001 /*
002 * Copyright (C) 2009 The Guava Authors
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 com.google.common.primitives;
018
019 import static com.google.common.base.Preconditions.checkArgument;
020 import static com.google.common.base.Preconditions.checkNotNull;
021
022 import com.google.common.annotations.VisibleForTesting;
023
024 import sun.misc.Unsafe;
025
026 import java.lang.reflect.Field;
027 import java.nio.ByteOrder;
028 import java.security.AccessController;
029 import java.security.PrivilegedAction;
030 import java.util.Comparator;
031
032 /**
033 * Static utility methods pertaining to {@code byte} primitives that interpret
034 * values as <i>unsigned</i> (that is, any negative value {@code b} is treated
035 * as the positive value {@code 256 + b}). The corresponding methods that treat
036 * the values as signed are found in {@link SignedBytes}, and the methods for
037 * which signedness is not an issue are in {@link Bytes}.
038 *
039 * @author Kevin Bourrillion
040 * @author Martin Buchholz
041 * @author Hiroshi Yamauchi
042 * @since 1
043 */
044 public final class UnsignedBytes {
045 private UnsignedBytes() {}
046
047 /**
048 * Returns the value of the given byte as an integer, when treated as
049 * unsigned. That is, returns {@code value + 256} if {@code value} is
050 * negative; {@code value} itself otherwise.
051 *
052 * @since 6
053 */
054 public static int toInt(byte value) {
055 return value & 0xFF;
056 }
057
058 /**
059 * Returns the {@code byte} value that, when treated as unsigned, is equal to
060 * {@code value}, if possible.
061 *
062 * @param value a value between 0 and 255 inclusive
063 * @return the {@code byte} value that, when treated as unsigned, equals
064 * {@code value}
065 * @throws IllegalArgumentException if {@code value} is negative or greater
066 * than 255
067 */
068 public static byte checkedCast(long value) {
069 checkArgument(value >> 8 == 0, "out of range: %s", value);
070 return (byte) value;
071 }
072
073 /**
074 * Returns the {@code byte} value that, when treated as unsigned, is nearest
075 * in value to {@code value}.
076 *
077 * @param value any {@code long} value
078 * @return {@code (byte) 255} if {@code value >= 255}, {@code (byte) 0} if
079 * {@code value <= 0}, and {@code value} cast to {@code byte} otherwise
080 */
081 public static byte saturatedCast(long value) {
082 if (value > 255) {
083 return (byte) 255; // -1
084 }
085 if (value < 0) {
086 return (byte) 0;
087 }
088 return (byte) value;
089 }
090
091 /**
092 * Compares the two specified {@code byte} values, treating them as unsigned
093 * values between 0 and 255 inclusive. For example, {@code (byte) -127} is
094 * considered greater than {@code (byte) 127} because it is seen as having
095 * the value of positive {@code 129}.
096 *
097 * @param a the first {@code byte} to compare
098 * @param b the second {@code byte} to compare
099 * @return a negative value if {@code a} is less than {@code b}; a positive
100 * value if {@code a} is greater than {@code b}; or zero if they are equal
101 */
102 public static int compare(byte a, byte b) {
103 return toInt(a) - toInt(b);
104 }
105
106 /**
107 * Returns the least value present in {@code array}.
108 *
109 * @param array a <i>nonempty</i> array of {@code byte} values
110 * @return the value present in {@code array} that is less than or equal to
111 * every other value in the array
112 * @throws IllegalArgumentException if {@code array} is empty
113 */
114 public static byte min(byte... array) {
115 checkArgument(array.length > 0);
116 int min = toInt(array[0]);
117 for (int i = 1; i < array.length; i++) {
118 int next = toInt(array[i]);
119 if (next < min) {
120 min = next;
121 }
122 }
123 return (byte) min;
124 }
125
126 /**
127 * Returns the greatest value present in {@code array}.
128 *
129 * @param array a <i>nonempty</i> array of {@code byte} values
130 * @return the value present in {@code array} that is greater than or equal
131 * to every other value in the array
132 * @throws IllegalArgumentException if {@code array} is empty
133 */
134 public static byte max(byte... array) {
135 checkArgument(array.length > 0);
136 int max = toInt(array[0]);
137 for (int i = 1; i < array.length; i++) {
138 int next = toInt(array[i]);
139 if (next > max) {
140 max = next;
141 }
142 }
143 return (byte) max;
144 }
145
146 /**
147 * Returns a string containing the supplied {@code byte} values separated by
148 * {@code separator}. For example, {@code join(":", (byte) 1, (byte) 2,
149 * (byte) 255)} returns the string {@code "1:2:255"}.
150 *
151 * @param separator the text that should appear between consecutive values in
152 * the resulting string (but not at the start or end)
153 * @param array an array of {@code byte} values, possibly empty
154 */
155 public static String join(String separator, byte... array) {
156 checkNotNull(separator);
157 if (array.length == 0) {
158 return "";
159 }
160
161 // For pre-sizing a builder, just get the right order of magnitude
162 StringBuilder builder = new StringBuilder(array.length * 5);
163 builder.append(toInt(array[0]));
164 for (int i = 1; i < array.length; i++) {
165 builder.append(separator).append(toInt(array[i]));
166 }
167 return builder.toString();
168 }
169
170 /**
171 * Returns a comparator that compares two {@code byte} arrays
172 * lexicographically. That is, it compares, using {@link
173 * #compare(byte, byte)}), the first pair of values that follow any common
174 * prefix, or when one array is a prefix of the other, treats the shorter
175 * array as the lesser. For example, {@code [] < [0x01] < [0x01, 0x7F] <
176 * [0x01, 0x80] < [0x02]}. Values are treated as unsigned.
177 *
178 * <p>The returned comparator is inconsistent with {@link
179 * Object#equals(Object)} (since arrays support only identity equality), but
180 * it is consistent with {@link java.util.Arrays#equals(byte[], byte[])}.
181 *
182 * @see <a href="http://en.wikipedia.org/wiki/Lexicographical_order">
183 * Lexicographical order article at Wikipedia</a>
184 * @since 2
185 */
186 public static Comparator<byte[]> lexicographicalComparator() {
187 return LexicographicalComparatorHolder.BEST_COMPARATOR;
188 }
189
190 @VisibleForTesting
191 static Comparator<byte[]> lexicographicalComparatorJavaImpl() {
192 return LexicographicalComparatorHolder.PureJavaComparator.INSTANCE;
193 }
194
195 /**
196 * Provides a lexicographical comparator implementation; either a Java
197 * implementation or a faster implementation based on {@link Unsafe}.
198 *
199 * <p>Uses reflection to gracefully fall back to the Java implementation if
200 * {@code Unsafe} isn't available.
201 */
202 @VisibleForTesting
203 static class LexicographicalComparatorHolder {
204 static final String UNSAFE_COMPARATOR_NAME =
205 LexicographicalComparatorHolder.class.getName() + "$UnsafeComparator";
206
207 static final Comparator<byte[]> BEST_COMPARATOR = getBestComparator();
208
209 @SuppressWarnings("unused") // only access this class via reflection!
210 enum UnsafeComparator implements Comparator<byte[]> {
211 INSTANCE;
212
213 static final boolean littleEndian =
214 ByteOrder.nativeOrder().equals(ByteOrder.LITTLE_ENDIAN);
215
216 /*
217 * The following static final fields exist for performance reasons.
218 *
219 * In UnsignedBytesBenchmark, accessing the following objects via static
220 * final fields is the fastest (more than twice as fast as the Java
221 * implementation, vs ~1.5x with non-final static fields, on x86_32)
222 * under the Hotspot server compiler. The reason is obviously that the
223 * non-final fields need to be reloaded inside the loop.
224 *
225 * And, no, defining (final or not) local variables out of the loop still
226 * isn't as good because the null check on the theUnsafe object remains
227 * inside the loop and BYTE_ARRAY_BASE_OFFSET doesn't get
228 * constant-folded.
229 *
230 * The compiler can treat static final fields as compile-time constants
231 * and can constant-fold them while (final or not) local variables are
232 * run time values.
233 */
234
235 static final Unsafe theUnsafe;
236
237 /** The offset to the first element in a byte array. */
238 static final int BYTE_ARRAY_BASE_OFFSET;
239
240 static {
241 theUnsafe = (Unsafe) AccessController.doPrivileged(
242 new PrivilegedAction<Object>() {
243 @Override
244 public Object run() {
245 try {
246 Field f = Unsafe.class.getDeclaredField("theUnsafe");
247 f.setAccessible(true);
248 return f.get(null);
249 } catch (NoSuchFieldException e) {
250 // It doesn't matter what we throw;
251 // it's swallowed in getBestComparator().
252 throw new Error();
253 } catch (IllegalAccessException e) {
254 throw new Error();
255 }
256 }
257 });
258
259 BYTE_ARRAY_BASE_OFFSET = theUnsafe.arrayBaseOffset(byte[].class);
260
261 // sanity check - this should never fail
262 if (theUnsafe.arrayIndexScale(byte[].class) != 1) {
263 throw new AssertionError();
264 }
265 }
266
267 /**
268 * Returns true if x1 is less than x2, when both values are treated as
269 * unsigned.
270 */
271 // TODO(kevinb): Should be a common method in primitives.UnsignedLongs.
272 static boolean lessThanUnsigned(long x1, long x2) {
273 return (x1 + Long.MIN_VALUE) < (x2 + Long.MIN_VALUE);
274 }
275
276 @Override public int compare(byte[] left, byte[] right) {
277 int minLength = Math.min(left.length, right.length);
278 int minWords = minLength / Longs.BYTES;
279
280 /*
281 * Compare 8 bytes at a time. Benchmarking shows comparing 8 bytes at a
282 * time is no slower than comparing 4 bytes at a time even on 32-bit.
283 * On the other hand, it is substantially faster on 64-bit.
284 */
285 for (int i = 0; i < minWords * Longs.BYTES; i += Longs.BYTES) {
286 long lw = theUnsafe.getLong(left, BYTE_ARRAY_BASE_OFFSET + (long) i);
287 long rw = theUnsafe.getLong(right, BYTE_ARRAY_BASE_OFFSET + (long) i);
288 long diff = lw ^ rw;
289
290 if (diff != 0) {
291 if (!littleEndian) {
292 return lessThanUnsigned(lw, rw) ? -1 : 1;
293 }
294
295 // Use binary search
296 int n = 0;
297 int y;
298 int x = (int) diff;
299 if (x == 0) {
300 x = (int) (diff >>> 32);
301 n = 32;
302 }
303
304 y = x << 16;
305 if (y == 0) {
306 n += 16;
307 } else {
308 x = y;
309 }
310
311 y = x << 8;
312 if (y == 0) {
313 n += 8;
314 }
315 return (int) (((lw >>> n) & 0xFFL) - ((rw >>> n) & 0xFFL));
316 }
317 }
318
319 // The epilogue to cover the last (minLength % 8) elements.
320 for (int i = minWords * Longs.BYTES; i < minLength; i++) {
321 int result = UnsignedBytes.compare(left[i], right[i]);
322 if (result != 0) {
323 return result;
324 }
325 }
326 return left.length - right.length;
327 }
328 }
329
330 enum PureJavaComparator implements Comparator<byte[]> {
331 INSTANCE;
332
333 @Override public int compare(byte[] left, byte[] right) {
334 int minLength = Math.min(left.length, right.length);
335 for (int i = 0; i < minLength; i++) {
336 int result = UnsignedBytes.compare(left[i], right[i]);
337 if (result != 0) {
338 return result;
339 }
340 }
341 return left.length - right.length;
342 }
343 }
344
345 /**
346 * Returns the Unsafe-using Comparator, or falls back to the pure-Java
347 * implementation if unable to do so.
348 */
349 static Comparator<byte[]> getBestComparator() {
350 try {
351 Class<?> theClass = Class.forName(UNSAFE_COMPARATOR_NAME);
352
353 // yes, UnsafeComparator does implement Comparator<byte[]>
354 @SuppressWarnings("unchecked")
355 Comparator<byte[]> comparator =
356 (Comparator<byte[]>) theClass.getEnumConstants()[0];
357 return comparator;
358 } catch (Throwable t) { // ensure we really catch *everything*
359 return lexicographicalComparatorJavaImpl();
360 }
361 }
362 }
363 }