SINGLE_CASE(kF, TF_ATTR_FLOAT, -1);
    SINGLE_CASE(kB, TF_ATTR_BOOL, -1);
    SINGLE_CASE(kType, TF_ATTR_TYPE, -1);
    SINGLE_CASE(kShape, TF_ATTR_SHAPE,
                attr->shape().unknown_rank() ? -1 : attr->shape().dim_size());
    SINGLE_CASE(kTensor, TF_ATTR_TENSOR, -1);
#undef SINGLE_CASE

    case tensorflow::AttrValue::kList:
      metadata.is_list = 1;
      metadata.list_size = 0;

  // scheduled on that same threadpool, causing a deadlock in cases where the
  // caller of event_mgr->ThenExecute() blocks on the completion of the callback
  // (as in the case of ConstOp kernel creation on GPU, which involves copying a
  // CPU tensor to GPU).
  // Setting a larger thread pool does not help with the Swift caller, as we use
  // a different TFE context for each thread of execution (for running graph

  // until the previously pending operation (a StartExecute without a Join) has
  // finished, if any.
  //
  // `cancellation_manager` must live until after `Join` finishes and pending
  // `is_async` operations finish. In addition to allowing the caller to cancel
  // the operation, its `StartCancel` method will be called if op execution
  // fails on any device in order to cancel the others.

TEST(CAPI, DataTypeEnum) {
  EXPECT_EQ(TF_FLOAT, static_cast<TF_DataType>(tensorflow::DT_FLOAT));
  EXPECT_EQ(TF_DOUBLE, static_cast<TF_DataType>(tensorflow::DT_DOUBLE));
  EXPECT_EQ(TF_INT32, static_cast<TF_DataType>(tensorflow::DT_INT32));
  EXPECT_EQ(TF_UINT8, static_cast<TF_DataType>(tensorflow::DT_UINT8));
  EXPECT_EQ(TF_INT16, static_cast<TF_DataType>(tensorflow::DT_INT16));
  EXPECT_EQ(TF_INT8, static_cast<TF_DataType>(tensorflow::DT_INT8));