I opened psycopg/psycopg#1023 to fix a "problem" is psycopg.

Should this also handle unions of literals? I notice if you narrow with is, it'll switch back to str again. Similarly if you explicitly annotate with Literal[SomeEnum.field], or a union of two enums.

@TeamSpen210 please, share your examples, I would be happy to support them as well :)

Here's an example of both:

class SomeEnum(Enum):
    A = 0
    B = 1
    C = 2

class AnotherEnum(Enum):
    ONE = 1
    TWO = 2
    THREE = 3

def func(value: SomeEnum, combo: SomeEnum | Literal[AnotherEnum.ONE, AnotherEnum.THREE]):
    reveal_type(value.name)  # Literal["A", "B", "C"]
    assert value is not SomeEnum.A
    reveal_type(value.name)  # Just str, but should be Literal["B", "C"]?
    # also str, but could be Literal["A", "B", "C", "ONE", "THREE"]
    reveal_type(combo.name)

Second is a bit less likely, but seems reasonable to support?

Ok, I see, supporting Literal[AnotherEnum.ONE, AnotherEnum.THREE].name seems reasonable to me. Literal values should produce other literals. But, SomeEnum | Literal[AnotherEnum.ONE, AnotherEnum.THREE] does not. It should be left as str, because looks like we have two different enums here. It might have duplicates, etc. Also, the use-case is not very common.

Done!

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅