MiniFuture

A monadic Future design pattern implementation in Swift language, using libdispatch and POSIX mutexes and condition variables. Design inspired from Scala’s scala.concurrent.Future.

Only the essential features are in place currently. We’re using the library in production, and it appears to work as expected. There’s a benchmark that acts as a stress test, see Performance below.

Requirements

• Mac OS X >= 10.9
• Xcode >= 9.0 (Swift 4.0)

Installation

CocoaPods

CocoaPods is a centralized dependency manager for Cocoa libraries. It integrates libraries as embedded frameworks to your project. This requires that the minimum deployment target of your project is iOS 8.0 or OS X 10.9.

MiniFuture is available as a pod. Add the following line to your project’s Podfile in order to have the library as a dependency:

pod 'MiniFuture', '~> 0.5.0'

And run pod install.

In your source code, import MiniFuture to use the library.

Manually

Copy all files under Source directory to your project. You have to figure out how to upgrade the library yourself.

Characteristics

A future is a promise of a computation that will complete eventually. We use Try<T> value type to wrap these computation results. It’s an enumeration with two members, success<T> and failure<Error>. The first is meant for the caller of the Future to signify successful computation, with the result of the computation as the associated value of type T. The latter member signifies computation failure, with Error as the associated value.

For example, the following future will complete with .success(1) eventually:

let fut = Future<Int>.async {
  Thread.sleep(forTimeInterval: 0.2)
  return .success(1)
}

assert(succeeded.get() == Try.success(1))

The basic operation to compose Futures is with Future#flatMap(_:). The method signature is:

flatMap<U>(f: T throws -> Future<U>) -> Future<U>

It takes in a closure, f. If the current future completes with a success value, f gets called with T as its argument. The return value of f is the next future computation. Later, if that future completes with a success value, the closure passed to that future gets called. This is the asynchronous call chain of the futures.

If the current future completes with a failure value, f will not be called. This short-circuits the call chain of the futures.

The closure f may throw an error when called. If that happens, the return value of f will be a future completed with a failure value containing the error.

For example:

enum AppError: Error {
  case deliberate(String)
}

let fut: Future<[Int]> = Future.async { .success(0) }
  .flatMap { Future.succeeded([$0, 1]) }
  .flatMap { throw AppError.deliberate(String(describing: $0)) }
  .flatMap { Future.succeeded($0 + [2]) }

assert(String(describing: fut.get()) == "Try.failure(deliberate(\"[0, 1]\"))")

Try<T> and the automatic handling of thrown errors is borrowed from Scala 2.10.

All asynchronous operations run in libdispatch’s default global concurrent queue. Closures passed to Future#flatMap(_:), Future#map(_:), Future#onComplete(_:), and Future.async(_:) are always executed in a queue worker thread. Use proper synchronization when accessing shared state via references captured in the closures.

Usage

To get a Future job running, use Future.succeeded(_:) and Future.failed(_:) to wrap immediate values. These return ImmediateFuture objects, a Future implementation class already completed with success or failure value.

Use Future.async(_:) for asynchronous jobs that compute the value later in a queue worker thread. You pass a block to async(_:) and return either a Try.success<T> or Try.failure<Error> value from it. The Future implementation class here is AsyncFuture.

For adapting existing asynchronous interfaces with Futures, use Future.promise(_:). This returns a PromiseFuture object, a promise kind of Future implementation class. Pass the Future to an existing asynchronous interface, and in the completion handler of the interface, complete the Future with success (Future#resolve(_:)) or failure (Future#reject(_:)). You can immediately return a PromiseFuture to code expecting Futures and let the PromiseFuture object complete later.

You can complete a PromiseFuture with the result of another future, too. Call PromiseFuture#completeWith(_:), passing another future as the argument. Once the future completes, the promise completes with the same result as the future.

When you get a handle to a Future, use Future#flatMap(_:) or Future#map(_:) to compose another Future that depends on the completed result of the previous Future. Use Future#get() to wait for the result of a Future. Use Future#onComplete(_:) to add a callback for side-effects to be run when the Future completes.

Example

extension String {
  var trimmed: String {
    return trimmingCharacters(in: CharacterSet.whitespacesAndNewlines)
  }

  func excerpt(_ maxLength: Int) -> String {
    precondition(maxLength >= 0, "maxLength must be positive")

    if maxLength == 0 {
      return ""
    }

    let length = characters.count

    if length <= maxLength {
      return self
    }

    return self[startIndex..<characters.index(startIndex, offsetBy: maxLength-1)].trimmed + "…"
  }
}

/**
 * Request a web resource asynchronously, immediately returning a handle to
 * the job as a promise kind of Future. When NSURLSession calls the completion
 * handler, we fullfill the promise. If the completion handler gets called
 * with the contents of the web resource, we resolve the promise with the
 * contents (the success case). Otherwise, we reject the promise with failure.
 */
func loadURL(_ url: URL) -> Future<Data> {
  let promise = Future<Data>.promise()
  let task = URLSession.shared.dataTask(with: url, completionHandler: { data, response, error in
    if let err = error {
      promise.reject(err)
    } else if let d = data {
      promise.resolve(d)
    } else {
      promise.reject(AppError.failedLoadingURL(url))
    }
  })
  task.resume()
  return promise
}

/**
 * Parse data as HTML document, finding specific contents from it with an
 * XPath query. We return a completed Future as the handle to the result. If
 * we can parse the data as an HTML document and the query succeeds, we return
 * a successful Future with the query result. Otherwise, we return failed
 * Future describing the error.
 *
 * Because this function gets called inside `Future#flatMap`, it's run in
 * background in a queue worker thread.
 */
func readXPathFromHTML(_ xpath: String, data: Data) throws -> Future<HTMLNode> {
  let doc = try HTMLDocument.readData(asUTF8: data)
  let node = try doc.rootHTMLNode()
  let found = try node.forXPath(xpath)
  return Future.succeeded(found)
}

let wikipediaURL = URL(string: "https://en.wikipedia.org/wiki/Main_Page")!
let featuredArticleXPath = "//*[@id='mp-tfa']"

let result = loadURL(wikipediaURL)
  /* Future composition (chaining): when this Future completes successfully,
   * pass its result to a function that does more work, returning another
   * Future. If this Future completes with failure, the chain short-circuits
   * and further flatMap methods are not called. Calls to flatMap are always
   * executed in a queue worker thread.
   */
  .flatMap { try readXPathFromHTML(featuredArticleXPath, data: $0) }
  /* Wait for Future chain to complete. This acts as a synchronization point.
   */
  .get()

switch result {
case .success(let value):
  let excerpt = value.textContents!.excerpt(78)
  print("Excerpt from today's featured article at Wikipedia: \(excerpt)")
case .failure(let error):
  print("Error getting today's featured article from Wikipedia: \(error)")
}

See more in Example/main.swift. You can run the examples:

$ make example
# xcodebuild output…

./build/Example
Excerpt from today's featured article at Wikipedia: Upper and Lower Table Rock are two prominent volcanic plateaus just north of…

Performance

There’s a benchmark in Benchmark/main.swift. It builds up complex nested Futures (the futEnd variable in the code) in a loop 2000 times (numberOfFutureCompositions) and chains them into one big composite Future (the fut variable). Then the benchmark waits for the Future to complete.

We repeat this 500 times (numberOfIterations) to get the arithmetic mean and standard deviation of time spent completing each composite Future.

Compile it with Release build configuration, which enables -O compiler flag. Then run it from the terminal.

Example run with MacBook Pro (MacBookPro11,3) 2.6 GHz Intel Core i7 Haswell, 16 GB 1600 MHz DDR3:

$ make benchmark
# xcodebuild output…

./build/Benchmark
iterations: 500, futures composed: 2000

warm up: 71 ms (± 2 ms)
measure: 71 ms (± 2 ms)

Apple Swift version 4.0.3 (swiftlang-900.0.74.1 clang-900.0.39.2)
Target: x86_64-apple-macosx10.9

Total memory consumption of the process stayed below 20 MB.

Future work

• Implement Future cancellation and timeouts
• Implement more composition operations on Futures

License

MiniFuture is released under the MIT license. See LICENSE.txt for details.