Subversion via Git

Subversion is a lot better than no version control system at all, but for those accustomed to distributed version control systems like Git or Mercurial, it can be pretty painful to lose features like cheap and flexible branching, intelligent merging and rebasing, and the snappy operations of Git when the need arises to work on code in a Subversion repository, perhaps for a legacy project or for an organisation with established repositories.

Fortunately, there’s an excellent compromise available in using the git-svn wrapper, which allows you to treat a Git repository as a working copy of a Subversion repository. This works transparently, meaning that others using the regular svn client on your team won’t have any difficulty working with your commits or branches, but your private Git workflow can meanwhile be as simple or complex as you like.


If you’re on Debian or Ubuntu, you can install the git-svn wrapper with:

# apt-get install git-svn

If you’re installing Git from source, it’s included in the default installation.

Cloning (Checkout)

To check out a copy of the trunk of the repository, you can clone it directly:

$ git svn clone svn:// project

This will provide you with a working copy of the repository’s trunk in the form of a Git repository, complete with a git log history imported from the Subversion commits. If you want to include tracking for the repository’s branches and tags as well, you can specify the paths to them on the command line:

$ git svn clone svn:// project \
    --trunk trunk --branches branches --tags tags

Typing git branch and git tag in the resulting repository will then show these as available branches and tags, as if they’d been created in Git. If the Subversion repository has a standard layout of folders, with branches, tags, and trunk, you can pass the --stdlayout or -s as a shortcut for the above:

$ git svn clone svn:// project --stdlayout

Fetching (Updating)

To pull the most recent changes from the Subversion repository into your Git “working copy”, use:

$ git svn fetch

If you have local changes in your repository that have not yet been committed, you may be prompted to temporarily cache them while you run the fetch operation. Git’s stash function works well for this:

$ git stash
$ git svn fetch
$ git stash apply


You can commit to your Git repository as normal with git commit. When you’re ready to send these commits to the Subversion repository, you can do so with:

$ git svn dcommit

This will also note information about the Subversion commit in the output of git log.


If you want to create a new branch in Git that tracks a similarly created branch in the Subversion repository, you can do this:

$ git svn branch experimental

It’s useful to note that if you don’t have any need to add the branch you’re creating to the Subversion repository, you can just use the usual git branch to keep a branch restricted to your Git repository. You should only need the git svn branch facility if others might need to use that branch before you merge it.


While it may be possible to conduct merges within the Subversion repository from the git-svn client, I think this particular task is probably best done using the usual svn merge tool. If you’re up to reading how this works in man git-svn, you should go for it, but I don’t think that getting a handle on the complexity of the rules for how these merges are run is really worth the effort in most cases, and given how brittle Subversion can be it’s likely not worth the risk of breaking things.


Aside from the above recommendation about using the native svn merge to conduct merges of Subversion branches, another limitation of the git-svn client is it ignores Subversion properties like the very useful svn:ignore, and furthermore doesn’t provide any way to set them. As a result, after the clone Git won’t know which file patterns a traditional Subversion working copy would be set up to ignore. You can emulate this by writing the properties file to a .gitignore or the .git/info/exclude files:

$ git svn show-ignore >> .gitignore
$ git svn show-ignore >> .git/info/exclude

Empty directories

Finally, Subversion and Git differ in how they treat empty directories. As a result, you may create an empty directory in a Git repository with intent to commit it into the Subversion repository, and find yourself unable to do so. The workaround here is to place some sort of file into the directory and commit that; a README file explaining the directory’s purpose seems to be a sensible choice.

These limitations, among others, show that the mapping from Subversion’s functionality to Git’s isn’t perfect, but for those who find working with Subversion a bit painful or imprecise, the functionality available in git-svn can certainly help, until such time as you’re able to convince your repository’s host to migrate everything to a more modern and capable revision control system.