Serialization.getFieldSetter(ImmutableSetMultimap.class, "emptySet");
  }

  @GwtIncompatible // java.io.ObjectInputStream
  @J2ktIncompatible
  // Serialization type safety is at the caller's mercy.
  @SuppressWarnings("unchecked")
  private void readObject(ObjectInputStream stream) throws IOException, ClassNotFoundException {
    stream.defaultReadObject();

 * percentiles together takes about 55% as long as computing them separately.
 *
 * <p>When calling {@link ScaleAndIndex#compute} (in {@linkplain ScaleAndIndexes#compute either
 * form}), the memory requirement is 8*N bytes for the copy of the dataset plus an overhead which is
 * independent of N (but depends on the quantiles being computed). When calling {@link

   * for {@code minimumBits}, will return identically-behaving {@link HashFunction} instances.
   *
   * @param minimumBits a positive integer. This can be arbitrarily large. The returned {@link
   *     HashFunction} instance may use memory proportional to this integer.
   * @return a hash function, described above, that produces hash codes of length {@code
   *     minimumBits} or greater
   */
  public static HashFunction goodFastHash(int minimumBits) {

 * them, even though they usually won't be needed. If you have such arguments, use the conventional
 * if/throw idiom instead.
 *
 * <p>Depending on your message arguments, memory may be allocated for boxing and varargs array
 * creation. However, the methods of this class have a large number of overloads that prevent such
 * allocations in many common cases.
 *

        comparator, keyFunction, valueFunction, mergeFunction);
  }

  /*
   * TODO(kevinb): Confirm that ImmutableSortedMap is faster to construct and
   * uses less memory than TreeMap; then say so in the class Javadoc.
   */
  private static final Comparator<?> NATURAL_ORDER = Ordering.natural();

  private static final ImmutableSortedMap<Comparable<?>, Object> NATURAL_EMPTY_MAP =

    if (file != null) {
      return new FileInputStream(file);
    } else {
      // requireNonNull is safe because we always have either `file` or `memory`.
      requireNonNull(memory);
      return new ByteArrayInputStream(memory.getBuffer(), 0, memory.getCount());
    }
  }

  /**
   * Calls {@link #close} if not already closed, and then resets this object back to its initial

    Fire<V> fire = new Fire<>(result);
    result.timer = scheduledExecutor.schedule(fire, time, unit);
    delegate.addListener(fire, directExecutor());
    return result;
  }

  /*
   * Memory visibility of these fields. There are two cases to consider.
   *
   * 1. visibility of the writes to these fields to Fire.run:
   *
   * The initial write to delegateRef is made definitely visible via the semantics of

 * <p>This class should not be assumed to be universally superior to {@code java.util.HashSet}.
 * Generally speaking, this class reduces object allocation and memory consumption at the price of
 * moderately increased constant factors of CPU. Only use this class when there is a specific reason
 * to prioritize memory over CPU.
 *
 * @author Dimitris Andreou
 * @author Jon Noack
 */
@GwtIncompatible // not worth using in GWT for now

   *       and virtual memory, this is not a problem - because it is mapped read-only, the kernel
   *       can always page it to disk "for free". However, on systems where killing processes
   *       happens all the time in normal conditions (i.e., android) the OS must make a tradeoff

   * map of public suffixes. The encoded trie string may be broken into multiple chunks to avoid the
   * 64k limit on string literal size. In-memory strings can be much larger (2G).
   */
  static ImmutableMap<String, PublicSuffixType> parseTrie(CharSequence... encodedChunks) {
    String encoded = DIRECT_JOINER.join(encodedChunks);
    return parseFullString(encoded);