[dzaima]

There are some hilariously ugly things for even trivial things like editing/replies:

Edits get a `"m.relates_to": {"event_id": ..., "rel_type": "m.replace"}` field in the body, and `"m.new_content": {...}`, containing the plain-text and HTML versions of the message (while also keeping a copy of both, typically with a prepended "*", outside "m.new_content", for backwards compatibility; yes, that's 4 (four) copies of the message text in an edit; for some while Element generated up to 10 (ten) copies IIRC, due to some proposed extension(s), but that seems to be gone thankfully).

Reply messages get `"m.relates_to": {"m.in_reply_to": {"event_id": ...}}` - an annoyingly different format from the edits. It might look like that that allows an edit to change which message is being replied to, but nope, last I checked, that's not supported. Oh and for backwards compatibility a <mx-reply> HTML element is to be prepended, containing a copy of the replied-to message and its info (and yes, that means that the replied-to message effectively can't be deleted as the reply will still contain its text; luckily clients can choose to do not generate such <mx-reply>, but Element still does). And if you want to actually get proper reliable info about the replied-to event, you have to just make an API request for every single one you want to display (unless you happen to already have a cached copy, which luckily for replies is reasonably common).

And then there's threads - again in the name of crappy backwards-compatibility there's a mess - within-thread replies imitate a reply to the last in-thread message, along with `"m.relates_to": {"event_id": ..., "rel_type": "m.thread", "is_falling_back": true, ...}`, that "is_falling_back" indicating that this isn't actually a reply (being false when you want an actual reply). And clients are "supposed" to also handle replies to in-thread messages without the "m.thread" relation (which'd come from clients not supporting threads), but as far as I can tell there's no way do to it while paginating without making an API request per every single message (and yes Element behaves quite buggily here).

And then there are some things that can't be reasonably handled - the context/message listing/pagination APIs don't give any reaction info (besides the reaction events themselves in chronological position of their addition), so reaction presence/counts in history view must be calculated by clients, and thus won't be able to show ones that were posted a while after the messages. (there used to be some aggregation provided, but it's since been removed!!!) I think the only way to do this properly is truly just making an API request of "list reactions" for every single message the client wants to show.

This may make it seem like Matrix has an extremely firm stance on backwards-compatibility, but nope - recently they deprecated & made non-functional the unauthenticated API retrieval of media, making it impossible for clients not supporting video/audio/image display to just open those in the browser, instead forcing them to have custom file downloading & saving code, and also making it impossible to link to media within a message. There was a window of 6 (or less?) months between the new API being finalized, and the old one being removed.

[jeltz]

While you have a point with the ugliness of threads and replies I do not like your example with the authenticated media. The reason for the very short deprecation cycle was that it was deemed to be almost a security fix. The Matrix team did not like how their and other's servers were used as CDNs.

[dzaima]

Right; it's not unreasonable to want to quickly transfer away from it, but it's still a rather short deprecation cycle compared to everything else I've had to touch never being deprecated; it's not an "example", it's a real thing that actually affects my usage of Matrix (due to being now unable to post images inline in a message (granted, including fixed-URL links wasn't the prettiest thing, but it did work), and having to fully download videos before viewing them from my client that doesn't include video playback, whereas before I could immediately open it in a browser and view it while it's streamed in). Granted, then again, it's the only such occurrence, and probably there's nothing else that could match it in the future.

However many years ago I started work on my matrix client I was rather surprised it ever allowed direct links to media in the first place, but then again Discord had done the same mistake. But at least Discord's solution is more sensible, providing temporary links.

[ptman]

Many of these problems stem from the fact that matrix is federated. And it's a protocol, not really even for messaging but for a federated graph database. In federation some messages may be missing and arrive out-of-order. And different clients have different capabilities. There's a really simple example client implemented in a couple of lines of bash.

[dzaima]

No, most of those are just plain and simple bad outcomes due to not having thought out extremely basic things ahead-of-time. And even if there are problems with actually no simple fix, that in no way means the problem ceases to be a problem.

Replies to threads from thread-unsupporting clients are the one potentially-hard thing federation-wise, as doing it properly would require the server to trace back the reply and from that handling it as in the thread, but that's not far from what servers already have to do with edits. Or you could also just not require/suggest that behavior, having replies without thread metadata always be outside of threads; would probably save on confusion too, as the illusion would break anyway when someone on a client without thread support would not do a reply on some message after having used replies on others.

There might be some graph core to it, but it's still primarily a messaging service. Opening the client spec will show you tons of messaging-specific APIs.

I utterly disagree that saving a half-dozen lines in largely-useless toy clients is worth making more feature-complete clients more complex, and doubling the size of every edit event.

[self_awareness]

> In federation some messages may be missing and arrive out-of-order.

Doesn't "some messages are missing" trait defeat the point of a reliable communication protocol?

[ptman]

Missing at any given moment. There's eventual consistency. But sometimes connectivity isn't 100%