Matrix 2.0: The Future of Matrix

You and I exchanged 1:1 around 8 years ago on MatrixHQ, when this whole thing was still somewhat new but the cracks were already apparent. Then, the room was crumbling, federation wasn't reliable at all, the overall performance for basic use cases was terrible and self hosting on affordable hardware would mean not federating with matrix.org. When I brought those observations up, you (kindly and approachably) promised that a new design was in the works and weeks away (something about fanning out, or whatever). Fast forward to today, all the points above still hold true in 2023. From a distance, Matrix has had synapse, dendrite, better synapse with workers, conduit, new components added left and right (sliding sync being the latest), and I'm not even talking about the community at large, taken hostage of the uninterrupted pace of breakage. And I couldn't resist laughing out loud at fosdem when hearing this year once again that we are weeks away from what we were promised 10 years ago.

I am sorry to be blunt, but I've never seen a project of this magnitude lacking so much in structure and focus. I admit that I regressed to being a complete outsider, and from my perspective on the sidelines it looks like this has no end in sight: Matrix is getting more and more complex and hard for hobbyists to deploy at a small scale/local level, New Vector's hosted instances have a de facto quasi monopoly on the user base and on the protocol implementation. I'd be curious to know how this dissonance fits in your mental model, because this insane technical complexity was sold as the price to pay for not letting servers hold too much network power.

Anyways, I don't doubt that this message will be met with a well written and kind reply, listing a bunch of fringe projects and ongoing initiatives to tackle this or that aspect of the same old problem. See you again in 10 years, maybe :)

I'd be first to agree that Matrix hasn't evolved as rapidly as it should, and we (specifically me) could and should have been more focused. Building out future-looking projects like P2P Matrix or Low Bandwidth or Third Room has helped inspire folks to build on Matrix, but it's come at the expense of focus on the core spec & implementations. And that's one of the main reasons (besides funding) that we've shelved them for now. It's also true that 8 years ago, everything was flakey as hell.

However, whether you like it or not, we fixed it. The federation problems were resolved before the time of Matrix 1.0 back in 2019, and since then we've focused on making everything go fast too - e.g. Faster Room Joins in Matrix 2.0. We experimented with fan-out routing back in 2015 or so (as yet another future-looking experiment), saw it had potential, but at the same time saw that empirically full-mesh federation was good enough for the vast majority of cases. So we focused on fixing things like consensus bugs and lazy-loading joins rather than premature optimisation on full-mesh routing. Meanwhile P2P Matrix took over better-than-full-mesh routing, but practically speaking it simply doesn't buy enough to prioritise it relative to other stuff, and it certainly makes Matrix (even) more complicated.

> Fast forward to today, all the points above still hold true in 2023

This is demonstrably false, and it's particularly obtuse to be whining about it in the comments on a post on Matrix 2.0 which is literally dedicated to spelling out how Matrix has improved. Your complaint seems to be simultaneously that we've improved both the implementations (e.g. scalable Synapse) and the spec (e.g. sliding sync, removing Matrix authentication in favour of OIDC)... while complaining that the project hasn't improved since 2015. Personally, I think it's an extremely good thing that the new APIs start off as separate projects, which can evolve and release as rapidly as possible and benefit all homeservers, before being embedded into Synapse or whatever once they're stable. But it's also easy to spin this as "new components added left and right" as if it were a bad thing.

> New Vector's hosted instances have a de facto quasi monopoly on the user base and on the protocol implementation. I'd be curious to know how this dissonance fits in your mental model, because this insane technical complexity was sold as the price to pay for not letting servers hold too much network power.

There are >65k servers on the public network, of which about 2000 are run by Element (New Vector). All the biggest deployments (100K+ users) aren't run by Element (other than the matrix.org server), given the most people who pick Matrix do so because they want to run it themselves. Community services like Mozilla/KDE/GNOME are tiny by comparison, and they only end up hosted by Element because those projects don't care about self-hosting and would rather someone else runs it for them. They could equally well host with providers like https://etke.cc/ or https://ungleich.ch/u/products/hosted-matrix-chat/ or https://federated.computer or whoever; nobody's forcing them to use Element.

> Anyways, I don't doubt that this message will be met with a well written and kind reply

I'm sure you're a lovely person, but you come across as a complete asshole here, imo. Sorry you feel let down by Matrix; meanwhile most other people on the thread seem to think it has some merit. shrug

Telegram has been adding a ton of features during the last 5 years, all while keeping the core experience pretty much the same (fast and seamless). Whatsapp is also doing this.

With some UX Design chops, you should be able to do it.

Meanwhile, we massively fixed Element Mobile's UI a year ago: https://element.io/blog/an-unrecognisable-improvement-elemen... - and Element X is essentially the same UI... but on a different engine. So I'd argue that we were very aware that Element's UI was crap, which is why we fixed it. Rewriting the Element mobile apps as Element X is unquestionably the right thing to do in terms of sharing the same rust codebase between the two platforms and letting the app layer focus on UI.

Agreed that our progress has been slow, and our focus been has very broad at points. But we got there in the end, and certainly have learned from the journey (and built Matrix to be a much broader and successful heterogenous ecosystem as a result).

I'm sure your projects are all much more successful though - I'm jealous! :D