Git submodules are pretty great. They allow you to have nested git repositories so that modular parts of your app can exist as separate repos but be worked with as one file tree. Another benefit is that when submodules are pushed to github they appear simply as links to the repos they represent.
If your not familiar, go ahead and google then come back – how submodules work overall is not the point of this post.
One of the ways I use submodules is to take modular pieces of my dotfiles repo and separate them out into single purpose, independently clonable repos for oh-my-zsh, vim and screen. A level down, inside the vim submodule itself, I use additional submodules in accordance with tpope’s awesome pathogen plugin to bundle the various vim plugins I use. At both of these levels there exist submodules of which I am the author and an active developer.
When I work on these submodules, I like to do so from within the parent repo (vs independently in some other directory). This is especially important in vim so that I can test out my changes immediately. If I didn’t do this, I would have to hack on the submodule, commit, push, go into the vim repo’s copy and pull – all before seeing the affects (Bret Victor would not be very happy with that workflow).
What this means is the submodule must be added with a pushable remote. And since I like to push using ssh keys and not enter my github credentials each time, I use the
git@github url for that. Problem is, when someone wants to clone my repo (that’s what it’s there for), they’re unable to recursively clone those submodules because they don’t have access to them using that url. For that to work, I would’ve had to have added the submodules using the
https protocol which allows anonymous checkouts.
As it turns out, due to the unexpected (but perfectly reasonable) behavior of a
git submodule add command, I can actually have my cake and eat it too.
You see, when you do a
git submodule add <url> <directory>, it writes that
.gitmodules. This is the file and url that’s used when you clone this repo anywhere else and init the submodules within. But this is not the url that’s used when you actually try to push or pull from within the submodules!
In addition to
.gitmodules, the url of the remote also gets written into the
.git/config of the submodule as the
origin remote (this is just normal
clone behavior). This is the url that’s used for push/pull from within the submodule. If you think about it, it makes perfect sense: you’re in a valid git repo; when executing a
push, you wouldn’t expect it to use anything but the
remote that was defined and stored in your
In some versions of git, I find that a submodule’s
.git is actually a file pointing to a
.git/modules/name/ directory in the parent repo.
Finally, the url/directory mapping for the submodule also gets written into the parent repo’s
.git/config. What purpose does that serve? If you figure it out, let me know…
So (however unlikely this is) if you find yourself in the same situation as I, this is how you do that:
Now anyone who clones (recursively) will get the anonymous checkout version (as defined in
.gitmodules), but the
origin remote in the local submodule has been changed to the
git@github version and is pushable using ssh keys.
I recently discovered that this can be solved much more elegantly by adding the following to ~/.gitconfig:
[url "email@example.com:pbrisbin/"] pushInsteadOf = "https://github.com/pbrisbin/"
git encounters the anonymous http remote, it’ll silently use the ssh-based url. Aces.