CodeFS is a filesystem for remote software development.

Remote development is a key component of software engineering. Whether developing for embedded devices or building large ML pipelines in the cloud, one often finds oneself needing to work jointly on a local laptop and on a desktop, embedded device, or virtual server.

There are already several approaches for this, here is a breakdown of their pros and cons:

CodeFS brings the POSIX interface of sshfs with the speed that comes with a dedicated server process.

CodeFS is in alpha testing. You should only use this on directories that are source controlled and you should git push often.

Use homebrew:

brew cask install osxfuse
brew install --HEAD MisterTea/codefs/codefs

First install the dependencies (either from a package manager or source):

1. Boost
2. Protobuf
3. GFlags
4. ZeroMQ
5. fswatch
6. FUSE (or OSXFUSE for mac) for the client. Build with -DBUILD_CLIENT=OFF to skip the client if you cannot install FUSE on the server.

Then:

git clone https://github.com/MisterTea/CodeFS.git --recurse-submodules
cd CodeFS
mkdir build
cd build
cmake ../
make -j4
make install

Note that, as of now, there is no security or encryption. This means that port 2298 should not be exposed to the outside. Instead, codefs should be run over a secure layer, such as Eternal Terminal: https://github.com/MisterTea/EternalTerminal

To connect to the server with port forwarding:

et -x -t="2298:2298" my_server.com

Then inside the et/ssh session, run:

codefsserver --path=/my/code/path --logtostdout

Where /my/code/path is the location of your code. For now, the server needs to be restarted every time the client (re)connects.

On the client, run:

codefs --path=/tmp/my_development_path --logtostdout

Where /tmp/my_development_path is some empty folder that will act like a mirror to the folder on the server.

In 0.0.1, the server needs to finish indexing the entire directory before the client will be able to connect to it.