ControlledGate equality seems too loose #5110
Description
Description of the issue
Equality in controlled gate is implemented as follows, intentionally ignoring the order of the controls.
def _value_equality_values_(self):
return (
self.sub_gate,
self.num_controls(),
frozenset(zip(self.control_values, self.control_qid_shape)),
)This means gates are considered equal even if their unitaries don't match.
def test_me():
g0 = cirq.ControlledGate(cirq.Z, control_values=[[1], [0]])
g1 = cirq.ControlledGate(cirq.Z, control_values=[[0], [1]])
assert g0 == g1 # passes
assert np.allclose(cirq.unitary(g0), cirq.unitary(g1)) # failsIn some cases the gate can't even be applied to the same qubits.
def test_me():
g0 = cirq.ControlledGate(cirq.Z, control_values=[[1], [1, 2]], control_qid_shape=[2, 3])
g1 = cirq.ControlledGate(cirq.Z, control_values=[[1, 2], [1]], control_qid_shape=[3, 2])
assert g0 == g1 # passes
qubits = cirq.LineQid.for_qid_shape((2, 3, 2))
op0 = g0.on(*qubits)
op1 = g1.on(*qubits) # raises "Wrong shape of qids"
assert op0 == op1So, while there's a relationship between such gates, calling that relationship "equality" doesn't feel right. Was this intentional?
Cirq version
0.14.0.dev