Saturday, November 1, 2014

String concatenation vs StringBuilder in Java

Lets consider two ways of concatenating Strings in Java.
The most common way to do so is to use operator +:
public String concatenateWithPlus(String s1, String s2){
  return "concatenated=" + s1 + s2;
}
Another way is to use StringBuilder:
public String concatenateWithStringBuilder(String s1, String s2){
  final StringBuilder result = new StringBuilder();
  result.append("concatenated=");
  result.append(s1);
  result.append(s2);
  return result.toString();
}
The question is which method is faster and consumes less resources?

It seems natural to think that + operator creates a new instance of a String every time it is applied. However, lets look into the bytecodes generated for these two snippets.
  public java.lang.String concatenateWithPlus(java.lang.String, java.lang.String);
    Code:
   0: new           #2  // class java/lang/StringBuilder
   3: dup
   4: invokespecial #3  // Method java/lang/StringBuilder."":()V
   7: ldc           #4  // String concatenated=
   9: invokevirtual #5  // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
  12: aload_1
  13: invokevirtual #5  // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
  16: aload_2
  17: invokevirtual #5  // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
  20: invokevirtual #6  // Method java/lang/StringBuilder.toString:()Ljava/lang/String;
  23: areturn
  public java.lang.String concatenateWithStringBuilder(java.lang.String, java.lang.String);
    Code:
   0: new           #2  // class java/lang/StringBuilder
   3: dup
   4: invokespecial #3  // Method java/lang/StringBuilder."":()V
   7: astore_3
   8: aload_3
   9: ldc           #4  // String concatenated=
  11: invokevirtual #5  // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
  14: pop
  15: aload_3
  16: aload_1
  17: invokevirtual #5  // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
  20: pop
  21: aload_3
  22: aload_2
  23: invokevirtual #5  // Method java/lang/StringBuilder.append:(Ljava/lang/String;)Ljava/lang/StringBuilder;
  26: pop
  27: aload_3
  28: invokevirtual #6  // Method java/lang/StringBuilder.toString:()Ljava/lang/String;
  31: areturn
As we see the compiler uses StringBuilder in the bytecode for both cases! So both variants don't create unnecessary String instances, and moreover, the second case requires two more instructions per concatenation (pop and aload_3), therefore is slower than the first one!

Take away from this example is that we should look into the bytecode more often and verify our assumptions with the compiler.

PS:
These are the commands to generate bytecode from a Java source:
javac MyClass.java
javap -c MyClass > MyClass.bc