setup.py

from setuptools import Extension, setup

import numpy as np
from Cython.Build import cythonize

"""
Prior to relying on PEP517/518 and using pyproject.toml, this setup.py used to
be an unintelligible mess. The main reason being that there were no clear
distinction between run-time and build-time dependencies, and since we didn't
want to make Cython a run-time dep, we had to enable a way to install the sdist
from the .c files instead of from the .pyx file.
Anyways. Now Cython is a build-time dep, not a run-time dep, since installing
from the sdist happens in an isolated env.

Creating the sdist still involves compiling the .pyx into .c because we're
executing this file. This is unnecessary but it doesn't matter. The .c files are
excluded from the sdist (in MANIFEST.in) anyway.

Release instruction:

Update changelog and contributors list. 

Basic local checks:
- tests run correctly
- doc compiles without warning (make clean first).

Check that the latest RTD build was OK: https://readthedocs.org/projects/surprise/builds/

Change __version__ in __init__.py to new version name. Also update the hardcoded
version in build_sdist.yml, otherwise the GA jobs will fail.

The sdist is built on Python 3.8. It should be installable from all Python
versions.
- check the sdist building process. It will (unnecessarily) compily the .pyx
  files and the .c files should be excluded from the archive.
- check the install jobs. This will compile the .pyx files again as well as the
  .c files. Look for compilation warnings.
- check test jobs for warnings etc.

Download the sdist from the CI job then upload it to test pypi: twine upload
blabla.tar.gz -r testpypi

Check that install works on testpypi, then upload to pypi and check again.
to install from testpypi:
pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple scikit-surprise  # noqa
Doesn't hurt to check that the tests pass after installing from testpypi.

If not already done, sync gh-pages with the master's README

push new release tag on github (commit last changes first if needed):
    git tag vX.Y.Z
    git push --tags

Check that RTD has updated 'stable' to the new release (may take a while).

In the mean time, upload to conda:
    - Compute SHA256 hash of the new .tar.gz archive (or check it up on PyPI)
    - update recipe/meta.yaml on feedstock fork consequently (only version and
      sha should be changed.  Maybe add some import tests).
    - Push changes, Then open pull request on conda-forge feedstock and merge it
      when all checks are OK. Access the conda-forge feedstock it by the link on
      GitHub 'forked from blah blah'.
    - Check on https://anaconda.org/conda-forge/scikit-surprise that new
      version is available for all platforms.

Then, maybe, celebrate.
"""

# This prevents Cython from using deprecated numpy C APIs
define_macros = [("NPY_NO_DEPRECATED_API", "NPY_1_7_API_VERSION")]

# We're using numpy C APIs in our Cython code so Cython will generate C code
# that requires the numpy headers. We need to tell the compiler where to find
# those headers.
# If you remove this and compilation still works, don't get fooled: it's
# probably only because the numpy headers are available in the default locations
# like /usr/include/numpy/ so they get found. But that wouldn't necessarily be
# the case on other users' machines building the sdist.
include_dirs = [np.get_include()]

extensions = [
    Extension(
        # name is where the .so will be placed, i.e. where the module will be
        # importable from.
        name="surprise.similarities",
        sources=["surprise/similarities.pyx"],
        include_dirs=include_dirs,
        define_macros=define_macros,
    ),
    Extension(
        name="surprise.prediction_algorithms.matrix_factorization",
        sources=["surprise/prediction_algorithms/matrix_factorization.pyx"],
        include_dirs=include_dirs,
        define_macros=define_macros,
    ),
    Extension(
        name="surprise.prediction_algorithms.optimize_baselines",
        sources=["surprise/prediction_algorithms/optimize_baselines.pyx"],
        include_dirs=include_dirs,
        define_macros=define_macros,
    ),
    Extension(
        name="surprise.prediction_algorithms.slope_one",
        sources=["surprise/prediction_algorithms/slope_one.pyx"],
        include_dirs=include_dirs,
        define_macros=define_macros,
    ),
    Extension(
        name="surprise.prediction_algorithms.co_clustering",
        sources=["surprise/prediction_algorithms/co_clustering.pyx"],
        include_dirs=include_dirs,
        define_macros=define_macros,
    ),
]

extensions = cythonize(
    extensions,
    compiler_directives={
        "language_level": 3,
        "boundscheck": False,
        "wraparound": False,
        "initializedcheck": False,
        "nonecheck": False,
    },
)

setup(ext_modules=extensions)