* perform well, {@code function} should be fast. To avoid lazy evaluation when the returned
   * multimap doesn't need to be a view, copy the returned multimap into a new multimap of your
   * choosing.
   *
   * @since 7.0
   */
  public static <
          K extends @Nullable Object, V1 extends @Nullable Object, V2 extends @Nullable Object>
      Multimap<K, V2> transformValues(

    }
  }

  // In theory we can use PartitionedCall if the XLA cluster does not have any
  // stateful operations.  However, for now we choose to be conservative since
  // we don't have any evidence that choosing a stateless partitioned call helps
  // for performance.
  ops::StatefulPartitionedCall call(
      root.WithOpName("stateful_partitioned_call"), args, n->output_types(),

include::sample[dir="snippets/dependencyManagement/modelingFeatures-crossProjectPublications-advanced/groovy",files="producer/build.gradle[tags=declare-outgoing-configuration]"]
====

[NOTE]
====
Choosing the right attributes to set is the hardest thing in this process, because they carry the semantics of the variant.

 * lock, but also maximizes memory footprint. On the other extreme, one could use a single lock for
 * all tasks, which minimizes memory footprint but also minimizes concurrency. Instead of choosing
 * either of these extremes, {@code Striped} allows the user to trade between required concurrency
 * and memory footprint. For example, if a set of tasks are CPU-bound, one could easily create a

   * addition, the cause of any {@code ExecutionException} is wrapped in order to ensure that the
   * new stack trace matches that of the current thread.
   *
   * <p>Instances of {@code exceptionClass} are created by choosing an arbitrary public constructor
   * that accepts zero or more arguments, all of type {@code String} or {@code Throwable}
   * (preferring constructors with at least one {@code String}, then preferring constructors with at

 * lock, but also maximizes memory footprint. On the other extreme, one could use a single lock for
 * all tasks, which minimizes memory footprint but also minimizes concurrency. Instead of choosing
 * either of these extremes, {@code Striped} allows the user to trade between required concurrency
 * and memory footprint. For example, if a set of tasks are CPU-bound, one could easily create a

	// i.e. the out object cannot be mutated without mutating the in object as well.
	// This method is similar to Convert() but handles specific details of choosing the correct
	// output version.
	ConvertToVersion(in Object, gv GroupVersioner) (out Object, err error)
	ConvertFieldLabel(gvk schema.GroupVersionKind, label, value string) (string, string, error)
}

// happens at the very beginning of the loop.
func TestIssue37499(t *testing.T) {
	// Choose u and v such that v is slightly larger than u >> N.
	// This tricks divBasic into choosing 1 as the first word
	// of the quotient. This works in both 32-bit and 64-bit settings.
	u := natFromString("0x2b6c385a05be027f5c22005b63c42a1165b79ff510e1706b39f8489c1d28e57bb5ba4ef9fd9387a3e344402c0a453381")

NODE_SIZE=${NODE_SIZE:-e2-standard-2}
NUM_NODES=${NUM_NODES:-3}
NUM_WINDOWS_NODES=${NUM_WINDOWS_NODES:-0}
MASTER_SIZE=${MASTER_SIZE:-e2-standard-$(get-master-size)}
MASTER_MIN_CPU_ARCHITECTURE=${MASTER_MIN_CPU_ARCHITECTURE:-} # To allow choosing better architectures.
export MASTER_DISK_TYPE=pd-ssd
MASTER_DISK_SIZE=${MASTER_DISK_SIZE:-$(get-master-disk-size)}
MASTER_ROOT_DISK_SIZE=${MASTER_ROOT_DISK_SIZE:-$(get-master-root-disk-size)}

   * well, {@code function} should be fast. To avoid lazy evaluation when the returned table doesn't
   * need to be a view, copy the returned table into a new table of your choosing.
   *
   * @since 10.0
   */
  public static <
          R extends @Nullable Object,
          C extends @Nullable Object,
          V1 extends @Nullable Object,
          V2 extends @Nullable Object>