false-positive: E0401 import-error for relative imports inside __init__.py (but only sometimes)

[Kwasniok]

Bug description

Description

Relative imports inside a __init__.pyfile cause random E0401.

E.g.:
Say one has a module A.B.C.D inside a folder src such that

• All the required __init__.py files exist for modules A, B, C and D
• src/A/B/C/D/some_moduleX.py exists and defines SomeClassX where X=1,2,3
• Content of src/A/B/C/D/__init__.py contains

from .some_module1 import SomeClass1
from .some_module2 import SomeClass2
from .some_module3 import SomeClass3

pylint will issue E0401 (pseudo-)randomly for some of the imports, even if all of them are similar. For which one the warnings are issued depends on the version/machine used - e.g. GitHub actions uses pylint v.2.11.1 and gives me different locations and a different amount of E0401.
If there are only a few relative imports across the entire project, pylint might not issue E0401 at all.

Python code using this module works flawlessly. E.g. via from A.B.C.D import SomeClass2

Workaround

Adding a .pylintrc

[MASTER]
init-hook='import sys; sys.path.append(".")'

fixes the issue.

Sad Note

I tried to come up with a minimalistic example with nested modules all containing relative imports in the __init__.py files importing dummy classes from the same level but for some reason this bug could not be provoked that way. I tried for quite a while but this is the best I can do.

Related

• False positive import-error #3984
• False positive import error on two files in the same directory with an __init__.py #4226

Configuration

No response

Command used

pylint src

Pylint output

************* Module src.A.B.C.D
src/A/B/C/D/__init__.py:2:0: E0401: Unable to import 'src.A.B.C.D.some_module2' (import-error)

# NOTE: The exact output is inconsistent across versions/machines.

Expected behavior

no warnings

Pylint version

pylint 2.10.2
astroid 2.7.3
Python 3.9.7 (default, Oct 10 2021, 15:13:22)
[GCC 11.1.0]

OS / Environment

ArchLinux 5.15.2-arch1-1 (Fri, 12 Nov 2021)

Additional dependencies

No response

[Kwasniok]

I noticed if I change the order of relative imports n a __init__.py, the exact same lines as before cause an issue.

[Kwasniok]

I wrote a script which generates a minimal example with N nested submodules each including M files each. Unfortunately, the issue does not occur in the minimal example.
I also checked for cyclic import in the original code but there is none and the critical sections of the code are also successfully interpreted and executed by python.

make_minimal_example_E0401.txt (note: change the extension to .py since GitHub does not allow to upload script files)

[jacobtylerwalls]

Thanks for the report. The indeterminacy is described in #3651, which appears to be fixed in recent pylint versions (anything shipping with astroid > 2.9.1). Feel free to reopen if still an issue.

[jacwalte]

It looks like this issue is still present - I have tried on 2.14.4 and 2.14.5 with the same issue ... sometimes happening

We are currently migrating from a Windows pipeline to a Linux pipeline in ADO when this issue first occured. I am not sure if it has to do with Linux or not, but we are only able to reproduce it there.

We have several repos where we have made this change recently with no issue. Also, we still have the Windows pipeline running and not seeing this issue.

astroid-2.11.7 pylama-8.3.8 isort-5.10.1 mccabe-0.7.0 pycodestyle-2.8.0 pydocstyle-6.1.1 pyflakes-2.4.0

I will continue to go down in versions to see if one is stable

@jacobtylerwalls for visibility

[jacobtylerwalls]

Thanks for testing, it's much appreciated! I'm going to take a note on #3651 that it may not be solved. (I declared victory too soon, it seems, since the regression test at #7117 isn't passing.) Still, unless there's a reason to treat these as separate issues, I'm thinking we can just keep #3651 open.