pypi2nix is a command line tool that generates Nix expressions from different python specific sources.

TODO: report bugs

Ping me @garbas if you get stuck.

Make sure Nix is installed.:

% curl https://nixos.org/nix/install | sh

Next clone pypi2nix repository.:

% git clone https://github.com/garbas/pypi2nix

And now install it using nix-env command.:

% cd pypi2nix
% nix-env -f release.nix -iA build."x86_64-linux"

The easiest way to generate a Nix expressions is to invoke.:

% pypi2nix -V "3.5" -e packageA -e packageB==0.1

If you also have requirements.txt file for you python project you can use -r option.:

% pypi2nix -V "3.5" -e packageA -e packageB==0.1 \
    -r requirements.txt -r requirements-dev.txt

If your project relies on zc.buildout you can give -b option a try.:

% pypi2nix -V "2.7" -b buildout.cfg

Option -V tells pypi2nix which python version to be used. To see which python versions are available consule pypi2nix --help.

Once Nix expressions are generated you should be able to see 3 new files:

• requirements_frozen.txt - full frozen set for your for you pypi2nix call. This is the output you would expect from pip freeze.
• requirements_generated.nix - this are the generated nix expressions
• requirements_override.nix - this is an empty file which is ment to override generated nix expressions.
• requirements.nix is a file which connects requirements_generated.nix and requirements_override.nix and exposes it for futher usage.

Build one package:

% nix-build requirements.nix -A packages.empy

Build all packages:

% nix-build requirements.nix -A packages

Build python interpreter with all packages loaded:

% nix-build requirements.nix -A interpreter
% ./result/bin/python -c "import empy"

Enter developent environemnt:

% nix-shell requirements.nix -A interpreter
(nix-shell) % python -c "import empy"

If you are working on a project where its dependencies are defined in requirements.txt then you can create a default.nix and add generated packages as buildInputs as demonstrated here:

{}:
let
  python = import ./requirements.nix { inherit pkgs; };
in python.mkDerivation {
  name = "ProjectA-1.0.0";
  src = ./.;
  buildInputs = [
    python.packages."coverage"
    python.packages."flake8"
    python.packages."mock"
    python.packages."pytest"
    python.packages."pytest-asyncio"
    python.packages."pytest-cov"
    python.packages."pytest-mock"
    python.packages."pytest-xdist"
    python.packages."virtualenv"
  ];
  propagatedBuildInputs = [
    python.packages."aiohttp"
    python.packages."arrow"
    python.packages."defusedxml"
    python.packages."frozendict"
    python.packages."jsonschema"
    python.packages."taskcluster"
    python.packages."virtualenv"
  ];
  ...
}

As you can see you can access all packages via python.packages."<name>". If you want to depend on all packages you can as well do:

propagatedBuildInputs = builtins.attrValues python.packages;

I hope nobody is expecting pypi2nix to do always a perfect job. In python packaging there are just too many different cases that we will never be able to cover. What pypi2nix tries to do is to get you very close.

When things go not as you expected, pypi2nix gives you an option to override anything that it was generated. Even add new packages this way.

An example how you would override a derivation would be adding extra build time dependencies which we can not detect with pypi2nix. As example lets add setuptools-src which got generated, but was not detected as build time dependency of execnet:

{ pkgs, python }:

self: super: {

  "execnet" = python.overrideDerivation super."execnet" (old: {
    buildInputs = old.buildInputs ++ [ self."setuptools-scm" ];
  });

}

This was you can add or remove any python package.

Nothing speaks better then an example:

{ }:

let
  pkgs = import <nixpkgs> {};
  python = import ./requirements.nix { inherit pkgs; };
in python.mkDerivation {
  name = "projectA-1.0.0";
  src = ./.;
  buildInputs = [
    python.packages."coverage"
    python.packages."flake8"
    python.packages."mock"
    python.packages."pytest"
    python.packages."pytest-asyncio"
    python.packages."pytest-cov"
    python.packages."pytest-mock"
    python.packages."pytest-xdist"
  ];
  propagatedBuildInputs = [
    python.packages."aiohttp"
    python.packages."arrow"
    python.packages."defusedxml"
    python.packages."frozendict"
    python.packages."jsonschema"
  ];
  checkPhase = ''
    export NO_TESTS_OVER_WIRE=1
    export PYTHONDONTWRITEBYTECODE=1

    flake8 src/
    py.test --cov=src -cov-report term-missing
    coverage html
  '';
}

Important to know here is that you instantiate all generated packages as python = import ./requirements.nix { inherit pkgs; }; which gives you a python environment with pypi2nix generated packages and some common utilities.

To create a package you would use python.mkDerivation as you are used to that pythonPackages.buildPythonPackage function in nixpkgs. All generated packages are available as one attribute set under python.packages.

One of future goals of pypi2nix project is to also improve the UX of our python tooling in nixpkgs. While this is very hard to do within nixpkgs it is almost trivial to experiment with this outside nixpkgs.

The file examples/Makefile contains specific instructions for packages like sentry, empy, lektor, awscli and rss2email.

Clone pypi2nix repository and using nix-shell command enter development environment.:

% git clone https://github.com/garbas/pypi2nix
% cd pypi2nix
% nix-shell

Code is located in src/pypi2nix.