如何在不安装“ extras_require”依赖项的情况下运行“ python setup.py test”？

Question

I have extras_require dependencies in a setup.cfg listed as:

[extras_require]
tensorflow = tensorflow
tensorflow_gpu = tensorflow-gpu

That way if my package is installed with pip install pkg[tensorflow_gpu] , it installs one version of the package, vs pip install pkg[tensorflow] which installs the other.

This is similar to how the Edward Package packages (following this git issue ).

However, when I run python setup.py test , it installs both tensorflow and tensorflow-gpu packages before executing my unit tests. That can be problematic once I import tensorflow in any unit tests, and is hardware-dependent.

Is there a way to selectively install a certain extras_require package when running python setup.py test ?

Alternately is there a way to install no extras_require packages? Because then I can install tensorflow before running the test command.

(Note: I understand I can run pip and pytest independently, but I'm looking for a solution that uses setuptools )

Answer 1

It appears this is not a setuptools issue, but rather a potential issue with the tool I am using called pyscaffold . I opened a question issue on their github.

When I run python setup.py test with the following setup.cfg , it works fine, as in it does not pull in tensorflow.

[metadata]
name = simple

[options]
package_dir = 
    =src
setup_requires = pytest-runner
install_requires = 
tests_require = pytest

[options.extras_require]
tf = tensorflow

[aliases]
test = pytest

[tool:pytest]
addopts = --verbose
testpaths = tests