.6 Git Tools - Submodules
It often happens that while working on one project, you need to use another project from within it. Perhaps it’s a library that a third party developed or that you’re developing separately and using in multiple parent projects. A common issue arises in these scenarios: you want to be able to treat the two projects as separate yet still be able to use one from within the other.
Here’s an example. Suppose you’re developing a web site and creating Atom feeds. Instead of writing your own Atom-generating code, you decide to use a library. You’re likely to have to either include this code from a shared library like a CPAN install or Ruby gem, or copy the source code into your own project tree. The issue with including the library is that it’s difficult to customize the library in any way and often more difficult to deploy it, because you need to make sure every client has that library available. The issue with vendoring the code into your own project is that any custom changes you make are difficult to merge when upstream changes become available.
Git addresses this issue using submodules. Submodules allow you to keep a Git repository as a subdirectory of another Git repository. This lets you clone another repository into your project and keep your commits separate.
Suppose you want to add the Rack library (a Ruby web server gateway interface) to your project, possibly maintain your own changes to it, but continue to merge in upstream changes. The first thing you should do is clone the external repository into your subdirectory. You add external projects as submodules with the
git submodule add command:
$ git submodule add git://github.com/chneukirchen/rack.git rack Initialized empty Git repository in /opt/subtest/rack/.git/ remote: Counting objects: 3181, done. remote: Compressing objects: 100% (1534/1534), done. remote: Total 3181 (delta 1951), reused 2623 (delta 1603) Receiving objects: 100% (3181/3181), 675.42 KiB | 422 KiB/s, done. Resolving deltas: 100% (1951/1951), done.
Now you have the Rack project under a subdirectory named
rack within your project. You can go into that subdirectory, make changes, add your own writable remote repository to push your changes into, fetch and merge from the original repository, and more. If you run
git status right after you add the submodule, you see two things:
$ git status # On branch master # Changes to be committed: # (use "git reset HEAD <file>..." to unstage) # # new file: .gitmodules # new file: rack #
First you notice the
.gitmodules file. This is a configuration file that stores the mapping between the project’s URL and the local subdirectory you’ve pulled it into:
$ cat .gitmodules [submodule "rack"] path = rack url = git://github.com/chneukirchen/rack.git
If you have multiple submodules, you’ll have multiple entries in this file. It’s important to note that this file is version-controlled with your other files, like your
.gitignore file. It’s pushed and pulled with the rest of your project. This is how other people who clone this project know where to get the submodule projects from.
The other listing in the
git status output is the rack entry. If you run
git diff on that, you see something interesting:
$ git diff --cached rack diff --git a/rack b/rack new file mode 160000 index 0000000..08d709f --- /dev/null +++ b/rack @@ -0,0 +1 @@ +Subproject commit 08d709f78b8c5b0fbeb7821e37fa53e69afcf433
rack is a subdirectory in your working directory, Git sees it as a submodule and doesn’t track its contents when you’re not in that directory. Instead, Git records it as a particular commit from that repository. When you make changes and commit in that subdirectory, the superproject notices that the HEAD there has changed and records the exact commit you’re currently working off of; that way, when others clone this project, they can re-create the environment exactly.
This is an important point with submodules: you record them as the exact commit they’re at. You can’t record a submodule at
master or some other symbolic reference.
When you commit, you see something like this:
$ git commit -m 'first commit with submodule rack' [master 0550271] first commit with submodule rack 2 files changed, 4 insertions(+), 0 deletions(-) create mode 100644 .gitmodules create mode 160000 rack
Notice the 160000 mode for the rack entry. That is a special mode in Git that basically means you’re recording a commit as a directory entry rather than a subdirectory or a file.
You can treat the
rack directory as a separate project and then update your superproject from time to time with a pointer to the latest commit in that subproject. All the Git commands work independently in the two directories:
$ git log -1 commit 0550271328a0038865aad6331e620cd7238601bb Author: Scott Chacon <firstname.lastname@example.org> Date: Thu Apr 9 09:03:56 2009 -0700 first commit with submodule rack $ cd rack/ $ git log -1 commit 08d709f78b8c5b0fbeb7821e37fa53e69afcf433 Author: Christian Neukirchen <email@example.com> Date: Wed Mar 25 14:49:04 2009 +0100 Document version change
Here you’ll clone a project with a submodule in it. When you receive such a project, you get the directories that contain submodules, but none of the files yet:
$ git clone git://github.com/schacon/myproject.git Initialized empty Git repository in /opt/myproject/.git/ remote: Counting objects: 6, done. remote: Compressing objects: 100% (4/4), done. remote: Total 6 (delta 0), reused 0 (delta 0) Receiving objects: 100% (6/6), done. $ cd myproject $ ls -l total 8 -rw-r--r-- 1 schacon admin 3 Apr 9 09:11 README drwxr-xr-x 2 schacon admin 68 Apr 9 09:11 rack $ ls rack/ $
rack directory is there, but empty. You must run two commands:
git submodule init to initialize your local configuration file, and
git submodule update to fetch all the data from that project and check out the appropriate commit listed in your superproject:
$ git submodule init Submodule 'rack' (git://github.com/chneukirchen/rack.git) registered for path 'rack' $ git submodule update Initialized empty Git repository in /opt/myproject/rack/.git/ remote: Counting objects: 3181, done. remote: Compressing objects: 100% (1534/1534), done. remote: Total 3181 (delta 1951), reused 2623 (delta 1603) Receiving objects: 100% (3181/3181), 675.42 KiB | 173 KiB/s, done. Resolving deltas: 100% (1951/1951), done. Submodule path 'rack': checked out '08d709f78b8c5b0fbeb7821e37fa53e69afcf433'
rack subdirectory is at the exact state it was in when you committed earlier. If another developer makes changes to the rack code and commits, and you pull that reference down and merge it in, you get something a bit odd:
$ git merge origin/master Updating 0550271..85a3eee Fast forward rack | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) [master*]$ git status # On branch master # Changes not staged for commit: # (use "git add <file>..." to update what will be committed) # (use "git checkout -- <file>..." to discard changes in working directory) # # modified: rack #
You merged in what is basically a change to the pointer for your submodule; but it doesn’t update the code in the submodule directory, so it looks like you have a dirty state in your working directory:
$ git diff diff --git a/rack b/rack index 6c5e70b..08d709f 160000 --- a/rack +++ b/rack @@ -1 +1 @@ -Subproject commit 6c5e70b984a60b3cecd395edd5b48a7575bf58e0 +Subproject commit 08d709f78b8c5b0fbeb7821e37fa53e69afcf433
This is the case because the pointer you have for the submodule isn’t what is actually in the submodule directory. To fix this, you must run
git submodule update again:
$ git submodule update remote: Counting objects: 5, done. remote: Compressing objects: 100% (3/3), done. remote: Total 3 (delta 1), reused 2 (delta 0) Unpacking objects: 100% (3/3), done. From firstname.lastname@example.org:schacon/rack 08d709f..6c5e70b master -> origin/master Submodule path 'rack': checked out '6c5e70b984a60b3cecd395edd5b48a7575bf58e0'
You have to do this every time you pull down a submodule change in the main project. It’s strange, but it works.
One common problem happens when a developer makes a change locally in a submodule but doesn’t push it to a public server. Then, they commit a pointer to that non-public state and push up the superproject. When other developers try to run
git submodule update, the submodule system can’t find the commit that is referenced, because it exists only on the first developer’s system. If that happens, you see an error like this:
$ git submodule update fatal: reference isn’t a tree: 6c5e70b984a60b3cecd395edd5b48a7575bf58e0 Unable to checkout '6c5e70b984a60b3cecd395edd5ba7575bf58e0' in submodule path 'rack'
You have to see who last changed the submodule:
$ git log -1 rack commit 85a3eee996800fcfa91e2119372dd4172bf76678 Author: Scott Chacon <email@example.com> Date: Thu Apr 9 09:19:14 2009 -0700 added a submodule reference I will never make public. hahahahaha!
Then, you e-mail that guy and yell at him.
Sometimes, developers want to get a combination of a large project’s subdirectories, depending on what team they’re on. This is common if you’re coming from CVS or Subversion, where you’ve defined a module or collection of subdirectories, and you want to keep this type of workflow.
A good way to do this in Git is to make each of the subdirectories a separate Git repository and then create superproject Git repositories that contain multiple submodules. A benefit of this approach is that you can more specifically define the relationships between the projects with tags and branches in the superprojects.
Using submodules isn’t without hiccups, however. First, you must be relatively careful when working in the submodule directory. When you run
git submodule update, it checks out the specific version of the project, but not within a branch. This is called having a detached HEAD — it means the HEAD file points directly to a commit, not to a symbolic reference. The issue is that you generally don’t want to work in a detached HEAD environment, because it’s easy to lose changes. If you do an initial
submodule update, commit in that submodule directory without creating a branch to work in, and then run
git submodule update again from the superproject without committing in the meantime, Git will overwrite your changes without telling you. Technically you won’t lose the work, but you won’t have a branch pointing to it, so it will be somewhat difficult to retrieve.
To avoid this issue, create a branch when you work in a submodule directory with
git checkout -b work or something equivalent. When you do the submodule update a second time, it will still revert your work, but at least you have a pointer to get back to.
Switching branches with submodules in them can also be tricky. If you create a new branch, add a submodule there, and then switch back to a branch without that submodule, you still have the submodule directory as an untracked directory:
$ git checkout -b rack Switched to a new branch "rack" $ git submodule add firstname.lastname@example.org:schacon/rack.git rack Initialized empty Git repository in /opt/myproj/rack/.git/ ... Receiving objects: 100% (3184/3184), 677.42 KiB | 34 KiB/s, done. Resolving deltas: 100% (1952/1952), done. $ git commit -am 'added rack submodule' [rack cc49a69] added rack submodule 2 files changed, 4 insertions(+), 0 deletions(-) create mode 100644 .gitmodules create mode 160000 rack $ git checkout master Switched to branch "master" $ git status # On branch master # Untracked files: # (use "git add <file>..." to include in what will be committed) # # rack/
You have to either move it out of the way or remove it, in which case you have to clone it again when you switch back—and you may lose local changes or branches that you didn’t push up.
The last main caveat that many people run into involves switching from subdirectories to submodules. If you’ve been tracking files in your project and you want to move them out into a submodule, you must be careful or Git will get angry at you. Assume that you have the rack files in a subdirectory of your project, and you want to switch it to a submodule. If you delete the subdirectory and then run
submodule add, Git yells at you:
$ rm -Rf rack/ $ git submodule add email@example.com:schacon/rack.git rack 'rack' already exists in the index
You have to unstage the
rack directory first. Then you can add the submodule:
$ git rm -r rack $ git submodule add firstname.lastname@example.org:schacon/rack.git rack Initialized empty Git repository in /opt/testsub/rack/.git/ remote: Counting objects: 3184, done. remote: Compressing objects: 100% (1465/1465), done. remote: Total 3184 (delta 1952), reused 2770 (delta 1675) Receiving objects: 100% (3184/3184), 677.42 KiB | 88 KiB/s, done. Resolving deltas: 100% (1952/1952), done.
Now suppose you did that in a branch. If you try to switch back to a branch where those files are still in the actual tree rather than a submodule — you get this error:
$ git checkout master error: Untracked working tree file 'rack/AUTHORS' would be overwritten by merge.
You have to move the
rack submodule directory out of the way before you can switch to a branch that doesn’t have it:
$ mv rack /tmp/ $ git checkout master Switched to branch "master" $ ls README rack
Then, when you switch back, you get an empty
rack directory. You can either run
git submodule update to reclone, or you can move your
/tmp/rack directory back into the empty directory.