Environment collision (?) using scikit-build with conda-build on Windows

[yosefm]

I'm working on transitioning the SfePy package to scikit-build (PR here). So far I managed to get it building in a shell when running setup.py install, even under a pip isolated build environment. However, building a conda package using conda-build (actually mamba-build, but I don't think that matters) fails on Windows.

I've narrowed it down to the following: When conda-build runs, it sets up a Visual Studio build environment using vcvarsall.bat. I can fail a shell build by doing this step myself. I understand that vcvarsall is not needed when ninja build is used, but I don't see how I can turn it off or affect it.

Deeper still: under vcvarsall, compiler outputs go to _skbuild.../Release/output....pyd, while without it the 'Release' part is omitted. This fails linking of extensions using other libraries in the build. If I knew what environmentvariable causes it, maybe I could use it in CMake, but I can't find it.

Has anyone tried to build in such settings, and can give me some directions?

[yosefm]

worked around this by ensuring Ninja is used for all platforms and settings. Still don't know exactly what it was.

[henryiii]

Sorry, missed this! At PyCon, will try to look early next week, please ping me if I forget!