How does auto boxing/unboxing work in Java?

Since JDK 5.0, auto boxing/unboxing was introduced in Java. The trick is simple and helpful, but when I started testing different conversions between wrapper classes and primitive types, I get really confused how the concept of auto boxing works in Java. For example:

Boxing

int intValue = 0;
Integer intObject = intValue;
byte byteValue = 0;
intObject = byteValue; // ==> Error

After trying different cases (short, long, float, double), the only case which is accepted by the compiler is when the type of the value on the right of affectation operator is int. When I looked inside the source of Integer.class I found that it implements only one constructor with int parameter.

So my conclusion is that the concept of auto boxing is based on constructor implemented in the wrapper class. I want to know if this conclusion is true or there is another concept used by auto boxing?

Unboxing

Integer intObject = new Integer(0);
byte byteValue = intObject; // ==> Error (the same Error with short)
int intValue = intObject; 
double doubleValue = intObject;

My conclusion about unboxing is that the wrapper class gives the value wrapped by the object in the corresponding type (Integer ==> int), then the compiler use the usual rules of converting primitive types (byte => short => int => long => float => double). I want to know if this conclusion is true or there is another concept used by auto unboxing?

Solution 1:

When in doubt, check the bytecode:

Integer n = 42;

becomes:

0: bipush        42
2: invokestatic  #16                 // Method java/lang/Integer.valueOf:(I)Ljava/lang/Integer;
5: astore_1      

So in actuality, valueOf() is used as opposed to the constructor (and the same goes for the other wrapper classes). This is beneficial since it allows for caching, and doesn't force the creation of a new object on each boxing operation.

The reverse is the following:

int n = Integer.valueOf(42);

which becomes:

0: bipush        42
2: invokestatic  #16                 // Method java/lang/Integer.valueOf:(I)Ljava/lang/Integer;
5: invokevirtual #22                 // Method java/lang/Integer.intValue:()I
8: istore_1      

i.e. intValue() is used (again, it's analogous for the other wrapper types as well). This is really all auto(un)boxing boils down to.

You can read about boxing and unboxing conversions in JLS §5.1.7 and JLS §5.1.8, respectively.

Solution 2:

This confusion can be cleared by using a switch of javac -XD-printflat which is very helpful in cases such as this one. So to unravel the mystery of boxing and unboxing you can write a simple program like following :

import java.util.*;

public class Boxing{
  public static void main(String[] args){
    Double d1 = 10.123;
    Float  f1 = 12.12f;
    Long   l1 = 1234L;
    Integer i1 = 55555;
    Short   s1 = 2345;
    Byte    b1 = 89;

    double d2 = d1;
    float  f2 = f1;
    long   l2 = l1;
    int    i2 = i1;
    short  s2 = s1;
    byte   b2 = b1;
  }
} 

and now we compile the above file as:

javac -XD-printflat -d src/ Boxing.java

output of this command is a java file with all the syntactic sugar (Generic types, enhanced for loop and in this case boxing-unboxing etc) removed. following is the output

import java.util.*;

public class Boxing {

    public Boxing() {
        super();
    }

    public static void main(String[] args) {
        Double d1 = Double.valueOf(10.123);
        Float f1 = Float.valueOf(12.12F);
        Long l1 = Long.valueOf(1234L);
        Integer i1 = Integer.valueOf(55555);
        Short s1 = Short.valueOf(2345);
        Byte b1 = Byte.valueOf(89);
        double d2 = d1.doubleValue();
        float f2 = f1.floatValue();
        long l2 = l1.longValue();
        int i2 = i1.intValue();
        short s2 = s1.shortValue();
        byte b2 = b1.byteValue();
    }
}

this is how java does boxing unboxing. using valueOf and ***Value methods.