Fix a bug in the /event_reports Admin API which meant that the total count could be larger than the number of results you can actually query for.

[reivilibre]

The two queries:

synapse/synapse/storage/databases/main/room.py

Lines 2004 to 2037 in 583f227

	sql = """
	SELECT COUNT(*) as total_event_reports
	FROM event_reports AS er
	{}
	""".format(
	where_clause
	)
	txn.execute(sql, args)
	count = cast(Tuple[int], txn.fetchone())[0]
	sql = """
	SELECT
	er.id,
	er.received_ts,
	er.room_id,
	er.event_id,
	er.user_id,
	er.content,
	events.sender,
	room_stats_state.canonical_alias,
	room_stats_state.name
	FROM event_reports AS er
	LEFT JOIN events
	ON events.event_id = er.event_id
	JOIN room_stats_state
	ON room_stats_state.room_id = er.room_id
	{where_clause}
	ORDER BY er.received_ts {order}
	LIMIT ?
	OFFSET ?
	""".format(
	where_clause=where_clause,
	order=order,
	)

need to match in both WHERE conditions and JOIN conditions (insofaras the joins can change the number of rows returned; the LEFT JOIN is safe because it won't remove any rows and neither will it lead to more rows being produced, since only one event per event ID can exist.)

[babolivier]

Is there any test for this endpoint? It looks like something we want to write a regression test for.

[DMRobertson]

Looks like #13521 has made the linting unhappy. I think the problem is that we're restoring a mypy 0.950 cache and mypy 0.971 is unhappy with it (or a similar problem with mypy-zope). Could try re-running this? Otherwise I guess it's time to fork/file a bug under https://github.com/AustinScola/mypy-cache-github-action.

Edit: re-running this failed :( Could you try merging in develop? If that doesn't help, let's revert #13521.

[reivilibre]

Is there any test for this endpoint? It looks like something we want to write a regression test for.

happy to add a test but it also feels like the kind of test that's a bit trivial / not really testing anything once it's fixed, but ah well

[babolivier]

not really testing anything once it's fixed

The point of a regression test is that it should only fail if we accidentally break it in the future (e.g. if we change the query to get the results but forget to change the one to get the count), so I would say it's still a helpful thing to have.