@swift-ci create

Any chance you can share the full definition of both the protocol and the class? Just want to make sure we fix the same thing you're seeing.

Comment by Amro Mousa (JIRA)

I'm not in a position to do that, but I'll do my best to reproduce the issue in a sample project today and attach it to this ticket.

My initial diagnosis was wrong; apparently the getter name is preferred (probably because of the common "isFoo" with a property name of "foo", where "isFoo" better matches the Swift conventions). But this code works in both master and 4.0:

class TestLong: Foo {
  var displayAsSelfThreadTweet: Bool { return true }
}

So something else funny is going on.

Also seems to work in Swift 3.1.

Comment by Amro Mousa (JIRA)

Thanks @belkadan. This is working for me in a sample project as well. I'll try to figure out what's different in our larger project. This is code that will be going away in the near future, but I'd still like to understand what's happening.

Comment by Amro Mousa (JIRA)

No luck narrowing this down to a sample project. I'll spend more time on that tomorrow.

Sorry, I just can't get this to fail locally. Going to have to send it back. :-(

Good news: an Apple-internal client ran into this too. I don't think the fix'll make 4.1, but at least your workaround should be safe.

Comment by Amro Mousa (JIRA)

That's great news! Thanks @belkadan.

Reduced:

@import Foundation;

@protocol TestProto
@property (nonatomic, readonly, getter=isFoo) BOOL foo;
@end

class Test: TestProto {
  var isFoo: Bool { return false }
}

protocol HasAssoc {
  associatedtype Assoc: TestProto
}

struct Impl: HasAssoc {
  typealias Assoc = Test
}

The error is emitted when compiling b.swift.

#14794

Unfortunately probably not worth pulling into 4.1 at this point, since there is a workaround.

Took me longer than expected to come to a viable solution, but it's in now! You should be able to test with a master snapshot in the next few days.