[Sema] Setter has incorrect mutating-ness inside class-constrained protocol extension

[theblixguy]

If we have a class-constrained protocol extension, where the protocol does not impose a class requirement, then the setter is incorrectly inferred to be mutating by default.

Resolves SR-11298.

[theblixguy]

cc @slavapestov

[jrose-apple]

Technically source-breaking, but I kinda hope no one is depending on this.

protocol SR_11298_P {
  init()
}

class SR_11298_C: SR_11298_P {
  var property: String = ""
  required init() {}
}

extension SR_11298_P where Self: SR_11298_C {
  var wrappingProperty: String {
    get { return property }
    set { self = .init() }
  }
}

let origInstance = SR_11298_C()
var instance = origInstance
instance.wrappingProperty = ""
print(instance === origInstance)

[jrose-apple]

@swift-ci Please test

[jrose-apple]

@swift-ci Please test source compatibility

[jrose-apple]

Thanks, @theblixguy!

[hamishknight]

Should we make a note of this in the changelog? Note that this change can also break code where the setter delegates to another mutating requirement or extension member, such as:

protocol P {
  var property: String { get set }
}

extension P where Self : AnyObject {
  var wrappingProperty: String {
    get { property }
    set { property = newValue }
  }
}

[jrose-apple]

Ah, hm. That might be a more serious concern, especially since people are already confused about why protocol methods can be mutating for classes.

[jrose-apple]

Reverting for now.

[slavapestov]

@theblixguy Can you create a PR that adds a test case for the current behavior, including @hamishknight's example? I'm not sure it's tested anywhere in our suite and we want to make sure any change is intentional.

[theblixguy]

Sure. I’ve also created a forum thread to discuss this.