Reconciling cat and dog - the most productive git-svn setup

I would like to present productive and effective setup for the git-svn bridge which removes most of the obstacles and fixes some problems which occur with this bridged approach.


Before we start - one remark: if you installed git on your machine first time ever, please remember about setting up proper user, email and one-more-thing just for convinience:

~ $ git config --global user.name "Your Name Here"
~ $ git config --global user.email "your_email@example.com"
~ $ git config --global credential.helper cache


At first, you need to clone (or checkout in SVN terminology) repository. You can do in three different ways:

 1 # Standard layout.
 2 ~ $ git svn clone -s $SVN_REPOSITORY_ROOT_URL
 4 # Non-standard layout.
 5 ~ $ git svn clone $SVN_REPOSITORY_ROOT_URL
 7 # 'git svn clone ...' is basically a shortcut for these commands:
 8 ~ $ mkdir REPOSITORY
 9 ~ $ cd REPOSITORY
10 ~ $ git svn init $SVN_REPOSITORY_TRUNK_URL .
11 ~ $ git svn fetch -r HEAD

Updating and commiting

Besides pull and push (or update and commit in SVN terminology) everything works like in classical git. First line represents svn update command, second - series of svn commit commands with messages collected inside local git repository.

1 ~ $ git svn rebase
2 ~ $ git svn dcommit


If you cannot store your .gitignore file in the SVN repository you can exclude these files locally:

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

But .gitignore file can be very convenient - you can easily share exclusions with other members of the team which also use this bridged approach. Moreover, this command is really slow for bigger repositories, so that is another advantage of having this file.


If your repository does not have standard layout you can still easily associate SVN branch in your .git/config by adding new svn-remote section:

[svn-remote "sample_branch"]
    url = https://svn_repository/branches/sample_branch
    fetch = :refs/remotes/sample_branch

Then you should run (and develop the habit of running):

~ $ git svn fetch --fetch-all

As the last step, you need to checkout new branch:

~ $ git checkout -b sample_branch remotes/sample_branch

After that your branch will be available like normal git branch. Remember that your trunk is actually a git master branch.

Removing empty directories

Regarding the git-svn bridge, dangling empty directories are my nightmare. Fortunately, someone thought about it and we can enable deleting empty directories on commit (globally via git-config or on demand just for certain commit):

~ $ git config svn.rmdir true
~ $ git svn dcommit --rmdir


There are still a few problems with the newest SVN clients for such combination. The most annoying is assertion … failed related to the renaming operation. This error looks like this example:

~ $ git svn dcommit
Committing to http://...
    C      path/to/file/a.js => other/path/to/file/b.js
assertion "svn_fspath__is_canonical(child_fspath)" failed: file "dirent_uri.c", line 2502, function: svn_fspath__skip_ancestor

At first glance you may think that the only (and the worst possible - lost information and mess in history) option is to split your commit into two. Happily, you can perform your commit normally but you need to turn off rename detection:

~ $ git svn dcommit -C1 -l1

Unfortunately you will still commit two changes, this time in one batch (as a deletion the old file and creation file with the new name), instead of just a file rename, so you will still loose information in the process.

To avoid this completely the only option is to downgrade SVN client to the latest version on 1.7.X branch.


Big thanks to @skremiec for undeceiving me in some cases and pointing out svn.rmdir option :wink:.


  1. Manual for git-svn on kernel.org
  2. Chapter about git-svn in Pro Git book