const char* node_name)
      : node_builder(node_name, op_type, g->graph.op_registry()), graph(g) {}

  tensorflow::NodeBuilder node_builder;
  TF_Graph* graph;
  std::set<tensorflow::string> colocation_constraints;
};

struct TF_Operation {
  tensorflow::Node node;

 private:
  ~TF_Operation() = default;
};

struct TF_Session {

.DS_Store
.ipynb_checkpoints
node_modules
/.bazelrc.user
/.tf_configure.bazelrc
/xla_configure.bazelrc
/bazel-*
/bazel_pip
/tools/python_bin_path.sh
/tensorflow/tools/git/gen
/pip_test
/_python_build
*.pyc
__pycache__
*.swp
.vscode/
cmake_build/
tensorflow/contrib/cmake/_build/
.idea/**
/build/
[Bb]uild/
/build_output/
/tensorflow/core/util/version_info.cc
/tensorflow/python/framework/fast_tensor_util.cpp

namespace tensorflow {

// Abstract interface to a TensorHandle.
//
// A TensorHandle is management class around a Tensor which may track additional
// metadata and synchronization.
//
// This allows us to hide concrete implementations of TensorHandle from header
// files. The interface lists the common functionality that must be provided by
// any concrete implementation. However, in cases where the true concrete class

// just nice to know that it compiles.
void* create(TF_OpKernelConstruction* ctx) {
  TF_DataType type;
  TF_Status* s = TF_NewStatus();
  TF_OpKernelConstruction_GetAttrType(ctx, "foobar", &type, s);
  TF_DeleteStatus(s);
  return NULL;
}

// A compute function. This will never actually get called in this test, it's
// just nice to know that it compiles.

  options->guard_size = 0;
  options->numa_node = -1;
}

TF_Thread* TF_StartThread(const TF_ThreadOptions* options,
                          const char* thread_name, void (*work_func)(void*),
                          void* param) {
  ::tensorflow::ThreadOptions cc_options;
  cc_options.stack_size = options->stack_size;
  cc_options.guard_size = options->guard_size;
  cc_options.numa_node = options->numa_node;

void TF_OutputListPushBack(TF_OutputList* o, TF_AbstractTensor* tensor,
                           TF_Status*);

// TF_ExecuteOperation will, if in eager mode, execute, if in graph mode, maybe
// capture some inputs and then add a node in the graph. The output tensors are
// returned through the provided TF_OutputList.
// Any active tape will observe the effects of this execution.
void TF_ExecuteOperation(TF_AbstractOp* op, int num_inputs,

bool IsPlaceholder(const tensorflow::NodeDef& node_def);

bool IsScalarConst(const tensorflow::NodeDef& node_def, int v);

bool IsAddN(const tensorflow::NodeDef& node_def, int n);

bool IsNeg(const tensorflow::NodeDef& node_def, const string& input);

bool GetGraphDef(TF_Graph* graph, tensorflow::GraphDef* graph_def);

bool GetNodeDef(TF_Operation* oper, tensorflow::NodeDef* node_def);

    "creators": [
        {
            "name": "TensorFlow Developers"
        }
    ],
    "access_right": "open",
    "notes": "Specific TensorFlow versions can be found in the \"Versions\" list on the right side of this page.<br>See the full list of authors <a href=\"https://github.com/tensorflow/tensorflow/graphs/contributors\">on GitHub</a>."

### Hardware attacks

Physical GPUs or TPUs can also be the target of attacks. [Published
research](https://scholar.google.com/scholar?q=gpu+side+channel) shows that it
might be possible to use side channel attacks on the GPU to leak data from other
running models or processes in the same system. GPUs can also have
implementation bugs that might allow attackers to leave malicious code running

                   "      output_arg {"
                   "        name: 'm'"
                   "        type: DT_FLOAT"
                   "      }"
                   "    }"
                   "    node_def {"
                   "      name: 'matmul'"
                   "      op: 'MatMul'"
                   "      input: 'a'"
                   "      input: 'b'"
                   "      device: '",