How to trim a java stringbuilder?
I've used Zaven's analysis approach and StringBuilder's delete(start, end) method which performs far better than the deleteCharAt(index) approach, but slightly worse than the substring() approach. This method also uses the array copy, but array copy is called far fewer times (only twice in the worst case). In addition, this avoids creating multiple instances of intermediate Strings in case trim() is called repeatedly on the same StringBuilder object.
public class Main {
public static String trimOriginal(StringBuilder sb) {
return sb.toString().trim();
}
public static String trimDeleteRange(StringBuilder sb) {
int first, last;
for (first = 0; first < sb.length(); first++)
if (!Character.isWhitespace(sb.charAt(first)))
break;
for (last = sb.length(); last > first; last--)
if (!Character.isWhitespace(sb.charAt(last - 1)))
break;
if (first == last) {
sb.delete(0, sb.length());
} else {
if (last < sb.length()) {
sb.delete(last, sb.length());
}
if (first > 0) {
sb.delete(0, first);
}
}
return sb.toString();
}
public static String trimSubstring(StringBuilder sb) {
int first, last;
for (first = 0; first < sb.length(); first++)
if (!Character.isWhitespace(sb.charAt(first)))
break;
for (last = sb.length(); last > first; last--)
if (!Character.isWhitespace(sb.charAt(last - 1)))
break;
return sb.substring(first, last);
}
public static void main(String[] args) {
runAnalysis(1000);
runAnalysis(10000);
runAnalysis(100000);
runAnalysis(200000);
runAnalysis(500000);
runAnalysis(1000000);
}
private static void runAnalysis(int stringLength) {
System.out.println("Main:runAnalysis(string-length=" + stringLength + ")");
long originalTime = 0;
long deleteTime = 0;
long substringTime = 0;
for (int i = 0; i < 200; i++) {
StringBuilder temp = new StringBuilder();
char[] options = {' ', ' ', ' ', ' ', 'a', 'b', 'c', 'd'};
for (int j = 0; j < stringLength; j++) {
temp.append(options[(int) ((Math.random() * 1000)) % options.length]);
}
String testStr = temp.toString();
StringBuilder sb1 = new StringBuilder(testStr);
StringBuilder sb2 = new StringBuilder(testStr);
StringBuilder sb3 = new StringBuilder(testStr);
long timer1 = System.currentTimeMillis();
trimOriginal(sb1);
originalTime += System.currentTimeMillis() - timer1;
long timer2 = System.currentTimeMillis();
trimDeleteRange(sb2);
deleteTime += System.currentTimeMillis() - timer2;
long timer3 = System.currentTimeMillis();
trimSubstring(sb3);
substringTime += System.currentTimeMillis() - timer3;
}
System.out.println(" original: " + originalTime + " ms");
System.out.println(" delete-range: " + deleteTime + " ms");
System.out.println(" substring: " + substringTime + " ms");
}
}
Output:
Main:runAnalysis(string-length=1000)
original: 0 ms
delete-range: 4 ms
substring: 0 ms
Main:runAnalysis(string-length=10000)
original: 4 ms
delete-range: 9 ms
substring: 4 ms
Main:runAnalysis(string-length=100000)
original: 22 ms
delete-range: 33 ms
substring: 43 ms
Main:runAnalysis(string-length=200000)
original: 57 ms
delete-range: 93 ms
substring: 110 ms
Main:runAnalysis(string-length=500000)
original: 266 ms
delete-range: 220 ms
substring: 191 ms
Main:runAnalysis(string-length=1000000)
original: 479 ms
delete-range: 467 ms
substring: 426 ms
Don't worry about having two strings. It's a microoptimization.
If you really have detected a bottleneck, you can have a nearly-constant-time trimming - just iterate the first N chars, until they are Character.isWhitespace(c)
You should not use the deleteCharAt approach.
As Boris pointed out, the deleteCharAt method copies the array over every time. The code in the Java 5 that does this looks like this:
public AbstractStringBuilder deleteCharAt(int index) {
if ((index < 0) || (index >= count))
throw new StringIndexOutOfBoundsException(index);
System.arraycopy(value, index+1, value, index, count-index-1);
count--;
return this;
}
Of course, speculation alone is not enough to choose one method of optimization over another, so I decided to time the 3 approaches in this thread: the original, the delete approach, and the substring approach.
Here is the code I tested for the orignal:
public static String trimOriginal(StringBuilder sb) {
return sb.toString().trim();
}
The delete approach:
public static String trimDelete(StringBuilder sb) {
while (sb.length() > 0 && Character.isWhitespace(sb.charAt(0))) {
sb.deleteCharAt(0);
}
while (sb.length() > 0 && Character.isWhitespace(sb.charAt(sb.length() - 1))) {
sb.deleteCharAt(sb.length() - 1);
}
return sb.toString();
}
And the substring approach:
public static String trimSubstring(StringBuilder sb) {
int first, last;
for (first=0; first<sb.length(); first++)
if (!Character.isWhitespace(sb.charAt(first)))
break;
for (last=sb.length(); last>first; last--)
if (!Character.isWhitespace(sb.charAt(last-1)))
break;
return sb.substring(first, last);
}
I performed 100 tests, each time generating a million-character StringBuffer with ten thousand trailing and leading spaces. The testing itself is very basic, but it gives a good idea of how long the methods take.
Here is the code to time the 3 approaches:
public static void main(String[] args) {
long originalTime = 0;
long deleteTime = 0;
long substringTime = 0;
for (int i=0; i<100; i++) {
StringBuilder sb1 = new StringBuilder();
StringBuilder sb2 = new StringBuilder();
StringBuilder sb3 = new StringBuilder();
for (int j=0; j<10000; j++) {
sb1.append(" ");
sb2.append(" ");
sb3.append(" ");
}
for (int j=0; j<980000; j++) {
sb1.append("a");
sb2.append("a");
sb3.append("a");
}
for (int j=0; j<10000; j++) {
sb1.append(" ");
sb2.append(" ");
sb3.append(" ");
}
long timer1 = System.currentTimeMillis();
trimOriginal(sb1);
originalTime += System.currentTimeMillis() - timer1;
long timer2 = System.currentTimeMillis();
trimDelete(sb2);
deleteTime += System.currentTimeMillis() - timer2;
long timer3 = System.currentTimeMillis();
trimSubstring(sb3);
substringTime += System.currentTimeMillis() - timer3;
}
System.out.println("original: " + originalTime + " ms");
System.out.println("delete: " + deleteTime + " ms");
System.out.println("substring: " + substringTime + " ms");
}
I got the following output:
original: 176 ms
delete: 179242 ms
substring: 154 ms
As we see, the substring approach provides a very slight optimization over the original "two String" approach. However, the delete approach is extremely slow and should be avoided.
So to answer your question: you are fine trimming your StringBuilder the way you suggested in the question. The very slight optimization that the substring method offers probably does not justify the excess code.