* without having to add {@code @SuppressWarnings}, the code can call this method.
   *
   * <p>Why <i>not</i> just add {@code SuppressWarnings}? The problem is that this method is
   * typically useful for {@code return} statements. That leaves the code with two options: Either
   * add the suppression to the whole method (which turns off checking for a large section of code),

   * without having to add {@code @SuppressWarnings}, the code can call this method.
   *
   * <p>Why <i>not</i> just add {@code SuppressWarnings}? The problem is that this method is
   * typically useful for {@code return} statements. That leaves the code with two options: Either
   * add the suppression to the whole method (which turns off checking for a large section of code),

   * without having to add {@code @SuppressWarnings}, the code can call this method.
   *
   * <p>Why <i>not</i> just add {@code SuppressWarnings}? The problem is that this method is
   * typically useful for {@code return} statements. That leaves the code with two options: Either
   * add the suppression to the whole method (which turns off checking for a large section of code),

   * Collections#reverseOrder(Comparator)}.
   *
   * <p><b>Java 8+ users:</b> Use {@code thisComparator.reversed()} instead.
   */
  // type parameter <S> lets us avoid the extra <String> in statements like:
  // Ordering<String> o = Ordering.<String>natural().reverse();
  @GwtCompatible(serializable = true)
  public <S extends T> Ordering<S> reverse() {
    return new ReverseOrdering<>(this);
  }

  /**

   *
   * <p>This method always throws an exception. The {@code RuntimeException} return type allows
   * client code to signal to the compiler that statements after the call are unreachable. Example
   * usage:
   *
   * <pre>
   * T doSomething() {
   *   try {
   *     return someMethodThatCouldThrowAnything();
   *   } catch (IKnowWhatToDoWithThisException e) {

   *
   * <p>This method always throws an exception. The {@code RuntimeException} return type allows
   * client code to signal to the compiler that statements after the call are unreachable. Example
   * usage:
   *
   * <pre>
   * T doSomething() {
   *   try {
   *     return someMethodThatCouldThrowAnything();
   *   } catch (IKnowWhatToDoWithThisException e) {

          while (top.hasNext()) {
            N element = top.next();
            // requireNonNull is safe because horizon contains only graph nodes.
            /*
             * TODO(cpovirk): Replace these two statements with one (`N element =
             * requireNonNull(top.next())`) once our checker supports it.
             *
             * (The problem is likely

    long guess = (long) Math.sqrt((double) x);
    // Note: guess is always <= FLOOR_SQRT_MAX_LONG.
    long guessSquared = guess * guess;
    // Note (2013-2-26): benchmarks indicate that, inscrutably enough, using if statements is
    // faster here than using lessThanBranchFree.
    switch (mode) {
      case UNNECESSARY:
        checkRoundingUnnecessary(guessSquared == x);
        return guess;
      case FLOOR:
      case DOWN:

   * without having to add {@code @SuppressWarnings}, the code can call this method.
   *
   * <p>Why <i>not</i> just add {@code SuppressWarnings}? The problem is that this method is
   * typically useful for {@code return} statements. That leaves the code with two options: Either
   * add the suppression to the whole method (which turns off checking for a large section of code),

   *
   * <p>The use of this class is considered an implementation detail of ServiceManager and as such
   * it is excluded from {@link #servicesByState}, {@link #startupTimes}, {@link #toString} and all
   * logging statements.
   */
  private static final class NoOpService extends AbstractService {
    @Override
    protected void doStart() {
      notifyStarted();
    }

    @Override
    protected void doStop() {