gh-150942: Speed up frame local item collection

[omkar-334]

Use _PyList_AppendTakeRef while collecting frame-local (name, value) pairs
instead of PyList_Append followed by Py_DECREF, removing an incref/decref
pair per returned item pair.

benchmarks

Benchmark	main	this PR	speedup
frame_locals_items_100locals	9.09 ms	8.61 ms	1.06x
frame_locals_items_1000locals	17.19 ms	16.59 ms	1.04x
frame_locals_items_5000locals	29.92 ms	28.94 ms	1.03x
geomean			1.04x

Benchmark script

"""Micro-benchmark for frame.f_locals.items() result-list building."""

from __future__ import annotations

import argparse
import json
import statistics
import sys
import time
from pathlib import Path


def make_frame(local_count: int):
    assignments = "\n".join(f"    v{i} = {i}" for i in range(local_count))
    src = (
        "def make_frame():\n"
        f"{assignments}\n"
        "    return sys._getframe()\n"
    )
    ns = {"sys": sys}
    exec(src, ns)
    return ns["make_frame"]()


def time_items(frame, loops: int) -> float:
    t0 = time.perf_counter_ns()
    for _ in range(loops):
        frame.f_locals.items()
    return (time.perf_counter_ns() - t0) / 1e9


def main() -> int:
    ap = argparse.ArgumentParser()
    ap.add_argument("--locals", type=int, default=1_000)
    ap.add_argument("--loops", type=int, default=2_000)
    ap.add_argument("--trials", type=int, default=15)
    ap.add_argument("--warmup", type=int, default=3)
    ap.add_argument("--label", default="run")
    ap.add_argument("--outdir", default="results/frame-locals-items")
    args = ap.parse_args()

    frame = make_frame(args.locals)
    item_count = len(frame.f_locals.items())

    for _ in range(args.warmup):
        time_items(frame, args.loops)

    trials = [time_items(frame, args.loops) for _ in range(args.trials)]

    outdir = Path(args.outdir)
    outdir.mkdir(parents=True, exist_ok=True)
    summary = {
        "label": args.label,
        "python": sys.executable,
        "locals": args.locals,
        "item_count": item_count,
        "loops": args.loops,
        "trials": args.trials,
        "warmup": args.warmup,
        "items_built": item_count * args.loops,
        "seconds": {
            "min": min(trials),
            "median": statistics.median(trials),
            "mean": statistics.fmean(trials),
            "stdev": statistics.pstdev(trials),
            "all": trials,
        },
        "items_per_sec_at_min": (item_count * args.loops) / min(trials),
    }
    out = outdir / f"bench_{args.label}.json"
    out.write_text(json.dumps(summary, indent=2))

    s = summary["seconds"]
    print(
        f"{args.label}: min={s['min']*1000:.2f}ms median={s['median']*1000:.2f}ms "
        f"mean={s['mean']*1000:.2f}ms stdev={s['stdev']*1000:.2f}ms "
        f"items/s@min={summary['items_per_sec_at_min']:.2e}"
    )
    print(f"wrote {out}")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())

• Issue: Improve performance by using reference stealing methods #150942

[sergey-miryanov]

LGTM

[corona10]

lgtm with the following perspectives.
: The PR makes code simpler with performance improvement.