An
Introduction to Reactive
Programming
Prof. Guido Salvaneschi
Guido Salvaneschi: introduction to reactive programming
Outline
• Intro to reactive applications
• The Observer pattern
• Event-based languages
• Reactive languages
Guido Salvaneschi: introduction to reactive programming
INTRO TO REACTIVE APPLICATIONS
Guido Salvaneschi: introduction to reactive programming
Software Taxonomy
• A transformational system
– Accepts input, performs computation on it, produces
output, and terminates
– Compilers, shell tools,
scientific computations
• A reactive system
– Continuously interacts with the environment
– Updates its state
Guido Salvaneschi: introduction to reactive programming
Use of State
• Transformational systems:
– Express transformations as incremental modifications of
the internal data structures
State is not necessary to describe the system
• Reactive systems:
– Represent the current state of interaction
– Reflect changes of the external world during interaction
State is essential to describe the system
Guido Salvaneschi: introduction to reactive programming
Reactive Applications
Interactive Applications Monitoring / Control
UI Systems
Guido Salvaneschi: introduction to reactive programming
Reactive Applications
• Many other examples
– Web applications
– Mobile apps
– Distributed computations
• Cloud
– …
• Typical operations
– Detect events/notifications and react
– Combine reactions
– Propagate updates/changes
Guido Salvaneschi: introduction to reactive programming
Reactive Applications
Why should we care?
• Event handling:
– 30% of code in desktop applications
– 50% of bugs reported during production cycle
Guido Salvaneschi: introduction to reactive programming
Reactive Programming
Now…
– Reactive applications are extremely common
– Can we design new language features
to specifically address this issue ?
• Think about the problems solved by exceptions,
visibility modifiers, inheritance, …
Guido Salvaneschi: introduction to reactive programming
REACTIVE PROGRAMMING
Guido Salvaneschi: introduction to reactive programming
Reactive Programming
Definition… ?
“Programming language abstractions
(techniques and patterns)
to develop reactive applications”
For example, abstractions to:
Represent event streams
Automatically propagate changes in the state
Combine events
…
Guido Salvaneschi: introduction to reactive programming
Reactive Programming
• Haskell: Fran, Yampa
• FrTime, Flapjax, REScala, Scala.react, …
• Angular.js, Bacon.js, Reactive.js, …
• Microsoft Reactive Extensions (Rx)
• Books 2014-16
Guido Salvaneschi: introduction to reactive programming
Reactive Programming
Guido Salvaneschi: introduction to reactive programming
Reactive Programming
Guido Salvaneschi: introduction to reactive programming
THE OBSERVER PATTERN
Guido Salvaneschi: introduction to reactive programming
The (good? old) Observer Pattern
Guido Salvaneschi: introduction to reactive programming
The (good? old) Observer Pattern
boolean highTemp; State
boolean smoke;
void Init() {
tempSensor.register(this);
smokeSensor.register(this); Registration
}
void notifyTempReading(TempEvent e ) { Callback
highTemp = e.getValue() > 45; functions
if (highTemp && smoke) {
alert.start(); Control
} statements
}
void notifySmokeReading(SmokeEvent e) { Callback
smoke = e.getIntensity() > 0.5; functions
if (highTemp && smoke) {
alert.start(); Control
} statements
}
Guido Salvaneschi: introduction to reactive programming
The Observer Pattern
• What about Java Swing ?
– javax.swing
BEEEEP!
Guido Salvaneschi: introduction to reactive programming
public class Beeper extends JPanel implements ActionListener {
JButton button;
public Beeper() {
super(new BorderLayout()); BEEEEP!
button = new JButton("Click Me");
button.setPreferredSize(new Dimension(200, 80));
add(button, BorderLayout.CENTER);
button.addActionListener(this);
}
public void actionPerformed(ActionEvent e) {
Toolkit.getDefaultToolkit().beep();
}
private static void createAndShowGUI() { // Create the GUI and show it.
JFrame frame = new JFrame("Beeper"); //Create and set up the window.
frame.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);
JComponent newContentPane = new Beeper(); //Create and set up the content pane.
newContentPane.setOpaque(true);
frame.setContentPane(newContentPane);
frame.pack(); //Display the window.
frame.setVisible(true);
}
public static void main(String[] args) {
javax.swing.SwingUtilities.invokeLater( new Runnable() { public void run() {createAndShowGUI();}});
}
}
Guido Salvaneschi: introduction to reactive programming
EVENT-BASED LANGUAGES
Guido Salvaneschi: introduction to reactive programming
Event-based Languages
Language-level support for events
• Events as object attributes
– Describe changes of the object's state
– Part of the interface
• Event-based languages are better!
– More concise, clear programming intention, …
– C#, Ptolemy, EScala, EventJava, …
Guido Salvaneschi: introduction to reactive programming
Example in C#
public class Drawing {
Collection<Figure> figures;
public event NoArgs Changed();
public virtual void Add(Figure figure) {
figures.Add(figure);
figure.Changed += OnChanged;
OnChanged();
}
public virtual void Remove(Figure figure) {
figures.Remove(figure);
figure.Changed -= OnChanged;
OnChanged();
}
protected virtual void OnChanged() {
if (Changed != null) { Changed(); }
}
...
}
Guido Salvaneschi: introduction to reactive programming
EVENTS IN SCALA
Guido Salvaneschi: introduction to reactive programming
REScala
• www.rescala-lang.com
– An advanced event-based system
– Abstractions for time-changing values
– Bridging between them
• Philosophy: foster a more declarative and functional
style without sacrificing the power of OO design
• Pure Scala
Guido Salvaneschi: introduction to reactive programming
Adding Events to Scala
• C# events are recognized by the compiler
– Scala does not support events by itself, but…
• Can we introduce events using the powerful Scala
support for DSLs?
• Can we do even better than C# ?
– E.g., event composition ?
Guido Salvaneschi: introduction to reactive programming
REScala events: Summary
• Different types of events: Imperative, declarative, …
• Events carry a value
– Bound to the event when the event is fired
– Received by all the handlers
• Events are parametric types.
– Event[T], Evt[T]
• All events are subtype of Event[T]
Guido Salvaneschi: introduction to reactive programming
Imperative Events
• Valid event declarations
val e1 = Evt[Unit]()
val e2 = Evt[Int]()
val e3 = Evt[String]()
val e4 = Evt[Boolean]()
val e5: Event[Int] = Evt[Int]()
class Foo
val e6 = Evt[Foo]()
Guido Salvaneschi: introduction to reactive programming
Imperative Events
• Multiple values for the same event
are expressed using tuples
val e1 = Evt[(Int,Int)]()
val e2 = Evt[(String,String)]()
val e3 = Evt[(String,Int)]()
val e4 = Evt[(Boolean,String,Int)]()
val e5: Evt[(Int,Int)] = Evt[(Int,Int)]()
Guido Salvaneschi: introduction to reactive programming
Handlers
• Handlers are executed when the event is fired
– The += operator registers the handler.
• The handler is a first class function
– The attached value is the function parameter.
var state = 0
val e = Evt[Int]()
e += { println(_) }
e += (x => println(x))
e += ((x: Int) => println(x))
e += (x => { // Multiple statements in the handler
state = x
println(x)
})
Guido Salvaneschi: introduction to reactive programming
Handlers
• The signature of the handler must conform the event
– E.g., Event[(Int,Int)] requires (Int,Int) =>Unit
– The handler:
• receives the attached value
• performs side effects.
val e = Evt[(Int,String)]()
e += (x => {
println(x._1)
println(x._2)
})
e += (x: (Int,String) => {
println(x)
})
Guido Salvaneschi: introduction to reactive programming
Handlers
• Events without arguments still need a
Unit argument in the handler.
val e = Evt[Unit]()
e += { x => println(“Fired!”) }
e += { (x: Unit) => println(“Fired!”) }
Guido Salvaneschi: introduction to reactive programming
Methods as Handlers
• Methods can be used as handlers.
– Partially applied functions syntax
– Types must be correct
def m1(x: Int) = {
val y = x + 1
println(y)
}
val e = Evt[Int]
e += m1 _
e(10)
Guido Salvaneschi: introduction to reactive programming
Firing Events
• Method call syntax
• The value is bound to the event occurrence
val e1 = Evt[Int]()
val e2 = Evt[Boolean]()
val e3 = Evt[(Int,String)]()
e1(10)
e2(false)
e3((10,"Hallo"))
Guido Salvaneschi: introduction to reactive programming
Firing Events
• Registered handlers are executed every time the
event is fired.
– The actual parameter is provided to the handler
val e = Evt[Int]()
e += { x => println(x) }
e(10)
e(11)
-- output ----
10
11
Guido Salvaneschi: introduction to reactive programming
Firing Events
• All registered handlers are executed
– The execution order is non deterministic
val e = Evt[Int]()
e += { x => println(x) }
e += { x => println("n: " + x)}
e(10)
e(11)
-- output ----
10
n: 10
11
n: 11
Guido Salvaneschi: introduction to reactive programming
Firing Events
• The .remove operator val e = Evt[Int]()
val handler1 = { x: Int => println(x) }
unregisters a handler val handler2 = { x: Int => println("n: " + x) }
via its handle val h1 = e += handler1
• The += operator also val h2 = e += handler2
e(10)
returns the handle h1.remove
e(10)
that will be used for h2.remove
unregistration e(10)
-- output ----
10
n: 10
n: 10
Guido Salvaneschi: introduction to reactive programming
Imperative Events
• Events can be referred to generically
val e1: Event[Int] = Evt[Int]()
Guido Salvaneschi: introduction to reactive programming
DECLARATIVE EVENTS
Guido Salvaneschi: introduction to reactive programming
The Problem
• Imperative events are fired by the programmer
• Conceptually, certain events depend on other events
• Examples:
– mouseClickE -> museClickOnShape
– mouseClose, keyboardClose -> closeWindow
• Can we solve this problem enhancing the language?
Guido Salvaneschi: introduction to reactive programming
Declarative Events
• Declarative events are defined by a combination of
other events.
• Some valid declarations:
val e1 = Evt[Int]()
val e2 = Evt[Int]()
val e3 = e1 || e2
val e4 = e1 && ((x: Int)=> x>10)
val e5 = e1 map ((x: Int)=> x.toString)
Guido Salvaneschi: introduction to reactive programming
OR events
• The event e1 || e2 is fired upon the occurrence of
one among e1 or e2.
– The events in the event expression have the same
parameter type
val e1 = Evt[Int]()
val e2 = Evt[Int]()
val e1_OR_e2 = e1 || e2
e1_OR_e2 += ((x: Int) => println(x))
e1(10)
e2(10)
-- output ----
10
10
Guido Salvaneschi: introduction to reactive programming
Predicate Events
• The event e && p is fired if e occurs and the
predicate p is satisfied.
– The predicate is a function that accepts the event
parameter as a formal and returns Boolean.
– && filters events using a parameter and a predicate.
val e = Evt[Int]()
val e_AND: Event[Int] = e && ((x: Int) => x>10)
e_AND += ((x: Int) => println(x))
e(5)
e(15)
-- output ----
15
Guido Salvaneschi: introduction to reactive programming
Map Events
• The event e map f is obtained by applying f to the
value carried by e.
– The map function takes the event parameter as a formal.
– The return type of map is the type parameter of the
resulting event.
val e = Evt[Int]()
val e_MAP: Event[String] = e map ((x: Int) => x.toString)
e_MAP += ((x: String) => println("Here: " + x))
e(5)
e(15)
-- output ----
Here: 5
Here: 15
Guido Salvaneschi: introduction to reactive programming
EXAMPLES OF RESCALA EVENTS
Guido Salvaneschi: introduction to reactive programming
Example: Figures
abstract class Figure {
val moved[Unit] = afterExecMoveBy
val resized[Unit]
val changed[Unit] = resized || moved || afterExecSetColor
val invalidated[Rectangle] = changed.map( _ => getBounds() )
...
val afterExecMoveBy = new Evt[Unit]
val afterExecSetColor = new Evt[Unit]
…
def moveBy(dx: Int, dy: Int) { position.move(dx, dy); afterExecMoveBy() }
def resize(s: Size) { size = s }
def setColor(col: Color) { color = col; afterExecSetColor() }
def getBounds(): Rectangle
...
}
Guido Salvaneschi: introduction to reactive programming
Example: Figures
class Connector(val start: Figure, val end: Figure) {
val h1 = start.changed += updateStart _
val h2 = end.changed += updateEnd _
...
def updateStart() { ... }
def updateEnd() { ... }
...
def dispose {
h1.remove
h2.remove
}
}
Guido Salvaneschi: introduction to reactive programming
Example: Figures
• Inherited events
abstract class Figure {
– May be overridden val moved[Unit] = afterExecMoveBy
val resized[Unit]
– Are late bound …
}
class RectangleFigure extends Figure {
val resized = afterExecResize || afterExecSetBounds
override val moved = super.moved || afterExecSetBounds
...
val afterExecResize = new Evt[Unit]
val afterExecSetBounds = new Evt[Unit]
…
def resize(s: Size) { ... ; afterExecResize() }
def setBounds(x1: Int, y1: Int, x2: Int, y2: Int) { ... ; afterExecSetBounds }
...
}
Guido Salvaneschi: introduction to reactive programming
Example: Temperature Sensor
class TemperatureSensor {
val tempChanged[Int] = new Evt[Int]
...
def run {
var currentTemp = measureTemp()
while(!stop) {
val newTemp = measureTemp()
if (newTemp != currentTemp) {
tempChanged(newTemp)
currentTemp = newTemp
}
sleep(100)
}
}
}
Guido Salvaneschi: introduction to reactive programming
REACTIVE LANGUAGES
Guido Salvaneschi: introduction to reactive programming
Events and Functional Dependencies
Events are often used for
val update = Evt[Unit]()
functional dependencies var a = 3
boolean highTemp := (temp.value > 45); var b = 7
var c = a + b // Functional dependency
update += ( _ =>{
var a = 3 c = a + b
var b = 7 })
val c = a + b
a = 4
a = 4 update()
b = 8 b = 8
update()
Guido Salvaneschi: introduction to reactive programming
Constraints
• What about expressing functional dependencies
as constraints ?
val a = 3 val a = 3
val b = 7 val b = 7
val c = a + b // Statement val c := a + b // Constraint
println(c) println(c)
> 10 > 10
a= 4 a= 4
println(c) println(c)
> 10 > 11
Guido Salvaneschi: introduction to reactive programming
EMBEDDING REACTIVE
PROGRAMMING IN SCALA
Guido Salvaneschi: introduction to reactive programming
Reactive Values
• Vars: primitive reactive values val a = Var(3)
– Updated “manually” val b = Var(7)
val c = Signal{ a() + b() }
println(c.now)
• Signals: reactive expressions > 10
– The constraints “automatically” a()= 4
enforced println(c.now)
> 11
Guido Salvaneschi: introduction to reactive programming
Reference Model
• Change propagation model i
– Dependency graph
f g h
– Push-driven evaluation
d e
…
val a = Var(3) d
val b = Var(7)
val c = Signal{ a() + b() }
val d = Signal { 2 * c() } c
…
a b
Guido Salvaneschi: introduction to reactive programming
SIGNALS AND VARS
Guido Salvaneschi: introduction to reactive programming
Vars
• Vars wrap normal Scala values
val a = Var(0)
• Var[T] is a parametric type. val b = Var("Hello World")
val c = Var(false)
val d: Var[Int] = Var(30)
val e: Var[String] = Var("REScala")
– The parameter T is the type val f: Var[Boolean] = Var(false)
the var wraps around
– Vars are assigned by the
a()= 3
“()=“ operator b()=“New World”
c()=true
Guido Salvaneschi: introduction to reactive programming
Signals
• Syntax: Signal{sigexpr}
– Sigexpr should be side-effect free
• Signals are parametric types.
– A signal Signal[T] carries a value of type T
Guido Salvaneschi: introduction to reactive programming
Signals: Collecting Dependencies
• A Var or a Signal called with () in a signal expression
is added to the dependencies of the defined signal
// Multiple vars
// in a signal expression c
val a = Var(0)
val b = Var(0)
val s = Signal{ a() + b() } a b
Guido Salvaneschi: introduction to reactive programming
Signals: Examples
val a = Var(0)
val b = Var(0)
val c = Var(0)
val r: Signal[Int] = Signal{ a() + 1 } // Explicit type in var decl
val s = Signal{ a() + b() } // Multiple vars is a signal expression
val t = Signal{ s() * c() + 10 } // Mix signals and vars in signal expressions
val u = Signal{ s() * t() } // A signal that depends on other signals
Guido Salvaneschi: introduction to reactive programming
Signals: Examples
val a = Var(0)
val b = Var(2)
val c = Var(true)
val s = Signal{ if (c()) a() else b() }
def factorial(n: Int) = ...
val a = Var(0)
val s: Signal[Int] = Signal{ // A signal expression can be any code block
val tmp = a() * 2
val k = factorial(tmp)
k + 2 // Returns an Int
}
Guido Salvaneschi: introduction to reactive programming
Signals
• Accessing reactive values: now
– Often used to return to a traditional computation
val a = Var(0)
val b = Var(2)
val c = Var(true)
val s: Signal[Int] = Signal{ a() + b() }
val t: Signal[Boolean] = Signal{ !c() }
val x: Int = a.now
val y: Int = s.now
val z: Boolean = t.now
println(z)
Guido Salvaneschi: introduction to reactive programming
EXAMPLES OF SIGNALS
Guido Salvaneschi: introduction to reactive programming
Example
Guido Salvaneschi: introduction to reactive programming
Example: Observer
/* Create the graphics */
title = "Reactive Swing App"
val button = new Button { /* The logic */
text = "Click me!" listenTo(button)
} var nClicks = 0
val label = new Label { reactions += {
text = "No button clicks registered" case ButtonClicked(b) =>
} nClicks += 1
contents = new BoxPanel(Orientation.Vertical) { label.text = "Number of button clicks: " + nClicks
contents += button if (nClicks > 0)
contents += label button.text = "Click me again"
} }
Guido Salvaneschi: introduction to reactive programming
Example: Signals
title = "Reactive Swing App"
val label = new ReactiveLabel
val button = new ReactiveButton
val nClicks = button.clicked.fold(0) {(x, _) => x + 1}
label.text = Signal { ( if (nClicks() == 0) "No" else nClicks() ) + " button clicks registered" }
button.text = Signal { "Click me" + (if (nClicks() == 0) "!" else " again " )}
contents = new BoxPanel(Orientation.Vertical) {
contents += button
contents += label
}
Guido Salvaneschi: introduction to reactive programming
Example: Smashing Particles
class Oval(center: Signal[Point], radius: Signal[Int]) { … }
val base = Var(0) // Increases indefinitely
val linearTime = base()
val cyclicTime = Signal{linearTime() % 200}
val point1 = Signal{ new Point(20+ cyclicTime (), 20+ cyclicTime ()) }
new Oval(point1, cyclicTime )
… // 4 times
Guido Salvaneschi: introduction to reactive programming 66
BASIC CONVERSION FUNCTIONS
Guido Salvaneschi: introduction to reactive programming
REScala design principles
• Signals (and events) are objects fields
– Inheritance, late binding, visibility modifiers, …
• Conversion functions
bridge signals and events SIGNALS
EVENTS
Guido Salvaneschi: introduction to reactive programming
Basic Conversion Functions
• Changed :: Signal[T] -> Event[T]
• Latest :: Event[T] -> Signal[T]
Guido Salvaneschi: introduction to reactive programming
Example: Changed
val SPEED = 10
val time = Var(0)
val space = Signal{ SPEED * time() }
while (true) {
Thread sleep 20
time() = time.now + 1
}
space.changed += ((x: Int) => println(x))
-- output --
10
20
30
40
...
Guido Salvaneschi: introduction to reactive programming
Example: Latest
val senseTmp = Evt[Int]() // Fahrenheit
val threshold = 40
val fahrenheitTmp = senseTmp.latest(0)
val celsiusTmp = Signal{ (fahrenheitTmp() – 32) / 1.8 }
val alert = Signal{ if (celsiusTmp() > threshold ) “Warning ” else “OK” }
Guido Salvaneschi: introduction to reactive programming
Quiz 1
val v1 = Var(4)
val v2 = Var(2)
val s1 = Signal{ v1() + v2() }
val s2 = Signal{ s1() / 3 }
assert(s2.now ==
assert(s2.now == 2)
v1()=1
assert(s2.now ==
assert(s2.now == 1)
Guido Salvaneschi: introduction to reactive programming
Quiz 2
var test = 0
val v1 = Var(4)
val v2 = Var(2)
val s1 = Signal{ v1() + v2() }
s1.changed += ((x: Int)=>{test+=1})
assert(test ==
assert(test == 0)
v1()=1
assert(test == 1)
assert(test ==
Guido Salvaneschi: introduction to reactive programming
Quiz 3
val e = Evt[Int]()
val v1 = Var(4)
val v2 = Var(2)
val s1 = e.latest(0)
val s2 = Signal{ v1() + v2() + s1() }
assert(s2.now ==
assert(s2.now == 6)
e(2)
assert(s2.now == 8)
assert(s2.now ==
e(1)
assert(s2.now ==
assert(s2.now == 7)
Guido Salvaneschi: introduction to reactive programming
TRUBLESHOOTING
Guido Salvaneschi: introduction to reactive programming
Common pitfalls
• Establishing dependencies c
– () creates a dependency. a b
Use only in signal expressions
– now returns the current value val a = Var(2)
val b = Var(3)
val c = Signal{ a.now + b() }
• Signals are not assignable.
– Depend on other signals and vars
– Are automatically updated
Guido Salvaneschi: introduction to reactive programming
Common pitfalls
• Avoid side effects in signal expressions
var c = 0 val c = Signal{
val s = Signal{ val sum = a() + b();
val sum = a() + b(); sum * 2
c = sum * 2 }
} ...
... foo(c.now)
foo(c)
• Avoid cyclic dependencies
val a = Var(0)
val s = Signal{ a() + t() }
val t = Signal{ a() + s() + 1 }
Guido Salvaneschi: introduction to reactive programming
Reactive Abstractions and Mutability
• Signals and vars hold references to objects, not the
objects themselves.
class Foo(init: Int){ class Foo(init: Int){ class Foo(x: Int) //Immutable
var x = init var x = init val foo = new Foo(1)
} }
val foo = new Foo(1) val foo = new Foo(1) val varFoo = Var(foo)
val s = Signal{
val varFoo = Var(foo) val varFoo = Var(foo) varFoo().x + 10
val s = Signal{ val s = Signal{ }
varFoo().x + 10 varFoo().x + 10 assert(s.now == 11)
} } varFoo()= new Foo(2)
assert(s. now== 11) assert(s.now == 11) assert(s.now == 12)
foo.x = 2 foo.x = 2
assert(s.now == 11) varFoo()=foo
assert(s.now == 11)
Guido Salvaneschi: introduction to reactive programming
QUESTIONS?
Guido Salvaneschi: introduction to reactive programming
