Wrong result with `If` node

[nomaddo]

Describe the bug
Wrong result is produce with the following ONNX.
https://gist.github.com/nomaddo/965614ac0684e1042850f6d84cc38970

See how to create
See how to run

This ONNX graph has one if node with very simple sub-graphs (having one Constant node).
The sub-graphs never return 0., but got wrong result as follows:

$ python3 run.py model.onnx 
[array([[[[0.]]]], dtype=float32)]

Urgency
None

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Debian GNU/Linux 10 (buster)
• ONNX Runtime installed from (source or binary): binary (via pip)
• ONNX Runtime version: 1.2.0
• Python version: 3.7.3
• Visual Studio version (if applicable):
• GCC/Compiler version (if compiling from source):
• CUDA/cuDNN version:
• GPU model and memory:

To Reproduce

• Describe steps/code to reproduce the behavior.

$ sudo docker run -it --rm debian:buster /bin/bash
$ apt update; apt install -y ssh git python3 python3-pip
$ git clone https://gist.github.com/965614ac0684e1042850f6d84cc38970.git bug_report
$ cd bug_report
$ pip3 install onnx onnxruntime
$ pip3 list
Package           Version
----------------- -------
asn1crypto        0.24.0 
cryptography      2.6.1  
entrypoints       0.3    
keyring           17.1.1 
keyrings.alt      3.1.1  
numpy             1.18.4 
onnx              1.7.0  
onnxruntime       1.2.0  
pip               18.1   
protobuf          3.11.3 
pycrypto          2.6.1  
PyGObject         3.30.4 
pyxdg             0.25   
SecretStorage     2.3.1  
setuptools        40.8.0 
six               1.12.0 
typing-extensions 3.7.4.2
wheel             0.32.3 
$ python3 gen_onnx.py  # create model.onnx
$ python3 run.py model.onnx 
[array([[[[0.]]]], dtype=float32)]

• Attach the ONNX model to the issue (where applicable) to expedite investigation.
  https://gist.github.com/nomaddo/965614ac0684e1042850f6d84cc38970#file-model-onnx

Expected behavior
Got the following result:

$ python3 run.py model.onnx 
[array([[[[1001.]]]], dtype=float32)]

Screenshots
None

Additional context
None

[hariharans29]

I couldn't repro this in current master (I used the onnx_test_runner app to run the model) -

I tried to trigger both conditional branches - if, and else and it outputs the expected results in both cases. I ll build the Python wheel and try.

[hariharans29]

I could repro it in the 1.2.0 wheel but with the wheel built off latest master, it seems to go away. So, with the upcoming release, you should be able to use it. Thanks for reporting.

[hariharans29]

#3912 had the fix for this