Clean up redundant complex type unions

[daxfohl]

Fixes #2375.

In type signatures, Union[complex, float, int] and such can be simplified to complex, and this PR does so. Mypy does not work well with numbers.Complex in type signatures, so this PR does not use it there.

In isinstance checks, (complex, numbers.Complex) is always redundant, and this PR replaces it with just isinstance(numbers.Complex).

isinstance(numbers.Complex) also returns true for float and int values, whereas isinstance(complex) returns false. Sometimes in the code isinstance(complex) was just used to differentiate number from symbolic or None. I changed those to isinstance(number.Complex) for broader applicability. Sometimes, isinstance(complex) was intentionally added to exclude float; for instance float does not have .imag or .real, and for those I left as-is. A follow-up item could be to replace val.real etc. with np.real(val) in those cases, which would allow standardizing on isinstance(numbers.Complex) everywhere.

The specific bug in #2375 is fixed with an addition to PauliString.__array_ufunc__.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.17%. Comparing base (d589bfc) to head (13d764d).
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7041      +/-   ##
==========================================
- Coverage   98.17%   98.17%   -0.01%     
==========================================
  Files        1085     1085              
  Lines       94672    94683      +11     
==========================================
+ Hits        92942    92952      +10     
- Misses       1730     1731       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[pavoljuhas]

LGTM with two small tweaks. Thank you for hunting these down!

[pavoljuhas]

Suggested change

	a: Union[complex, Iterable[float], np.ndarray], period: float, *, atol: float = 1e-8
	a: Union[complex, Iterable[complex], np.ndarray], period: float, *, atol: float = 1e-8

[daxfohl]

I actually went the opposite direction. The function fails on complex numbers (there's a mod operation in it), so I'm not sure why complex was an option at all.

[pavoljuhas]

I am getting True for issubclass(complex, numbers.Number) -

Suggested change

	def __mul__(self, other: Union[complex, numbers.Number]) -> 'cirq.PauliString[TKey]':
	def __mul__(self, other: numbers.Number) -> 'cirq.PauliString[TKey]':

[daxfohl]

Here also I went the opposite direction. numbers types are supposed to be for instanceof, but don't work well with mypy. SupportsComplex is the proper way to do it with mypy, but SupportsComplex comes with overhead. It only means it has __complex__ on it, which means you've got to call complex(other) before you can safely use it; code and time overhead. The only thing it improves is that mypy won't complain when using np.int64 or other tangential types, though they still work at run time regardless. So, just leaving this as complex, to be consistent with all our other complex parameter types.

[pavoljuhas]

A follow-up item could be to replace val.real etc. with np.real(val) in those cases, which would allow standardizing on isinstance(numbers.Complex) everywhere.

We should keep in mind here that attribute access is much faster than np.real -

In [3]: %timeit c.real
39.5 ns ± 1.52 ns per loop (mean ± std. dev. of 7 runs, 10,000,000 loops each)
In [4]: %timeit np.real(c)
231 ns ± 16.2 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)