spec is missing documentation of /_matrix/federation/v2/send_{join,leave}

[turt2live]

This is due to a legacy bug in Synapse from before the federation spec was a thing. It's correctly specified in the spec, but we need a /v2 endpoint of some things to remove the thing - similar to the invite endpoint.

This issue primarily exists as a place to point people who notice the weirdness.

[aaronraimist]

Wasn't that #1418?

[turt2live]

indeed, I knew I wasn't crazy in thinking this was fixed. Looks like it needs a spec PR - I'll keep this open as a personal reminder to write that spec before we cut a release (as clearly it being a merged PR with a label isn't good enough).

[richvdh]

as clearly it being a merged PR with a label isn't good enough

if that's true, isn't it also true of other MSCs under spec-pr-missing ?

[iinuwa]

I think I'm a little confused on this issue. So this is a spec bug and not actually implemented in Synapse/Construct? So I should use just res instead of [200, res] as the response body in the v1 endpoint for our server library?

[richvdh]

the v1 endpoints return [200, res]. There also exist v2 versions of these endpoints which just return res: v2/invite has existed since s2s spec r0.1.0, and MSC1802 proposes the introduction of v2/send_join and v2/send_leave, though these have not yet made it into a released spec (though this is now just a formality).

Recent versions of synapse implement all three v2 endpoints, as well as the v1 versions for compatibility with older implementations. For outbound requests, Synapse will first try the v2 endpoints and then fall back to the v1 versions for compatibility with older servers.

I can't speak to what construct does here.

To be spec-compliant, you should follow Synapse's behaviour.

If this sounds like a mess: yes, fixing bugs in federated protocols is hard and takes ages :/.

[turt2live]

as clearly it being a merged PR with a label isn't good enough

if that's true, isn't it also true of other MSCs under spec-pr-missing ?

Almost certainly. Usually I try to keep an eye on that label but this particular MSC missed the queue somehow.

[iinuwa]

Does the same thing apply for the send transaction endpoint?

[turt2live]

That would be #2236

[iinuwa]

Ah, thank you. I didn't find it in the other events listed in the MSCs linked here, but a simple search probably would have worked. Sorry about that.