A Swift Reactive Programming Kit
Swift Other
Latest commit 3600e31 Jan 17, 2017 @srdanrasic srdanrasic Up version


Platform Build Status Join Us on Gitter Twitter

ReactiveKit is a Swift framework for reactive and functional reactive programming.

The framework is best used in a combination with Bond that provides UIKit and AppKit bindings, reactive delegates and data sources.

Reactive Programming

Apps transform data. They take some data as input or generate data by themselves, transform that data into another data and output new data to the user. An app could take computer-friendly response from an API, transform it to a user-friendly text with a photo or video and render an article to the user. An app could take readings from the magnetometer, transform them into an orientation angle and render a nice needle to the user. There are many examples, but the pattern is obvious.

Basic premise of reactive programming is that the output should be derived from the input in such way that whenever the input changes, the output is changed too. Whenever new magnetometer readings are received, needle is updated. In addition to that, if the input is derived into the output using functional constructs like pure or higher-order functions one gets functional reactive programming.

ReactiveKit is a framework that provides mechanisms for leveraging functional reactive paradigm. It is based on ReactiveX API and provides Signal type that is generic both over the elements it generates and over the errors it can terminate with. ReactiveKit places great importance on errors and enforces you to handle them in compile time.

ReactiveKit aims to be the simplest yet complete framework for functional reactive programming in Swift. The goal is that you can dive in into each operator's implementation and understand it in under a minute. This makes ReactiveKit very lightweight and easy to learn, but also focused just on reactive paradigm. To get the best of the ReactiveKit in Cocoa / Cocoa Touch development use it with the Bond framework that provides reactive delegates, data sources and binding extensions for various UIKit and AppKit objects. Bond v5 is built on top of ReactiveKit.


Main type that ReactiveKit provides is Signal. It's used to represent a signal of events. Event can be anything from a button tap to a voice command or network response.

Signal event is defined by Event type and looks like this:

public enum Event<Element, Error: Swift.Error> {
  case next(Element)
  case failed(Error)
  case completed

Valid signals produce zero or more .next events and always terminate with either a .completed event or a .failed event in case of an error. Each .next event contains an associated element - the actual value or object produced by the signal.

Creating Signals

There are many ways to create signals. Main one is by using the constructor that accepts a producer closure. The closure has one argument - an observer to which you send events. To send next element, use next method of the observer. When there are no more elements to be generated, send completion event using completed method. For example, to create a signal that produces first three positive integers do:

let counter = Signal<Int, NoError> { observer in

  // send first three positive integers

  // complete

  return NonDisposable.instance

Producer closure expects you to return a disposable. More about disposables can be found here.

Notice how we defined signal as Signal<Int, NoError>. First generic argument specifies that the signal emits elements of type Int. Second one specifies the error type that the signal can error-out with. NoError is a type without a constructor so it cannot be initialized. It is used to create signals that cannot error-out, so called non-failable signals. This is so common type so ReactiveKit provides a typealias SafeSignal defined as

public typealias SafeSignal<Element> = Signal<Element, NoError>

That means that instead of Signal<Int, NoError> you can write just SafeSignal<Int>.

The type name SafeSignal might not be the happiest name, but we expect Swift 4 to introduce default generic arguments so we will be able to use just Signal<Int>.

When the producer fails to produce the element, you can signal an error. For example, mapping network request could looks like this:

let getUser = Signal<User, NetworkError> { observer in

  let task = api.getUser { result in
    switch result {
      case .success(let user):
      case .failure(let error):


  return BlockDisposable {

The example also shows to use a disposable. When the signal is disposed, the BlockDisposable will call its closure and cancel the task.

These were examples of how to manually create signals. There are few operators in the framework that you can use to create convenient signals. For example, when you need to convert a sequence to a signal, you will use following constructor:

let counter = SafeSignal.sequence([1, 2, 3])

To create a signal that produces an integer every second, do

let counter = SafeSignal<Int>.interval(1)

For more constructors, refer to the code reference.

Observing Signals

Signal is only useful if it's being observed. To observe signal, use observe method:

counter.observe { event in

That will print following:


Most of the time we are interested only in the elements that the signal produces. Elements are associated with .next events and to observe just them you can do:

counter.observeNext { element in

That will print:


Observing the signal actually starts the production of events. In other words, that producer closure we passed in the constructor is called only when you register an observer. If you register more that one observer, producer closure will be called once for each of them.

Observers will be by default invoked on the thread (queue) on which the producer generates events. You can change that behaviour by passing another execution context using the observeOn method.

Transforming Signals

Signals can be transformed into another signals. Methods that transform signals are often called operators. For example, to convert our signal of positive integers into a signal of positive even integers we can do

let evenCounter = counter.map { $0 * 2 }

or to convert it to a signal of integers divisible by three

let divisibleByThree = counter.filter { $0 % 3 == 0 }

or to convert each element to another signal that just triples that element and merge those new signals by concatenating them one after another

let tripled = counter.flatMapConcat { number in
  return SafeSignal.sequence(Array(count: 3, repeatedValue: number))

and so on... There are many operators available. For more info on them, check out code reference.

Handling Errors

One way to try to recover from an error is to just retry the signal again. To do so, just do

let betterFetch = fetchImage(url: ...).retry(3)

and smile thinking about how many number of lines would that take in the imperative paradigm.

Errors that cannot be handled with retry will happen eventually. Worst way to handle those is to just ignore and log any error that happens:

let image = fetchImage(url: ...).suppressError(logging: true)

Better way is to provide a default value in case of an error:

let image = fetchImage(url: ...).recover(with: Assets.placeholderImage)

Most powerful way is to flatMapError into another signal:

let image = fetchImage(url: ...).flatMapError { error in
  return SafeSignal<UIImage> ...

There is no best way. Errors suck.

Sharing Results

Whenever the observer is registered, the signal producer is executed all over again. To share results of a single execution, use shareReplay method.

let sharedCounter = counter.shareReplay()

Disposing Signals

Signals produce events until they complete with either a .completed or .failed event. When that event happens, a signal is disposed. Disposing means cleaning up and releasing all of the resources the signal might have been using.

However, some signals might never complete while others might complete when we don't care about them anymore. For example, a signal representing network request that is observed by a view controller might complete after the view controller is dismissed. Such signal can be called a dangling signal. Dangling signals can be dangerous and take up resources that could be put to better use.

So, how do we ensure that all signals are eventually disposed?

A way to ensure that is to leverage a disposable that is returned by any observe* method.

let disposable = aSignal.observeNext { ... }

Such disposable can then be used to dispose the observed signal. Just call dispose on it.


From that point on the signal will not send any more events, the underlying task will be cancelled and resources cleaned up.

A general rule is to dispose all observations you make. It's recommended to keep a dispose bag where you can put all of your disposables. The bag will automatically dispose all disposables you put in when it is deallocated.

class X {
  let disposeBag = DisposeBag()

  func y() {
    aSignal.observeNext { _ in
    }.dispose(in: disposeBag)

If you are using Bond framework and your class is a subclass or a descendent of NSObject, Bond provides the bag as an extension property reactive.bag that you can use out of the box.

Another way to ensure signal disposition is by using bindings instead of observations where possible. They handle everything automatically so you don't have to manually dispose signals.

Hot Signals

If you need hot signals, i.e. signals that can generate events regardless of the observers, you can use PublishSubject type:

let numbers = PublishSubject<Int, NoError>()

numbers.observerNext { num in

numbers.next(1) // prints: 1
numbers.next(2) // prints: 2


Property wraps mutable state into an object that enables observation of that state. Whenever the state changes, an observer can be notified. Just like the PublishSubject, it represents a bridge into the imperative paradigm.

To create the property, just initialize it with the initial value.

let name = Property("Jim")

nil is valid value for properties that wrap optional type.

Properties are signals just like signals of Signal type. They can be transformed into another signals, observed and bound in the same manner as signals can be.

For example, you can register an observer with observe or observeNext methods.

name.observeNext { value in
  print("Hi \(value)!")

When you register an observer, it will be immediately invoked with the current value of the property so that snippet will print "Hi Jim!".

To change value of the property afterwards, just set the value property.

name.value = "Jim Kirk" // Prints: Hi Jim Kirk!


ReactiveKit uses simple concept of execution contexts inspired by BrightFutures to handle threading.

When you want to receive events on a specific dispatch queue, just use context extension of dispatch queue type DispatchQueue, for example: DispatchQueue.main.context, and pass it to the observeOn signal operator.


  • iOS 8.0+ / macOS 10.9+ / tvOS 9.0+ / watchOS 2.0+
  • Xcode 8


  • If you'd like to ask a general question, use Stack Overflow.
  • If you'd like to ask a quick question or chat about the project, try Gitter.
  • If you found a bug, open an issue.
  • If you have a feature request, open an issue.
  • If you want to contribute, submit a pull request (include unit tests).
  • You can track project plan and progress on Waffle.

Additional Documentation


Bond is optional, but recommended for Cocoa / Cocoa touch development.


pod 'ReactiveKit', '~> 3.2'
pod 'Bond', '~> 6.0'


github "ReactiveKit/ReactiveKit" ~> 3.2
github "ReactiveKit/Bond" ~> 6.0


Migration from v2.x to v3.x

There are some big changes in v3. Major one is that ReactiveKit is joining forces with Bond to make great family of frameworks for functional reactive programming. Some things have been moved out of ReactiveKit to Bond in order to make ReactiveKit simpler and focused on FRP, while Bond has been reimplemented on top of ReactiveKit in order to provide great extensions like bindings, reactive delegates or observable collections.

What that means for you? Well, nothing has changed conceptually so your migration should come down to renaming. Stream and Operation had to be renamed because of conflicts with types from Foundation framework that's now lost NS prefix. A number of operators has been renamed to match Swift 3 syntax. CollectionProperty and reactive delegates are now part of Bond framework so make sure you import Bond in places where you use those. Binding extensions provided by ReactiveUIKit framework are now provided by Bond. Just import Bond instead of ReactiveUIKit and change extension prefixes from r to bnd.

  • Type Operation has been renamed to Signal.
  • Type Stream is now implemented as a typealieas to non-failable Signal and named Signal1. Just replace all occurrences of Stream with Signal1 in your project.
  • Operator flatMap(_ strategy:) has been replaced with flatMapLatest, flatMapMerge and flatMapConcat operators.
  • Operator toSignal that returned stream of elements and stream of errors has been renamed to branchOutError().
  • Operator toSignal(justLogError:) has been renamed to suppressError(logging:)
  • Operators like takeLast, skipLast, feedNextInto, bindTo have been renamed to take(last:), skip(last:), feedNext(into:), bind(to:) etc.
  • Each of the operators combineLatest, zip, merge and amb now has overloads for 6 arguments.
  • PushStream and PushOperation have been replaced by SafePublishSubject and PublishSubject.
  • Queue has been removed. Use DispatchQueue.
  • CollectionProperty has been moved to Bond framework and implemented as three types: ObservableArray, ObservableDictionary and ObservableSet.
  • ProtocolProxy and other Foundation extensions have been moved to Bond framework. Prefix of extensions has been changed to reactive. For example rBag is renamed to reactive.bag.


The MIT License (MIT)

Copyright (c) 2015-2017 Srdan Rasic (@srdanrasic)

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.