I am no longer developing this project. For a promising alternative, see https://bazil.org/

---

Here is sharebox, a project of a distributed synchronizing filesystem.

The problem

We want a filesystem that will synchronize arbitrary data between several machines. This filesystem has to be simple, and to provide as efficiently as possible a set of minimum features:

• Offline access: it should be possible to modify data while not being connected to the internet, and get the data synchronized when the connection is restored.
• Conflict handling: if the same document is modified in several places, the system should keep both versions and yet provide a simple way to choose between the two versions. Ideally the remote conflicting version should be shown as a hidden file. Removing this file should resolve the conflict in favor of the local version. Moving this file as the new version should resolve the conflict in favor of the remote version. True merges should be avoided as they could be both confusing and difficult to handle. Solved conflicts should propagate: if the conflict between A and B is solved on A, B should follow the same choice.
• Versioning: It should be able to automatically keep several versions of document.
• Efficient storage: Versioning must not induce a forever growing size. It should provide tools to automatically clean the filesystem.
• Useability: Users should not have to be aware of the internals. Preferably, we should avoid creating special commands. Browsing the history should be done through regular 'ls' and 'cd' commands. Cleaning old versions of deleting snapshot should be doable through 'rm'. Moving content between machines should be doable through 'mv'.
• Battery friendliness: The filesystem should not force the user to run updates every time a file is modified. Instead, it should let the user schedule the synchronizations. Still, callbacks have to be provided in for those who wish to have these instant updates.

Planned interface

Here is how the user should see the filesystem:

mnt/
   |-regular files
   |-.sharebox/
              |-history/
              |-peers/

The test suite gives a sequence of use cases that may be interesting to browse if you want to understand how sharebox should be used in the end.

Internal storage

Storage should be done through git-annex. Default for each file should be to be versioned through git-annex. Transfers should happen only when files are actually opened (through the open() syscall). In order to allow the user to use git inside the filesystem, the .git directory of sharebox has to be removed from the tree. This is achieved easily by moving the files in a subtree:

fs/
  |-.git/
  |-files/-regular files (links handled through git-annex)