Counting words and characters in java with high / low surrogates?

Question

I know there are some SO's about this topic, but all the solutions presented seem take a different approach than an example I've seen worked out in javascript.

Here's the javascript example that counts paragraphs, sentences words and characters typed in a text string, which includes checks for high / low surrogates for counting the characters specifically:

javascript version

count(text);

function count(original) {
    var trimmed = original.replace(/[\u200B]+/, '').trim();
    return {
        paragraphs: trimmed ? (trimmed.match(/\n+/g) || []).length + 1 : 0,
        sentences: trimmed ? (trimmed.match(/[.?!…\n]+./g) || []).length + 1 : 0,
        words: trimmed ? (trimmed.replace(/['";:,.?¿\-!¡]+/g, '').match(/\S+/g) || []).length : 0,
        characters: trimmed ? _decode(trimmed.replace(/\s/g, '')).length : 0,
        all: _decode(original).length
    };
};

function _decode(string) {
    var output = [],
        counter = 0,
        length = string.length,
        value, extra;
    while (counter < length) {
        value = string.charCodeAt(counter++);
        if (value >= 0xD800 && value <= 0xDBFF && counter < length) {
            // High surrogate, and there is a next character.
            extra = string.charCodeAt(counter++);
            if ((extra & 0xFC00) === 0xDC00) {
                // Low surrogate.
                output.push(((value & 0x3FF) << 10) + (extra & 0x3FF) + 0x10000);
            } else {
                // unmatched surrogate; only append this code unit, in case the next
                // code unit is the high surrogate of a surrogate pair
                output.push(value, extra);
                counter--;
            }
        } else {
            output.push(value);
        }
    }
    return output;
}

Demo below and in jsfiddle

 var text = 'This is a paragraph. This is the 2nd sentence in the 1st paragraph.\\nThis is another paragraph.'; var count = doCount(text); document.body.innerHTML = '<pre>' + text + '</pre><hr>'; for (i in count) { document.body.innerHTML += '<p>'+ i +': ' + count[i] + '</p>'; } /* COUNTING LIBRARY */ /** * Extracted from https://github.com/RadLikeWhoa/Countable/, which in * turn uses `ucs2decode` function from the punycode.js library. */ function doCount(original) { var trimmed = original.replace(/[\​]+/, '').trim(); return { paragraphs: trimmed ? (trimmed.match(/\\n+/g) || []).length + 1 : 0, sentences: trimmed ? (trimmed.match(/[.?!…\\n]+./g) || []).length + 1 : 0, words: trimmed ? (trimmed.replace(/['";:,.?¿\\-!¡]+/g, '').match(/\\S+/g) || []).length : 0, characters: trimmed ? _decode(trimmed.replace(/\\s/g, '')).length : 0, all: _decode(original).length }; }; /** * `ucs2decode` function from the punycode.js library. * * Creates an array containing the decimal code points of each Unicode * character in the string. While JavaScript uses UCS-2 internally, this * function will convert a pair of surrogate halves (each of which UCS-2 * exposes as separate characters) into a single code point, matching * UTF-16. * * @see <http://goo.gl/8M09r> * @see <http://goo.gl/u4UUC> * * @param {String} string The Unicode input string (UCS-2). * * @return {Array} The new array of code points. */ function _decode(string) { var output = [], counter = 0, length = string.length, value, extra; while (counter < length) { value = string.charCodeAt(counter++); if (value >= 0xD800 && value <= 0xDBFF && counter < length) { // High surrogate, and there is a next character. extra = string.charCodeAt(counter++); if ((extra & 0xFC00) === 0xDC00) { // Low surrogate. output.push(((value & 0x3FF) << 10) + (extra & 0x3FF) + 0x10000); } else { // unmatched surrogate; only append this code unit, in case the next // code unit is the high surrogate of a surrogate pair output.push(value, extra); counter--; } } else { output.push(value); } } return output; } 

I'm not well-versed in character encoding schemes and things like high / low surrogates, but is this not needed when counting using java?

I'm happy with the javascript implementation's results and I would like to do the counting on my java back-end, but I'm not sure if the same approach is even needed or how it should be done.

Answer 1

So what the javascript version does is reading surrogate pairs as one character if they appear in the text that is being decoded. This is possible in Javascript, because depending on the Javascript engine both UCS-2 and UTF-16 are allowed and UTF-16 supports high surrogates, meaning a single visible character is encoded using to code points. To calculate the length properly, the library accounts for the extra code points so they are counted as one.

In Java you have a similar problem, except in Java you can have many more encoding schemes. Fortunately, Java already returns the right length for a String that contains high surrogates. Still, if you want to separate the combined code points or even remove them, Java offers the Normalizer ( example of removing diacritics from a text).

string = Normalizer.normalize(string, Normalizer.Form.NFD);