Consider moving repository out of GitHub
Should we move the public code repository away from GitHub?
GitHub is owned by Microsoft and it's servers are in the USA. The political environment in the USA is rather volatile at present and an edict that cripples the project's public repository is well within the realms of possibility.
When raised on the Corporation BAC call earlier today, two committee members pointed out that the instability is already having an impact on them:
Randall noted that he's had to cancel a trip to the USA (from Canada) because travel insurance wasn't possible; and
James noted that they are moving their code to https://codeberg.org/ because they cannot risk having hosting issues.
It was also mentioned that:
OpenSSL has a GitHub Enterprise instance that is hosted in Europe (in Germany unless it's changed).
There are free options available that are not USA based.
Tim said that Cryptsoft's customers hadn't raised this as a concern.
This is unlikely to be a concern for USA based companies or institutions.
Cryptography has been restricted in the past and having restrictions re-introduced is very plausible.
An open source project isn't really a thing if it's not on GitHub. Slightly overstated but …
Even if migration away from GitHub doesn't happen, having a contingency in place would be prudent.
Thoughts folks?
Note: this is not about "Microsoft is evil, we can't trust them". All large businesses are evil and none can be trusted. It's highly unlikely Microsoft would decide to cripple GitHub on a whim. They would, however, comply with any edict.
I'll post this in the other BAC as a discussion since the overlap isn't 100%.
I don't see how this addresses the concern. Microsoft's reassurance isn't going to change their response to a request.
As far as we are publicly aware, there have only been a small number of large tech companies that have even contested government requests.
I don't think it's worth moving from GitHub - but definitely worth having several more backup locations.
OpenSSL has a GitHub Enterprise instance that is hosted in Europe (in Germany unless it's changed).
FYI, OpenSSL's github instance is hosted by Google. Anyone can see that with two commands:
- 'host github.openssl.org'
- 'whois {IP address}' (IP address taken from the previous command)
Do note that anyone is free to mirror github.com/openssl/openssl. In terms of resilience, we can set up clones of the repository anywhere, on any forge. It's in the nature of git.
This is not the actual problem.
The problem is redirection of forge specific features, such as handling issues and patches (pull-requests, merge-requests, ...), as well as workflows.
The effort required is something we do need to consider but it shouldn't be the sole (or even primary) concern in the decision. This is a BAC discussion, not a TAC one. We need to ask:
Is this a credible threat?
What's the likelihood of it happening?
What's the damage?
Basic risk analysis. My estimates of these are:
yes
insufficient information to assess properly
-
potentially catastrophic
@Paul Dale I think evaluating the costs of closing our GH repo(s) properly is the most important part of getting to any well informed decision.
From a very personal standpoint, I'm not convinced that another forge with the same centralizing format (which https://codeberg.org/ currently has, to be a bit crass, but I'm getting back to that). In the end, it's a matter of where things are hosted, and even if Germany (where codeberg is located) is a good enough country, we might find it displeasing some time in the future.
Rince.
Repeat.
With all that in mind, and if we're even considering these things, I would take a closer look at more decentralized models.
There is a model that relies on ActivityPub... it's unfortunately still more theoretical, or at most experimental. I know that Forgejo, which is the forge implementation that Codeberg uses, is working on an implementation, and it seems to be going forward, but development takes its due time, and they aren't there yet.
There is a model that relies on a mix of gossip protocol and essential git protocols for replication, implemented in Radicle. It's still young, so while it has all the basic features (issue and patch handling), it has some rough corners.
There is a model on top of ATproto (the protocol that drives Bluesky, among others), called Tangled, which seems to make some strides forward, but I haven't looked more closely.
I personally find this an interesting field, well worth a look, but can understand if we conclude that "yeah ok, interesting and worth keeping an eye on, but perhaps not quite going there yet".
Just to be clear: all of what I said above are personal reflections, not to be misunderstood as statements in a Foundation capacity.
Exploring other options apart from Codeberg is prudent, that was just an example not a commitment. Something more distributed would be good. Nonetheless, this is premature IMO.
For now, we need to figure out if we want to start looking into such a transition. Then we can figure out the specifics and the level of pain (i.e. cost) we're willing to endure to enact it.
This hasn't been discussed for a while... meanwhile, I see different projects move away from github (for example ziglang, just a few days ago), while others choose to remain (Daniel of curl discussed this publically almost four months ago, albeit he also chose to set up a codeberg mirror a few days later).
As far as I can read the space and diverse decisions out there, there seems to be one common major sticking point: CI. OpenSSL uses github's CI infrastructure a lot, with github hosted runners (we did have a small amount of self-hosted runners at some point, but it seems that github is now hosting everything we need). Moving to another forge is very likely to mean dropping quite a lot of CI runs, or self-hosting a lot more of that, or relying on a network of volunteer worker nodes. Is that worth the while? I'd say that's one of the bigger questions.
These are only my views an opinions, I haven't asked anyone else, albeit I'm writing with my Foundation hat on for perspective.
This being the Foundation BAC, I gotta ask you guys, is this something you have deliberated with your respective communities? Is there a collective opinion in the matter, with consideration for pros and cons?
Perhaps, given how large the OpenSSL project is, and how important it is to Microsoft and GitHub, that we can get a commitment statement from their senior management.