[swift-evolution] Ownership on protocol property requirements

[Greg Spiers]

On Mon, May 8, 2017 at 7:57 AM, David Hart <david at hartbit.com> wrote:
> Sounds great! It should be an easy one to get through,

Thanks David, appreciate it :) I've created a draft of the proposal.
Any feedback would be very welcome. I wasn't sure if there was an
effect on ABI stability or API resilience. I don't think there would
be as this is only correcting syntax.

Also, just adding the link again where I had a try at implementing the
warning/error. Hopefully that can be helpful in getting feedback about
the proposal as well.
https://github.com/gspiers/swift/commit/ecde3ec5f61f259f8a396618e9973bac04536fd0


# Remove ownership keyword support in protocols

* Proposal: [SE-NNNN](NNNN-remove-ownership-keyword-support-in-protocols.md)
* Authors: [Greg Spiers](https://github.com/gspiers)
* Review Manager: TBD
* Status:
* Bug: [SR-479](https://bugs.swift.org/browse/SR-479)

## Introduction

This proposal removes support for the keywords `weak` and `unowned`
for property declarations in a protocol.

Swift-evolution thread: [Ownership on protocol property
requirements](https://lists.swift.org/pipermail/swift-evolution/Week-of-Mon-20170501/036495.html)
thread. (TODO: Add link to rest of discussion)

## Motivation

Currently it's possible to use the weak/unowned keywords for a
property requirement in a protocol. This can lead to confusion as
specifying one of these keywords does not enforce or raise any
warnings in the adopting type of that protocol:

```swift

class A {}

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

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

```

This can lead to unexpected and surprising behaviour from the point of
view of users. They keywords themselves are currently meaningless
inside of a protocol but look like they would have an effect when the
protocol is adopted.

## Proposed solution

Although the case could be made that the keywords should have meaning
in a protocol, as they are currently implemented today they don't have
an effect. This proposal aims to cleanup the misleading syntax and
isn't meant to remove functionality only correct to existing
behaviour.

This proposal suggests removing support for `weak` and `unowned` in a protocol.

## Detailed design

The compiler will flag the use of `weak` and `unowned` in a protocol
and suggest a fix to remove the keyword.

## Source compatibility

This is a source breaking change but one that would only correct code
that is already has broken assumptions. For existing use the compiler
will raise a compilation error. When running in Swift 3 mode a warning
can be generated instead of an error. It could be possible to address
source compatibility through source migration as well.

## Effect on ABI stability

This proposal does not affect ABI stability.

## Effect on API resilience

This proposal should not effect API resilience.

## Alternatives considered

There is an argument in making `weak` and `unowned` have meaning in a
protocol but this does open up other questions and is probably better
as a topic of a separate discussion/proposal. As this would be
additive it can be addressed at a later point when we have a clearer
understanding.