How are scalars stored in tensor protobuf files

[kraiskil]

I am trying to understand the ConstantOfShape operator and scalar 'tensors'. The example/backend test int32_shape_zero produces an output that should be a scalar of value 0 (if I understand it correctly).

How do I find this value of the scalar output tensor from output_0.pb in the test data: https://github.com/onnx/onnx/tree/main/onnx/backend/test/data/node/test_constantofshape_int_shape_zero/test_data_set_0?

In that directory, I read and print out the test result tensor using python:

import onnx

tensor=onnx.TensorProto()
with open('test_data_set_0/output_0.pb', 'rb') as f:
  tensor.ParseFromString(f.read())

print(tensor)

The output is:

dims: 0
data_type: 6
name: "y"
raw_data: ""

I.e. the "tensor" is a scalar (dims: 0), and uses raw data storage. However, there is no data stored in raw_data (or in int32_data for that matter). Should there be?