for (const auto* edge : oper->node.out_edges()) {
    if (edge->IsControlEdge() && !edge->dst()->IsSink()) {
      ++count;
    }
  }
  return count;
}

int TF_OperationGetControlOutputs(TF_Operation* oper,
                                  TF_Operation** control_outputs,
                                  int max_control_outputs) {
  int count = 0;
  for (const auto* edge : oper->node.out_edges()) {

void TF_OutputListSetNumOutputs(TF_OutputList* o, int num_outputs,
                                TF_Status* s) {
  unwrap(o)->expected_num_outputs = num_outputs;
  unwrap(o)->outputs.clear();
  unwrap(o)->outputs.resize(num_outputs);
}
int TF_OutputListNumOutputs(TF_OutputList* o) {
  return unwrap(o)->outputs.size();
}
TF_AbstractTensor* TF_OutputListGet(TF_OutputList* o, int i) {
  return wrap(unwrap(o)->outputs[i]);
}

  ops->filesystem_ops = static_cast<TF_FilesystemOps*>(
      plugin_memory_allocate(TF_FILESYSTEM_OPS_SIZE));
  ops->filesystem_ops->init = tf_posix_filesystem::Init;
  ops->filesystem_ops->cleanup = tf_posix_filesystem::Cleanup;
  ops->filesystem_ops->new_random_access_file =
      tf_posix_filesystem::NewRandomAccessFile;

int TF_CheckpointReaderHasTensor(TF_CheckpointReader* reader,
                                 const char* name) {
  return reader->HasTensor(name);
}

const char* TF_CheckpointReaderGetVariable(TF_CheckpointReader* reader,
                                           int index) {
  return reader->variable_list[index].c_str();
}

int TF_CheckpointReaderSize(TF_CheckpointReader* reader) {

  }
}

TEST_F(CApiAttributesTest, EmptyList) {
  auto desc = init("list(int)");
  TF_SetAttrIntList(desc, "v", nullptr, 0);
  auto oper = TF_FinishOperation(desc, s_);
  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
  EXPECT_TF_META("v", 0, TF_ATTR_INT, -1);
}

TEST_F(CApiAttributesTest, Names) {
  auto desc = init("string");
  TF_SetAttrString(desc, "v", "bunny", 5);

  auto s = RunModel(grad_model, ctx, inputs, absl::MakeSpan(outputs),
                    /*use_function=*/use_function);
  ASSERT_EQ(errors::OK, s.code()) << s.message();

  for (int i = 0; i < num_inputs; ++i) {
    if (!outputs[i]) continue;

    AbstractTensorHandlePtr numerical_grad;
    {
      AbstractTensorHandle* numerical_grad_raw;
      s = CalcNumericalGrad(ctx, model, inputs,

 public:
  Status Compute(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    for (int i = 0; i < grad_outputs.size(); i++) {
      auto grad_input = grad_outputs[i];
      // TODO(srbs): Should we add a copy contructor to AbstractTensorHandle
      // that takes care of this similar to `Tensor`?

    const TF_Graph* fn_body, const char* fn_name, int ninputs,
    const TF_Output* inputs, std::vector<OutputTensor>* input_tensors,
    std::unordered_map<const Node*, std::vector<int>>* input_nodes)
    TF_EXCLUSIVE_LOCKS_REQUIRED(fn_body->mu) {
  input_tensors->reserve(ninputs);
  for (int i = 0; i < ninputs; ++i) {
    Node* node = inputs[i].oper ? &inputs[i].oper->node : nullptr;
    int idx = inputs[i].index;

  }
  TFE_OpAddAttrs(op_.get(), attributes);
  for (int input_index = 0; input_index < inputs.size(); ++input_index) {
    TFE_OpAddInput(op_.get(), inputs[input_index], status);
    if (TF_GetCode(status) != TF_OK) return;
  }
  std::vector<TFE_TensorHandle*> unwrapped_results(expected_max_outputs);
  int real_num_outputs = expected_max_outputs;

  for (int32_t i = start; i < end; i += step) {
    (*data)[i] = i;
  }
}

// Fills out_dims with the dimensions of the given tensor.
void GetDims(const TF_Tensor* t, int64_t* out_dims) {
  int num_dims = TF_NumDims(t);
  for (int i = 0; i < num_dims; i++) {
    out_dims[i] = TF_Dim(t, i);
  }
}

// Runs model as is if output is a scalar,
// else sums the output tensor before returning.