001 /*
002 * Copyright 2010-2015 JetBrains s.r.o.
003 *
004 * Licensed under the Apache License, Version 2.0 (the "License");
005 * you may not use this file except in compliance with the License.
006 * You may obtain a copy of the License at
007 *
008 * http://www.apache.org/licenses/LICENSE-2.0
009 *
010 * Unless required by applicable law or agreed to in writing, software
011 * distributed under the License is distributed on an "AS IS" BASIS,
012 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
013 * See the License for the specific language governing permissions and
014 * limitations under the License.
015 */
016
017 package org.jetbrains.kotlin.serialization.jvm;
018
019 import org.jetbrains.annotations.NotNull;
020
021 import java.util.ArrayList;
022 import java.util.List;
023
024 import static org.jetbrains.kotlin.serialization.jvm.UtfEncodingKt.MAX_UTF8_INFO_LENGTH;
025
026 public class BitEncoding {
027 private static final boolean FORCE_8TO7_ENCODING = "true".equals(System.getProperty("kotlin.jvm.serialization.use8to7"));
028
029 private static final char _8TO7_MODE_MARKER = (char) -1;
030
031 private BitEncoding() {
032 }
033
034 /**
035 * Converts a byte array of serialized data to an array of {@code String} satisfying JVM annotation value argument restrictions:
036 * <ol>
037 * <li>Each string's length should be no more than 65535</li>
038 * <li>UTF-8 representation of each string cannot contain bytes in the range 0xf0..0xff</li>
039 * </ol>
040 */
041 @NotNull
042 public static String[] encodeBytes(@NotNull byte[] data) {
043 // TODO: try both encodings here and choose the best one (with the smallest size)
044 if (!FORCE_8TO7_ENCODING) {
045 return UtfEncodingKt.bytesToStrings(data);
046 }
047 byte[] bytes = encode8to7(data);
048 // Since 0x0 byte is encoded as two bytes in the Modified UTF-8 (0xc0 0x80) and zero is rather common to byte arrays, we increment
049 // every byte by one modulo max byte value, so that the less common value 0x7f will be represented as two bytes instead.
050 addModuloByte(bytes, 1);
051 return splitBytesToStringArray(bytes);
052 }
053
054 /**
055 * Converts a byte array to another byte array, every element of which is in the range 0x0..0x7f.
056 *
057 * The conversion is equivalent to the following: input bytes are combined into one long bit string. This big string is then split into
058 * groups of 7 bits. Each resulting 7-bit chunk is then converted to a byte (with a leading bit = 0). The last chunk may have less than
059 * 7 bits, it's prepended with zeros to form a byte. The result is then the array of these bytes, each of which is obviously in the
060 * range 0x0..0x7f.
061 *
062 * Suppose the input of 4 bytes is given (bytes are listed from the beginning to the end, each byte from the least significant bit to
063 * the most significant bit, bits within each byte are numbered):
064 *
065 * 01234567 01234567 01234567 01234567
066 *
067 * The output for this kind of input will be of the following form ('#' represents a zero bit):
068 *
069 * 0123456# 7012345# 6701234# 5670123# 4567####
070 */
071 @NotNull
072 private static byte[] encode8to7(@NotNull byte[] data) {
073 // ceil(data.length * 8 / 7)
074 int resultLength = (data.length * 8 + 6) / 7;
075 byte[] result = new byte[resultLength];
076
077 // We maintain a pointer to the bit in the input, which is represented by two numbers: index of the current byte in the input and
078 // the index of a bit inside this byte (0 is least significant, 7 is most significant)
079 int byteIndex = 0;
080 int bit = 0;
081
082 // Write all resulting bytes except the last one. To do this we need to collect exactly 7 bits, starting from the current, into a
083 // byte. In almost all cases these 7 bits can be collected from two parts: the first is several (at least one) most significant bits
084 // from the current byte, the second is several (maybe zero) least significant bits from the next byte. The special case is when the
085 // current bit is the first (least significant) bit in its byte (bit == 0): then the 7 needed bits are just the 7 least significant
086 // of the current byte.
087 for (int i = 0; i < resultLength - 1; i++) {
088 if (bit == 0) {
089 result[i] = (byte) (data[byteIndex] & 0x7f);
090 bit = 7;
091 continue;
092 }
093
094 int firstPart = (data[byteIndex] & 0xff) >>> bit;
095 int newBit = (bit + 7) & 7;
096 int secondPart = (data[++byteIndex] & ((1 << newBit) - 1)) << 8 - bit;
097 result[i] = (byte) (firstPart + secondPart);
098 bit = newBit;
099 }
100
101 // Write the last byte, which is just several most significant bits of the last byte in the input, padded with zeros
102 if (resultLength > 0) {
103 assert bit != 0 : "The last chunk cannot start from the input byte since otherwise at least one bit will remain unprocessed";
104 assert byteIndex == data.length - 1 : "The last 7-bit chunk should be encoded from the last input byte: " +
105 byteIndex + " != " + (data.length - 1);
106 result[resultLength - 1] = (byte) ((data[byteIndex] & 0xff) >>> bit);
107 }
108
109 return result;
110 }
111
112 private static void addModuloByte(@NotNull byte[] data, int increment) {
113 for (int i = 0, n = data.length; i < n; i++) {
114 data[i] = (byte) ((data[i] + increment) & 0x7f);
115 }
116 }
117
118 /**
119 * Converts a big byte array into the array of strings, where each string, when written to the constant pool table in bytecode, produces
120 * a byte array of not more than MAX_UTF8_INFO_LENGTH. Each byte, except those which are 0x0, occupies exactly one byte in the constant
121 * pool table. Zero bytes occupy two bytes in the table each.
122 *
123 * When strings are constructed from the array of bytes here, they are encoded in the platform's default encoding. This is fine: the
124 * conversion to the Modified UTF-8 (which here would be equivalent to replacing each 0x0 with 0xc0 0x80) will happen later by ASM, when
125 * it writes these strings to the bytecode
126 */
127 @NotNull
128 private static String[] splitBytesToStringArray(@NotNull byte[] data) {
129 List<String> result = new ArrayList<String>();
130
131 // The offset where the currently processed string starts
132 int off = 0;
133
134 // The effective length the bytes of the current string would occupy in the constant pool table.
135 // 2 because the first char is -1 which denotes the encoding mode and occupies two bytes in Modified UTF-8
136 int len = 2;
137
138 boolean encodingModeAdded = false;
139
140 for (int i = 0, n = data.length; i < n; i++) {
141 // When the effective length reaches at least MAX - 1, we add the current string to the result. Note that the effective length
142 // is at most MAX here: non-zero bytes occupy 1 byte and zero bytes occupy 2 bytes, so we couldn't jump over more than one byte
143 if (len >= MAX_UTF8_INFO_LENGTH - 1) {
144 assert len <= MAX_UTF8_INFO_LENGTH : "Produced strings cannot contain more than " + MAX_UTF8_INFO_LENGTH + " bytes: " + len;
145 String string = new String(data, off, i - off);
146 if (!encodingModeAdded) {
147 encodingModeAdded = true;
148 result.add(_8TO7_MODE_MARKER + string);
149 }
150 else {
151 result.add(string);
152 }
153 off = i;
154 len = 0;
155 }
156
157 if (data[i] == 0) {
158 len += 2;
159 }
160 else {
161 len++;
162 }
163 }
164
165 if (len >= 0) {
166 result.add(new String(data, off, data.length - off));
167 }
168
169 return result.toArray(new String[result.size()]);
170 }
171
172 /**
173 * Converts encoded array of {@code String} obtained by {@link BitEncoding#encodeBytes(byte[])} back to a byte array.
174 */
175 @NotNull
176 public static byte[] decodeBytes(@NotNull String[] data) {
177 if (data.length > 0 && !data[0].isEmpty()) {
178 char possibleMarker = data[0].charAt(0);
179 if (possibleMarker == UtfEncodingKt.UTF8_MODE_MARKER) {
180 return UtfEncodingKt.stringsToBytes(dropMarker(data));
181 }
182 if (possibleMarker == _8TO7_MODE_MARKER) {
183 data = dropMarker(data);
184 }
185 }
186
187 byte[] bytes = combineStringArrayIntoBytes(data);
188 // Adding 0x7f modulo max byte value is equivalent to subtracting 1 the same modulo, which is inverse to what happens in encodeBytes
189 addModuloByte(bytes, 0x7f);
190 return decode7to8(bytes);
191 }
192
193 @NotNull
194 private static String[] dropMarker(@NotNull String[] data) {
195 // Clone because the clients should be able to use the passed array for their own purposes.
196 // This is cheap because the size of the array is 1 or 2 almost always.
197 String[] result = data.clone();
198 result[0] = result[0].substring(1);
199 return result;
200 }
201
202 /**
203 * Combines the array of strings resulted from encodeBytes() into one long byte array
204 */
205 @NotNull
206 private static byte[] combineStringArrayIntoBytes(@NotNull String[] data) {
207 int resultLength = 0;
208 for (String s : data) {
209 assert s.length() <= MAX_UTF8_INFO_LENGTH : "String is too long: " + s.length();
210 resultLength += s.length();
211 }
212
213 byte[] result = new byte[resultLength];
214 int p = 0;
215 for (String s : data) {
216 for (int i = 0, n = s.length(); i < n; i++) {
217 result[p++] = (byte) s.charAt(i);
218 }
219 }
220
221 return result;
222 }
223
224 /**
225 * Decodes the byte array resulted from encode8to7().
226 *
227 * Each byte of the input array has at most 7 valuable bits of information. So the decoding is equivalent to the following: least
228 * significant 7 bits of all input bytes are combined into one long bit string. This bit string is then split into groups of 8 bits,
229 * each of which forms a byte in the output. If there are any leftovers, they are ignored, since they were added just as a padding and
230 * do not comprise a full byte.
231 *
232 * Suppose the following encoded byte array is given (bits are numbered the same way as in encode8to7() doc):
233 *
234 * 01234567 01234567 01234567 01234567
235 *
236 * The output of the following form would be produced:
237 *
238 * 01234560 12345601 23456012
239 *
240 * Note how all most significant bits and leftovers are dropped, since they don't contain any useful information
241 */
242 @NotNull
243 private static byte[] decode7to8(@NotNull byte[] data) {
244 // floor(7 * data.length / 8)
245 int resultLength = 7 * data.length / 8;
246
247 byte[] result = new byte[resultLength];
248
249 // We maintain a pointer to an input bit in the same fashion as in encode8to7(): it's represented as two numbers: index of the
250 // current byte in the input and index of the bit in the byte
251 int byteIndex = 0;
252 int bit = 0;
253
254 // A resulting byte is comprised of 8 bits, starting from the current bit. Since each input byte only "contains 7 bytes", a
255 // resulting byte always consists of two parts: several most significant bits of the current byte and several least significant bits
256 // of the next byte
257 for (int i = 0; i < resultLength; i++) {
258 int firstPart = (data[byteIndex] & 0xff) >>> bit;
259 byteIndex++;
260 int secondPart = (data[byteIndex] & ((1 << (bit + 1)) - 1)) << 7 - bit;
261 result[i] = (byte) (firstPart + secondPart);
262
263 if (bit == 6) {
264 byteIndex++;
265 bit = 0;
266 }
267 else {
268 bit++;
269 }
270 }
271
272 return result;
273 }
274 }