* other callers have added to it.
       *
       * This read is guaranteed to get us the right value because we only set this once (here).
       *
       * requireNonNull is safe because either our compareAndSet succeeded or it failed because
       * another thread did it for us.
       */
      seenExceptionsLocal = requireNonNull(seenExceptions);
    }
    return seenExceptionsLocal;

    return wrapList(key, (List<V>) collection, null);
  }

  // Following Javadoc copied from ListMultimap.

  /**
   * {@inheritDoc}
   *
   * <p>Because the values for a given key may have duplicates and follow the insertion ordering,
   * this method returns a {@link List}, instead of the {@link Collection} specified in the {@link
   * Multimap} interface.
   */
  @Override

    return wrapList(key, (List<V>) collection, null);
  }

  // Following Javadoc copied from ListMultimap.

  /**
   * {@inheritDoc}
   *
   * <p>Because the values for a given key may have duplicates and follow the insertion ordering,
   * this method returns a {@link List}, instead of the {@link Collection} specified in the {@link
   * Multimap} interface.
   */
  @Override

    /**
     * This event means that the artifactScope has NOT been updated to a farther node artifactScope because current
     * node is in the first level pom
     *
     * @param artifact     current node artifact, the one in the first level pom
     * @param ignoredScope artifactScope that was ignored because artifact was in first level pom
     */
    void updateScopeCurrentPom(Artifact artifact, String ignoredScope);

  /**
   * {@inheritDoc}
   *
   * <p>Because a {@code SetMultimap} has unique values for a given key, this method returns a
   * {@link Set}, instead of the {@link java.util.Collection} specified in the {@link Multimap}
   * interface.
   */
  @Override
  Set<V> get(@ParametricNullness K key);

  /**
   * {@inheritDoc}
   *

     * least old versions), so we mostly do. This is useful because we don't actually run our CI on
     * Windows under Java 8, at least as of this writing.
     *
     * Under Windows in particular, we want to test that:
     *
     * - Under Java 9+, createTempDir() succeeds because it can look up the *real* username, rather
     * than relying on the one from the system property.
     *

        "A device attached to the system is not functioning.",
        "The process cannot access the file because it is being used by another process.",
        "The process cannot access the file because it is being used by another process.",
        "The process cannot access the file because another process has locked a portion of the file.",
        "The disk is full.",
        "A duplicate name exists on the network.",

   *   even back to `this.seq`. However, that may suggest that we're defending against concurrent
   *   mutation, which is not an actual risk because we use `synchronized`.
   * - Make `remaining` require a non-null `seq` argument. But this is a bit weird because the
   *   method, while it would avoid the instance field `seq` would still access the instance field
   *   `pos`.
   */

  @Override

 * Contrast the method above to:
 *
 * <ul>
 *   <li>methods whose return type is a type variable but which can never return {@code null},
 *       typically because the type forbids nullable type arguments: For example, {@code
 *       ImmutableList.get} returns {@code E}, but that value is never {@code null}. (Accordingly,

 * Contrast the method above to:
 *
 * <ul>
 *   <li>methods whose return type is a type variable but which can never return {@code null},
 *       typically because the type forbids nullable type arguments: For example, {@code
 *       ImmutableList.get} returns {@code E}, but that value is never {@code null}. (Accordingly,