*       cancelled, cancellation never propagates to a task that has started to run -- neither to
 *       the callable itself nor to any {@code Future} returned by an {@code AsyncCallable}.
 *       (However, cancellation can prevent an <i>unstarted</i> task from running.) Therefore, the
 *       next task will wait for any running callable (or pending {@code Future} returned by an

   * key set contains a key if and only if this multimap maps that key to at least one value.
   *
   * <p>Changes to the returned set will update the underlying multimap, and vice versa. However,
   * <i>adding</i> to the returned set is not possible.
   */
  Set<K> keySet();

  /**
   * Returns a view collection containing the key from each key-value pair in this multimap,

       *
       * My impression is that an AtomicReferenceFieldUpdater in a static field is similarly fast to
       * Unsafe on modern JVMs (if perhaps not quite as fast as VarHandle?). However, I'm not sure
       * exactly what we've benchmarked, and we certainly haven't benchmarked as far back as JDK 8.
       * (We also haven't benchmarked under Android. We continue to use UnsafeAtomicHelper there so

   * bug</b> as described in the deprecation text.
   *
   * <p>The C++ equivalent is the MurmurHash3_x86_32 function (Murmur3A), which however does not
   * have the bug.
   *
   * @deprecated This implementation produces incorrect hash values from the {@link
   *     HashFunction#hashString} method if the string contains non-BMP characters. Use {@link

   * xStats().populationVariance() > 0.0 && yStats().populationVariance() > 0.0}). The result is not
   * guaranteed to be exactly +/-1 even when the data are perfectly (anti-)correlated, due to
   * numerical errors. However, it is guaranteed to be in the inclusive range [-1, +1].
   *
   * <h3>Non-finite values</h3>
   *
   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link

   *
   * <p>This method is significantly faster than {@link String#equalsIgnoreCase} and should be used
   * in preference if at least one of the parameters is known to contain only ASCII characters.
   *
   * <p>Note however that this method does not always behave identically to expressions such as:
   *
   * <ul>
   *   <li>{@code string.toUpperCase().equals("UPPER CASE ASCII")}
   *   <li>{@code string.toLowerCase().equals("lower case ascii")}

   * <p>This is guaranteed to return zero if the dataset contains only exactly one finite value. It
   * is not guaranteed to return zero when the dataset consists of the same value multiple times,
   * due to numerical errors. However, it is guaranteed never to return a negative result.
   *
   * <h3>Non-finite values</h3>
   *
   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link

      if (currentSubscribers == null || !currentSubscribers.removeAll(listenerMethodsForType)) {
        // if removeAll returns true, all we really know is that at least one subscriber was
        // removed... however, barring something very strange we can assume that if at least one
        // subscriber was removed, all subscribers on listener for that event type were... after

   * <p>This is guaranteed to return zero if the dataset contains only exactly one finite value. It
   * is not guaranteed to return zero when the dataset consists of the same value multiple times,
   * due to numerical errors. However, it is guaranteed never to return a negative result.
   *
   * <h3>Non-finite values</h3>
   *
   * <p>If the dataset contains any non-finite values ({@link Double#POSITIVE_INFINITY}, {@link

 * {@code asMap} has a few subtleties. As long as the set of keys remains unchanged, adding or
 * removing mappings does not affect the key iteration order. However, if you remove all values
 * associated with a key and then add the key back to the multimap, that key will come last in the
 * key iteration order.
 *