so (for fun) i'm trying to backport a cuda library to run on my older card. often one imagines this and peters out. [!!!!!!] one issue i ran into is that the cuda sdk appears to provide a polyfill that is already provided by the library in its incomplete backward compatibility code, so i'm lookiing into how to properly add preprocessor defines to it it seems ot make sense to check for the sdk version, but i don't see the presence of the polyfill documented, and the file appears to only reside within the sdk blob on the internet. so i'm downloading the rpms and making a git history of them :D yup here's what I did: first I downloaded all the rpms: curl https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/ | sed -ne "s/.*'\(cuda-cudart-devel[^']*\)'.*/\1/p" | while read file do wget https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/$file done This worked because the webpath happened to return an index page that listed all the files. I made the sed pattern specific for the output. There are existing tools to scrape these paths too. Then I made a git repository and moved all the rpms into an rpms/ subdir, and then added all the contents as git commits: for rpm in rpms/* do git rm -rf usr rpm2cpio $rpm | cpio -idv && git add usr && git commit -m "$rpm" || break done so useful !!!!! ............ ok but now i look at it, and each release has everything in a different usr/local/cuda-* subfolder >( maybe i can rename them before commit