I love the idea behind Matrix and dearly want it to succeed, but unlike IRC and XMPP as well as (to a lesser extent) email and netnews, it has the inherent problem that it implements a genuinely difficult thing: a large partition-tolerant distributed database without a central authority and even a bit of mistrust between nodes. Now that I’ve put this in writing, I wonder if there were any attempts to do this at all before the current generation of systems. (IRC and netnews basically YOLO their consistency away because they can’t afford it, which was indeed correct when they were designed, but people did a lot of distributed systems work between then and now...) That implies that a Matrix server is necessarily somewhat heavy and rather difficult to implement.

This makes me think that the federated network must by necessity exist in a fragile and tender state while Matrix is young, and need to be tended to with attention and care. The current stewards of the protocol don’t seem to be giving it that, growing features (even if good and competently designed ones) while treating the surroundings with something of a “they will come” approach. (To put it uncharitably: Look, there are now a whole two comprehensive server implementations, and the second one you could even afford a server for! Yeah, the Gnome people gave up on independently implementing a secure client, but surely an Electron one should be enough?..)

I’ve spent a very limited time studying the situation so I am not going to make any dire-sounding prophecies; besides, I actively don’t want any such prophecies to come true. What I am is worried. It doesn’t take much for a protocol to become a non-interoperable tangle of implementation details, and Matrix is an extraordinarily ambitious one.