How I've decided to coordinate multiple git repos for a single project

September 3, 2015

I'm increasingly using git for my own projects (partly because I keep putting them on Github), and this has brought up a problem. On the one hand, I like linear VCS histories (even if they're lies); I don't plan on having branches be visible in the history of my own repos unless it's clearly necessary. On the other hand, I routinely have multiple copies of my repos spread across multiple machines. In theory I always keep all repos synchronized with each other before I start working in one and make commits. In practice, well, not necessarily, and the moment I screw that up a straightforward git pull/push workflow to propagate changes around creates merges.

My current solution goes like this. First, I elect one repo as the primary repo; this is the repo which I use to push changes to Github, for example. To avoid merge commits ever appearing in it, I set it to only allow fast-forward merges when I do 'git pull', with:

git config pull.ff only

This insures that if the primary repo and a secondary repo wind up with different changes, a pull from the secondary into the primary will fail instead of throwing me into creating a merge commit that I don't want. To avoid creating merge commits when I pull the primary into secondaries, all other repos are set to rebase on pulls following my standard recipe. This is exactly what I want; if I pull new changes from the primary into a secondary, any changes in the secondary are rebased on top of the primary's stuff and linear history is preserved. I can then turn around and pull the secondary's additional changes back into the primary as a fast-forward.

If I use 'git push' to move commits from one repo to another I'm already safe by default, because git push normally refuses to do anything except fast-forward updates of the remote. If it complains, the secondary repo involved needs a rebase. I can either do the rebase with 'git pull' in the secondary repo, or in the primary repo I can push to the remote tracking branch in the secondary with 'git push <machine>:<directory> master:origin/master' and then do a 'git rebase' on the secondary.

(Using a push from the primary usually means that my ssh activity flows the right way. And if I'm pushing frequently I should configure a remote for the secondary or something. I'm not quite hep on git repo remotes and remote tracking branches just yet, though, so that's going to take a bit of fumbling around when I get to it.)

Written on 03 September 2015.
« Thinking about the different models of supplying computing
Consistency and durability in the context of filesystems »

Page tools: View Source, Add Comment.
Login: Password:
Atom Syndication: Recent Comments.

Last modified: Thu Sep 3 00:55:03 2015
This dinky wiki is brought to you by the Insane Hackers Guild, Python sub-branch.