Use ssh to host a remote git repo

First, NEVER use a file synchronization service such as Nextcloud or Dropbox to synchronize git repos among multiple users. These and similar services should NEVER be used to synchronize git repo’s because they don’t synchronize everyone simultaneously.

For example, if I push to a local repo in my ~/Nextcloud directory, another developer could be pushing to their local repo at about the same time, and there will be a race between our Nextcloud clients to see whose updates make it to the Nextcloud server first. The best outcome is that Nextcloud will notify the loser of a sync collision (a file was modified locally at the same time it was modified on the server), and that the collision will need to be corrected manually. The worst outcome is that the loser will lose their commit entirely (their local repo will be overwritten by the copy on the server from the winning user), or the server repo will be corrupted with some files from the first user and some from the second user. Uggg! This is NOT what you want to have happen.

Only a git push command should be used to update an upstream server. The upstream repo will be locked and the push will be done in an atomic action. This is the only way to preserve repo integrity!

Here is my setup, note that it requires having ssh login capability to the server. I use FreeBSD for my server OS, but this procedure applies equally to Linux.

Step 1. Create a login user, the git repo will be stored in their user directory. You can put more than one git repo in the user directory, but all the repos will have common access (i.e. control is to the user directory, not to the repos in the directory).

Step 2. Create a public/private ssh key to access the git repo with (more correctly, to access the user account that has the git repos), and add the public key to the /home/username/.ssh/authorized_keys file.

You can use a single ssh key and give the private key to each real user who needs access, or better, each real user has their own ssh key and you add all their public keys to the authorized_keys file (this way you can easily remove access from a single user by removing their public key from the authorized_keys file).

Step 3. Create a bare repo in the user directory. E.g. sudo -u username git init –bare repo_name.git

Now you have two choices, depending on whether you have started writing code yet.

Step 4a. If you have not started development yet, commit a file from the server so the repo has at least one commit. Then from your dev workstation, clone the repo using the ssh protocol. After that you write code, commit locally, and push your commits to the upstream origin repo as you would with any other cloned project.

Step 4b. If you have started writing code already (i.e. if you already have a repo with commits), add an origin or upstream ssh protocol remote to your local repo and push it to the empty repo on the server.

This approach maintains complete control over the repo and is easy to setup. It has worked well for me with small projects and a few developers. Compared to GitHub, the primary technical difference is the loss of pull requests (which is a GitHub workflow, not a feature of Git), but in a small team everyone is likely committing directly anyway.

Happy coding.


Start a Windows MPLAB X Project by Cloning a Git Repo

MPLAB X includes great built-in support for Git but it assumes projects are created first from within MPLAB X (File > New Project ) before managing them using Git. This doesn’t help if you’re starting by cloning a remote repo.

I’m using MPLAB X on Windows. The standard Git install for Windows from the Git project includes Git GUI, a basic Git client GUI app. I’ll use Git GUI to clone the remote repo, and then use MPLAB X’s built-in Git client for everything else.

Start Git GUI

and select Clone Existing Repository.

Clone the remote master repository locally (the master repo for me is on a Windows network drive).

When the remote repo has been cloned, Git GUI will show the current status of the new local repository – which should show no modified files (i.e. no files either staged or unstaged).

To confirm the repository was cloned correctly, you can check the commit history by accessing menu Repository > Visualize master’s History.

Now we can open the cloned local repo using MPLAB X.

Git is primarily accessed using a project’s context menu from the Project window.

However, be aware there are a few Git commands under the Team menu, including a Repository Browser that isn’t available from the project context menu.

Try accessing Show History for the project (you will need to click “Search” in the Show History view that opens).

Now I’ll ready to start hacking on the source. I’ll keep it simple and just add some TODO’s to the README.txt file. Notice how the editor shows my changes compared to the last commit in real-time!

When I’m ready to commit my changes, I can commit just the modified README.txt file,

or I can commit all modified files in the project.

To push a local commit back to the upstream origin repo, use the general-purpose Remote > Push… selection from the Project context menu.

Origin is the only remote repo configured in the local clone, so it should be indicated automatically in the Push to Remote Repository wizard.

There, all done! Wasn’t that easy?

I didn’t do show it here, but standard practice is to always Pull from the upstream repo (e.g. origin) before you try pushing code. If someone has pushed code to the master repo since your last pull (or clone), you’ll need to merge their changes from the master repo into your local repo first, then push your combined changes back to the upstream repo.

Configuring Git

I needed to configure Git on a new server recently (no GUI), and couldn’t remember my typical configuration.

Disable Output Color-Coding

Many developers can’t live without color-coded command-line output, but I find at best it’s hard to read and distracting, and at worst absolutely incompressible with high ambient lighting and some screen glare. To disable color-coded command line output from Git:

$ git config --global color.ui false
$ git config --global color.diff false
$ git config --global color.status false
$ git config --global color.branch false
$ git config --global color.interactive false

Ignore File-Mode Changes

Git may report that executable files (e.g. shell scripts) have been modified based on differences in file mode interpretation between Unix and Windows systems. If the mode of a file is set to executable and committed to a Git repository in a Unix environment, and then the repository cloned into a Windows environment, the file will be reported by Git in Windows as having been modified – based on its mode. This is the result of subtle differences between a Unix file system and a Windows file system. Committing the “modified” file in Windows and pushing the repository changes back to the Unix repository will result in the file not being executable in Unix (until its file mode is set back to executable).

If this is an issue for you, set your Windows global Git config (~/.gitconfig) to ignore file mode changes (but first, check that your global configuration will not be overridden by a repository configuration).

Check your global and local configs:

$ git config --global core.filemode
$ cd gitrepo
$ git config core.filemode

Set configuration to ignore file mode changes:

$ git config --global core.filemode false
$ cd gitrepo
$ git config core.filemode false