[SR-2661] Usability regression re: fatalError() within function literals

[swift-ci]

Previous ID	SR-2661
Radar	None
Original Reporter	Darkmoon_UK (JIRA User)
Type	Bug
Status	Resolved
Resolution	Done

Environment

Swift 3.0

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug, 3.0Regression, TypeChecker
Assignee	@slavapestov
Priority	Medium

md5: d96afbd5d3223a6d24b62985a71c1cb6

Issue Description:

The transition to using the 'Never' type as the underlying mechanism for no-return functions such as fatalError() has caused a serious usability regression preventing their use within function literals, where it was possible before.

Consider the following cases:

// This compiles in both Swift 2 and 3
func test1() -> Int {
    fatalError()
}

// This compiled in Swift2 but no longer does in Swift 3
let test2 : () -> Int = {
    fatalError()
}

// Compiles in Swift 3
let test2 : () -> Int = {
    if false { return 0 }
    fatalError()
}

I filed a bug report with Apple which was closed as 'intended behaviour' but I am unsatisfied by this answer. The third case in particular highlights this as a complex inconsistency that should be considered for a potential fix, or at least deserves a detailed explanation.

[belkadan]

Yes, the second case seems reasonable to allow.

[belkadan]

What was your Radar?

[swift-ci]

Comment by Christopher Hatton (JIRA)

The ticket number is 28269358.

[belkadan]

Thanks.

[swift-ci]

Comment by Christopher Hatton (JIRA)

Reddit discussion

[slavapestov]

This should be straightforward to fix, I'll take a look.

The problem stems from the fact that a closure body can either be a series of statements, or a single expression. If it is a single expression, its type must match the return type. The fatalError() call triggers the expression path, leading to a type mismatch.

We already have a special case where if the closure return type is () but the expression type is something else, we treat the closure as a statement, not an expression. Similarly, if the expression type is Never, we should treat the closure body as a statement.

[slavapestov]

Done: #4951

[swift-ci]

Comment by Christopher Hatton (JIRA)

Thanks Slava! :-D