Federation 'send_join' is insufficiently pedantic (SYN-507)

[matrixbot]

In the process of writing a wire-level federation test in sytest to check the room join dance of synapse, I have ended up with a "minimal viable case" test (https://github.com/matrix-org/sytest/blob/paul/federation/tests/50federation/30room-join.pl).

This test appears to be too small in places, and is missing things that we feel synapse ought to require. This bug exists to track those things it should be checking, that this test demonstrates it currently does not.

synapse does not notice any of the following mistakes:

• missing m.room.member event for the room creator.
• missing m.room.power_levels event.
• m.room.member event of joining user has depth of 1.
• all auth_events have a depth of 0.
• joining server does not hash or sign the event submitted to send_join.
• the state key returned by send_join can be an empty object, or an empty list.

(Imported from https://matrix.org/jira/browse/SYN-507)

(Reported by @leonerd)

[matrixbot]

Jira watchers: @leonerd

[matrixbot]

Actually, this seems not quite true.

While send_join appears to be happy, and causes the joining HS to return a 200 OK to the requesting client along with the room ID, any subsequent queries by the user's client on their HS seem to reply with 403 "user is not in the room".

-- @leonerd

[matrixbot]

Ugh. I retract my previous comment. It was working fine if I actually use the room ID, not the room alias.

-- @leonerd

[DMRobertson]

• m.room.member event of joining user has depth of 1.

• all auth_events have a depth of 0.

depth is a hint at best; it's not to be relied upon. Note that state res v2 is no longer involves depth.