当他们从PyPI pip安装软件包时，从setup.py中访问用户激活的虚拟环境

Question

My package Foo builds an extension libFoo.so that depends on another package Bar's extension libBar.so, and I don't know a priori where the dependency libBar.so is already installed. I need to know the dependency's directory in order to set libFoo.so's rpath so that it can find libBar.so at runtime.

I thought a good way to do this might be to import the dependency package Bar in my setup.py and inspect the module to get the directory, since I know where libBar.so will be relative to the Bar package's directory. That way if they installed from a virtual environment, I would locate the appropriate path.

However, when I try to do this, ie, when I activate a virtual environment and install my package from PyPI, I find that import Bar in the setup.py ends up importing a version from a DIFFERENT virtual environment (I can tell by inspecting the built libFoo.so's rpath after installation).

On the other hand, if I install the Foo package locally by navigating to the source directory and doing pip install -e . with my virtual environment activated, the rpath is set correctly...

How can I get the right behavior when installing from PyPI? In case it helps, here is the relevant code from setup.py:

import Bar
barDir = os.path.dirname(inspect.getfile(Bar))

When I activate my environment and install from PyPI, barDir is

/Users/me/anaconda/envs/WRONGENVIRONMENT/lib/python2.7/site-packages/bar

When I activate my environment and install locally, barDir is

/Users/me/anaconda/envs/RIGHTENVIRONMENT/lib/python2.7/site-packages/bar

I'm running OS X 10.10.5 and using conda environments

Answer 1

Try using pkg_resources from setuptools : https://pythonhosted.org/setuptools/pkg_resources.html .

Something like:

from pkg_resources import resource_string
libbar = resource_string('Bar', 'libBar.so')

Answer 2

It turned out to be a stupid caching problem (pip install was not getting the latest package from PyPI, but using a cached copy of an older version I had been testing with). I had to do

pip install --no-cache-dir Foo

to ensure I was getting a fresh version