📄 base64.java
字号:
// GZip? if( gzip == GZIP ) { gzos = new java.util.zip.GZIPOutputStream( b64os ); oos = new java.io.ObjectOutputStream( gzos ); } // end if: gzip else oos = new java.io.ObjectOutputStream( b64os ); oos.writeObject( serializableObject ); } // end try catch( java.io.IOException e ) { e.printStackTrace(); return null; } // end catch finally { try{ oos.close(); } catch( Exception e ){} try{ gzos.close(); } catch( Exception e ){} try{ b64os.close(); } catch( Exception e ){} try{ baos.close(); } catch( Exception e ){} } // end finally // Return value according to relevant encoding. try { return new String( baos.toByteArray(), PREFERRED_ENCODING ); } // end try catch (java.io.UnsupportedEncodingException uue) { return new String( baos.toByteArray() ); } // end catch } // end encode /** * Encodes a byte array into Base64 notation. * Does not GZip-compress data. * * @param source The data to convert * @since 1.4 */ public static String encodeBytes( byte[] source ) { return encodeBytes( source, 0, source.length, NO_OPTIONS ); } // end encodeBytes /** * Encodes a byte array into Base64 notation. * <p> * Valid options:<pre> * GZIP: gzip-compresses object before encoding it. * DONT_BREAK_LINES: don't break lines at 76 characters * <i>Note: Technically, this makes your encoding non-compliant.</i> * </pre> * <p> * Example: <code>encodeBytes( myData, Base64.GZIP )</code> or * <p> * Example: <code>encodeBytes( myData, Base64.GZIP | Base64.DONT_BREAK_LINES )</code> * * * @param source The data to convert * @param options Specified options * @see Base64#GZIP * @see Base64#DONT_BREAK_LINES * @since 2.0 */ public static String encodeBytes( byte[] source, int options ) { return encodeBytes( source, 0, source.length, options ); } // end encodeBytes /** * Encodes a byte array into Base64 notation. * Does not GZip-compress data. * * @param source The data to convert * @param off Offset in array where conversion should begin * @param len Length of data to convert * @since 1.4 */ public static String encodeBytes( byte[] source, int off, int len ) { return encodeBytes( source, off, len, NO_OPTIONS ); } // end encodeBytes /** * Encodes a byte array into Base64 notation. * <p> * Valid options:<pre> * GZIP: gzip-compresses object before encoding it. * DONT_BREAK_LINES: don't break lines at 76 characters * <i>Note: Technically, this makes your encoding non-compliant.</i> * </pre> * <p> * Example: <code>encodeBytes( myData, Base64.GZIP )</code> or * <p> * Example: <code>encodeBytes( myData, Base64.GZIP | Base64.DONT_BREAK_LINES )</code> * * * @param source The data to convert * @param off Offset in array where conversion should begin * @param len Length of data to convert * @param options Specified options * @see Base64#GZIP * @see Base64#DONT_BREAK_LINES * @since 2.0 */ public static String encodeBytes( byte[] source, int off, int len, int options ) { // Isolate options int dontBreakLines = ( options & DONT_BREAK_LINES ); int gzip = ( options & GZIP ); // Compress? if( gzip == GZIP ) { java.io.ByteArrayOutputStream baos = null; java.util.zip.GZIPOutputStream gzos = null; Base64.OutputStream b64os = null; try { // GZip -> Base64 -> ByteArray baos = new java.io.ByteArrayOutputStream(); b64os = new Base64.OutputStream( baos, ENCODE | dontBreakLines ); gzos = new java.util.zip.GZIPOutputStream( b64os ); gzos.write( source, off, len ); gzos.close(); } // end try catch( java.io.IOException e ) { e.printStackTrace(); return null; } // end catch finally { try{ gzos.close(); } catch( Exception e ){} try{ b64os.close(); } catch( Exception e ){} try{ baos.close(); } catch( Exception e ){} } // end finally // Return value according to relevant encoding. try { return new String( baos.toByteArray(), PREFERRED_ENCODING ); } // end try catch (java.io.UnsupportedEncodingException uue) { return new String( baos.toByteArray() ); } // end catch } // end if: compress // Else, don't compress. Better not to use streams at all then. else { // Convert option to boolean in way that code likes it. boolean breakLines = dontBreakLines == 0; int len43 = len * 4 / 3; byte[] outBuff = new byte[ ( len43 ) // Main 4:3 + ( (len % 3) > 0 ? 4 : 0 ) // Account for padding + (breakLines ? ( len43 / MAX_LINE_LENGTH ) : 0) ]; // New lines int d = 0; int e = 0; int len2 = len - 2; int lineLength = 0; for( ; d < len2; d+=3, e+=4 ) { encode3to4( source, d+off, 3, outBuff, e ); lineLength += 4; if( breakLines && lineLength == MAX_LINE_LENGTH ) { outBuff[e+4] = NEW_LINE; e++; lineLength = 0; } // end if: end of line } // en dfor: each piece of array if( d < len ) { encode3to4( source, d+off, len - d, outBuff, e ); e += 4; } // end if: some padding needed // Return value according to relevant encoding. try { return new String( outBuff, 0, e, PREFERRED_ENCODING ); } // end try catch (java.io.UnsupportedEncodingException uue) { return new String( outBuff, 0, e ); } // end catch } // end else: don't compress } // end encodeBytes /* ******** D E C O D I N G M E T H O D S ******** */ /** * Decodes four bytes from array <var>source</var> * and writes the resulting bytes (up to three of them) * to <var>destination</var>. * The source and destination arrays can be manipulated * anywhere along their length by specifying * <var>srcOffset</var> and <var>destOffset</var>. * This method does not check to make sure your arrays * are large enough to accomodate <var>srcOffset</var> + 4 for * the <var>source</var> array or <var>destOffset</var> + 3 for * the <var>destination</var> array. * This method returns the actual number of bytes that * were converted from the Base64 encoding. * * * @param source the array to convert * @param srcOffset the index where conversion begins * @param destination the array to hold the conversion * @param destOffset the index where output will be put * @return the number of decoded bytes converted * @since 1.3 */ private static int decode4to3( byte[] source, int srcOffset, byte[] destination, int destOffset ) { // Example: Dk== if( source[ srcOffset + 2] == EQUALS_SIGN ) { // Two ways to do the same thing. Don't know which way I like best. //int outBuff = ( ( DECODABET[ source[ srcOffset ] ] << 24 ) >>> 6 ) // | ( ( DECODABET[ source[ srcOffset + 1] ] << 24 ) >>> 12 ); int outBuff = ( ( DECODABET[ source[ srcOffset ] ] & 0xFF ) << 18 ) | ( ( DECODABET[ source[ srcOffset + 1] ] & 0xFF ) << 12 ); destination[ destOffset ] = (byte)( outBuff >>> 16 ); return 1; } // Example: DkL= else if( source[ srcOffset + 3 ] == EQUALS_SIGN ) { // Two ways to do the same thing. Don't know which way I like best. //int outBuff = ( ( DECODABET[ source[ srcOffset ] ] << 24 ) >>> 6 ) // | ( ( DECODABET[ source[ srcOffset + 1 ] ] << 24 ) >>> 12 ) // | ( ( DECODABET[ source[ srcOffset + 2 ] ] << 24 ) >>> 18 ); int outBuff = ( ( DECODABET[ source[ srcOffset ] ] & 0xFF ) << 18 ) | ( ( DECODABET[ source[ srcOffset + 1 ] ] & 0xFF ) << 12 ) | ( ( DECODABET[ source[ srcOffset + 2 ] ] & 0xFF ) << 6 ); destination[ destOffset ] = (byte)( outBuff >>> 16 ); destination[ destOffset + 1 ] = (byte)( outBuff >>> 8 ); return 2; } // Example: DkLE else { try{ // Two ways to do the same thing. Don't know which way I like best. //int outBuff = ( ( DECODABET[ source[ srcOffset ] ] << 24 ) >>> 6 ) // | ( ( DECODABET[ source[ srcOffset + 1 ] ] << 24 ) >>> 12 ) // | ( ( DECODABET[ source[ srcOffset + 2 ] ] << 24 ) >>> 18 ) // | ( ( DECODABET[ source[ srcOffset + 3 ] ] << 24 ) >>> 24 ); int outBuff = ( ( DECODABET[ source[ srcOffset ] ] & 0xFF ) << 18 ) | ( ( DECODABET[ source[ srcOffset + 1 ] ] & 0xFF ) << 12 ) | ( ( DECODABET[ source[ srcOffset + 2 ] ] & 0xFF ) << 6) | ( ( DECODABET[ source[ srcOffset + 3 ] ] & 0xFF ) ); destination[ destOffset ] = (byte)( outBuff >> 16 ); destination[ destOffset + 1 ] = (byte)( outBuff >> 8 ); destination[ destOffset + 2 ] = (byte)( outBuff ); return 3; }catch( Exception e){ System.out.println(""+source[srcOffset]+ ": " + ( DECODABET[ source[ srcOffset ] ] ) ); System.out.println(""+source[srcOffset+1]+ ": " + ( DECODABET[ source[ srcOffset + 1 ] ] ) ); System.out.println(""+source[srcOffset+2]+ ": " + ( DECODABET[ source[ srcOffset + 2 ] ] ) ); System.out.println(""+source[srcOffset+3]+ ": " + ( DECODABET[ source[ srcOffset + 3 ] ] ) ); return -1; } //e nd catch } } // end decodeToBytes /** * Very low-level access to decoding ASCII characters in * the form of a byte array. Does not support automatically * gunzipping or any other "fancy" features. * * @param source The Base64 encoded data * @param off The offset of where to begin decoding * @param len The length of characters to decode * @return decoded data * @since 1.3 */ public static byte[] decode( byte[] source, int off, int len ) { int len34 = len * 3 / 4; byte[] outBuff = new byte[ len34 ]; // Upper limit on size of output int outBuffPosn = 0; byte[] b4 = new byte[4]; int b4Posn = 0; int i = 0; byte sbiCrop = 0; byte sbiDecode = 0; for( i = off; i < off+len; i++ ) { sbiCrop = (byte)(source[i] & 0x7f); // Only the low seven bits sbiDecode = DECODABET[ sbiCrop ]; if( sbiDecode >= WHITE_SPACE_ENC ) // White space, Equals sign or better { if( sbiDecode >= EQUALS_SIGN_ENC ) { b4[ b4Posn++ ] = sbiCrop; if( b4Posn > 3 ) { outBuffPosn += decode4to3( b4, 0, outBuff, outBuffPosn ); b4Posn = 0; // If that was the equals sign, break out of 'for' loop if( sbiCrop == EQUALS_SIGN ) break; } // end if: quartet built } // end if: equals sign or better } // end if: white space, equals sign or better else { System.err.println( "Bad Base64 input character at " + i + ": " + source[i] + "(decimal)" ); return null; } // end else: } // each input character byte[] out = new byte[ outBuffPosn ]; System.arraycopy( outBuff, 0, out, 0, outBuffPosn ); return out; } // end decode /** * Decodes data from Base64 notation, automatically * detecting gzip-compressed data and decompressing it. * * @param s the string to decode * @return the decoded data
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -