[alkonaut]

BitTorrent the protocol is decentralized but again when one of the largest torrent sites goes down it shows. That the protocol was decentralized didn’t make the usage pattern decentralized because it’s so convenient with centralization. And the same again with git: when github goes down we see that a distributed version control protocol isn’t very different from a centralized one if we all use centralized servers and usage patterns (which almost everyone does).

[davnicwil]

On the github point, agree that I've never come across a non centralised workflow with git - what are the decentralised git workflows?

The issues involved with everyone pulling from everyone, dealing with merge conflicts, etc in a distributed way with no agreed upon, central, source of truth seems fairly implausible for any reasonably sized team but I'd love to know if there are gitflow style workflows for this that make it easier than it first appears.

[danShumway]

> I've never come across a non centralised workflow with git

Even in centralized workflows, the decentralized core of Git makes things better.

You've never had a central repository go down and then pulled from the person sitting next to you? At the last organization I worked with, this happened multiple times. And when Microsoft bought Github and a bunch of people panicked, a lot of them were able to move to Gitlab (even without deleting their Github repos) specifically because the technology behind Git allowed you to add multiple remotes.

The article makes the point:

> No system is simply decentralized, full-stop. We shouldn’t expect any to be.

Git is the same way. Even with Linux, where patches are sent around using Email and there is no central repository that has every in-progress effort, there's still ultimately one person signing off on the final release. But that doesn't mean that there aren't decentralized parts of Git that are very, very useful, specifically because they are decentralized or because they allow you to fall back to decentralized workflows when necessary.

One of the things this article is getting at is that people treat decentralized as a binary state -- you either have one source of truth that coordinates everything, or you have complete chaos. Most decentralized systems are in the middle; they allow centralization, but also support fallbacks for when that centralization fails.

Git's branch strategy is another good example of this. Most Git systems use a master branch. But the value of branching is that you can have diverging code states that aren't yet centralized into a single location. You can fork off of master and have a completely parallel codebase, if for some reason you need to. You can have your release in a decentralized state for a week, where multiple people are working on separate branches and there is no single branch that has everyone's changes, and then you can re-centralize everything to master or a staging once you feel comfortable with the features you want.

[mcguire]

On the other hand, it's trivial to switch centralized cores for git workflows.

[grumbel]

If github goes down you upload your repo to gitlab and are back in action within 5min. The issue isn't git, but the fact that github does a lot more than just git (issue tracking, push requests and so on) and all of that other stuff centralized.

There is in theory nothing that prevents an issue tracker to be decentralized in the same way git is, but so far that hasn't materialized. You can export issues from github as .json, which allows you to transfer some things over, but users only have a username in those bug reports, not an email, which makes it impossible to contact them after a move to gitlab.

[johnisgood]

> BitTorrent the protocol is decentralized but again when one of the largest torrent sites goes down it shows.

When one of the largest torrent sites go down, the torrents do not go down with them. You can still seed them and indeed many people continue doing that. Others can still download them provided they have the torrent file or the magnet links. These sites are usually archived, you can still have access to such magnet links. Plus, there are many torrent search engines, and so forth. You can use as many trackers as you want, or none! These torrent sites are really just aggregators for better discoverability. Plus there are trackerless torrents nowadays that use DHT in which every peer acts as a tracker. This is called decentralized tracking, a trackerless system. I do not think that comparing this to git or GitHub makes any sense.