overriding method signatures in extensions seems to produce unpredictable results in certain cases. The following example demonstrates two different results with a similar pattern.
class A: UIViewController {
func doThing() {
print("dothing super class")
}
override func viewDidLoad() {
print("viewdidload superclass")
super.viewDidLoad()
}
}
class B: A { }
extension B {
override func doThing() {
print("dothing sub class")
super.doThing()
}
override func viewDidLoad() {
print("viewdidload subclass")
super.viewDidLoad()
}
}
let a: A = B()
a.doThing()
let vc: UIViewController = B()
vc.viewDidLoad()
This prints :
dothing super class
viewdidload subclass
viewdidload superclass
You can see this skips the B's implementation of doThing when it is cast as A, however includes both implementations of viewDidLoad when cast as UIViewController. Is this the expected behavior? If so, what is the reason for this?
ENV: Xcode 7.3, Playground
The surprise here is that the compiler permits the override in the extension. This doesn't compile:
class A {
func doThing() {
print("dothing super class")
}
}
class B: A {
}
extension B {
override func doThing() { // error: declarations in extensions cannot override yet
print("dothing sub class")
super.doThing()
}
}
In your example, it appears that the compiler gives you a pass because A derives from NSObject — presumably in order to allow this class to interact with Objective-C. This does compile:
class A : NSObject {
func doThing() {
print("dothing super class")
}
}
class B: A {
}
extension B {
override func doThing() {
print("dothing sub class")
super.doThing()
}
}
My guess is that the fact you're allowed to do this override at all is itself possibly a bug. The docs say:
Extensions can add new functionality to a type, but they cannot override existing functionality.
And overriding is nowhere listed as one of the things an extension can do. So it seems like this should not compile. However, perhaps this is permitted deliberately for compatibility with Objective-C, as I said before. Either way, we are then exploring an edge case, and you have very nicely elicited its edginess.
In particular, the preceding code still doesn't cause dynamic dispatch to become operational. That's why you either have to declare doThing as dynamic, as suggested by #jtbandes, or put it in the actual class rather than the extension — if you want polymorphism to operate. Thus, this works the way you expect:
class A : NSObject {
dynamic func doThing() {
print("dothing super class")
}
}
class B: A {
}
extension B {
override func doThing() {
print("dothing sub class")
super.doThing()
}
}
And so does this:
class A : NSObject {
func doThing() {
print("dothing super class")
}
}
class B: A {
override func doThing() {
print("dothing sub class")
super.doThing()
}
}
My conclusion would be: Very nice example; submit it to Apple as a possible bug; and Don't Do That. Do your overriding in the class, not in the extension.
Related
How can we call class functions with a dynamic class name?
Assume the following example where I have two class with methods with same signature
class Foo{
class func doSomething()
}
class Foobar {
class func doSomething()
}
class ActualWork{
//call following method with a variable type so that it accepts dynamic class name
func callDynamicClassMethod(x: dynamicClass)
x.doSomething()
}
How can this be implemented so that x accepts values at run time
Edit: Sorry, I missed to mention that I was looking for any other ways other than protocol oriented approach. This is more of an exploratory question to explore if there is a more direct approach/pods/libraries to achieve this.
I liked this question, because it made me to think a lit'bit outside of the box.
I'll answer it, by dividing it into a few parts.
First
call class functions
Class function is basically a Type methods, which can be achieved using the static word inside the class context.
Taking that into account, you can get a simple solution, using protocol and passing the class reference (conforming to that protocol) like this:
protocol Aaa{
static func doSomething();
}
class Foo : Aaa{
static func doSomething() {
print("Foo doing something");
}
}
class FooBar : Aaa{
static func doSomething() {
print("FooBar doing something");
}
}
class ActualWork{
//Using class (static) method
func callDynamicClassMethod <T: Aaa> (x: T.Type) {
x.doSomething();
}
}
//This is how you can use it
func usage(){
let aw = ActualWork();
aw.callDynamicClassMethod(x: Foo.self);
aw.callDynamicClassMethod(x: Foo.self);
}
Second
In case you don't really need the method on the class context, you may consider using instance methods. In that case the solution would be even simpler, like this:
protocol Bbb{
func doSomething();
}
class Bar : Bbb{
func doSomething() {
print("Bar instance doing something");
}
}
class BarBar : Bbb{
func doSomething() {
print("BarBar instance doing something");
}
}
class ActualWork{
//Using instance (non-static) method
func callDynamicInstanceMethod <T: Bbb> (x: T){
x.doSomething();
}
}
//This is how you can use it
func usage(){
let aw = ActualWork();
aw.callDynamicInstanceMethod(x: Bar());
aw.callDynamicInstanceMethod(x: BarBar());
}
Third
If you need to use the class func syntax, as OP originally did:
class func doSomething()
You CANNOT simply use a protocol. Because protocol is not a class...
So compiler won't allow it.
But it's still possible, you can achieve that by using
Selector with NSObject.perform method
like this:
class ActualWork : NSObject{
func callDynamicClassMethod<T: NSObject>(x: T.Type, methodName: String){
x.perform(Selector(methodName));
}
}
class Ccc : NSObject{
#objc class func doSomething(){
print("Ccc class Doing something ");
}
}
class Ddd : NSObject{
#objc class func doSomething(){
print("Ccc class Doing something ");
}
#objc class func doOther(){
print("Ccc class Doing something ");
}
}
//This is how you can use it
func usage() {
let aw = ActualWork();
aw.callDynamicClassMethod(x: Ccc.self, methodName: "doSomething");
aw.callDynamicClassMethod(x: Ddd.self, methodName: "doSomething");
aw.callDynamicClassMethod(x: Ddd.self, methodName: "doOther");
}
Generics and Protocol oriented programming will do the job:
protocol Doable {
static func doSomething()
}
class Foo: Doable {
static func doSomething() {
debugPrint("Foo")
}
}
class Foobar: Doable {
static func doSomething() {
debugPrint("Foobar")
}
}
class ActualWork {
func callDynamicClassMethod<T: Doable>(x: T.Type) {
x.doSomething()
}
}
let work = ActualWork()
work.callDynamicClassMethod(x: Foo.self)
work.callDynamicClassMethod(x: Foobar.self)
you can achieve this with help of Protocol
protocol common {
static func doSomething()
}
class Foo : common{
static func doSomething() {
print("Foo")
}
}
class Foobar : common {
static func doSomething() {
print("Foobar")
}
}
class ActualWork{
//call following method with a variable type so that it accepts dynamic class name
func callDynamicClassMethod(x: common.Type) {
x.doSomething()
}
}
let fooObj : common = Foo()
let Foobarobj : common = Foobar()
let workObk = ActualWork()
workObk.callDynamicClassMethod(x:Foo.self)
workObk.callDynamicClassMethod(x:Foobar.self)
I think, there are three solutions. I shared an sample below.
Use "protocol" that has "doSomething()" function requirements.
Create a function which gets function definition as a parameter.
Use reflection. you can use EVReflection that is good Api for reflection.
sample code:
protocol FooProtocol {
static func doSomething()
}
class Foo: FooProtocol {
class func doSomething() {
print("Foo:doSomething")
}
}
class Foobar: FooProtocol {
class func doSomething() {
print("Foobar:doSomething")
}
}
class ActualWork {
func callDynamicClassMethod<T: FooProtocol>(x: T.Type) {
x.doSomething()
}
func callDynamicClassMethod(x: #autoclosure () -> Void) {
x()
}
func callDynamicClassMethod(x: () -> Void) {
x()
}
}
ActualWork().callDynamicClassMethod(x: Foo.self)
ActualWork().callDynamicClassMethod(x: Foobar.self)
print("\n")
ActualWork().callDynamicClassMethod(x: Foo.doSomething())
ActualWork().callDynamicClassMethod(x: Foobar.doSomething())
print("\n")
ActualWork().callDynamicClassMethod(x: Foo.doSomething)
ActualWork().callDynamicClassMethod(x: Foobar.doSomething)
Looks like you are searching for duck typing, and this is harder to achieve in a statically typed language (with some exceptions, listed in the linked Wikipedia page).
This is because dynamically calling a method requires knowledge about the layout of the target object, thus either inheritance of the class declaring the method, or conformance to a protocol that requires that method.
Starting with Swift 4.2, and the introduction of dynamic member lookup, there is another approach to solve your problem, however it also involves some ceremony:
// This needs to be used as base of all classes that you want to pass
// as arguments
#dynamicMemberLookup
class BaseDynamicClass {
subscript(dynamicMember member: String) -> () -> Void {
return { /* empty closure do nothing */ }
}
}
// subclasses can choose to respond to member queries any way they like
class Foo: BaseDynamicClass {
override subscript(dynamicMember member: String) -> () -> Void {
if member == "doSomething" { return doSomething }
return super[dynamicMember: member]
}
func doSomething() {
print("Dynamic from Foo")
}
}
class Bar: BaseDynamicClass {
override subscript(dynamicMember member: String) -> () -> Void {
if member == "doSomething" { return doSomething }
return super[dynamicMember: member]
}
func doSomething() {
print("Dynamic from Bar")
}
}
func test(receiver: BaseDynamicClass) {
receiver.doSomething()
}
test(receiver: Bar()) // Dynamic from Bar
To conclude, in the current Swift version there is no way to have both the argument and the method dynamic, some common ground needs to be set.
I got a struct :
struct ErrorResultType: ErrorType {
var description: String
var code: Int
}
and a protocol:
protocol XProtocol {
func dealError(error: ErrorResultType)
}
Now I want to make an extention of UIViewController:
extension UIViewController: XProtocol {
func dealError(error: ErrorResultType) {
// do something
}
}
So I can subclass from this and override the function like:
class ABCViewController: UIViewController {
--->override func dealError(error: ErrorResultType) {
super.dealError(error)
// do something custom
}
}
But it goes wrong with: Declarations from extensions cannot be overridden yet
It doesn't make any sense to me. When I replace all ErrorResultType with AnyObject, the error won't appear any more.
Anything I missed?
For now the method in the extension must be marked with #objc to allow overriding it in subclasses.
extension UIViewController: XProtocol {
#objc
func dealError(error: ErrorResultType) {
// do something
}
}
But that requires all types in the method signature to be Objective-C compatible which your ErrorResultType is not.
Making your ErrorResultType a class instead of a struct should work though.
If i am not making mistake this is connected with Swift official extension mechanism for adding methods to classes.
Conclusion :
At the moment, it's not possible to override entities declared in
extension by subclassing, like so:
class Base { }
extension Base {
var foo: String { return "foo" }
}
class Sub: Base {
override var foo: String { return "FOO" } // This is an error
}
Please check this resource for more information : https://github.com/ksm/SwiftInFlux/blob/master/README.md#overriding-declarations-from-extensions
While playing around in playground with protocol extension I came to strange error. More precisely I declared a protocol and used it as type in the following manner :
protocol InvokeProtocol{
func invokeA()
func invokeB()
}
class Controller{
var invoker : InvokeProtocol?
func methodA(){
invoker?.invokeA()
}
func methodB(){
invoker?.invokeB()
}
}
What I did next was creating a subclass of Controller and tried to extend the InvokeProtocol for that particular type
extension InvokeProtocol where Self: SpecificController{
func invokeC(){
}
}
class SpecificController : Controller {
override func methodA() {
super.methodA()
}
override func methodB() {
super.methodB()
}
func methodC(){
invoker?.invokeC()
}
}
But this give me the following compile time error while I try to call invokeC on SpecificController
#opened("70A8B09C-65E3-11E5-9A8E-6C40088AF546") InvokeProtocol' is not a subtype of 'SpecificController'
I just jumped to Swift 2.0 and this is my first time I see errors starting with #opened. What does this mean? Is this a known bug waiting to be fixed? If so do you guys have any workaround about this?
To reply to your comment. It is achievable. You just haven't implemented the protocol in the first place. The following will work:
protocol InvokeProtocol {
func invokeA()
func invokeB()
}
class Controller : InvokeProtocol {
func invokeA(){
}
func invokeB() {
}
}
extension InvokeProtocol where Self: SpecificController{
func invokeC() {
}
}
class SpecificController : Controller {
override func invokeA() {
super.invokeA()
}
override func invokeB() {
super.invokeB()
}
func invokeC() {
}
}
I guess the where Self: SpecificController, the Self means var invoke: InvokeProtocol?
if invoke conform the protocol and invoke is SpecificController, that invoke can call the method named invokeC()
I'm facing a problem regarding protocols methods dispatch.
I have a class hierarchy that looks like that:
protocol E {
func test()
}
extension E {
func test() {
print("jello")
}
}
class A: E {
}
class B: A {
func test() {
print("hello")
}
}
But when I call test on an instance of class B statically forced to be typed A, "jello" gets printed, not "hello".
let b: A = B() // prints "jello" not "hello"
b.test()
My understanding is that test method printing "jello" gets "integrated" into instances of A (since A conforms to E protocol). I'm then providing another implementation of test inside B (that inherits form A). I thought polymorphism would work here and calling test on B instance that are stored inside A references would print hello. What's happening here?
It's perfectly working when not using any protocol:
class A {
func test() {
print("jello")
}
}
class B: A {
override func test() {
print("hello")
}
}
let b: A = B() // prints "hello"
b.test()
What's different from adopting a protocol that adds new methods to my parent class and providing a new implementation in a subclass, than having directly written this method in the parent class and then overriding it in a subclass?
Do you guys have any workaround?
Smells like a bug.
The only workaround I came up with was very ugly...
protocol E {
func test()
}
func E_test(_s: E) {
print("jello")
}
extension E {
func test() { E_test(self) }
}
class A: E {
func test() { E_test(self) }
}
class B: A {
override func test() {
print("hello")
}
}
let b: A = B()
b.test()
It is a bug indeed. Here is the link to it: https://bugs.swift.org/browse/SR-103
I have an extension on UIView implementing a protocol
protocol SomeProtocol {
var property : Int
}
extension UIView : SomeProtocol {
var property : Int {
get {
return 0
}
set {
// do nothing
}
}
}
in a concrete subclass I want to override this extension method:
class Subclass : UIView, SomeProtocol {
var _property : Int = 1
var property : Int {
get { return _property}
set(val) {_property = val}
}
}
I set breakpoints and see that the extension method is called and not the concrete subclass method:
var subclassObject = Subclass()
someObject.doSomethingWithConcreteSubclassObject(subclassObject)
// other code;
fun doSomethingWithConcreteSuclassObject(object : UIView) {
var value = object.property // always goes to extension class get/set
}
As others have noted, Swift does not (yet) allow you to override a method declared in a class extension. However, I'm not sure whether you'll ever get the behavior you want even if/when Swift someday allows you to override these methods.
Consider how Swift deals with protocols and protocol extensions. Given a protocol to print some metasyntactic variable names:
protocol Metasyntactic {
func foo() -> String
func bar() -> String
}
An extension to provide default implementations:
extension Metasyntactic {
func foo() -> String {
return "foo"
}
func bar() -> String {
return "bar"
}
}
And a class that conforms to the protocol:
class FooBar : Metasyntactic {
func foo() -> String {
return "FOO"
}
func bar() -> String {
return "BAR"
}
}
Swift will use dynamic dispatch to call the appropriate implementations of foo() and bar() based on each variable's runtime type rather than on the type inferred by the compiler:
let a = FooBar()
a.foo() // Prints "FOO"
a.bar() // Prints "BAR"
let b: Metasyntactic = FooBar()
b.foo() // Prints "FOO"
b.bar() // Prints "BAR"
If, however, we extend the protocol further to add a new method:
extension Metasyntactic {
func baz() -> String {
return "baz"
}
}
And if we override our new method in a class that conforms to the protocol:
class FooBarBaz : Metasyntactic {
func foo() -> String {
return "FOO"
}
func bar() -> String {
return "BAR"
}
func baz() -> String {
return "BAZ"
}
}
Swift will now use static dispatch to call the appropriate implementation of baz() based on the type inferred by the compiler:
let a = FooBarBaz()
a.baz() // Prints "BAZ"
let b: Metasyntactic = FooBarBaz()
b.baz() // Prints "baz"
Alexandros Salazar has a fantastic blog post explaining this behavior in depth, but suffice it to say that Swift only uses dynamic dispatch for methods declared in the original protocol, not for methods declared in protocol extensions. I imagine the same would be true of class extensions, as well.
I know this question has been asked a while ago. But this will be handy for someone who looking for an easier way. There is a way of overriding an extension methods. I know its bit hacky but it does the job beautifully.
If you declare your protocol with #objc
#objc protocol MethodOverridable {
func overrideMe()
}
In Extension
extension MainClass: MethodOverridable {
func overrideMe() {
print("Something useful")
}
}
Subclass - You can able to override it in your subclass. It works like a magic. Well, not really when adding #objc it exposes your protocol to Objective-C and its Runtime. That allows your subclass to override.
class SubClass: MainClass {
override func overrideMe() {
print("Something more useful")
}
}
Swift 5
class Class
{
#objc dynamic func make() { print("make from class") }
}
class SubClass: Class {}
extension SubClass {
override func make() {
print("override")
}
}
It looks like you can override property for 2nd super class property. For example, you can access UIView property by making extension to the UILabel wanting to override frame property of UIView. This sample works for me in Xcode 6.3.2
extension UILabel {
override public var frame: CGRect {
didSet {
println("\(frame)")
}
}
}
You can't do this through normal means.
It's in Apple's docs that you can't override a method in an extension in a subclass.
Also, extensions can add new functionality to a type, but they cannot override existing functionality.
https://docs.swift.org/swift-book/LanguageGuide/Extensions.html
I think you forgot to override the superclass property in your subclass:
class Subclass : UIView {
var _property : Int = 1
override var property : Int {
get { return _property}
set(val) {_property = val}
}
}