Prefer last same-named function in a class rather than first in igetattr()

[nelfin]

Steps

• For new features or bug fixes, add a ChangeLog entry describing what your PR does.
• Write a good description on what the PR does.

Description

Ref #1015. When there are multiple statements defining some attribute
ClassDef.igetattr filters out later definitions if they are not in the
same scope as the first (to support cases like "self.a = 1; self.a = 2"
or "self.items = []; self.items += 1"). However, it checks the scope of
the first attribute against the parent scope of the later attributes.
For mundane statements this makes no difference, but for
scope-introducing statements such as FunctionDef these are not the same.

Type of Changes

	Type
✓	🐛 Bug fix

Related Issue

Closes #1015

[cdce8p]

I'm wondering if

class A:
    def foo(self): ...
    def foo(self):
        yield

should be inferred as Const and Generator as Python will always use the last definition. Wouldn't it be better to only infer it as Generator?

--
Some test cases to be aware of

if some_test:
    def func():
        return 42
else:
    def func():
        return bool

# should still return two results for `func()`

Functools.singledispatch should still work:
https://docs.python.org/3/library/functools.html#functools.singledispatch

[nelfin]

Wouldn't it be better to only infer it as Generator?

My thinking leans the same way as it's the more "useful" definition.

Python will always use the last definition

Pretty much, but I wonder what to do about monkeypatching, or this (contrived) example:

class A:
    def foo(self):
        return 'first'
    bar = foo(object())
    def foo(self):
        return 'second'

A().foo()  # 'second'
A().bar  # 'first'

[cdce8p]

Just tested your example: This will already be inferred incorrectly. A().foo() ->'first'. Possibly another bug?

As for the issue at hand. Would it work to filter previous definitions only if no other attributes are in between?

[Pierre-Sassoulas]

The case with two foo function one redefined are a little contrived and probably not seen a lot in real "sane" code. The @overload use case though seems interesting to test. Do you think this is really equivalent ?

[DanielNoord]

@nelfin Just going through some stale PRs. Have you given the questions raised here some thoughts? Perhaps somewhat related is my comment here which also relates to "picking the correct definition".

[jacobtylerwalls]

We're closing tons of duplicates for this in pylint. I think special-casing functions (sans properties) is good enough.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 92.75%. Comparing base (12ed435) to head (06a5f2b).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1173   +/-   ##
=======================================
  Coverage   92.75%   92.75%           
=======================================
  Files          94       94           
  Lines       11080    11084    +4     
=======================================
+ Hits        10277    10281    +4     
  Misses        803      803           

Flag	Coverage Δ
linux	92.56% <100.00%> (+<0.01%)	⬆️
pypy	90.76% <100.00%> (+<0.01%)	⬆️
windows	92.34% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
astroid/nodes/scoped_nodes/scoped_nodes.py	92.52% <100.00%> (+0.02%)	⬆️