tika-parsers/tika-parsers-standard/tika-parsers-standard-modules/tika-parser-text-module/src/main/java/org/apache/tika/parser/txt/CharsetRecog_UTF8.java

// © 2016 and later: Unicode, Inc. and others.
// License & terms of use: http://www.unicode.org/copyright.html#License
/**
 * ******************************************************************************
 * Copyright (C) 2005 - 2014, International Business Machines Corporation and  *
 * others. All Rights Reserved.                                                *
 * ******************************************************************************
 */
package org.apache.tika.parser.txt;

/**
 * Charset recognizer for UTF-8
 */
class CharsetRecog_UTF8 extends CharsetRecognizer {

    String getName() {
        return "UTF-8";
    }

    /* (non-Javadoc)
     * @see com.ibm.icu.text.CharsetRecognizer#match(com.ibm.icu.text.CharsetDetector)
     */
    CharsetMatch match(CharsetDetector det) {
        boolean hasBOM = false;
        int numValid = 0;
        int numInvalid = 0;
        byte[] input = det.fRawInput;
        int i;
        int trailBytes = 0;
        int confidence;

        if (det.fRawLength >= 3 && (input[0] & 0xFF) == 0xef && (input[1] & 0xFF) == 0xbb &&
                (input[2] & 0xFF) == 0xbf) {
            hasBOM = true;
        }

        // Scan for multi-byte sequences
        for (i = 0; i < det.fRawLength; i++) {
            int b = input[i];
            if ((b & 0x80) == 0) {
                continue;   // ASCII
            }

            // Hi bit on char found.  Figure out how long the sequence should be
            if ((b & 0x0e0) == 0x0c0) {
                trailBytes = 1;
            } else if ((b & 0x0f0) == 0x0e0) {
                trailBytes = 2;
            } else if ((b & 0x0f8) == 0xf0) {
                trailBytes = 3;
            } else {
                numInvalid++;
                continue;
            }

            // Verify that we've got the right number of trail bytes in the sequence
            for (; ; ) {
                i++;
                if (i >= det.fRawLength) {
                    break;
                }
                b = input[i];
                if ((b & 0xc0) != 0x080) {
                    numInvalid++;
                    break;
                }
                if (--trailBytes == 0) {
                    numValid++;
                    break;
                }
            }
        }

        // Cook up some sort of confidence score, based on presense of a BOM
        //    and the existence of valid and/or invalid multi-byte sequences.
        confidence = 0;
        if (hasBOM && numInvalid == 0) {
            confidence = 100;
        } else if (hasBOM && numValid > numInvalid * 10) {
            confidence = 80;
        } else if (numValid > 3 && numInvalid == 0) {
            confidence = 100;
        } else if (numValid > 0 && numInvalid == 0) {
            confidence = 80;
        } else if (numValid == 0 && numInvalid == 0) {
            // Plain ASCII. Confidence must be > 10, it's more likely than UTF-16, which
            //              accepts ASCII with confidence = 10.
            // TODO: add plain ASCII as an explicitly detected type.
            confidence = 15;
        } else if (numValid > numInvalid * 10) {
            // Probably corruput utf-8 data.  Valid sequences aren't likely by chance.
            confidence = 25;
        }
        return confidence == 0 ? null : new CharsetMatch(det, this, confidence);
    }

}