copium 0.1.0a0

Fast drop-in replacement for copy.deepcopy()

copium

Fast drop-in replacement for copy.deepcopy().

[!WARNING]
This is alpha version, see project status.

Benchmarked on GitHub Actions using tools/run_benchmark.py, charted with tools/generate_chart.py

Installation

pip install copium

Usage

Simply export COPIUM_PATCH_DEEPCOPY=1 before running your application. You don't have to change a single line of code:

cat example.py

from copy import deepcopy

assert deepcopy(x := []) is not x

COPIUM_PATCH_DEEPCOPY=1 python example.py

---

To enable patch manually:

import copium.patch

copium.patch.enable()

---

To use manually:

from copium import deepcopy

assert deepcopy(x := []) is not x

copium module includes all public declarations of stdlib copy module, so it's generally safe to:

- from copy import deepcopy
+ from copium import deepcopy

---

Project status

• passes all tests in CPython/Lib/test/test_copy.py
• behaves exactly the same as stdlib copy in all cases declared in datamodelzoo
• consumes less or equal amount of memory as stdlib copy
• always faster than stdlib copy
• passes all tests on free-threaded Python builds
• has tests for refcounts/recursion/threading edge cases

If you're interested in using this, simply add it to your project and run tests with COPIUM_PATCH_DEEPCOPY=1.

If it fails — open an issue with detailed bugreport.

If it works as expected, consider whether the speedup in your case worth of using alpha version.

Speedups on Python 3.14

tuple

• () ~5.66x
• tuple[int, ...] ~12.46x (20 items)
• tuple[int, ...] ~28.28x (5000 items)
• tuple[tuple[str, tuple[tuple[str, tuple[tuple[str, tuple[tuple[str, tuple[int, ...]]]]]]]]]] ~14.54x

list

• [] ~4.97x
• list[int] ~9.58x (20 items)
• list[int] ~16.23x (5000 items)
• list[str | list[str | list[str | list[str | list[int]]]]] ~9.19x

dict

• {} ~4.59x
• dict[str, int] ~5.95x (20 items)
• dict[str, int] ~3.94x (5000 items)
• dict[str, dict[str, ...]] ~6.19x

set

• set() ~15.87x
• set[int] ~6.55x (20 items)
• set[int] ~3.69x (5000 items)
• set[frozenset[frozenset[int]]] ~13.72x

other

• int/str/bytes ~4.86x
• dataclasses ~2.74x

Expect much higher speedups on Python 3.13 and lower.