Is there the possibility of building packages with dummy shims for proprietary dynamic libraries that could be substituted at runtime (i.e. a package BRs dummy-libthing-devel, and dynamic linker paths provide the nasty real libthing at runtime)? Obviously there are potential technical and legal issues involved. I'm interested in packages which are useful without the proprietary bits (such as parallel performance tools with multiple targets or computational programs that otherwise run on the CPU). I'm thinking specifically about CUDA and NVIDIA management and performance stuff, but also more generally. Unfortunately, at least until AMD GGPUs get well established, NVIDIA support is pretty much de rigeur in HPC and machine learning application (if that isn't an HPC subset). I thought about offload in libgomp as an example, but I found what's involved difficult to follow; it's not immediately clear to me what interfaces are used, and how. I suppose the first question is whether dummy libcuda etc. already exists that could be used for packaging. I've looked without luck, but maybe someone here knows. Second, is the legal position on producing such a thing from header information clear? I guess that would be a question for the legal list, but I've failed to get mail to that in the past, and I guess the answer is generally known. Thanks. _______________________________________________ devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/devel@xxxxxxxxxxxxxxxxxxxxxxx
