[Property Wrappers] Don't allow wrappedValue and projectedValue to have dynamic Self type.

[hborla]

Currently, when wrappedValue or projectedValue have dynamic Self type, the synthesized properties when using the wrapper will also get a dynamic Self type, but in a new context. This bug could manifest as really misleading diagnostics, or a crash. For now, we will disallow this, but we can re-visit implementing correct support for this in the future.

Resolves: rdar://problem/53258469

[hborla]

@swift-ci please smoke test

[hborla]

@swift-ci please smoke test OS X platform

[theblixguy]

Hmm, I suppose this is only problematic when working with classes? For example, currently this code works fine:

@propertyWrapper
struct Foo {
  let wrappedValue: String
  var projectedValue: Self { self }

  init(wrappedValue: String) {
    self.wrappedValue = wrappedValue
  }

  func bar() {}
}

struct UsesFoo {
  @Foo var property = "hello"

  func test() {
    $property.bar() // Ok
  }
}

although if you change Foo to be a class then I get value of type 'UsesFoo' has no member 'bar' which is wrong, so I am wondering whether we should only disallow it when the wrapper type is a class, although I understand simply banning it outright is a simpler rule.

[hborla]

@theblixguy you're right that Self is not problematic when the property wrapper is a struct, and this code still allows that to compile. Self in a struct is replaced with the type of that struct rather than a DynamicSelfType. So, this code does only disallow Self when the property wrapper is a class.

The followup errors when using the property wrapper are annoying, and they happen because the backing storage and projected value properties aren't synthesized when a property wrapper has an invalid PropertyWrapperTypeInfo. This can also happen if the property wrapper is missing a non-static wrappedValue property, for example. Synthesizing invalid properties in this case might be good for diagnostics...

[theblixguy]

Ah, I didn't know that, I thought it'd still be DynamicSelfType. Thank you for the explanation! I think it might be worth adding a test case for the struct case (if there's isn't one) to make sure we don't regress.

(btw I think it should work for final class too as its equivalent to Self but it's not supported right now.).

[hborla]

@swift-ci please smoke test