TF_Operation** control_inputs,
                                 int max_control_inputs) {
  int count = 0;
  for (const auto* edge : oper->node.in_edges()) {
    if (edge->IsControlEdge() && !edge->src()->IsSource()) {
      if (count < max_control_inputs) {
        control_inputs[count] = ToOperation(edge->src());
      }
      ++count;
    }
  }

    *   Deprecate `tf.contrib.learn`. Please check contrib/learn/README.md for
        instructions on how to convert existing code.
    *   `tf.contrib.data`
    *   Remove deprecated `tf.contrib.data.Dataset`, `tf.contrib.data.Iterator`,
        `tf.contrib.data.FixedLengthRecordDataset`,
        `tf.contrib.data.TextLineDataset`, and `tf.contrib.data.TFRecordDataset`
        classes.

are skipped, and any checks implemented inside the shape inference code are not
executed.

The security impact of skipping those checks should be low, since the attack
scenario would require a malicious user to be able to control the model which as
stated above is already equivalent to code execution. In any case, the
recommendation is not to serve models using Eager mode since it also has
performance limitations.

## Multi-Tenant environments

          << "Failed to find expected control edge " << e.ToString()
          << " in fdef: " << fdef.DebugString();
    }

    // If caller specified all edges, check that we have seen all
    if (is_exact_edges) {
      ASSERT_EQ(e_edges.size() + c_edges.size(), a_edges.size())
          << "Expected edges: " << ToString(e_edges)
          << " Expected Control edges: " << ToString(c_edges)

  }

  Status Compute(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    // TODO(vnvo2409): Add control dependency
    /* Given upstream grad U and a Log1p op: Y = log(1 + X), the gradients are:
     *
     *    dX = U / (1 + X)
     *
     */

    AbstractTensorHandle* upstream_grad = grad_outputs[0];

    srcs_version = "PY3",
    visibility = ["//visibility:public"],
    deps = [
        ":tensorflow_py_no_contrib",
        "//tensorflow/lite/python:analyzer",
        "//tensorflow/lite/python:lite",
        "//tensorflow/lite/python/authoring",
    ],
)

py_library(
    name = "tensorflow_py_no_contrib",
    srcs = select({
        "api_version_2": [":tf_python_api_gen_v2"],

      output_names_vec.push_back(string(output_names[i]));
    }
  }

  // Process control output names.
  std::vector<string> control_output_names_vec;
  if (control_output_names) {
    control_output_names_vec.reserve(ncontrol_outputs);
    for (int i = 0; i < ncontrol_outputs; ++i) {
      control_output_names_vec.push_back(string(control_output_names[i]));
    }
  }

  // Compute body nodes.

// Get the number of control inputs to an operation.
TF_CAPI_EXPORT extern int TF_OperationNumControlInputs(TF_Operation* oper);

// Get list of all control inputs to an operation.  `control_inputs` must
// point to an array of length `max_control_inputs` (ideally set to
// TF_OperationNumControlInputs(oper)).  Returns the number of control
// inputs (should match TF_OperationNumControlInputs(oper)).

                         std::unique_ptr<tensorflow::Graph>* graph,
                         tensorflow::FunctionLibraryDefinition* flib_def,
                         std::vector<std::string>* control_ret_node_names,
                         bool* control_rets_updated) override {
    // Inject failure to function instantiation if finding a node that contains
    // the given node name (error_node_) and requested device (error_device_).

  if (TF_GetCode(status) != TF_OK) return nullptr;
  if (parallel_tensor->num_tensors() == 1) {
    // Copy-off for single-device tensors is allowed to make debugging dynamic
    // control flow easier.
    return TFE_TensorHandleCopySharingTensor(parallel_tensor->tensor(0),
                                             status);
  } else {
    TF_SetStatus(
        status, TF_UNIMPLEMENTED,