Bob, the hash invariant isn't a mere implementation detail, it is critical to making hash based data structures work properly - if two equal objects (say the integer zero and the float zero) ever end up in different hash bins, then the uniqueness property of dictionary keys and sets breaks down.

The three proposed mitigation strategies (using SipHash for string hashing, a tunable collision counting hash map and providing a non-hash based mapping container in the standard library) are all reasonable approaches to the problem and, most importantly, they're *orthogonal* approaches to the problem. There's nothing stopping us doing all three if someone is willing and able to provide the code.