}

TEST_F(ShapeInferenceTest, InfersShapesFromInputTensors) {
  // Prepare some tensors for shape.
  TF_Tensor* tensor_1X6 = Int32Tensor({1, 6});
  CHECK_EQ(TF_OK, TF_GetCode(status_)) << TF_Message(status_);
  TF_Tensor* tensor_1X1X6 = Int32Tensor({1, 1, 6});
  CHECK_EQ(TF_OK, TF_GetCode(status_)) << TF_Message(status_);

    *   Add `RaggedTensor.merge_dims()`.
    *   Added new `uniform_row_length` row-partitioning tensor to
        `RaggedTensor`.
    *   Add `shape` arg to `RaggedTensor.to_tensor`; Improve speed of
        `RaggedTensor.to_tensor`.
    *   `tf.io.parse_sequence_example` and `tf.io.parse_single_sequence_example`
        now support ragged features.
    *   Fix `while_v2` with variables in custom gradient.

    ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
    TF_Tensor* expected_out0 = expected_csession->output_tensor(0);
    TF_Tensor* expected_out1 = expected_csession->output_tensor(1);

    CompareTensors(out0, expected_out0);
    CompareTensors(out1, expected_out1);
  }

  void CompareTensors(TF_Tensor* a, TF_Tensor* b) {
    float* a_data = static_cast<float*>(TF_TensorData(a));

    name = "tf_tensor_hdrs",
    hdrs = [
        "tf_tensor.h",
        "tf_tensor_helper.h",
    ],
    visibility = ["//visibility:public"],
    deps = [
        ":c_api_macros_hdrs",
        ":tf_datatype_hdrs",
        ":tf_status_headers",
        "//tensorflow/core/platform:status",
    ],
)

cc_library(
    name = "tf_tensor",
    srcs = ["tf_tensor.cc"],
    hdrs = [
        "tf_tensor.h",

}  // namespace tensorflow

static void TF_Run_Setup(int noutputs, TF_Tensor** c_outputs,
                         TF_Status* status) {
  status->status = absl::OkStatus();
  for (int i = 0; i < noutputs; ++i) {
    c_outputs[i] = nullptr;
  }
}

// TF_TensorToTensorV1 decodes a string serialization to DT_RESOURCE.
// In the TFv1 convention, TF_Tensor can hold a string serialization of

                                                     TF_Tensor** value,
                                                     TF_Status* status);

// Fills in `values` with the TF_Tensor values of the attribute `attr_name` of
// `oper`. `values` must point to an array of TF_Tensor* of length at least
// `max_values` (ideally set to TF_AttrMetadata.list_size from

        binary_proto_buf.data(), binary_proto_buf.size(), status);
    if (!status->status.ok()) return {};
    ret.push_back(UniqueFuncPtr(func, TF_DeleteFunction));
  }
  return ret;
}

TF_Tensor* TF_DequeueNamedTensor(TF_Session* session, int tensor_id,
                                 TF_Status* status) {
  assert(session);
  {
    tensorflow::mutex_lock c(session->graph->mu);

// empty, this call is blocked.
//
// Tensors are enqueued via the corresponding TF enqueue op.
// TODO(hongm): Add support for `timeout_ms`.
TF_CAPI_EXPORT extern TF_Tensor* TF_DequeueNamedTensor(TF_Session* session,
                                                       int tensor_id,
                                                       TF_Status* status);

  cp tensorflow/c/c_api.h \
    tensorflow/c/tf_attrtype.h \
    tensorflow/c/tf_buffer.h  \
    tensorflow/c/tf_datatype.h \
    tensorflow/c/tf_status.h \
    tensorflow/c/tf_tensor.h \
    tensorflow/c/tf_tensor_helper.h \
    tensorflow/c/tf_tstring.h \
    tensorflow/c/tf_file_statistics.h \
    tensorflow/c/tensor_interface.h \
    tensorflow/c/c_api_macros.h \

  ~CApiFunctionTest() override {
    TF_DeleteFunction(func_);
    TF_DeleteGraph(host_graph_);
    TF_DeleteGraph(func_graph_);
    TF_DeleteStatus(s_);
  }

  void Run(const std::vector<std::pair<TF_Operation*, TF_Tensor*>>& inputs,
           TF_Operation* output, int32_t expected_result) {
    Run(inputs, {{output, 0}}, {expected_result});
  }

  // Run the host graph, which now contains a function and check that