A non-federated decentralized social protocol based on Git | Hacker News Reader

A non-federated decentralized social protocol based on Git

github.com

103 pointspetar42 comments

I'm sorry but I believe I'm missing something here. How does this solve any of the problems that people commonly have with decentralized, federated social networks?

* You're still either hosting your own, or at the whim of whomever hosts your repository. Mastodon.social or GitHub.

* Hosting your own Git is not particularly easier than hosting your own GoTo Social or Akkoma.

* What if you end up with either a big following, or a big follower list? Aren't you going to be rate limited by GitHub?

* Is signing up for GitHub easier than signing up for mastodon.social, especially if Mastodon et al already have good mobile clients?

* What about moderation?

* What about media?

And I mean... Isn't Git federated by nature? Multiple machines store multiple copies of the data. That's defederation isn't it?

But OK, let's put the federated social networks we do have aside for a moment.

The site says: Every user stores their application state in a git repo they own and control.

But you don't do that if you're on GitHub, right? Not really, anyway. What is the benefit from doing this over git? What if I want to delete something? What is the overhead of the git protocol?

If it's just a toy, then that's totally cool with me. But it says that Microsoft Research is involved. I'm a bit confused. What is this?

Git makes it easy to make copies, but its nature of having one designated "origin" remote per repository that serves as upstream source of truth (on platforms like github also across forks) makes it not really federated/decentralized the way Mastodon et al. are.

And I don't reaaally see how this tool solves that problem. Or any of the other problems you mentioned.

But that said, Mastodon/ActivityPub don't really feel like mature solutions yet. It's incredibly opaque how instances interact, and self-hosting is an exercise in frustration as you try to figure out why A sees your reactions but B does not and you can get notifications for C's reactions but not A's, but B's replies and not… I gave up after a while. It's more opaque than even email self-hosting, at least exim and postfix have fairly verbose error logging.

> one designated "origin" remote

Git is not limited to a single remote. The default remote is called "origin" but there's nothing particularly special about it.

To be more precise, git clone adds the origin cloned from as a remote named "origin", unless you override this with a configuration setting or the -o option.

The default is no remotes at all.

But git gives you nothing to handle more complicated topologies, all remotes you interact with have to result in a linear history with everything on all other remotes… unless you feel like doing a manual merge every time you update your social media feed. Making that work automagically needs another layer on top of git, none of which seems documented here.

> Git makes it easy to make copies, but its nature of having one designated "origin" remote per repository that serves as upstream source of truth (on platforms like github also across forks) makes it not really federated/decentralized the way Mastodon et al. are.

That makes sense. Each Activitypub server might also be considered the source of truth for any given post, but since you're usually going to fetch them from your own instance that your account lives on you're not getting it directly from there unless you specifically do it yourself.

> But that said, Mastodon/ActivityPub don't really feel like mature solutions yet.

Yeah, that makes sense. I think this is partly because Activitypub is both quite opinionated, but also gives you quite some wiggle room with how you implement different functionality, like defederation. And of course the most well known software in that space is effectively what all others have to follow.

Git is probably a more stable protocol, but then again ActivityPub is served over HTTP, which is also pretty stable. I believe the same issues would befall this as well, since Git, to me, doesn't seem to be the biggest factor in this, and more the append-only nature of it. I've never been a big fan of append-only social networks.

I've been running a Mastodon server for a couple of years now and have had relatively few federation issues, luckily. But I do agree that trudging through the logs and figuring out what's going on is not exactly ideal, not to mention debugging other issues like if something went wrong during the key exchange with your server and a remote one, or if you accidentally lose your key and suddenly can't talk to any other instance since they won't trust your old domain with the new key. But you might run into similar issues with this, right?

Edit: If it came across as downplaying the project or it's author then that wasn't my intention at all. These were just some thoughts that immediately popped into my head when I saw this. They may well have thought through answers to my not thought through questions, and I certainly don't mean to pretend like I know better.

Kinda does look like round peg square hole situation.

"Federated network" could be modeled by a stream of posts (comments, tweets, whatever else) that are either a start ("root") of conversation or reply to one or more previous post(s), so it does form a graph. Then a bunch of "indexes" like "each post with this tag", usually ordered via some kind of filter (time, how popular the twat posting it is, removing any offensive or NSFW ones etc).

But there is really nothing to merge and not much else common to git aside from sitting on its store and protocol

> "Hosting your own Git is not particularly easier than hosting your own GoTo Social or Akkoma."

Easiest way I've found (so far) to "host your own git" is https://github.com/charmbracelet/soft-serve

>But you don't do that if you're on GitHub, right?

How do you push to github without a repo you own and control? Are people pushing patches through the web interface or something?

If you want decentralized without hosting your own servers, you can store data on a public blockchain. Each interaction pays a small amount of gas that grants them hosting indefinitely. Protocols are a progression from platforms since they can be ownerless. Blockchains enable protocols by acting as a general purpose database.

This is what I'm doing with https://nonphysical.systems

It's still under construction so it's on Polygon mumbai testnet therefore gas is free. The source is linked in the first message board.

It's a more difficult to use version of rss.

RSS really feels where the whole distributed thing peaked. I can subscribe to who I want where I want and if I want to see some new stuff either follow "what people I follow link to in their blogs" or get RSS channel of some topical federator.

All it missed was similiar distributed way to conversate instead of just reading and maybe leaving a comment on a blog

The author is well-known for creating Kademlia [1], a distributed hash table used to resolve BitTorrent magnet links.

[1] https://en.wikipedia.org/wiki/Kademlia

> Every user stores their application state in a git repo they own and control. No other infrastructure is necessary!

As a long-time user (and writer) of static site generators, I welcome these ideas, as I think these would allow us to organize static sites into some kind of decentralized social networks. Think of this as a bulkier version of RSS, because - for most practical cases - it is relative cheap to just fetch everything from a source and present it in a consistent way.

Even if people were using centralized git repositories, a separate naming service could allow easy (and cheap/free) redirects, so data migration wouldn't be an issue.

Can this be done with mastadon? Of course. But it requires a database server and other moving parts that need occasional maintenance. Treating the social network as a network of static sites makes operations (and federation) so much easier.

Would this be this specific project that wins us over? Maybe not, we shall see. But I think the concept could evolve into something useful.

I know Petar Maymoukov and met w him when we started designing Intercloud.

I wanted to know the design decisions behind how Kademlia worked in detail.

Glad to see this… wondering why we need it if we have Dat/Hypercore and beaker browser, which already uses Secure Kademlia DHT for swarming and follows up with SLEEP protocol instead of Git. It is encrypted and secure, to boot.

But those saying this is federated are wrong. It is peer to peer, just like Git is. Email is federated (the domain part is even a centrally controlled database). Having said that, git can be hosted even on your own computer!

Unless I misunderstand, this is federated - each instance stores data related to its user, and fetches from other instances the data it's interested in (from other users the user is following.)

It also doesn't do much to address the reasons users in federated systems tend to congregate around large(r) servers, like discoverability and reliability.

It's a neat PoC but I think the communication needs some work.

When the federation is per-user, that's typically called "peer-to-peer" rather than federated...?

Also, git doesn't work p2p, it uses centralized servers. If i understand correctly, the only real difference between Twitter and this is that you have your data downloaded, so switching to a new platform is a matter of adding a new remote and pushing it there.

Git was designed from the beginning to be used p2p. It’s how the Linux kernel development is done.

My understanding of this implementation is you can use arbitrary git urls to follow. That could point to any process serving data. P2P is 100% possible.

It would be smart to use a personal domain for the git url you give your followers, so you can update the hosting location without causing any disruption.

You can certainly work with git in a p2p fashion.

I'm pretty sure git-daemon does work p2p, but maybe that's a different p2p you have in mind.

If its based on git then more than likely, copies can be made and shared separate from the user instance. Users own their signature so changes cannot be forged but valid data can be proxied and cached. User data is fully portable from one vendor to the next.

Would you call git federated or decentralized?

So far as I understand, Git is both federated and decentralized - but I could easily be misunderstanding the definition of "federated" in this context.

The underlying software architecture and the motivation behind it are outlined in the whitepaper of the (more complex) gov4git project: https://github.com/gov4git/doc/blob/main/whitepaper/whitepap...

I am always highly skeptical of these applications that use git to store non-code data. Often many of the key features of git aren't applicable to other problem spaces and this makes everything less efficient.

There is a big exception here. You sign commits, so each update comes with some proof of the identity of the author. I like how simple this is, and generally prefer it over the way Mastodon and other federated services handle identity.

I certainly support the idea that alternatives to what we currently have are needed. But I must admit even as a software developer I feel confused when looking at https://github.com/social4git/social4git/blob/main/doc/walkt... how this even works.

Maybe some illustration would help to explain data and/or control flow?

The number of pilot users would still deserve growing: https://github.com/social4git/social4git/blob/main/doc/direc...

There are a few ideas that I really like here:

- Managing data (push/pull/etc) in a common, widely-supported protocol (git)

- which has a large, stable, and free-as-in-beer implementation (GitHub)

- as well as some major competitors (Bitbucket/GitLab)

- and is possible to self-host (gitea, OSS GitLab)

What I'd love to see:

- posts are currently in JSON format. Make it something more readable (like Markdown+YAML), at least optionally. That way people can interact _directly_ in GitHub, instead of using a janky client

- The main problem with decentralized frameworks is always "search and discoverability". Someone needs to aggregate all the data in a centralized database and make it easy for others to pick through

Are we reinventing Usenet?

Any time I hear the word "federated" in tech contexts I am led to presume someone is attempting to reinvent the wheel, and badly at that.

That's what it seems like: Usenet built on Git instead of NNTP, which was, itself, a reinvention of Usenet built on UUCP.

I still miss it

What's the advantage of this over a design like Scuttlebutt / Farcaster? They have the same properties (signed messages, Merkel trees to ensure data isn't changed) but are engineered for social media.

Nostr is another one that is more lightweight without the Merkel tree but is seeing some traction.

How do notifications work with this?

I’ll let you know.

But will I get notified?

See! It’s already working!

maybe by emailing pull request to all your contacts ;-)

Seems like it all gets wiped by one git force-push.

Reply on news.ycombinator.com