A lot of people want to setup a shared repository to allow a team of developers to publish their code on a server. This centralized approach is the model behind CVS/Subversion-style, but is also one of the many workflows supported by git.

Setting up such a repository in git involves a few simple steps. For simplicity's sake, I will assume that:

  • you have sorted out the UNIX permissions issues (e.g. putting all developers in a common group and adjusting permissions on the appropriate directories)
  • and that the repository is located on the same machine (replace /tmp/myrepo.git by ssh://hostname/repo.git if you need to access a different machine).

First of all, you need to create the bare repository where your users will push their changes and pull in updates:

mkdir /tmp/myrepo.git  
cd /tmp/myrepo.git  
git --bare init --shared

A bare repository in git is one that doesn't have an associated code checkout. It's essentially like the .git subdirectory of a normal repository.

Then you need to create a normal repository from which you will push to the central one you just created:

mkdir /tmp/myrepo1  
cd /tmp/myrepo1  
git init

You can't push yet because this local repository is completely empty (hence there is nothing to push). So let's create an initial commit:

touch test  
git add test  
git commit

Now, you can push your commit to the central repository:

git push /tmp/myrepo.git master

Other developers can clone the shared code using the usual command:

cd /tmp  
git clone /tmp/myrepo.git myrepo2

If you want to be able to track the central repository (i.e. use "git pull" and "git push" directly), you will need your local master branch to track the remote branch.

To do this, you can either throw away your local repo and clone again or simply run the following:

git remote add origin /tmp/myrepo.git  
git config branch.master.remote origin  
git config branch.master.merge refs/heads/master

That's it. If you want a longer explanation however, have a look at Aaron Toponce's tutorial or see the Debian Alioth instructions.

The initialisation of the bare remote repo is actually slightly easier than you suggest. Instead of doing "git init" to create your local workspace, do "git clone /tmp/myrepo.git". This will complain "warning: You appear to have cloned an empty repository" but your local repo will be set up as a slave (i.e. "git remote -v show" shows /tmp/myrepo.git as origin). This means you can commit to your local master branch, then "git push origin master" and everything will work as expected. Git even sets up the local master branch as a tracker for origin/master. Magic!
Comment by kbro