[swift-evolution] [Pitch] Never as a bottom type

[Xiaodi Wu]

Clearly many uses for this; would like to see it some day, definitely.
However, IIRC, the last day for any _code_ to be incorporated in Swift 4 is
June 1. There's simply no chance that this can be considered for the
current phase of Swift evolution.

When it does come into scope (which will depend on the core team's
identified priorities), the one thing I would comment is that there is no
need for a name such as "undefined": we already have "fatalError()".


On Sun, May 14, 2017 at 17:27 André Videla via swift-evolution <
swift-evolution at swift.org> wrote:

> Hi Swift-Evolution,
>
> At the end of SE-0102 (
> https://github.com/apple/swift-evolution/blob/master/proposals/0102-noreturn-bottom-type.md)
> it is mentioned that the Never type could be used as a universal bottom
> type but that is currently not the case.
>
> As of today the Never type is used as a return type of `fatalError` and as
> function which do not return by the virtue of calling `fatalError`.
> This allows for very useful behaviours for example:
>
> let anything: Any = …
>
> switch anything {
> case let view as UIView: …
> case let str as String: …
> default: fatalError("we got an unexpected type")
> }
>
> But it has its limits, most notably, it cannot be used as an expression
>
> let dunno: Int = fatalError("what should I put here?") // cannot convert
> value of type 'Never'
>
> It makes sense because Never is not a bottom type. If it were, this
> statement would be absolutely valid.
>
> Having a bottom type and a value for it has several advantages:
>
> - More informative error messages with forced unwrap:
>
> protocol WonkyAPI {
>     func apiCall() -> Int? //this always return an Int, why is it
> optional?
> }
>
> func mycode() {
>     let valueFromAPI = apiCall() ?? fatalError("The API used to always
> return a value, but now it does not!")
>     …
> }
>
> It sometimes happen that some part of the code uses an optional but in
> your particular codepath, the optional is always containing a value (for
> example after an assignment).
> As of today, you can write
> guard let value = value else { fatalError("something terrible happened") }
> for the same effect with a more verbose syntax.
>
> - Use as a hole/placeholder during development
> During development it is very likely that you will want to write a piece
> of functionality but be stuck on an intermediate part of the computation.
> Assume we have an identifier `undefined` of type `Never` which would
> represent an impossible value as a bottom type. We would ben able to write:
>
> func transform(point: CGPoint) -> CGPoint {
>     let translation =  Matrix3D(1, 0, 2,
>                                 0, 1, -2,
>                                 0, 0, 1)
>     let rotation: Matrix3D = undefined //what was it? I forgot
>     return (translation * rotation * point.homogenous).toPoint()
> }
>
> We can debate on the right naming for undefined. Haskell uses 'undefined',
> Scala uses `???`. `unimplemented`, `impossible`, `void`are all valid
> contenders.
>
> - Eliminate type annotations for generic covariant types
> As of today this is not valid
>
> struct Person {
>     let name: String
> }
>
> var maybeSomeone = nil
> maybeSomeone = Person(name: "Doug”)
>
> Even though it is clear that maybeSomeone is of type Optional<Person>.
> That is because the compiler cannot guess which type the Optional wraps
> when `maybeSomeone` is declared. But with a bottom type, a naked nil can be
> mapped to `Optional<Never>` until the type inference figures out from the
> context what is the type of Optional. If it cannot because no use or no
> assignment is done, the compiler could emit an “unreachable” warning just
> like it does for
>
> func unreach() {
>     fatalError("stop here")
>     print("not printed”) // warning: will never be executed
> }
>
> Should I write a proposal?
>
> André Videla
> _______________________________________________
> swift-evolution mailing list
> swift-evolution at swift.org
> https://lists.swift.org/mailman/listinfo/swift-evolution
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20170515/3cf3a92c/attachment.html>