Development environment with nix-shell
Nix can be used to provides some kind of virtual environment through the nix-shell
command.
If you already have a nix package definition of your project it's easy: Just use nix-shell
instead of nix-build
and you will end up in a bash shell that reproduce the build-environment of your package. You can also override[1] your package in a shell.nix
file to add test and coverage dependencies, that are not necessary for the actual build of the package, but that you want for your development environment.
But, if you don't (or you don't want to) have a package definition you can still use a nix-shell to provide a reproducible development environment. To do so, you have to create a shell.nix
file at the root of your repository. For example, if you want to have Ruby 2.3 and not one provided by your system you can write:
{ pkgs ? import <nixpkgs> {} }:
pkgs.mkShell {
# nativeBuildInputs is usually what you want -- tools you need to run
nativeBuildInputs = [ pkgs.buildPackages.ruby_2_3 ];
}
Then just run:
$ nix-shell
Or, to be more explicit:
$ nix-shell shell.nix
Now you got ruby 2.3 in available in your shell:
$ ruby --version
ruby 2.3.7p456 (2018-03-28) [x86_64-linux]
To be sure that the tools installed on your system will not interfere with the dependencies that you've defined in the shell you can use the --pure
option.
If you'd like to load a local nix expression into a shell you can do it by modifying the earlier example a little bit:
{ pkgs ? import <nixpkgs> {} }:
pkgs.mkShell {
# buildInputs is for dependencies you'd need "at run time",
# were you to to use nix-build not nix-shell and build whatever you were working on
buildInputs = [
(import ./my-expression.nix { inherit pkgs; })
];
}
Small cross env note
The comments in the code snippets on nativeBuildInputs
and buildInputs
above might seem pendantic --- who cares about build-time vs run-time when we're just making a dev environment, not a real package! However, the distinction becomes of practical importance if one wants a cross compilation development environment. In that case one would begin file with something like:
{ pkgs ? import <nixpkgs> { crossSystem.config = "exotic_arch-unknown-exotic_os"; } }:
and nativeBuildInputs
would be for the native platform, while buildInputs
would be for the foreign platform. That's a much more practical distinction: any tool that's miscategorized one won't be able to run, and any library that's miscategorized one won't be able to link!
Using Direnv
One of the limitations of nix-shell is that you can't use a shell other than bash. Thankfully, there is Direnv [[2]] with the support of Nix[[3]] to overcome this limitation. Also, Direnv provides some nice features like loading the environment automatically when you enter your project directory and show the loaded variables to you (explicit is always better;-)).
First, install Direnv:
nix-env -i direnv
Now, you need to add one more file in the root of your repository named .envrc
that contains only this:
use_nix
Then depending on the shell you are using, you need to add a line in your configuration file. See the Setup section of the doc[4]. For example, for Zsh put in your ~/.zshrc.local
:
eval "$(direnv hook zsh)"
Then, still at the root of your repository, run:
$ direnv allow .
direnv: loading .envrc
direnv: using nix
[...]
+SIZE +SOURCE_DATE_EPOCH +STRINGS +STRIP +TEMP +TEMPDIR +TMP +TMPDIR +_PATH +buildInputs +builder +checkPhase +cmakeFlags +configureFlags +doCheck +enableParallelBuilding +name +nativeBuildInputs +out +postCheck +preCheck +preConfigure +propagatedBuildInputs +propagatedNativeBuildInputs +shell +src +stdenv +system +testInputs +version ~PATH
Bonus: you can see all the variables set by the nix-shell :)
Now you can leave your project and the environment will be unloaded:
$ cd ..
direnv: unloading
No need to use direnv allow
anymore, the next time you go to your project the environment will be loaded!
More explanation and configuration tweaks can be found in the Direnv wiki [5].
Troubleshooting
When compiling software which links against local files (e.g. when compiling with rust's cargo), you may encounter the following problem:
= note: impure path `/[...]' used in link
This happens due to a specialty in nix: ld
is wrapped in a shell script which refuses to link against files not residing in the nix store, to ensure the purity of builds. Obviously this is not useful when building locally, for example in your home directory. To disable this behavior simply set
NIX_ENFORCE_PURITY=0
in the nix-shell.
No GSettings schemas are installed on the system
When working with gtk, the XDG_DATA_DIRS
must contain a path to the gtk schemas, if not an application may crash with the error above.
For packages we use wrapGAppsHook
in nativeBuildInputs
, however in nix-shell this is not working as expected.
To get your application to work in nix-shell you will need to add the following to your mkShell
expression:
mkShell {
...
buildInputs = [ gtk3 ];
shellHook = ''
XDG_DATA_DIRS=$GSETTINGS_SCHEMA_PATH
'';
}
This may also called: $GSETTINGS_SCHEMAS_PATH
.
Icons not working
Similar to the Gsettings issue, icons can be added with XDG_DATA_DIRS:
XDG_DATA_DIRS=...:${hicolor-icon-theme}/share:${gnome3.adwaita-icon-theme}/share