int version = TFE_CUSTOM_DEVICE_VERSION;
  // Method to copy a tensor to the custom device.
  TFE_TensorHandle* (*copy_tensor_to_device)(TFE_Context* context,
                                             TFE_TensorHandle* tensor,
                                             TF_Status* status,
                                             void* device_info);

  // Method to copy a tensor from the custom device to a target device.

#include <memory>

#include "absl/types/optional.h"
#include "absl/types/span.h"
#include "tensorflow/c/eager/abstract_operation.h"
#include "tensorflow/c/eager/immediate_execution_tensor_handle.h"
#include "tensorflow/c/tensor_interface.h"
#include "tensorflow/core/framework/cancellation.h"
#include "tensorflow/core/framework/device_attributes.pb.h"
#include "tensorflow/core/framework/op_def.pb.h"

  TF_DeleteStatus(s);
  return NULL;
}

// A compute function. This will never actually get called in this test, it's
// just nice to know that it compiles.
void compute(void* kernel, TF_OpKernelContext* ctx) {
  TF_Tensor* input;
  TF_Status* s = TF_NewStatus();
  TF_GetInput(ctx, 0, &input, s);
  TF_DeleteTensor(input);
  TF_DeleteStatus(s);
}

// Exercises tensorflow's C API.
int main(int argc, char** argv) {

    std::vector<TF_Output> graph_outputs;
    graph_outputs.reserve(outputs->outputs.size());
    for (auto* abstract_output : outputs->outputs) {
      GraphTensor* output = dyn_cast<GraphTensor>(abstract_output);
      if (!output) {
        return errors::Unimplemented(
            "Returning a non-graph tensor from a function has not "
            "been implemented yet.");

#include "tensorflow/c/tf_status_helper.h"
#include "tensorflow/core/platform/env.h"
#include "tensorflow/core/platform/path.h"
#include "tensorflow/core/platform/types.h"

struct TF_StringStream {
  std::vector<::tensorflow::string>* list;
  size_t position;
};

void TF_CreateDir(const char* dirname, TF_Status* status) {
  TF_SetStatus(status, TF_OK, "");
  ::tensorflow::Set_TF_Status_from_Status(

      }
    }
  }

 private:
  // TODO(b/174778737): Only hold needed inputs.
  vector<AbstractTensorHandle*> forward_inputs_;
};

class DivNoNanGradientFunction : public GradientFunction {
 public:
  explicit DivNoNanGradientFunction(vector<AbstractTensorHandle*> f_inputs,
                                    vector<AbstractTensorHandle*> f_outputs)
      : forward_inputs_(f_inputs), forward_outputs_(f_outputs) {

    return nullptr;
  }

  return new TF_Tensor{t};
}

TFE_TensorHandle* TFE_NewTensorHandleFromTensor(TFE_Context* ctx, TF_Tensor* t,
                                                TF_Status* status) {
  return tensorflow::wrap(
      tensorflow::unwrap(ctx)->CreateLocalHandle(t->tensor));
}

TFE_TensorHandle* TFE_CreatePackedTensorHandle(TFE_Context* ctx,

int TF_OutputListNumOutputs(TF_OutputList* o);
// Return the `i`th output in the list.
TF_AbstractTensor* TF_OutputListGet(TF_OutputList* o, int i);
// Append a tensor at the end of the output list, growing its size by one.
void TF_OutputListPushBack(TF_OutputList* o, TF_AbstractTensor* tensor,
                           TF_Status*);

// TF_ExecuteOperation will, if in eager mode, execute, if in graph mode, maybe

        ],
    }),
)

cc_library(
    name = "immediate_execution_tensor_handle",
    srcs = ["immediate_execution_tensor_handle.cc"],
    hdrs = ["immediate_execution_tensor_handle.h"],
    visibility = [
        "//tensorflow:internal",
    ],
    deps = [
        ":abstract_tensor_handle",
        "//tensorflow/c:tensor_interface",
        "//tensorflow/core:framework",
        "//tensorflow/core:lib",

  }

  TF_Tensor* numerical_tensor;
  s = GetValue(numerical_grad.get(), &numerical_tensor);
  ASSERT_EQ(errors::OK, s.code()) << s.message();
  auto num_elem_numerical = TF_TensorElementCount(numerical_tensor);
  ASSERT_EQ(num_elem_numerical, num_grad);

  float* dnumerical = new float[num_elem_numerical]{0};
  memcpy(&dnumerical[0], TF_TensorData(numerical_tensor),
         TF_TensorByteSize(numerical_tensor));