[swift-evolution] Disallow local parameter names in protocols

Dave Abrahams dabrahams at apple.com
Tue Jul 19 17:30:03 CDT 2016


on Tue Jul 19 2016, James Froggatt <swift-evolution at swift.org> wrote:

> Made a draft proposal to start this discussion, and help define the
> problem. I'm not too biased either way, this isn't a particularly
> major aspect of the language, but obviously the proposal has to
> justify the potential change to the language. This is something due
> for re-evaluation since Swift 3 removed the ‘first-argument
> hide-by-default’ rule, which now makes this simplification possible.
>
> Hoping to get an idea of what people's opinions are on this, and the
> subject of local parameter names leaking to the user in
> general. Please let me know if my reasoning is flawed, anywhere.
>
> ## Introduction
>
> Swift's current treatment of local parameter names is rather
> odd. They're hidden from the caller most of the timer, where an
> external parameter label is specified, but otherwise (if a ‘_’ is
> used), they actually leak out into Xcode's autocomplete. External
> labels are intended to add clarity, so the local names seem like they
> should be an implementation detail. Otherwise, why don't local names
> show up for labelled parameters, too?
>
> The main point in the language affected by this ambiguity is in the
> protocol definition, where local names can be specified, but are only
> in any way useful for autocomplete when implementing the protocol,
> making them useless to the language itself. 

They're useful in documentation comments, which you might have realized
if you had included them in your examples ;-)

> This proposal, if accepted, would simplify the language by removing
> the ability to declare local parameter names from protocols, reducing
> their function signatures to just the actual requirements - label and
> type.

Code completion and doc comments are good enough reasons (for me) to
keep this capability.  I am opposed to any change here.

> ## Motivation
>
> Consider the following code:
>
> ````swift
> protocol ExampleProtocol {
>
> 	func move(from source: Int, to destination: Int)
>
> 	func max(_ a: Int, _ b: Int) -> Int
>
> 	func find(_: String, in: String)
> }
>
> struct ExampleStruct1: ExampleProtocol {
>
> 	func move(from source: Int, to destination: Int) { return }
>
> 	func max(_ a: Int, _ b: Int) -> Int { return 0 }
>
> 	func find(_: String, in: String) { return }
> }
>
> //still conforms, despite differences to the protocol signature
> struct ExampleStruct2: ExampleProtocol {
>
> 	func move(from: Int, to: Int) { return }
>
> 	func max(_ first: Int, _ second: Int) -> Int { return 0 }
>
> 	func find(_ substring: String, in string: String) { return }
> }
> ````
>
> ExampleStruct2 conforms to ExampleProtocol, yet clearly has some
> differences to the signatures defined in the protocol. In
> ExampleStruct2, its `move` function has no local parameter names,
> despite their presence in the protocol definition. Its `max` function
> has differing local names, and `find` has names which aren't present
> in the protocol.
>
> In functions, it is the purpose of external parameter labels serve to
> add clarity to the call site. The local names become variables in the
> function's scope, and are generally invisible to the caller. While
> these local names can be declared in the protocol, they aren't
> enforced, and are effectively useless. A confused user may expect that
> this will enforce the local names, or otherwise specify a requirement,
> similar to other protocol semantics.
>
> ## Proposed solution
>
> Remove the ability to specify local parameter names from protocols.
>
> ## Detailed design
>
> Specification of local parameter names in protocols would trigger a
> compiler error, explaining that local parameter names are not
> enforced, with a fixit to remove the local name from the parameter
> list.
>
> ## Impact on existing code
>
> This will break existing protocols which declare local parameter
> names, but should be easily fixable with the migrator, or the
> previously mentioned fixit.
>
> ## Alternatives considered
>
> Leave protocols as they are now, or enforce local parameter names where specified.
>
> _______________________________________________
> swift-evolution mailing list
> swift-evolution at swift.org
> https://lists.swift.org/mailman/listinfo/swift-evolution
>

-- 
Dave



More information about the swift-evolution mailing list