Why do we use autoboxing and unboxing in Java?

Question

Some context is required to fully understand the main reason behind this.

Primitives versus classes

Primitive variables in Java contain values (an integer, a double-precision floating point binary number, etc). Because these values may have different lengths, the variables containing them may also have different lengths (consider float versus double).

On the other hand, class variables contain references to instances. References are typically implemented as pointers (or something very similar to pointers) in many languages. These things typically have the same size, regardless of the sizes of the instances they refer to (Object, String, Integer, etc).

This property of class variables makes the references they contain interchangeable (to an extent). This allows us to do what we call substitution: broadly speaking, to use an instance of a particular type as an instance of another, related type (use a String as an Object, for example).

Primitive variables aren't interchangeable in the same way, neither with each other, nor with Object. The most obvious reason for this (but not the only reason) is their size difference. This makes primitive types inconvenient in this respect, but we still need them in the language (for reasons that mainly boil down to performance).

Generics and type erasure

Generic types are types with one or more type parameters (the exact number is called generic arity). For example, the generic type definition List<T> has a type parameter T, which can be Object (producing a concrete type List<Object>), String (List<String>), Integer (List<Integer>) and so on.

Generic types are a lot more complicated than non-generic ones. When they were introduced to Java (after its initial release), in order to avoid making radical changes to the JVM and possibly breaking compatibility with older binaries, the creators of Java decided to implement generic types in the least invasive way: all concrete types of List<T> are, in fact, compiled to (the binary equivalent of) List<Object> (for other types, the bound may be something other than Object, but you get the point). Generic arity and type parameter information are lost in this process, which is why we call it type erasure.

Putting the two together

Now the problem is the combination of the above realities: if List<T> becomes List<Object> in all cases, then T must always be a type that can be directly assigned to Object. Anything else can't be allowed. Since, as we said before, int, float and double aren't interchangeable with Object, there can't be a List<int>, List<float> or List<double> (unless a significantly more complicated implementation of generics existed in the JVM).

But Java offers types like Integer, Float and Double which wrap these primitives in class instances, making them effectively substitutable as Object, thus allowing generic types to indirectly work with the primitives as well (because you can have List<Integer>, List<Float>, List<Double> and so on).

The process of creating an Integer from an int, a Float from a float and so on, is called boxing. The reverse is called unboxing. Because having to box primitives every time you want to use them as Object is inconvenient, there are cases where the language does this automatically - that's called autoboxing.

Auto Boxing is used to convert primitive data types to their wrapper class objects. Wrapper class provide a wide range of function to be performed on the primitive types. The most common example is :

int a = 56;
Integer i = a; // Auto Boxing

It is needed because of programmers easy to be able to directly write code and JVM will take care of the Boxing and Unboxing.

Auto Boxing also comes in handy when we are working with java.util.Collection types. When we want to create a Collection of primitive types we cannot directly create a Collection of a primitive type , we can create Collection only of Objects. For Example :

ArrayList<int> al = new ArrayList<int>(); // not supported 

ArrayList<Integer> al = new ArrayList<Integer>(); // supported 
al.add(45); //auto Boxing 

Wrapper Classes

Each of Java's 8 primitive type (byte,short,int,float,char,double,boolean,long) hava a seperate Wrapper class Associated with them. These Wrapper class have predefined methods for preforming useful operations on primitive data types.

Use of Wrapper Classes

String s = "45";
int a = Integer.parseInt(s); // sets the value of a to 45.

There are many useful functions that Wrapper classes provide. Check out the java docs here

Unboxing is opposite of Auto Boxing where we convert the wrapper class object back to its primitive type. This is done automatically by JVM so that we can use a the wrapper classes for certain operation and then convert them back to primitive types as primitives result int faster processing. For Example :

Integer s = 45;
int a = s; auto UnBoxing;

In case of Collections which work with objects only auto unboxing is used. Here's how :

ArrayList<Integer> al = new ArrayList<Integer>();
al.add(45);

int a = al.get(0); // returns the object of Integer . Automatically Unboxed . 

The primitive (non-object) types have there justification in efficiency.

The primitive types int, boolean, double are immediate data, whereas Objects are references. Hence fields (or variables)

int i;
double x;
Object s;

would need local memory 4+8+8? where for the object only the reference (address) to memory is stored.

Using the Object wrappers Integer, Double and others, one would introduce an indirection, reference to some Integer/Double instance in the heap memory.

Why boxing is needed?

That is a question of relative scope. In a future java it is planned to be able to have an ArrayList<int>, lifting primitive types.

Answer: For now an ArrayList only works for Object, reserving room for an object reference, and managing garbage collection likewise. Hence generic types are Object children. So if one wanted an ArrayList of floating point values, one needed to wrap a double in a Double object.

Here Java differs from the traditional C++ with its templates: C++ classes vector<string>, vector<int> would create two compilation products. Java design went for having one ArrayList.class, not needing for every parameter type a new compiled product.

So without boxing to Object one would need to compile classes for every occurrence of a parameter type. In concreto: every collection or container class would need a version for Object, int, double, boolean. The version for Object would handle all child classes.

In fact, the need for such diversification already existed in Java SE for IntBuffer, CharBuffer, DoubleBuffer, ... which operate on int, char, double. It was solved in a hacky way by generating these sources from a common one.

Starting with JDK 5, java has added two important functions: autoboxing and autounboxing. AutoBoxing is the process for which a primitive type is automatically encapsulated in the equivalent wrapper whenever such an object is needed. You do not have to explicitly construct an object. Auto-unboxing is the process whereby the value of an encapsulated object is automatically extracted from a type wrapper when its value is required. You do not need to call a method such as intValue() or doubleValue().

The addition of autoboxing and auto-unboxing greatly simplifies writing algorithms, eliminating the bait manually boxing and unboxing of values. It is also helpful to avoid mistakes. It is also very important for generics, who only operate on objects. Lastly, autoboxing facilitates work with the Collections Framework.