False positive for undefined variable with assignment expression

[hauntsaninja]

Steps to reproduce

# ~/tmp λ cat test4.py
def i(am: str = (the := "walrus")) -> str:
    return "goo goo g'joob"
print(the)

Current behavior

~/tmp λ pylint -E test4.py
************* Module test4
test4.py:5:6: E0602: Undefined variable 'the' (undefined-variable)

Expected behavior

pylint recognises that the is defined. (Or alternatively, pylint yells at me for doing this weird thing in the first place ;-) )

pylint --version output

~/tmp λ pylint --version
pylint 2.5.3
astroid 2.4.2
Python 3.8.3 (default, May 27 2020, 20:54:22) 
[Clang 11.0.3 (clang-1103.0.32.59)]

Pretty minor, especially given that assignment expressions aren't a priority for pylint (#3275 (comment)).

[bersbersbers]

Here's another case. Interestingly,

x = next(i for i in range(10) if (y := i * i))
print(y)

raises this false-positive, while

next(i for i in range(10) if (y := i * i))
print(y)

does not.

[egormm]

A couple more interesting cases with false-positive behaviour:

print({'key': value if (value := 'something') else 'anything'})

raises used-before-assigment, while

value = val if (val := 'something') else 'anything'
print({'key': value})

doesn't. Excepting behaviour - no errors.
Same with classes and dataclasses:

class Dummy:
    def __init__(self, value):
        self.value = value

dummy = Dummy(value=val if (val := 'something') else 'anything')
print(dummy.value)

raises used-before-assigment, while

class Dummy:
    def __init__(self, value):
        self.value = value

value = val if (val := 'something') else 'anything'
dummy = Dummy(value)
print(dummy.value)

doesn't. Excepting behaviour - no errors.

[DanielNoord]

If have created a new issue for the last two issues of @egormm's comments, as they are related to a different message.

[hauntsaninja]

@Pierre-Sassoulas @DanielNoord I don't think #5213 fixes the original case mentioned here (use of walrus in function default), should this issue be re-opened?

[DanielNoord]

It should be fixed with #5188. I separated this issue in two PR's as they covered different cases.

[hauntsaninja]

Awesome, missed that, thank you so much!!