[Pneumaticat]

Matrix isn't really hostile to competing server implementations. If it were, then they wouldn't have spent a long time and a lot of effort to finalize the Server-to-Server spec r0.1, which anyone can implement [1].

And just to add my personal anecdote, Matthew (CEO of New Vector, co-founder of Matrix.org) is present quite often in the Matrix HQ chat room, and willing to field questions from anyone. Perhaps with slightly less willingness when people start such accusations.

Also, to address some of the technical issues, I think what they're talking about as "The blocks have hashes but their implementation does not check the hash" in the DAG is being fixed with the next (v3) room version [2]. Again, this development is all being done in public: see e.g. https://github.com/matrix-org/matrix-doc/pull/1659 for the v3 room proposal.

[1]: https://matrix.org/blog/2019/02/04/matrix-at-fosdem-2019/ [2]: https://matrix.org/docs/spec/rooms/v3.html

[fro0116]

To quote one of the follow up comments from the author:

> If you go to matrix.org and look at the list of about a dozen or so servers: you will find that none of them actually work except the reference implementation, and maybe sometimes Construct. Even thus, the phrase "able to build" is questionable. I have spent months reverse-engineering their software and its interactions before, and after, it was at all documented in this so-called standard (by the way, it's just documentation of their software -- errata and all (and rather poor)).

> Construct server is the single survivor out of the ones listed and even more who have attempted and given up early which we don't know about. That being said, it is still incomplete.

A spec that "Anyone can implement" doesn't have much value if it's so bad/incomplete that in practice nobody else is actually able to create a complete working implementation of it as the post claims.

And regardless of what they're planning on doing about any security issue, there's never a good reason for a discussion of a potential security issue in an "open" protocol to end in "good luck talking to your own federation." That's the behavior I'm pointing to that's hostile to competing server implementations.

Of course, the quote could be taken out of context, or straight up made up. But until proven otherwise, I'm willing to give the benefit of the doubt to the small independent developer who seems to genuinely care enough about the openness of the protocol to build their own server implementation.

[Arathorn]

The "good luck talking to your own federation" comment is being taken out of context. I can't find it in my history (so I assume it's been paraphrased), plus from memory I think it was the final line in an argument over how some bit of the protocol should be implemented. After failing to convince the guy that our approach was better than his proposal, I gave up with "okay, you can always do it your way, but you'll only be speaking to yourself" or something similar. I'm not sure this is evidence of being hostile to alternative servers, but if anything an encouragement for people to experiment with variations on the spec if they think they can do better :|

See https://news.ycombinator.com/item?id=19418111 for a detailed response to these accusations.

[StudentStuff]

It'd be great if there were usable non-Riot clients that fully supported end to end cryptography. At this point Signal has better alternative clients in Pyre.chat and Signal-Weechat than Matrix has, as the existing alternative clients are almost all MVPs missing major features like e2e.

The lack of a simple, common library is a real sticking point. Each client & bot shouldn't have to implement the baseline network protocol and then build e2e atop it, rather the clients should be focused on what to do with the messages after receiving them, how to display them, etc.

I say this as a daily non-Matrix.org user of Riot, the functionality of the Riot ecosystem on its own is good, but I'd be hard pressed to port my own Signal bots to Matrix, let alone build a featureful client in a performant toolkit (eg: Not Electron) for Matrix today.

There isn't a viable ecosystem outside Riot/Synapse currently, which makes it feel like Matrix is just a self-hosted Slack/Hipchat solution with just Riot as usable clients. IMO there needs to be diversity in the Matrix ecosystem just like the Fediverse has with Mastodon, Pleroma, PeerTube, PixelFed, NextCloud, Rustodon, etc all inter-operating in a sane, functional manner.

[Arathorn]

So, other clients with fully working E2E encryption include: Seaglass, Weechat, Pattle and nheko (although it needs to implement attachments, but they is easy). The weechat and nheko implementations are independent implementations (ie not the same as the Riot ones).

We are starting work this week on closing the E2E gap for all the rest.

[jcgruenhage]

I don't think Pattle has e2ee support, from what I can see on my phone. While it is built on the Matrix Android SDK which has e2ee support, the app itself doesn't seem to use that support yet.

[mindB]

Oh neat. Is the plan to write a proxy sitting between client and server and decrypting/encrypting all events so any client can be used in e2ee rooms?

[Arathorn]

precisely. should have news about it later this week :)

[feanaro]

There's weechat-matrix[1] which supports E2EE and the accompanying python library, nio[2].

[1]: https://github.com/poljar/weechat-matrix/

[2]: https://github.com/poljar/matrix-nio/

[iknowstuff]

Good idea. A library written in Rust and compiled to WASM for Riot would be a good fit.

[maxidorius]

You're right, we need to do better, and we have started doing so under a fork of the protocol that will respect its user's privacy: The Grid - Come talk to us at #thegrid:libremonde.org

[ADogWithPurpose]

(Full disclosure - I'm contracted by New Vector to work on their integration services, but the below experiences are about my open source work which is unfunded and unrelated to my day job)

I wanted to chime in as another developer in the server area. I've recently been working on a service that sits atop of an existing homeserver (like Synapse) to handle outgoing federation in .NET rather than Python (which has proven to give quite an nice performance boost). The project implements it's own federation and signing code.

So far I've had to do almost no special casing around Synapse and followed the new S2S spec without issues, which has all been done in the open. The special casing I did have to do was brought before the team, judged a Synapse bug and will be fixed in an upcoming update.

I don't have any evidence to point to whether the spec can be used to write a complete homeserver implementation, but everything so far has been done painlessly following the docs online. I believe that I could, with enough time.

The project is developed by myself and another, and we've experienced no hostility from the team when asking questions about the protocol or the above bug. Quite the contrary, they were quick to help and the project has been promoted by matrix.org every week in the news blog posts.

On the project itself, we hope one day to branch out from working on services that sit on top, to being a standalone homeserver (which should progress naturally. I don't foresee any spec related problems with doing this from reading the spec and chatting to folks in the community.

So I felt it would be worth sharing my experiences as another server developer, albeit one that hasn't gotten as far in their journey yet.

[mindB]

The server-to-server specification was only formally released on February 5 of this year. Before that, it was certainly necessary to do reverse-engineering of synapse in order to build a HS, but it no longer is. The lack of server implementations just over a month after the release of the spec isn't good evidence of a bad spec. I strongly disagree with Jason's (who has been openly hostile to most of the matrix community including me) characterization of the spec in its current form.

Disclosure: I like matrix and have hacked on a few matrix-related projects as a hobby activity (Jason would refer to this as buying into Matthew's (the CEO of New Vector, the company primarily driving matrix) personal cult). That's it.

[maxidorius]

As the author of mxhsd [1] and Gridepo [2], I can sadly confirm Jason's claims that it is STILL necessary to reverse engineer to talk to the existing federation.

Synapse doesn't follow the spec, which is a fact I'm experiencing every day. But more to that, there is no desire from its developers to ever comply with the spec, rather than changing the spec on a whim.

And that are the reasons why making a HS is not possible: endless moving target.

[1]: https://github.com/kamax-matrix/mxhsd [2]: https://gitlab.com/kamax-io/grid/gridepo

[Arathorn]

if anyone's curious about the unspecific complaints here, it transpires that the actual complaint is synapse had a bug where it some endpoints were called with a trailing slash when the spec said it shouldn't have one... it got fixed in https://github.com/matrix-org/synapse/pull/4840 and https://github.com/matrix-org/synapse/pull/4935.

[Pneumaticat]

At least from my perspective (I've never implemented the spec), I wouldn't call it just documentation [1]. Sure, there are likely some holes, but overall it seems like they've put a pretty good-faith effort into making it detailed enough for someone to implement.

I feel like the difficulty comes mainly from the breadth/size of the spec. Which is a major issue (the Matrix.org team's own alternative server, Dendrite, has been slow to progress as well), but ultimately I'm not sure what they can really do about that considering one of their key tenets is to contain all the functionality out-of-the-box, so to speak (no extensions like XMPP).

[1]: https://matrix.org/docs/spec/server_server/latest.html

[maxidorius]

I am the author of the now-frozen mxhsd project [1] and its resurected version, Gridepo [2]. I have contributed several times to the specification, reported issues for the past two years and still, to this day, I can confirm that it is not possible to implement a Matrix server with the current specification that can talk to the reference implementation.

Most fundamental problem: synapse doesn't sign the events like the spec tell us to do it, making every event invalid. Whatever you sent that is spec compliant will be refused by synapse.

So talking "synapse" is possible. Talking Matrix and be able to federate with the ecosystem is not. I wish people would not blindly believe what a for-profit is saying, especially one that is in full control of the said protocol, and start checking for themselves.

[1]: https://github.com/kamax-matrix/mxhsd [2]: https://gitlab.com/kamax-io/grid/gridepo

[Arathorn]

I'm not aware of a bug in how synapse signs events, and whilst it's possible there's a bug there, we can't fix stuff which we're unaware of which hasn't been disclosed. (Max would probably claim he can't disclose because he's been banned from our Github for consistently being obstructive; unfortunately this is a case in point).

Edit: he might be talking about https://github.com/matrix-org/synapse/issues/4787? Except that was fixed a month ago...

In terms of "don't blindly believe what a for-profit is saying, especially one that is in full control of the said protocol", I'd agree, were it true. However:

* These days the protocol is controlled entirely by the non-profit Matrix Foundation.

* The majority (3 out of 5) of the Foundation's directors have zero affiliation to New Vector, the company that the core Matrix set up to try to keep the project funded.

* The governance of the foundation & protocol is entirely open, as per https://github.com/matrix-org/matrix-doc/blob/matthew/msc177...

* We're announcing the final board once formally appointed; due in the next few weeks.

Meanwhile, it's also true to say that you shouldn't believe the skewed perspective of a disgruntled ex-community member with a chip on their shoulder, who is trying to plug their hostile fork: https://news.ycombinator.com/item?id=19473694.