Fix bug conversions of function returning function failed

[kntkymt]

Description

resolves #82397

let a: () -> () -> Void = {{}}
let _: (() -> () -> Void)? = a // 🔴 Cannot convert value of type '() -> Void' to specified type '() -> () -> Void'

regression of #78377

- if (convertTo->is<FunctionType>())
+ if (convertTo->is<FunctionType>() && !resultType->is<FunctionType>())
    return false;

Since the new condition doesn’t handle the case where both lhs and rhs have the same curry levels, conversions of function-returning functions are incorrectly repaired by repairByInsertingExplicitCall. It should be resolved by ValueToOptionalConversion by returning false in repairByInsertingExplicitCall.

Solution

improve guard condition at repairByInsertingExplicitCall, specially: check curry levels of lhs and rhs. if lhs doesn't have more curry levels than rhs, it shouldn't be repaired by InsertingExplicitCall.

note that since the existing logic (!resultType->is<FunctionType>()) is designed to ensure that

• lhs has 2 (or more) curry levels and
• rhs has 1 (or more) curry level

, The new guard logic (lhs.curryLevels <= rhs.curryLevels) include this existing logic.

[xedin]

Sorry but I don't think this is the right fix. Before attempting this repair we need to make sure that there are no restrictions recorded (other locations in repairFailures do it with hasAnyRestriction) otherwise we end up in a situation like this where the fix is greedily applied before optional injection could be attempted (if that was allowed to proceed the types would line up exactly).

[kntkymt]

Thanks for quick review!

we need to make sure that there are no restrictions recorded

I see, it sounds like related to the overall flow of reparingByInserting, it may be difficult for me to fix it 😅 let me close this PR and leave to owners 🙇 (Since I'm still a beginner of compiler)