* As such, this field must be cleared before every map mutation.
   *
   * Note about volatile: volatile doesn't make it safe to read from a mutable graph in one thread
   * while writing to it in another. All it does is help with _reading_ from multiple threads
   * concurrently. For more information, see AbstractNetworkTest.concurrentIteration.
   */
  @CheckForNull private transient volatile Entry<K, V> cacheEntry;

  /**
   * Adds {@code edge} connecting {@code nodeU} to {@code nodeV}.
   *
   * <p>If the graph is directed, {@code edge} will be directed in this graph; otherwise, it will be
   * undirected.
   *
   * <p><b>{@code edge} must be unique to this graph</b>, just as a {@code Map} key must be. It must
   * also be non-null.
   *
   * <p>If {@code nodeU} and {@code nodeV} are not already present in this graph, this method will

 * using a lock, however it is dangerous to dispatch listeners while holding a lock because they
 * might run on the {@code directExecutor()} or be otherwise re-entrant (call back into your
 * object). So it is important to not call {@link #dispatch} while holding any locks. This is why
 * {@link #enqueue} and {@link #dispatch} are 2 different methods. It is expected that the decision

 * </ul>
 *
 * <p>Consumers of this annotation include:
 *
 * <ul>
 *   <li>Kotlin, for which it makes the type-variable usage (a) a Kotlin platform type when the type
 *       argument is non-nullable and (b) nullable when the type argument is nullable. We use this
 *       to "undo" {@link ElementTypesAreNonnullByDefault}. It is the best we can do for Kotlin
 *       under our current constraints.
 *   <li>NullAway, which will <a

 * </ul>
 *
 * <p>Consumers of this annotation include:
 *
 * <ul>
 *   <li>Kotlin, for which it makes the type-variable usage (a) a Kotlin platform type when the type
 *       argument is non-nullable and (b) nullable when the type argument is nullable. We use this
 *       to "undo" {@link ElementTypesAreNonnullByDefault}. It is the best we can do for Kotlin
 *       under our current constraints.
 *   <li>NullAway, which will <a

 * </ul>
 *
 * <p>Consumers of this annotation include:
 *
 * <ul>
 *   <li>Kotlin, for which it makes the type-variable usage (a) a Kotlin platform type when the type
 *       argument is non-nullable and (b) nullable when the type argument is nullable. We use this
 *       to "undo" {@link ElementTypesAreNonnullByDefault}. It is the best we can do for Kotlin
 *       under our current constraints.
 *   <li>NullAway, which will <a

 * </ul>
 *
 * <p>Consumers of this annotation include:
 *
 * <ul>
 *   <li>Kotlin, for which it makes the type-variable usage (a) a Kotlin platform type when the type
 *       argument is non-nullable and (b) nullable when the type argument is nullable. We use this
 *       to "undo" {@link ElementTypesAreNonnullByDefault}. It is the best we can do for Kotlin
 *       under our current constraints.
 *   <li>NullAway, which will <a

  /*
   * If I understand correctly:
   *
   * This needs to be a @JsMethod so that J2CL knows to look for a JavaScript implemention of
   * it in Platform.native.js. (The JavaScript implementation inline below is visible to *GWT*, but
   * *J2CL* doesn't look at it.)
   *
   * However, once it's a @JsMethod, GWT produces a warning. That's because (a) the *other* purpose

 * </ul>
 *
 * <p>Consumers of this annotation include:
 *
 * <ul>
 *   <li>Kotlin, for which it makes the type-variable usage (a) a Kotlin platform type when the type
 *       argument is non-nullable and (b) nullable when the type argument is nullable. We use this
 *       to "undo" {@link ElementTypesAreNonnullByDefault}. It is the best we can do for Kotlin
 *       under our current constraints.
 *   <li>NullAway, which will <a

  // Override hashCode() to access delegate directly (so that it doesn't trigger the validate() call
  // via delegate()); it seems inappropriate to throw ISE on this method.
  @Override
  public int hashCode() {
    return delegate.hashCode();
  }

  private void validate() {
    // Don't use checkState(), because we don't want the overhead of generating the error message