[swift-evolution] [Draft] Remove support for final in protocol extensions

Rod Brown rodney.brown6 at icloud.com
Thu Mar 9 08:32:29 CST 2017 

There has been a lot of discussion around this design decision. Personally, I’m with you: this should be allowed. Protocol extensions should be defaults, nothing more.

The rationale mentioned in Swift Evolution for discouraging this behaviour tends to be that if you conform to the protocol, you should conform and adhere to all its extensions as well, and that not doing so in the same way will be inconsistent.

I personally think this comes to the Type-first vs Protocol-first approach and I think instances of types should have the final say in the behaviour of the operation. While this would perform slightly worse, and could potentially be unsafe, I think it is consistent with the fact that not all implementers of a protocol behave exactly the same; indeed, if they did, we’d only have one type per protocol, in which case, what is the point of a protocol at all? Just do it with types.

I love POP and protocol extensions but I’m tempered by the fact that not all types implement protocols in the same ways, and we can’t predict ahead of time where those differences will be needed.


> On 10 Mar 2017, at 12:48 am, Adrian Zubarev via swift-evolution <swift-evolution at swift.org> wrote:
> 
> My question is why cannot the Base type override the default implementation? I might want to override it, by calling the default implementation and modifying the result for my needs.
> 
> Something like that:
> 
> protocol P {
>     func foo() -> Int
> }
> 
> extension P {
>     func foo() -> Int {
>         return 42
>     }
> }
> 
> class Base : P {
>     override func foo() -> {
>          
>         return default.foo() * 100
>     }
> }
> The example is kept simple.
> 
> 
> 
> 
> -- 
> Adrian Zubarev
> Sent with Airmail
> 
> Am 9. März 2017 um 14:37:08, Matthew Johnson via swift-evolution (swift-evolution at swift.org <mailto:swift-evolution at swift.org>) schrieb:
> 
>> 
>> 
>> Sent from my iPad
>> 
>> On Mar 9, 2017, at 2:23 AM, Slava Pestov via swift-evolution <swift-evolution at swift.org <mailto:swift-evolution at swift.org>> wrote:
>> 
>>> I think the fact that the type checker permits ‘final’ to appear inside protocol extensions is an oversight and this probably does not even warrant a proposal. I don’t think allowing this was ever part of the conceptual model of protocol extensions at any point in time (if you recall they were introduced ‘recently’, in Swift 2). If someone put together a PR which makes ‘final’ in protocol extensions an error in Swift 4 mode (and a warning in 3), I would merge it.
>>> 
>>> FWIW that there is one restriction around the direct dispatch here we want to lift, but it’s not related to this proposal.
>>> 
>>> If you have a base class conforming to a protocol using default requirements, eg
>>> 
>>>   protocol Proto { func f() }
>>>   extension Proto { func f() { } }
>>> 
>>>   class Base : Proto {}
>>> 
>>> Currently the witness table for Base : Proto directly references the extension method Proto.f.
>>> 
>>> We want to allow this, at least inside the module:
>>> 
>>> class Derived {
>>>   override func f() {} // does not work currently
>>> }
>>> 
>>> This will mean that ‘Proto.f’ will appear in the vtable of ‘Base’, pointing at the extension method. The conformance will dispatch through the vtable instead of directly calling the extension method.
>> 
>> Would this allow the override to call `Proto.f` through super?
>> 
>>> 
>>> Slava
>>> 
>>>> On Mar 7, 2017, at 7:23 PM, Brian King via swift-evolution <swift-evolution at swift.org <mailto:swift-evolution at swift.org>> wrote:
>>>> 
>>>> Hey Folks, This draft proposal addresses starter bug SR-1762. I believe this is in scope for Swift 4 since it impacts source compatibility. It's not a very exciting proposal, but I think it will help make Swift a little more consistent.
>>>> 
>>>> https://gist.github.com/KingOfBrian/6f20c566114ac0ef54c8092d80e54ee7 <https://gist.github.com/KingOfBrian/6f20c566114ac0ef54c8092d80e54ee7>
>>>> https://bugs.swift.org/browse/SR-1762 <https://bugs.swift.org/browse/SR-1762>
>>>> 
>>>> Thanks
>>>> 
>>>> Brian
>>>> Introduction
>>>> 
>>>> This proposal suggests removing support for the final keyword when declaring a function in a protocol extension. The presence of the final keyword does not currently generate an error message, and it does not actually modify the dispatch behavior in any way.
>>>> 
>>>>  <https://gist.github.com/KingOfBrian/6f20c566114ac0ef54c8092d80e54ee7#motivation>Motivation
>>>> 
>>>> In the original protocol model of Swift, a developer could use the final keyword when declaring a function in a protocol extension to ensure the function could not be overridden. This keyword has no use in Swift's current protocol model, since functions in protocol extensions can not be overridden and will always use direct dispatch.
>>>> 
>>>>  <https://gist.github.com/KingOfBrian/6f20c566114ac0ef54c8092d80e54ee7#detailed-design>Detailed design
>>>> 
>>>> The compiler should generate an error or warning when the final keyword is used on a function declaration inside of a protocol extension. This is consistent with the use of final in structs and enumerations.
>>>> 
>>>>  <https://gist.github.com/KingOfBrian/6f20c566114ac0ef54c8092d80e54ee7#source-compatibility>Source compatibility
>>>> 
>>>> This change will impact source compatibility. To maintain compatibility with Swift 3, a warning will be generated in Swift 3 mode instead of an error message.
>>>> 
>>>>  <https://gist.github.com/KingOfBrian/6f20c566114ac0ef54c8092d80e54ee7#effect-on-abi-stability>Effect on ABI stability
>>>> 
>>>> This has no effect on ABI stability
>>>> 
>>>>  <https://gist.github.com/KingOfBrian/6f20c566114ac0ef54c8092d80e54ee7#effect-on-api-resilience>Effect on API resilience
>>>> 
>>>> This has no effect on API resilience
>>>> 
>>>>  <https://gist.github.com/KingOfBrian/6f20c566114ac0ef54c8092d80e54ee7#alternatives-considered>Alternatives considered
>>>> 
>>>> The only alternative would be to not fix this bug
>>>> 
>>>> _______________________________________________
>>>> swift-evolution mailing list
>>>> swift-evolution at swift.org <mailto:swift-evolution at swift.org>
>>>> https://lists.swift.org/mailman/listinfo/swift-evolution <https://lists.swift.org/mailman/listinfo/swift-evolution>
>>> 
>>> _______________________________________________
>>> swift-evolution mailing list
>>> swift-evolution at swift.org <mailto:swift-evolution at swift.org>
>>> https://lists.swift.org/mailman/listinfo/swift-evolution <https://lists.swift.org/mailman/listinfo/swift-evolution>
>> _______________________________________________
>> swift-evolution mailing list
>> swift-evolution at swift.org
>> https://lists.swift.org/mailman/listinfo/swift-evolution
> 
> 
> _______________________________________________
> swift-evolution mailing list
> swift-evolution at swift.org <mailto:swift-evolution at swift.org>
> https://lists.swift.org/mailman/listinfo/swift-evolution <https://lists.swift.org/mailman/listinfo/swift-evolution>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20170310/a96324cc/attachment.html>

More information about the swift-evolution mailing list