Build Wheels

[jameskermode]

I'd like to setup cibuildwheel to build binary wheels using GitHub Actions that we can release on PyPI. This would also help with issues like #57 requesting support on a wider range of platforms. I've had good success doing this with f90wrap and quippy, and matscipy should be simpler as there's no Fortran (although there is C++, so #96 would still be valuable).

[jameskermode]

@pastewka Is C++11 sufficient for everything we use? If not there are tricks to enable C++14 compliant compilers: https://cibuildwheel.readthedocs.io/en/stable/cpp_standards/

[pastewka]

The C++ parts are very bare bones, C++11 should be sufficient.

[jameskermode]

Done for MacOS and Linux in #107. Will leave this open until tested and as a reminder to add Windows support.

[ilyes319]

Hey @jameskermode,

Putting a comment here for the windows wheels. I am trying to install matscipy on windows. I think I am using MSVC compiler. When doing pip install matscipy I get the following error:
matscipy_error_pip.txt

Not sure I understand correctly the error. Maybe the wheels does not work with MSVC and I should use GCC.

[jameskermode]

OK thanks. This is the same error I get in GitHub Actions so at least it's consistent.

[jameskermode]

There do seem to be some portability problems with our C++ code to Visual C++. Working on it in the windows-wheels branch, but iteration is a bit slow testing only via CI.

[ilyes319]

The new windows wheel works great, but when I import matscipy.neighbours I get the following error:

  File "<stdin>", line 1, in <module>
  File "C:\Users\Lilyes\anaconda3\envs\torch_nightly\lib\site-packages\matscipy\neighbours.py", line 35, in <module>
    from . import ffi
  File "C:\Users\Lilyes\anaconda3\envs\torch_nightly\lib\site-packages\matscipy\ffi.py", line 38, in <module>
    from _matscipy import *  # noqa
ModuleNotFoundError: No module named '_matscipy'

[pastewka]

I think this could be related to moving the location of the extension module @prs513rosewood

[jameskermode]

Agreed, it looks related to that change. Can we fix or revert please?

[prs513rosewood]

@ilyes319 Could you list your site-packages\ and site-packages\matscipy directories? The changes I've made should pick up _matscipy.* installed in either of these locations.

[prs513rosewood]

I investigated a bit more and the true error message is the following:

Traceback (most recent call last):
  File "C:\Users\lucas\AppData\Roaming\Python\Python311\site-packages\matscipy\ffi.py", line 36, in <module>
    from ._matscipy import *  # noqa
    ^^^^^^^^^^^^^^^^^^^^^^^^
ImportError: DLL load failed while importing _matscipy: The specified module could not be found.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 2, in <module>
  File "C:\Users\lucas\AppData\Roaming\Python\Python311\site-packages\matscipy\ffi.py", line 38, in <module>
    from _matscipy import *  # noqa
    ^^^^^^^^^^^^^^^^^^^^^^^
ModuleNotFoundError: No module named '_matscipy'

I looked into the DLL load failed error and the source could be either:

1. Missing C++ runtime components
2. The compiled extension dynamically links to GCC libs that are missing, which can happen with Meson

After looking at the dependencies of the compiled extension, it seems the second option is more likely, since it links to libgcc_s_seh-1.dll and libstdc++-6.dll, so the error is unrelated to the location of the _matscipy module.

[prs513rosewood]

@ilyes319 The latest wheels should now work for windows.