* <i>similar</i>.
 *
 * <h3>Nullability</h3>
 *
 * <p>A converter always converts {@code null} to {@code null} and non-null references to non-null
 * references. It would not make sense to consider {@code null} and a non-null reference to be
 * "different representations of the same information", since one is distinguishable from

         * don't _want_ our new TypeVariable to be equal to the JDK TypeVariable because it has
         * _different bounds_ than the JDK TypeVariable. And it wouldn't make sense for our new
         * TypeVariable to be equal to any _other_ JDK TypeVariable, either, because any other JDK
         * TypeVariable must have a different declaration or name. The only TypeVariable that our

   * no service under management. Listeners would never fire because the ServiceManager was healthy
   * and stopped at the same time. This test ensures that listeners fire and isHealthy makes sense.
   */
  public void testEmptyServiceManager() {
    Logger logger = Logger.getLogger(ServiceManager.class.getName());
    logger.setLevel(Level.FINEST);
    TestLogHandler logHandler = new TestLogHandler();

   * This implements a "bursty" RateLimiter, where storedPermits are translated to zero throttling.
   * The maximum number of permits that can be saved (when the RateLimiter is unused) is defined in
   * terms of time, in this sense: if a RateLimiter is 2qps, and this time is specified as 10
   * seconds, we can save up to 2 * 10 = 20 permits.
   */
  static final class SmoothBursty extends SmoothRateLimiter {

 * supported. The elements can be any objects.
 *
 * <p>{@code contains(x)}, {@code add(x)} and {@code remove(x)}, are all (expected and amortized)
 * constant time operations. Expected in the hashtable sense (depends on the hash function doing a
 * good job of distributing the elements to the buckets to a distribution not far from uniform), and
 * amortized since some operations can trigger a hash table resize.
 *

 * <i>similar</i>.
 *
 * <h3>Nullability</h3>
 *
 * <p>A converter always converts {@code null} to {@code null} and non-null references to non-null
 * references. It would not make sense to consider {@code null} and a non-null reference to be
 * "different representations of the same information", since one is distinguishable from

 * Stats}, {@link PairedStatsAccumulator}, and {@link PairedStats}.
 *
 * @author Pete Gillin
 */
class StatsTesting {
  // TODO(cpovirk): Convince myself that this larger error makes sense.
  static final double ALLOWED_ERROR = isAndroid() ? .25 : 1e-10;

  // Inputs and their statistics:

  static final double ONE_VALUE = 12.34;

  static final double OTHER_ONE_VALUE = -56.78;

       * more information, see the comments in that class.
       *
       * We already know that that's how it behaves, and subclasses of Converter can't change that
       * behavior. So there's no sense in making all subclass authors exclude the method from any
       * NullPointerTester tests that they have.
       */
      ignoredMembers.add(Converter.class.getMethod("apply", Object.class));

   * smaller than a large array. (This assumes that in the lazy case, most stripes are unused. As
   * always, if many stripes are in use, a non-lazy striped makes more sense.)
   */
  private static final int LARGE_LAZY_CUTOFF = 1024;

  private Striped() {}

  /**
   * Returns the stripe that corresponds to the passed key. It is always guaranteed that if {@code

              <groupId>com.toasttab.android</groupId>
              <artifactId>gummy-bears-api-19</artifactId>
              <version>0.6.1</version>
              <!-- TODO(cpovirk): In principle, it would make sense to *also* test compatibility with JDK 1.8, since guava-android also has JRE users. -->
            </signature>
            <ignores>
              <!-- Unsafe isn't part of the documented Android API, but it is available.