Skip to content

Latest commit

 

History

History
492 lines (373 loc) · 12.4 KB

CommonPatterns.md

File metadata and controls

492 lines (373 loc) · 12.4 KB

Common Patterns

One feature of promises that makes them particularly useful is that they are composable. This fact enables complex, yet safe asynchronous patterns that would otherwise be quite intimidating when implemented with traditional methods.

Chaining

The most common pattern is chaining:

firstly {
    fetch()
}.then {
    map($0)
}.then {
    set($0)
    return animate()
}.ensure {
    // something that should happen whatever the outcome
}.catch {
    handle(error: $0)
}

If you return a promise in a then, the next then waits on that promise before continuing. This is the essence of promises.

Promises are easy to compose, so they encourage you to develop highly asynchronous apps without fear of the spaghetti code (and associated refactoring pains) of asynchronous systems that use completion handlers.

APIs That Use Promises

Promises are composable, so return them instead of accepting completion blocks:

class MyRestAPI {
    func user() -> Promise<User> {
        return firstly {
            URLSession.shared.dataTask(.promise, with: url)
        }.compactMap {
            try JSONSerialization.jsonObject(with: $0.data) as? [String: Any]
        }.map { dict in
            User(dict: dict)
        }
    }

    func avatar() -> Promise<UIImage> {
        return user().then { user in
            URLSession.shared.dataTask(.promise, with: user.imageUrl)
        }.compactMap {
            UIImage(data: $0.data)
        }
    }
}

This way, asynchronous chains can cleanly and seamlessly incorporate code from all over your app without violating architectural boundaries.

Note: We provide promises for Alamofire too!

Background Work

class MyRestAPI {
    func avatar() -> Promise<UIImage> {
        let bgq = DispatchQueue.global(qos: .userInitiated)

        return firstly {
            user()
        }.then(on: bgq) { user in
            URLSession.shared.dataTask(.promise, with: user.imageUrl)
        }.compactMap(on: bgq) {
            UIImage(data: $0)
        }
    }
}

All PromiseKit handlers take an on parameter that lets you designate the dispatch queue on which to run the handler. The default is always the main queue.

PromiseKit is entirely thread safe.

Tip: With caution, you can have all then, map, compactMap, etc., run on a background queue. See PromiseKit.conf. Note that we suggest only changing the queue for the map suite of functions, so done and catch will continue to run on the main queue, which is usually what you want.

Failing Chains

If an error occurs mid-chain, simply throw an error:

firstly {
    foo()
}.then { baz in
    bar(baz)
}.then { result in
    guard !result.isBad else { throw MyError.myIssue }
    //…
    return doOtherThing()
}

The error will surface at the next catch handler.

Since promises handle thrown errors, you don't have to wrap calls to throwing functions in a do block unless you really want to handle the errors locally:

foo().then { baz in
    bar(baz)
}.then { result in
    try doOtherThing()
}.catch { error in
    // if doOtherThing() throws, we end up here
}

Tip: Swift lets you define an inline enum Error inside the function you are working on. This isn’t great coding practice, but it's better than avoiding throwing an error because you couldn't be bothered to define a good global Error enum.

Abstracting Away Asynchronicity

var fetch = API.fetch()

override func viewDidAppear() {
    fetch.then { items in
        //…
    }
}

func buttonPressed() {
    fetch.then { items in
        //…
    }
}

func refresh() -> Promise {
    // ensure only one fetch operation happens at a time

    if fetch.isResolved {
        startSpinner()
        fetch = API.fetch().ensure {
            stopSpinner()
        }
    }
    return fetch
}

With promises, you don’t need to worry about when your asynchronous operation finishes. Just act like it already has.

Above, we see that you can call then as many times on a promise as you like. All the blocks will be executed in the order they were added.

Chaining Sequences

When you have a series of tasks to perform on an array of data:

// fade all visible table cells one by one in a “cascading” effect

let fade = Guarantee()
for cell in tableView.visibleCells {
    fade = fade.then {
        UIView.animate(.promise, duration: 0.1) {
            cell.alpha = 0
        }
    }
}
fade.done {
    // finish
}

Or if you have an array of promises:

var foo = Promise()
for nextPromise in arrayOfPromises {
    foo = foo.then { nextPromise }
}
foo.done {
    // finish
}

Note: You usually want when(), since when executes all of its component promises in parallel and so completes much faster. Use the pattern shown above in situations where tasks must be run sequentially; animation is a good example.

We also provide when(concurrently:), which lets you schedule more than one promise at a time if you need to.

Timeout

let fetches: [Promise<T>] = makeFetches()
let timeout = after(seconds: 4)

race(when(fulfilled: fetches).asVoid(), timeout).then {
    //…
}

race continues as soon as one of the promises it is watching finishes.

Make sure the promises you pass to race are all of the same type. The easiest way to ensure this is to use asVoid().

Note that if any component promise rejects, the race will reject, too.

Minimum Duration

Sometimes you need a task to take at least a certain amount of time. (For example, you want to show a progress spinner, but if it shows for less than 0.3 seconds, the UI appears broken to the user.)

let waitAtLeast = after(seconds: 0.3)

firstly {
    foo()
}.then {
    waitAtLeast
}.done {
    //…
}

The code above works because we create the delay before we do work in foo(). By the time we get to waiting on that promise, either it will have already timed out or we will wait for whatever remains of the 0.3 seconds before continuing the chain.

Cancellation

Promises don’t have a cancel function, but they do support cancellation through a special error type that conforms to the CancellableError protocol.

func foo() -> (Promise<Void>, cancel: () -> Void) {
    let task = Task()
    var cancelme = false

    let promise = Promise<Void> { seal in
        task.completion = { value in
            guard !cancelme else { return reject(PMKError.cancelled) }
            seal.fulfill(value)
        }
        task.start()
    }

    let cancel = {
        cancelme = true
        task.cancel()
    }

    return (promise, cancel)
}

Promises don’t have a cancel function because you don’t want code outside of your control to be able to cancel your operations--unless, of course, you explicitly want to enable that behavior. In cases where you do want cancellation, the exact way that it should work will vary depending on how the underlying task supports cancellation. PromiseKit provides cancellation primitives but no concrete API.

Cancelled chains do not call catch handlers by default. However you can intercept cancellation if you like:

foo.then {
    //…
}.catch(policy: .allErrors) {
    // cancelled errors are handled *as well*
}

Important: Canceling a promise chain is not the same as canceling the underlying asynchronous task. Promises are wrappers around asynchronicity, but they have no control over the underlying tasks. If you need to cancel an underlying task, you need to cancel the underlying task!

The library CancellablePromiseKit extends the concept of Promises to fully cover cancellable tasks.

Retry / Polling

func attempt<T>(maximumRetryCount: Int = 3, delayBeforeRetry: DispatchTimeInterval = .seconds(2), _ body: @escaping () -> Promise<T>) -> Promise<T> {
    var attempts = 0
    func attempt() -> Promise<T> {
        attempts += 1
        return body().recover { error -> Promise<T> in
            guard attempts < maximumRetryCount else { throw error }
            return after(delayBeforeRetry).then(on: nil, attempt)
        }
    }
    return attempt()
}

attempt(maximumRetryCount: 3) {
    flakeyTask(parameters: foo)
}.then {
    //…
}.catch { _ in
    // we attempted three times but still failed
}

In most cases, you should probably supplement the code above so that it re-attempts only for specific error conditions.

Wrapping Delegate Systems

Be careful with Promises and delegate systems, as they are not always compatible. Promises complete once, whereas most delegate systems may notify their delegate many times. This is why, for example, there is no PromiseKit extension for a UIButton.

A good example of an appropriate time to wrap delegation is when you need a single CLLocation lookup:

extension CLLocationManager {
    static func promise() -> Promise<CLLocation> {
        return PMKCLLocationManagerProxy().promise
    }
}

class PMKCLLocationManagerProxy: NSObject, CLLocationManagerDelegate {
    private let (promise, seal) = Promise<[CLLocation]>.pending()
    private var retainCycle: PMKCLLocationManagerProxy?
    private let manager = CLLocationManager()

    init() {
        super.init()
        retainCycle = self
        manager.delegate = self // does not retain hence the `retainCycle` property

        promise.ensure {
            // ensure we break the retain cycle
            self.retainCycle = nil
        }
    }

    @objc fileprivate func locationManager(_: CLLocationManager, didUpdateLocations locations: [CLLocation]) {
        seal.fulfill(locations)
    }

    @objc func locationManager(_: CLLocationManager, didFailWithError error: Error) {
        seal.reject(error)
    }
}

// use:

CLLocationManager.promise().then { locations in
    //…
}.catch { error in
    //…
}

Please note: we provide this promise with our CoreLocation extensions at https://github.com/PromiseKit/CoreLocation

Recovery

Sometimes you don’t want an error to cascade. Instead, you want to supply a default result:

CLLocationManager.requestLocation().recover { error -> Promise<CLLocation> in
    guard error == MyError.airplaneMode else {
        throw error
    }
    return .value(CLLocation.savannah)
}.done { location in
    //…
}

Be careful not to ignore all errors, though! Recover only those errors that make sense to recover.

Promises for Modal View Controllers

class ViewController: UIViewController {

    private let (promise, seal) = Guarantee<…>.pending()  // use Promise if your flow can fail

    func show(in: UIViewController) -> Promise<…> {
        in.show(self, sender: in)
        return promise
    }

    func done() {
        dismiss(animated: true)
        seal.fulfill()
    }
}

// use:

ViewController().show(in: self).done {
    //…
}.catch { error in
    //…
}

This is the best approach we have found, which is a pity as it requires the presentee to control the presentation and requires the presentee to dismiss itself explicitly.

Nothing seems to beat storyboard segues for decoupling an app's controllers.

Saving Previous Results

Let’s say you have:

login().then { username in
    fetch(avatar: username)
}.done { image in
    //…
}

What if you want access to both username and image in your done?

The most obvious way is to use nesting:

login().then { username in
    fetch(avatar: username).done { image in
        // we have access to both `image` and `username`
    }
}.done {
    // the chain still continues as you'd expect
}

However, such nesting reduces the clarity of the chain. Instead, we could use Swift tuples:

login().then { username in
    fetch(avatar: username).map { ($0, username) }
}.then { image, username in
    //…
}

The code above simply maps Promise<String> into Promise<(UIImage, String)>.

Waiting on Multiple Promises, Whatever Their Result

Use when(resolved:):

when(resolved: a, b).done { (results: [Result<T>]) in
    // `Result` is an enum of `.fulfilled` or `.rejected`
}

// ^^ cannot call `catch` as `when(resolved:)` returns a `Guarantee`

Generally, you don't want this! People ask for it a lot, but usually because they are trying to ignore errors. What they really need is to use recover on one of the promises. Errors happen, so they should be handled; you usually don't want to ignore them.