com.unboundid.util.Base64 Maven / Gradle / Ivy
                 Go to download
                
        
                    Show more of this group  Show more artifacts with this name
Show all versions of unboundid-ldapsdk-commercial-edition Show documentation
                Show all versions of unboundid-ldapsdk-commercial-edition Show documentation
      The UnboundID LDAP SDK for Java is a fast, comprehensive, and easy-to-use
      Java API for communicating with LDAP directory servers and performing
      related tasks like reading and writing LDIF, encoding and decoding data
      using base64 and ASN.1 BER, and performing secure communication.  This
      package contains the Commercial Edition of the LDAP SDK, which includes
      all of the general-purpose functionality contained in the Standard
      Edition, plus additional functionality specific to UnboundID server
      products.
    
                
            /*
 * Copyright 2007-2016 UnboundID Corp.
 * All Rights Reserved.
 */
/*
 * Copyright (C) 2008-2016 UnboundID Corp.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License (GPLv2 only)
 * or the terms of the GNU Lesser General Public License (LGPLv2.1 only)
 * as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, see .
 */
package com.unboundid.util;
import java.io.IOException;
import java.text.ParseException;
import static com.unboundid.util.UtilityMessages.*;
import static com.unboundid.util.Validator.*;
/**
 * This class provides methods for encoding and decoding data in base64 as
 * defined in RFC 4648.  It
 * provides a relatively compact way of representing binary data using only
 * printable characters.  It uses a six-bit encoding mechanism in which every
 * three bytes of raw data is converted to four bytes of base64-encoded data,
 * which means that it only requires about a 33% increase in size (as compared
 * with a hexadecimal representation, which requires a 100% increase in size).
 * 
 * Base64 encoding is used in LDIF processing as per
 * RFC 2849 to represent data
 * that contains special characters or might otherwise be ambiguous.  It is also
 * used in a number of other areas (e.g., for the ASCII representation of
 * certificates) where it is desirable to deal with a string containing only
 * printable characters but the raw data may contain other characters outside of
 * that range.
 * 
 * This class also provides support for the URL-safe variant (called base64url)
 * as described in RFC 4648 section 5.  This is nearly the same as base64,
 * except that the '+' and '/' characters are replaced with '-' and '_',
 * respectively.  The padding may be omitted if the context makes the data size
 * clear, but if padding is to be used then the URL-encoded "%3d" will be used
 * instead of "=".
 * 
 * Example
 * The following examples demonstrate the process for base64-encoding raw data,
 * and for decoding a string containing base64-encoded data back to the raw
 * data used to create it:
 * 
 * // Base64-encode some raw data:
 * String base64String = Base64.encode(rawDataBytes);
 *
 * // Decode a base64 string back to raw data:
 * byte[] decodedRawDataBytes;
 * try
 * {
 *   decodedRawDataBytes = Base64.decode(base64String);
 * }
 * catch (ParseException pe)
 * {
 *   // The string did not represent a valid base64 encoding.
 *   decodedRawDataBytes = null;
 * }
 * 
 */
@ThreadSafety(level=ThreadSafetyLevel.COMPLETELY_THREADSAFE)
public final class Base64
{
  /**
   * The set of characters in the base64 alphabet.
   */
  private static final char[] BASE64_ALPHABET =
       ("ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz" +
        "0123456789+/").toCharArray();
  /**
   * The set of characters in the base64url alphabet.
   */
  private static final char[] BASE64URL_ALPHABET =
       ("ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz" +
        "0123456789-_").toCharArray();
  /**
   * Prevent this class from being instantiated.
   */
  private Base64()
  {
    // No implementation is required.
  }
  /**
   * Encodes the UTF-8 representation of the provided string in base64 format.
   *
   * @param  data  The raw data to be encoded.  It must not be {@code null}.
   *
   * @return  The base64-encoded representation of the provided data.
   */
  public static String encode(final String data)
  {
    ensureNotNull(data);
    return encode(StaticUtils.getBytes(data));
  }
  /**
   * Encodes the provided data in base64 format.
   *
   * @param  data  The raw data to be encoded.  It must not be {@code null}.
   *
   * @return  The base64-encoded representation of the provided data.
   */
  public static String encode(final byte[] data)
  {
    ensureNotNull(data);
    final StringBuilder buffer = new StringBuilder(4*data.length/3+1);
    encode(BASE64_ALPHABET, data, 0, data.length, buffer, "=");
    return buffer.toString();
  }
  /**
   * Appends a base64-encoded version of the contents of the provided buffer
   * (using a UTF-8 representation) to the given buffer.
   *
   * @param  data    The raw data to be encoded.  It must not be {@code null}.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   */
  public static void encode(final String data, final StringBuilder buffer)
  {
    ensureNotNull(data);
    encode(StaticUtils.getBytes(data), buffer);
  }
  /**
   * Appends a base64-encoded version of the contents of the provided buffer
   * (using a UTF-8 representation) to the given buffer.
   *
   * @param  data    The raw data to be encoded.  It must not be {@code null}.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   */
  public static void encode(final String data, final ByteStringBuffer buffer)
  {
    ensureNotNull(data);
    encode(StaticUtils.getBytes(data), buffer);
  }
  /**
   * Appends a base64-encoded representation of the provided data to the given
   * buffer.
   *
   * @param  data    The raw data to be encoded.  It must not be {@code null}.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   */
  public static void encode(final byte[] data, final StringBuilder buffer)
  {
    encode(BASE64_ALPHABET, data, 0, data.length, buffer, "=");
  }
  /**
   * Appends a base64-encoded representation of the provided data to the given
   * buffer.
   *
   * @param  data    The array containing the raw data to be encoded.  It must
   *                 not be {@code null}.
   * @param  off     The offset in the array at which the data to encode begins.
   * @param  length  The number of bytes to be encoded.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   */
  public static void encode(final byte[] data, final int off, final int length,
                            final StringBuilder buffer)
  {
    encode(BASE64_ALPHABET, data, off, length, buffer, "=");
  }
  /**
   * Appends a base64-encoded representation of the provided data to the given
   * buffer.
   *
   * @param  data    The raw data to be encoded.  It must not be {@code null}.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   */
  public static void encode(final byte[] data, final ByteStringBuffer buffer)
  {
    encode(BASE64_ALPHABET, data, 0, data.length, buffer, "=");
  }
  /**
   * Appends a base64-encoded representation of the provided data to the given
   * buffer.
   *
   * @param  data    The raw data to be encoded.  It must not be {@code null}.
   * @param  off     The offset in the array at which the data to encode begins.
   * @param  length  The number of bytes to be encoded.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   */
  public static void encode(final byte[] data, final int off, final int length,
                            final ByteStringBuffer buffer)
  {
    encode(BASE64_ALPHABET, data, off, length, buffer, "=");
  }
  /**
   * Retrieves a base64url-encoded representation of the provided data to the
   * given buffer.
   *
   * @param  data  The raw data to be encoded.  It must not be {@code null}.
   * @param  pad   Indicates whether to pad the URL if necessary.  Padding will
   *               use "%3d", as the URL-escaped representation of the equal
   *               sign.
   *
   * @return  A base64url-encoded representation of the provided data to the
   *          given buffer.
   */
  public static String urlEncode(final String data, final boolean pad)
  {
    return urlEncode(StaticUtils.getBytes(data), pad);
  }
  /**
   * Retrieves a base64url-encoded representation of the provided data to the
   * given buffer.
   *
   * @param  data    The raw data to be encoded.  It must not be {@code null}.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   * @param  pad     Indicates whether to pad the URL if necessary.  Padding
   *                 will use "%3d", as the URL-escaped representation of the
   *                 equal sign.
   */
  public static void urlEncode(final String data, final StringBuilder buffer,
                               final boolean pad)
  {
    final byte[] dataBytes = StaticUtils.getBytes(data);
    encode(BASE64_ALPHABET, dataBytes, 0, dataBytes.length, buffer,
         (pad ? "%3d" : null));
  }
  /**
   * Retrieves a base64url-encoded representation of the provided data to the
   * given buffer.
   *
   * @param  data    The raw data to be encoded.  It must not be {@code null}.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   * @param  pad     Indicates whether to pad the URL if necessary.  Padding
   *                 will use "%3d", as the URL-escaped representation of the
   *                 equal sign.
   */
  public static void urlEncode(final String data, final ByteStringBuffer buffer,
                               final boolean pad)
  {
    final byte[] dataBytes = StaticUtils.getBytes(data);
    encode(BASE64_ALPHABET, dataBytes, 0, dataBytes.length, buffer,
         (pad ? "%3d" : null));
  }
  /**
   * Retrieves a base64url-encoded representation of the provided data to the
   * given buffer.
   *
   * @param  data  The raw data to be encoded.  It must not be {@code null}.
   * @param  pad   Indicates whether to pad the URL if necessary.  Padding will
   *               use "%3d", as the URL-escaped representation of the equal
   *               sign.
   *
   * @return  A base64url-encoded representation of the provided data to the
   *          given buffer.
   */
  public static String urlEncode(final byte[] data, final boolean pad)
  {
    final StringBuilder buffer = new StringBuilder(4*data.length/3+6);
    encode(BASE64URL_ALPHABET, data, 0, data.length, buffer,
         (pad ? "%3d" : null));
    return buffer.toString();
  }
  /**
   * Appends a base64url-encoded representation of the provided data to the
   * given buffer.
   *
   * @param  data    The raw data to be encoded.  It must not be {@code null}.
   * @param  off     The offset in the array at which the data to encode begins.
   * @param  length  The number of bytes to be encoded.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   * @param  pad     Indicates whether to pad the URL if necessary.  Padding
   *                 will use "%3d", as the URL-escaped representation of the
   *                 equal sign.
   */
  public static void urlEncode(final byte[] data, final int off,
                               final int length, final StringBuilder buffer,
                               final boolean pad)
  {
    encode(BASE64URL_ALPHABET, data, off, length, buffer, (pad ? "%3d" : null));
  }
  /**
   * Appends a base64url-encoded representation of the provided data to the
   * given buffer.
   *
   * @param  data    The raw data to be encoded.  It must not be {@code null}.
   * @param  off     The offset in the array at which the data to encode begins.
   * @param  length  The number of bytes to be encoded.
   * @param  buffer  The buffer to which the base64-encoded data is to be
   *                 written.
   * @param  pad     Indicates whether to pad the URL if necessary.  Padding
   *                 will use "%3d", as the URL-escaped representation of the
   *                 equal sign.
   */
  public static void urlEncode(final byte[] data, final int off,
                               final int length, final ByteStringBuffer buffer,
                               final boolean pad)
  {
    encode(BASE64URL_ALPHABET, data, off, length, buffer, (pad ? "%3d" : null));
  }
  /**
   * Appends a base64-encoded representation of the provided data to the given
   * buffer.
   *
   * @param  alphabet  The alphabet of base64 characters to use for the
   *                   encoding.
   * @param  data      The raw data to be encoded.  It must not be {@code null}.
   * @param  off       The offset in the array at which the data to encode
   *                   begins.
   * @param  length    The number of bytes to be encoded.
   * @param  buffer    The buffer to which the base64-encoded data is to be
   *                   written.
   * @param  padStr    The string to use for padding.  It may be {@code null} if
   *                   no padding should be applied.
   */
  private static void encode(final char[] alphabet, final byte[] data,
                             final int off, final int length,
                             final Appendable buffer, final String padStr)
  {
    ensureNotNull(data);
    ensureTrue(data.length >= off);
    ensureTrue(data.length >= (off+length));
    if (length == 0)
    {
      return;
    }
    try
    {
      int pos = off;
      for (int i=0; i < (length / 3); i++)
      {
        final int intValue = ((data[pos++] & 0xFF) << 16) |
             ((data[pos++] & 0xFF) << 8) |
             (data[pos++] & 0xFF);
        buffer.append(alphabet[(intValue >> 18) & 0x3F]);
        buffer.append(alphabet[(intValue >> 12) & 0x3F]);
        buffer.append(alphabet[(intValue >> 6) & 0x3F]);
        buffer.append(alphabet[intValue & 0x3F]);
      }
      switch ((off+length) - pos)
      {
        case 1:
          int intValue = (data[pos] & 0xFF) << 16;
          buffer.append(alphabet[(intValue >> 18) & 0x3F]);
          buffer.append(alphabet[(intValue >> 12) & 0x3F]);
          if (padStr != null)
          {
            buffer.append(padStr);
            buffer.append(padStr);
          }
          return;
        case 2:
          intValue = ((data[pos++] & 0xFF) << 16) | ((data[pos] & 0xFF) << 8);
          buffer.append(alphabet[(intValue >> 18) & 0x3F]);
          buffer.append(alphabet[(intValue >> 12) & 0x3F]);
          buffer.append(alphabet[(intValue >> 6) & 0x3F]);
          if (padStr != null)
          {
            buffer.append(padStr);
          }
          return;
      }
    }
    catch (final IOException ioe)
    {
      Debug.debugException(ioe);
      // This should never happen.
      throw new RuntimeException(ioe.getMessage(), ioe);
    }
  }
  /**
   * Decodes the contents of the provided base64-encoded string.
   *
   * @param  data  The base64-encoded string to decode.  It must not be
   *               {@code null}.
   *
   * @return  A byte array containing the decoded data.
   *
   * @throws  ParseException  If the contents of the provided string cannot be
   *                          parsed as base64-encoded data.
   */
  public static byte[] decode(final String data)
         throws ParseException
  {
    ensureNotNull(data);
    final int length = data.length();
    if (length == 0)
    {
      return new byte[0];
    }
    if ((length % 4) != 0)
    {
      throw new ParseException(ERR_BASE64_DECODE_INVALID_LENGTH.get(), length);
    }
    int numBytes = 3 * (length / 4);
    if (data.charAt(length-2) == '=')
    {
      numBytes -= 2;
    }
    else if (data.charAt(length-1) == '=')
    {
      numBytes--;
    }
    final byte[] b = new byte[numBytes];
    int stringPos = 0;
    int arrayPos  = 0;
    while (stringPos < length)
    {
      int intValue = 0x00;
      for (int i=0; i < 4; i++)
      {
        intValue <<= 6;
        switch (data.charAt(stringPos++))
        {
          case 'A':
            intValue |= 0x00;
            break;
          case 'B':
            intValue |= 0x01;
            break;
          case 'C':
            intValue |= 0x02;
            break;
          case 'D':
            intValue |= 0x03;
            break;
          case 'E':
            intValue |= 0x04;
            break;
          case 'F':
            intValue |= 0x05;
            break;
          case 'G':
            intValue |= 0x06;
            break;
          case 'H':
            intValue |= 0x07;
            break;
          case 'I':
            intValue |= 0x08;
            break;
          case 'J':
            intValue |= 0x09;
            break;
          case 'K':
            intValue |= 0x0A;
            break;
          case 'L':
            intValue |= 0x0B;
            break;
          case 'M':
            intValue |= 0x0C;
            break;
          case 'N':
            intValue |= 0x0D;
            break;
          case 'O':
            intValue |= 0x0E;
            break;
          case 'P':
            intValue |= 0x0F;
            break;
          case 'Q':
            intValue |= 0x10;
            break;
          case 'R':
            intValue |= 0x11;
            break;
          case 'S':
            intValue |= 0x12;
            break;
          case 'T':
            intValue |= 0x13;
            break;
          case 'U':
            intValue |= 0x14;
            break;
          case 'V':
            intValue |= 0x15;
            break;
          case 'W':
            intValue |= 0x16;
            break;
          case 'X':
            intValue |= 0x17;
            break;
          case 'Y':
            intValue |= 0x18;
            break;
          case 'Z':
            intValue |= 0x19;
            break;
          case 'a':
            intValue |= 0x1A;
            break;
          case 'b':
            intValue |= 0x1B;
            break;
          case 'c':
            intValue |= 0x1C;
            break;
          case 'd':
            intValue |= 0x1D;
            break;
          case 'e':
            intValue |= 0x1E;
            break;
          case 'f':
            intValue |= 0x1F;
            break;
          case 'g':
            intValue |= 0x20;
            break;
          case 'h':
            intValue |= 0x21;
            break;
          case 'i':
            intValue |= 0x22;
            break;
          case 'j':
            intValue |= 0x23;
            break;
          case 'k':
            intValue |= 0x24;
            break;
          case 'l':
            intValue |= 0x25;
            break;
          case 'm':
            intValue |= 0x26;
            break;
          case 'n':
            intValue |= 0x27;
            break;
          case 'o':
            intValue |= 0x28;
            break;
          case 'p':
            intValue |= 0x29;
            break;
          case 'q':
            intValue |= 0x2A;
            break;
          case 'r':
            intValue |= 0x2B;
            break;
          case 's':
            intValue |= 0x2C;
            break;
          case 't':
            intValue |= 0x2D;
            break;
          case 'u':
            intValue |= 0x2E;
            break;
          case 'v':
            intValue |= 0x2F;
            break;
          case 'w':
            intValue |= 0x30;
            break;
          case 'x':
            intValue |= 0x31;
            break;
          case 'y':
            intValue |= 0x32;
            break;
          case 'z':
            intValue |= 0x33;
            break;
          case '0':
            intValue |= 0x34;
            break;
          case '1':
            intValue |= 0x35;
            break;
          case '2':
            intValue |= 0x36;
            break;
          case '3':
            intValue |= 0x37;
            break;
          case '4':
            intValue |= 0x38;
            break;
          case '5':
            intValue |= 0x39;
            break;
          case '6':
            intValue |= 0x3A;
            break;
          case '7':
            intValue |= 0x3B;
            break;
          case '8':
            intValue |= 0x3C;
            break;
          case '9':
            intValue |= 0x3D;
            break;
          case '+':
            intValue |= 0x3E;
            break;
          case '/':
            intValue |= 0x3F;
            break;
          case '=':
            switch (length - stringPos)
            {
              case 0:
                // The string ended with a single equal sign, so there are only
                // two bytes left.  Shift the value eight bits to the right and
                // read those two bytes.
                intValue >>= 8;
                b[arrayPos++] = (byte) ((intValue >> 8) & 0xFF);
                b[arrayPos]   = (byte) (intValue & 0xFF);
                return b;
              case 1:
                // The string ended with two equal signs, so there is only one
                // byte left.  Shift the value ten bits to the right and read
                // that single byte.
                intValue >>= 10;
                b[arrayPos] = (byte) (intValue & 0xFF);
                return b;
              default:
                throw new ParseException(ERR_BASE64_DECODE_UNEXPECTED_EQUAL.get(
                                              (stringPos-1)),
                                         (stringPos-1));
            }
          default:
            throw new ParseException(ERR_BASE64_DECODE_UNEXPECTED_CHAR.get(
                                          data.charAt(stringPos-1)),
                                     (stringPos-1));
        }
      }
      b[arrayPos++] = (byte) ((intValue >> 16) & 0xFF);
      b[arrayPos++] = (byte) ((intValue >> 8) & 0xFF);
      b[arrayPos++] = (byte) (intValue & 0xFF);
    }
    return b;
  }
  /**
   * Decodes the contents of the provided base64-encoded string to a string
   * containing the raw data using the UTF-8 encoding.
   *
   * @param  data  The base64-encoded string to decode.  It must not be
   *               {@code null}.
   *
   * @return  A string containing the decoded data.
   *
   * @throws  ParseException  If the contents of the provided string cannot be
   *                          parsed as base64-encoded data using the UTF-8
   *                          encoding.
   */
  public static String decodeToString(final String data)
         throws ParseException
  {
    ensureNotNull(data);
    final byte[] decodedBytes = decode(data);
    return StaticUtils.toUTF8String(decodedBytes);
  }
  /**
   * Decodes the contents of the provided base64url-encoded string.
   *
   * @param  data  The base64url-encoded string to decode.  It must not be
   *               {@code null}.
   *
   * @return  A byte array containing the decoded data.
   *
   * @throws  ParseException  If the contents of the provided string cannot be
   *                          parsed as base64url-encoded data.
   */
  public static byte[] urlDecode(final String data)
         throws ParseException
  {
    ensureNotNull(data);
    final int length = data.length();
    if (length == 0)
    {
      return new byte[0];
    }
    int stringPos = 0;
    final ByteStringBuffer buffer = new ByteStringBuffer(length);
decodeLoop:
    while (stringPos < length)
    {
      int intValue = 0x00;
      for (int i=0; i < 4; i++)
      {
        // Since the value may not be padded, then we need to handle the
        // possibility of missing characters.
        final char c;
        if (stringPos >= length)
        {
          c = '=';
          stringPos++;
        }
        else
        {
          c = data.charAt(stringPos++);
        }
        intValue <<= 6;
        switch (c)
        {
          case 'A':
            intValue |= 0x00;
            break;
          case 'B':
            intValue |= 0x01;
            break;
          case 'C':
            intValue |= 0x02;
            break;
          case 'D':
            intValue |= 0x03;
            break;
          case 'E':
            intValue |= 0x04;
            break;
          case 'F':
            intValue |= 0x05;
            break;
          case 'G':
            intValue |= 0x06;
            break;
          case 'H':
            intValue |= 0x07;
            break;
          case 'I':
            intValue |= 0x08;
            break;
          case 'J':
            intValue |= 0x09;
            break;
          case 'K':
            intValue |= 0x0A;
            break;
          case 'L':
            intValue |= 0x0B;
            break;
          case 'M':
            intValue |= 0x0C;
            break;
          case 'N':
            intValue |= 0x0D;
            break;
          case 'O':
            intValue |= 0x0E;
            break;
          case 'P':
            intValue |= 0x0F;
            break;
          case 'Q':
            intValue |= 0x10;
            break;
          case 'R':
            intValue |= 0x11;
            break;
          case 'S':
            intValue |= 0x12;
            break;
          case 'T':
            intValue |= 0x13;
            break;
          case 'U':
            intValue |= 0x14;
            break;
          case 'V':
            intValue |= 0x15;
            break;
          case 'W':
            intValue |= 0x16;
            break;
          case 'X':
            intValue |= 0x17;
            break;
          case 'Y':
            intValue |= 0x18;
            break;
          case 'Z':
            intValue |= 0x19;
            break;
          case 'a':
            intValue |= 0x1A;
            break;
          case 'b':
            intValue |= 0x1B;
            break;
          case 'c':
            intValue |= 0x1C;
            break;
          case 'd':
            intValue |= 0x1D;
            break;
          case 'e':
            intValue |= 0x1E;
            break;
          case 'f':
            intValue |= 0x1F;
            break;
          case 'g':
            intValue |= 0x20;
            break;
          case 'h':
            intValue |= 0x21;
            break;
          case 'i':
            intValue |= 0x22;
            break;
          case 'j':
            intValue |= 0x23;
            break;
          case 'k':
            intValue |= 0x24;
            break;
          case 'l':
            intValue |= 0x25;
            break;
          case 'm':
            intValue |= 0x26;
            break;
          case 'n':
            intValue |= 0x27;
            break;
          case 'o':
            intValue |= 0x28;
            break;
          case 'p':
            intValue |= 0x29;
            break;
          case 'q':
            intValue |= 0x2A;
            break;
          case 'r':
            intValue |= 0x2B;
            break;
          case 's':
            intValue |= 0x2C;
            break;
          case 't':
            intValue |= 0x2D;
            break;
          case 'u':
            intValue |= 0x2E;
            break;
          case 'v':
            intValue |= 0x2F;
            break;
          case 'w':
            intValue |= 0x30;
            break;
          case 'x':
            intValue |= 0x31;
            break;
          case 'y':
            intValue |= 0x32;
            break;
          case 'z':
            intValue |= 0x33;
            break;
          case '0':
            intValue |= 0x34;
            break;
          case '1':
            intValue |= 0x35;
            break;
          case '2':
            intValue |= 0x36;
            break;
          case '3':
            intValue |= 0x37;
            break;
          case '4':
            intValue |= 0x38;
            break;
          case '5':
            intValue |= 0x39;
            break;
          case '6':
            intValue |= 0x3A;
            break;
          case '7':
            intValue |= 0x3B;
            break;
          case '8':
            intValue |= 0x3C;
            break;
          case '9':
            intValue |= 0x3D;
            break;
          case '-':
            intValue |= 0x3E;
            break;
          case '_':
            intValue |= 0x3F;
            break;
          case '=':
          case '%':
            switch ((stringPos-1) % 4)
            {
              case 2:
                // The string should have two padding tokens, so only a single
                // byte of data remains.  Shift the value ten bits to the right
                // and read that single byte.
                intValue >>= 10;
                buffer.append((byte) (intValue & 0xFF));
                break decodeLoop;
              case 3:
                // The string should have a single padding token, so two bytes
                // of data remain.  Shift the value eight bits to the right and
                // read those two bytes.
                intValue >>= 8;
                buffer.append((byte) ((intValue >> 8) & 0xFF));
                buffer.append((byte) (intValue & 0xFF));
                break decodeLoop;
            }
            // If we've gotten here, then that must mean the string had padding
            // when none was needed, or it had an invalid length.  That's an
            // error.
            throw new ParseException(ERR_BASE64_URLDECODE_INVALID_LENGTH.get(),
                 (stringPos-1));
          default:
            throw new ParseException(
                 ERR_BASE64_DECODE_UNEXPECTED_CHAR.get(
                      data.charAt(stringPos-1)),
                 (stringPos-1));
        }
      }
      buffer.append((byte) ((intValue >> 16) & 0xFF));
      buffer.append((byte) ((intValue >> 8) & 0xFF));
      buffer.append((byte) (intValue & 0xFF));
    }
    return buffer.toByteArray();
  }
  /**
   * Decodes the contents of the provided base64-encoded string to a string
   * containing the raw data using the UTF-8 encoding.
   *
   * @param  data  The base64-encoded string to decode.  It must not be
   *               {@code null}.
   *
   * @return  A string containing the decoded data.
   *
   * @throws  ParseException  If the contents of the provided string cannot be
   *                          parsed as base64-encoded data using the UTF-8
   *                          encoding.
   */
  public static String urlDecodeToString(final String data)
         throws ParseException
  {
    ensureNotNull(data);
    final byte[] decodedBytes = urlDecode(data);
    return StaticUtils.toUTF8String(decodedBytes);
  }
}
     © 2015 - 2025 Weber Informatics LLC | Privacy Policy