Apparent compiler bug relating to Optional witnessing failable initializer requirement

[jeremyabannister]

Description

I think this must be a bug. The following test passes if the Optional initializer is declared as a failable initializer, but fails if the ? is removed from the Optional initializer.

The Test

@Test
func foo() {
    #expect(
        Demo<Optional<Int>>.test()
    )
}
struct Demo<T: Foo> {
    static func test() -> Bool {
        if let a = T.init() {
            return true
        }
        return false
    }
}
protocol Foo {
    init?()
}

Passes If Optional Conforms Like This:

extension Optional: Foo {
    init?() {
        self = Self.none
    }
}

Fails If Optional Conforms Like This:

extension Optional: Foo {
    init() {
        self = Self.none
    }
}

Original Swift Forums Post: https://forums.swift.org/t/apparent-compiler-bug-relating-to-optional-witnessing-failable-initializer-requirement/76759

Reproduction

@Test
func foo() {
    #expect(
        Demo<Optional<Int>>.test()
    )
}
struct Demo<T: Foo> {
    static func test() -> Bool {
        if let a = T.init() {
            return true
        }
        return false
    }
}
protocol Foo {
    init?()
}
extension Optional: Foo {
    init() {
        self = Self.none
    }
}

Expected behavior

I expected that the unwrapping if let a = T.init() { } would succeed since in this case T == Optional<Int> and Optional unconditionally succeeds in initializing itself to .none.

Environment

Swift 6

Additional information

No response

[AnthonyLatsis]

This simpler example might help. A failable initializer returns an optional of the parent type — a double optional in this case. Since I am not returning nil, neither is the resulting Int?? value nil:

extension Optional {
  init?(failable: Void) {
    self = Self.none
  }
  init(nonfailable: Void) {
    self = Self.none
  }
}

do {
  let _ = Int?.init(failable: ())    // == Int??.some(Int?.none)
  let _ = Int?.init(nonfailable: ()) // == Int?.none
}

[AnthonyLatsis]

Sorry, I was not paying enough attention. I see now.

[jeremyabannister]

Hi @AnthonyLatsis , any progress on this bug?