[java] Scanner vs. StringTokenizer vs. String.Split

I just learned about Java's Scanner class and now I'm wondering how it compares/competes with the StringTokenizer and String.Split. I know that the StringTokenizer and String.Split only work on Strings, so why would I want to use the Scanner for a String? Is Scanner just intended to be one-stop-shopping for spliting?

One important difference is that both String.split() and Scanner can produce empty strings but StringTokenizer never does it.

For example:

String str = "ab cd  ef";

StringTokenizer st = new StringTokenizer(str, " ");
for (int i = 0; st.hasMoreTokens(); i++) System.out.println("#" + i + ": " + st.nextToken());

String[] split = str.split(" ");
for (int i = 0; i < split.length; i++) System.out.println("#" + i + ": " + split[i]);

Scanner sc = new Scanner(str).useDelimiter(" ");
for (int i = 0; sc.hasNext(); i++) System.out.println("#" + i + ": " + sc.next());

Output:

//StringTokenizer
#0: ab
#1: cd
#2: ef
//String.split()
#0: ab
#1: cd
#2: 
#3: ef
//Scanner
#0: ab
#1: cd
#2: 
#3: ef

This is because the delimiter for String.split() and Scanner.useDelimiter() is not just a string, but a regular expression. We can replace the delimiter " " with " +" in the example above to make them behave like StringTokenizer.

StringTokenizer was always there. It is the fastest of all, but the enumeration-like idiom might not look as elegant as the others.

split came to existence on JDK 1.4. Slower than tokenizer but easier to use, since it is callable from the String class.

Scanner came to be on JDK 1.5. It is the most flexible and fills a long standing gap on the Java API to support an equivalent of the famous Cs scanf function family.

String.split seems to be much slower than StringTokenizer. The only advantage with split is that you get an array of the tokens. Also you can use any regular expressions in split. org.apache.commons.lang.StringUtils has a split method which works much more faster than any of two viz. StringTokenizer or String.split. But the CPU utilization for all the three is nearly the same. So we also need a method which is less CPU intensive, which I am still not able to find.

Split is slow, but not as slow as Scanner. StringTokenizer is faster than split. However, I found that I could obtain double the speed, by trading some flexibility, to get a speed-boost, which I did at JFastParser https://github.com/hughperkins/jfastparser

Testing on a string containing one million doubles:

Scanner: 10642 ms
Split: 715 ms
StringTokenizer: 544ms
JFastParser: 290ms

Let's start by eliminating StringTokenizer. It is getting old and doesn't even support regular expressions. Its documentation states:

StringTokenizer is a legacy class that is retained for compatibility reasons although its use is discouraged in new code. It is recommended that anyone seeking this functionality use the split method of String or the java.util.regex package instead.

So let's throw it out right away. That leaves split() and Scanner. What's the difference between them?

For one thing, split() simply returns an array, which makes it easy to use a foreach loop:

for (String token : input.split("\\s+") { ... }

Scanner is built more like a stream:

while (myScanner.hasNext()) {
    String token = myScanner.next();
    ...
}

or

while (myScanner.hasNextDouble()) {
    double token = myScanner.nextDouble();
    ...
}

(It has a rather large API, so don't think that it's always restricted to such simple things.)

This stream-style interface can be useful for parsing simple text files or console input, when you don't have (or can't get) all the input before starting to parse.

Personally, the only time I can remember using Scanner is for school projects, when I had to get user input from the command line. It makes that sort of operation easy. But if I have a String that I want to split up, it's almost a no-brainer to go with split().

I recently did some experiments about the bad performance of String.split() in highly performance sensitive situations. You may find this useful.

http://eblog.chrononsystems.com/hidden-evils-of-javas-stringsplit-and-stringr

The gist is that String.split() compiles a Regular Expression pattern each time and can thus slow down your program, compared to if you use a precompiled Pattern object and use it directly to operate on a String.

For the default scenarios I would suggest Pattern.split() as well but if you need maximum performance (especially on Android all solutions I tested are quite slow) and you only need to split by a single char, I now use my own method:

public static ArrayList<String> splitBySingleChar(final char[] s,
        final char splitChar) {
    final ArrayList<String> result = new ArrayList<String>();
    final int length = s.length;
    int offset = 0;
    int count = 0;
    for (int i = 0; i < length; i++) {
        if (s[i] == splitChar) {
            if (count > 0) {
                result.add(new String(s, offset, count));
            }
            offset = i + 1;
            count = 0;
        } else {
            count++;
        }
    }
    if (count > 0) {
        result.add(new String(s, offset, count));
    }
    return result;
}

Use "abc".toCharArray() to get the char array for a String. For example:

String s = "     a bb   ccc  dddd eeeee  ffffff    ggggggg ";
ArrayList<String> result = splitBySingleChar(s.toCharArray(), ' ');

If you have a String object you want to tokenize, favor using String's split method over a StringTokenizer. If you're parsing text data from a source outside your program, like from a file, or from the user, that's where a Scanner comes in handy.

String.split() works very good but has its own boundaries, like if you wanted to split a string as shown below based on single or double pipe (|) symbol, it doesn't work. In this situation you can use StringTokenizer.

ABC|IJK