Use C API to implement Operation.control_inputs

Operation.control_inputs must return Python Operation instances,
but C API returns TF_Operation*s. After retrieving TF_Operation*s,
we get the Python Operation object from the graph using TF_Operation's
name.

There is currently just one test for this functionality, namely
ControlDependenciesTest.testBasic. We can't enable other tests
because C API does not accept unregistered operations.
In following changes we will deal with this problem thereby enabling
more tests.

PiperOrigin-RevId: 161106091

tensorflow/python/client/tf_session.i

@@ -68,6 +68,27 @@ tensorflow::ImportNumpy();
   $result = PyUnicode_FromString($1);
 }
 
+// We use TF_OperationGetControlInputs_wrapper instead of
+// TF_OperationGetControlInputs
+%ignore TF_OperationGetControlInputs;
+%unignore TF_OperationGetControlInputs_wrapper;
+// See comment for "%noexception TF_SessionRun_wrapper;"
+%noexception TF_OperationGetControlInputs_wrapper;
+
+// Build a Python list of TF_Operation* and return it.
+%typemap(out) std::vector<TF_Operation*> tensorflow::TF_OperationGetControlInputs_wrapper {
+  $result = PyList_New($1.size());
+  if (!$result) {
+    SWIG_exception_fail(SWIG_MemoryError, "$symname: couldn't create list");
+  }
+
+  for (size_t i = 0; i < $1.size(); ++i) {
+    PyList_SET_ITEM($result, i, SWIG_NewPointerObj(
+                            $1[i], SWIGTYPE_p_TF_Operation, 0));
+  }
+}
+
+
 ////////////////////////////////////////////////////////////////////////////////
 // BEGIN TYPEMAPS FOR tensorflow::TF_Run_wrapper()
 ////////////////////////////////////////////////////////////////////////////////

tensorflow/python/client/tf_session_helper.cc

@@ -766,4 +766,12 @@ void TF_SessionPRun_wrapper(TF_Session* session, const char* handle,
   ClearDecrefCache();
 }
 
+std::vector<TF_Operation*> TF_OperationGetControlInputs_wrapper(
+    TF_Operation* oper) {
+  std::vector<TF_Operation*> control_inputs(TF_OperationNumControlInputs(oper));
+  TF_OperationGetControlInputs(oper, control_inputs.data(),
+                               control_inputs.size());
+  return control_inputs;
+}
+
 }  // namespace tensorflow

tensorflow/python/client/tf_session_helper.h

@@ -158,6 +158,11 @@ void TF_SessionPRun_wrapper(TF_Session* session, const char* handle,
                             TF_Status* out_status,
                             std::vector<PyObject*>* py_outputs);
 
+// Retrieves control inputs of this operation.
+// control_inputs should be empty.
+std::vector<TF_Operation*> TF_OperationGetControlInputs_wrapper(
+    TF_Operation* oper);
+
 }  // namespace tensorflow
 
 #endif  // TENSORFLOW_PYTHON_CLIENT_TF_SESSION_HELPER_H_

tensorflow/python/framework/ops.py

@@ -1583,7 +1583,14 @@ class Operation(object):
       A list of `Operation` objects.
 
     """
-    return self._control_inputs
+    if _USE_C_API:
+      control_c_ops = c_api.TF_OperationGetControlInputs_wrapper(self._c_op)
+      # pylint: disable=protected-access
+      return [self.graph._get_operation_by_name_unsafe(
+          c_api.TF_OperationName(c_op)) for c_op in control_c_ops]
+      # pylint: enable=protected-access
+    else:
+      return self._control_inputs
 
   @property
   def type(self):
@@ -2781,6 +2788,29 @@ class Graph(object):
                       % type(name).__name__)
     return self.as_graph_element(name, allow_tensor=False, allow_operation=True)
 
+  def _get_operation_by_name_unsafe(self, name):
+    """Returns the `Operation` with the given `name`.
+
+    This is a internal unsafe version of get_operation_by_name. It skips many
+    checks and does not have user friedly error messages but runs considerably
+    faster. This method may be called concurrently from multiple threads.
+
+    Args:
+      name: The name of the `Operation` to return.
+
+    Returns:
+      The `Operation` with the given `name`.
+
+    Raises:
+      KeyError: If `name` does not correspond to an operation in this graph.
+    """
+
+    if self._finalized:
+      return self._nodes_by_name[name]
+
+    with self._lock:
+      return self._nodes_by_name[name]
+
   def get_tensor_by_name(self, name):
     """Returns the `Tensor` with the given `name`.