[jlokier]

> But calculating a SHA256 takes time linear in the length of the thing you're hashing.

Ahem, so does hashing.

And so does sort comparison.

The GP's idea is actually quite effective for medium-to-large objects.

As a bonus, you can cache and store the SHA256 of each object for future in-set testing and uniqueness-finding without looking at the objects themselves. (You cannot do this with non-cryptographic hashes or sorting by comparisons.)

This is basically what Git does.