False positives when using methods that pylint saw being once overridden at instance level

[jstasiak]

Steps to reproduce

1 Have the following file (xyz.py):

class A:
    def get_stuff(self, x):
        return x


def test1():
    def g():
        pass
    a = A()
    a.get_stuff = g


def test2():
    a = A()
    value = a.get_stuff(1)

2 Run pylint -f parseable xyz 2> /dev/null | grep "\[E"

Current behavior

Pylint detects errors where it shouldn't:

xyz.py:15: [E1111(assignment-from-no-return), test2] Assigning to function call which doesn't return
xyz.py:15: [E1121(too-many-function-args), test2] Too many positional arguments for function call

Expected behavior

No errors reported (well, maybe pylint could detect incompatible type on a.get_stuff = g assignment)

pylint --version output

No config file found, using default configuration
pylint 1.6.4, 
astroid 1.4.8
Python 3.5.2 (default, Jun 29 2016, 13:43:58) 
[GCC 4.2.1 Compatible Apple LLVM 7.3.0 (clang-703.0.31)]

Keywords for visibility: type inference

[clavedeluna]

I looked into this issue and it's decidedly an inference issue in astroid.

In the assignment-from-no-return case the inference returned in function_node = safe_infer(node.value.func) references <FunctionDef.g l.7 at 0x111e19370>.

Similarly, in the too-many-function-args, the inference returned in called = safe_infer(node.func) references <FunctionDef.g l.7 at 0x10225b0a0>

Both of these should've returned <ClassDef.A l.1 at 0x10ab643d0>. Astroid probably has this implementation to handle some other cases when getting the overriding value is correct, but in this case this isn't the correct behavior.