Conda-installed packages not importable under test

Friday, I discovered I was no longer able to test my provenance parsing code. This wasn’t entirely unexpected - in an effort to get my head around the rpy2 package for a side project, I’d been mucking around pretty agressively with conda environments. As a result, all of the packages I needed to install and test provenance_lib were shown in conda list, but networkx and yaml were unavailable when I ran pytest.

By checking which version of Python was available to different conda environments*, I figured out that the python in the traceback lined up with the python version used by my base environment, not my subject environment. I must have accidentally installed pytest in my base env somewhere along the line. Because of this, it wasn’t installed in the subject environments (requirement already satisfied). Acting as the the test runner, I suspect it grabbed its local python version, which didn’t know anything about the site packages in my subject env. Removing pytest from base and reinstalling per subject env solved the problem, and left things feeling much cleaner.

* You can check on this with which python, or by starting a python session, importing sys, and checking sys.executable. This latter approach can be useful for troubleshooting, because it’s possible to get different sys.executable values from different programs. If it’s installed in base, for example, IPython may be running a different Python than the “raw” python in your current conda environment. I suspect it’s possible to get different results from different systems as well. Homebrew, for example, used to mishandle things like this.