Consider the following playground:
import Foundation
protocol StringInitable {
init( string:String )
}
class A : StringInitable {
var stored:String
required init ( string:String ) {
stored = string
}
}
class B : A /*, StringInitable */ {
var another_stored:String
required init ( string:String ) {
another_stored = "B-store"
super.init(string: string)
}
}
func maker<T:StringInitable>(string:String) -> T {
return T(string: string)
}
let instanceA = A(string: "test-maker-A")
let instanceB = B(string: "test-maker-B")
let makerA:A = maker("test-maker-A")
let makerB:B = maker("test-maker-B")
let typeInstanceA = _stdlib_getTypeName(instanceA)
let typeMakerA = _stdlib_getTypeName(makerA)
let typeInstanceB = _stdlib_getTypeName(instanceB)
let typeMakerB = _stdlib_getTypeName(makerB)
From the results the compiler seems to have inferred the correct types but has failed to call the correct initializers. How come I have to explicitly implement StringInitable in the B-class (test by removing the comment in the B class definition) to have the generic function "maker" call the correct initializer?
That smells like a compiler bug for one simple reason: makerB
is a variable of B
type, but it is assigned an instance of A
. This should not be possible, and in fact if you try to print, and more generally to access to, the another_stored
property of the makerB
variable, a runtime exception is raised, and I wouldn't expecting nothing else.
That's because if B
is a subclass of A
, an instance of A
cannot be assigned to a variable of B
type (whereas the opposite is possible).
Assigning a variable of A
type to a variable of B
type is possible though, but only under these conditions:
A
to B
is done (the compiler should error otherwise)A
variable is actually an instance of B
(a runtime exception should be raised otherwise)Note that the compiler didn't just fail to call the correct initializer - it called the initializer of another class