For the BayesHive project, we have code split over a number of different Git repositories, all hosted on GitHub. We use GitHub for issue tracking and keeping notes on wikis.
Some of the splitting into separate repositories makes sense (our web app is in its own repository, for example), but some of the splitting is mostly historical and makes things a little inconvenient – the infrastructure for the Baysig statistical modelling language lives in three different repositories, which made sense at one time, but doesn’t now. Having these closely related bits of code in different repositories makes it more difficult to track down regressions (
git bisect doesn’t work across multiple repositories) and can make branching messy (if you’re implementing a feature that requires work in multiple different repositories, you need to remember to create a branch in each of them, otherwise it’s easy to commit inconsistent changes to the master branch in one repo and your topic branch in another repo).
As a result of all this, we decided to merge some of our repositories together. Based on a bit of Googling, this seems to be a relatively common requirement, and the advice that’s on offer out there is a little conflicting and confusing. In this article, I’ll try to reduce some of the confusion related to repository merging, as well as showing how to migrate GitHub issues from the pre-existing multiple repositories into the new merged repository. I’ve written a couple of small Haskell programs to do these tasks – they’re available here.
What we want to do here is to take a number of pre-existing repositories and merge them all into a single repository, with the contents of each pre-existing repository in a separate sub-directory, and a single common history of commits reflecting the historical order of changes made to each of the pre-existing repositories.
You will find lots of pages talking about Git sub-projects, sub-tree merging, and other slightly exotic approaches. That’s not what’s needed here. The best explanation I’ve seen of a simple approach that (almost) works is here. Apart from one small error, this does the trick.
Assuming that we have repositories called
repo2, … at URLs
url2, …, the merge goes in two stages (automated in the
merge-repos.hs program in my git-utils repository):
Step 1: pre-processing
For each repository, we want to create a copy of the repository that has its Git index rewritten to make it so that all of the items in the repository live in the sub-directory that we want to use for this individual repository in the merged repository. (Rewriting the Git index this way makes it look as though these things have always lived in this sub-directory, which is exactly what we want.)
For each repository, we do this (taken verbatim from here):
git clone url1 repo1_copy cd repo1_copy git filter-branch --prune-empty --tree-filter ' if [[ ! -e repo1 ]]; then mkdir -p repo1 git ls-tree --name-only $GIT_COMMIT | xargs -I files mv files repo1 fi'
The result of this is that we will have one
_copy repository for each of our “input” repositories, with the contents reorganised to live under a sub-directory.
Step 2: merging
We now want to merge the individual copy repositories. Start by cloning the first one:
git clone repo1_copy merged cd merged
Now, for each of the remaining repositories, we do the following:
git remote add -f repo2 ../repo2_copy git merge -s ours --no-commit repo2/master git read-tree --prefix=/ -u repo2/master git commit -m "Merged repo2" git pull -s subtree repo2 master
(The minor mistake in the instructions I followed is in the
git read-tree command had the wrong value for the
prefix argument, which was leading to extra levels of sub-directories in the merged repository.)
Once you’ve done this for each of your pre-existing repositories, you’ll end up with a single repository with a common history, which you can then push to a new remote on GitHub or wherever for people to use.
If you use GitHub, merging your repositories is only part of the story. You probably also want to take all of the issues on your pre-existing repositories and migrate them to the merged repository. And you don’t want to do that by hand! Fortunately, GitHub has a web API that makes this pretty easy to do. I did it using Haskell (the
merge-issues.hs program in my git-utils repository), but you could do it using whatever tools your comfortable with. The Haskell github library is very convenient though.
I did this in two stages: first, getting all the issues and associated comments from the pre-existing repositories, then applying these issues to the new repository.
Cutting out the details of error handling, getting all the issues for a given repository is simple (here,
ghAuth is our authentication for connecting to GitHub,
ghRepoUser is the GitHub user name of the owner of the repository we’re accessing and
r is the repository name):
(We get the open and closed issues separately, just because the GitHub API defaults to giving open issues only.) Then we can get all the comments for a given issue, represented by the Haskell value
Once we have all the information about all the issues in the pre-existing repositories, we merge that information to produce a ful list of issues for the new repository sorted globally by issue creation date. We can then run through this global list of issues, creating them on the new merged repository one by one. Here, we create a single issue, for a repository
r (the names like
issueBody, etc. are fields in the structure describing the new issue, brought into scope with a record wildcard):
We add an extra label based on the original repository name to allow us to distinguish the origin of issues in the new merged repository and we copy the original assignee and milestone (if any) to the new issue (it’s important that the assignee should already be added as a collaborator on the new merged repository, otherwise we won’t be able to assign issues to them). We can then add comments to the new issue, running over the list of comments we retrieved from the original repository in chronological order:
Finally, for closed issues, after we’ve created the new issue and added any relevant comments, we close it by editing the issue state:
The end result is that all of the issues from the pre-existing repositories are migrated to the new merged repository.
There are a couple of warts to the process. First, all of the issues and comments on the new repository appear under the GitHub user name you use for connecting and running the scripts. That’s pretty unavoidable though, since there’s no way (and neither should there be any way) of spoofing another user’s GitHub ID to generate issues or comments that appear to come from someone else. Second, the issue numbers change compared to those in the original repositories. Again, that’s pretty unavoidable since you’re merging multiple numbered lists into one.
The couple of tools I’ve written to help with this process don’t automate everything (you still need to create yourself a new GitHub repo, add it as a remote for your new merged repo and push to GitHub by hand; you need to set up any collaborators manually so that you can assign issues to them; you need to merge wiki contents by hand; etc.), but this is something that you need to do infrequently enough that that doesn’t really matter.