Performance version: 1.11.0Python version: 3.14.0a0 (64-bit) revision 1a84bdc237Report on macOS-14.5-arm64-arm-64bit-Mach-ONumber of logical CPUs: 8Start date: 2024-07-02 01:23:14.690599End date: 2024-07-02 01:23:51.844329### deepcopy ###Mean +- std dev: 123 us +- 12 us -> 120 us +- 1 us: 1.03x fasterSignificant (t=2.25)### deepcopy_memo ###Mean +- std dev: 14.1 us +- 0.2 us -> 14.1 us +- 0.3 us: 1.00x slowerNot significant### deepcopy_reduce ###Mean +- std dev: 1.22 us +- 0.06 us -> 1.21 us +- 0.01 us: 1.00x fasterNot significant

from dataclasses import dataclass@dataclassclass DC:    a : list = []    x=DC()

import attrs@attrs.defineclass DC:    a : list = []    x=DC()y=DC()x.a.append(1)print(x)print(y)

DC(a=[1])DC(a=[1])

import pyperfrunner = pyperf.Runner()setup="""import copyfrom dataclasses import dataclass@dataclassclass A:    a : list    b : str    c : bool    d: tuple    e : set    dc=A([1,2,3], 'hello', True, (), {})empty_tuple = tuple()empty_list = []"""runner.timeit(name="deepcopy empty tuple", stmt="b=copy.deepcopy(empty_tuple)", setup=setup)runner.timeit(name="deepcopy empty list", stmt="b=copy.deepcopy(empty_list)", setup=setup)runner.timeit(name="deepcopy dataclass", stmt="b=copy.deepcopy(dc)", setup=setup)

deepcopy empty tuple: Mean +- std dev: [dc_main] 827 ns +- 37 ns -> [dc_fastpath_empty] 203 ns +- 7 ns: 4.06x fasterdeepcopy empty list: Mean +- std dev: [dc_main] 1.00 us +- 0.09 us -> [dc_fastpath_empty] 233 ns +- 24 ns: 4.29x fasterdeepcopy dataclass: Mean +- std dev: [dc_main] 8.02 us +- 0.12 us -> [dc_fastpath_empty] 8.22 us +- 0.07 us: 1.02x slowerGeometric mean: 2.57x faster

(env312) c:\projects\misc\cpython\PCbuild>python -m pyperf compare_to dc_main.json dc_c.jsondeepcopy empty tuple: Mean +- std dev: [dc_main] 827 ns +- 37 ns -> [dc_c] 90.6 ns +- 9.8 ns: 9.13x fasterdeepcopy empty list: Mean +- std dev: [dc_main] 1.00 us +- 0.09 us -> [dc_c] 216 ns +- 24 ns: 4.63x fasterdeepcopy dataclass: Mean +- std dev: [dc_main] 8.02 us +- 0.12 us -> [dc_c] 4.06 us +- 0.43 us: 1.97x fasterGeometric mean: 4.37x faster

fromdataclassesimportdataclass,as_dict@dataclassclassSome:field:set[int]as_dict(Some(set()))

from dataclasses import dataclass, asdict@dataclass class Some:    a: str    t: tuple[int]    field: set[int] # benefits    l : list[int]    d : dict[str, str]    fs: frozenset[int] # benefits    nonempty : set[int]s=Some('x', tuple(), set(), [], d={}, fs=frozenset(), nonempty={1, 2})asdict(s)

importpyperfrunner=pyperf.Runner()setup="""from dataclasses import dataclass, asdict@dataclassclass Some:    a: str    t: tuple[int]    field: set[int] # benefits    l : list[int]    d : dict[str, str]    fs: frozenset[int] # benefits    nonempty : set[int]s = Some('x', tuple(), set(), [], d={}, fs=frozenset(), nonempty={1, 2})"""runner.timeit(name="dataclass asdict",stmt=f"b = asdict(s)",setup=setup)

Mean +- std dev: [asdict-baseline] 6.26 us +- 0.07 us -> [asdict-opt] 3.39 us +- 0.18 us: 1.85x faster