I have a large table in MySQL with a slow query partly due to a condition like WHERE sent_at IS NULL. sent_at is a nullable datetime column but here we only want to check if it’s null so a standard index across all values would be wasteful.

I’ve discovered that creating a functional index with ISNULL() or IS NULL works but it is never used for queries.
E.g.

ALTER TABLE users ADD INDEX index_users_sent_at_is_null ((ISNULL(sent_at)));

Followed by the query:

SELECT COUNT(*) FROM users WHERE (sent_at IS NULL);

Will have a (JSON) EXPLAIN output with access_type: "ALL" and no possible or selected keys listed.

Doing the equivalent for YEAR(sent_at) works as expected by using the relevant functional index, so there is something special about IS NULL even though the virtual hidden column the functional index produces correctly contains 0 or 1.

Is there an elegant way to make this work?