Objects turn the structured data into a stream of integers; hashers turn the stream into a numeric hash.

One problem? The hash-functions use blocks instead of streaming these days. On the contrary, the Hasher API needs to stream unsigned integers. So in order to use this API a byte must work with the hasher. It collides with the current block where 128 bits can be hashed at the same time...

The bottom line is: hashing a product type can only be efficient if it’s linearized.