[Sema] Add warning for ambiguous value assignment when using Optional

[theblixguy]

This issue is best explained with an example:

enum Foo {
  case bar
  case none
}

// The assigned value is (silently) Optional.none
// rather than Foo.none as you would expect.
let _: Foo? = .none

let test: Foo? = Foo.none
// False, as it matches Optional.none, rather than Foo.none.
let result: Bool = test == .none 

// Same with a struct // 

struct Foo {
  static let bar = Foo()
  static let none = Foo()
}

let _: Foo? = .none // uh-oh

This PR fixes this issue by diagnosing it as a warning and offering two fixits: prefix Optional or prefix the explicit type to silence the warning.

Resolves SR-2176, SR-3711, SR-6044, SR-8283, SR-8451 & SR-8515.
Resolves rdar://problem/26126801.

[theblixguy]

cc @jrose-apple @xedin

[xedin]

Overall, I think this is supposed to be an error instead of a warning but warning is better than nothing.

It seems like it could only happen in expression context, so maybe it would make sense for this warning to be produced by the constraint solver instead? You can take a look at ExprRewritter::visitUnresolvedMemberExpr, I think it would be reasonable to put that check there if the base was optional type.

Also it might be reasonable to mention member name as well e.g. enum case .noneexists in both 'E' andOptional, picked the one from 'Optional<E>' or something like that...

[slavapestov]

A warning is our only choice since an error would create a source compatibility issue. Here's a test case demonstrating that it needs to happen in CSApply's ExprRewriter and not when checking the initial value of a property:

enum E {
  case none
}

func foo(_: E?) {}

foo(.none) // should be the same warning?

[theblixguy]

@slavapestov @xedin done 👍

[beccadax]

I have a lot of suggestions, but this is a really solid start!

[theblixguy]

@slavapestov can you invoke the CI again?

[xedin]

@swift-ci please smoke test

[theblixguy]

Seems like I should've attached the fix-it to the expected-note rather than the expected-warning. Also, there was a crash because it was looking up the enum case in Optional extensions or tuple types, which I have now fixed.

@xedin can you invoke the CI now? The tests should now pass 🙏

[xedin]

@swift-ci please smoke test

[theblixguy]

The column numbers I used for the fixits differed by 1 🤦‍♂️ @xedin can you invoke the CI once again? Thanks 😄

[xedin]

@swift-ci please smoke test

[theblixguy]

All tests have passed 🎉 @xedin

[theblixguy]

@brentdax @xedin could you invoke the CI now?

[beccadax]

@swift-ci please smoke test

[theblixguy]

Good to merge now? @brentdax @xedin