Pylint: False-positive E1137 (unsupported-assignment-operation) with np.empty_like

[bersbersbers]

Steps to reproduce

pylint this file:

"""X"""
import numpy as np

X = np.empty_like([1])
X[:] = 2

Current behavior

************* Module bug
bug.py:5:0: E1137: 'X' does not support item assignment (unsupported-assignment-operation)

Expected behavior

No error, code runs fine in python bug.py

pylint --version output

pylint 2.2.2
astroid 2.1.0
Python 3.7.2 (tags/v3.7.2:9a3ffc0492, Dec 23 2018, 23:09:28) [MSC v.1916 64 bit (AMD64)]`

[PCManticore]

Hi @bersbersbers thanks for reporting the issue.

[martinholmer]

Same problem as reported by @bersbersbers in #2767 occurs when using np.zeros_like(...).

Also, use of np.logical_or(..., ...) generates an assignment-from-no-return pylint error.

Neither of these errors were generated by earlier versions of pylint.

cc @PCManticore

[hippo91]

@martinholmer @bersbersbers thank for the report.
In fact it's very close to #2746 #2694 #2784.

In early times of pylint the brain used to deal with numpy was quite simple. We, then, had a lot of no-member false positives. They were due to the fact that some numpy modules are written in c language, especially those at the core of numpy and thus astroid (the pylint inference engine) can not infer any object defined in those modules.

To correct this, i wrote python ghosts of those numpy c modules in order astroid/pylint to deal with them correctly (PyCQA/astroid#567 PyCQA/astroid#486). It implied studying numpy deeply and wrote a python ghost for every function/class that numpy defines in c language. At this point almost all classes or functions are defined and thus almost all no-member false positives should have gone away. However the call to such functions/methods cannot yet be inferred. That's why you are facing such problems.

I'am currently working on an improvment of astroids numpy brain. It's quite a hard work and have not so much time to spent on it but i hope it will be finished quite soon.

Sorry for that long message but i wanted to explain why you and other numpy + pylint users are facing such problems.

[martinholmer]

@hippo91 said in issue #2767:

Sorry for that long message but I wanted to explain why you and other numpy + pylint users are facing such problems.

Thanks for the helpful explanation, and thanks especially for all your ongoing work on this problem.

[jordyjwilliams]

not sure if this helps:
but I found that the following throwed a unsupported-assignment-operation

my_array = np.zeros_like(another_array)
condition_array = _np.where(_np.less(different_array, my_array))
my_array[condition_array] = values

that could be solved by calling np.zeros(another_array.shape)
instead of zeros_like

[martinholmer]

@jordyjwilliams said in issue #2767:

I found that the following throwed a unsupported-assignment-operation pylint warning:

my_array = np.zeros_like(another_array)
condition_array = _np.where(_np.less(different_array, my_array))
my_array[condition_array] = values

that could be solved by calling np.zeros(another_array.shape) instead of np.zeros_like(another_array).

Thanks for the useful tip!

[hippo91]

@martinholmer , @jordyjwilliams
The problem you were facing should have disappeared with, at least, the 2.3 version of astroid.
That's why i'm closing this issue.
Feel free to reopen it if necessary.