• 比sun.misc.Encoder()/Decoder()的base64更高效的mxBase64算法


    package com.mxgraph.online;
    
    import java.util.Arrays;
    
    /** A very fast and memory efficient class to encode and decode to and from BASE64 in full accordance
     * with RFC 2045.<br><br>
     * On Windows XP sp1 with 1.4.2_04 and later ;), this encoder and decoder is about 10 times faster
     * on small arrays (10 - 1000 bytes) and 2-3 times as fast on larger arrays (10000 - 1000000 bytes)
     * compared to <code>sun.misc.Encoder()/Decoder()</code>.<br><br>
     *
     * On byte arrays the encoder is about 20% faster than Jakarta Commons Base64 Codec for encode and
     * about 50% faster for decoding large arrays. This implementation is about twice as fast on very small
     * arrays (&lt 30 bytes). If source/destination is a <code>String</code> this
     * version is about three times as fast due to the fact that the Commons Codec result has to be recoded
     * to a <code>String</code> from <code>byte[]</code>, which is very expensive.<br><br>
     *
     * This encode/decode algorithm doesn't create any temporary arrays as many other codecs do, it only
     * allocates the resulting array. This produces less garbage and it is possible to handle arrays twice
     * as large as algorithms that create a temporary array. (E.g. Jakarta Commons Codec). It is unknown
     * whether Sun's <code>sun.misc.Encoder()/Decoder()</code> produce temporary arrays but since performance
     * is quite low it probably does.<br><br>
     *
     * The encoder produces the same output as the Sun one except that the Sun's encoder appends
     * a trailing line separator if the last character isn't a pad. Unclear why but it only adds to the
     * length and is probably a side effect. Both are in conformance with RFC 2045 though.<br>
     * Commons codec seem to always att a trailing line separator.<br><br>
     *
     * <b>Note!</b>
     * The encode/decode method pairs (types) come in three versions with the <b>exact</b> same algorithm and
     * thus a lot of code redundancy. This is to not create any temporary arrays for transcoding to/from different
     * format types. The methods not used can simply be commented out.<br><br>
     *
     * There is also a "fast" version of all decode methods that works the same way as the normal ones, but
     * har a few demands on the decoded input. Normally though, these fast verions should be used if the source if
     * the input is known and it hasn't bee tampered with.<br><br>
     *
     * If you find the code useful or you find a bug, please send me a note at base64 @ miginfocom . com.
     *
     * Licence (BSD):
     * ==============
     *
     * Copyright (c) 2004, Mikael Grev, MiG InfoCom AB. (base64 @ miginfocom . com)
     * All rights reserved.
     *
     * Redistribution and use in source and binary forms, with or without modification,
     * are permitted provided that the following conditions are met:
     * Redistributions of source code must retain the above copyright notice, this list
     * of conditions and the following disclaimer.
     * Redistributions in binary form must reproduce the above copyright notice, this
     * list of conditions and the following disclaimer in the documentation and/or other
     * materials provided with the distribution.
     * Neither the name of the MiG InfoCom AB nor the names of its contributors may be
     * used to endorse or promote products derived from this software without specific
     * prior written permission.
     *
     * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
     * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
     * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     * IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
     * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
     * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA,
     * OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
     * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY
     * OF SUCH DAMAGE.
     *
     * @version 2.2
     * @author Mikael Grev
     *         Date: 2004-aug-02
     *         Time: 11:31:11
     */
    
    public class mxBase64
    {
        private static final char[] CA = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"
                .toCharArray();
    
        private static final int[] IA = new int[256];
        static
        {
            Arrays.fill(IA, -1);
            for (int i = 0, iS = CA.length; i < iS; i++)
                IA[CA[i]] = i;
            IA['='] = 0;
        }
    
        // ****************************************************************************************
        // *  char[] version
        // ****************************************************************************************
    
        /** Encodes a raw byte array into a BASE64 <code>char[]</code> representation i accordance with RFC 2045.
         * @param sArr The bytes to convert. If <code>null</code> or length 0 an empty array will be returned.
         * @param lineSep Optional "
    " after 76 characters, unless end of file.<br>
         * No line separator will be in breach of RFC 2045 which specifies max 76 per line but will be a
         * little faster.
         * @return A BASE64 encoded array. Never <code>null</code>.
         */
        public final static char[] encodeToChar(byte[] sArr, boolean lineSep)
        {
            // Check special case
            int sLen = sArr != null ? sArr.length : 0;
            if (sLen == 0)
                return new char[0];
    
            int eLen = (sLen / 3) * 3; // Length of even 24-bits.
            int cCnt = ((sLen - 1) / 3 + 1) << 2; // Returned character count
            int dLen = cCnt + (lineSep ? (cCnt - 1) / 76 << 1 : 0); // Length of returned array
            char[] dArr = new char[dLen];
    
            // Encode even 24-bits
            for (int s = 0, d = 0, cc = 0; s < eLen;)
            {
                // Copy next three bytes into lower 24 bits of int, paying attension to sign.
                int i = (sArr[s++] & 0xff) << 16 | (sArr[s++] & 0xff) << 8
                        | (sArr[s++] & 0xff);
    
                // Encode the int into four chars
                dArr[d++] = CA[(i >>> 18) & 0x3f];
                dArr[d++] = CA[(i >>> 12) & 0x3f];
                dArr[d++] = CA[(i >>> 6) & 0x3f];
                dArr[d++] = CA[i & 0x3f];
    
                // Add optional line separator
                if (lineSep && ++cc == 19 && d < dLen - 2)
                {
                    dArr[d++] = '
    ';
                    dArr[d++] = '
    ';
                    cc = 0;
                }
            }
    
            // Pad and encode last bits if source isn't even 24 bits.
            int left = sLen - eLen; // 0 - 2.
            if (left > 0)
            {
                // Prepare the int
                int i = ((sArr[eLen] & 0xff) << 10)
                        | (left == 2 ? ((sArr[sLen - 1] & 0xff) << 2) : 0);
    
                // Set last four chars
                dArr[dLen - 4] = CA[i >> 12];
                dArr[dLen - 3] = CA[(i >>> 6) & 0x3f];
                dArr[dLen - 2] = left == 2 ? CA[i & 0x3f] : '=';
                dArr[dLen - 1] = '=';
            }
            return dArr;
        }
    
        /** Decodes a BASE64 encoded char array. All illegal characters will be ignored and can handle both arrays with
         * and without line separators.
         * @param sArr The source array. <code>null</code> or length 0 will return an empty array.
         * @return The decoded array of bytes. May be of length 0. Will be <code>null</code> if the legal characters
         * (including '=') isn't divideable by 4.  (I.e. definitely corrupted).
         */
        public final static byte[] decode(char[] sArr)
        {
            // Check special case
            int sLen = sArr != null ? sArr.length : 0;
            if (sLen == 0)
                return new byte[0];
    
            // Count illegal characters (including '
    ', '
    ') to know what size the returned array will be,
            // so we don't have to reallocate & copy it later.
            int sepCnt = 0; // Number of separator characters. (Actually illegal characters, but that's a bonus...)
            for (int i = 0; i < sLen; i++)
                // If input is "pure" (I.e. no line separators or illegal chars) base64 this loop can be commented out.
                if (IA[sArr[i]] < 0)
                    sepCnt++;
    
            // Check so that legal chars (including '=') are evenly divideable by 4 as specified in RFC 2045.
            if ((sLen - sepCnt) % 4 != 0)
                return null;
    
            int pad = 0;
            for (int i = sLen; i > 1 && IA[sArr[--i]] <= 0;)
                if (sArr[i] == '=')
                    pad++;
    
            int len = ((sLen - sepCnt) * 6 >> 3) - pad;
    
            byte[] dArr = new byte[len]; // Preallocate byte[] of exact length
    
            for (int s = 0, d = 0; d < len;)
            {
                // Assemble three bytes into an int from four "valid" characters.
                int i = 0;
                for (int j = 0; j < 4; j++)
                { // j only increased if a valid char was found.
                    int c = IA[sArr[s++]];
                    if (c >= 0)
                        i |= c << (18 - j * 6);
                    else
                        j--;
                }
                // Add the bytes
                dArr[d++] = (byte) (i >> 16);
                if (d < len)
                {
                    dArr[d++] = (byte) (i >> 8);
                    if (d < len)
                        dArr[d++] = (byte) i;
                }
            }
            return dArr;
        }
    
        /** Decodes a BASE64 encoded char array that is known to be resonably well formatted. The method is about twice as
         * fast as {@link #decode(char[])}. The preconditions are:<br>
         * + The array must have a line length of 76 chars OR no line separators at all (one line).<br>
         * + Line separator must be "
    ", as specified in RFC 2045
         * + The array must not contain illegal characters within the encoded string<br>
         * + The array CAN have illegal characters at the beginning and end, those will be dealt with appropriately.<br>
         * @param sArr The source array. Length 0 will return an empty array. <code>null</code> will throw an exception.
         * @return The decoded array of bytes. May be of length 0.
         */
        public final static byte[] decodeFast(char[] sArr)
        {
            // Check special case
            int sLen = sArr.length;
            if (sLen == 0)
                return new byte[0];
    
            int sIx = 0, eIx = sLen - 1; // Start and end index after trimming.
    
            // Trim illegal chars from start
            while (sIx < eIx && IA[sArr[sIx]] < 0)
                sIx++;
    
            // Trim illegal chars from end
            while (eIx > 0 && IA[sArr[eIx]] < 0)
                eIx--;
    
            // get the padding count (=) (0, 1 or 2)
            int pad = sArr[eIx] == '=' ? (sArr[eIx - 1] == '=' ? 2 : 1) : 0; // Count '=' at end.
            int cCnt = eIx - sIx + 1; // Content count including possible separators
            int sepCnt = sLen > 76 ? (sArr[76] == '
    ' ? cCnt / 78 : 0) << 1 : 0;
    
            int len = ((cCnt - sepCnt) * 6 >> 3) - pad; // The number of decoded bytes
            byte[] dArr = new byte[len]; // Preallocate byte[] of exact length
    
            // Decode all but the last 0 - 2 bytes.
            int d = 0;
            for (int cc = 0, eLen = (len / 3) * 3; d < eLen;)
            {
                // Assemble three bytes into an int from four "valid" characters.
                int i = IA[sArr[sIx++]] << 18 | IA[sArr[sIx++]] << 12
                        | IA[sArr[sIx++]] << 6 | IA[sArr[sIx++]];
    
                // Add the bytes
                dArr[d++] = (byte) (i >> 16);
                dArr[d++] = (byte) (i >> 8);
                dArr[d++] = (byte) i;
    
                // If line separator, jump over it.
                if (sepCnt > 0 && ++cc == 19)
                {
                    sIx += 2;
                    cc = 0;
                }
            }
    
            if (d < len)
            {
                // Decode last 1-3 bytes (incl '=') into 1-3 bytes
                int i = 0;
                for (int j = 0; sIx <= eIx - pad; j++)
                    i |= IA[sArr[sIx++]] << (18 - j * 6);
    
                for (int r = 16; d < len; r -= 8)
                    dArr[d++] = (byte) (i >> r);
            }
    
            return dArr;
        }
    
        // ****************************************************************************************
        // *  byte[] version
        // ****************************************************************************************
    
        /** Encodes a raw byte array into a BASE64 <code>byte[]</code> representation i accordance with RFC 2045.
         * @param sArr The bytes to convert. If <code>null</code> or length 0 an empty array will be returned.
         * @param lineSep Optional "
    " after 76 characters, unless end of file.<br>
         * No line separator will be in breach of RFC 2045 which specifies max 76 per line but will be a
         * little faster.
         * @return A BASE64 encoded array. Never <code>null</code>.
         */
        public final static byte[] encodeToByte(byte[] sArr, boolean lineSep)
        {
            // Check special case
            int sLen = sArr != null ? sArr.length : 0;
            if (sLen == 0)
                return new byte[0];
    
            int eLen = (sLen / 3) * 3; // Length of even 24-bits.
            int cCnt = ((sLen - 1) / 3 + 1) << 2; // Returned character count
            int dLen = cCnt + (lineSep ? (cCnt - 1) / 76 << 1 : 0); // Length of returned array
            byte[] dArr = new byte[dLen];
    
            // Encode even 24-bits
            for (int s = 0, d = 0, cc = 0; s < eLen;)
            {
                // Copy next three bytes into lower 24 bits of int, paying attension to sign.
                int i = (sArr[s++] & 0xff) << 16 | (sArr[s++] & 0xff) << 8
                        | (sArr[s++] & 0xff);
    
                // Encode the int into four chars
                dArr[d++] = (byte) CA[(i >>> 18) & 0x3f];
                dArr[d++] = (byte) CA[(i >>> 12) & 0x3f];
                dArr[d++] = (byte) CA[(i >>> 6) & 0x3f];
                dArr[d++] = (byte) CA[i & 0x3f];
    
                // Add optional line separator
                if (lineSep && ++cc == 19 && d < dLen - 2)
                {
                    dArr[d++] = '
    ';
                    dArr[d++] = '
    ';
                    cc = 0;
                }
            }
    
            // Pad and encode last bits if source isn't an even 24 bits.
            int left = sLen - eLen; // 0 - 2.
            if (left > 0)
            {
                // Prepare the int
                int i = ((sArr[eLen] & 0xff) << 10)
                        | (left == 2 ? ((sArr[sLen - 1] & 0xff) << 2) : 0);
    
                // Set last four chars
                dArr[dLen - 4] = (byte) CA[i >> 12];
                dArr[dLen - 3] = (byte) CA[(i >>> 6) & 0x3f];
                dArr[dLen - 2] = left == 2 ? (byte) CA[i & 0x3f] : (byte) '=';
                dArr[dLen - 1] = '=';
            }
            return dArr;
        }
    
        /** Decodes a BASE64 encoded byte array. All illegal characters will be ignored and can handle both arrays with
         * and without line separators.
         * @param sArr The source array. Length 0 will return an empty array. <code>null</code> will throw an exception.
         * @return The decoded array of bytes. May be of length 0. Will be <code>null</code> if the legal characters
         * (including '=') isn't divideable by 4. (I.e. definitely corrupted).
         */
        public final static byte[] decode(byte[] sArr)
        {
            // Check special case
            int sLen = sArr.length;
    
            // Count illegal characters (including '
    ', '
    ') to know what size the returned array will be,
            // so we don't have to reallocate & copy it later.
            int sepCnt = 0; // Number of separator characters. (Actually illegal characters, but that's a bonus...)
            for (int i = 0; i < sLen; i++)
                // If input is "pure" (I.e. no line separators or illegal chars) base64 this loop can be commented out.
                if (IA[sArr[i] & 0xff] < 0)
                    sepCnt++;
    
            // Check so that legal chars (including '=') are evenly divideable by 4 as specified in RFC 2045.
            if ((sLen - sepCnt) % 4 != 0)
                return null;
    
            int pad = 0;
            for (int i = sLen; i > 1 && IA[sArr[--i] & 0xff] <= 0;)
                if (sArr[i] == '=')
                    pad++;
    
            int len = ((sLen - sepCnt) * 6 >> 3) - pad;
    
            byte[] dArr = new byte[len]; // Preallocate byte[] of exact length
    
            for (int s = 0, d = 0; d < len;)
            {
                // Assemble three bytes into an int from four "valid" characters.
                int i = 0;
                for (int j = 0; j < 4; j++)
                { // j only increased if a valid char was found.
                    int c = IA[sArr[s++] & 0xff];
                    if (c >= 0)
                        i |= c << (18 - j * 6);
                    else
                        j--;
                }
    
                // Add the bytes
                dArr[d++] = (byte) (i >> 16);
                if (d < len)
                {
                    dArr[d++] = (byte) (i >> 8);
                    if (d < len)
                        dArr[d++] = (byte) i;
                }
            }
    
            return dArr;
        }
    
        /** Decodes a BASE64 encoded byte array that is known to be resonably well formatted. The method is about twice as
         * fast as {@link #decode(byte[])}. The preconditions are:<br>
         * + The array must have a line length of 76 chars OR no line separators at all (one line).<br>
         * + Line separator must be "
    ", as specified in RFC 2045
         * + The array must not contain illegal characters within the encoded string<br>
         * + The array CAN have illegal characters at the beginning and end, those will be dealt with appropriately.<br>
         * @param sArr The source array. Length 0 will return an empty array. <code>null</code> will throw an exception.
         * @return The decoded array of bytes. May be of length 0.
         */
        public final static byte[] decodeFast(byte[] sArr)
        {
            // Check special case
            int sLen = sArr.length;
            if (sLen == 0)
                return new byte[0];
    
            int sIx = 0, eIx = sLen - 1; // Start and end index after trimming.
    
            // Trim illegal chars from start
            while (sIx < eIx && IA[sArr[sIx] & 0xff] < 0)
                sIx++;
    
            // Trim illegal chars from end
            while (eIx > 0 && IA[sArr[eIx] & 0xff] < 0)
                eIx--;
    
            // get the padding count (=) (0, 1 or 2)
            int pad = sArr[eIx] == '=' ? (sArr[eIx - 1] == '=' ? 2 : 1) : 0; // Count '=' at end.
            int cCnt = eIx - sIx + 1; // Content count including possible separators
            int sepCnt = sLen > 76 ? (sArr[76] == '
    ' ? cCnt / 78 : 0) << 1 : 0;
    
            int len = ((cCnt - sepCnt) * 6 >> 3) - pad; // The number of decoded bytes
            byte[] dArr = new byte[len]; // Preallocate byte[] of exact length
    
            // Decode all but the last 0 - 2 bytes.
            int d = 0;
            for (int cc = 0, eLen = (len / 3) * 3; d < eLen;)
            {
                // Assemble three bytes into an int from four "valid" characters.
                int i = IA[sArr[sIx++]] << 18 | IA[sArr[sIx++]] << 12
                        | IA[sArr[sIx++]] << 6 | IA[sArr[sIx++]];
    
                // Add the bytes
                dArr[d++] = (byte) (i >> 16);
                dArr[d++] = (byte) (i >> 8);
                dArr[d++] = (byte) i;
    
                // If line separator, jump over it.
                if (sepCnt > 0 && ++cc == 19)
                {
                    sIx += 2;
                    cc = 0;
                }
            }
    
            if (d < len)
            {
                // Decode last 1-3 bytes (incl '=') into 1-3 bytes
                int i = 0;
                for (int j = 0; sIx <= eIx - pad; j++)
                    i |= IA[sArr[sIx++]] << (18 - j * 6);
    
                for (int r = 16; d < len; r -= 8)
                    dArr[d++] = (byte) (i >> r);
            }
    
            return dArr;
        }
    
        // ****************************************************************************************
        // * String version
        // ****************************************************************************************
    
        /** Encodes a raw byte array into a BASE64 <code>String</code> representation i accordance with RFC 2045.
         * @param sArr The bytes to convert. If <code>null</code> or length 0 an empty array will be returned.
         * @param lineSep Optional "
    " after 76 characters, unless end of file.<br>
         * No line separator will be in breach of RFC 2045 which specifies max 76 per line but will be a
         * little faster.
         * @return A BASE64 encoded array. Never <code>null</code>.
         */
        public final static String encodeToString(byte[] sArr, boolean lineSep)
        {
            // Reuse char[] since we can't create a String incrementally anyway and StringBuffer/Builder would be slower.
            return new String(encodeToChar(sArr, lineSep));
        }
    
        /** Decodes a BASE64 encoded <code>String</code>. All illegal characters will be ignored and can handle both strings with
         * and without line separators.<br>
         * <b>Note!</b> It can be up to about 2x the speed to call <code>decode(str.toCharArray())</code> instead. That
         * will create a temporary array though. This version will use <code>str.charAt(i)</code> to iterate the string.
         * @param str The source string. <code>null</code> or length 0 will return an empty array.
         * @return The decoded array of bytes. May be of length 0. Will be <code>null</code> if the legal characters
         * (including '=') isn't divideable by 4.  (I.e. definitely corrupted).
         */
        public final static byte[] decode(String str)
        {
            // Check special case
            int sLen = str != null ? str.length() : 0;
            if (sLen == 0)
                return new byte[0];
    
            // Count illegal characters (including '
    ', '
    ') to know what size the returned array will be,
            // so we don't have to reallocate & copy it later.
            int sepCnt = 0; // Number of separator characters. (Actually illegal characters, but that's a bonus...)
            for (int i = 0; i < sLen; i++)
                // If input is "pure" (I.e. no line separators or illegal chars) base64 this loop can be commented out.
                if (IA[str.charAt(i)] < 0)
                    sepCnt++;
    
            // Check so that legal chars (including '=') are evenly divideable by 4 as specified in RFC 2045.
            if ((sLen - sepCnt) % 4 != 0)
                return null;
    
            // Count '=' at end
            int pad = 0;
            for (int i = sLen; i > 1 && IA[str.charAt(--i)] <= 0;)
                if (str.charAt(i) == '=')
                    pad++;
    
            int len = ((sLen - sepCnt) * 6 >> 3) - pad;
    
            byte[] dArr = new byte[len]; // Preallocate byte[] of exact length
    
            for (int s = 0, d = 0; d < len;)
            {
                // Assemble three bytes into an int from four "valid" characters.
                int i = 0;
                for (int j = 0; j < 4; j++)
                { // j only increased if a valid char was found.
                    int c = IA[str.charAt(s++)];
                    if (c >= 0)
                        i |= c << (18 - j * 6);
                    else
                        j--;
                }
                // Add the bytes
                dArr[d++] = (byte) (i >> 16);
                if (d < len)
                {
                    dArr[d++] = (byte) (i >> 8);
                    if (d < len)
                        dArr[d++] = (byte) i;
                }
            }
            return dArr;
        }
    
        /** Decodes a BASE64 encoded string that is known to be resonably well formatted. The method is about twice as
         * fast as {@link #decode(String)}. The preconditions are:<br>
         * + The array must have a line length of 76 chars OR no line separators at all (one line).<br>
         * + Line separator must be "
    ", as specified in RFC 2045
         * + The array must not contain illegal characters within the encoded string<br>
         * + The array CAN have illegal characters at the beginning and end, those will be dealt with appropriately.<br>
         * @param s The source string. Length 0 will return an empty array. <code>null</code> will throw an exception.
         * @return The decoded array of bytes. May be of length 0.
         */
        public final static byte[] decodeFast(String s)
        {
            // Check special case
            int sLen = s.length();
            if (sLen == 0)
                return new byte[0];
    
            int sIx = 0, eIx = sLen - 1; // Start and end index after trimming.
    
            // Trim illegal chars from start
            while (sIx < eIx && IA[s.charAt(sIx) & 0xff] < 0)
                sIx++;
    
            // Trim illegal chars from end
            while (eIx > 0 && IA[s.charAt(eIx) & 0xff] < 0)
                eIx--;
    
            // get the padding count (=) (0, 1 or 2)
            int pad = s.charAt(eIx) == '=' ? (s.charAt(eIx - 1) == '=' ? 2 : 1) : 0; // Count '=' at end.
            int cCnt = eIx - sIx + 1; // Content count including possible separators
            int sepCnt = sLen > 76 ? (s.charAt(76) == '
    ' ? cCnt / 78 : 0) << 1
                    : 0;
    
            int len = ((cCnt - sepCnt) * 6 >> 3) - pad; // The number of decoded bytes
            byte[] dArr = new byte[len]; // Preallocate byte[] of exact length
    
            // Decode all but the last 0 - 2 bytes.
            int d = 0;
            for (int cc = 0, eLen = (len / 3) * 3; d < eLen;)
            {
                // Assemble three bytes into an int from four "valid" characters.
                int i = IA[s.charAt(sIx++)] << 18 | IA[s.charAt(sIx++)] << 12
                        | IA[s.charAt(sIx++)] << 6 | IA[s.charAt(sIx++)];
    
                // Add the bytes
                dArr[d++] = (byte) (i >> 16);
                dArr[d++] = (byte) (i >> 8);
                dArr[d++] = (byte) i;
    
                // If line separator, jump over it.
                if (sepCnt > 0 && ++cc == 19)
                {
                    sIx += 2;
                    cc = 0;
                }
            }
    
            if (d < len)
            {
                // Decode last 1-3 bytes (incl '=') into 1-3 bytes
                int i = 0;
                for (int j = 0; sIx <= eIx - pad; j++)
                    i |= IA[s.charAt(sIx++)] << (18 - j * 6);
    
                for (int r = 16; d < len; r -= 8)
                    dArr[d++] = (byte) (i >> r);
            }
    
            return dArr;
        }
    }
  • 相关阅读:
    Centos 7.X 安装JDK1.8
    CentOS 7.X 搭建时间服务器 --- chrony
    CentOS 7.X 设置系统时间
    CentOS 7.X 关闭SELinux
    C11工具类:时间处理
    C11内存管理之道:智能指针
    C11性能之道:标准库优化
    C11性能之道:转移和转发
    C11性能之道:右值引用
    C11简洁之道:tupe元祖
  • 原文地址:https://www.cnblogs.com/rgqancy/p/9084039.html
Copyright © 2020-2023  润新知