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 * <p>See the Guava User Guide article on <a href=
040 * "http://code.google.com/p/guava-libraries/wiki/PrimitivesExplained">
041 * primitive utilities</a>.
042 *
043 * @author Kevin Bourrillion
044 * @author Martin Buchholz
045 * @author Hiroshi Yamauchi
046 * @since 1.0
047 */
048 public final class UnsignedBytes {
049 private UnsignedBytes() {}
050
051 /**
052 * The largest power of two that can be represented as an unsigned {@code byte}.
053 *
054 * @since 10.0
055 */
056 public static final byte MAX_POWER_OF_TWO = (byte) (1 << 7);
057
058 /**
059 * Returns the value of the given byte as an integer, when treated as
060 * unsigned. That is, returns {@code value + 256} if {@code value} is
061 * negative; {@code value} itself otherwise.
062 *
063 * @since 6.0
064 */
065 public static int toInt(byte value) {
066 return value & 0xFF;
067 }
068
069 /**
070 * Returns the {@code byte} value that, when treated as unsigned, is equal to
071 * {@code value}, if possible.
072 *
073 * @param value a value between 0 and 255 inclusive
074 * @return the {@code byte} value that, when treated as unsigned, equals
075 * {@code value}
076 * @throws IllegalArgumentException if {@code value} is negative or greater
077 * than 255
078 */
079 public static byte checkedCast(long value) {
080 checkArgument(value >> 8 == 0, "out of range: %s", value);
081 return (byte) value;
082 }
083
084 /**
085 * Returns the {@code byte} value that, when treated as unsigned, is nearest
086 * in value to {@code value}.
087 *
088 * @param value any {@code long} value
089 * @return {@code (byte) 255} if {@code value >= 255}, {@code (byte) 0} if
090 * {@code value <= 0}, and {@code value} cast to {@code byte} otherwise
091 */
092 public static byte saturatedCast(long value) {
093 if (value > 255) {
094 return (byte) 255; // -1
095 }
096 if (value < 0) {
097 return (byte) 0;
098 }
099 return (byte) value;
100 }
101
102 /**
103 * Compares the two specified {@code byte} values, treating them as unsigned
104 * values between 0 and 255 inclusive. For example, {@code (byte) -127} is
105 * considered greater than {@code (byte) 127} because it is seen as having
106 * the value of positive {@code 129}.
107 *
108 * @param a the first {@code byte} to compare
109 * @param b the second {@code byte} to compare
110 * @return a negative value if {@code a} is less than {@code b}; a positive
111 * value if {@code a} is greater than {@code b}; or zero if they are equal
112 */
113 public static int compare(byte a, byte b) {
114 return toInt(a) - toInt(b);
115 }
116
117 /**
118 * Returns the least value present in {@code array}.
119 *
120 * @param array a <i>nonempty</i> array of {@code byte} values
121 * @return the value present in {@code array} that is less than or equal to
122 * every other value in the array
123 * @throws IllegalArgumentException if {@code array} is empty
124 */
125 public static byte min(byte... array) {
126 checkArgument(array.length > 0);
127 int min = toInt(array[0]);
128 for (int i = 1; i < array.length; i++) {
129 int next = toInt(array[i]);
130 if (next < min) {
131 min = next;
132 }
133 }
134 return (byte) min;
135 }
136
137 /**
138 * Returns the greatest value present in {@code array}.
139 *
140 * @param array a <i>nonempty</i> array of {@code byte} values
141 * @return the value present in {@code array} that is greater than or equal
142 * to every other value in the array
143 * @throws IllegalArgumentException if {@code array} is empty
144 */
145 public static byte max(byte... array) {
146 checkArgument(array.length > 0);
147 int max = toInt(array[0]);
148 for (int i = 1; i < array.length; i++) {
149 int next = toInt(array[i]);
150 if (next > max) {
151 max = next;
152 }
153 }
154 return (byte) max;
155 }
156
157 /**
158 * Returns a string containing the supplied {@code byte} values separated by
159 * {@code separator}. For example, {@code join(":", (byte) 1, (byte) 2,
160 * (byte) 255)} returns the string {@code "1:2:255"}.
161 *
162 * @param separator the text that should appear between consecutive values in
163 * the resulting string (but not at the start or end)
164 * @param array an array of {@code byte} values, possibly empty
165 */
166 public static String join(String separator, byte... array) {
167 checkNotNull(separator);
168 if (array.length == 0) {
169 return "";
170 }
171
172 // For pre-sizing a builder, just get the right order of magnitude
173 StringBuilder builder = new StringBuilder(array.length * 5);
174 builder.append(toInt(array[0]));
175 for (int i = 1; i < array.length; i++) {
176 builder.append(separator).append(toInt(array[i]));
177 }
178 return builder.toString();
179 }
180
181 /**
182 * Returns a comparator that compares two {@code byte} arrays
183 * lexicographically. That is, it compares, using {@link
184 * #compare(byte, byte)}), the first pair of values that follow any common
185 * prefix, or when one array is a prefix of the other, treats the shorter
186 * array as the lesser. For example, {@code [] < [0x01] < [0x01, 0x7F] <
187 * [0x01, 0x80] < [0x02]}. Values are treated as unsigned.
188 *
189 * <p>The returned comparator is inconsistent with {@link
190 * Object#equals(Object)} (since arrays support only identity equality), but
191 * it is consistent with {@link java.util.Arrays#equals(byte[], byte[])}.
192 *
193 * @see <a href="http://en.wikipedia.org/wiki/Lexicographical_order">
194 * Lexicographical order article at Wikipedia</a>
195 * @since 2.0
196 */
197 public static Comparator<byte[]> lexicographicalComparator() {
198 return LexicographicalComparatorHolder.BEST_COMPARATOR;
199 }
200
201 @VisibleForTesting
202 static Comparator<byte[]> lexicographicalComparatorJavaImpl() {
203 return LexicographicalComparatorHolder.PureJavaComparator.INSTANCE;
204 }
205
206 /**
207 * Provides a lexicographical comparator implementation; either a Java
208 * implementation or a faster implementation based on {@link Unsafe}.
209 *
210 * <p>Uses reflection to gracefully fall back to the Java implementation if
211 * {@code Unsafe} isn't available.
212 */
213 @VisibleForTesting
214 static class LexicographicalComparatorHolder {
215 static final String UNSAFE_COMPARATOR_NAME =
216 LexicographicalComparatorHolder.class.getName() + "$UnsafeComparator";
217
218 static final Comparator<byte[]> BEST_COMPARATOR = getBestComparator();
219
220 @VisibleForTesting
221 enum UnsafeComparator implements Comparator<byte[]> {
222 INSTANCE;
223
224 static final boolean littleEndian =
225 ByteOrder.nativeOrder().equals(ByteOrder.LITTLE_ENDIAN);
226
227 /*
228 * The following static final fields exist for performance reasons.
229 *
230 * In UnsignedBytesBenchmark, accessing the following objects via static
231 * final fields is the fastest (more than twice as fast as the Java
232 * implementation, vs ~1.5x with non-final static fields, on x86_32)
233 * under the Hotspot server compiler. The reason is obviously that the
234 * non-final fields need to be reloaded inside the loop.
235 *
236 * And, no, defining (final or not) local variables out of the loop still
237 * isn't as good because the null check on the theUnsafe object remains
238 * inside the loop and BYTE_ARRAY_BASE_OFFSET doesn't get
239 * constant-folded.
240 *
241 * The compiler can treat static final fields as compile-time constants
242 * and can constant-fold them while (final or not) local variables are
243 * run time values.
244 */
245
246 static final Unsafe theUnsafe;
247
248 /** The offset to the first element in a byte array. */
249 static final int BYTE_ARRAY_BASE_OFFSET;
250
251 static {
252 theUnsafe = (Unsafe) AccessController.doPrivileged(
253 new PrivilegedAction<Object>() {
254 @Override
255 public Object run() {
256 try {
257 Field f = Unsafe.class.getDeclaredField("theUnsafe");
258 f.setAccessible(true);
259 return f.get(null);
260 } catch (NoSuchFieldException e) {
261 // It doesn't matter what we throw;
262 // it's swallowed in getBestComparator().
263 throw new Error();
264 } catch (IllegalAccessException e) {
265 throw new Error();
266 }
267 }
268 });
269
270 BYTE_ARRAY_BASE_OFFSET = theUnsafe.arrayBaseOffset(byte[].class);
271
272 // sanity check - this should never fail
273 if (theUnsafe.arrayIndexScale(byte[].class) != 1) {
274 throw new AssertionError();
275 }
276 }
277
278 @Override public int compare(byte[] left, byte[] right) {
279 int minLength = Math.min(left.length, right.length);
280 int minWords = minLength / Longs.BYTES;
281
282 /*
283 * Compare 8 bytes at a time. Benchmarking shows comparing 8 bytes at a
284 * time is no slower than comparing 4 bytes at a time even on 32-bit.
285 * On the other hand, it is substantially faster on 64-bit.
286 */
287 for (int i = 0; i < minWords * Longs.BYTES; i += Longs.BYTES) {
288 long lw = theUnsafe.getLong(left, BYTE_ARRAY_BASE_OFFSET + (long) i);
289 long rw = theUnsafe.getLong(right, BYTE_ARRAY_BASE_OFFSET + (long) i);
290 long diff = lw ^ rw;
291
292 if (diff != 0) {
293 if (!littleEndian) {
294 return UnsignedLongs.compare(lw, rw);
295 }
296
297 // Use binary search
298 int n = 0;
299 int y;
300 int x = (int) diff;
301 if (x == 0) {
302 x = (int) (diff >>> 32);
303 n = 32;
304 }
305
306 y = x << 16;
307 if (y == 0) {
308 n += 16;
309 } else {
310 x = y;
311 }
312
313 y = x << 8;
314 if (y == 0) {
315 n += 8;
316 }
317 return (int) (((lw >>> n) & 0xFFL) - ((rw >>> n) & 0xFFL));
318 }
319 }
320
321 // The epilogue to cover the last (minLength % 8) elements.
322 for (int i = minWords * Longs.BYTES; i < minLength; i++) {
323 int result = UnsignedBytes.compare(left[i], right[i]);
324 if (result != 0) {
325 return result;
326 }
327 }
328 return left.length - right.length;
329 }
330 }
331
332 enum PureJavaComparator implements Comparator<byte[]> {
333 INSTANCE;
334
335 @Override public int compare(byte[] left, byte[] right) {
336 int minLength = Math.min(left.length, right.length);
337 for (int i = 0; i < minLength; i++) {
338 int result = UnsignedBytes.compare(left[i], right[i]);
339 if (result != 0) {
340 return result;
341 }
342 }
343 return left.length - right.length;
344 }
345 }
346
347 /**
348 * Returns the Unsafe-using Comparator, or falls back to the pure-Java
349 * implementation if unable to do so.
350 */
351 static Comparator<byte[]> getBestComparator() {
352 try {
353 Class<?> theClass = Class.forName(UNSAFE_COMPARATOR_NAME);
354
355 // yes, UnsafeComparator does implement Comparator<byte[]>
356 @SuppressWarnings("unchecked")
357 Comparator<byte[]> comparator =
358 (Comparator<byte[]>) theClass.getEnumConstants()[0];
359 return comparator;
360 } catch (Throwable t) { // ensure we really catch *everything*
361 return lexicographicalComparatorJavaImpl();
362 }
363 }
364 }
365 }
366