[swift-evolution] Ownership on protocol property requirements

[Adrian Zubarev]

I had a short conversation on Twitter with Joe Groff, here is what he said about it: “nowned/weak are meaningless inside a protocol”.



-- 
Adrian Zubarev
Sent with Airmail

Am 7. Mai 2017 um 13:54:28, Greg Spiers via swift-evolution (swift-evolution at swift.org) schrieb:

Hello,

I hope this is the appropriate place to raise this discussion and see if it's worth a proposal around which direction to take.

I've recently had some confusion around specifying weak for a property in a protocol and surprised later of not being warned if the adopting type does not also specify weak for the property. There is an open bug around this behaviour [SR-479](https://bugs.swift.org/browse/SR-479). Example:

class A {}

protocol P {
    weak var weakVar: A? { get set }
}

class B: P {
    var weakVar: A? // Not declared weak, no compiler warning/error
}

Jordan Rose left the comment, "We need to figure out what weak means in a protocol. We should probably ban it for now."

It is a little inconsistent to allow an ownership attribute in a protocol but not enforce the same attribute in the adopting type I think. I guess the point could be made that it's to document/hint that the type adopting the protocol should use the ownership attribute suggested? The Swift Programming Language reference has this to say about Property Requirements in Protocols:

“A protocol can require any conforming type to provide an instance property or type property with a particular name and type. The protocol doesn’t specify whether the property should be a stored property or a computed property—it only specifies the required property name and type.”

There’s no mention if the protocol can specify weak/unowned for a property.

My concern is that this is a warning when this situation happens in Objective-C and there's another bug that makes this point [SR-1494](https://bugs.swift.org/browse/SR-1494).

Would that be desirable behaviour to warn/error in Swift in the adopting type? Or does it make more sense as Jordan commented on the original bug to ban it in protocols (technically a source breaking change)?

I've done some initial investigation on how difficult it would be to [implement banning it in protocols](https://github.com/gspiers/swift/commit/ecde3ec5f61f259f8a396618e9973bac04536fd0). This would be my first contribution so sorry the code will need some work but I'm happy to try and get it in shape and work on a proposal if people think it's worth fixing.

Thanks,

Greg Spiers

_______________________________________________
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/20170507/b5c984a0/attachment-0001.html>