pkgs.hello
</syntaxhighlight>The above will provide a derivation result for the hello derivation that can run on an <code>aarch64-linux</code> system. This can sometimes be tedious especially for common <code>hostPlatform</code> targets. Fortunately, Nixpkgs exposes a <code>pkgsCross</code> attribute that provides pre-configured cross compiling targets. The snippet above converted to using <code>pkgsCross</code> can be shorted to:<syntaxhighlight lang="nix">
let
  pkgs = import <nixpkgs> {
     localSystem = "x86_64-linux";
   };
   };
};
in pkgs.pkgsCross.aarch64-multiplatform.hello
 
</syntaxhighlight>You can perform the same operations using the CLI, and Nix will correctly evaluate the <code>localSystem</code> based on your current system:<syntaxhighlight lang="bash">
mkShell {
nix-build '<nixpkgs>' -A pkgsCross.aarch64-multiplatform.hello # nix-legacy
   buildInputs = [ zlib ]; # your dependencies here
nix build nixpkgs#pkgsCross.aarch64-multiplatform.hello # nix3
}
</syntaxhighlight>All of the above snippets will resolve to the exact same derivation result, which will provide a binary for GNU Hello that can execute only on an <code>aarch64</code> system. There are many other systems <code>pkgsCross</code> has defined, you can see an exhaustive list of all of them on your system:<syntaxhighlight lang="bash">
$ nix-instantiate --eval --expr 'builtins.attrNames (import <nixpkgs> {}).pkgsCross' --json | nix-shell -p jq --command 'jq' # nix-legacy
$ nix eval --impure --expr 'builtins.attrNames (import <nixpkgs> {}).pkgsCross' --json | nix run nixpkgs#jq # nix3
</syntaxhighlight>If you instead prefer to write your systems directly, through <code>localSystem</code> and <code>crossSystem</code>, you can refer to [https://github.com/NixOS/nixpkgs/blob/master/lib/systems/examples.nix nixpkgs/lib/systems/examples.nix] for examples of platforms exposed as attributes. These can be directly used in-place for the aforementioned arguments:<syntaxhighlight lang="nix">
let
   lib = import <nixpkgs/lib>;
  pkgs = import <nixpkgs> {
    #localSystem = ...;
    crossSystem = lib.systems.examples.aarch64-multiplatform;
  };
in pkgs.hello
</syntaxhighlight>
</syntaxhighlight>
and then use it to obtain a shell:
{{commands|nix-shell crossShell.nix}}
The resulting shell contains a cross toolchain and zlib in this example. Note that contrary to native shells, the compiler and some other tools are prefixed: there is no <code>gcc</code> but a <code>aarch64-unknown-linux-gnu-gcc</code>. Some convenience environment variables expand to the prefixed version of tools: <code>$CC</code>, <code>$LD</code>...


Examples of how to specify your target system can be found in [https://github.com/NixOS/nixpkgs/blob/master/lib/systems/examples.nix lib/systems/examples.nix]. If the exact system you are targeting is available in this file then you can use the existing definition as in the following example:
== Development ==
<syntaxhighlight lang="nix">
let pkgs = import <nixpkgs> {
    crossSystem = (import <nixpkgs/lib>).systems.examples.armv7l-hf-multiplatform;
};
in
mkShell {}
</syntaxhighlight>


Even shorter:
=== Basics ===
<syntaxhighlight lang="nix">
let pkgs = import <nixpkgs> {}; in
pkgs.pkgsCross.armv7l-hf-multiplatform.mkShell {}
</syntaxhighlight>


The examples above do not work as is with build dependencies (<code>nativeBuildInputs</code>). A solution is to use <code>callPackage</code> to enable splicing:
Using the same ideas as above, we can create development environments which provide us with a compilation suite that can perform cross-compilation for us. A very simple [[Development environment with nix-shell|development shell]] (colloquially called a "devshell") can be written as:<syntaxhighlight lang="nix">
<syntaxhighlight lang="nix">
# shell.nix
let pkgs = import <nixpkgs> {
{
  crossSystem = {
  pkgs ? import <nixpkgs> {
     config = "aarch64-unknown-linux-gnu";
    localSystem = "x86_64-linux";
   };
     crossSystem = "aarch64-linux";
};
   },
in
}:
  pkgs.callPackage (
pkgs.mkShell {
    {mkShell, pkg-config, zlib}:
  # By default this provides gcc, ar, ld, and some other bare minimum tools
    mkShell {
}
      nativeBuildInputs = [ pkg-config ]; # you build dependencies here
</syntaxhighlight>Entering this development shell via <code>nix-shell shell.nix</code> will add the relevant compiler tools to your PATH temporarily. Similar to other Linux systems, all cross-compiling tools are prefixed with relevant platform prefixes, which means simply typing <code>gcc</code> will not work. However, the provided <code>mkShell</code> will introduce environment variables for your devshell, such as <code>$CC</code>, <code>$AR</code>, <code>$LD</code>, and more. At the time of writing, official documentation on an exhaustive list of these variables does not exist, but you can view them for your devshell through the command-line:<syntaxhighlight lang="bash">
      buildInputs = [ zlib ]; # your dependencies here
$ $EDITOR $(nix-build ./shell.nix) # opens your EDITOR with a massive bash script full of declare -x ...
    }
</syntaxhighlight>Given these environment variables, you can run compile your software using the exact same commands with fairly minimal changes (changing hardcoded <code>gcc</code> values into $CC, for example):<syntaxhighlight lang="bash">
  ) {}
$ $CC -o main src/main.c
</syntaxhighlight>
$ file main
See also {{issue|49526}}.
main: ELF 64-bit LSB executable, ARM aarch64, version 1 (SYSV), dynamically linked, interpreter /nix/store/qa51m8r8rjnigk5hf7sxv0hw7qr7l4bc-glibc-aarch64-unknown-linux-gnu-2.39-52/lib/ld-linux-aarch64.so.1, for GNU/Linux 3.10.0, not stripped
</syntaxhighlight>The above snippet will have minor differences depending on when you run it, but the main thing to notice is <code>ARM aarch64</code>, which tells us our software was able to successfully cross compile.


== Lazy cross-compiling ==
=== Declaring dependencies ===
If you try to declare build-time dependencies within the devshell (such as <code>pkgs.cmake</code>), you will quickly realize that these derivations are actually being built for the <code>crossSystem</code>, making them unusable on your system architecture (see [https://github.com/NixOS/nixpkgs/issues/49526 #49526]). There are ways around this, but in general once you've gotten to this point you should prefer [https://nixos.org/manual/nixpkgs/unstable/#chap-stdenv writing a derivation], which will make it not only easier to write both derivations, but will allow you to follow the recommended practices for using Nix.


If you target "aarch64-unknown-linux-gnu", there is a nice way to reduce amount of cross-compiling and side-step journey to fix cross errors. The idea is to fetch non-essential dependencies from binary cache of regular aarch64 binaries.
If you would prefer to continue building within the devshell, you can use [https://nixos.org/guides/nix-pills/13-callpackage-design-pattern callPackage], which will ''magically'' resolve the dependencies for the correct architecture, provided you place them in the correct attributes:<syntaxhighlight lang="nix">
# shell.nix
{
  pkgs ? import <nixpkgs> {
    #localSystem = ...;
    crossSystem = "aarch64-linux";
  },
}:
pkgs.callPackage (
  {
    mkShell,


Say we are building SDL2.
    pkg-config,
    libGL,
  }:
  mkShell {
    # Derivations that must run on the buildPlatform.
    nativeBuildInputs = [
      pkg-config
    ];


<syntaxhighlight lang="nix">
     # Derivations that must link with the targetPlatform.
let
     buildInputs = [
     # this will use aarch64 binaries from binary cache, so no need to build those
      libGL
     pkgsArm = import <nixpkgs> {
    ];
        config = {};
  }
        overlays = [];
) {}
        system = "aarch64-linux";
</syntaxhighlight>The above snippet will drop you into a devshell that provides <code>pkg-config</code> as a native binary (accessible through <code>$PKG_CONFIG</code>), while also allowing linking to a valid <code>libGL</code> for the <code>crossSystem</code>.
    };


    # these will be your cross packages
For more information regarding the above, namely the usage of <code>nativeBuildInputs</code> and <code>buildInputs</code>, see [https://nixos.org/manual/nixpkgs/stable/#ssec-stdenv-dependencies stdenv dependencies] for a in-depth explanation. Alternatively, a simplified explanation can be found in a comment on the [https://github.com/NixOS/nixpkgs/pull/50881 Nixpkgs repo].
    pkgsCross = import <nixpkgs> {


      overlays = [(self: super: {
== Tips and tricks ==


        # we want to hack on SDL, don't want to hack on those. Some even don't cross-compile
=== Executing cross compiled binaries ===
        inherit (pkgsArm)
By using [[QEMU]], we can natively execute a cross-compiled binary through an emulation layer. This will result in degraded performance but is very suitable for testing the functionality of a binary.
          xorg libpulseaudio libGL guile systemd libxkbcommon
        ;
      })];
      crossSystem = {
        config = "aarch64-unknown-linux-gnu";
      };
    };


in pkgsCross.SDL2.override {
If you are on NixOS, this functionality can be provided automatically on any cross-compiled binary by setting [https://nixos.org/manual/nixos/unstable/options#opt-boot.binfmt.emulatedSystems boot.binfmt.emulatedSystems] in your configuration. After rebuilding, attempting to run a cross-compiled binary will automatically invoke <code>qemu</code> indirectly through the [https://www.kernel.org/doc/html/latest/admin-guide/binfmt-misc.html binfmt_misc kernel feature].<syntaxhighlight lang="bash">
      # those shouldn't be neither pkgsCross, nor pkgsArm
$ ./result
      # because those trigger
Hello World!
      #    cannot execute binary file: Exec format error
$ ./result-aarch64-linux
      # in this case it was enough to just use buildPackages variants
Hello World!
      # but in general there may be problems
</syntaxhighlight>Otherwise, you can use the <code>pkgs.qemu-user</code> to download qemu user space programs (or use any installed by your distro) to run your package easily.<syntaxhighlight lang="bash">
      inherit (pkgsCross.buildPackages)
$ ./result
        wayland wayland-protocols
Hello World!
      ;
$ qemu-aarch64 ./result-aarch64-linux
  }
Hello World!
</syntaxhighlight>
</syntaxhighlight>


=== Leveraging the binary cache ===
You will likely have noticed that resolving derivations through either pkgsCross or a configured Nixpkgs instance results in your system needing to build the binary. This is because cross-compiled binaries are not cached on the official [[Binary Cache|binary cache]]. Fortunately, there are a small set of systems that are actively built and cached officially. At the time of writing, this only includes <code>aarch64-linux</code>, <code>aarch64-darwin</code>, <code>i686-linux</code>, <code>x86_64-linux</code>, and <code>x86_64-darwin</code>. If your platform targets include these, you may be able to leverage a slight hack to avoid large-scale builds.<blockquote>Please note that this is not recommended, as it hacks around some internal details of Nixpkgs which are subject to change at any time.</blockquote>An example of this using <code>pkgs.SDL2</code>:<syntaxhighlight lang="nix">
let
  # this will use aarch64 binaries from binary cache, so no need to build those
  pkgsArm = import <nixpkgs> {
    system = "aarch64-linux";
  };


== How to specify dependencies ==
  # these will be your cross packages
 
  pkgsCross = import <nixpkgs> {
Depending in which if packages are required at build time or at runtime they need to go to different inputs the derivation.
    overlays = [
 
      (self: super: {
* If it is used at build-time it's <code>depsBuildXXX</code>
        # we want to hack on SDL, don't want to hack on those. Some even don't cross-compile
** compiler producing native binaries go to <code>depsBuildBuild</code>
        inherit (pkgsArm)
** compiler producing cross binaries, all setup hooks and programs executed by the builder go to <code>depsBuildHost</code>
          xorg libpulseaudio libGL guile systemd libxkbcommon
*** common examples: <code>pkg-config, autoreconfHook, makeWrapper, intltool, bison, flex</code>
          ;
* If it is used at run-time it's <code>depsHostXXX</code>. [Stack linking doesn't effect this, even if it allows us to forget where things came from.]
      })
** if it’s an interpreter that will be needed by an installed script, it should go in <code>depsHostTarget</code>.
    ];
** otherwise it is probably only needed at build time and can go in <code>depsBuildHost</code>
    #localSystem = ...;
* If it is a tool and "acts" (e.g. helps build) on build-time stuff, then it's <code>depsXXXBuild</code>
    crossSystem = "aarch64-linux";
* If it is a tool and "acts" on run-time stuff, then it's <code>depsXXXHost</code>
  };
* if it is not a tool, it's <code>depsXXX(XXX+1)</code>(build + 1 == host, host +1 == target) for backwards compatibility, use <code>nativeBuildInputs</code> instead of <code>depsBuildHost</code> and <code>buildInputs</code> instead of <code>depsHostTarget</code>.
in pkgsCross.SDL2.override {
 
  # These should be neither pkgsCross, nor pkgsArm
Source: https://github.com/NixOS/nixpkgs/pull/50881#issuecomment-440772499
  # because those trigger
 
  # > cannot execute binary file: Exec format error
  # In this case it was enough to just use buildPackages variants,
  # but in general, there may be problems
  inherit (pkgsCross.buildPackages)  
    wayland wayland-protocols
    ;
}
</syntaxhighlight>


== References ==
== See also ==


* Nixpkgs manual on [https://nixos.org/nixpkgs/manual/#chap-cross cross compiling]
* [[NixOS on ARM]]
* Nixpkgs manual on [https://nixos.org/nixpkgs/manual/#ssec-stdenv-dependencies Specifying dependencies]
* [[Packaging/32bit Applications]]
* [[Cheatsheet#Cross-compile packages]]
* [https://nixos.org/nixpkgs/manual/#chap-cross Nixpkgs manual on cross compiling]


* [https://matthewbauer.us/blog/beginners-guide-to-cross.html Introduction to Cross Compilation with nix by Matthew Bauer]
* [https://matthewbauer.us/blog/beginners-guide-to-cross.html Introduction to Cross Compilation with nix by Matthew Bauer]
* [https://docs.google.com/presentation/d/11B3Igxj0KmsxgT_UQvKufd5El_oH-T__Sl3JNCZeOnY/edit?usp=sharing Slides from the cross-compilation workshop on 35c3]


== Additional resources ==
* [https://logs.nix.samueldr.com/nixos/2018-08-03#1533327247-1533327971; 2018-08-03 discussion on #nixos] ([https://matrix.to/#/!AinLFXQRxTuqNpXyXk:matrix.org/$15333274371713496LOAor:matrix.org Mirror of chat on Matrix.org])


* [https://docs.google.com/presentation/d/11B3Igxj0KmsxgT_UQvKufd5El_oH-T__Sl3JNCZeOnY/edit?usp=sharing Slides from the cross-compilation workshop on 35c3]
[[Category:nix]]
* [https://logs.nix.samueldr.com/nixos/2018-08-03#1533327247-1533327971; 2018-08-03 discussion on #nixos] ([https://matrix.to/#/!AinLFXQRxTuqNpXyXk:matrix.org/$15333274371713496LOAor:matrix.org Mirror of chat on Matrix.org])
[[Category:Development]]

Latest revision as of 19:13, 13 October 2024

Nixpkgs provides excellent support in configuring it for cross-platform compiling tasks since 18.09[citation needed].

In order to prepare Nixpkgs for a cross-compiling environment, it needs to be aware of both the platform that performs the build-step, and the platform that will execute the resulting binaries. The former is referred to as the buildPlatform, while the latter is hostPlatform.

If you were compiling a program from your system for a Raspberry PI, you would be the buildPlatform whereas the Raspberry PI would be the hostPlatform.

Furthermore, in order to provide more granular control to declaring dependencies in these environments, Nixpkgs derivations expose an exhaustive set of attributes that can explicitly define when are where dependencies are required. A full reference to these can be found in the Nixpkgs manual.

Getting Started

Nixpkgs exposes two configuration attributes that map internally to the expected behaviors of the build/host platforms as described above. These attributes can be set when importing Nixpkgs as a Nix expression:

let
  pkgs = import <nixpkgs> {
    localSystem = "x86_64-linux"; # buildPlatform
    crossSystem = "aarch64-linux"; # hostPlatform
  };
in pkgs.hello

The above will provide a derivation result for the hello derivation that can run on an aarch64-linux system. This can sometimes be tedious especially for common hostPlatform targets. Fortunately, Nixpkgs exposes a pkgsCross attribute that provides pre-configured cross compiling targets. The snippet above converted to using pkgsCross can be shorted to:

let
  pkgs = import <nixpkgs> {
    localSystem = "x86_64-linux";
  };
in pkgs.pkgsCross.aarch64-multiplatform.hello

You can perform the same operations using the CLI, and Nix will correctly evaluate the localSystem based on your current system:

nix-build '<nixpkgs>' -A pkgsCross.aarch64-multiplatform.hello # nix-legacy
nix build nixpkgs#pkgsCross.aarch64-multiplatform.hello # nix3

All of the above snippets will resolve to the exact same derivation result, which will provide a binary for GNU Hello that can execute only on an aarch64 system. There are many other systems pkgsCross has defined, you can see an exhaustive list of all of them on your system:

$ nix-instantiate --eval --expr 'builtins.attrNames (import <nixpkgs> {}).pkgsCross' --json | nix-shell -p jq --command 'jq' # nix-legacy
$ nix eval --impure --expr 'builtins.attrNames (import <nixpkgs> {}).pkgsCross' --json | nix run nixpkgs#jq # nix3

If you instead prefer to write your systems directly, through localSystem and crossSystem, you can refer to nixpkgs/lib/systems/examples.nix for examples of platforms exposed as attributes. These can be directly used in-place for the aforementioned arguments:

let
  lib = import <nixpkgs/lib>;
  pkgs = import <nixpkgs> {
    #localSystem = ...;
    crossSystem = lib.systems.examples.aarch64-multiplatform;
  };
in pkgs.hello

Development

Basics

Using the same ideas as above, we can create development environments which provide us with a compilation suite that can perform cross-compilation for us. A very simple development shell (colloquially called a "devshell") can be written as:

# shell.nix
{
  pkgs ? import <nixpkgs> {
    localSystem = "x86_64-linux";
    crossSystem = "aarch64-linux";
  },
}:
pkgs.mkShell {
  # By default this provides gcc, ar, ld, and some other bare minimum tools
}

Entering this development shell via nix-shell shell.nix will add the relevant compiler tools to your PATH temporarily. Similar to other Linux systems, all cross-compiling tools are prefixed with relevant platform prefixes, which means simply typing gcc will not work. However, the provided mkShell will introduce environment variables for your devshell, such as $CC, $AR, $LD, and more. At the time of writing, official documentation on an exhaustive list of these variables does not exist, but you can view them for your devshell through the command-line:

$ $EDITOR $(nix-build ./shell.nix) # opens your EDITOR with a massive bash script full of declare -x ...

Given these environment variables, you can run compile your software using the exact same commands with fairly minimal changes (changing hardcoded gcc values into $CC, for example):

$ $CC -o main src/main.c
$ file main
main: ELF 64-bit LSB executable, ARM aarch64, version 1 (SYSV), dynamically linked, interpreter /nix/store/qa51m8r8rjnigk5hf7sxv0hw7qr7l4bc-glibc-aarch64-unknown-linux-gnu-2.39-52/lib/ld-linux-aarch64.so.1, for GNU/Linux 3.10.0, not stripped

The above snippet will have minor differences depending on when you run it, but the main thing to notice is ARM aarch64, which tells us our software was able to successfully cross compile.

Declaring dependencies

If you try to declare build-time dependencies within the devshell (such as pkgs.cmake), you will quickly realize that these derivations are actually being built for the crossSystem, making them unusable on your system architecture (see #49526). There are ways around this, but in general once you've gotten to this point you should prefer writing a derivation, which will make it not only easier to write both derivations, but will allow you to follow the recommended practices for using Nix.

If you would prefer to continue building within the devshell, you can use callPackage, which will magically resolve the dependencies for the correct architecture, provided you place them in the correct attributes:

# shell.nix
{
  pkgs ? import <nixpkgs> {
    #localSystem = ...;
    crossSystem = "aarch64-linux";
  },
}:
pkgs.callPackage (
  {
    mkShell,

    pkg-config,
    libGL,
  }:
  mkShell {
    # Derivations that must run on the buildPlatform.
    nativeBuildInputs = [
      pkg-config
    ];

    # Derivations that must link with the targetPlatform.
    buildInputs = [
      libGL
    ];
  }
) {}

The above snippet will drop you into a devshell that provides pkg-config as a native binary (accessible through $PKG_CONFIG), while also allowing linking to a valid libGL for the crossSystem.

For more information regarding the above, namely the usage of nativeBuildInputs and buildInputs, see stdenv dependencies for a in-depth explanation. Alternatively, a simplified explanation can be found in a comment on the Nixpkgs repo.

Tips and tricks

Executing cross compiled binaries

By using QEMU, we can natively execute a cross-compiled binary through an emulation layer. This will result in degraded performance but is very suitable for testing the functionality of a binary.

If you are on NixOS, this functionality can be provided automatically on any cross-compiled binary by setting boot.binfmt.emulatedSystems in your configuration. After rebuilding, attempting to run a cross-compiled binary will automatically invoke qemu indirectly through the binfmt_misc kernel feature.

$ ./result
Hello World!
$ ./result-aarch64-linux
Hello World!

Otherwise, you can use the pkgs.qemu-user to download qemu user space programs (or use any installed by your distro) to run your package easily.

$ ./result
Hello World!
$ qemu-aarch64 ./result-aarch64-linux
Hello World!

Leveraging the binary cache

You will likely have noticed that resolving derivations through either pkgsCross or a configured Nixpkgs instance results in your system needing to build the binary. This is because cross-compiled binaries are not cached on the official binary cache. Fortunately, there are a small set of systems that are actively built and cached officially. At the time of writing, this only includes aarch64-linux, aarch64-darwin, i686-linux, x86_64-linux, and x86_64-darwin. If your platform targets include these, you may be able to leverage a slight hack to avoid large-scale builds.

Please note that this is not recommended, as it hacks around some internal details of Nixpkgs which are subject to change at any time.

An example of this using pkgs.SDL2:

let
  # this will use aarch64 binaries from binary cache, so no need to build those
  pkgsArm = import <nixpkgs> {
    system = "aarch64-linux";
  };

  # these will be your cross packages
  pkgsCross = import <nixpkgs> {
    overlays = [
      (self: super: {
        # we want to hack on SDL, don't want to hack on those. Some even don't cross-compile
        inherit (pkgsArm)
          xorg libpulseaudio libGL guile systemd libxkbcommon
          ;
      })
    ];
    #localSystem = ...;
    crossSystem = "aarch64-linux";
  };
in pkgsCross.SDL2.override { 
  # These should be neither pkgsCross, nor pkgsArm
  # because those trigger
  # > cannot execute binary file: Exec format error
  # In this case it was enough to just use buildPackages variants,
  # but in general, there may be problems
  inherit (pkgsCross.buildPackages) 
    wayland wayland-protocols
    ;
}

See also