[5.8] Switch expression causes build failure because of type inference change

[keith]

With Swift 5.7 this code compiles fine:

class Foo {
    private var action2: () -> Void = {}
    func bar() {
        let action = { [weak self] in
            switch true {
            case true:
                self?.bar()
            case false:
                self?.bar()
            }
        }

        self.action2 = action
    }
}

With Swift 5.8 it fails to compile because action is inferred as () -> ()? instead of () -> Void. I assume this is because of switch changing to be an expression, and now the optional result of self?.bar() is used as the return value, but since it breaks the build I figured I would report it anyways in case that's unexpected.

[xedin]

Due to SE-0380, switch is treated as a single-expression and optional chaining produces optional Void result type. The fix here is straightforward - self.action2 should get closure assigned to it directly and thus propagate Void into the body but I think behavior change needs to be formally discussed by the language group (cc @hamishknight).

[keith]

In our case the easier fix is to explicitly annotate let action as let action: () -> Void because our real instances are more complicated than inlining the definition.

[xedin]

Sure or add _ = to one of the self?... expressions.

[hamishknight]

FWIW SE-0380 isn't implemented in 5.8, it was in the release/5.8 branch for a week, but was reverted out in #63649. Are you using a 5.8 toolchain build that happened to include that change? Otherwise there might be something else at play here.

[keith]

This is with Xcode 14.3 beta 2, so I guess it shipped with a version pre-revert? It looks like it because this code compiles and prints true:

func a() -> String {
    switch true {
    case true:
        "true"
    case false:
        "false"
    }
}

print(a())