Use a single pass for core narrowing logic, add comments

[hauntsaninja]

This PR should not change semantics

We refactor to avoid the separate loops for type_targets and value_targets, which allows the logic to be more consolidated, helps with future changes and is probably faster

[github-actions]

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

[hauntsaninja]

Merging, but comments are very welcome!

[cdce8p]

This change seems to have introduced a small regression.

fromenumimportStrEnumclassCapability(StrEnum):
COLOR_CONTROL="colorControl"COLOR_TEMPERATURE="colorTemperature"deffunc(capability: Capability|str):
ifcapabilityin (Capability.COLOR_CONTROL, Capability.COLOR_TEMPERATURE):
reveal_type(capability)

# beforeCapability# nowCapability|str

It reverses part of the improvement from #20602.

[hauntsaninja]

Thanks for the post, but this is actually an improvement!

from enum import StrEnum
class Capability(StrEnum):
COLOR_CONTROL = "colorControl"
COLOR_TEMPERATURE = "colorTemperature"
def func(capability: Capability | str):
if capability in (Capability.COLOR_CONTROL, Capability.COLOR_TEMPERATURE):
print(capability, type(capability))
func(Capability.COLOR_CONTROL)
func("colorControl")

colorControl <enum 'Capability'>
colorControl <class 'str'>

I have tests for this fix in my stack, but it looks like I split the tests for this change into a different commit

Ideally we could narrow to a literal, but in general mypy is missing a bunch of logic around the values of enums

[hauntsaninja]

#20672 includes more tests here

[cdce8p]

Thanks for the post, but this is actually an improvement!

Good point, missed that! Thanks for working on it.