Installing numpy before using numpy.distutils.core.setup

Question

I am using numpy.distutils to setup a package (mypackage) that has a frotran module. The problem is that if I do pip install mypackage on an environment that does not have numpy, I get the following error:

ModuleNotFoundError: No module named 'numpy'

The easy solution is to ask users (if I manage to have any) to pip install numpy before they install my package, but I do not think this is a very elegant solution.

I came up with the idea of calling setuptools.setup with only setup_requires=['numpy'] before I import numpy and it seems to work well. This is my setup.py:

import setuptools

setuptools.setup(
    setup_requires=[
        'numpy'
    ],)

from numpy.distutils.core import setup, Extension

mod = Extension(name='mypackage.amodule', sources=['source/a.f90'])

setup(name='mypackage',
      packages=['mypackage'],
      ext_modules=[mod],)

I honestly don't fully understand what it implies to call an empty setup() (no name, no package). Is this a good solution? Is this somehow a bad practice?

Answer 1

It is a common issue. How to install a build-time dependency? You might want to use a pyproject.toml file and take advantage of the build-system feature. See PEP517. And an example here:

[build-system]
build-backend = "setuptools.build_meta"
requires = ["setuptools", "numpy"]

Use the pep517 tool to build the distributions (sdist and wheel).