[Bug]: Elementwise function cannot be applied or does not return the correct value to Python float when ivy.set_array_mode(False)

[34j]

Bug Explanation

Calculating log(1):

python -c "import ivy; ivy.set_backend('torch'); ivy.set_array_mode(False); print(ivy.log(1))"
venv/lib/python3.11/site-packages/ivy/func_wrapper.py:440: UserWarning: In the case of Compositional function, operators might cause inconsistent behavior when array_mode is set to False
  warnings.warn(
tensor([])

python -c "import ivy; ivy.set_backend('numpy'); ivy.set_array_mode(False); print(ivy.log(1))"
(long output)
ivy.utils.exceptions.IvyBackendException: numpy: is_array: numpy: is_ivy_array: numpy: is_native_array: numpy: nested_argwhere: numpy: nested_argwhere: maximum recursion depth exceeded

These works:

python -c "import ivy; import numpy as np; ivy.set_backend('numpy'); ivy.set_array_mode(False); print(ivy.log(np.array(1)))"
0.0

python -c "import ivy; import torch; ivy.set_backend('torch'); ivy.set_array_mode(False); print(ivy.log(torch.tensor(1)))"
tensor(0.)

Perhaps this kind of usage is not expected, but I would just like to report it anyway.

Steps to Reproduce Bug

python -c "import ivy; ivy.set_backend('numpy'); ivy.set_array_mode(False); print(ivy.log(1))"
python -c "import ivy; ivy.set_backend('torch'); ivy.set_array_mode(False); print(ivy.log(1))"

Environment

Ubuntu 22.04

Ivy Version

1.0.0.0

Backend

• NumPy
• TensorFlow
• PyTorch
• JAX

Device

CPU

[YushaArif99]

Hi @34j. Thanks alot for opening the issue. Upon some close inspection, it turns out the actual bug is originating from printing the result of ivy.log(1) which invokes the __repr__ method of the ivy.Array class. I have opened a PR with the relevant fix, which should hopefully also resolve this issue you were experiencing. Let me know if this works or if the issue is still persisting on your end 😄

[34j]

@YushaArif99 Thank you very much for your swift response and fix.

I ran pip install -U --force-reinstall git+https://github.com/ivy-llc/ivy.git and ran the commands again, and the Numpy issue seems to be resolved.

$ python -c "import ivy; ivy.set_backend('numpy'); ivy.set_array_mode(False); print(ivy.log(1))"
venv/lib/python3.11/site-packages/ivy/func_wrapper.py:440: UserWarning: In the case of Compositional function, operators might cause inconsistent behavior when array_mode is set to False
  warnings.warn(
ivy.array(0.)

However, the PyTorch one still persists.
I may be doing something wrong, so could you please run the latter command on your end to check this?

As you said, the Numpy one seems to be related to print, but I think the PyTorch one is a different issue as the shape is already wrong.

$ python -c "import ivy; ivy.set_backend('torch'); ivy.set_array_mode(True); print(ivy.log(1).shape)"
ivy.Shape()
$ python -c "import ivy; ivy.set_backend('torch'); ivy.set_array_mode(False); print(ivy.log(1).shape)"
venv/lib/python3.11/site-packages/ivy/func_wrapper.py:440: UserWarning: In the case of Compositional function, operators might cause inconsistent behavior when array_mode is set to False
  warnings.warn(
torch.Size([0])

[YushaArif99]

Hi @34j. Apologies for overlooking the 2nd example, totally slipped under my radar 😅

I have opened another PR with the torch fix as well which should hopefully resolve this 2nd issue.

These are the outputs I am getting:

- import ivy; ivy.set_backend('numpy'); ivy.set_array_mode(False); print(ivy.log(1))
>>> 0.0

- import ivy; ivy.set_backend('numpy'); ivy.set_array_mode(True); print(ivy.log(1))
>>> ivy.array(0.)

- import ivy; ivy.set_backend('torch'); ivy.set_array_mode(False); print(ivy.log(1))
>>> tensor(0.)

- import ivy; ivy.set_backend('torch'); ivy.set_array_mode(True); print(ivy.log(1))
>>> ivy.array(0.)