For Optional[T] or T, infer the type T instead of Optional[T].

[gvanrossum]

Fixes #1817, fixes #1733.

[gvanrossum]

Is there an equivalent example for and? I can't think of any.

FWIW this is kind of blocking further explorations of --strict-optional on mypy itself.

[JukkaL]

Maybe test or with both operands optional?

[JukkaL]

Also, could test left operand with a complex union type such as Union[int, str, None].

[gvanrossum]

Done; I've also added special-casing for `None or T` (see tests for what I mean).

[ddfisher]

I feel like we shouldn't need this special casing and this should just fall out of the find_isinstance_check like everything else. I'm a bit surprised that that isn't the case already -- going to briefly investigate why that is.

[ddfisher]

Looks like that was actually pretty straightforward to do. Let's consider #1828 instead of this?

[gvanrossum]

Closing in favor of #1828.

[gvanrossum]

Cool. Maybe someday I'll understand how this works -- the name (find_isinstance_check()) and the resulting variables (right_map, left_map -- why the transposition?) were all kind of confusing.

[ddfisher]

Agree it's all pretty confusing. The function might be be better named find_conditional_type_information or similar. The transposition is because for and we care about the type of the left hand expression if it's falsey but for or we care about the type of the left hand expression when it's truthy (and find_isinstance_check returns a (conditional type map if true, conditional type map if false) tuple).