LOG(INFO) << "There are " << num_devices << " devices.";
  for (int i = 0; i < num_devices; ++i) {
    const char* device_name = TF_DeviceListName(device_list, i, s);
    CHECK_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    const char* device_type = TF_DeviceListType(device_list, i, s);
    CHECK_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    LOG(INFO) << "Device " << i << " has name " << device_name << ", type "

`tf` instead of `tensorflow`, following the common practice of importing
TensorFlow via the command `import tensorflow as tf`.

The primary function of this module is to import all of the public TensorFlow
interfaces into a single place. The interfaces themselves are located in
sub-modules, as described below.

Note that the file `__init__.py` in the TensorFlow source code tree is actually

    initiates CI/CD tests.
-   We can't move forward if these tests fail.
-   In such situations, we may request you to make further changes to your PR
    for the tests to pass.
-   Once the tests pass, we now bring all the code into the internal code base,
    using a job called "copybara".

**5. Copy to Google Internal codebase and run internal CI**

-   Once the PR is in the Google codebase, we make sure it integrates well with

    const Graph& graph = session->graph->graph;

    const string& mutation_warning = session->graph->sessions[session];
    if (!mutation_warning.empty()) {
      // TODO(b/74949947): turn this back into an error status
      LOG(WARNING) << mutation_warning;
      session->graph->sessions[session].clear();
    }

    const auto num_nodes = graph.num_node_ids();
    if (session->last_num_graph_nodes < num_nodes) {

namespace tensorflow {

// Register the POSIX filesystems statically.
// Return value will be unused
bool StaticallyRegisterLocalFilesystems() {
  TF_FilesystemPluginInfo info;
  TF_InitPlugin(&info);
  Status status = filesystem_registration::RegisterFilesystemPluginImpl(&info);
  if (!status.ok()) {
    VLOG(0) << "Static POSIX filesystem could not be registered: " << status;
    return false;
  }
  return true;
}

namespace {
class CustomDeviceAPI : public tensorflow::CustomDevice {
 public:
  CustomDeviceAPI(TFE_Context* context, TFE_CustomDevice device, void* info,
                  string name)
      : context_(context), device_(device), info_(info), name_(name) {}

  ~CustomDeviceAPI() override { device_.delete_device(info_); }

  const string& name() override { return name_; }

  tensorflow::Status CopyTensorToDevice(

*   TensorFlow Debugger (tfdbg) CLI: ncurses-based CLI for tfdbg v1 was removed.

    ],
    deps = [
        ":immediate_execution_operation",
        "//tensorflow/c:conversion_macros",
    ],
)

cc_library(
    name = "tfe_tensor_debug_info_internal",
    hdrs = ["tfe_tensor_debug_info_internal.h"],
    visibility = ["//visibility:private"],
    deps = [
        "//tensorflow/core:lib",
    ],
)

cc_library(
    name = "tfe_tensorhandle_internal",

        "Unable to find function with name: ", function_name);
    return;
  }

  tensorflow::GraphDebugInfo debug_info =
      tensorflow::StackTracesMapToGraphDebugInfo(
          function_record->stack_traces());

  string str = debug_info.SerializeAsString();
  void* data = tensorflow::port::Malloc(str.length());
  str.copy(static_cast<char*>(data), str.length(), 0);
  buf->data = data;

auto& kUnknownDim = shape_inference::InferenceContext::kUnknownDim;
auto& kUnknownRank = shape_inference::InferenceContext::kUnknownRank;

// GraphTensor wraps a `TF_Output`, i.e. a pointer to TF_Operation and the index
// into the list of outputs for the operation.
class GraphTensor : public TracingTensorHandle {
 public:
  explicit GraphTensor(TF_Output output, TF_Graph* graph)
      : TracingTensorHandle(kGraph), output_(output), graph_(graph) {}