PSA: If you're using gh-pages to host your CI-generated documentation, make sure you don't store any history for that branch

This is something we've been it with on the Smithay repository, and I'm thinking this might be of interest to others, so sharing what we found in case that helps. :)

For some time the Smithay git repo was becoming very slow to clone, and digging into this we realized it was almost 450 MB in size. It ended up that almost all of that space was taken by our gh-pages branch, on which our CI script uploads the api docs for Smithay and its public dependencies.

As it appears, Smithay is kinda hitting a few bad cases so that rustdoc output ends up pretty large: the target/doc/ folder we serve on github pages is ~220 MB in size. Keeping git history on such large content, generating huge diffs at each commit, really made the git repo blow up both in disk size and in processing time required to clone it.

As a first-step measure, we changed the CI script to overwrite the gh-pages branch at every run, rather than just appending a new commit. We use this gh-pages action, so it was just a matter of adding a force_orphan: true parameter.

Turns out that this improved the situation a lot: when it no longer needs to keep history, git manages to compress that 220MB of documentation very well, and now the whole Smithay git repo is only ~15 MB!

So if you're using gh-pages to host your documentation, you might want to check if you can have some similar gains on this side. :)