shell.nix

{ pkgs ? import <nixpkgs> { } }:
with pkgs;
let
  subprocess32 = python38Packages.buildPythonPackage rec {
    pname = "subprocess32";
    version = "3.5.4";

    src = python38Packages.fetchPypi {
      inherit pname version;
      sha256 =
        "eb2937c80497978d181efa1b839ec2d9622cf9600a039a79d0e108d1f9aec79d";
    };

    doCheck = false;

    meta = {
      homepage = "https://github.com/google/python-subprocess32";
      description =
        "A backport of the Python 3 subprocess module for use on Python 2";
    };
  };
  wandb = python38Packages.buildPythonPackage rec {
    pname = "wandb";
    version = "0.10.22";

    src = python38Packages.fetchPypi {
      inherit pname version;
      sha256 =
        "676a0b78864527ba7e2ed0a5fc34e04c34010f2ee5be87900d704092c86380da";
    };

    doCheck = false;

    propagatedBuildInputs = with python38Packages; [
      GitPython
      pyyaml
      click
      configparser
      dateutil
      docker_pycreds
      pathtools
      promise
      protobuf
      psutil
      requests
      sentry-sdk
      shortuuid
      six
      subprocess32
    ];

    meta = {
      homepage = "https://github.com/wandb/client";
      description =
        "🔥 A tool for visualizing and tracking your machine learning experiments. This repo contains the CLI and Python API";
    };
  };
  local-python = python38.withPackages (ps:
    with ps; [
      fastapi
      flake8
      ipython
      mypy
      pytorch
      transformers
      uvicorn
      wandb
    ]);
in mkShell { buildInputs = [ local-python ]; }

I am not familiar with python, but it seems to me that the version bound on subprocess32 inside ./requirements.txt is incorrect, at least when building with python3.

Try removing pkgs.python27Packages.subprocess32 and setuptools from propagatedBuildInputs, and add (next to propagatedBuildInputs), patchPhase = "sed -i '/^subprocess32>=.*/d' ./requirements.txt";. This makes wandb build for me.

I don't know of the ramifications of doing that though! Maybe that is not the right approach at all...

Thanks so much @aforemny. I've got this working now by writing a derivation for subprocess32 as well. I'll update the gist with what worked.