* Null row keys, columns keys, and values are not supported.
 *
 * <p>Lookups by row key are often faster than lookups by column key, because the data is stored in
 * a {@code Map<R, Map<C, V>>}. A method call like {@code column(columnKey).get(rowKey)} still runs
 * quickly, since the row key is provided. However, {@code column(columnKey).size()} takes longer,

          double z = aa.getAndAccumulate(i, y, Double::sum);
          assertBitEquals(x, z);
          assertBitEquals(x + y, aa.get(i));
        }
      }
    }
  }

  /** getAndAccumulate with max stores max of given value to current, and returns previous value */
  public void testGetAndAccumulateWithMax() {
    AtomicDoubleArray aa = new AtomicDoubleArray(SIZE);
    for (int i : new int[] {0, SIZE - 1}) {

   * {@link Set}, instead of the {@link java.util.Collection} specified in the {@link Multimap}
   * interface.
   *
   * <p>Any duplicates in {@code values} will be stored in the multimap once.
   */
  @CanIgnoreReturnValue
  @Override
  Set<V> replaceValues(@ParametricNullness K key, Iterable<? extends V> values);

  /**
   * {@inheritDoc}
   *

  //   completion preferably and AQS is non-deterministic (depends on where in the queue the waiter
  //   is). If we wanted to be strict about it, we could store the unpark() time in the Waiter node
  //   and we could use that to make a decision about whether or not we timed out prior to being
  //   unparked.

  /**
   * {@inheritDoc}
   *

  }

  /*
   * We don't define getPredecessor+getSuccessor and setPredecessor+setSuccessor here because
   * they're defined above -- including logic to add and subtract 1 to map between the values stored
   * in the predecessor/successor arrays and the indexes in the elements array that they identify.
   */

  }

  /*
   * We don't define getPredecessor+getSuccessor and setPredecessor+setSuccessor here because
   * they're defined above -- including logic to add and subtract 1 to map between the values stored
   * in the predecessor/successor arrays and the indexes in the elements array that they identify.
   */

    implements Serializable {

  /**
   * Creates an {@code ArrayTable} filled with {@code null}.
   *
   * @param rowKeys row keys that may be stored in the generated table
   * @param columnKeys column keys that may be stored in the generated table
   * @throws NullPointerException if any of the provided keys is null
   * @throws IllegalArgumentException if {@code rowKeys} or {@code columnKeys} contains duplicates

    static final int DEFAULT_INITIAL_CAPACITY = 4;

    static int expandedCapacity(int oldCapacity, int minCapacity) {
      if (minCapacity < 0) {
        throw new AssertionError("cannot store more than MAX_VALUE elements");
      }
      // careful of overflow!
      int newCapacity = oldCapacity + (oldCapacity >> 1) + 1;
      if (newCapacity < minCapacity) {

   * To maintain this code, make a change for the strong reference type. Then, cut and paste, and
   * replace "Strong" with "Soft" or "Weak" within the pasted text. The primary difference is that
   * strong entries store the key reference directly while soft and weak entries delegate to their
   * respective superclasses.
   */

  /** Used for strongly-referenced keys. */

   * by {@code factory}. Those objects should not be manually updated and they should not use soft,
   * weak, or phantom references.
   *
   * @param backingMap place to store the mapping from each row key to its corresponding column key
   *     / value map
   * @param factory supplier of new, empty maps that will each hold all column key / value mappings
   *     for a given row key