Add nullability annotations to RACSubscriber

[sibljon]

In general, it seems helpful to have nullability annotations specified within RACSubscriber.h, though perhaps there's a strong reason for which this file hasn't already been annotated that I'm not aware of.

However, according to SE-0140:

when an Optional is intentionally passed into Objective-C as a nonnull object, we should bridge some value by bridging the wrapped value, and bridge nones to a singleton such as NSNull

With SE-0140, which has already been implemented in Swift 3.0.1 (available in the Xcode 8.1 beta 3), it is important that nullability annotations be specified on RACSubscriber. Otherwise, when RACSubscriber.sendNext() is passed an Optional.none from within Swift, Swift will (post Swift 3.0.1) be bridge this to NSNull, when previously (pre-Swift 3.0.1) it was visible within the Objective-C sendNext: implementation simply as nil. This results in a change in behavior when ReactiveObjC is compiled with the Swift 3.0.1 compiler

@raylillywhite and I observed this change in behavior in Xcode 8.1 beta 3, and have verified that adding nullability annotations restores the previous behavior that users of ReactiveObjC are relying on.

Here's the difference in the Swift generated interface before and after this change:

Previous Swift generated interface (from Xcode 8.1 beta 3 with Swift 3.0.1)

public protocol RACSubscriber : NSObjectProtocol {

    public func sendNext(_ value: Any!)

    public func sendError(_ error: Error!)

    public func sendCompleted()

    public func didSubscribe(with disposable: RACCompoundDisposable!)
}

New Swift generated interface (from Xcode 8.1 beta 3 with Swift 3.0.1):

public protocol RACSubscriber : NSObjectProtocol {

    public func sendNext(_ value: Any?)

    public func sendError(_ error: Error)

    public func sendCompleted()

    public func didSubscribe(with disposable: RACCompoundDisposable)
}

Relatedly, there seems to be a bug or limitation in the Swift compiler that causes ReactiveObjCBridging's RACSignal.toRACSignal() function to bridge Optional.none to NSNull (in next events' values) even after added the nullability annotation to -[RACSubscriber sendNext:]. @raylillywhite and I are in the process of investigating this and have created ReactiveCocoa/ReactiveObjCBridge/pull/5 to address this.

[mdiep]

This is great! Just one minor change I think we should make.

[mdiep]

Could we use NS_ASSUME_NONNULL_BEGIN and NS_ASSUME_NONNULL_END around the protocol instead of explicitly specifying _Nonnull here?

[sibljon]

👍 @mdiep I just amended my commit to use NS_ASSUME_NONNULL_* as suggested.

[erichoracek]

Shouldn't it be sendNext:(nullable id)value and sendError:(nullable NSError *)error;? As far as I can tell from the docs, these can both be nil.

[mdiep]

@erichoracek is correct. NS_ASSUME_NONNULL_* should only have replaced the _Nonnull uses.

[sibljon]

Yep, good catch @erichoracek. Too much rushing on my part. Updated.

[erichoracek]

Just a quick nit: the NS_ASSUME_NONNULL_* family of macros doesn't need a trailing semicolon in my experience.

[erichoracek]

Same comment re: semicolon usage applies here.

[mdiep]

Thanks!