Does

@OverloadResolutionByLambdaReturnType

not cover the case where one of the return types is a subtype of the other? For instance, this surprisingly fails (playground):

@OptIn(kotlin.experimental.ExperimentalTypeInference::class)
@OverloadResolutionByLambdaReturnType 
inline fun printIf(condition: Boolean, message: () -> Any?) { 
    if(condition) println(message()) 
}
inline fun printIf(condition: Boolean, message: () -> Int) { 
    if(condition) println(message()) 
}
fun main() {
    printIf(true) { 42 }
    printIf(false) { "hello world" } // Type mismatch: inferred type is String but Int was expected
}

This can obv be used to avoid boxing primitives like Int when unneeded. I know the cost of boxing is marginal, but in a tight loop it can be detrimental.