Any way to convert AnySchedulerOf<RunLoop> back to RunLoop?

[nsoojin]

Describe the bug
Not a bug
Hi guys, I'm using CADisplayLink in my project and want to inject AnyScheduler to test the behavior. However, CADisplayLink takes in RunLoop instead of Scheduler as a parameter to add(to:,forMode:) method.
So I cannot use AnySchedulerOf that I injected.

Is there a way to convert RunLoop.main.eraseToAnyScheduler() back to RunLoop type? Is it possible?

To Reproduce

// And/or enter code that reproduces the behavior here.
private let mainLoop: AnySchedulerOf<RunLoop>

let displayLink = CADisplayLink(target: self, selector: #selector(update))
displayLink.add(to: self.mainLoop, forMode: .default) // <- Error Above

Expected behavior

Screenshots

Environment

• Xcode 12.5.1
• Swift 5.3
• OS (if applicable): [e.g. iOS 13]

Additional context
Add any more context about the problem here.

[mbrandonw]

Hi @nsoojin, unfortunately this is not possible and for good reason.

If you have an instance of AnySchedulerOf<RunLoop> it may not necessarily be powered by an actual run loop under the hood. In tests you may want to substitute in an immediate scheduler or a test scheduler, and so there wouldn't be a RunLoop to extract out of the AnyScheduler.

However, one of the main reasons one uses AnySchedulers is for testability, but the moment you use a CADisplayLink you are forgoing testability. Because that API takes a concrete run loop it is just impossible to control the flow of time like you can with TestScheduler.

You can see the same problem with Timer.publisher that ships with Foundation. It takes a concrete run loop, and so therefore is not testable at all. It's for this reason that this library ships its own Publishers.Timer which works with any scheduler, so that one can write deterministic tests.

If you share some details how you plan on using CADisplayLink in your code base we may be able to give you some hints on how you can make it a bit more testable.

You may also be interested in this similar discussion: #36

Also I'm going to convert this issue to a discussion since it's not a bug with the library.

[nsoojin]

Hey @mbrandonw, thanks for a very detailed response. It makes sense why it can't be done.

I use CADisplayLink to implement a type of 'video replay' in my app(in development). I have a sequence of images that needs to be displayed according each image's timestamp. The data structure is like this.

struct ReplayLog {
  let timestamp: TimeInterval
  let frameID: String // a unique token to retrieve the frame from a fetcher.
}

So I'm running a display link on main runloop, then for each cycle, I render the frame that needs to be shown based on the timestamp.

In hopes of writing deterministic tests that don't need waiting some arbitrary amount of time, I wanted to test if I pass Array to my player, it requests the correct number of frames to the fetcher. For example, if I pass an array of three items, test if the player requested for three correct frameIDs.

[mbrandonw]

I think your only option for making that code testable is to design a dependency wrapper so that you can hide the CADisplayLink. Then the live version of the dependency can use CADisplayLink and the mock/test-friendly version can just use a plain timer that works with schedulers.

[nsoojin]

thanks for the guidance