001/* 002 * Copyright (C) 2011 The Guava Authors 003 * 004 * Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except 005 * in compliance with the License. You may obtain a copy of the License at 006 * 007 * http://www.apache.org/licenses/LICENSE-2.0 008 * 009 * Unless required by applicable law or agreed to in writing, software distributed under the License 010 * is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express 011 * or implied. See the License for the specific language governing permissions and limitations under 012 * the License. 013 */ 014 015package com.google.common.hash; 016 017import static com.google.common.base.Preconditions.checkArgument; 018import static com.google.common.base.Preconditions.checkNotNull; 019import static java.lang.Math.max; 020 021import com.google.common.annotations.Beta; 022import com.google.common.annotations.VisibleForTesting; 023import com.google.common.base.Objects; 024import com.google.common.base.Predicate; 025import com.google.common.hash.BloomFilterStrategies.LockFreeBitArray; 026import com.google.common.math.DoubleMath; 027import com.google.common.math.LongMath; 028import com.google.common.primitives.SignedBytes; 029import com.google.common.primitives.UnsignedBytes; 030import com.google.errorprone.annotations.CanIgnoreReturnValue; 031import java.io.DataInputStream; 032import java.io.DataOutputStream; 033import java.io.IOException; 034import java.io.InputStream; 035import java.io.InvalidObjectException; 036import java.io.ObjectInputStream; 037import java.io.OutputStream; 038import java.io.Serializable; 039import java.math.RoundingMode; 040import java.util.stream.Collector; 041import javax.annotation.CheckForNull; 042import org.checkerframework.checker.nullness.qual.Nullable; 043 044/** 045 * A Bloom filter for instances of {@code T}. A Bloom filter offers an approximate containment test 046 * with one-sided error: if it claims that an element is contained in it, this might be in error, 047 * but if it claims that an element is <i>not</i> contained in it, then this is definitely true. 048 * 049 * <p>If you are unfamiliar with Bloom filters, this nice <a 050 * href="http://llimllib.github.io/bloomfilter-tutorial/">tutorial</a> may help you understand how 051 * they work. 052 * 053 * <p>The false positive probability ({@code FPP}) of a Bloom filter is defined as the probability 054 * that {@linkplain #mightContain(Object)} will erroneously return {@code true} for an object that 055 * has not actually been put in the {@code BloomFilter}. 056 * 057 * <p>Bloom filters are serializable. They also support a more compact serial representation via the 058 * {@link #writeTo} and {@link #readFrom} methods. Both serialized forms will continue to be 059 * supported by future versions of this library. However, serial forms generated by newer versions 060 * of the code may not be readable by older versions of the code (e.g., a serialized Bloom filter 061 * generated today may <i>not</i> be readable by a binary that was compiled 6 months ago). 062 * 063 * <p>As of Guava 23.0, this class is thread-safe and lock-free. It internally uses atomics and 064 * compare-and-swap to ensure correctness when multiple threads are used to access it. 065 * 066 * @param <T> the type of instances that the {@code BloomFilter} accepts 067 * @author Dimitris Andreou 068 * @author Kevin Bourrillion 069 * @since 11.0 (thread-safe since 23.0) 070 */ 071@Beta 072@ElementTypesAreNonnullByDefault 073public final class BloomFilter<T extends @Nullable Object> implements Predicate<T>, Serializable { 074 /** 075 * A strategy to translate T instances, to {@code numHashFunctions} bit indexes. 076 * 077 * <p>Implementations should be collections of pure functions (i.e. stateless). 078 */ 079 interface Strategy extends java.io.Serializable { 080 081 /** 082 * Sets {@code numHashFunctions} bits of the given bit array, by hashing a user element. 083 * 084 * <p>Returns whether any bits changed as a result of this operation. 085 */ 086 <T extends @Nullable Object> boolean put( 087 @ParametricNullness T object, 088 Funnel<? super T> funnel, 089 int numHashFunctions, 090 LockFreeBitArray bits); 091 092 /** 093 * Queries {@code numHashFunctions} bits of the given bit array, by hashing a user element; 094 * returns {@code true} if and only if all selected bits are set. 095 */ 096 <T extends @Nullable Object> boolean mightContain( 097 @ParametricNullness T object, 098 Funnel<? super T> funnel, 099 int numHashFunctions, 100 LockFreeBitArray bits); 101 102 /** 103 * Identifier used to encode this strategy, when marshalled as part of a BloomFilter. Only 104 * values in the [-128, 127] range are valid for the compact serial form. Non-negative values 105 * are reserved for enums defined in BloomFilterStrategies; negative values are reserved for any 106 * custom, stateful strategy we may define (e.g. any kind of strategy that would depend on user 107 * input). 108 */ 109 int ordinal(); 110 } 111 112 /** The bit set of the BloomFilter (not necessarily power of 2!) */ 113 private final LockFreeBitArray bits; 114 115 /** Number of hashes per element */ 116 private final int numHashFunctions; 117 118 /** The funnel to translate Ts to bytes */ 119 private final Funnel<? super T> funnel; 120 121 /** The strategy we employ to map an element T to {@code numHashFunctions} bit indexes. */ 122 private final Strategy strategy; 123 124 /** Natural logarithm of 2, used to optimize calculations in Bloom filter sizing. */ 125 private static final double LOG_TWO = Math.log(2); 126 127 /** Square of the natural logarithm of 2, reused to optimize the bit size calculation. */ 128 private static final double SQUARED_LOG_TWO = LOG_TWO * LOG_TWO; 129 130 /** Creates a BloomFilter. */ 131 private BloomFilter( 132 LockFreeBitArray bits, int numHashFunctions, Funnel<? super T> funnel, Strategy strategy) { 133 checkArgument(numHashFunctions > 0, "numHashFunctions (%s) must be > 0", numHashFunctions); 134 checkArgument( 135 numHashFunctions <= 255, "numHashFunctions (%s) must be <= 255", numHashFunctions); 136 this.bits = checkNotNull(bits); 137 this.numHashFunctions = numHashFunctions; 138 this.funnel = checkNotNull(funnel); 139 this.strategy = checkNotNull(strategy); 140 } 141 142 /** 143 * Creates a new {@code BloomFilter} that's a copy of this instance. The new instance is equal to 144 * this instance but shares no mutable state. 145 * 146 * @since 12.0 147 */ 148 public BloomFilter<T> copy() { 149 return new BloomFilter<>(bits.copy(), numHashFunctions, funnel, strategy); 150 } 151 152 /** 153 * Returns {@code true} if the element <i>might</i> have been put in this Bloom filter, {@code 154 * false} if this is <i>definitely</i> not the case. 155 */ 156 public boolean mightContain(@ParametricNullness T object) { 157 return strategy.mightContain(object, funnel, numHashFunctions, bits); 158 } 159 160 /** 161 * @deprecated Provided only to satisfy the {@link Predicate} interface; use {@link #mightContain} 162 * instead. 163 */ 164 @Deprecated 165 @Override 166 public boolean apply(@ParametricNullness T input) { 167 return mightContain(input); 168 } 169 170 /** 171 * Puts an element into this {@code BloomFilter}. Ensures that subsequent invocations of {@link 172 * #mightContain(Object)} with the same element will always return {@code true}. 173 * 174 * @return true if the Bloom filter's bits changed as a result of this operation. If the bits 175 * changed, this is <i>definitely</i> the first time {@code object} has been added to the 176 * filter. If the bits haven't changed, this <i>might</i> be the first time {@code object} has 177 * been added to the filter. Note that {@code put(t)} always returns the <i>opposite</i> 178 * result to what {@code mightContain(t)} would have returned at the time it is called. 179 * @since 12.0 (present in 11.0 with {@code void} return type}) 180 */ 181 @CanIgnoreReturnValue 182 public boolean put(@ParametricNullness T object) { 183 return strategy.put(object, funnel, numHashFunctions, bits); 184 } 185 186 /** 187 * Returns the probability that {@linkplain #mightContain(Object)} will erroneously return {@code 188 * true} for an object that has not actually been put in the {@code BloomFilter}. 189 * 190 * <p>Ideally, this number should be close to the {@code fpp} parameter passed in {@linkplain 191 * #create(Funnel, int, double)}, or smaller. If it is significantly higher, it is usually the 192 * case that too many elements (more than expected) have been put in the {@code BloomFilter}, 193 * degenerating it. 194 * 195 * @since 14.0 (since 11.0 as expectedFalsePositiveProbability()) 196 */ 197 public double expectedFpp() { 198 return Math.pow((double) bits.bitCount() / bitSize(), numHashFunctions); 199 } 200 201 /** 202 * Returns an estimate for the total number of distinct elements that have been added to this 203 * Bloom filter. This approximation is reasonably accurate if it does not exceed the value of 204 * {@code expectedInsertions} that was used when constructing the filter. 205 * 206 * @since 22.0 207 */ 208 public long approximateElementCount() { 209 long bitSize = bits.bitSize(); 210 long bitCount = bits.bitCount(); 211 212 /** 213 * Each insertion is expected to reduce the # of clear bits by a factor of 214 * `numHashFunctions/bitSize`. So, after n insertions, expected bitCount is `bitSize * (1 - (1 - 215 * numHashFunctions/bitSize)^n)`. Solving that for n, and approximating `ln x` as `x - 1` when x 216 * is close to 1 (why?), gives the following formula. 217 */ 218 double fractionOfBitsSet = (double) bitCount / bitSize; 219 return DoubleMath.roundToLong( 220 -Math.log1p(-fractionOfBitsSet) * bitSize / numHashFunctions, RoundingMode.HALF_UP); 221 } 222 223 /** Returns the number of bits in the underlying bit array. */ 224 @VisibleForTesting 225 long bitSize() { 226 return bits.bitSize(); 227 } 228 229 /** 230 * Determines whether a given Bloom filter is compatible with this Bloom filter. For two Bloom 231 * filters to be compatible, they must: 232 * 233 * <ul> 234 * <li>not be the same instance 235 * <li>have the same number of hash functions 236 * <li>have the same bit size 237 * <li>have the same strategy 238 * <li>have equal funnels 239 * </ul> 240 * 241 * @param that The Bloom filter to check for compatibility. 242 * @since 15.0 243 */ 244 public boolean isCompatible(BloomFilter<T> that) { 245 checkNotNull(that); 246 return this != that 247 && this.numHashFunctions == that.numHashFunctions 248 && this.bitSize() == that.bitSize() 249 && this.strategy.equals(that.strategy) 250 && this.funnel.equals(that.funnel); 251 } 252 253 /** 254 * Combines this Bloom filter with another Bloom filter by performing a bitwise OR of the 255 * underlying data. The mutations happen to <b>this</b> instance. Callers must ensure the Bloom 256 * filters are appropriately sized to avoid saturating them. 257 * 258 * @param that The Bloom filter to combine this Bloom filter with. It is not mutated. 259 * @throws IllegalArgumentException if {@code isCompatible(that) == false} 260 * @since 15.0 261 */ 262 public void putAll(BloomFilter<T> that) { 263 checkNotNull(that); 264 checkArgument(this != that, "Cannot combine a BloomFilter with itself."); 265 checkArgument( 266 this.numHashFunctions == that.numHashFunctions, 267 "BloomFilters must have the same number of hash functions (%s != %s)", 268 this.numHashFunctions, 269 that.numHashFunctions); 270 checkArgument( 271 this.bitSize() == that.bitSize(), 272 "BloomFilters must have the same size underlying bit arrays (%s != %s)", 273 this.bitSize(), 274 that.bitSize()); 275 checkArgument( 276 this.strategy.equals(that.strategy), 277 "BloomFilters must have equal strategies (%s != %s)", 278 this.strategy, 279 that.strategy); 280 checkArgument( 281 this.funnel.equals(that.funnel), 282 "BloomFilters must have equal funnels (%s != %s)", 283 this.funnel, 284 that.funnel); 285 this.bits.putAll(that.bits); 286 } 287 288 @Override 289 public boolean equals(@CheckForNull Object object) { 290 if (object == this) { 291 return true; 292 } 293 if (object instanceof BloomFilter) { 294 BloomFilter<?> that = (BloomFilter<?>) object; 295 return this.numHashFunctions == that.numHashFunctions 296 && this.funnel.equals(that.funnel) 297 && this.bits.equals(that.bits) 298 && this.strategy.equals(that.strategy); 299 } 300 return false; 301 } 302 303 @Override 304 public int hashCode() { 305 return Objects.hashCode(numHashFunctions, funnel, strategy, bits); 306 } 307 308 /** 309 * Returns a {@code Collector} expecting the specified number of insertions, and yielding a {@link 310 * BloomFilter} with false positive probability 3%. 311 * 312 * <p>Note that if the {@code Collector} receives significantly more elements than specified, the 313 * resulting {@code BloomFilter} will suffer a sharp deterioration of its false positive 314 * probability. 315 * 316 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 317 * is. 318 * 319 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 320 * ensuring proper serialization and deserialization, which is important since {@link #equals} 321 * also relies on object identity of funnels. 322 * 323 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 324 * @param expectedInsertions the number of expected insertions to the constructed {@code 325 * BloomFilter}; must be positive 326 * @return a {@code Collector} generating a {@code BloomFilter} of the received elements 327 * @since 23.0 (but only since 33.4.0 in the Android flavor) 328 */ 329 public static <T extends @Nullable Object> Collector<T, ?, BloomFilter<T>> toBloomFilter( 330 Funnel<? super T> funnel, long expectedInsertions) { 331 return toBloomFilter(funnel, expectedInsertions, 0.03); 332 } 333 334 /** 335 * Returns a {@code Collector} expecting the specified number of insertions, and yielding a {@link 336 * BloomFilter} with the specified expected false positive probability. 337 * 338 * <p>Note that if the {@code Collector} receives significantly more elements than specified, the 339 * resulting {@code BloomFilter} will suffer a sharp deterioration of its false positive 340 * probability. 341 * 342 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 343 * is. 344 * 345 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 346 * ensuring proper serialization and deserialization, which is important since {@link #equals} 347 * also relies on object identity of funnels. 348 * 349 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 350 * @param expectedInsertions the number of expected insertions to the constructed {@code 351 * BloomFilter}; must be positive 352 * @param fpp the desired false positive probability (must be positive and less than 1.0) 353 * @return a {@code Collector} generating a {@code BloomFilter} of the received elements 354 * @since 23.0 (but only since 33.4.0 in the Android flavor) 355 */ 356 public static <T extends @Nullable Object> Collector<T, ?, BloomFilter<T>> toBloomFilter( 357 Funnel<? super T> funnel, long expectedInsertions, double fpp) { 358 checkNotNull(funnel); 359 checkArgument( 360 expectedInsertions >= 0, "Expected insertions (%s) must be >= 0", expectedInsertions); 361 checkArgument(fpp > 0.0, "False positive probability (%s) must be > 0.0", fpp); 362 checkArgument(fpp < 1.0, "False positive probability (%s) must be < 1.0", fpp); 363 return Collector.of( 364 () -> BloomFilter.create(funnel, expectedInsertions, fpp), 365 BloomFilter::put, 366 (bf1, bf2) -> { 367 bf1.putAll(bf2); 368 return bf1; 369 }, 370 Collector.Characteristics.UNORDERED, 371 Collector.Characteristics.CONCURRENT); 372 } 373 374 /** 375 * Creates a {@link BloomFilter} with the expected number of insertions and expected false 376 * positive probability. 377 * 378 * <p>Note that overflowing a {@code BloomFilter} with significantly more elements than specified, 379 * will result in its saturation, and a sharp deterioration of its false positive probability. 380 * 381 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 382 * is. 383 * 384 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 385 * ensuring proper serialization and deserialization, which is important since {@link #equals} 386 * also relies on object identity of funnels. 387 * 388 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 389 * @param expectedInsertions the number of expected insertions to the constructed {@code 390 * BloomFilter}; must be positive 391 * @param fpp the desired false positive probability (must be positive and less than 1.0) 392 * @return a {@code BloomFilter} 393 */ 394 public static <T extends @Nullable Object> BloomFilter<T> create( 395 Funnel<? super T> funnel, int expectedInsertions, double fpp) { 396 return create(funnel, (long) expectedInsertions, fpp); 397 } 398 399 /** 400 * Creates a {@link BloomFilter} with the expected number of insertions and expected false 401 * positive probability. 402 * 403 * <p>Note that overflowing a {@code BloomFilter} with significantly more elements than specified, 404 * will result in its saturation, and a sharp deterioration of its false positive probability. 405 * 406 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 407 * is. 408 * 409 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 410 * ensuring proper serialization and deserialization, which is important since {@link #equals} 411 * also relies on object identity of funnels. 412 * 413 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 414 * @param expectedInsertions the number of expected insertions to the constructed {@code 415 * BloomFilter}; must be positive 416 * @param fpp the desired false positive probability (must be positive and less than 1.0) 417 * @return a {@code BloomFilter} 418 * @since 19.0 419 */ 420 public static <T extends @Nullable Object> BloomFilter<T> create( 421 Funnel<? super T> funnel, long expectedInsertions, double fpp) { 422 return create(funnel, expectedInsertions, fpp, BloomFilterStrategies.MURMUR128_MITZ_64); 423 } 424 425 @VisibleForTesting 426 static <T extends @Nullable Object> BloomFilter<T> create( 427 Funnel<? super T> funnel, long expectedInsertions, double fpp, Strategy strategy) { 428 checkNotNull(funnel); 429 checkArgument( 430 expectedInsertions >= 0, "Expected insertions (%s) must be >= 0", expectedInsertions); 431 checkArgument(fpp > 0.0, "False positive probability (%s) must be > 0.0", fpp); 432 checkArgument(fpp < 1.0, "False positive probability (%s) must be < 1.0", fpp); 433 checkNotNull(strategy); 434 435 if (expectedInsertions == 0) { 436 expectedInsertions = 1; 437 } 438 /* 439 * TODO(user): Put a warning in the javadoc about tiny fpp values, since the resulting size 440 * is proportional to -log(p), but there is not much of a point after all, e.g. 441 * optimalM(1000, 0.0000000000000001) = 76680 which is less than 10kb. Who cares! 442 */ 443 long numBits = optimalNumOfBits(expectedInsertions, fpp); 444 int numHashFunctions = optimalNumOfHashFunctions(fpp); 445 try { 446 return new BloomFilter<>(new LockFreeBitArray(numBits), numHashFunctions, funnel, strategy); 447 } catch (IllegalArgumentException e) { 448 throw new IllegalArgumentException("Could not create BloomFilter of " + numBits + " bits", e); 449 } 450 } 451 452 /** 453 * Creates a {@link BloomFilter} with the expected number of insertions and a default expected 454 * false positive probability of 3%. 455 * 456 * <p>Note that overflowing a {@code BloomFilter} with significantly more elements than specified, 457 * will result in its saturation, and a sharp deterioration of its false positive probability. 458 * 459 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 460 * is. 461 * 462 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 463 * ensuring proper serialization and deserialization, which is important since {@link #equals} 464 * also relies on object identity of funnels. 465 * 466 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 467 * @param expectedInsertions the number of expected insertions to the constructed {@code 468 * BloomFilter}; must be positive 469 * @return a {@code BloomFilter} 470 */ 471 public static <T extends @Nullable Object> BloomFilter<T> create( 472 Funnel<? super T> funnel, int expectedInsertions) { 473 return create(funnel, (long) expectedInsertions); 474 } 475 476 /** 477 * Creates a {@link BloomFilter} with the expected number of insertions and a default expected 478 * false positive probability of 3%. 479 * 480 * <p>Note that overflowing a {@code BloomFilter} with significantly more elements than specified, 481 * will result in its saturation, and a sharp deterioration of its false positive probability. 482 * 483 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 484 * is. 485 * 486 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 487 * ensuring proper serialization and deserialization, which is important since {@link #equals} 488 * also relies on object identity of funnels. 489 * 490 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 491 * @param expectedInsertions the number of expected insertions to the constructed {@code 492 * BloomFilter}; must be positive 493 * @return a {@code BloomFilter} 494 * @since 19.0 495 */ 496 public static <T extends @Nullable Object> BloomFilter<T> create( 497 Funnel<? super T> funnel, long expectedInsertions) { 498 return create(funnel, expectedInsertions, 0.03); // FYI, for 3%, we always get 5 hash functions 499 } 500 501 // Cheat sheet: 502 // 503 // m: total bits 504 // n: expected insertions 505 // b: m/n, bits per insertion 506 // p: expected false positive probability 507 // 508 // 1) Optimal k = b * ln2 509 // 2) p = (1 - e ^ (-kn/m))^k 510 // 3) For optimal k: p = 2 ^ (-k) ~= 0.6185^b 511 // 4) For optimal k: m = -nlnp / ((ln2) ^ 2) 512 513 /** 514 * Computes the optimal number of hash functions (k) for a given false positive probability (p). 515 * 516 * <p>See http://en.wikipedia.org/wiki/File:Bloom_filter_fp_probability.svg for the formula. 517 * 518 * @param p desired false positive probability (must be between 0 and 1, exclusive) 519 */ 520 @VisibleForTesting 521 static int optimalNumOfHashFunctions(double p) { 522 // -log(p) / log(2), ensuring the result is rounded to avoid truncation. 523 return max(1, (int) Math.round(-Math.log(p) / LOG_TWO)); 524 } 525 526 /** 527 * Computes m (total bits of Bloom filter) which is expected to achieve, for the specified 528 * expected insertions, the required false positive probability. 529 * 530 * <p>See http://en.wikipedia.org/wiki/Bloom_filter#Probability_of_false_positives for the 531 * formula. 532 * 533 * @param n expected insertions (must be positive) 534 * @param p false positive rate (must be 0 < p < 1) 535 */ 536 @VisibleForTesting 537 static long optimalNumOfBits(long n, double p) { 538 if (p == 0) { 539 p = Double.MIN_VALUE; 540 } 541 return (long) (-n * Math.log(p) / SQUARED_LOG_TWO); 542 } 543 544 private Object writeReplace() { 545 return new SerialForm<T>(this); 546 } 547 548 private void readObject(ObjectInputStream stream) throws InvalidObjectException { 549 throw new InvalidObjectException("Use SerializedForm"); 550 } 551 552 private static class SerialForm<T extends @Nullable Object> implements Serializable { 553 final long[] data; 554 final int numHashFunctions; 555 final Funnel<? super T> funnel; 556 final Strategy strategy; 557 558 SerialForm(BloomFilter<T> bf) { 559 this.data = LockFreeBitArray.toPlainArray(bf.bits.data); 560 this.numHashFunctions = bf.numHashFunctions; 561 this.funnel = bf.funnel; 562 this.strategy = bf.strategy; 563 } 564 565 Object readResolve() { 566 return new BloomFilter<T>(new LockFreeBitArray(data), numHashFunctions, funnel, strategy); 567 } 568 569 private static final long serialVersionUID = 1; 570 } 571 572 /** 573 * Writes this {@code BloomFilter} to an output stream, with a custom format (not Java 574 * serialization). This has been measured to save at least 400 bytes compared to regular 575 * serialization. 576 * 577 * <p>Use {@linkplain #readFrom(InputStream, Funnel)} to reconstruct the written BloomFilter. 578 */ 579 public void writeTo(OutputStream out) throws IOException { 580 // Serial form: 581 // 1 signed byte for the strategy 582 // 1 unsigned byte for the number of hash functions 583 // 1 big endian int, the number of longs in our bitset 584 // N big endian longs of our bitset 585 DataOutputStream dout = new DataOutputStream(out); 586 dout.writeByte(SignedBytes.checkedCast(strategy.ordinal())); 587 dout.writeByte(UnsignedBytes.checkedCast(numHashFunctions)); // note: checked at the c'tor 588 dout.writeInt(bits.data.length()); 589 for (int i = 0; i < bits.data.length(); i++) { 590 dout.writeLong(bits.data.get(i)); 591 } 592 } 593 594 /** 595 * Reads a byte stream, which was written by {@linkplain #writeTo(OutputStream)}, into a {@code 596 * BloomFilter}. 597 * 598 * <p>The {@code Funnel} to be used is not encoded in the stream, so it must be provided here. 599 * <b>Warning:</b> the funnel provided <b>must</b> behave identically to the one used to populate 600 * the original Bloom filter! 601 * 602 * @throws IOException if the InputStream throws an {@code IOException}, or if its data does not 603 * appear to be a BloomFilter serialized using the {@linkplain #writeTo(OutputStream)} method. 604 */ 605 @SuppressWarnings("CatchingUnchecked") // sneaky checked exception 606 public static <T extends @Nullable Object> BloomFilter<T> readFrom( 607 InputStream in, Funnel<? super T> funnel) throws IOException { 608 checkNotNull(in, "InputStream"); 609 checkNotNull(funnel, "Funnel"); 610 int strategyOrdinal = -1; 611 int numHashFunctions = -1; 612 int dataLength = -1; 613 try { 614 DataInputStream din = new DataInputStream(in); 615 // currently this assumes there is no negative ordinal; will have to be updated if we 616 // add non-stateless strategies (for which we've reserved negative ordinals; see 617 // Strategy.ordinal()). 618 strategyOrdinal = din.readByte(); 619 numHashFunctions = UnsignedBytes.toInt(din.readByte()); 620 dataLength = din.readInt(); 621 622 Strategy strategy = BloomFilterStrategies.values()[strategyOrdinal]; 623 624 LockFreeBitArray dataArray = new LockFreeBitArray(LongMath.checkedMultiply(dataLength, 64L)); 625 for (int i = 0; i < dataLength; i++) { 626 dataArray.putData(i, din.readLong()); 627 } 628 629 return new BloomFilter<>(dataArray, numHashFunctions, funnel, strategy); 630 } catch (IOException e) { 631 throw e; 632 } catch (Exception e) { // sneaky checked exception 633 String message = 634 "Unable to deserialize BloomFilter from InputStream." 635 + " strategyOrdinal: " 636 + strategyOrdinal 637 + " numHashFunctions: " 638 + numHashFunctions 639 + " dataLength: " 640 + dataLength; 641 throw new IOException(message, e); 642 } 643 } 644 645 private static final long serialVersionUID = 0xcafebabe; 646}