Gcc 13 : GLIBCXX_3.4.32 not found... when using docker run --gpus all

[chavid]

Hi.

I usually add --gpus all when starting a gcc docker image, so to grant access to my NVidia card. Starting with the 13.1, and now 13.2 release of gcc, I observe that the presence of --gpus all prevents from running the simplest hello_world.cpp program:

/usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.32' not found (required by ./hello_world.exe)

Some ldd inspection on hello_world.exe shows that without --gpus all, the need with the dynamic library libstdc++.so.6 is correctly solved to /usr/local/lib64/libstdc++.so.6. But when the container is started with --gpus all, this is wrongly solved to /usr/lib/x86_64-linux-gnu/libstdc++.so.6.

Any explanation and fix ?

[tianon]

Sorry for the delay! I think --gpus injects extra things into the container (perhaps NVIDIA shared libraries, for example?) that might cause conflicts, especially with an image like this one that already has version conflicts on libraries like libstdc++ given the nature of the image. I'd suggest you're probably better off being FROM debian:xxx or FROM buildpack-deps:xxx if you just need GCC for compiling applications and don't actually need a specific version of GCC.