How does github upload file feature works?

Question

I came across a reddit post sometime back about how does github upload a file feature doesn't actually clone the project, instead it modifies the git history through low-level api. I'm unable to find that reddit post again :-(

I have a similar requirement to add & push a file to a git repo but cloning the entire repo is very costly (files are huge). Hence, I'm looking for low-level git api commands to use in order to add file to repository and push without cloning.

FYI - Shallow cloning won't work for me, since all the large files are located in the project root.

There are other similar questions, but have no solution - everybody says it's impossible. Considering github did it, there should be a way.

Answer 1

GitHub has the repository, so it's just a matter, on GitHub, of extracting the latest commit, writing the file, updating the index, and making a new commit. ¹ You can do the same thing on any server that has the entire repository: accept a file, check out the desired commit, add the file, and make the new commit.

---

¹ In fact, all you really need to do is read the commit into an index file with git read-tree , write a blob object with git hash-object -w , then update the index in place and use git write-tree , git commit-tree , and git update-ref to insert the new commit. These plumbing commands do not require a separate work-tree directory. It's easier if you have a work-tree directory, though, as that makes the user-interface commands work.