Declaring local variable as final in a loop

Question

I know that very similar questions have been asked and answered already, I read the ones I was able to locate and still not 100% clear.

Considering this code snippet:

public static void fooMethod {

   while(<...>) {
     ....
     final int temp = <something>;
     ....
   }
}

No inner classes, nothing else special or unusual. Seems counter-intuitive to me.

Does declaring a local variable final in the above sample serve any purpose whatsoever?

Do I understand correctly that with or without final here compiler will produce exactly the same byte-code?

Am I missing something here? If it's an RTFM case, please point me in the right direction.

Follow-up Question (if I may)

What do I gain and/or lose by re-writing like this (with the understanding that temp does not have to be a primitive)?

public static void fooMethod2 {

   int temp;
   while(<...>) {
     ....
     temp = <something>;
     ....
   }
}

Answer 1

In a few words: The final keyword, when used in local variables and parameters, does not make it to the generated bytecode (.class file) and, as expected, its use has no effect during runtime. _{(Compile-time, it could make a diference, though, check below.)}

In those cases, when not enforced due to anonymous inner classes, it is merely a style choice, useful in documenting the intended scope of the variable.

The tests below confirm that information.

---

1: If the compiler can make something of it, using final makes difference:

Look at this snippet:

boolean zZ = true;
while (zZ) {
    int xX = 1001;         // <------------- xX
    int yY = 1002;         // <------------- yY
    zZ = (xX == yY);
}

Two int variables, xX and yY. First time declared both as final and second time, took away the final from both. Here are the generated bytecodes (printed with javap -c):

Both final:

     0: iconst_1             // pushes int 1 (true) onto the stack
     1: istore_1             // stores the int on top of the stack into var zZ
     2: goto          15
     5: sipush        1001   // pushes 1001 onto the operand stack
     8: istore_2             // stores on xX
     9: sipush        1002   // pushes 1002 onto the operand stack
    12: istore_3             // stores on yY
    13: iconst_0             // pushes 0 (false): does not compare!! <---------
    14: istore_1             // stores on zZ
    15: iload_1              // loads zZ
    16: ifne          5      // goes to 5 if top int (zZ) is not 0
    19: return        

Both non-final:

    // 0: to 12: all the same
    13: iload_2              // pushes xX onto the stack
    14: iload_3              // pushes yY onto the stack
    15: if_icmpne     22     // here it compares xX and yY! <------------
    18: iconst_1      
    19: goto          23
    22: iconst_0      
    23: istore_1      
    24: iload_1       
    25: ifne          5
    28: return        

In the case above, when they are final, the compiler knows that they are not equal and never compares them (false is generated in the bytecode wherever xX == yY is).

From this, we can conclude, bytecode-wise, the compiler does can do some optimization on the generated code when using final. (I'm not saying they are meaningful, but for sure final is not only a style choice here.)

2: If the compiler can't conclude anything, using final on local vars is just a design choice:

Now take the following code:

boolean zZ = true;
int aA = 1001;
int bB = 1002;
while (zZ) {
    final int xX = aA;   // <------- took away the "final" here, didnt matter
    final int yY = bB;   // <------- took away the "final" here, didnt matter
    zZ = (xX == yY);
}

In this case, even using final, the compiler cannot tell compiler-time if xX and yY are equal, right?

Because of this, we can see: the generated bytecode is exactly the same (same MD5!) when we generate the class with or without final.

While, in the general case, some say and others disagree that there are performance benefits of using final, in local blocks, final is definitely only a style choice.

3: Local variables inside or outside loops - no difference at all:

The generated bytecode for this snippet...

boolean zZ = true;
int aA = 1001, bB = 1002;
while (zZ) {
    int xX = aA;                      // <--- declaration is inside WHILE
    int yY = bB;
    zZ = (xX == yY);
}

...and the generated bytecode for this snippet...

boolean zZ = true;
int aA = 1001, bB = 1002;
int xX, yY;                           // <--- declaration is outside WHILE
while (zZ) {
    xX = aA;
    yY = bB;
    zZ = (xX == yY);
}

...are exactly the same (only the line numbers changed, of course).

Other tests using objects (not only primitive typed variables) showed the same behaviour.

It is safe to conclude, then, if not used elsewhere, declaring local variables inside or outside loops is pretty much a design choice, with no bytecode effects.

_{Note: All tests were made under Oracle's JRE, version 1.7.0_13.}