When installed, Python packages have the typical structure with include, lib, and bin folders. When loaded by spack, their include directory is added to the CPATH environment variable. Most python packages do not expose C headers. However, python packages that do have headers will often install them in include/pythonX (where X is the python version), rather than in include (for instance on my machine with Python 3.6.8 they are installed in include/python3.6m). Hence any code that requires these headers will need to #include <python3.6m/path/to/the/header.h> (so we should know ahead of time the python version) instead of simply #include <path/to/the/header.h>, OR it should figure out the header path by a means other than simply relying on CPATH.

It would be good, when loading a Python package, that its include/pythonX also be added to CPATH (with X being replaced appropriately).