Also, opinions on whether field naming should be taken into account when hashing? Currently structs

{"a": 1, "b": 2}

and

{"c": 1, "d": 2}

are hashed the same way. Within a column this doesn’t matter because the schema is always going to be the same, but if hashes from different schemas are compared then it might lead to lots of collisions.

The other approach would be to make the datatype itself hashable, and take that into account during hashing?