search

tudo-astroparticlephysics / PROPOSAL

Monte Carlo Simulation propagating charged Leptons through Media as C++ Library
GNU Lesser General Public License v3.0
35 stars 21 forks source link

WIP: use scikit-build-core #389

Open maxnoe opened 7 months ago

maxnoe commented 7 months ago

This should get you started.

More metadata can be added and I recommend to also include setuptools_scm for automatic version handling, this would require a top-level proposal/ pure python package though, instead of just shipping the extension library.

This can be made backwards compatible by importing everything from proposal.....so in proposal/__init__.py.

Jean1995 commented 6 months ago

If I understand this concept correctly, calling conan and pip install is now separated - which I would prefer to the old solution where conan gets called from within setup.py.

I see from the CI how this can work when locally cloing the repository. But would this also be possible via pypi? So that one can do something like

conan install . -o with_python=True -o with_testing=False --build=missing # optional: fetch dependencies
pip install proposal # install proposal

or how would this work?

maxnoe commented 6 months ago

I see from the CI how this can work when locally cloing the repository. But would this also be possible via pypi?

No... because the conanfile is of course only there once you downloaded the package.

This is one of the reasons why this is still "draft". Installing the source distribution from pypi would require having all dependencies already on the system.

One solution around this would be to offer pre-built wheels for the most common platforms, e.g. via cibuildwheel. Then most users will just install the compiled binary, without ever needing conan or even compiling proposal itself.

If we really want to call conan, we could call it from CMakeLists.txt I think, in an if(SKBUILD AND USE_CONAN) block or similar