It seems that in the latest version of pip, we can no longer pass in a class into include_dirs that eventually resolves into a string. This was previously done because running setup.py in a fresh environment would fail as numpy had not yet been installed. Therefore, we used a utility get_numpy_include class which would import numpy only when it was called, after numpy was already installed. This is apparently no longer supported.
Related issue: #248
Failing tests in PR #246
Description of changes
This is most likely no longer needed because we now build using pyproject.toml, where we can pre-specify the build requirements, which will always be installed before the setup.py script is run. Therefore, we no longer need any checks if numpy is installed.
Similarly, the have_cython check is no longer needed in setup.py either, as we specify cython as a build dependency in pyproject.toml.
Issue
It seems that in the latest version of
pip, we can no longer pass in a class intoinclude_dirsthat eventually resolves into a string. This was previously done because runningsetup.pyin a fresh environment would fail as numpy had not yet been installed. Therefore, we used a utilityget_numpy_includeclass which would import numpy only when it was called, after numpy was already installed. This is apparently no longer supported.Related issue: #248 Failing tests in PR #246
Description of changes
This is most likely no longer needed because we now build using
pyproject.toml, where we can pre-specify the build requirements, which will always be installed before thesetup.pyscript is run. Therefore, we no longer need any checks if numpy is installed. Similarly, thehave_cythoncheck is no longer needed insetup.pyeither, as we specify cython as a build dependency inpyproject.toml.Includes