cirq.Result type validation for measurement argument

[balopat]

cirq.Result is supposed to be supplied measurement results as a dict from string to a multidimensional numpy array. However it is possible to instead supply the value as a list of lists. In this case cirq.Result equality will still work, but it will fail to when converted to a string (since __str__ uses numpy slicing).

I'm not quite sure why mypy didn't yell about the incorrect type here. I guess the value in a dict is covariant? I thought it was invariant.

mypy doesn't recognize numpy types.
All of this is accepted by default

Result(params=ParamResolver({}), measurements={"a":[1,2,3]}) # should fail but it won't 
Result(params=ParamResolver({}), measurements={"a":"b"})  # should fail but it won't 
Result(params=ParamResolver({}), measurements={"a":[[1,0],[0,1]]})  # should fail but it won't 

Result(params=ParamResolver({}), measurements={"a":np.array([1,2,3])})  # OK 
Result(params=ParamResolver({}), measurements={"a":np.array([[1,0],[0,1]])}) # OK 

I tried with pip install data-science-types - that project creates type stubs for numpy, matplotlib, pandas.
With this, mypy catches the first three:

cirq/bla.py:5: error: Dict entry 0 has incompatible type "str": "List[int]"; expected "str": "ndarray[Any]"
cirq/bla.py:6: error: Dict entry 0 has incompatible type "str": "str"; expected "str": "ndarray[Any]"
cirq/bla.py:7: error: Dict entry 0 has incompatible type "str": "List[List[int]]"; expected "str": "ndarray[Any]"

However, it generates a lot of other noise:

Found 1337 errors in 154 files (checked 763 source files)

That includes errors due to missing type stubs in the package e.g.:

cirq/contrib/acquaintance/gates_test.py:20: error: Module 'numpy.random' has no attribute 'poisson'
cirq/contrib/acquaintance/executor_test.py:93: error: Module has no attribute "allclose"
cirq/contrib/acquaintance/executor_test.py:94: error: Module has no attribute "absolute"
cirq/contrib/acquaintance/executor_test.py:116: error: Module has no attribute "random"; maybe "rand"?
cirq/contrib/acquaintance/executor_test.py:177: error: Unexpected keyword argument "verbose" for "assert_allclose"

So I think we'll just have to live with this not being in mypy. Maybe we could / should do runtime type validation in the constructor though for the measurement argument!

Originally posted by @balopat in #3710 (comment)

[95-martin-orion]

Related to #3767.

[dstrain115]

We can probably close this after numpy requirements go past 1.20 where they have mypy stubs.