Enable manifpy compilation when ROBOTOLOGY_USES_PYTHON is enabled

[traversaro]

Differently from most C++/Python projects that we support, manif does not permit to install from CMake its Python bindings (see https://github.com/artivis/manif/blob/0.0.4/python/CMakeLists.txt, and do not get afraid for the EIGEN_DEFAULT_TO_ROW_MAJOR as I was : ) ). Instead, the "installation" is happening in the setup.py by setting the CMAKE_LIBRARY_OUTPUT_DIRECTORY variable to the desired installation location.

So the options are:

• Install manifpy by invoking the setup.py and specify manually the location where it should be installed
• Patching the CMake build script of manif to permit to install the Python bindings directly from C++

Any opinion @GiulioRomualdi @diegoferigo ?

[traversaro]

Sister issue: conda-forge/manif-feedstock#3 .

[diegoferigo]

I'm not really an expert of packaging for conda. If I understood correctly, instead of getting files from the CMake install tree as we do for other projects, we have to pass through setuptools that will install the Python files in the enabled site-packages, and then package this folder with conda. Correct?

If this is ok in the conda-side, it seems to be the easiest solution at the moment. Note that I recently opened artivis/manif#228 that, depending if and how it will get implemented, could involve a new CMake machinery similar to what we are are used to have in our hybrid projects.

[traversaro]

To clarify, the issue on manifpy packaging in conda is conda-forge/manif-feedstock#3 . This one is to install manifpy as part of the robotology-superbuild.

[traversaro]

Calling a setup.py from the Buildmanif.cmake file or even creating a Buildmanifpy.cmake seems non obvious, as we would need to use features of ycm_ep_helper that I do not really know if they have been tested a lot in the past. Probably just modifying manifpy to support installing its Python bindings is the best option.

[diegoferigo]

Maybe we can wait artivis/manif#233? However, this means that we cannot support the newly tagged release.

[traversaro]

Maybe we can wait artivis/manif#233? However, this means that we cannot support the newly tagged release.

Cool, I did not saw that! This is not a big problem, if the PR is not merged soon I can quickly create a internal release on a fork and use that one, the proper release was just necessary for conda-forge.