Phase out SupportsPhase protocol

The use case I have in mind is defining this protocol for SWAP. This is useful, for instance, for the EjectZ optimizer because one can pass a Z gate through a SWAP as follows:

0: ───Z^theta───×───
                │
1: ─────────────×───

=

0: ───×─────────────
      │
1: ───×───Z^theta───

So I think the phase_by function signature should be changed from

def phase_by(val: Any,
             phase_turns: float,
             qubit_index: int,
             default: TDefault = RaiseTypeErrorIfNotProvided):

to

def phase_by(val: Any,
             phase_turns: float,
             qubit_index: int=None,
             qubits: Sequence[cirq.Qid]=None,
             phased_qubit: cirq.Qid=None,
             default: TDefault = RaiseTypeErrorIfNotProvided):

Then phase_by for SWAP would return an OP_TREE.

[kevinsung]

@dabacon I think what happened was that we discussed this issue during a Cirq sync and then decided to instead remove the SupportsPhase protocol using the plan in @mpharrigan’s comment. Then I changed the title of this issue.

[dstrain115]

It was decided to keep this protocol after all. The consensus in cirq cync was that this didn’t really affect anyone negatively and did provide some small value, so that we would keep it in.

The above functionality to propagate Z’s through swaps had already been added to EjectZ.

Closing this issue.