dataclass.__eq__ no longer errors on missing attributes in Python 3.13

[willhcr]

Bug report

Bug description:

Description

Equality with self now returns True in 3.13 for dataclasses with deleted attributes. Previously, this would raise an exception.

This may be related to changes discussed in #128294. However, unlike
other reports (#120645, #116647) where something that was True in 3.12 is now False or an Exception,
this issue is the inverse. In 3.13, the equality test silently returns True when previously it would raise.

This class is badly behaved, it should probably still raise an error.

Test Case

import dataclasses


@dataclasses.dataclass
class Test:
    a: int

    def __init__(self):
        self.a = 1

    def del_a(self):
        del self.a


o1 = Test()
o2 = Test()

o1.del_a()
o2.del_a()

print(o1 == o1)  # True (3.13), raises AttributeError (3.12)
print(o1 == o2)  # raises AttributeError (3.13, 3.12)
# AttributeError: 'Test' object has no attribute 'a'

CPython versions tested on:

3.13, 3.12

Operating systems tested on:

Linux

[ZeroIntensity]

I'm not sure there's anything we can do at this point.

[willhcr]

A note in the 3.13 What's New would be helpful.

For context, I discovered this when debugging behavior differences in async library internals between 3.12 and 3.13. A dataclass with deleted attributes was stored in a contextvar:
https://github.com/django/asgiref/blob/5eff04dce46b74172a1d6d1c5b02409f0afe33b6/asgiref/sync.py#L44-L53

[F18-Ray]

Bug report

Bug description:

Description

Equality with self now returns True in 3.13 for dataclasses with deleted attributes. Previously, this would raise an exception.

This may be related to changes discussed in #128294. However, unlike other reports (#120645, #116647) where something that was True in 3.12 is now False or an Exception, this issue is the inverse. In 3.13, the equality test silently returns True when previously it would raise.

This class is badly behaved, it should probably still raise an error.

Test Case

import dataclasses

@dataclasses.dataclass
class Test:
a: int

def __init__(self):
    self.a = 1

def del_a(self):
    del self.a

o1 = Test()
o2 = Test()

o1.del_a()
o2.del_a()

print(o1 == o1) # True (3.13), raises AttributeError (3.12)
print(o1 == o2) # raises AttributeError (3.13, 3.12)

AttributeError: 'Test' object has no attribute 'a'

CPython versions tested on:

3.13, 3.12

Operating systems tested on:

Linux

I watched your code and I think I fix your bug.
`import dataclasses
@dataclasses.dataclass
class Test:
#a: int # why use "a:int"

def __init__(self):
    self.a = 1


def del_a(self):
    del self.a

o1 = Test()
o2 = Test()

o1.del_a()
o2.del_a()

print(o1 == o1)
print(o1 == o2)`
You may try this, and I actually don't know why you use "a:int", It's use less.

[rhettinger]

This class is badly behaved, it should probably still raise an error.

FWIW no error is raised if o1 is stored in a list instead of dataclass:

>>> s = [o1]
>>> s == s
True

[rhettinger]

I'm not sure there's anything we can do at this point.

And given that other tools like list do not raise an exception, there's not anything we should do at the point.

Closing for now. If this proves problematic in practice, we can reopen.