Why does the following work fine?
String str;
while (condition) {
str = calculateStr();
.....
}
But this one is said to be dangerou
In your example I presume str
is not used outside of the while
loop, otherwise you would not be asking the question, because declaring it inside the while
loop would not be an option, since it would not compile.
So, since str
is not used outside the loop, the smallest possible scope for str
is within the while loop.
So, the answer is emphatically that str
absolutely ought to be declared within the while loop. No ifs, no ands, no buts.
The only case where this rule might be violated is if for some reason it is of vital importance that every clock cycle must be squeezed out of the code, in which case you might want to consider instantiating something in an outer scope and reusing it instead of re-instantiating it on every iteration of an inner scope. However, this does not apply to your example, due to the immutability of strings in java: a new instance of str will always be created in the beginning of your loop and it will have to be thrown away at the end of it, so there is no possibility to optimize there.
EDIT: (injecting my comment below in the answer)
In any case, the right way to do things is to write all your code properly, establish a performance requirement for your product, measure your final product against this requirement, and if it does not satisfy it, then go optimize things. And what usually ends up happening is that you find ways to provide some nice and formal algorithmic optimizations in just a couple of places which make our program meet its performance requirements instead of having to go all over your entire code base and tweak and hack things in order to squeeze clock cycles here and there.
One solution to this problem could be to provide a variable scope encapsulating the while loop:
{
// all tmp loop variables here ....
// ....
String str;
while(condition){
str = calculateStr();
.....
}
}
They would be automatically de-reference when the outer scope ends.
Please skip to the updated answer...
For those who care about performance take out the System.out and limit the loop to 1 byte. Using double (test 1/2) and using String (3/4) the elapsed times in milliseconds is given below with Windows 7 Professional 64 bit and JDK-1.7.0_21. Bytecodes (also given below for test1 and test2) are not the same. I was too lazy to test with mutable & relatively complex objects.
double
Test1 took: 2710 msecs
Test2 took: 2790 msecs
String (just replace double with string in the tests)
Test3 took: 1200 msecs
Test4 took: 3000 msecs
Compiling and getting bytecode
javac.exe LocalTest1.java
javap.exe -c LocalTest1 > LocalTest1.bc
public class LocalTest1 {
public static void main(String[] args) throws Exception {
long start = System.currentTimeMillis();
double test;
for (double i = 0; i < 1000000000; i++) {
test = i;
}
long finish = System.currentTimeMillis();
System.out.println("Test1 Took: " + (finish - start) + " msecs");
}
}
public class LocalTest2 {
public static void main(String[] args) throws Exception {
long start = System.currentTimeMillis();
for (double i = 0; i < 1000000000; i++) {
double test = i;
}
long finish = System.currentTimeMillis();
System.out.println("Test1 Took: " + (finish - start) + " msecs");
}
}
Compiled from "LocalTest1.java"
public class LocalTest1 {
public LocalTest1();
Code:
0: aload_0
1: invokespecial #1 // Method java/lang/Object."<init>":()V
4: return
public static void main(java.lang.String[]) throws java.lang.Exception;
Code:
0: invokestatic #2 // Method java/lang/System.currentTimeMillis:()J
3: lstore_1
4: dconst_0
5: dstore 5
7: dload 5
9: ldc2_w #3 // double 1.0E9d
12: dcmpg
13: ifge 28
16: dload 5
18: dstore_3
19: dload 5
21: dconst_1
22: dadd
23: dstore 5
25: goto 7
28: invokestatic #2 // Method java/lang/System.currentTimeMillis:()J
31: lstore 5
33: getstatic #5 // Field java/lang/System.out:Ljava/io/PrintStream;
36: new #6 // class java/lang/StringBuilder
39: dup
40: invokespecial #7 // Method java/lang/StringBuilder."<init>":()V
43: ldc #8 // String Test1 Took:
45: invokevirtual #9 // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
48: lload 5
50: lload_1
51: lsub
52: invokevirtual #10 // Method java/lang/StringBuilder.append:(J)Ljava/lang/StringBuilder;
55: ldc #11 // String msecs
57: invokevirtual #9 // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
60: invokevirtual #12 // Method java/lang/StringBuilder.toString:()Ljava/lang/String;
63: invokevirtual #13 // Method java/io/PrintStream.println:(Ljava/lang/String;)V
66: return
}
Compiled from "LocalTest2.java"
public class LocalTest2 {
public LocalTest2();
Code:
0: aload_0
1: invokespecial #1 // Method java/lang/Object."<init>":()V
4: return
public static void main(java.lang.String[]) throws java.lang.Exception;
Code:
0: invokestatic #2 // Method java/lang/System.currentTimeMillis:()J
3: lstore_1
4: dconst_0
5: dstore_3
6: dload_3
7: ldc2_w #3 // double 1.0E9d
10: dcmpg
11: ifge 24
14: dload_3
15: dstore 5
17: dload_3
18: dconst_1
19: dadd
20: dstore_3
21: goto 6
24: invokestatic #2 // Method java/lang/System.currentTimeMillis:()J
27: lstore_3
28: getstatic #5 // Field java/lang/System.out:Ljava/io/PrintStream;
31: new #6 // class java/lang/StringBuilder
34: dup
35: invokespecial #7 // Method java/lang/StringBuilder."<init>":()V
38: ldc #8 // String Test1 Took:
40: invokevirtual #9 // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
43: lload_3
44: lload_1
45: lsub
46: invokevirtual #10 // Method java/lang/StringBuilder.append:(J)Ljava/lang/StringBuilder;
49: ldc #11 // String msecs
51: invokevirtual #9 // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
54: invokevirtual #12 // Method java/lang/StringBuilder.toString:()Ljava/lang/String;
57: invokevirtual #13 // Method java/io/PrintStream.println:(Ljava/lang/String;)V
60: return
}
It's really not easy to compare performance with all the JVM optimizations. However, it is somewhat possible. Better test and detailed results in Google Caliper
This is not identical to the code above. If you just code a dummy loop JVM skips it, so at least you need to assign and return something. This is also recommended in Caliper documentation.
@Param int size; // Set automatically by framework, provided in the Main
/**
* Variable is declared inside the loop.
*
* @param reps
* @return
*/
public double timeDeclaredInside(int reps) {
/* Dummy variable needed to workaround smart JVM */
double dummy = 0;
/* Test loop */
for (double i = 0; i <= size; i++) {
/* Declaration and assignment */
double test = i;
/* Dummy assignment to fake JVM */
if(i == size) {
dummy = test;
}
}
return dummy;
}
/**
* Variable is declared before the loop.
*
* @param reps
* @return
*/
public double timeDeclaredBefore(int reps) {
/* Dummy variable needed to workaround smart JVM */
double dummy = 0;
/* Actual test variable */
double test = 0;
/* Test loop */
for (double i = 0; i <= size; i++) {
/* Assignment */
test = i;
/* Not actually needed here, but we need consistent performance results */
if(i == size) {
dummy = test;
}
}
return dummy;
}
Summary: declaredBefore indicates better performance -really tiny- and it's against the smallest scope principle. JVM should actually do this for you
Declaring inside the loop limits the scope of the respective variable. It all depends on the requirement of the project on the scope of the variable.
These two examples result in the same thing. However, the first provides you with using the str
variable outside of the while loop; the second is not.
I compared the byte code of those two (similar) examples:
Let's look at 1. example:
package inside;
public class Test {
public static void main(String[] args) {
while(true){
String str = String.valueOf(System.currentTimeMillis());
System.out.println(str);
}
}
}
after javac Test.java
, javap -c Test
you'll get:
public class inside.Test extends java.lang.Object{
public inside.Test();
Code:
0: aload_0
1: invokespecial #1; //Method java/lang/Object."<init>":()V
4: return
public static void main(java.lang.String[]);
Code:
0: invokestatic #2; //Method java/lang/System.currentTimeMillis:()J
3: invokestatic #3; //Method java/lang/String.valueOf:(J)Ljava/lang/String;
6: astore_1
7: getstatic #4; //Field java/lang/System.out:Ljava/io/PrintStream;
10: aload_1
11: invokevirtual #5; //Method java/io/PrintStream.println:(Ljava/lang/String;)V
14: goto 0
}
Let's look at 2. example:
package outside;
public class Test {
public static void main(String[] args) {
String str;
while(true){
str = String.valueOf(System.currentTimeMillis());
System.out.println(str);
}
}
}
after javac Test.java
, javap -c Test
you'll get:
public class outside.Test extends java.lang.Object{
public outside.Test();
Code:
0: aload_0
1: invokespecial #1; //Method java/lang/Object."<init>":()V
4: return
public static void main(java.lang.String[]);
Code:
0: invokestatic #2; //Method java/lang/System.currentTimeMillis:()J
3: invokestatic #3; //Method java/lang/String.valueOf:(J)Ljava/lang/String;
6: astore_1
7: getstatic #4; //Field java/lang/System.out:Ljava/io/PrintStream;
10: aload_1
11: invokevirtual #5; //Method java/io/PrintStream.println:(Ljava/lang/String;)V
14: goto 0
}
The observations shows that there is no difference among those two examples. It's the result of JVM specifications...
But in the name of best coding practice it is recommended to declare the variable in the smallest possible scope (in this example it is inside the loop, as this is the only place where the variable is used).