of those modules are ones that it uses but forgets to list in its own <inherits>, we'd
            like to get an error. Currently we do, but if we add the extra <inherits> lines, we
            won't.

            I have one idea for a better approach, but it's painful, and I haven't tested it: We
            could postprocess Collect.gwt.xml to add <skip> lines for all the files that should be

 * com.google.common.base.FinalizableReference}.
 *
 * <p>While this class is public, we consider it to be *internal* and not part of our published API.
 * It is public so we can access it reflectively across class loaders in secure environments.
 *
 * <p>This class can't depend on other Guava code. If we were to load this class in the same class

 * </ul>
 *
 * To force selection of our fallback strategies we load {@link AbstractFuture} (and all of {@code
 * com.google.common.util.concurrent}) in degenerate class loaders which make certain platform
 * classes unavailable. Then we construct a test suite so we can run the normal AbstractFutureTest
 * test methods in these degenerate classloaders.
 */

    // we pad because the dummy hash function we use to test this, merely copies the input into
    // the output, so the input must be at least 32 bits, since the output has to be that long
    assertPutString(new char[] {'p', HashTestUtils.randomLowSurrogate(new Random())});
  }

  public void testPutStringWithHighSurrogate() {
    // we pad because the dummy hash function we use to test this, merely copies the input into

    // we pad because the dummy hash function we use to test this, merely copies the input into
    // the output, so the input must be at least 32 bits, since the output has to be that long
    assertPutString(new char[] {'p', HashTestUtils.randomLowSurrogate(new Random())});
  }

  public void testPutStringWithHighSurrogate() {
    // we pad because the dummy hash function we use to test this, merely copies the input into

   * reclaimed.
   *
   * If any of this fails along the way, we fall back to loading Finalizer directly in the
   * application class loader.
   *
   * NOTE: The tests for this behavior (FinalizableReferenceQueueClassLoaderUnloadingTest) fail
   * strangely when run in JDK 9. We are considering this a known issue. Please see
   * https://github.com/google/guava/issues/3086 for more information.

   * reclaimed.
   *
   * If any of this fails along the way, we fall back to loading Finalizer directly in the
   * application class loader.
   *
   * NOTE: The tests for this behavior (FinalizableReferenceQueueClassLoaderUnloadingTest) fail
   * strangely when run in JDK 9. We are considering this a known issue. Please see
   * https://github.com/google/guava/issues/3086 for more information.

      }
      updateLastKnownIndex();
      /*
       * If the entry has been removed from the map, we return null, even though that might not be a
       * valid value. That's the best we can do, short of holding a reference to the most recently
       * seen value. And while we *could* do that, we aren't required to: Map.Entry explicitly says

      tmp += UnsignedLongs.remainder(longs[j], divisors[j]);
    }
    return tmp;
  }

  @Benchmark
  long parseUnsignedLong(int reps) {
    long tmp = 0;
    // Given that we make three calls per pass, we scale reps down in order
    // to do a comparable amount of work to other measurements.
    int scaledReps = reps / 3 + 1;
    for (int i = 0; i < scaledReps; i++) {
      int j = i & ARRAY_MASK;

  }

  /*
   * TODO(b/242884182): Figure out why this helps produce the same class file when we compile most
   * of common.collect a second time with the results of the first compilation on the classpath. Or
   * just back this out once we stop doing that (which we'll do after our internal GWT setup
   * changes).
   */
  @Override
  public ImmutableMap<Object, Collection<Object>> asMap() {