1

Not sure exactly how to phrase this or if it's possible, but essentially what I'm wanting to do is this. I would like to have a protocol that includes a function that takes a protocol as a parameter, but swap out that protocol with a different one, which also conforms to said protocol.

e.g.

//Top level protocol
protocol ApplicationState: Equatable {}

//Feature specific protocols
protocol UserState: ApplicationState {
  var userDidChange: Bool { get set }
}

protocol FreezeState: ApplicationState {
  var freezeState: String { get set }
}

protocol Foo {
  func update(_ value: any ApplicationState)
}

//Implementation

struct Bar: Foo { 
  func update(_ value: any ApplicationState) { } <--------- Conformance as defined in the protocol.
  func update(_ value: any UserState) { } <--------- Ideally what I would like to do since UserState also conforms to ApplicationState.
}

I am able to achieve what I want by casting the type any ApplicationState as? any UserState but I would ultimately like to remove that casting step since I would need to that in all the locations where my Foo protocol is used.

Is this possible?

2

You would typically use protocol-based dispatch for this for this:

protocol ApplicationState {
  func beUpdatedBy(_ foo: Foo)
}

extension Foo {
  func update(_ value: ApplicationState) {
    value.beUpdatedBy(self)
  }
}
3

Not sure if that's exactly what I'm looking for. That doesn't seem to allow me to change to the protocol that I ideally want to use since it won't include the properties associated with UserState or FreezeState.

Here is a more detailed Bar implementation on what I wanting to do.

struct Bar: Foo {
  struct State: UserState {
    var userDidChange: Bool = false
  }
  var userState: State = State()
  
  mutating func update(_ value: any ApplicationState) {
    guard let value = value as? any UserState else { <---------- Other parts of the app would use a different protocol here, depending the needs.
      return
    }
    userState.userDidChange = value.userDidChange
  }
}

struct InternalAppState: UserState, FreezeState { <-------- Internal type that maintains state relevant to the App. It also conforms to the same protocols defined in different features.
  var userDidChange: Bool = false
  var freezeState: String = ""
}

//Again a very crude implementation of what I'd like to do.
let internalState = InternalState()
var bar = Bar()
bar.update(internalState) <------ Since "internalState" conforms to `ApplicationState`, I can pass it into the function, but it's opaque and the exact values I'm interested aren't readily available unless I cast as the `protocol` the receiving feature needs. I'm hoping to make this is reusable and reduce the amount of casting I need, if possible.

For more context, the protocol's will be feature specific and unaware of each other, besides the InternalState which will have access the different protocols.

4

The issue here is that the protocol Foo is a contract that any conforming type must uphold, specifically saying that you must be able to accept any type conforming to ApplicationState. If you want to handle a specific type, you could use an associatedtype State: ApplicationState requirement, which allows each conforming type to choose the specific type it allows to be passed. (You won’t be able to use a type like any FreezeState here, though, because it doesn’t conform to ApplicationState)

2 Likes
5

Yep, I get that. I explored using an associatedtype in the protocol as well and what I'm attempting to do also didn't work.

protocol Foo {
  associatedtype: State: ApplicationState
  func update(_ value: State)
}

This gets me the API that I want, but the call site is where it breaks down. The type I'll be passing in is some ApplicationState and casting fails.

//Rough implementation
struct FeatureContainer<Feature: Foo> {
  let feature: Foo

  func receivedUpdate(_ value: some ApplicationState) {
    feature.update(value as? Foo.State) <------ This casting will fail since the concrete type passed in is not concrete Foo.State,
    but DOES conform to the protocol that constrained ` Foo.State` (protocol UserState).
    But at this point, I don't know which `ApplicationState` protocol the `Foo.State` is using. (FreezeState or UserState).
  }
}

I guessing what I want isn't possible to accomplish since the type could have multiple protocol conformances that also conform to ApplicationState.

6

Do constraints solve the problem?

extension Foo {
  mutating func update(_ value: some UserState) where State: UserState { }
  mutating func update(_ value: some FreezeState) where State: FreezeState { }
}

extension FeatureContainer {
  mutating func receivedUpdate(_ value: some UserState) where Feature.State: UserState {
    feature.update(value)
  }

  mutating func receivedUpdate(_ value: some FreezeState) where Feature.State: FreezeState {
    feature.update(value)
  }
}
7

@Quedlinbug Not a bad idea but I don't think that will work in my current app.

The core ApplicationState concrete type will be a single type that is composed of different protocols defined by the different features, i.e. the properties defined in the examples protocols FreezeState and UserState, to maintain a single source of truth. When a property on the ApplicationState concrete type changes (e.g. appState.userDidChange = true), it will post itself to the FeatureContainer, which will determines if any property defined in the properties it cares about have changed.

8

In classes the feature you are talking about is called "covariant return types" and "contravariant parameter types":

class SuperClass {}

class MainClass: SuperClass {}

class SubClass: MainClass {}

class Foo {
    func update(_ value: MainClass) -> MainClass {
        MainClass()
    }
}

class Bar: Foo {
    override func update(_ value: SuperClass) -> SubClass {
        SubClass()
    }
}

Swift protocols don't support this (could be said they require "invariant" return and parameter types. I think something like this could be considered for Swift, although it would make the language more complicated.