SimulatedLocalProcessor `.run()` vs `.get_sampler().run()`

[mpharrigan]

proc = cg.engine.SimulatedLocalProcessor(
    processor_id='hey ma',
    sampler=cirq.Simulator(),
    device=cirq.UNCONSTRAINED_DEVICE,
)
samp = proc.get_sampler()
circ = cirq.Circuit(cirq.measure(cirq.LineQubit(0)))

print(repr(proc.run(circ)))
print(repr(samp.run(circ)))

cirq_google.EngineResult(params=cirq.ParamResolver({}), records={'0': np.array([[[0]]], dtype=np.int8)}, job_id='projects/fake_project/processors/hey ma/job/2', job_finished_time=datetime.datetime(2022, 5, 13, 0, 40, 23, 917546, tzinfo=datetime.timezone.utc))
cirq.ResultDict(params=cirq.ParamResolver({}), records={'0': np.array([[[0]]], dtype=np.int8)})

The processor knows about jobs and such and returns the new (#5152) EngineResult result subtype. If you just get sampler, it doesn't. cc @dstrain115

[mpharrigan]

If you use EngineProcessor, both code paths return an EngineResult.

[mpharrigan]

EngineProcessor gives you a QuantumEngineSampler whereas the SimulatedLocalProcessor gives you a normal cirq.Sampler (albeit wrapped in validation logic).

Can we have SimulatedLocalProcessor.get_sampler() also return a QuantumEngineSampler? In that case it would be backed by a simulated local engine instead of a real engine.

[mpharrigan]

there is no simulated local engine. there is, but it's None in my simulatedlocalprocessor. If I make a simulatedlocalengine, it doesn't have any of the methods needed by QuantumEngineSampler since AbstractEngine doesn't have any running methods.

Can we modify QuantumEngineSampler to be backed by an [Abstract]Processor instead of an engine? and then it can be shared with SimulatedLocalProcessor and EngineProcessor?

[mpharrigan]

QuantumEngineSampler wants to defer to engine methods so it can dispatch to one of many processor_ids specified as a list (which I never liked: #2075)

[mpharrigan]

See #5361 for a sampler that works with SimulatedLocal and real Engine processors