Could not build wheels for XXX which use PEP 517

[albireox]

• I am on the latest Poetry version.
• I have searched the issues of this repo and believe that this is not a duplicate.

• OS version and name: macOS 10.15
• Poetry version: 1.0.0b4

Issue

#11 provides a workaround for custom build systems (e.g., Cython extensions) that works well with poetry. However, when you try to do pip install . on a project that has a custom build.py the build process fails with

ERROR: Could not build wheels for seiya which use PEP 517 and cannot be installed directly

This means, among other things, that one cannot use ReadTheDocs to document the code. This happens with my code but I have also tested it with pendulum.

Is there a workaround for this?

[albireox]

Doing some more Googling on this I came up with the solution described here. One can create a setup.py file such as

from distutils.core import setup

import numpy

from Cython.Build import cythonize
from setuptools.extension import Extension


global setup_kwargs


extensions = [
    Extension('seiya.cube.cubify',
              ['seiya/cube/cubify.pyx'],
              include_dirs=[numpy.get_include()])
]


setup_kwargs = {}
setup_kwargs.update({'ext_modules': cythonize(extensions, annotate=True)})


setup(**setup_kwargs)

and remove the need of using build.py. As far as I can see this works fine (even allows editable installs!). When you do poetry install it creates an UNKNOWN.egg-info directory with PKG-INFO in which all the fields are UNKNOWN. That does not seem to affect poetry but I'm not sure what will happen when you build a distribution (I have run poetry build and the results look ok, with the correct metadata, but I haven't tested it all the way by uploading it to PyPI and installing from a fresh environment).

IMHO poetry really needs to define a way to deal with extensions. There are many projects that depend on a custom build system and this kind of workarounds are getting old ...

[albireox]

@sdispater Any comments on this? I'm seeing problems with this workaround with a recent project that requires compilation of extensions and being able to install using pip install . and I cannot find a good solution.

[flokno]

+1 from me and probably related to #2214

I use numpy to compile a small fortran module. Installation works fine with poetry install, but with pip install ., pip complains that it does not find a setup.py. From what I can see this should be easy to fix since poetry install builds a temporary setup.py as far as I understand.