RegisterLoggingDevice(context.get(), custom0,
                        /*strict_scope_placement=*/false, &arrived, &executed,
                        status.get());
  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  RegisterLoggingDevice(context.get(), custom1,
                        /*strict_scope_placement=*/true, &arrived, &executed,
                        status.get());

#include "tensorflow/c/eager/c_api.h"
#include "tensorflow/c/eager/c_api_experimental.h"
#include "tensorflow/c/tf_status.h"

void RegisterLoggingDevice(TFE_Context* context, const char* name,
                           bool strict_scope_placement, bool* arrived_flag,
                           bool* executed_flag, TF_Status* status);
void AllocateLoggingDevice(const char* name, bool* arrived_flag,

          const tensorflow::Scope& scope,
          const std::vector<tensorflow::Output>& inputs,
          std::vector<tensorflow::Output>* outputs) {
        DCHECK_EQ(scope.graph(), &parent->graph);
        TF_RETURN_IF_ERROR(
            CopyGraph(&params->body_graph->graph, &parent->graph,
                      &parent->refiner, params->body_inputs, inputs,
                      scope.impl()->name(), scope.impl()->control_deps(),

  device->arrived_flag = arrived_flag;
  device->executed_flag = executed_flag;
  device->device_name = name;
  device->underlying_device = "/job:localhost/replica:0/task:0/device:CPU:0";
  device->strict_scope_placement = strict_scope_placement;
  TFE_RegisterCustomDevice(context, custom_device, name, device, status);
}

TFE_TensorHandle* UnpackTensorHandle(TFE_TensorHandle* logged_tensor_handle,

// need for `Env::Default()`. At that time, we might decide to remove the need
// for `Env::Default()` altogether, but that's a different project, not in
// scope for now. I'm just mentioning this here as that transition will mean
// removal of the registration part from `Env` and adding it here instead: we
// will need tables to hold for each scheme the function tables that implement

## Scope

// efficiency of forward-mode autodiff. Instead, language bindings should call
// DeleteGradient as soon as a Tensor which was `Watch`ed or was an output
// Tensor passed to Accumulate goes out of scope.
//
// Not thread-safe.
template <typename Gradient, typename BackwardFunction, typename TapeTensor>
class ForwardAccumulator {
 public:
  // Does not take ownership of `vspace`, which must outlive the

    assignment to variables. In particular one can write
    `var[1:3].assign([1,11,111])`.
*   Deprecated `tf.op_scope` and `tf.variable_op_scope` in favor of a unified
    `tf.name_scope` and `tf.variable_scope`. The new argument order of
    `tf.variable_scope` is incompatible with previous versions.
*   Introducing `core/util/tensor_bundle` module: a module to efficiently

TEST_F(CApiGradientsTest, GradientsPrefix_TwoGradientsInSameScope) {
  BuildGraphAndAddGradientsWithPrefixes("scope/gradients", "scope/gradients_1");
  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
}

TEST_F(CApiGradientsTest, GradientsPrefix_TwoGradientsInDifferentScopes) {
  BuildGraphAndAddGradientsWithPrefixes("scope/gradients", "scope_1/gradients");
  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
}

            "//tensorflow/c/experimental/filesystem:modular_filesystem",
            "//tensorflow/cc:grad_ops",
            "//tensorflow/cc:gradients",
            "//tensorflow/cc:ops",
            "//tensorflow/cc:scope_internal",
            "//tensorflow/cc:while_loop",
            "//tensorflow/cc/saved_model:loader_lite",
            "//tensorflow/compiler/mlir/tfr:graph_decompose_pass",