Rootgit is a format I've made to handle system configuration.

For quite a while now I've kept my user configuration in a git repository and kept things synced with a little script. Some time later, I decided my system configurations deserved the same, at which point I did various ad-hoc linking into git repositories. Then I decided to standardize what I was doing.

The script rootgit-linkup does symlinking (and moves any existing files that it replaces in $ROOTGIT/overwrites.rootgit/. IE the rootgit mirrors the root directory structure, and all files in the root that also has a corresponding file in the rootgit will be a symlink into the rootgit. Files in a .git directory or with a name or path that includes .rootgit are ignored when linking.

The script rootgit-run-scripts runs all executables in the rootgit named *.rootgit. The motivation of this is to be able to have scripts to set ownership and permission of the files, which can be run any time the git repository is updated. In other words, since git only preserves the executable bit of Unix permissions and all the files in the rootgit will originally be owned by root, yet you may want files in /etc or such to be owned by an app user or have different permissions (eg private keys or sensitive data), you can add these scripts to keep those permissions even when the git repo messes them up.

Each *.rootgit script runs with a $PWD of the directory it's in, so you can chown/chmod directly without absolute paths.

I've made the scripts work using relative links, and needing a path to the root so it can also work for chroot environments.

Sometimes you want to link a directory instead of a file - for instance if you want any files automatically added to that directory to go in your rootgit (and give you untracked file messages, etc). To do this, append .rg-linkdir to the file name. This extra extension will be stripped from the symlink name. So if you have a directory at /rootgit/etc/foo.linkdir (and your rootgit is at /rootgit), then you will have a symlink at /etc/foo that points to the directory.

If you really want the file to be copied rather than symlinked, you can append .rg-copy to the file name in the rootgit. This may be useful for programs that can't read symlinked files due to being chrooted or something.

Some versions of systemd apparently doesn't play nice with unit files that are symlinks, and will purposely fail. By this I mean that if you run systemctl <enable/start/etc> foo.service and /etc/systemd/system/foo.service is a symlink to something else (say, in a rootgit), it will fail. A workaround is to either use systemctl <enable/start/etc> <absolute-path-to-unit-file> or by setting the environment variable SYSTEMD_UNIT_PATH to include the path to the actual directory your unit file lives in.

I wanted a clean, standard way to manage this. I know there are big tools for managing server configs, but really I just want the simplicity of files in a git repo, because I'm only doing this for a few computers that have different purposes.

If you find it useful or want to improve it, have at it!

• Configurable name transformers. For example, a lot of files need to start with a dot in the real file hierarchy, but I want them not to be hidden in the rootgit. So I want to be able to say "DOT" or something will be translated to "." in the real tree.
• This and perhaps other per-rootgit configuration. It should go in $ROOTGIT/conf.rootgit.

MIT license -- do what you want, don't blame me for problems.