this.buffer = ByteBuffer.allocate(bufferSize + 7).order(ByteOrder.LITTLE_ENDIAN);
    this.bufferSize = bufferSize;
    this.chunkSize = chunkSize;
  }

  /** Processes the available bytes of the buffer (at most {@code chunk} bytes). */
  protected abstract void process(ByteBuffer bb);

  /**
   * This is invoked for the last bytes of the input, which are not enough to fill a whole chunk.

   *
   * Note well that if, for this function, we chose a horizontal line, at height of exactly (1/QPS),
   * then the effect of the function is non-existent: we serve storedPermits at exactly the same
   * cost as fresh ones (1/QPS is the cost for each). We use this trick later.
   *
   * If we pick a function that goes /below/ that horizontal line, it means that we reduce the area

   * are guarded by a lock and operations are applied in batches to avoid lock contention. The
   * penalty of applying the batches is spread across threads so that the amortized cost is slightly
   * higher than performing just the operation without enforcing the capacity constraint.
   *
   * This implementation uses a per-segment queue to record a memento of the additions, removals,

 *       scalability or other performance factors (see the separate "jtreg" tests for a set intended
 *       to check these for the most central aspects of functionality.) So, most tests use the
 *       smallest sensible numbers of threads, collection sizes, etc needed to check basic
 *       conformance.

@GwtCompatible
@ElementTypesAreNonnullByDefault
public abstract class ForwardingCollection<E extends @Nullable Object> extends ForwardingObject
    implements Collection<E> {
  // TODO(lowasser): identify places where thread safety is actually lost

  /** Constructor for use by subclasses. */
  protected ForwardingCollection() {}

  @Override
  protected abstract Collection<E> delegate();

  @Override
  public Iterator<E> iterator() {

        if (!(localValue instanceof SetFuture)) {
          // obj cannot be null at this point, because value can only change from null to non-null.
          // So if value changed (and it did since we lost the CAS), then it cannot be null and
          // since it isn't a SetFuture, then the future must be done and we should exit the loop
          break;
        }
      }
    }
    return rValue;
  }

 *       in any other way than how {@code Stopwatch} uses them.
 *   <li>An alternative source of nanosecond ticks can be substituted, for example for testing or
 *       performance reasons, without affecting most of your code.
 * </ul>
 *
 * <p>The one downside of {@code Stopwatch} relative to {@link System#nanoTime()} is that {@code
 * Stopwatch} requires object allocation and additional method calls, which can reduce the accuracy

   * MoreExecutors#directExecutor}. Otherwise, avoid it: See the warnings on the docs for {@code
   * directExecutor}.
   *
   * <p>This is the most general listener interface. For common operations performed using
   * listeners, see {@link Futures}. For a simplified but general listener interface, see {@link
   * Futures#addCallback addCallback()}.
   *

      checkEmpty(entrySet);
      checkEmpty(cache);
    }
  }

  /* ---------------- Local utilities -------------- */

  /** Most of the tests in this class run against every one of these caches. */
  private Iterable<LoadingCache<Object, Object>> caches() {
    // lots of different ways to configure a LoadingCache

   * comparable</a>. If you pass objects that are not mutually comparable, this method may throw an
   * exception. (The reason for this decision is lost to time, but the reason <i>might</i> be that
   * we wanted to support legacy classes that implement the raw type {@code Comparable} (instead of