Pylint: Incorrect inference on Path objects

[pylint-bot]

Originally reported by: Antony Lee (BitBucket: anntzer, GitHub: @anntzer?)

---

Linting

#!python
from pathlib import Path
Path("foo").with_name("bar").open("rb")

gives

E:  2,29: Instance of 'PurePath' has no 'open' member (no-member)

but with_name actually returns a Path here (more generally, an object with the same class as the initial object), not a PurePath, so the inference and the error message are wrong.

---

• Bitbucket: https://bitbucket.org/logilab/pylint/issue/224

[pylint-bot]

_Original comment by_ Claudiu Popa (BitBucket: PCManticore, GitHub: @PCManticore):

---

I'll have a look soon.

[xrr]

Will it be fixed any time soon ? Thanks.

[PCManticore]

Probably not as soon as I was expecting it to be. Most probably it requires a special astroid brain tip for Path objects.

[rogalski]

Minimal astroid examples that expose a bug:

def test1(self):
    ast = extract_node("""
    class C:
        @classmethod
        def factory(cls):
            return cls()

        def m(self):
            return self.__class__.factory()

    class D(C):
        pass

    C().m() #@
    D().m() #@
    """, 'm')
    should_be_C = list(ast[0].infer())
    should_be_D = list(ast[1].infer())
    self.assertEqual(1, len(should_be_C))
    self.assertEqual(1, len(should_be_D))
    self.assertEqual('m.C', should_be_C[0].qname())
    self.assertEqual('m.D', should_be_D[0].qname())

def test2(self):
    ast = extract_node("""
    class C:
        @classmethod
        def factory(cls):
            return object.__new__(cls)

        def m(self):
            return self.__class__.factory()

    class D(C):
        pass

    C().m() #@
    D().m() #@
    """, 'm')
    should_be_C = list(ast[0].infer())
    should_be_D = list(ast[1].infer())
    self.assertEqual(1, len(should_be_C))
    self.assertEqual(1, len(should_be_D))
    self.assertEqual('m.C', should_be_C[0].qname())
    self.assertEqual('m.D', should_be_D[0].qname())

test1 passes, test2 fails. The only difference is a way how new instance of object is created in factory.

It seems like dealing with a new callcontext results in lack of data needed to infer these types properly. I don't know how to fix it (yet), so I'll just leave it there, maybe someone can figure this out.

See pathlib.py source for reference.

[butla]

Any update on this? I fail the tests on Pylint issues and the need to disable the error on every use of .resolve() is quite annoying.

[PCManticore]

No one is currently working on this. We have lots of issues in our backlog and we're all volunteers :-) We'll get to it at some point.

[butla]

@PCManticore Of course. And your sacrifices are much appreciated :) I didn't want to sound like giving orders or requesting you to fix it right now. I was just pointing out that this is still a problem, and these little things can drive people away from Pylint. But I myself am not switching to flake8 :)