dataclass re-decoration timing

main.py

# See comment below for an explanation of what this script is testing!

from dataclasses import dataclass, is_dataclass
from timeit import timeit


@dataclass
class A:
    a: int = 10


class B:
    a: int = 10


assert is_dataclass(A)
assert not is_dataclass(B)

n = 1_000

print(timeit('dataclass(A)', globals=globals(), number=n))  # 0.138
print(timeit('dataclass(B)', globals=globals(), number=n))  # 0.134

Notes

Ideally, the performance timing of a call to dataclass(A) should be in O(1) time, and be as close to zero as possible - which indicates that the dataclass function is a no-op in scenarios where a class is already a dataclass.

However, what we instead observe, is that it takes the same time to call the @dataclass decorator (or function), regardless if the input class is already a dataclass or not. Thus, timings for dataclass(A) ~= dataclass(B), which means that we should avoid repeat calls to the dataclass() function if at all possible.