Query:
I came across PECS (short for Producer extends and Consumer super) while reading up on generics.
Can someone explain to me how to use PECS to resolve confusion between extends and super?
What is PECS? Answer #1:
“PECS” is from the collection’s point of view. If you are only pulling items from a generic collection, it is a producer and you should use extends; if you are only stuffing items in, it is a consumer and you should use super. If you do both with the same collection, you shouldn’t use either extends or super.
Suppose you have a method that takes as its parameter a collection of things, but you want it to be more flexible than just accepting a Collection<Thing>.
Case 1: You want to go through the collection and do things with each item.
Then the list is a producer, so you should use a Collection<? extends Thing>.
The reasoning is that a Collection<? extends Thing> could hold any subtype of Thing, and thus each element will behave as a Thing when you perform your operation. (You actually cannot add anything (except null) to a Collection<? extends Thing>, because you cannot know at runtime which specific subtype of Thing the collection holds.)
Case 2: You want to add things to the collection.
Then the list is a consumer, so you should use a Collection<? super Thing>.
The reasoning here is that unlike Collection<? extends Thing>, Collection<? super Thing> can always hold a Thing no matter what the actual parameterized type is. Here you don’t care what is already in the list as long as it will allow a Thing to be added; this is what ? super Thing guarantees.
Answer #2:
The principles behind this in computer science is called
- Covariance:
? extends MyClass, - Contravariance:
? super MyClassand - Invariance/non-variance:
MyClass
The picture below should explain the concept. Picture courtesy: Andrey Tyukin
Answer #3:
PECS (Producer extends and Consumer super)
mnemonic → Get (extend) and Put (Super) principle.
- This principle states that:
- Use an
extendswildcard when you only get values out of a structure. - Use a
superwildcard when you only put values into a structure. - And don’t use a wildcard when you both get and put.
- Use an
Example in Java:
class Super {
Number testCoVariance() {
return null;
}
void testContraVariance(Number parameter) {
}
}
class Sub extends Super {
@Override
Integer testCoVariance() {
return null;
} //compiles successfully i.e. return type is don't care(Integer is subtype of Number)
@Override
void testContraVariance(Integer parameter) {
} //doesn't support even though Integer is subtype of Number
}
The Liskov Substitution Principle (LSP) states that “objects in a program should be replaceable with instances of their subtypes without altering the correctness of that program”.
Within the type system of a programming language, a typing rule
- covariant if it preserves the ordering of types (≤), which orders types from more specific to more generic;
- contravariant if it reverses this ordering;
- invariant or nonvariant if neither of these applies.
- Read-only data types (sources) can be covariant;
- write-only data types (sinks) can be contravariant.
- Mutable data types which act as both sources and sinks should be invariant.
To illustrate this general phenomenon, consider the array type. For the type Animal we can make the type Animal[]
- covariant: a Cat[] is an Animal[];
- contravariant: an Animal[] is a Cat[];
- invariant: an Animal[] is not a Cat[] and a Cat[] is not an Animal[].
Java Examples:
Object name= new String("prem"); //works
List<Number> numbers = new ArrayList<Integer>();//gets compile time error
Integer[] myInts = {1,2,3,4};
Number[] myNumber = myInts;
myNumber[0] = 3.14; //attempt of heap pollution i.e. at runtime gets java.lang.ArrayStoreException: java.lang.Double(we can fool compiler but not run-time)
List<String> list=new ArrayList<>();
list.add("prem");
List<Object> listObject=list; //Type mismatch: cannot convert from List<String> to List<Object> at Compiletime
bounded(i.e. heading toward somewhere) wildcard : There are 3 different flavours of wildcards:
- In-variance/Non-variance:
?or? extends Object– Unbounded Wildcard. It stands for the family of all types. Use when you both get and put. - Co-variance:
? extends T( Reign ofTdescendants) – a wildcard with an upper bound.Tis the upper-most class in the inheritance hierarchy. Use anextendswildcard when you only Get values out of a structure. - Contra-variance:
? super T( Reign ofTancestor) – a wildcard with a lower bound.Tis the lower-most class in the inheritance hierarchy. Use asuperwildcard when you only Put values into a structure.
Note: wildcard ? means zero or one time, represents an unknown type. The wildcard can be used as the type of a parameter, never used as a type argument for a generic method invocation, a generic class instance creation.(i.e. when used wildcard that reference not used in elsewhere in program like we use T)
import java.util.ArrayList;
import java.util.List;
class Shape { void draw() {}}
class Circle extends Shape {void draw() {}}
class Square extends Shape {void draw() {}}
class Rectangle extends Shape {void draw() {}}
public class Test {
public static void main(String[] args) {
//? extends Shape i.e. can use any sub type of Shape, here Shape is Upper Bound in inheritance hierarchy
List<? extends Shape> intList5 = new ArrayList<Shape>();
List<? extends Shape> intList6 = new ArrayList<Cricle>();
List<? extends Shape> intList7 = new ArrayList<Rectangle>();
List<? extends Shape> intList9 = new ArrayList<Object>();//ERROR.
//? super Shape i.e. can use any super type of Shape, here Shape is Lower Bound in inheritance hierarchy
List<? super Shape> inList5 = new ArrayList<Shape>();
List<? super Shape> inList6 = new ArrayList<Object>();
List<? super Shape> inList7 = new ArrayList<Circle>(); //ERROR.
//-----------------------------------------------------------
Circle circle = new Circle();
Shape shape = circle; // OK. Circle IS-A Shape
List<Circle> circles = new ArrayList<>();
List<Shape> shapes = circles; // ERROR. List<Circle> is not subtype of List<Shape> even when Circle IS-A Shape
List<? extends Circle> circles2 = new ArrayList<>();
List<? extends Shape> shapes2 = circles2; // OK. List<? extends Circle> is subtype of List<? extends Shape>
//-----------------------------------------------------------
Shape shape2 = new Shape();
Circle circle2= (Circle) shape2; // OK. with type casting
List<Shape> shapes3 = new ArrayList<>();
List<Circle> circles3 = shapes3; //ERROR. List<Circle> is not subtype of List<Shape> even Circle is subetype of Shape
List<? super Shape> shapes4 = new ArrayList<>();
List<? super Circle> circles4 = shapes4; //OK.
}
/*
* Example for an upper bound wildcard (Get values i.e Producer `extends`)
*
* */
public void testCoVariance(List<? extends Shape> list) {
list.add(new Object());//ERROR
list.add(new Shape()); //ERROR
list.add(new Circle()); // ERROR
list.add(new Square()); // ERROR
list.add(new Rectangle()); // ERROR
Shape shape= list.get(0);//OK so list act as produces only
/*
* You can't add a Shape,Circle,Square,Rectangle to a List<? extends Shape>
* You can get an object and know that it will be an Shape
*/
}
/*
* Example for a lower bound wildcard (Put values i.e Consumer`super`)
* */
public void testContraVariance(List<? super Shape> list) {
list.add(new Object());//ERROR
list.add(new Shape());//OK
list.add(new Circle());//OK
list.add(new Square());//OK
list.add(new Rectangle());//OK
Shape shape= list.get(0); // ERROR. Type mismatch, so list acts only as consumer
Object object= list.get(0); //OK gets an object, but we don't know what kind of Object it is.
/*
* You can add a Shape,Circle,Square,Rectangle to a List<? super Shape>
* You can't get an Shape(but can get Object) and don't know what kind of Shape it is.
*/
}
}
Covariance and contravariance determine compatibility based on types. In either case, variance is a directed relation. Covariance can be translated as “different in the same direction,” or with-different, whereas contravariance means “different in the opposite direction,” or against-different. Covariant and contravariant types are not the same, but there is a correlation between them. The names imply the direction of the correlation.
Answer #4:
Let’s assume this hierarchy:
class Creature{}// X
class Animal extends Creature{}// Y
class Fish extends Animal{}// Z
class Shark extends Fish{}// A
class HammerSkark extends Shark{}// B
class DeadHammerShark extends HammerSkark{}// C
Let’s clarify PE – Producer Extends:
List<? extends Shark> sharks = new ArrayList<>();
Why you cannot add objects that extend “Shark” in this list? like:
sharks.add(new HammerShark());//will result in compilation error
Since you have a list that can be of type A, B or C at runtime, you cannot add any object of type A, B or C in it because you can end up with a combination that is not allowed in java.
In practice, the compiler can indeed see at compiletime that you add a B:
sharks.add(new HammerShark());
…but it has no way to tell if at runtime, your B will be a subtype or supertype of the list type. At runtime the list type can be any of the types A, B, C. So you cannot end up adding HammerSkark (super type) in a list of DeadHammerShark for example.
*You will say: “OK, but why can’t I add HammerSkark in it since it is the smallest type?”. Answer: It is the smallest you know. But HammerSkark can be extended too by somebody else and you end up in the same scenario.
Let’s clarify CS – Consumer Super:
In the same hierarchy we can try this:
List<? super Shark> sharks = new ArrayList<>();
What and why you can add to this list?
sharks.add(new Shark());
sharks.add(new DeadHammerShark());
sharks.add(new HammerSkark());
You can add the above types of objects because anything below shark(A,B,C) will always be subtypes of anything above shark (X,Y,Z). Easy to understand.
You cannot add types above Shark, because at runtime the type of added object can be higher in hierarchy than the declared type of the list(X,Y,Z). This is not allowed.
But why you cannot read from this list? (I mean you can get an element out of it, but you cannot assign it to anything other than Object o):
Object o;
o = sharks.get(2);// only assignment that works
Animal s;
s = sharks.get(2);//doen't work
At runtime, the type of list can be any type above A: X, Y, Z, … The compiler can compile your assignment statement (which seems correct) but, at runtime the type of s (Animal) can be lower in hierarchy than the declared type of the list(which could be Creature, or higher). This is not allowed.
To sum up
We use <? super T> to add objects of types equal or below T to the List. We cannot read from it.
We use <? extends T> to read objects of types equal or below T from list. We cannot add element to it.
Answer #5:
This is the clearest, simplest way for me to think of extends vs. super:
extendsis for readingsuperis for writing
I find “PECS” to be a non-obvious way to think of things regarding who is the “producer” and who is the “consumer”. “PECS” is defined from the perspective of the data collection itself – the collection “consumes” if objects are being written to it (it is consuming objects from calling code), and it “produces” if objects are being read from it (it is producing objects to some calling code). This is counter to how everything else is named though. Standard Java APIs are named from the perspective of the calling code, not the collection itself. For example, a collection-centric view of java.util.List should have a method named “receive()” instead of “add()” – after all, the calling code adds the element, but the list itself receives the element.
I think it’s more intuitive, natural and consistent to think of things from the perspective of the code that interacts with the collection – does the code “read from” or “write to” the collection? Following that, any code writing to the collection would be the “producer”, and any code reading from the collection would be the “consumer”.
Hope you learned something from this post.
Follow Programming Articles for more!