Pylint gets TensorFlow's tf.split all wrong

[bersbersbers]

Steps to reproduce

import tensorflow as tf
tf.split([0, 0], num_or_size_splits=[1, 1], axis=-1)
print("Done.")

Current behavior

Code runs fine, but

bug.py:3:0: E1123: Unexpected keyword argument 'num_or_size_splits' in function call (unexpected-keyword-arg)
bug.py:3:0: E1124: Argument 'axis' passed by position and keyword in function call (redundant-keyword-arg)
bug.py:3:0: E1120: No value for argument 'value' in function call (no-value-for-parameter)
bug.py:3:0: E1120: No value for argument 'num_split' in function call (no-value-for-parameter)

Expected behavior

Perfect score

pylint --version output

tensorflow 2.2.0
pylint 2.5.2
astroid 2.4.1
Python 3.8.2 (default, Feb 26 2020, 09:52:54) 
[GCC 7.5.0]

[bersbersbers]

Maybe related:

import tensorflow as tf
print(tf.random.uniform(()))

yields

bug.py:3:6: E1120: No value for argument 'dtype' in function call (no-value-for-parameter)

although the code runs fine.

[jxi24]

Maybe related:

import tensorflow as tf

x = tf.convert_to_tensor([1, 2, 3])
print(tf.cast(x, dtype=tf.float64))
y = tf.convert_to_tensor([1, 2, 3])
print(tf.concat([x,y], axis=-1))
print(-tf.math.log(x))

yields

bug.py:4:0: E1123: Unexpected keyword argument 'dtype' in function call (unexpected-keyword-arg)
bug.py:4:0: E1120: No value for argument 'DstT' in function call (no-value-for-parameter)
bug.py:6:6: E1123: Unexpected keyword argument 'axis' in function call (unexpected-keyword-arg)
bug.py:6:6: E1120: No value for argument 'values' in function call (no-value-for-parameter)
bug.py:7:6: E1130: bad operand type for unary -: object (invalid-unary-operand-type)

but outputs the following expected result:

tf.Tensor([1. 2. 3.], shape=(3,), dtype=float64)
tf.Tensor([1. 2. 3. 1. 2. 3.], shape=(6,), dtype=float32)
tf.Tensor([-0.        -0.6931472 -1.0986123], shape=(3,), dtype=float32)

[cohama]

Same issue.
I noticed tensorflow==2.2.0 causes this problem but tensorflow==2.1.0 dose not.

I also try to specify ignored-modules but it didn't work.

"""a"""
import tensorflow as tf


tf.concat([1, 2], axis=0)

with the following .pylintrc

[TYPECHECK]
ignored-modules=tensorflow

yields

************* Module main
main.py:5:0: E1123: Unexpected keyword argument 'axis' in function call (unexpected-keyword-arg)
main.py:5:0: E1120: No value for argument 'values' in function call (no-value-for-parameter)

My environment is as follows.

pylint 2.5.2
astroid 2.4.1
Python 3.8.0 (default, Oct 16 2019, 10:08:31)
[GCC 9.2.0]

[fel-thomas]

Same for me with concat and other functions from tensorflow, since 2.2.0.

Env

pylint 2.5.2
astroid 2.4.1
Python 3.7.5 (default, Nov  7 2019, 10:50:52) 
[GCC 8.3.0]

[charmasaur]

Looks as though the combination of TF 2.2.0 and pylint 2.5 is picking up the internal gen_array_ops functions (e.g. def split(axis, value, num_split, name=None)) instead of the exported ones.

I don't know enough about TF's exporting, nor about pylint, to have any idea why that would be happening though...

[charmasaur]

Did some bisecting on the pylint side and came up with this: 01dfa52

I think what's happening is:

• astroid is inferring multiple possible functions for the split (or similar) call, due to various wildcard imports in TF (e.g. https://github.com/tensorflow/tensorflow/blob/v2.2.0/tensorflow/python/ops/array_ops.py#L40)
• all those functions have the same pytype (they're all just functions), so safe_infer is returning the first hit, which happens to be the one from gen_array_ops

Maybe safe_infer needs to be a little more circumspect when deciding whether a list of inferred values is ambiguous?

[charmasaur]

+@PCManticore, any ideas about what we can do here? (see comment above)

[atyshka]

Also complaining about missing args to tf.sparse.reorder in v2.3.0

[alfonso-landin]

This does not happen with the pylint version installed with my package manager (2.2.2):

pylint3 2.2.2
astroid 2.1.0
Python 3.7.3 (default, Jul 25 2020, 13:03:44) 
[GCC 8.3.0]

And still happens in pylint 2.6.0 installed in a virtualenv:

pylint 2.6.0
astroid 2.4.2
Python 3.7.3 (default, Dec 20 2019, 18:57:59) 
[GCC 8.3.0]

In case it is useful to know, this also affects the functions tf.one_hot, tf.gather and tf.concat in tensorflow v2.2.0

[andraghetti]

Any update on this?

[Pierre-Sassoulas]

Hello @andraghetti , no one worked on it, but we'll gladly review a merge request addressing this issue.