Default EQUALITY_OPERATORS in Autograph

[bhack]

fix #55278

/cc @mdanatg Let me know where do you want to move the test so that I write it like an integration test as where is now it will go to fail for missing variable conversion (ag__.ld).

[bhack]

I've extended an high level integration test.
Let me know if we want to remove or we could modify the lowlevel one to pass.

[bhack]

Copybara again

[mdanatg]

BTW, I finally exported the integration tests - they're now under autograph/tests.

[bhack]

What is see here is that here in the test this a python bool.
Where in a standard gist like this is a bool Tensor

def test(a, b):
  print(a == b)
  return a == b

a = tf.constant([1])
b = 2
test(a, b)

tf.Tensor([ True], shape=(1,), dtype=bool)

[mdanatg]

That's in a standalone shell, right? The unit test might be running in different conditions, likely in graph mode. I fear the best way to debug the error is to run the test, with both --test_env=TF2_BEHAVIOR=1 and =0...

[bhack]

That's in a standalone shell, right

it was on colab of course in graph mode was Tensor("Equal:0", shape=(1,), dtype=bool) that is the similar when in the test I use another operator like e.g. >=

--test_env=TF2_BEHAVIOR=1 and =0

It always print a Python bool in the test

[bhack]

I have committed an alternative test and it is passing. You can compare the results in the log.

If there is something not working in ops.enable_tensor_equality() I suppose that it need to be covered with an additional use case in its test file:
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/eager/core_test.py#L92

[bhack]

@mdanatg I suppose this was merged with b462db3.
But I have totally lost my attribution on git:

git show -s --format="%ae" b462db3

gardener@tensorflow.org

[bhack]

P.s. Ok I found it on git show -s --format="%ce %ae" e2cc982. I suppose we could close this.

[mdanatg]

Yes, it's strange that GH didn't automatically marked it as merged. Probably due to the internal revisions.