setup.py

#!/usr/bin/env python
try:
    from skbuild import setup
except ImportError:
    print("Failed to find scikit-build, please run `pip install scikit-build`")
    raise

import os

ITEMS = [
    "-DGOOFIT_PYTHON=ON",
    "-DGOOFIT_TESTS=OFF",
    "-DGOOFIT_CERNROOT=OFF",
    "-DGOOFIT_EXAMPLES=OFF",
    "-DCMAKE_UNITY_BUILD=ON",  # Faster build on CMake 3.16+
]

# Add GOOFIT_* from env.
for item in os.environ:
    if item.startswith("GOOFIT_"):
        ITEMS.append(f"-D{item}={os.environ[item]}")


setup(
    name="goofit",
    version="2.3.0",
    description="GooFit fitting package",
    provides=["goofit"],
    cmake_args=ITEMS,
    packages=["goofit"],
    long_description="""\
GooFit for Python
-----------------

GooFit is a highly parallel fitting framework originally designed for High Energy Physics.

Installation basics
===================

This package can be installed with pip, but uses SciKit-Build, and is build,
fully optimized, on your system. Because of this, there are a few caveats when
running a pip install if you use an old version of pip. When you build, you
should also use pip's ``-v`` flag, so that you can see it build (and observe
the configuration options). Otherwise, you might wait a very long time without
output (especially if CUDA was found).


Installation: pip
=================

Using pip 10+::

    pip install -v goofit

Using pip < 10::

    pip install scikit-build # optionally cmake ninja
    pip install -v goofit


GooFit will automatically look for CUDA, and build in GPU mode if it finds CUDA. You can pick a specific version by passing through a CMake option (see below), or by setting an environment variable, `GOOFIT_DEVICE` before building. You may want to build with OpenMP as a backend to avoid using your GPU, or you might want the CPP version if you are using Anaconda on macOS. Here are the three common backends::

    GOOFIT_DEVICE=CUDA pip install -v goofit
    GOOFIT_DEVICE=OMP pip install -v goofit
    GOOFIT_DEVICE=CPP pip install -v goofit

The lines above use environment variables; GooFit will find any environment variables that start with ``GOOFIT_*`` and set them as CMake defines. If you want to send arbitrary commands to CMake through PIP, you will need to pass each option through, starting with a ``--`` option. Pip will try to reuse the built version if you do not pass options, but will rebuild if you pass options, so this works for a rebuild, unlike the lines above. This is how you would do this to set OMP as the backend::

    pip install -v goofit --install-option="--" --install-option="-DGOOFIT_DEVICE=OMP"
    # OR
    PIP_INSTALL_OPTION="-- -DGOOFIT_DEVICE=OMP" pip install -v goofit


Installation: local
===================

If you want to add PDFs to GooFit, or use GooFit packages, you should be working in a local directory using git. In the following example, I'm assuming you've set up SSH keys with GitHub; you can use https instead if you prefer by changing the URL to ``https://github.com/GooFit/GooFit.git``::

    git clone --recursive git@github.com:GooFit/GooFit.git
    cd goofit

Local pip
~~~~~~~~~

The normal install here works, though as usual you should include verbose output and you should be in a virtual environment (standard practice)::

    pip install -v .

""",
)


# Building a source package from git
# ==================================
#
# For developers only:
#
# To make a source package, start with a clean (such as new) git GooFit package with all submodules checked out::
#
#     git clone --branch=master --recursive --depth=10 git@github.com:GooFit/GooFit.git
#     cd goofit
#     pipx run build --sdist
#     pipx run twine upload dist/*
#
# To make a binary package, use instead::
#
#     GOOFIT_OPTI="" pipx run build --wheel
#
# To set this up on Docker for linux, use::
#
#    docker run -it quay.io/pypa/manylinux1_x86_64 -v goofit-py:goofit-py
#    export PATH=/opt/python/cp36-cp36m/bin:$PATH
#    cd goofit-py
#    python -m pip install scikit-build cmake build
#    python setup.py bdist_wheel -- -DGOOFIT_OPTI="-march=core2"