MyPy fails on Union[str, int] < Union[str, int] when underlying type is known

Bug Report

Given the following script I'm receiving several unexpected errors:

test.py:22: error: Unsupported operand types for < ("str" and "int")
test.py:22: error: Unsupported operand types for < ("int" and "str")
test.py:22: note: Both left and right operands are unions
test.py:24: error: Unsupported operand types for < ("str" and "int")
test.py:24: error: Unsupported operand types for < ("int" and "str")
test.py:24: note: Both left and right operands are unions
Found 4 errors in 1 file (checked 1 source file)

Even though the if-block and assert statement just before the comparison operator usage on a and b clearly filters out the case of mismatching types. I've only added the assert statement in the hope of telling mypy that these variables have the same type, but no luck there.

from typing import Union

def semver_prerelease_compare(lhs: str, rhs: str) -> bool:
    a: Union[str, int]
    b: Union[str, int]
    for a, b in zip(lhs.split("."), rhs.split(".")):
        try:
            a = int(a)
        except ValueError:
            pass
        try:
            b = int(b)
        except ValueError:
            pass

        if isinstance(a, int) and not isinstance(b, int):
            # Numeric identifiers sort before non-numeric ones
            return True
        if type(a) != type(b):
            return False
        if a < b:
            return True
        elif b < a:
            return False

    return len(lhs) < len(rhs)

Reproduziere die gemeldeten Diagnosen, indem du mypy auf test.py ausführst, und verfolge anschließend, wie sich die Guard-Bedingung type(a) != type(b) auf die Eingrenzung für die Vergleiche a < b und b < a auswirkt. Erledigt, wenn die Prüfung auf äquivalente Typen nicht mehr die gemeldeten str/int-Fehler erzeugt und das Verhalten durch einen Regressionstest abgedeckt ist.