I'm moving my projects off GitHub
Wednesday, November 16, 2022
It's time for me to leave GitHub behind and move to another forge. I'm not necessarily advocating for anyone else to do the same, but if my reasons resonate with you then you may want to consider it. I also don't expect this post to... matter, if that makes sense1. I'm not a major open-source maintainer or contributor. I'm just somebody who likes to write code and likes to put it out there.
So, why am I moving my projects off of GitHub?
My Issues with GitHub
It ultimately comes down to some issues I have with GitHub, both as a product and philosophically.
The tangible one that tipped me to finally move: I'm upset about GitHub Copilot. It's fairly well known that Copilot can reproduce significant pieces of open-source code, stripped of their license2. I'm moving to make it a little bit harder to have Copilot train on my code. This is perhaps a futile protest, but it's what I can do an individual. Writing about this is another aspect of what I can do.
I hope that the ongoing litigation gets us some clarity in what is legal here. In the US, a lot of "is this legal" is deferred to the run-time execution of contracts by courts, so this is our chance to find out what is legal or not. Hopefully this goes in the direction of defending open source and requiring attribution and copyright. As it stands, Microsoft/GitHub have basically washed their hands of it for users, saying reproduction of code is rare and the users must make sure it doesn't happen. Which... they're supposed to do how, exactly?
Anyway, I don't want this post to be a full-on rant about just Copilot (I've got plenty of other things to rant about 😉) Copilot was just the tipping point. There are plenty of other issues I have with GitHub which are more significant for the decision to leave.
First, I think that open-source code should use an open-source forge. It's deeply ironic that the biggest forge for open-source code is itself proprietary. (And ironic that one of its biggest competitors, GitLab, is open-source and hosts tons of proprietary code.) I think this is not healthy for open-source in the long term. It gives a lot of control over open-source to Microsoft, and concentrating that power in one entity is not good, regardless of who that one entity is. I think they've mostly acted as good stewards so far, and this is about mitigating a risk and addressing a philosophical issue.
I also don't like how GitHub changes my behavior. This one is somewhat on me (personal responsibility) but this also comes down to how many modern tools are designed. Modern web design leverages and exploits human psychology to achieve the outcomes it wants (ultimately, increasing revenue, usually by driving usage). GitHub in particular is pretty effective at doling out dopamine hits to me. As a GitHub user, I was always seeking green squares to try to make sure I had activity every day. This led me to change my workflow to generate more green squares, not for whatever is maximally effective. Having visible stars and followers also turns it into a sort of popularity contest. There's a dopamine hit when you get one of those, so it creates a strong reward function for attention-seeking behavior.
Tools should be designed to help the user, not to help the company3. Many of GitHub's features can be defended (and I'm sure readers of this post will do so!). They certainly don't work for me, though. Using GitHub changes my behavior in a way that, ultimately, I find to be negative. So: bye, GitHub!
Another small reason is that I believe in paying for my tools. This is why I pay for email with Fastmail instead of using Gmail. The incentives are clearer when you're a paying customer than when you're a free user. I mean, they have to make money off of you somehow. The innocent explanation for free usage is as a loss leader to funnel people into an eventual enterprise sales cycle. The cynical explanation is to take more control over open-source and also to make a massive dataset to power Copilot and other products. If you're not paying, you're the product, after all.
Why I Chose Sourcehut
After deciding to leave GitHub, I had to pick a new forge. I settled on Sourcehut for a variety of reasons.
The non-negotiable criterion was that it's open-source. The platform itself is licensed under the AGPL (mostly) and you can self-host it. They also provide consulting on open-source projects to get some revenue, and they don't require copyright assignment for contributions to Sourcehut from volunteers or employees. All of this is pretty strongly aligned with my philosophy and I appreciate it.
Another big reason is that Sourcehut is explicitly designed to not dispense dopamine. My brain's response to dopamine is widely exploited by our industry, which is why I'm not on any social media anymore. GitHub dispenses a lot of dopamine and it makes me change how I work to get those nice little green squares. Sourcehut rejects features that are only for dopamine hits without utility on their own, and is generally designed in a humane way that doesn't exploit human psychology. This makes my life tangibly better.
I also feel like the platform's direction is understandable. Some people criticize the maintainer (Drew DeVault) for having strong opinions. He does have those, and Sourcehut reflects it. This lets you have a pretty good understanding of where things are going and what he won't compromise on. In contrast, with a proprietary platform like GitHub, you can't quite be sure of the long-term direction. It depends on the company strategy and what metrics they're trying to optimize and who's making design decisions. Drew is transparent with things and even though he has strong opinions, it's a big tent. Unless you're working on blockchain projects that waste energy for imaginary magic beans. Then you can get out of this tent—and indeed, your data is portable, and you can easily migrate off to somewhere else!
As an example of how much lighter it is, let's look at one of my repos on both platforms.
- On GitHub, my config repo takes 933 ms to load and downloads 2.5 MB.
- On Sourcehut, my config repo takes 158 ms to load and downloads 138 kB.
One aspect of forges today is CI or build tool. Having CI on the platform was an important part of me moving to Sourcehut (where Gitea, for instance, doesn't have CI built in). The delightful surprise was how good Sourcehut's builds are. I find GitHub Actions fairly confusing and difficult to get working. It feels like whenever I'm updating a GitHub Action, I have a string of ten or more commits that are all of the form "whoops, now does it work?". The only way to update it is to keep pushing to the repo, so you're left with a string of ugly commits while you iterate on your CI. Maybe I'm just dense, but this is an experience I've heard from others as well.
In contrast, Sourcehut's builds are quite easy for me to understand and use. You write a minimal YAML file describing the build (which OS, what packages you want installed, and a list of shell scripts to run) and then you get a VM that is running that for you! The model itself is pretty easy to understand, and the absolute delight is how easy it is to debug builds. With GitHub Actions, you keep pushing commits to debug. With Sourcehut Builds, you can do that, but you have two more powerful options:
- You can run ad hoc builds. This lets you keep iterating on a build and test it before you commit at all, which keeps your commit history clean and is a much nicer workflow.
- You can SSH into your build VM. If you have a failed build, the VM sticks around for a bit and you can connect in to try things and figure out what you should change to make the build work next time. This is such a helpful tool.
The elephant in the room with Sourcehut is, of course, its contribution workflow. Projects on Sourehut use git-send-email to accept and review patches. In fact, basically everything on Sourcehut can be done through emails, like discussions and issue handling.
I think that this is Sourcehut's biggest weakness in getting people to switch, because the flow is different and often unfamiliar. I don't think that it's harder than the PR flow, but it's significant friction. It makes intuitive sense for discussions and issue reporting, though, and I hope people will give it a shot.
It took me some time to get comfortable with the
git send-email flow, because it's different and it was intimidating at first.
Ultimately, though, I'm a big fan of the email workflow.
This lets me spend less time in my browser, which is a big win for someone with attention/distraction issues.
It's easier to stay focused on submitting, reviewing, and merging patches when I'm doing that in a dedicated email client instead of in my browser, one click away from dopamine.
The submission workflow is probably not a big deal for the projects I have on Sourcehut. I've gotten a few issue reports on my repos of the years, and a sprinkling of contributions, but they're far from having active communities. I do hope that the chess database I'm working on will be able to grow a community, if it ends up being useful, and I'm looking forward to growing that on Sourcehut.
Ultimately, though, I don't see hosting on Sourcehut as being an impediment to contribution.
There are many ways to contribute and you don't even have to use
git send-email if you don't want.
You can email in the patch, of course.
Or you can mirror the code to another forge and say "please pull it from here".
In either case, contributors don't need yet another account to make a contribution.
This was one of my factors in switching, because I don't want to force people to create more accounts on more platforms.
Why not X?
There are a few other major forges. It's worth talking about why I didn't pick those. This just deserves a few bullet points.
- it's super slow in my experience
- the features always felt half-baked and there are a lot of features
- it's very similar to GitHub and I wanted a change
- contributors have to make yet another account (or OAuth with a GitHub account, I guess? meh.)
- reasonably fast, actually! seems good on that front
- doesn't have build support for CI that I could find
- feels like GitHub Lite
- contributors have to make yet another account (or OAuth with a GitHub account, I guess? meh.)
When I'll still use GitHub
I'm going to keep my account on GitHub, and I'll still use it for three reasons:
My workplace uses GitHub. I like my job, and we use GitHub, so I will of course keep using it there. There's no reason to push for a change. It's a perfectly fine platform for our needs.
To make contributions to projects on GitHub. My projects are going to be hosted elsewhere but sometimes I'll contribute to projects on GitHub! Maybe bugfixes for things I run into at work. Maybe contributions while pairing at Recurse Center. But occasions will arise when I make contributions to projects on GitHub, and that's going to use my GitHub account.
I have a few projects I'm not moving. Old projects that aren't active, I'm not really investing the time or energy in moving. If these ever have activity, then I'll of course use my GitHub account to manage them.
Should you leave GitHub?
If this post resonated with you and you're thinking about leaving GitHub behind, you might want to! You can always diversify your forge use without committing 100% to leaving one behind and using the other. That's what I'm doing.
But there are also some very legitimate reasons to stay on GitHub as your primary forge. If you don't have philosophical objections to staying, it's the place to be. For worse, it's the place that employers expect to find activity from candidates. If you're looking for a job, an active GitHub profile can help with that. (If you're responsible for hiring decisions: Please let's talk about changing this!)
I think more people diversifying their forge use would be good for the industry long term. GitHub controls much of the open-source world and they also control much of the software industry. Regardless of their current actions, this is a major risk in the long term.
I don't fault any individual for staying on GitHub. But let's normalize using more forges. If you have the ability to switch, please consider it!
So why write a blog post if I don't think it'll "matter"? Because this is my blog, and I write what I want! I'm not writing to achieve any goal or effect change. I do think it will be interesting to someone else, and writing publicly is a hobby I enjoy.
See Drew Devault's post "GitHub Copilot and open source laundering" for a good post on this topic.
This is another reason I want to be on a FOSS forge. Incentives are better aligned when there is less profit motive and better philosophical alignment. Everything boils down to behavior and incentives.
If this post was enjoyable or useful for you, please share it! If you have comments, questions, or feedback, you can email my personal email. To get new posts, subscribe to the newsletter or use the RSS feed.
Want to become a better programmer? Join the Recurse Center!