Unit test failure: TypeError: EndVector() missing 1 required positional argument: 'vectorNumElems'

[buddhapuneeth]

Describe the bug

Running the tf2onnx tests on TF 2.10 failing the below Unit test

tests/test_tflite_utils.py::TFListUtilsTests::test_parse_tflite_graph - TypeError: EndVector() missing 1 required positional argument: 'vectorNumElems'

Urgency

not a immediate blocker. But look for fix by 10/15.

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 18.04*): Cent OS, Kernel: Linux 5.4.207-126.363.amzn2int.x86_6
• TensorFlow Version: 2.10 (installed from pip)
• Python version: 3.8 (conda env)
• ONNX version (if applicable, e.g. 1.11*):1.10.2
• ONNXRuntime version (if applicable, e.g. 1.11*):

To Reproduce

Followed steps from https://github.com/onnx/tensorflow-onnx#build-and-install-latest-from-source-for-development
Then

python setup.py test

Screenshots

Additional context

tests/test_tflite_utils.py:50:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
/home/user/anaconda3/envs/tf210_open_tf2onnx/lib/python3.8/site-packages/tensorflow/lite/python/lite.py:2808: in convert
    return super(TFLiteConverter, self).convert()
/home/user/anaconda3/envs/tf210_open_tf2onnx/lib/python3.8/site-packages/tensorflow/lite/python/lite.py:930: in wrapper
    return self._convert_and_export_metrics(convert_func, *args, **kwargs)
/home/user/anaconda3/envs/tf210_open_tf2onnx/lib/python3.8/site-packages/tensorflow/lite/python/lite.py:922: in _convert_and_export_metrics
    return flatbuffer_utils.convert_object_to_bytearray(model_object)
/home/user/anaconda3/envs/tf210_open_tf2onnx/lib/python3.8/site-packages/tensorflow/lite/tools/flatbuffer_utils.py:84: in convert_object_to_bytearray
    model_offset = model_object.Pack(builder)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <tensorflow.lite.python.schema_py_generated.ModelT object at 0x7f94221d7c40>, builder = <flatbuffers.builder.Builder object at 0x7f94e1ff94a0>

    def Pack(self, builder):
        if self.operatorCodes is not None:
            operatorCodeslist = []
            for i in range(len(self.operatorCodes)):
                operatorCodeslist.append(self.operatorCodes[i].Pack(builder))
            ModelStartOperatorCodesVector(builder, len(self.operatorCodes))
            for i in reversed(range(len(self.operatorCodes))):
                builder.PrependUOffsetTRelative(operatorCodeslist[i])
>           operatorCodes = builder.EndVector()
E           TypeError: EndVector() missing 1 required positional argument: 'vectorNumElems'

[yan12125]

That appears caused by old FlatBuffers.

• FlatBuffers < 2.0: EndVector() requires an argument vectorNumElems
• FlatBuffers 2.0.0 ~ 2.0.6: EndVector() accepts no argument [1]
• FlatBuffers >= 2.0.7: EndVector() accepts optional vectorNumElems again, while it's deprecated [2]

From the error message, I assume you're using FlatBuffers < 2.0. On the other hand, TensorFlow 2.10 appears to use FlatBuffers >= 2.0. Installing newer FlatBuffers should fix the issue.

Note that current tensorflow-onnx pins FlatBuffers < 2.0 [3][4]. You will need to ignore that version constraint.

[1] google/flatbuffers#6328
[2] google/flatbuffers#7246
[3] #1841
[4] #1977

[buddhapuneeth]

Yes, it worked after updating flatbuffers to 2.0.0 and relaxing the constraint on flatbuffers in setup.py. Thanks a lot!