The slow enshittification of GitHub has had me thinking about trying to move off of it for my personal projects. I don’t think it’s entirely likely that they’re going to yoink my ability to have simple repos hosted there, but the experience of doing so just keeps getting worse.
But whenever I start pondering a move like this it occurs to me just how many things GitHub provides these days. Their business plan seems to be the one stop shop for everything in your SDLC, and they’ve gotten pretty far. In terms of features GitHub offers you’ve got:
That’s a lot of functionality to replace. Even if I don’t use all of that stuff, I use enough of it to make it hard to find a drop-in replacement. There are a couple platforms out there that are basically clones of GitHub, but that seems like I get all the same problems plus the cost of switching.
So I decided to take inspiration from Moneyball.
Guys, you’re still trying to replace Giambi. I told you we can’t do it, and we can’t do it. Now, what we might be able to do is re-create him. Re-create him in the aggregate.
Is there a single platform that will commit to all the principles of openness and sound UI that I value and also support all of the functionality that GitHub provides? No. Can I use a reasonable collection of such platforms/tools to get the functionality I truly need? Probably.
I think the convenience of a GitHub-style workflow for contributions is probably too practical to pass up. I have romantic ideas about old school distributed workflows with mailing lists and such, but there’s a reason very few new projects choose that model anymore. Plus, precious little of what I publish gets any attention from would-be contributors anyway, so it’s probably best to make it simple and recognizable for anybody that does want to send a patch.
Hosting an instance of something like Forgejo or SourceHut is feasible, but it’s a big lift and these kinds of platforms benefit from a network effect. It seems better to find one whose principles are in line with my own and chip in for the hosting costs and such. I think that Codeberg is attractive in that regard as they have a governance structure that seems like it’s designed for longevity and FOSS. However, being designed for open source, Codeberg does not generally allow private repos, which I do use for some things (infrastructure repos where I might have left secrets lying around, interview questions, a couple for private writing). For most of those the git repo is really just about having a backup, so I’d probably have to look into some alternative place to store those (or continue to store them on GitHub for the time being).
So an alternative forge gives me my “core” git operations, web UI, ticketing, and contibutor workflow. The next big feauture I cannot live without is some kind of CI/CD pipeline. I probably would like to bring my own worker(s) because the compute resources to run the jobs is usually what you end up paying for anyway. Woodpecker seems promising, but if it’s too much of a PITA to set up I might cave and run a personal Jenkins node. Either way the main need for me is just to make sure that things don’t get to run jobs unless I approve them, which both of these have the power to do.
Artifact publishing is prickly because it’s very much necessary but there’s a whole bunch of ways to do it and they tend to vary depending on what it is you’re building. Languages tend to have repositories for packaged libraries and sometimes packaged tools/binaries, like for example gems in Ruby. And then there’s also projects that can be distributed by language-agnostic means like container images, OS packages, and so on. Most of these boil down to some kind of binary file + metadata so it seems like it should be possible to distribute from one central service but there aren’t a ton of options. There’s some proprietary vendor products that support many different formats like Artifactory and then there’s the likes of GitHub which offer a similar service but hosted.
I think for this the best way forward is again to find existing stable platforms with a commitment to open principles. For now what I mostly need is publishing for gems and for container images (Go modules and Homebrew formulae both install straight from source control by default, which is nice). rubygems.org and DockerHub are not without issues but they have wide support and they allow free public hosting with minimal restrictions so they’ll do for now.
This is already my longest blog post yet and it’s really more of a plan for what I want to do than anything I have actually already done. It is nice to lay things out and see that what I truly need is doable though. More thoughts to come on implementation.