// instantiation).
  virtual Status AddFunctionDefWithStackTraces(
      const FunctionDef& fdef, const StackTracesMap& stack_traces) = 0;

  // Find and return a added function by its name.
  virtual const FunctionDef* FindFunctionDef(const string& name) const = 0;

  // Find and return a function record added by its name.
  virtual core::RefCountPtr<FunctionRecord> FindRecord(

using tensorflow::string;

void TFE_OpReset(TFE_Op* op_to_reset, const char* op_or_function_name,
                 const char* raw_device_name, TF_Status* status) {
  if (op_to_reset) {
    tensorflow::ImmediateExecutionOperation* op =
        tensorflow::unwrap(op_to_reset);
    op->Clear();
    status->status = op->Reset(op_or_function_name, raw_device_name);
  } else {
    TF_SetStatus(status, TF_INVALID_ARGUMENT,

                          const char* name = "scalar");

TF_Operation* ScalarConst(double v, TF_Graph* graph, TF_Status* s,
                          const char* name = "scalar");

TF_Operation* ScalarConst(float v, TF_Graph* graph, TF_Status* s,
                          const char* name = "scalar");

TF_Operation* Add(TF_Operation* l, TF_Operation* r, TF_Graph* graph,
                  TF_Status* s, const char* name = "add");

  }
};

// An abstract operation describes an operation by its type, name, and
// attributes. It can be "executed" by the context with some input tensors.
// It is allowed to reusing the same abstract operation for multiple execution
// on a given context, with the same or different input tensors.
class TracingOperation : public AbstractOperation {
 protected:

}

tensorflow::ServerDef GetServerDef(const string& job_name, int num_tasks) {
  tensorflow::ServerDef server_def;
  server_def.set_protocol("grpc");
  server_def.set_job_name(job_name);
  server_def.set_task_index(0);
  tensorflow::ClusterDef* cluster_def = server_def.mutable_cluster();
  tensorflow::JobDef* job_def = cluster_def->add_job();
  job_def->set_name(job_name);
  for (int i = 0; i < num_tasks; i++) {

  std::vector<IOSpec> v;
  for (const string& name : names) {
    v.push_back(IOSpec(name, DT_INVALID));
  }
  return v;
}

// Specification for an expected edge.
// src is either:
// - input name (as it appears in FunctionDef)
// - name of output tensor (in nested "add:z:0" format)
// dst is either:
// - output name (as it appears in FunctionDef)
// - <name_of_node>:<index_of_this_input_into_node> (this looks the same as

  }
  if (remote) {
    TFE_OpSetDevice(matmul, task1_name, status);
    ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  } else if (!async) {
    // Set the local device to CPU to easily validate mirroring
    string cpu_device_name;
    ASSERT_TRUE(GetDeviceName(ctx, &cpu_device_name, "CPU"));
    TFE_OpSetDevice(matmul, cpu_device_name.c_str(), status);

    ],
)

cc_library(
    name = "abstract_op_attrs",
    hdrs = ["abstract_op_attrs.h"],
    visibility = [
        "//tensorflow:internal",
    ],
    deps = [
        "//tensorflow/core:lib",
        "//tensorflow/core:protos_all_cc",
        "@com_google_absl//absl/container:inlined_vector",
    ],
)

cc_library(
    name = "abstract_operation",
    hdrs = ["abstract_operation.h"],

  virtual bool HasCustomDeviceInput() const = 0;

  virtual const tensorflow::OpDef* OpDef() const = 0;

  virtual Status InputLength(const char* input_name, int* length) = 0;
  virtual Status OutputLength(const char* output_name, int* length) = 0;

  // Set stack trace to be used for potential async error reporting.
  virtual void SetStackTrace(ManagedStackTrace stack_trace) = 0;

// Set an AttrValue on the op. Doesn't handle the list types.
void SetOpAttrValueScalar(TFE_Context* ctx, TFE_Op* op,
                          const tensorflow::AttrValue& default_value,
                          const char* attr_name, TF_Status* status);
}  // namespace tensorflow