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 com.google.errorprone.annotations.InlineMe; 032import java.io.DataInputStream; 033import java.io.DataOutputStream; 034import java.io.IOException; 035import java.io.InputStream; 036import java.io.InvalidObjectException; 037import java.io.ObjectInputStream; 038import java.io.OutputStream; 039import java.io.Serializable; 040import java.math.RoundingMode; 041import java.util.stream.Collector; 042import org.jspecify.annotations.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 072public final class BloomFilter<T extends @Nullable Object> implements Predicate<T>, Serializable { 073 /** 074 * A strategy to translate T instances, to {@code numHashFunctions} bit indexes. 075 * 076 * <p>Implementations should be collections of pure functions (i.e. stateless). 077 */ 078 interface Strategy extends java.io.Serializable { 079 080 /** 081 * Sets {@code numHashFunctions} bits of the given bit array, by hashing a user element. 082 * 083 * <p>Returns whether any bits changed as a result of this operation. 084 */ 085 <T extends @Nullable Object> boolean put( 086 @ParametricNullness T object, 087 Funnel<? super T> funnel, 088 int numHashFunctions, 089 LockFreeBitArray bits); 090 091 /** 092 * Queries {@code numHashFunctions} bits of the given bit array, by hashing a user element; 093 * returns {@code true} if and only if all selected bits are set. 094 */ 095 <T extends @Nullable Object> boolean mightContain( 096 @ParametricNullness T object, 097 Funnel<? super T> funnel, 098 int numHashFunctions, 099 LockFreeBitArray bits); 100 101 /** 102 * Identifier used to encode this strategy, when marshalled as part of a BloomFilter. Only 103 * values in the [-128, 127] range are valid for the compact serial form. Non-negative values 104 * are reserved for enums defined in BloomFilterStrategies; negative values are reserved for any 105 * custom, stateful strategy we may define (e.g. any kind of strategy that would depend on user 106 * input). 107 */ 108 int ordinal(); 109 } 110 111 /** The bit set of the BloomFilter (not necessarily power of 2!) */ 112 private final LockFreeBitArray bits; 113 114 /** Number of hashes per element */ 115 private final int numHashFunctions; 116 117 /** The funnel to translate Ts to bytes */ 118 private final Funnel<? super T> funnel; 119 120 /** The strategy we employ to map an element T to {@code numHashFunctions} bit indexes. */ 121 private final Strategy strategy; 122 123 /** Natural logarithm of 2, used to optimize calculations in Bloom filter sizing. */ 124 private static final double LOG_TWO = Math.log(2); 125 126 /** Square of the natural logarithm of 2, reused to optimize the bit size calculation. */ 127 private static final double SQUARED_LOG_TWO = LOG_TWO * LOG_TWO; 128 129 /** Creates a BloomFilter. */ 130 private BloomFilter( 131 LockFreeBitArray bits, int numHashFunctions, Funnel<? super T> funnel, Strategy strategy) { 132 checkArgument(numHashFunctions > 0, "numHashFunctions (%s) must be > 0", numHashFunctions); 133 checkArgument( 134 numHashFunctions <= 255, "numHashFunctions (%s) must be <= 255", numHashFunctions); 135 this.bits = checkNotNull(bits); 136 this.numHashFunctions = numHashFunctions; 137 this.funnel = checkNotNull(funnel); 138 this.strategy = checkNotNull(strategy); 139 } 140 141 /** 142 * Creates a new {@code BloomFilter} that's a copy of this instance. The new instance is equal to 143 * this instance but shares no mutable state. 144 * 145 * @since 12.0 146 */ 147 public BloomFilter<T> copy() { 148 return new BloomFilter<>(bits.copy(), numHashFunctions, funnel, strategy); 149 } 150 151 /** 152 * Returns {@code true} if the element <i>might</i> have been put in this Bloom filter, {@code 153 * false} if this is <i>definitely</i> not the case. 154 */ 155 public boolean mightContain(@ParametricNullness T object) { 156 return strategy.mightContain(object, funnel, numHashFunctions, bits); 157 } 158 159 /** 160 * @deprecated Provided only to satisfy the {@link Predicate} interface; use {@link #mightContain} 161 * instead. 162 */ 163 @InlineMe(replacement = "this.mightContain(input)") 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(@Nullable 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 33.4.0 (but since 23.0 in the JRE flavor) 328 */ 329 @SuppressWarnings("Java7ApiChecker") 330 @IgnoreJRERequirement // Users will use this only if they're already using streams. 331 public static <T extends @Nullable Object> Collector<T, ?, BloomFilter<T>> toBloomFilter( 332 Funnel<? super T> funnel, long expectedInsertions) { 333 return toBloomFilter(funnel, expectedInsertions, 0.03); 334 } 335 336 /** 337 * Returns a {@code Collector} expecting the specified number of insertions, and yielding a {@link 338 * BloomFilter} with the specified expected false positive probability. 339 * 340 * <p>Note that if the {@code Collector} receives significantly more elements than specified, the 341 * resulting {@code BloomFilter} will suffer a sharp deterioration of its false positive 342 * probability. 343 * 344 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 345 * is. 346 * 347 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 348 * ensuring proper serialization and deserialization, which is important since {@link #equals} 349 * also relies on object identity of funnels. 350 * 351 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 352 * @param expectedInsertions the number of expected insertions to the constructed {@code 353 * BloomFilter}; must be positive 354 * @param fpp the desired false positive probability (must be positive and less than 1.0) 355 * @return a {@code Collector} generating a {@code BloomFilter} of the received elements 356 * @since 33.4.0 (but since 23.0 in the JRE flavor) 357 */ 358 @SuppressWarnings("Java7ApiChecker") 359 @IgnoreJRERequirement // Users will use this only if they're already using streams. 360 public static <T extends @Nullable Object> Collector<T, ?, BloomFilter<T>> toBloomFilter( 361 Funnel<? super T> funnel, long expectedInsertions, double fpp) { 362 checkNotNull(funnel); 363 checkArgument( 364 expectedInsertions >= 0, "Expected insertions (%s) must be >= 0", expectedInsertions); 365 checkArgument(fpp > 0.0, "False positive probability (%s) must be > 0.0", fpp); 366 checkArgument(fpp < 1.0, "False positive probability (%s) must be < 1.0", fpp); 367 return Collector.of( 368 () -> BloomFilter.create(funnel, expectedInsertions, fpp), 369 BloomFilter::put, 370 (bf1, bf2) -> { 371 bf1.putAll(bf2); 372 return bf1; 373 }, 374 Collector.Characteristics.UNORDERED, 375 Collector.Characteristics.CONCURRENT); 376 } 377 378 /** 379 * Creates a {@link BloomFilter} with the expected number of insertions and expected false 380 * positive probability. 381 * 382 * <p>Note that overflowing a {@code BloomFilter} with significantly more elements than specified, 383 * will result in its saturation, and a sharp deterioration of its false positive probability. 384 * 385 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 386 * is. 387 * 388 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 389 * ensuring proper serialization and deserialization, which is important since {@link #equals} 390 * also relies on object identity of funnels. 391 * 392 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 393 * @param expectedInsertions the number of expected insertions to the constructed {@code 394 * BloomFilter}; must be positive 395 * @param fpp the desired false positive probability (must be positive and less than 1.0) 396 * @return a {@code BloomFilter} 397 */ 398 public static <T extends @Nullable Object> BloomFilter<T> create( 399 Funnel<? super T> funnel, int expectedInsertions, double fpp) { 400 return create(funnel, (long) expectedInsertions, fpp); 401 } 402 403 /** 404 * Creates a {@link BloomFilter} with the expected number of insertions and expected false 405 * positive probability. 406 * 407 * <p>Note that overflowing a {@code BloomFilter} with significantly more elements than specified, 408 * will result in its saturation, and a sharp deterioration of its false positive probability. 409 * 410 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 411 * is. 412 * 413 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 414 * ensuring proper serialization and deserialization, which is important since {@link #equals} 415 * also relies on object identity of funnels. 416 * 417 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 418 * @param expectedInsertions the number of expected insertions to the constructed {@code 419 * BloomFilter}; must be positive 420 * @param fpp the desired false positive probability (must be positive and less than 1.0) 421 * @return a {@code BloomFilter} 422 * @since 19.0 423 */ 424 public static <T extends @Nullable Object> BloomFilter<T> create( 425 Funnel<? super T> funnel, long expectedInsertions, double fpp) { 426 return create(funnel, expectedInsertions, fpp, BloomFilterStrategies.MURMUR128_MITZ_64); 427 } 428 429 @VisibleForTesting 430 static <T extends @Nullable Object> BloomFilter<T> create( 431 Funnel<? super T> funnel, long expectedInsertions, double fpp, Strategy strategy) { 432 checkNotNull(funnel); 433 checkArgument( 434 expectedInsertions >= 0, "Expected insertions (%s) must be >= 0", expectedInsertions); 435 checkArgument(fpp > 0.0, "False positive probability (%s) must be > 0.0", fpp); 436 checkArgument(fpp < 1.0, "False positive probability (%s) must be < 1.0", fpp); 437 checkNotNull(strategy); 438 439 if (expectedInsertions == 0) { 440 expectedInsertions = 1; 441 } 442 /* 443 * TODO(user): Put a warning in the javadoc about tiny fpp values, since the resulting size 444 * is proportional to -log(p), but there is not much of a point after all, e.g. 445 * optimalM(1000, 0.0000000000000001) = 76680 which is less than 10kb. Who cares! 446 */ 447 long numBits = optimalNumOfBits(expectedInsertions, fpp); 448 int numHashFunctions = optimalNumOfHashFunctions(fpp); 449 try { 450 return new BloomFilter<>(new LockFreeBitArray(numBits), numHashFunctions, funnel, strategy); 451 } catch (IllegalArgumentException e) { 452 throw new IllegalArgumentException("Could not create BloomFilter of " + numBits + " bits", e); 453 } 454 } 455 456 /** 457 * Creates a {@link BloomFilter} with the expected number of insertions and a default expected 458 * false positive probability of 3%. 459 * 460 * <p>Note that overflowing a {@code BloomFilter} with significantly more elements than specified, 461 * will result in its saturation, and a sharp deterioration of its false positive probability. 462 * 463 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 464 * is. 465 * 466 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 467 * ensuring proper serialization and deserialization, which is important since {@link #equals} 468 * also relies on object identity of funnels. 469 * 470 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 471 * @param expectedInsertions the number of expected insertions to the constructed {@code 472 * BloomFilter}; must be positive 473 * @return a {@code BloomFilter} 474 */ 475 public static <T extends @Nullable Object> BloomFilter<T> create( 476 Funnel<? super T> funnel, int expectedInsertions) { 477 return create(funnel, (long) expectedInsertions); 478 } 479 480 /** 481 * Creates a {@link BloomFilter} with the expected number of insertions and a default expected 482 * false positive probability of 3%. 483 * 484 * <p>Note that overflowing a {@code BloomFilter} with significantly more elements than specified, 485 * will result in its saturation, and a sharp deterioration of its false positive probability. 486 * 487 * <p>The constructed {@code BloomFilter} will be serializable if the provided {@code Funnel<T>} 488 * is. 489 * 490 * <p>It is recommended that the funnel be implemented as a Java enum. This has the benefit of 491 * ensuring proper serialization and deserialization, which is important since {@link #equals} 492 * also relies on object identity of funnels. 493 * 494 * @param funnel the funnel of T's that the constructed {@code BloomFilter} will use 495 * @param expectedInsertions the number of expected insertions to the constructed {@code 496 * BloomFilter}; must be positive 497 * @return a {@code BloomFilter} 498 * @since 19.0 499 */ 500 public static <T extends @Nullable Object> BloomFilter<T> create( 501 Funnel<? super T> funnel, long expectedInsertions) { 502 return create(funnel, expectedInsertions, 0.03); // FYI, for 3%, we always get 5 hash functions 503 } 504 505 // Cheat sheet: 506 // 507 // m: total bits 508 // n: expected insertions 509 // b: m/n, bits per insertion 510 // p: expected false positive probability 511 // 512 // 1) Optimal k = b * ln2 513 // 2) p = (1 - e ^ (-kn/m))^k 514 // 3) For optimal k: p = 2 ^ (-k) ~= 0.6185^b 515 // 4) For optimal k: m = -nlnp / ((ln2) ^ 2) 516 517 /** 518 * Computes the optimal number of hash functions (k) for a given false positive probability (p). 519 * 520 * <p>See http://en.wikipedia.org/wiki/File:Bloom_filter_fp_probability.svg for the formula. 521 * 522 * @param p desired false positive probability (must be between 0 and 1, exclusive) 523 */ 524 @VisibleForTesting 525 static int optimalNumOfHashFunctions(double p) { 526 // -log(p) / log(2), ensuring the result is rounded to avoid truncation. 527 return max(1, (int) Math.round(-Math.log(p) / LOG_TWO)); 528 } 529 530 /** 531 * Computes m (total bits of Bloom filter) which is expected to achieve, for the specified 532 * expected insertions, the required false positive probability. 533 * 534 * <p>See http://en.wikipedia.org/wiki/Bloom_filter#Probability_of_false_positives for the 535 * formula. 536 * 537 * @param n expected insertions (must be positive) 538 * @param p false positive rate (must be 0 < p < 1) 539 */ 540 @VisibleForTesting 541 static long optimalNumOfBits(long n, double p) { 542 if (p == 0) { 543 p = Double.MIN_VALUE; 544 } 545 return (long) (-n * Math.log(p) / SQUARED_LOG_TWO); 546 } 547 548 private Object writeReplace() { 549 return new SerialForm<T>(this); 550 } 551 552 private void readObject(ObjectInputStream stream) throws InvalidObjectException { 553 throw new InvalidObjectException("Use SerializedForm"); 554 } 555 556 private static class SerialForm<T extends @Nullable Object> implements Serializable { 557 final long[] data; 558 final int numHashFunctions; 559 final Funnel<? super T> funnel; 560 final Strategy strategy; 561 562 SerialForm(BloomFilter<T> bf) { 563 this.data = LockFreeBitArray.toPlainArray(bf.bits.data); 564 this.numHashFunctions = bf.numHashFunctions; 565 this.funnel = bf.funnel; 566 this.strategy = bf.strategy; 567 } 568 569 Object readResolve() { 570 return new BloomFilter<T>(new LockFreeBitArray(data), numHashFunctions, funnel, strategy); 571 } 572 573 private static final long serialVersionUID = 1; 574 } 575 576 /** 577 * Writes this {@code BloomFilter} to an output stream, with a custom format (not Java 578 * serialization). This has been measured to save at least 400 bytes compared to regular 579 * serialization. 580 * 581 * <p>Use {@linkplain #readFrom(InputStream, Funnel)} to reconstruct the written BloomFilter. 582 */ 583 public void writeTo(OutputStream out) throws IOException { 584 // Serial form: 585 // 1 signed byte for the strategy 586 // 1 unsigned byte for the number of hash functions 587 // 1 big endian int, the number of longs in our bitset 588 // N big endian longs of our bitset 589 DataOutputStream dout = new DataOutputStream(out); 590 dout.writeByte(SignedBytes.checkedCast(strategy.ordinal())); 591 dout.writeByte(UnsignedBytes.checkedCast(numHashFunctions)); // note: checked at the c'tor 592 dout.writeInt(bits.data.length()); 593 for (int i = 0; i < bits.data.length(); i++) { 594 dout.writeLong(bits.data.get(i)); 595 } 596 } 597 598 /** 599 * Reads a byte stream, which was written by {@linkplain #writeTo(OutputStream)}, into a {@code 600 * BloomFilter}. 601 * 602 * <p>The {@code Funnel} to be used is not encoded in the stream, so it must be provided here. 603 * <b>Warning:</b> the funnel provided <b>must</b> behave identically to the one used to populate 604 * the original Bloom filter! 605 * 606 * @throws IOException if the InputStream throws an {@code IOException}, or if its data does not 607 * appear to be a BloomFilter serialized using the {@linkplain #writeTo(OutputStream)} method. 608 */ 609 @SuppressWarnings("CatchingUnchecked") // sneaky checked exception 610 public static <T extends @Nullable Object> BloomFilter<T> readFrom( 611 InputStream in, Funnel<? super T> funnel) throws IOException { 612 checkNotNull(in, "InputStream"); 613 checkNotNull(funnel, "Funnel"); 614 int strategyOrdinal = -1; 615 int numHashFunctions = -1; 616 int dataLength = -1; 617 try { 618 DataInputStream din = new DataInputStream(in); 619 // currently this assumes there is no negative ordinal; will have to be updated if we 620 // add non-stateless strategies (for which we've reserved negative ordinals; see 621 // Strategy.ordinal()). 622 strategyOrdinal = din.readByte(); 623 numHashFunctions = UnsignedBytes.toInt(din.readByte()); 624 dataLength = din.readInt(); 625 626 Strategy strategy = BloomFilterStrategies.values()[strategyOrdinal]; 627 628 LockFreeBitArray dataArray = new LockFreeBitArray(LongMath.checkedMultiply(dataLength, 64L)); 629 for (int i = 0; i < dataLength; i++) { 630 dataArray.putData(i, din.readLong()); 631 } 632 633 return new BloomFilter<>(dataArray, numHashFunctions, funnel, strategy); 634 } catch (IOException e) { 635 throw e; 636 } catch (Exception e) { // sneaky checked exception 637 String message = 638 "Unable to deserialize BloomFilter from InputStream." 639 + " strategyOrdinal: " 640 + strategyOrdinal 641 + " numHashFunctions: " 642 + numHashFunctions 643 + " dataLength: " 644 + dataLength; 645 throw new IOException(message, e); 646 } 647 } 648 649 private static final long serialVersionUID = 0xdecaf; 650}