targets_.emplace_back(t);
  }
}

void CSession::Run(TF_Status* s) {
  if (inputs_.size() != input_values_.size()) {
    ADD_FAILURE() << "Call SetInputs() before Run()";
    return;
  }
  ResetOutputValues();
  output_values_.resize(outputs_.size(), nullptr);

  const TF_Output* inputs_ptr = inputs_.empty() ? nullptr : &inputs_[0];
  TF_Tensor* const* input_values_ptr =

    for (auto const& x : out) {
      inputs_.push_back(x);
    }
  }

  typename std::vector<Input>::iterator begin() { return inputs_.begin(); }
  typename std::vector<Input>::iterator end() { return inputs_.end(); }
  typename std::vector<Input>::const_iterator begin() const {
    return inputs_.begin();
  }
  typename std::vector<Input>::const_iterator end() const {
    return inputs_.end();
  }

 private:

  std::unordered_set<int> stop_backprop_nodes =
      GetStopBackpropNodes(reachable_nodes, output_nodes);

  // Populate `input_nodes_` from Outputs in `inputs_`.
  input_nodes_.reserve(inputs_.size());
  for (size_t i = 0; i < inputs_.size(); ++i) {
    input_nodes_.insert({inputs_[i], i});
  }

  // TODO(andydavis) Consider a more efficient data structure for `pending_` to

  }

  void Run(const std::vector<TF_Output>& run_outputs,
           std::initializer_list<int> input_values) {
    DCHECK_EQ(inputs_.size(), input_values.size());
    std::vector<std::pair<TF_Operation*, TF_Tensor*>> inputs(inputs_.size());
    int i = 0;
    for (int v : input_values) {
      inputs[i] = {inputs_[i].oper, Int32Tensor(v)};
      ++i;
    }
    // TODO(skyewm): use std::make_unique or absl::make_unique when possible.

      }
      graph_outputs.push_back(output->output_);
    }

    auto s = TF_NewStatus();
    auto func = TF_GraphToFunction(graph_.get(), name_.data(), 0, -1, nullptr,
                                   inputs_.size(), inputs_.data(),
                                   graph_outputs.size(), graph_outputs.data(),
                                   nullptr, nullptr, name_.data(), s);
    *f = new GraphFunction(std::move(func->record->fdef()));

  TF_EXPECT_OK(InitOp());

  // Add inputs.
  Tensor* a = CreateDeviceTensor<int32>(TensorShape({1, 3}), {1, 2, 3});
  Tensor* b = CreateDeviceTensor<int32>(TensorShape({1, 3}), {4, 5, 6});
  inputs_.push_back({nullptr, a});
  inputs_.push_back({nullptr, b});

  CreateContext();

  std::vector<XlaCompiler::Argument> args(2);
  args[0].kind = XlaCompiler::Argument::kParameter;
  args[0].type = DT_INT32;

  TFE_TensorHandle* input_eager = TestScalarTensorHandle(eager_ctx, 2.0f);
  TF_AbstractTensor* input_t =
      TF_CreateAbstractTensorFromEagerTensor(input_eager, status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

  TF_ExecuteOperation(fn_op, 1, &input_t, add_outputs, status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

	}
}

// TestSignV4TrimAll - tests the logic of TrimAll() function
func TestSignV4TrimAll(t *testing.T) {
	testCases := []struct {
		// Input.
		inputStr string
		// Expected result.
		result string
	}{
		{"本語", "本語"},
		{" abc ", "abc"},
		{" a b ", "a b"},
		{"a b ", "a b"},
		{"a  b", "a b"},
		{"a   b", "a b"},
		{"   a   b  c   ", "a b c"},

    handle.set_maybe_type_name(type_index.name());

    Tensor* input = new Tensor(host_allocator_, DT_RESOURCE, TensorShape({}));
    input->scalar<ResourceHandle>()() = handle;
    tensors_.push_back(input);
    inputs_.push_back({nullptr, input});
  }

 protected:
  core::RefCountPtr<DeviceContext> device_context_;
  Allocator* host_allocator_;
  Allocator* device_allocator_;

  XlaCompiler::Options compiler_options_;

            dtype: DT_STRING
            tensor_shape {
            }
            string_val: "model"
          }
        }
      }
    }
    node {
      name: "save/StringJoin/inputs_1"
      op: "Const"
      attr {
        key: "_output_shapes"
        value {
          list {
            shape {
            }
          }
        }
      }
      attr {