@daviddoji and I ran into a situation where using a custom environment with a package that has binary dependencies (numba) threw confusing load errors:
The problem is that the amore modules set the LD_LIBRARY_PATH variables, which caused Python to find an incompatible libstdc++ from the amore/mid module instead of the one that the offending shared object (libLLVM-14.so) was actually linked against in our current environment:
Curiously, import pandas, numba hits that error, but import numba, pandas doesn't. So loading pandas is setting up some global state which is part of the problem. :confused:
@daviddoji and I ran into a situation where using a custom environment with a package that has binary dependencies (numba) threw confusing load errors:
The problem is that the
amore
modules set theLD_LIBRARY_PATH
variables, which caused Python to find an incompatible libstdc++ from theamore/mid
module instead of the one that the offending shared object (libLLVM-14.so) was actually linked against in our current environment:Presumably the child
context_python
process is inheriting the environment variables from the parent, so we hacked around it with a shim:But it would be nice if there was a way to specify environment variables for
context_python
, or a command to activate the environment properly.