dataclass(slots=True) does not account for slots in base classes

[ariebovenberg]

BPO	46382
Nosy	@ericvsmith, @hynek, @serhiy-storchaka, @TeamSpen210, @sobolevn, @AlexWaygood, @ariebovenberg
PRs	bpo-46382 dataclass(slots=True) now takes inherited slots into account #31980

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/ericvsmith'
closed_at = <Date 2022-03-19.21:01:51.297>
created_at = <Date 2022-01-14.19:31:43.335>
labels = ['type-bug', 'library', '3.10', '3.11']
title = 'dataclass(slots=True) does not account for slots in base classes'
updated_at = <Date 2022-03-19.21:01:51.297>
user = 'https://github.com/ariebovenberg'

bugs.python.org fields:

activity = <Date 2022-03-19.21:01:51.297>
actor = 'eric.smith'
assignee = 'eric.smith'
closed = True
closed_date = <Date 2022-03-19.21:01:51.297>
closer = 'eric.smith'
components = ['Library (Lib)']
creation = <Date 2022-01-14.19:31:43.335>
creator = 'ariebovenberg'
dependencies = []
files = []
hgrepos = []
issue_num = 46382
keywords = ['patch']
message_count = 19.0
messages = ['410591', '410601', '410628', '410630', '410641', '410642', '410820', '410841', '411010', '415393', '415399', '415400', '415401', '415419', '415453', '415471', '415474', '415570', '415571']
nosy_count = 7.0
nosy_names = ['eric.smith', 'hynek', 'serhiy.storchaka', 'Spencer Brown', 'sobolevn', 'AlexWaygood', 'ariebovenberg']
pr_nums = ['31980']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue46382'
versions = ['Python 3.10', 'Python 3.11']

[ariebovenberg]

@DataClass(slots=True) adds slots to dataclasses. It adds a slot per field.
However, it doesn't account for slots already present in base classes:

>>> class Base:
...     __slots__ = ('a', )
...
>>> @dataclass(slots=True)
... class Foo(Base):
...     a: int
...     b: float
...
>>> Foo.__slots__
('a', 'b')  # should be: ('b', )

The __slots__ documentation says:

If a class defines a slot also defined in a base class, the instance variable 
defined by the base class slot is inaccessible (except by retrieving its descriptor 
directly from the base class). This renders the meaning of the program undefined. 
In the future, a check may be added to prevent this.

Solution: don't add slots which are already defined in any base classes:

>>> @dataclass
... class Bla(Base):
...     __slots__ = ('b', )
...     a: int
...     b: float
...
>>> Bla(4, 5.65)
Bla(a=4, b=5.65)

If you agree, I'd like to submit a PR to fix this. I already have a prototype working.

[ericvsmith]

I'll have to do some more research. But your analysis looks correct to me, so far.

[ariebovenberg]

There are already 2 complexities I can think of:

1. This behavior may break some people's code, if they use slots to iterate over
   the fields of a dataclass. Solution: explicitly mention in the docs that
   not every field may get a slot on the new class. Advise them to use
   fields() to iterate over the fields.
2. It's technically allowed for slots to be an iterator (which will then be
   exhausted at class creation). Finding the slots of such a class
   may require more elaborate introspection.

[sobolevn]

Arie, can you please explain what is the technical difference between these two cases:

class A:
    __slots__ = ('a', )
    # fields

class B(A):
    __slots__ = ('a', 'b')
    # fields

And:

class C:
    __slots__ = ('a', )
    # fields

class D(C):
    __slots__ = ('b', )
    # fields

?

[TeamSpen210]

Both will function, but class B will add its slots after A's, causing there to be an extra unused slot in the object that you can only access by directly using the A.a descriptor. So all slotted inheriting dataclasses cause the object to use more memory than necessary.