handle_shapes_and_types->at(0).shape;
        TensorShapeProto shape_proto;
        context->ShapeHandleToProto(handle, &shape_proto);
        if (!shape_proto.unknown_rank()) {
          NodeDef const_def;
          const_def.set_op("Const");
          Node* var_node;
          TF_RETURN_IF_ERROR(n->input_node(0, &var_node));
          const_def.set_name(

  auto u = ops::Placeholder(scope.WithOpName("U"), DT_RESOURCE);
  auto v = ops::Placeholder(scope.WithOpName("V"), DT_RESOURCE);
  auto w = ops::Placeholder(scope.WithOpName("W"), DT_RESOURCE);

  NodeDef def;
  TF_CHECK_OK(NodeDefBuilder("launch0", function, &flib_def)
                  .Input(a.node()->name(), 0, DT_INT32)
                  .Input(b.node()->name(), 0, DT_FLOAT)

  TF_FinishOperation(desc, s_);
  ASSERT_EQ(TF_INVALID_ARGUMENT, TF_GetCode(s_));
  string msg(TF_Message(s_));
  EXPECT_NE(msg.find("Input 'inputs' passed int32 expected float while "
                     "building NodeDef 'float_op'"),
            msg.npos);
  TF_AbortWhile(params_.get());
}

// This is a basic test to make sure the C++ gradient code can handle while

    node_names->push_back(flag.name());
    // TOCO doesn't required `data_type` to be filled for every input.
    // If it's not filled, make it an empty string so the importer will use
    // the data type in the NodeDef.
    auto toco_data_type = flag.data_type();
    if (toco_data_type == ::toco::IODataType::IO_DATA_TYPE_UNKNOWN) {
      node_dtypes->push_back("");
    } else {
      node_dtypes->push_back(

	kl.recorder.Eventf(kl.nodeRef, eventType, event, "Node %s status is now: %s", kl.nodeName, event)
}

// recordEvent records an event for this node, the Kubelet's nodeRef is passed to the recorder
func (kl *Kubelet) recordEvent(eventType, event, message string) {
	kl.recorder.Eventf(kl.nodeRef, eventType, event, message)
}

// record if node schedulable change.

                                                     const TF_DataType* values,
                                                     int num_values);

// Checks the tensorflow::NodeDef built via the methods above to see if it can
// run on device_type.
TF_CAPI_EXPORT extern void TF_AttrBuilderCheckCanRunOnDevice(
    TF_AttrBuilder* builder, const char* device_type, TF_Status* status);

      op_, name_mapper_.GetUniqueName(op_),
      /*ignore_unregistered_attrs=*/true);
  if (!nodedef_or.ok()) {
    return op_->emitRemark() << "failed to convert op to NodeDef: "
                             << nodedef_or.status().ToString();
  }

  if (failed(PrepareParams())) return failure();

  std::shared_ptr<const tensorflow::NodeProperties> props;

gtl::FlatMap<StringPiece, const AttrValueMap*, StringPieceHasher> NodeToAttrMap(
    const tensorflow::GraphDef& graphdef) {
  gtl::FlatMap<StringPiece, const AttrValueMap*, StringPieceHasher> result;
  for (const tensorflow::NodeDef& node : graphdef.node()) {
    result[node.name()] = &node.attr();
  }
  return result;
}

gtl::FlatMap<StringPiece, const tensorflow::FunctionDef*, StringPieceHasher>

    DeviceCompilationProfiler* profiler,
    const XlaCompiler::CompilationResult** out_compilation_result,
    ExecutableType** out_executable) {
  const NodeDef& def = ctx->op_kernel().def();
  NameAttrList name;
  name.set_name(def.op());
  *name.mutable_attr() = def.attr();
  // Remove the "_class" attribute from the attribute set used to create the

  if (materialize_derived_attrs_) {
    for (const auto& attr : derived_attrs) {
      // Add or update the derived attribute with the value. Skip the fixed
      // element type attributes, in case they are present in the NodeDef.
      if (!fixed_elt_type_attrs_.contains(attr.first())) {
        new_op->setAttr(attr.first(), attr.second);
      }
    }
  }
  // Create the tfr.cast ops on the results and replace the uses of the