Remove 'room_id' field from m.typing, m.receipt and m.fully_read examples and schema

[anoadragon453]

It was discovered here that the spec had an erroneous room_id field in a m.typing EDU entry of /sync. After some further research, room_id fields also appear in m.read receipts in /sync, and m.fully_read room account data objects in the spec. None of these are necessary nor used in practice.

Checking part of the ecosystem for whether clients look for, or homeservers include, these room_id fields, I found that:

• Element does not require them, nor does Synapse include them.
• Ruma does not include them.
• Dendrite does not include them.
• nheko/mtxclient does not look for them.

This change removes room_id from the example and OpenAPI schema in each case mentioned above. It only affects the Client-Server spec - the Server-Server spec text remains unchanged.

The field was initially introduced in 0f28f83#diff-8d917764f05b0823bc6aee1f9a74895d38df32fb1eeb10c18a55d9120cf23cb2R3.

Preview: https://pr3679--matrix-org-previews.netlify.app

[turt2live]

This change feels wrong - a room_edu is in a room, so needs a room ID. Not all EDUs use a top-level room_id though, so possibly just a badly written example somewhere specific?

[richvdh]

This change feels wrong - a room_edu is in a room, so needs a room ID.

I think the confusion here stems from the fact that typing events (and, for that matter, EDUs in general) have a different format when sent over the federation API (where they do have a room_id) than when sent in a /sync response (where the room is implied by the position in the /sync response).

I haven't checked, but I think this file is only used when rendering the /sync response, so is probably correct?

[anoadragon453]

@richvdh I believe so, though CI seems to be unhappy.

We should perhaps rename this in some way so that it's explicitly client-specific?

[turt2live]

The CI is failing because the schema and the example don't match - this PR only touches the example, not the schema.

[anoadragon453]

I have updated the schema, and CI now passes. I also discovered that m.fully_read is affected by this. So room_id has been removed in the appropriate places as well.

I've only been verifying against Synapse/Element here. I need to check other implementations to see if these fields are being relied upon.

[aaronraimist]

This should fix #3183

[turt2live]

lgtm - just need a changelog to denote the bugfix in the spec prose

[richvdh]

lgtm

done