Two git remotes with different history

Question

We have an internal corporate github repository that we want to open-source to github.com soon. In order to do so safely we want to remove the git history upon the initial push to the public repo. Therefore we need to:

1. Create a public repo that we add as 2nd remote to our local github repo
2. Push a history-truncated version of our local code to the public repo
3. Ensure that all follow-up commits end up in both the local repo and the public repo

As an example, if the local history at the time of the initial public push is:

A -> B -> C -> D

then the public repo after the initial push should look like:

D

After adding two new commits, E and F, the two repos should looks like this respectively:

Local:

A -> B -> C -> D -> E -> F

Public:

D -> E -> F

How can we achieve all of that simultaneously as easily as possible? I realize that there are lots of questions/answers regarding truncating history or having more than one remote, but I haven't found anything about both of those features simultaneously.

Answer 1

One solution is to create a initial commit for your public repo, and use git replace --graft to sort of bind your internal and public histories together. I have personally found this solution to be great in a similar situation as you described in your question. In this case the histories will look like this:

History in your internal Github repo:

A -> B -> C -> D -> I (replaced) -> E -> F

History in your public Github repo:

I -> E -> F

where commits E and F are new development and I is the initial commit for public repo.

Steps to achieve this (assuming master is the main branch in all repos):

• At your stating point you are in commit D
• Copy the files to a new folder and create the initial commit I for you public repo
• Push the initial commit to your (empty) public repo
• Go back to your old local repo and add the new public remote
• Pull the initial commit from public repo to a new branch
• Check the hashes of commits D and I
• On the new branch, run git replace --graft hash_of_commit_I hash_of_commit_D
• When you run git show-ref you'll see ref refs/replace/*
• git log will show replaced mark on commit I
• Merge new branch to master (I to master)
• Push new reference to your internal repo git push internal_remote 'refs/replace/*'
• Push initial commit to your internal repo git push internal_remote HEAD:master
• At this point you can delete the new branch
• Pushing to public repo should not do anything git push public_remote master, there should only be the initial commit I and no previous history
• Now you can make new development and push to both remotes without messing up the history!
• git commit
• git push internal_remote master
• git push public_remote master