None comparison is not possible in new versions

[bloussou]

Hello,
In 1.1.224 it was possible to have the following code but it is not possible with 1.1.258. Is this an expected behavior ?

from typing import Any

class B:
    pass

class A:
    def __eq__(self, other: Any) -> B:  # type: ignore[override]
        print("b")
        return B()
# It was not failing before because I think it was inferring the __eq__ was overridden
if A() == None: # Condition will always evaluate to False since the types "A" and "None" have no overlap
    ...

[erictraut]

The current behavior is "as designed" even if it's not entirely expected.

I see that you have enabled the strict-mode check reportUnnecessaryComparison. This check needs to make some assumptions about the way comparisons work. These assumptions work in most use cases, but there are edge cases where the heuristic will fail.

You're overriding __eq__ in a very unusual way in this code. In fact, you're overriding the __eq__ method for object in an incompatible manner and then ignoring the error using a # type: ignore comment.

I've fixed many bugs and made other changes since version 1.1.224, so I can't say specifically what caused this to work previously. If you can find the specific version where this behavior changed, I'm willing to investigate further.

[bloussou]

Hello,
Thanks for your answer.
This behavior changed in version 1.1.129

[erictraut]

Thanks. Yes, I introduced a bug fix in 1.1.129 that would explain why you didn't see this behavior prior to that.

The check is working as expected. The assumption is that a comparison of some class instance with None will always produce a False result. Of course, with Python you can always override __eq__ and __ne__ in weird and unusual ways, but checks like this one need to make assumptions that they will be used in a standard manner. If you find that the reportUnnecessaryComparison check doesn't meet your needs, you can disable it.

[bloussou]

indeed, we should think of something like pandas isnull() probably. Thanks a lot for your explanations.