Pycall

[samrose]

Many Julia projects use Pycall (at least I am finding).

What is a good Julia2nix..jl way to handle the installing or build of pycall, and packages that rely on Pycall?

[johnnychen94]

I'm not familiar with Nix. But for self-managed PyCall, there're a few things to notice to build reproducible Python dependencies for Julia:

• preinstall conda (and python dependencies), and set CONDA_JL_HOME to tell Conda.jl to use the pre-built version. (PyCall relies on Conda.jl to find python)
• I use conda-lock to manage the conda/python deps.

[GTrunSec]

you can simply enable Python support from attributions of enable.python, https://github.com/JuliaCN/Julia2Nix.jl/blob/main/templates/dev/flake.nix#L33

of course, you can also introduce any pythonEnv here, such as poetry2nix, python.withPakcages.

      enable = {
            python =
              pkgs.python3.buildEnv.override
              { 
                extraLibs = with pkgs.python3Packages; [xlrd matplotlib pyqt5] ;
                # ignoreCollisions = true;
              };
          };
        };

patches: https://github.com/JuliaCN/Julia2Nix.jl/blob/main/nix/julia2nix/overlays/builder/patches.nix#L13