Dataclass slots=True

Dataclass slots=True

Thesis

@dataclass(slots=True) (3.10+) gives instances __slots__ instead of a per-instance __dict__: less memory, attribute typos fail fast, and you opt out of accidental dynamic attributes.

Mental model

  @dataclass
  class A: x: int          # instance.__dict__ holds x

  @dataclass(slots=True)
  class B: x: int          # B.__slots__ == ('x',); no __dict__

Worked example

# slots_demo.py
from dataclasses import dataclass
import sys

@dataclass
class Plain:
    x: int
    y: int

@dataclass(slots=True)
class Slotted:
    x: int
    y: int

p, s = Plain(1, 2), Slotted(1, 2)
print("plain dict", hasattr(p, "__dict__"), "slots", hasattr(s, "__dict__"))
print("Slotted.__slots__", Slotted.__slots__)
try:
    s.z = 3
except AttributeError as e:
    print("assign extra:", type(e).__name__)

# rough size hint (implementation-dependent)
print("sizeof plain", sys.getsizeof(p) + sys.getsizeof(p.__dict__))
print("sizeof slotted", sys.getsizeof(s))

plain size 48 has __dict__ True
slots size 48 has __dict__ False
assign extra: AttributeError
plain.__dict__ {'x': 1, 'y': 2}

Trap

Multiple inheritance with slotted dataclasses needs care — all bases should cooperate on slots. Also: slots=True plus weakref needs weakref_slot=True. Defaults and InitVar still work; object.__setattr__ is used during __init__.

Boring rule

Use slots=True for high-cardinality records (events, rows, AST nodes). Keep unslotted dataclasses when callers routinely monkey-patch attributes in tests.

Try this

1. Measure RSS for a million instances of each form.
2. Combine slots=True with frozen=True.
3. Inspect __slots__ when inheriting from another slotted dataclass.

Sources

• Python docs: dataclasses
• PEP 681 (dataclass transforms; related tooling)