CSession csession(session);

  // Retrieve the regression signature from meta graph def.
  const auto signature_def_map = metagraph_def.signature_def();
  const auto signature_def = signature_def_map.at("regress_x_to_y");

  const string input_name =
      signature_def.inputs().at(tensorflow::kRegressInputs).name();
  const string output_name =
      signature_def.outputs().at(tensorflow::kRegressOutputs).name();

  ASSERT_EQ(func_->record->fdef().signature().attr().size(), 2);
  EXPECT_EQ(func_->record->fdef().signature().attr(0).name(), "v1");
  EXPECT_EQ(func_->record->fdef().signature().attr(0).type(), "int");
  EXPECT_EQ(func_->record->fdef().signature().attr(1).name(), "v2");
  EXPECT_EQ(func_->record->fdef().signature().attr(1).type(), "int");
}

  GraphErrorInjectionPass::enabled_ = false;
}

string VariableAddFunctionWithGraphError() {
  string signature = VariableAddFunctionSignature();
  // Replace the node 'read0' with 'read0_maybe_with_graph_error', so that the
  // error injecting pass can identify and introduce graph pass errors.
  signature = std::regex_replace(signature, std::regex("read0"),
                                 "read0_maybe_with_graph_error");

    Name("TestCommUnavailable").Device(tensorflow::DEVICE_DEFAULT),
    TestUnavailableErrorOp);

string FunctionWithErrorOp(const tensorflow::StringPiece op_name) {
  const std::string& func_str =
      "    signature {"
      "      name: 'FunctionWith__OP_NAME__'"
      "      output_arg {"
      "        name: 'out'"
      "        type: DT_STRING"
      "      }"
      "    }"
      "    node_def {"

  if (TF_GetCode(status) != TF_OK) return;

  tensorflow::GradientDef gdef;
  gdef.set_function_name(func->record->fdef().signature().name());
  gdef.set_gradient_func(grad->record->fdef().signature().name());
  status->status = g->graph.AddGradientDef(std::move(gdef));
}

int TF_GraphNumFunctions(TF_Graph* g) {
  tensorflow::mutex_lock l(g->mu);

      return tensorflow::errors::InvalidArgument("Unsupported Type Codes: ",
                                                 dtype.code);
  }
}

// Wraps the deleter function of DLManagedTensor to match the function signature
// TFE_NewTensorHandleFromDeviceMemory.
void DeallocatorWrapperFunc(void* data, size_t len, void* dlmt_vptr) {
  TFE_CallDLManagedTensorDeleter(dlmt_vptr);
}

       *   `tflite::Interpreter::GetSignatureRunner`
       *   `tflite::Interpreter::signature_keys`
       *   `tflite::Interpreter::signature_inputs`
       *   `tflite::Interpreter::signature_outputs`
       *   `tflite::Interpreter::input_tensor_by_signature`
       *   `tflite::Interpreter::output_tensor_by_signature`

        ":c_test_util",
        ":test_op_kernel",
        ":tf_buffer",
        ":tf_buffer_internal",
        "//tensorflow/cc:cc_ops",
        "//tensorflow/cc:grad_ops",
        "//tensorflow/cc/saved_model:signature_constants",
        "//tensorflow/cc/saved_model:tag_constants",
        "//tensorflow/compiler/jit",
        "//tensorflow/core:array_ops_op_lib",
        "//tensorflow/core:bitwise_ops_op_lib",

      if (TF_GetCode(status) != TF_OK) return -1;
      if (!gcs_file->file_block_cache->ValidateAndUpdateFileSignature(
              path, stat.generation_number)) {
        TF_VLog(
            1,
            "File signature has been changed. Refreshing the cache. Path: %s",
            path.c_str());
      }
      read = gcs_file->file_block_cache->Read(path, offset, n, buffer, status);
    } else {

/// TensorFlow but can be used by each filesystem plugin to represent internal
/// data.
///
/// We prefer to have these structures instead of passing `void*` into
/// method signatures to have some type of type safety: for example, operations
/// that are only valid on random access files have a `TF_RandomAccessFile`
/// argument.
///