Description
Hi,
I have a scikit-build-based / nanobind project .
When I try to build it into an activated conda env where the libstd++ conda env lib is different from the one of my system, the build system takes into account my system libstd++ at link stage. As a result it could crash at import time if my system libstdc++ ABI is > to the one in the conda env (../lib/libstdc++.so.6: version 'GLIBCXX_3.4.30' not found (required by....)
. One option is to repair the wheel, but it's not optimal. I noticed that Setuptools based builds seems to be aware of running into a conda env by setting correct -L
and -rpath
flags. So my question is pretty simple : am I missing something in sk-build-core configuration or is it a design choice and in this case should I rely on "pure" cmake mecanisms to detect and handle the compilation into an activated conda env?
thanks in advance!