[swift-evolution] [Proposal draft] Generalized Naming for Any Function

[Douglas Gregor]

>> On Dec 27, 2015, at 10:37 AM, Joe Groff <jgroff at apple.com> wrote:
>> 
>> 
>> On Dec 26, 2015, at 11:22 PM, Douglas Gregor via swift-evolution <swift-evolution at swift.org> wrote:
>> 
>> Hi all,
>> 
>> Here’s a proposal draft to allow one to name any function in Swift. In effect, it’s continuing the discussion of retrieving getters and setters as functions started by Michael Henson here:
>> 
>> 	https://lists.swift.org/pipermail/swift-evolution/Week-of-Mon-20151207/002168.html
>> 
>> the proposal follows, and is available here as well:
>> 
>> 	https://github.com/DougGregor/swift-evolution/blob/generalized-naming/proposals/0000-generalized-naming.md
>> 
>> Comments appreciated!
>> 
>> Generalized Naming for Any Function
>> 
[snip]
> 
>> Getters and setters can be written using dotted syntax within the back-ticks:
>> 
>> let specificTitle = button.`currentTitle.get` // has type () -> String?
>> let otherTitle = UIButton.`currentTitle.get`  // has type (UIButton) -> () -> String?
>> let setTintColor = button.`tintColor.set`     // has type (UIColor!) -> ()
>> The same syntax works with subscript getters and setters as well, using the full name of the subscript:
>> 
>> extension Matrix {
>>   subscript (row row: Int) -> [Double] {
>>     get { ... }
>>     set { ... }
>>   }
>> }
>> 
>> let getRow = someMatrix.`subscript(row:).get` // has type (Int) -> () -> [Double]
>> let setRow = someMatrix.`subscript(row:).set` // has type (Int) -> ([Double]) -> ()
> At least as far as pure Swift is concerned, for unapplied access, like `UIButton.currentTitle`, I think it would be more consistent with the way method references works for that to give you the getter (or lens) without decoration. instance.instanceMethod has type Args -> Ret, and Type.instanceMethod has type Self -> Args -> Ret; by analogy, since instance.instanceProperty has type Ret or inout Ret, it's reasonable to expect Type.instanceProperty to have type Self -> [inout] Ret.

Yes, that seems reasonable.

> Forming a getter or setter partially applied to an instance feels unmotivated to me—{ button.currentTitle } or { button.currentTitle = $0 } already work, and are arguably clearer than this syntax.

I’m not strongly motivated by it in and of itself; rather, I like the idea of being able to get at all of the functions (for completeness/consistency), partly because of the Objective-C selector issue.

> I acknowledge that this leaves forming selectors from setters out to dry, but I feel like that's something that could be incorporated into a "lens" design along with typed selectors. As a rough sketch, we could say that the representation of @convention(selector) T -> inout U is a pair of getter/setter selectors,

I should weigh in over on a typed-selectors thread, but my personal view is that typed selectors are a well-designed feature that isn't worth doing: one would probably not use them outside of interoperability with Objective-C. To make that work, we'd need a Clang feature as well (to express the types), then all of the relevant Objective-C APIs would need to adopt it for us to see the benefits. On iOS, we are talking about a relatively small number of APIs (100-ish), and many of those have blocks/closure-based variants that are preferred. 

> and provide API on Selector to grab the individual selectors from that, maybe Selector(getterFor: UIView.currentTitle)/(setterFor: UIView.currentTitle)

Sure. I suspect that retrieving the selector of a getter/setter will be fairly rare, so I'm fine with that operation being ugly. 

> . I don't think get/set is a good interface for working with Swift properties, so I don't like the idea of building in language support to codify it beyond what's needed for ObjC interaction.

That is an excellent point. I think you've convinced me to drop the getter/setter part of this: lenses are the right abstraction for working with properties, and we can handle ObjC getter/setter in some other way. 

>> Can we drop the back-ticks? It's very tempting to want to drop the back-ticks entirely, because something like
>> 
>> let fn = someView.insertSubview(_:at:)
>> can be correctly parsed as a reference to insertSubview(_:at:). However, it breaks down at the margins, e.g., with getter/setter references or no-argument functions:
>> 
>> extension Optional {
>>   func get() -> T { return self! }
>> }
>> 
>> let fn1 = button.currentTitle.get   // getter or Optional<String>.get?
>> let fn2 = set.removeAllElements()   // call or reference?
> From what I remember, the bigger concern with allowing foo(bar:bas:) without backticks is parser error recovery. The unambiguity with call syntax depends on having the `:)` token pair at the end. The edit distance between foo(bar:bas:) and a call foo(bar: bas) or work-in-progress call foo(bar: x, bas: ) is pretty slight, and would be tricky to give good diagnostics for. If we felt confident we could give good diagnostics, I'd support removing the backticks.
> 
> -Joe

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20151229/f472381f/attachment.html>