Repoman: the life of a repo man is always intense

If you work across more than one machine, you know the shape of this problem. A repo you need is on the other laptop. A clone you thought was current is eleven commits behind. Something has uncommitted changes and you can’t remember which machine, or why. And the .env file that makes any of it actually run isn’t in git at all, because it never should have been.

Repoman is a single Go binary that keeps track of all that. It’s named after the 1984 punk/cult classic film by Alex Cox, and the joke stopped being a joke around the time it started repossessing things I’d forgotten I owned.

The manifest

One TOML file lists your repositories and which machines each belongs on:

[defaults]
root = "~/src"

[[repos]]
name = "acme-api"
remote = "git@github.com:you/acme-api.git"
hosts = ["desktop", "laptop"]

repoman init writes it for you by scanning ~/src. Run it on your second machine and it merges — repos already listed simply gain that host, nothing is overwritten. The file lives in its own (private) git repo, so repoman sync-manifest is how machines learn about each other.

The daily commands

repoman status --problems   # what's dirty, unpushed, or missing — in parallel
repoman pull                # fast-forward everything; skips dirty trees
repoman clone               # clone whatever this machine is missing

Moving a repo onto another machine is three steps that read the way you’d describe it out loud:

repoman assign acme-api     # I want this here
repoman sync-manifest       # tell the other machines
repoman clone               # bring it down

The half git doesn’t carry

Your remotes already protect your code, so Repoman doesn’t back that up. What it does back up is everything else a checkout needs and git deliberately ignores: .env files, local overrides, certificates, and the project directories full of PDFs and database dumps that were never repo material.

repoman backup      # restic snapshot of exactly that
repoman snapshots   # newest backup per machine; non-zero exit if any is stale

That last one matters more than it looks. A backup nobody checks is a backup that has already failed, so snapshots is built to be run by a cron job and shout when a machine goes quiet.

There’s also archive and restore for shelving a finished project — it takes a snapshot, removes the local clone, and marks it archived, then puts all of it back months later. It refuses to delete anything the remote can’t replace, and verifies its own backup before removing a single file.

Try it

go install github.com/klokie/repoman/cmd/repoman@latest
repoman init && repoman status

MIT licensed, works on macOS and Linux, no daemon and no service to sign up for. Source and README: github.com/klokie/repoman.