Inward vs outward negation from performance point of view

[Anand Kane]

Hi All

Here is a question I was asked in one of the interviews. Which of the following two performs better,

var!=null or !(var==null)?

The interviewer told me that the latter performs better. I haven't heard or read any such thing before. So I will appreciate if someone can confirm the interviewer's opinion.

Thanks
Anand

[Ninad Kuchekar]

Hi Anand,

Welcome to the Ranch!

That is actually a very good question for interviews. Although, personally, I would never use

 !(var == null)   

I would prefer,

 (var != null) 

It provides better readability. However, if anyone has some other insights regarding performance I would be very interested in understanding them.

[Rob Spoor]

To be honest, unless you're programming time-critical code, you won't notice any difference. Changing var != null into !(var == null) because it may be microseconds (more likely nanoseconds) faster is micro-optimization.

[Rob Spoor]

OK, I wanted to know myself, so I wrote this little test program:

public class Test
{
    public static void main(String[] args) throws Exception
    {
        String s = null;
        if (s != null)
        {
            System.out.println(s);
        }
        if (!(s == null))
        {
            System.out.println(s);
        }
    }
}

I then used javap -c to disassemble it:

Compiled from "Test.java"
public class Test extends java.lang.Object{
public Test();
  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:	aconst_null
   1:	astore_1
   2:	aload_1
   3:	ifnull	13
   6:	getstatic	#2; //Field java/lang/System.out:Ljava/io/PrintStream;
   9:	aload_1
   10:	invokevirtual	#3; //Method java/io/PrintStream.println:(Ljava/lang/String;)V
   13:	aload_1
   14:	ifnull	24
   17:	getstatic	#2; //Field java/lang/System.out:Ljava/io/PrintStream;
   20:	aload_1
   21:	invokevirtual	#3; //Method java/io/PrintStream.println:(Ljava/lang/String;)V
   24:	return

}

Both steps use ifnull, getstatic, aload_1 and invokevirtual. In fact, when I use JAD to decompile it, the !(s == null) is transformed into s != null. So my conclusion is: there is no difference.

[Mike Simmons]

Furthermore, on many systems there will be absolutely no difference at all. I just tested using javac 1.6.0_20 on my MacBook Pro, compiling the following program:

public class Negation {
  public void test1() {
    if (condition() != null) {
      System.out.println("true");
    }
  }
  public void test2() {
    if (!(condition() == null)) {
      System.out.println("true");
    }
  }

  private Object condition() {
    return Math.random() >= .5 ? this : null;
  }
}

I then ran "javap -c Negation" and got the following output:

> javap -c Negation
Compiled from "Negation.java"
public class Negation extends java.lang.Object{
public Negation();
  Code:
   0:	aload_0
   1:	invokespecial	#1; //Method java/lang/Object."<init>":()V
   4:	return

public void test1();
  Code:
   0:	aload_0
   1:	invokespecial	#2; //Method condition:()Ljava/lang/Object;
   4:	ifnull	15
   7:	getstatic	#3; //Field java/lang/System.out:Ljava/io/PrintStream;
   10:	ldc	#4; //String true
   12:	invokevirtual	#5; //Method java/io/PrintStream.println:(Ljava/lang/String;)V
   15:	return

public void test2();
  Code:
   0:	aload_0
   1:	invokespecial	#2; //Method condition:()Ljava/lang/Object;
   4:	ifnull	15
   7:	getstatic	#3; //Field java/lang/System.out:Ljava/io/PrintStream;
   10:	ldc	#4; //String true
   12:	invokevirtual	#5; //Method java/io/PrintStream.println:(Ljava/lang/String;)V
   15:	return

}

Evidently the methods test1() and test2() compiled to exactly the same code. There will be no performance difference between these two methods. Maybe on some other systems there will be, if the compiler is not smart enough to recognize that the code is equivalent. But I think your interviewer gave a bad answer in this case, and I am very suspicious of the reasoning they were using.

[Mike Simmons]

Curse you, Robtimus.

Well, this is further evidence that the code is equivalent on more than one platform. Well, unless Rob is also using a Mac with JDK 6.

[Anand Kane]

Thank you all for posting responses with clear evidence. I posted this particular question because I wasn't sure of it. There were few more to make me realize that there was no point arguing with that interviewer. And the company is one the biggest brandnames across the globe.

[Campbell Ritchie]

Mike Simmons wrote:Curse you, Robtimus. . . .

You only say that sort of thing when he beats you by 4 seconds, not 4 minutes. Though, come to think of it, I did manage to beat him by about two hours on Tuesday . . .

[Rob Spoor]

Mike Simmons wrote:Curse you, Robtimus.

To be honest, I prefer your example. Because you used two different methods it's clearer to see the boundaries of both pieces of code.

Well, this is further evidence that the code is equivalent on more than one platform. Well, unless Rob is also using a Mac with JDK 6.

Windows XP with JDK 6.