q1 == q1 but hash(q1) != hash(q2)

[mpharrigan]

I'm having trouble reproducing this, but I'm in a situation where the thing in the title is happening with GridQubit. I've narrowed it down to the fact that the cached cirq.GridQubit._hash values are different and recomputing them with cirq.Qid.__hash__(q1) gives correct, identical hash functions.

My leading theory is that I'm loading a previously-pickled gridqubit, a copy is being made somewhere and the hash is recomputed. The hash value of Qid (the class) is part of the hash input, which could differ over python instantiations

[mpharrigan]

xref #3375

relevant: https://docs.python.org/3/library/pickle.html#handling-stateful-objects

[maffoo]

We should probably make the hash deterministic, e.g.

self._hash = hash(self.row) ^ hash(self.col)