There’s a case where a property (or function) override in a subclass’ extension 
leads to unexpected behavior:

class Base: NSObject {
    var directProperty: String { return "This is Base" }
    var indirectProperty: String { return directProperty }
}

class Sub : Base { }

extension Sub {
    override var directProperty: String { return "This is Sub" }
}

Accessing “directProperty” works as expected:

Base().directProperty // “This is Base“
Sub().directProperty // “This is Sub“

Accessing “indirectProperty”, shows the unexpected behavior:

Base().indirectProperty // “This is Base“
Sub().indirectProperty // “This is Base“ <- Unexpected!

Adding the “dynamic” keyword to Base.directProperty resolves the issue, so it 
seems like the implementation of “indirectProperty” always accesses 
“directProperty” of the Base class directly instead of dispatching dynamically.

I think this is a bug, either because this doesn’t work as expected at runtime, 
or because it compiles at all. Because the Language Guide for Swift 2.1, 
chapter “Extensions” states in a note near the beginning:
> “Extensions can add new functionality to a type, but they cannot override 
> existing functionality.”

https://developer.apple.com/library/ios/documentation/Swift/Conceptual/Swift_Programming_Language/Extensions.html#//apple_ref/doc/uid/TP40014097-CH24-ID151

That sounds to me like the “override” keyword must not appear in an extension. 
Curiously enough, the above code compiles and some parts of it even run without 
issues despite violating the Language Guide. If “Base” does not inherit from 
NSObject but becomes a root class instead, the compiler produces the following 
error in the extension:

> Declarations in extensions cannot override yet


Note the “yet”. Seems like this is just not done yet. Also, I don’t understand 
how this feature could work except if overriding properties and methods in a 
subclass’ extension will only be supported if the original property or method 
was declared as dynamic. Otherwise the compiler could inline their 
implementation and the extension’s implementation would never be called – 
assuming the extension isn’t known at compile time, as could be the case for 
extensions to classes of system frameworks. Everything’s probably fine with 
Whole Module Optimization if the module contains all the classes and extensions.

Anyways, if this will be supported for non-Objective-C classes in the future, 
the above bug could point to something that doesn’t work correctly.

Also note that this could be related to (but not the same as) a bug reported 
already where methods implemented in a protocol extension cannot be overridden 
by a conforming class’ subclass:
https://bugs.swift.org/browse/SR-103
_______________________________________________
swift-users mailing list
swift-users@swift.org
https://lists.swift.org/mailman/listinfo/swift-users

Reply via email to