Add missing type hints to replication.http.

[clokep]

Fully types synapse.replication.http.

I have a follow-up that does the rest of synapse.replication and forbids untyped defs.

[clokep]

The only caller of this seems to only pass a room_id, I'm unsure if this is fixing a bug or not:

synapse/synapse/handlers/federation.py

Lines 1379 to 1389 in 251b556

	async def _clean_room_for_join(self, room_id: str) -> None:
	"""Called to clean up any data in DB for a given room, ready for the
	server to join the room.
	Args:
	room_id
	"""
	if self.config.worker.worker_app:
	await self._clean_room_for_join_client(room_id)
	else:
	await self.store.clean_room_for_join(room_id)

[DMRobertson]

Anything in sentry for this? Maybe it's just never called?

[clokep]

It seems to be called via do_invite_join which is definitely called, it seems to eventually come from either RoomMemberMaster or RoomMemberWorker, maybe we're not routing anything that calls the worker version to the worker?

[DMRobertson]

I think there is a clue here:

synapse/synapse/handlers/federation.py

Lines 452 to 474 in 251b556

	# TODO: We should be able to call this on workers, but the upgrading of
	# room stuff after join currently doesn't work on workers.
	assert self.config.worker.worker_app is None
	logger.debug("Joining %s to %s", joinee, room_id)
	origin, event, room_version_obj = await self._make_and_verify_event(
	target_hosts,
	room_id,
	joinee,
	"join",
	content,
	params={"ver": KNOWN_ROOM_VERSIONS},
	)
	# This shouldn't happen, because the RoomMemberHandler has a
	# linearizer lock which only allows one operation per user per room
	# at a time - so this is just paranoia.
	assert room_id not in self._federation_event_handler.room_queues
	self._federation_event_handler.room_queues[room_id] = []
	await self._clean_room_for_join(room_id)

I couldn't see any other call site of _clean_room_for_join. So yeah, I think this is never called on workers as you say. Given that, let's go ahead with the fix as written. (Strictly speaking it probably ought to be a separate change, but I don't mind sneaking it in here.)

[DMRobertson]

Sorry, I picked this from my email queue and only now have noticed it's a draft. Looks like CI is unhappy :(

Otherwise I didn't spot anything horrible. Looks like it was a fairly routine transcription sort of type hinting?

[DMRobertson]

Anything in sentry for this? Maybe it's just never called?

[clokep]

@DMRobertson I think this is back to you? I'm not sure what to do about the _serialize_payload change discussed above. The rest I'm fairly comfortable with!

[DMRobertson]

LGTM---many thanks.

(Still marked as draft but I don't think you have further changes planned here?)