tf.Tensor used as a type is coming up as a tensor instance

[khatchad]

Consider the following code:

from tensorflow import Tensor


def f(a):
    assert isinstance(a, Tensor)

I am seeing this in the tensor analysis:

[Node: synthetic < PythonLoader, Lwala/builtin/isinstance, do()LRoot; > Context: CallStringContext: [ script src.tf2_test_module3a.py.f.do()LRoot;@2 ], v2][{[D:Symbolic,n, D:Compound,[D:Constant,28, D:Constant,28]] of pixel}]

Here is the corresponding summary:

• ML/com.ibm.wala.cast.python.ml/data/tensorflow.xml

  Lines 233 to 235 in d9a3d6b

  	<new def="Tensor" class="Ltensorflow/functions/Tensor" />
  	<putfield class="LRoot" field="Tensor" fieldType="LRoot" ref="x" value="Tensor" />
  	<putfield class="LRoot" field="Tensor" fieldType="LRoot" ref="ops" value="Tensor" />

• ML/com.ibm.wala.cast.python.ml/data/tensorflow.xml

  Lines 502 to 511 in d9a3d6b

  	<class name="Tensor" allocatable="true">
  	<method name="read_data" descriptor="()LRoot;">
  	<new def="x" class="Ltensorflow/python/framework/ops/Tensor" />
  	<return value="x" />
  	</method>
  	<method name="do" descriptor="()LRoot;" numArgs="3" paramNames="op value_index dtype">
  	<call class="LRoot" name="read_data" descriptor="()LRoot;" type="virtual" arg0="arg0" def="x" />
  	<return value="x" />
  	</method>
  	</class>

That looks right. I'm unsure why it's being picked up as function, though.