import resolution incorrect with same-named file and external module

[silverwind]

Bug description

When a python file and a external module have the same name, python resolves a import for that name to the external module, but pylint resolves it to the current file which causes false positives in rules import-self, no-name-in-module, no-member and more module-related rules.

To reproduce:

• Install redis pypi module using your favourite package manager
• Create a file redis.py with content import redis
• Lint the file using pylint redis.py

Pylint output

************* Module redis
redis.py:1:0: W0406: Module import itself (import-self)

Expected behavior

No error

Pylint version

pylint 2.11.1
astroid 2.8.2
Python 3.9.7 (default, Sep  3 2021, 12:45:31)
[Clang 12.0.0 (clang-1200.0.32.29)]

OS / Environment

macOS 11

Additional dependencies

redis==3.2.0

[timmartin]

I think if you create an file called redis.py in the current directory, with the content:

import redis

and that file is not part of a package then that file is importing itself, because the directory of the currently-executing script will always be first on sys.path, and so the import will find the local file redis.py before it searches libraries.

Where there might be an issue is with a file called redis.py within a package. In that case, the directory containing the redis.py will no longer be on that path. But in that case, I can't reproduce a case where pylint will complain about it importing itself. If I have the file be in a package (just add an __init__.py to the directory) then pylint stops complaining about importing self.

I suppose if the file in question is imported from another Python file, and is not in a package but is imported because its directory has been added to sys.path after the standard import locations, then pylint will report it as import self and Python will not, although I haven't tested this. I don't think pylint can know about runtime modification to sys.path.

Is there a more specific case that we should be investigating here?

[jacobtylerwalls]

I suggest closing this in favor of #7289 and retitling it -- that one focuses on the case where we have redis or gzip instead of redis.py.

[DanielNoord]

Sorry, I missed the investigation by Tim. That said, I can still get an incorrect import-self to trigger when I create a pylint.py file in the root of the pylint project with a import pylint line.

pylint-dev/astroid#1747 fixes this.

Edit: Never mind, pylint-dev/astroid#1747 doesn't fix this. Needs some investigation.

[jacobtylerwalls]

Thanks @DanielNoord. Can you include a snippet from a python session that verifies that python is not importing self? E.g. import pylint; print(pylint) and show the output?

[DanielNoord]

❯ cat pylint.py
import pylint

print(pylint.modify_sys_path)
❯ python pylint.py
<function modify_sys_path at 0x101dde050>

[DanielNoord]

Oh wait, pylint-dev/astroid#1747 does fix this. As the no-member warnings go away. As I explained in #7289 (comment) for the import-self to go away we need to actually change the import-self check.

But the successful removal of no-member warnings on pylint.modify_sys_path shows that this works.

[jacobtylerwalls]

I see something different:

$ cat gzip.py
import gzip
print(gzip)

$ python3 gzip.py
<module 'gzip' from '/Users/jwalls/somefolder/gzip.py'>
<module 'gzip' from '/Users/jwalls/somefolder/gzip.py'>

[DanielNoord]

Yeah, this is different than the original issue but I think there is still an issue here.
If there is a folder pylint which is a normal python package and there is a pylint.py file that imports pylint then the pylint directory gets precedence. However, due to how we cache modules this doesn't work in astroid. pylint-dev/astroid#1747 does fix this.

[DanielNoord]

You can also recreate this with:

cd /tmp
mkdir pylint
cd pylint
echo -e "X = 1" > __init__.py
cd ..
echo -e "import pylint\nprint(pylint.X)" > pylint.py

Without the fix in astroid this will warn about no-member, while Python executes this fine.

[DanielNoord]

Proposal, let pylint-dev/astroid#1747 fix this issue as this one was originally about import order resolution which I clearly what we are fixing there.

Rescope #7289 to fix import-self after we have access to that astroid update so that we can use the Module object instead of doing a string comparison for import-self.

@jacobtylerwalls Does that sound right?

[DanielNoord]

This was fixed in pylint-dev/astroid#1747. We don't need regression tests in pylint but this should be closed with the bump to astroid 2.13 in pylint at some point.

[jacobtylerwalls]

This was fixed with pylint 2.16's upgrade to astroid 2.13; please reopen if still an issue.