*   When you contribute a new feature to TensorFlow, the maintenance burden is
    (by default) transferred to the TensorFlow team. This means that the benefit
    of the contribution must be compared against the cost of maintaining the
    feature.
*   Full new features (e.g., a new op implementing a cutting-edge algorithm)
    typically will live in

TFE_MonitoringCounter0* TFE_MonitoringNewCounter0(const char* name,
                                                  TF_Status* status,
                                                  const char* description) {
  auto* result = new TFE_MonitoringCounter0({name, description});
  tsl::Set_TF_Status_from_Status(status, result->counter->GetStatus());
  if (!result->counter->GetStatus().ok()) {
    delete result;
    return nullptr;
  }
  return result;

  return new SqrtGradientFunction(op.outputs[0]);
}

GradientFunction* NegRegisterer(const ForwardOperation& op) {
  return new NegGradientFunction;
}

GradientFunction* SubRegisterer(const ForwardOperation& op) {
  return new SubGradientFunction;
}

GradientFunction* MulRegisterer(const ForwardOperation& op) {
  return new MulGradientFunction(op.inputs);
}

 public:
  explicit GraphContext(const char* name)
      : TracingContext(kGraph),
        graph_(new TF_Graph(), TF_DeleteGraph),
        name_(name) {}

  void Release() override { delete this; }

  TracingOperation* CreateOperation() override {
    return new GraphOperation(graph_.get());
  }

  Status AddParameter(DataType dtype, const PartialTensorShape& shape,

                  TF_Status* status) {
  tensorflow::ImmediateExecutionOperation* new_op =
      tensorflow::unwrap(ctx)->CreateOperation();
  status->status = new_op->Reset(op_or_function_name, nullptr);
  if (!status->status.ok()) {
    new_op->Release();
    new_op = nullptr;
  }
  return tensorflow::wrap(new_op);
}

void TFE_DeleteOp(TFE_Op* op) {
  if (op == nullptr) {
    return;
  }

  ops->filesystem_ops->cleanup = tf_posix_filesystem::Cleanup;
  ops->filesystem_ops->new_random_access_file =
      tf_posix_filesystem::NewRandomAccessFile;
  ops->filesystem_ops->new_writable_file = tf_posix_filesystem::NewWritableFile;
  ops->filesystem_ops->new_appendable_file =
      tf_posix_filesystem::NewAppendableFile;
  ops->filesystem_ops->new_read_only_memory_region_from_file =
      tf_posix_filesystem::NewReadOnlyMemoryRegionFromFile;

  }

  // We may need to allocate zero inputs for trainable dtypes we don't have JVPs
  // for. Make sure they get cleaned up.
  std::vector<Gradient*> new_zeros;
  auto delete_new_zeros = gtl::MakeCleanup([&new_zeros, this] {
    for (Gradient* tensor : new_zeros) {
      this->vspace_.DeleteGradient(tensor);
    }
  });
  std::vector<Gradient*> in_grads;
  in_grads.reserve(input_tensors.size());

        done.Notify();
      });
  done.WaitForNotification();
}

TF_ShapeAndTypeList* TF_NewShapeAndTypeList(int num_items) {
  TF_ShapeAndTypeList* result = new TF_ShapeAndTypeList;
  result->num_items = num_items;
  result->items = (num_items == 0) ? nullptr : new TF_ShapeAndType[num_items]();
  return result;
}

void TF_ShapeAndTypeListSetShape(TF_ShapeAndTypeList* shape_list, int index,

  tensorflow::StackTracesMap stack_traces;
  for (const Node* n : fn_body->graph.nodes()) {
    stack_traces[n->name()] = n->GetStackTrace();
  }

  TF_Function* tf_function = new TF_Function();
  tf_function->record = new tensorflow::FunctionRecord(
      std::move(fdef), std::move(stack_traces), false);

  return tf_function;
}

TF_Function* TF_GraphToFunction(const TF_Graph* fn_body, const char* fn_name,