*     directly, or use a combination of {@link #throwIfUnchecked} and {@code throw new
   *     RuntimeException(e)}. But consider whether users would be better off if your API threw a
   *     different type of exception. For background on the deprecation, read <a
   *     href="https://github.com/google/guava/wiki/Why-we-deprecated-Throwables.propagate">Why we

 *       method to obtain a correctly-sized copy.
 *   <li>The performance of using the associated {@code Builder} class can be assumed to be no
 *       worse, and possibly better, than creating a mutable collection and copying it.
 *   <li>Implementations generally do not cache hash codes. If your element or key type has a slow
 *       {@code hashCode} implementation, it should cache it itself.
 * </ul>

  }

  /**
   * Returns a new builder with a hint for how many distinct keys are expected to be added. The
   * generated builder is equivalent to that returned by {@link #builder}, but may perform better if
   * {@code expectedKeys} is a good estimate.
   *
   * @throws IllegalArgumentException if {@code expectedKeys} is negative
   * @since 33.3.0
   */

            like to get an error. Currently we do, but if we add the extra <inherits> lines, we
            won't.

            I have one idea for a better approach, but it's painful, and I haven't tested it: We
            could postprocess Collect.gwt.xml to add <skip> lines for all the files that should be
            covered by testModule.gwt.xml. Maybe I'll try it someday.

    interface to track metrics and monitor HTTP requests' size and duration.
 *  New: `okhttp-dnsoverhttps` is an experimental API for doing DNS queries over HTTPS. Using HTTPS
    for DNS offers better security and potentially better performance. This feature is a preview:
    the API is subject to change.
 *  New: `okhttp-sse` is an early preview of Server-Sent Events (SSE). This feature is incomplete

   * structure (to make it possible to throw ConcurrentModificationException in the iterator). Note
   * that we choose not to make this volatile, so we do less of a "best effort" to track such
   * errors, for better performance.
   */
  private transient int metadata;

  /** The number of elements contained in the set. */
  private transient int size;

  /** Constructs a new empty instance of {@code CompactHashSet}. */

   *
   * <p><b>Performance note:</b> When feasible, {@code initialCapacity} should be the exact number
   * of values that will be added, if that knowledge is readily available. It is better to guess a
   * value slightly too high than slightly too low. If the value is not exact, the {@link
   * ImmutableLongArray} that is built will very likely occupy more memory than strictly necessary;

   *
   * <p><b>Performance note:</b> When feasible, {@code initialCapacity} should be the exact number
   * of values that will be added, if that knowledge is readily available. It is better to guess a
   * value slightly too high than slightly too low. If the value is not exact, the {@link
   * ImmutableIntArray} that is built will very likely occupy more memory than strictly necessary;

  }

  /**
   * Not supported. Use {@link #naturalOrder}, which offers better type-safety, instead. This method
   * exists only to hide {@link ImmutableSet#builder} from consumers of {@code ImmutableSortedSet}.
   *
   * @throws UnsupportedOperationException always
   * @deprecated Use {@link ImmutableSortedSet#naturalOrder}, which offers better type-safety.
   */
  @DoNotCall("Use naturalOrder")
  @Deprecated

delete marker replication and delete replication should match to avoid inconsistent picture between the clusters. It is not recommended to turn on asymmetric replication - for e.g. if three sites A,B,C are participating in replication, it would be better to avoid replication setups like A -> [B, C], B -> A. In this particular example, an object uploaded to A will be replicated to B,C. If replica metadata sync is turned on in site B, any metadata updates on a replica version made in B would reflect...