[SR-7422] Protocol extension with a same type non-final class constraint cannot be conformed to by said class

[hamishknight]

Previous ID	SR-7422
Radar	None
Original Reporter	@hamishknight
Type	Bug
Status	Resolved
Resolution	Done

Environment

Swift version 4.2-dev (LLVM 064f71a292, Clang 238f385323, Swift 18b28b9)
Target: x86_64-apple-darwin17.5.0

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug, 4.1Regression
Assignee	None
Priority	Medium

md5: 6015a5fb7960efaf9e0cda5af8f7a9e8

Issue Description:

The following compiles in 4.0.3, but doesn't in 4.1:

class C {}

protocol P {
  func foo()
}

extension P where Self == C {
  func foo() {}
}

extension C : P {} // Type 'C' does not conform to protocol 'P'

I believe this is because of #12174 which changed the witness thunk for foo() from (C) -> Void to <Self : C>(Self) -> Void, thus making it not match the extension implementation signature of (C) -> Void.

I'm unsure if this regression is intentional or not (given we have no covariant Self returns in play here), filing here for confirmation.

[hamishknight]

cc @slavapestov

[slavapestov]

This is intentional. Try 'extension P where Self : C'.

[hamishknight]

Thanks – would it be worth adding a test case for this to ensure changes like covariant witnesses don't change this behaviour?

[belkadan]

We could probably use a customized diagnostic too.

[slavapestov]

@hamishknight even without a covariant Self return, we still want any references to the Self metadata inside the extension method to have the right behavior. I still consider the 4.1 behavior to be correct, and we didn't really anticipate people would define protocol extensions with a same-type constraint on Self since that is equivalent to an extension on the concrete type itself!

I'll add a test to ensure the behavior doesn't change again though: #15924