"one", 1,
            "two", 2,
            "three", 3,
            "four", 4,
            "five", 5,
            "six", 6,
            "seven", 7,
            "eight", 8),
        "eight",
        8,
        "five",
        5,
        "four",
        4,
        "one",
        1,
        "seven",
        7,
        "six",
        6,
        "three",

    if (supportsRemove) {
      // We use a copy of "entryToRemove" in the assertion because "entryToRemove" might be
      // invalidated and have undefined behavior after entrySet.removeAll(entriesToRemove),
      // for example entryToRemove.getValue() might be null.
      Entry<K, V> entryToRemoveCopy =
          Helpers.mapEntry(entryToRemove.getKey(), entryToRemove.getValue());

          Arrays.sort(sortedKs, comparator);
          Object[] sortedValues = new Object[size];

          // We might, somehow, be able to reorder values in-place.  But it doesn't seem like
          // there's a way around creating the separate sortedKeys array, and if we're allocating
          // one array of size n, we might as well allocate two -- to say nothing of the allocation
          // done in Arrays.sort.

    this.iterator = iterator;
  }

  @Override
  public Iterator<E> iterator() {
    if (iterator == null) {
      // TODO: throw something else? Do we worry that people's code and tests
      // might be relying on this particular type of exception?
      throw new IllegalStateException();
    }
    try {
      return iterator;
    } finally {
      iterator = null;
    }
  }

     * known-non-null type? But that particular problem might not arise here, since we're not
     * actually reading from the fields in any case in which they might be null (as proven by the
     * requireNonNull checks below). Plus, we're *already* lying here, since newHeader passes a null
     * key and value, which we pass to the superconstructor, even though the key and value type for

      throw death;
    } catch (Throwable t) {
      /*
       * This is not one of AppEngine's allowed classes, so even in Sun JDKs, this can fail with
       * a NoClassDefFoundError. Other apps might deny access to sun.misc packages.
       */
      return null;
    }
  }

  /**
   * Returns the Method that can be used to resolve an individual StackTraceElement, or null if that

      throw death;
    } catch (Throwable t) {
      /*
       * This is not one of AppEngine's allowed classes, so even in Sun JDKs, this can fail with
       * a NoClassDefFoundError. Other apps might deny access to sun.misc packages.
       */
      return null;
    }
  }

  /**
   * Returns the Method that can be used to resolve an individual StackTraceElement, or null if that

  /**
   * Verify thread safety by using a collection whose size() may be inconsistent with the actual
   * number of elements and whose elements may change over time.
   *
   * <p>This test might fail in GWT because the GWT emulations might count on the input collection
   * not to change during the copy. It is safe to do so in GWT because javascript is
   * single-threaded.
   */
  @GwtIncompatible // GWT is single threaded

  /**
   * Verify thread safety by using a collection whose size() may be inconsistent with the actual
   * number of elements and whose elements may change over time.
   *
   * <p>This test might fail in GWT because the GWT emulations might count on the input collection
   * not to change during the copy. It is safe to do so in GWT because javascript is
   * single-threaded.
   */
  @GwtIncompatible // GWT is single threaded

  private final transient int size;

  /*
   * We have some considerable complexity in these create methods because of
   * Builder.buildKeepingLast(). The same Builder might be called with buildKeepingLast() and then
   * buildOrThrow(), or vice versa. So in particular, if we modify alternatingKeysAndValues to
   * eliminate duplicate keys (for buildKeepingLast()) then we have to ensure that a later call to