/state_ids returns huge amounts of pointless data

[richvdh]

Currently, /state_ids is specced to return 'The full set of authorization events that make up the state of the room, and their authorization events, recursively.'

Thankfully it appears to mean the ids of those events rather than the events themselves, but still, this can be a huge amount of data (causing synapse to OOM: matrix-org/synapse#6597), and I don't understand why any of it is helpful. The auth events of a given event are embedded in the (signed and hashed part of) the event itself, so if we have the event itself, we know its auth events. The auth event ids are no use without the event itself, so... what is the point of this?

Unfortunately it's not entirely easy to fix this because, at least up until Synapse 1.7.0, Synapse actually relies on the auth events being present to make backfill work correctly; I guess we'll need a /v2/state_ids.

[richvdh]

On reflection, this seems to be correct behaviour.