Make igetattr() idempotent

[jacobtylerwalls]

Type of Changes

	Type
✓	🐛 Bug fix

Description

See regression test. The second time an instance's attribute is inferred with the same context (which pylint has legitimate use cases for, see existing test that was improved!), we still need to be able to get a successful inference result. Instead, we were getting Uninferable due to a performance optimization added in 2d7a87b.

This addresses some reports of varying results when running pylint with --jobs.

Closes pylint-dev/pylint#4356
Refs #7

More info

The original inconsistency added in 2d7a87b targeted a report in #7, but we have no source for the original code being tested.

I attempted to guess what the bug report in #7 was. I came up with the following based on #7 (comment), and it lints fine:

class A:
    def __init__(self) -> None:
        self.transaction = self.one() or self.two() or self.three() or self.four() or self.five()
        self.transaction #@

    def one(self):
        self.transaction = trans
        trans = self.transaction

    def two(self):
        self.transaction = trans
        trans = self.transaction

    def three(self):
        self.transaction = trans
        trans = self.transaction

    def four(self):
        self.transaction = trans
        trans = self.transaction

    def five(self):
        self.transaction = trans
        trans = self.transaction

I don't want to commit this file to the tests, though, because it's a wild guess.

[codecov]

Codecov Report

Merging #2208 (df30870) into main (4493399) will increase coverage by 0.00%.
The diff coverage is n/a.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2208   +/-   ##
=======================================
  Coverage   92.68%   92.68%           
=======================================
  Files          94       94           
  Lines       10820    10828    +8     
=======================================
+ Hits        10028    10036    +8     
  Misses        792      792           

Flag	Coverage Δ
linux	92.44% <ø> (+<0.01%)	⬆️
pypy	87.64% <ø> (-0.01%)	⬇️
windows	92.27% <ø> (+<0.01%)	⬆️

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

Impacted Files	Coverage Δ
astroid/bases.py	87.90% <ø> (-0.07%)	⬇️

... and 1 file with indirect coverage changes

[Pierre-Sassoulas]

It seems like the original reason to do that was for performance. Did you check the performance or just the tests suite result when testing in pylint ? (We're probably going to merge anyway as correctness is more important than performance especially if we can't parralelize without fixing this.. but is it really bad ?)

[jacobtylerwalls]

It seems like the original reason to do that was for performance.

Right, but only for the pathological case where there are self-references created like in #7 (comment). So I wouldn't expect much of a performance impact.

Did you check the performance or just the tests suite result when testing in pylint ?

Just the test suite, and then also trying to reproduce the original pathological case, but couldn't. I just now linted a couple packages and there was no significant difference in the total time.

[Pierre-Sassoulas]

Thank you for checking !

[jacobtylerwalls]

Thanks for the review!