*
 * <ul>
 *   <li>whether threads should be marked as {@linkplain Thread#setDaemon daemon} threads
 *   <li>a {@linkplain ThreadFactoryBuilder#setNameFormat naming format}
 *   <li>a {@linkplain Thread#setPriority thread priority}
 *   <li>an {@linkplain Thread#setUncaughtExceptionHandler uncaught exception handler}
 *   <li>a {@linkplain ThreadFactory#newThread backing thread factory}
 * </ul>
 *

   *
   * <p>Among the improvements that this method and {@code LoadingCache.get(K)} both provide are:
   *
   * <ul>
   *   <li>{@linkplain LoadingCache#get(Object) awaiting the result of a pending load} rather than
   *       starting a redundant one
   *   <li>eliminating the error-prone caching boilerplate
   *   <li>tracking load {@linkplain #stats statistics}
   * </ul>
   *

 * and performant but introduces some testing difficulties. This test exercises the two fallback
 * strategies.
 *
 * <ul>
 *   <li>SafeAtomicHelper: uses Atomic FieldsUpdaters to implement synchronization
 *   <li>SynchronizedHelper: uses {@code synchronized} blocks for synchronization
 * </ul>
 *
 * To force selection of our fallback strategies we load {@link AggregateFutureState} (and all of

 * bimap containing the same entries as this bimap but with reversed keys and values.
 *
 * <h3>Implementations</h3>
 *
 * <ul>
 *   <li>{@link ImmutableBiMap}
 *   <li>{@link HashBiMap}
 *   <li>{@link EnumBiMap}
 *   <li>{@link EnumHashBiMap}
 * </ul>
 *
 * <p>See the Guava User Guide article on <a href=
 * "https://github.com/google/guava/wiki/NewCollectionTypesExplained#bimap">{@code BiMap}</a>.

 * actions taken by the java garbage collection system:
 *
 * <ul>
 *   <li>invoking the {@code finalize} methods of unreachable objects
 *   <li>clearing weak references to unreachable referents
 *   <li>enqueuing weak references to unreachable referents in their reference queue
 * </ul>
 *

 *   <li>Convert multiple instances "forward" using {@code converter.convertAll(as)}.
 *   <li>Convert in the "backward" direction using {@code converter.reverse().convert(b)} or {@code
 *       converter.reverse().convertAll(bs)}.
 *   <li>Use {@code converter} or {@code converter.reverse()} anywhere a {@link
 *       java.util.function.Function} is accepted (for example {@link java.util.stream.Stream#map
 *       Stream.map}).

 * webservers, RPC servers and timers.
 *
 * <p>The normal lifecycle of a service is:
 *
 * <ul>
 *   <li>{@linkplain State#NEW NEW} -&gt;
 *   <li>{@linkplain State#STARTING STARTING} -&gt;
 *   <li>{@linkplain State#RUNNING RUNNING} -&gt;
 *   <li>{@linkplain State#STOPPING STOPPING} -&gt;
 *   <li>{@linkplain State#TERMINATED TERMINATED}
 * </ul>
 *

   *
   * <ol>
   *   <li>The state of the RateLimiter (storedPermits) is a vertical line in this figure.
   *   <li>When the RateLimiter is not used, this goes right (up to maxPermits)
   *   <li>When the RateLimiter is used, this goes left (down to zero), since if we have
   *       storedPermits, we serve from those first

 *
 * <p>Some uses of this class include
 *
 * <ul>
 *   <li>As a method return type, as an alternative to returning {@code null} to indicate that no
 *       value was available
 *   <li>To distinguish between "unknown" (for example, not present in a map) and "known to have no
 *       value" (present in the map, with value {@code Optional.absent()})

 *       defined this way, but enough to live with.
 *   <li>If you override {@link #setUp} or {@link #tearDown}, make sure to invoke {@code
 *       super.setUp} and {@code super.tearDown} within them. These methods are used to clear and
 *       check for thread assertion failures.
 *   <li>All delays and timeouts must use one of the constants {@code SHORT_DELAY_MS}, {@code