I don't know where the site_packages property is used, so it isn't obvious to me, why only the first site_package is stored and why dist-packages are never stored when site_packages where found.

The site_packages property is used in the InstalledRepository class (see https://github.com/python-poetry/poetry/blob/master/poetry/repositories/installed_repository.py#L32).

We only need the "site-packages" directory where pip will install the dependencies which is, as far as I know, the site-packages in virtual environments and most OS and dist-packages for Debian-based distributions.

Ah, ok I got it. I wonder if something with sysconfig would be more reliable? But as far as I can see, we must be able to detect if we are inside a virtual environment and of we are on a posix or nt system, to get the correct path (purelib seems to be the one we are interested in), are we?

Thanks for pointing me to sysconfig! Do you know if there are some systems where purelib and platlib are different? If there is none we could indeed use purelib and avoid the heuristics we are currently doing.

I'm not sure. I've just discovered that module as well. There is something in PEP427 about the difference.

In the end, it seems that all installed packages will reside in purelib and/or platlib, so to get populate the InstalledRepository we can just check both, I think. I'll test it out and see where it goes.

After testing it out it seems that using sysconfig won't help us. While it works in some cases, for Debian-based systems, for instance, it won't return the directory we need. We could use site.getsitepackages() for non-virtual environments but we would need to find a way to replicate it for virtual environments which would lead to using the same heuristic this PR already uses.

I think we can leave the PR as-is for now since I tested it on Ubuntu, MacOS and Windows without any issue and it fixes the issues I mentioned above.