CS: Handle unbound references to @objc optional methods

[AnthonyLatsis]

Mostly a cleanup and refactoring so that the target change fits nicely. The dynamic lookup case needs more work.

Resolves rdar://69761566

@objc protocol P {
  @objc optional func f()
}

let x: (P) -> (() -> Void)? = P.f

[AnthonyLatsis]

@swift-ci please smoke test macOS

[slavapestov]

CC @beccadax

[AnthonyLatsis]

@swift-ci please smoke test macOS

[hborla]

Awesome, I've been wanting to get rid of buildPropertyWrapperFnThunk by generalizing the thunk-building code in CSApply for a long time! I'll try to take a look tomorrow 🙂

[AnthonyLatsis]

@swift-ci please smoke test macOS

[AnthonyLatsis]

@swift-ci please smoke test macOS

[hborla]

The property wrapper changes look great, thank you!

[hborla]

inout wrapped parameters were intentionally left out of SE-0293: https://github.com/apple/swift-evolution/blob/main/proposals/0293-extend-property-wrappers-to-function-and-closure-parameters.md#support-inout-in-wrapped-function-parameters

Good point about variadics, presumably this should work and it doesn't today

@propertyWrapper struct Wrapper {
  var wrappedValue: [Int]
}

func test(@Wrapper value: Int...) {
  print(value)
}

let f = test(value:)
f(1, 2, 3) // error: extra arguments in call

[AnthonyLatsis]

I see, will mark inout as a TODO: then! Was it supposed to be an error to have an inout wrapped parameter?

Good point about variadics, presumably this should work and it doesn't today

Interesting, I was expecting @Wrapper value: Int... to mean [Wrapper<Int>] rather than Wrapper<[Int]>.

[hborla]

I think a variadic wrapped parameter would have to mean the wrapped value is an array type because of the restriction that the parameter to init(wrappedValue:) must either exactly match the wrappedValue type or be an autoclosure returning the wrappedValue type.

Was it supposed to be an error to have an inout wrapped parameter?

Oops, yeah I think that should be an error at the declaration. Hopefully no code is actually using an inout wrapped parameter, because the synthesized accessors inside the body don't actually let you mutate the parameter anyway. I also suspect the compiler will crash in the case where the wrapper has init(projectedValue:)

[hborla]

Oh, I misread your comment about variadics as Wrapper<Int>, I understand what you're saying now. Interesting! My brain definitely interpreted @Wrapper value: Int... as a single wrapper.

[AnthonyLatsis]

@swift-ci please smoke test

[AnthonyLatsis]

@swift-ci please smoke test

[AnthonyLatsis]

@swift-ci please smoke test

[AnthonyLatsis]

@swift-ci please smoke test

[AnthonyLatsis]

@swift-ci please test source compatibility

[slavapestov]

Is it worth adding an executable test as well, since there are so many moving parts here?

[AnthonyLatsis]

Is it worth adding an executable test as well, since there are so many moving parts here?

Just for optional methods, or function references in general?

UPD: Added some executable tests for optional methods.

[AnthonyLatsis]

@swift-ci please smoke test

[AnthonyLatsis]

@swift-ci please test source compatibility

[AnthonyLatsis]

@slavapestov Must've been tiring to even skim through all of this, thank you!