sergey-miryanov

import gc
import json
import os
import sys
import _gc_monitor as m
import time

if len(sys.argv) == 2:
    pid = int(sys.argv[-1])
else:

sergey-miryanov

import collections.abc
import gc
import random
import sys
import statistics
import time
import pyperf

# Adapted from https://gist.github.com/stedolan/4369a0fa27820e27d6e56bee5e412896
# See https://github.com/ocaml/ocaml/pull/10195

sergey-miryanov

2026-01-28 results

I compare different versions:

• main (29f1e778faaa31479e4ef84e2fb37a7a53984550)
• main+revert(+ https://github.com/sergey-miryanov/cpython/commit/efd2a4c0447496396445a5b5acc63a0482f1790d)
• main+gc-fix(+ https://github.com/sergey-miryanov/cpython/commits/gc-fix?author=sergey-miryanov)

I did comparison of pyperformance and gc-bench. gc-bench is pyperf-ed version of yours gc_big and gc_big_tree. Results are following:

sergey-miryanov

import httpx
import psutil
import asyncio
import gc
import os


def get_rss():
    p = psutil.Process()
    mem_info = p.memory_info()

sergey-miryanov

#!/usr/bin/env ./python
import contextlib
import os
import sys
import shutil
import subprocess
import time
import venv
from pathlib import Path
import tempfile

sergey-miryanov

import gc
import time
import statistics

# Adapted from https://gist.github.com/stedolan/4369a0fa27820e27d6e56bee5e412896
# See https://github.com/ocaml/ocaml/pull/10195
#
# In Python (free threading), this uses closer to 1.9 GiB of memory isntead of 800 MiB

# perf stat -e L1-dcache-loads,L1-dcache-load-misses,LLC-loads,LLC-load-misses ./python gc_big_800m.py

sergey-miryanov

import sys
import gc
import random
import time
import collections.abc
import statistics

# perf stat -e L1-dcache-loads,L1-dcache-load-misses,LLC-loads,LLC-load-misses ./python gc_big_tree.py
# perf stat -e l3_request_g1.caching_l3_cache_accesses,l3_comb_clstr_state.request_miss ./python gc_big_tree.py

sergey-miryanov

asyncio

async generator's finalization is broken

import asyncio
work_done = False

async def cursor():
 try:

sergey-miryanov

+--------------------------+----------+------------------------+
| Benchmark                | win-main | main                   |
+==========================+==========+========================+
| 2to3                     | 342 ms   | 171 ms: 2.01x faster   |
+--------------------------+----------+------------------------+
| async_generators         | 436 ms   | 269 ms: 1.62x faster   |
+--------------------------+----------+------------------------+
| asyncio_tcp              | 659 ms   | 301 ms: 2.19x faster   |
+--------------------------+----------+------------------------+
| asyncio_tcp_ssl          | 1.82 sec | 801 ms: 2.27x faster   |

sergey-miryanov

+--------------------------+----------+------------------------+
| Benchmark                | win-main | main                   |
+==========================+==========+========================+
| 2to3                     | 342 ms   | 174 ms: 1.97x faster   |
+--------------------------+----------+------------------------+
| async_generators         | 436 ms   | 262 ms: 1.66x faster   |
+--------------------------+----------+------------------------+
| asyncio_tcp              | 659 ms   | 304 ms: 2.17x faster   |
+--------------------------+----------+------------------------+
| asyncio_tcp_ssl          | 1.82 sec | 805 ms: 2.26x faster   |