1
0
mirror of https://github.com/bec-project/bec_widgets.git synced 2026-04-23 16:50:44 +02:00
Files
bec_widgets/.github/scripts/compare_benchmarks.py
T
2026-04-21 13:58:38 +02:00

452 lines
14 KiB
Python

#!/usr/bin/env python3
"""Compare benchmark JSON files and write a GitHub Actions summary.
The script supports JSON emitted by hyperfine, JSON emitted by pytest-benchmark,
and a compact mapping format generated by ``aggregate_benchmarks.py``. Timing
formats prefer median values and fall back to mean values when median values are
not present.
"""
from __future__ import annotations
import argparse
import json
import math
from dataclasses import dataclass
from pathlib import Path
from typing import Any
@dataclass(frozen=True)
class Benchmark:
"""Normalized benchmark result.
Attributes:
name (str): Stable benchmark name used to match baseline and current results.
value (float): Numeric benchmark value used for comparison.
unit (str): Display unit for the value, for example ``"s"``.
metric (str): Source metric name, for example ``"median"`` or ``"mean"``.
"""
name: str
value: float
unit: str
metric: str = "value"
@dataclass(frozen=True)
class Comparison:
"""Comparison between one baseline benchmark and one current benchmark.
Attributes:
name (str): Benchmark name.
baseline (float): Baseline benchmark value.
current (float): Current benchmark value.
delta_percent (float): Percent change from baseline to current.
unit (str): Display unit for both values.
metric (str): Current result metric used for comparison.
regressed (bool): Whether the change exceeds the configured threshold in
the worse direction.
improved (bool): Whether the change exceeds the configured threshold in
the better direction.
"""
name: str
baseline: float
current: float
delta_percent: float
unit: str
metric: str
regressed: bool
improved: bool
def _read_json(path: Path) -> Any:
"""Read JSON data from a file.
Args:
path (Path): Path to the JSON file.
Returns:
Any: Parsed JSON value.
"""
with path.open("r", encoding="utf-8") as stream:
return json.load(stream)
def _as_float(value: Any) -> float | None:
"""Convert a value to a finite float.
Args:
value (Any): Value to convert.
Returns:
float | None: Converted finite float, or ``None`` if conversion fails.
"""
try:
result = float(value)
except (TypeError, ValueError):
return None
if math.isfinite(result):
return result
return None
def _extract_hyperfine(data: dict[str, Any]) -> dict[str, Benchmark]:
"""Extract normalized benchmarks from hyperfine JSON.
Args:
data (dict[str, Any]): Parsed hyperfine JSON object.
Returns:
dict[str, Benchmark]: Benchmarks keyed by command name.
"""
benchmarks: dict[str, Benchmark] = {}
for result in data.get("results", []):
if not isinstance(result, dict):
continue
name = str(result.get("command") or result.get("name") or "").strip()
metric = "median"
value = _as_float(result.get(metric))
if value is None:
metric = "mean"
value = _as_float(result.get(metric))
if name and value is not None:
benchmarks[name] = Benchmark(name=name, value=value, unit="s", metric=metric)
return benchmarks
def _extract_pytest_benchmark(data: dict[str, Any]) -> dict[str, Benchmark]:
"""Extract normalized benchmarks from pytest-benchmark JSON.
Args:
data (dict[str, Any]): Parsed pytest-benchmark JSON object.
Returns:
dict[str, Benchmark]: Benchmarks keyed by full benchmark name.
"""
benchmarks: dict[str, Benchmark] = {}
for benchmark in data.get("benchmarks", []):
if not isinstance(benchmark, dict):
continue
name = str(benchmark.get("fullname") or benchmark.get("name") or "").strip()
stats = benchmark.get("stats", {})
value = None
metric = "median"
if isinstance(stats, dict):
value = _as_float(stats.get(metric))
if value is None:
metric = "mean"
value = _as_float(stats.get(metric))
if name and value is not None:
benchmarks[name] = Benchmark(name=name, value=value, unit="s", metric=metric)
return benchmarks
def _extract_simple_mapping(data: dict[str, Any]) -> dict[str, Benchmark]:
"""Extract normalized benchmarks from a compact mapping JSON object.
Args:
data (dict[str, Any]): Parsed mapping where each benchmark is either a
raw number or an object containing ``value``, ``unit``, and ``metric``.
Returns:
dict[str, Benchmark]: Benchmarks keyed by mapping key.
"""
benchmarks: dict[str, Benchmark] = {}
for name, raw_value in data.items():
if name in {"version", "context", "commit", "timestamp"}:
continue
value = _as_float(raw_value)
unit = ""
metric = "value"
if value is None and isinstance(raw_value, dict):
value = _as_float(raw_value.get("value"))
unit = str(raw_value.get("unit") or "")
metric = str(raw_value.get("metric") or "value")
if value is not None:
benchmarks[str(name)] = Benchmark(name=str(name), value=value, unit=unit, metric=metric)
return benchmarks
def extract_benchmarks(path: Path) -> dict[str, Benchmark]:
"""Extract normalized benchmarks from a supported JSON file.
Args:
path (Path): Path to a hyperfine, pytest-benchmark, or compact mapping
JSON file.
Returns:
dict[str, Benchmark]: Normalized benchmarks keyed by name.
Raises:
ValueError: If the JSON root is not an object or no supported benchmark
entries can be extracted.
"""
data = _read_json(path)
if not isinstance(data, dict):
raise ValueError(f"{path} must contain a JSON object")
extractors = (_extract_hyperfine, _extract_pytest_benchmark, _extract_simple_mapping)
for extractor in extractors:
benchmarks = extractor(data)
if benchmarks:
return benchmarks
raise ValueError(f"No supported benchmark entries found in {path}")
def compare_benchmarks(
baseline: dict[str, Benchmark],
current: dict[str, Benchmark],
threshold_percent: float,
higher_is_better: bool,
) -> tuple[list[Comparison], list[str], list[str]]:
"""Compare baseline benchmarks with current benchmarks.
Args:
baseline (dict[str, Benchmark]): Baseline benchmarks keyed by name.
current (dict[str, Benchmark]): Current benchmarks keyed by name.
threshold_percent (float): Regression threshold in percent.
higher_is_better (bool): If ``True``, lower current values are treated as
regressions. If ``False``, higher current values are treated as
regressions.
Returns:
tuple[list[Comparison], list[str], list[str]]: Comparisons for common
benchmark names, names missing from current results, and names newly
present in current results.
"""
comparisons: list[Comparison] = []
missing_in_current: list[str] = []
new_in_current: list[str] = []
for name, baseline_benchmark in sorted(baseline.items()):
current_benchmark = current.get(name)
if current_benchmark is None:
missing_in_current.append(name)
continue
if baseline_benchmark.value == 0:
delta_percent = 0.0
else:
delta_percent = (
(current_benchmark.value - baseline_benchmark.value)
/ abs(baseline_benchmark.value)
* 100
)
if higher_is_better:
regressed = delta_percent <= -threshold_percent
improved = delta_percent >= threshold_percent
else:
regressed = delta_percent >= threshold_percent
improved = delta_percent <= -threshold_percent
comparisons.append(
Comparison(
name=name,
baseline=baseline_benchmark.value,
current=current_benchmark.value,
delta_percent=delta_percent,
unit=current_benchmark.unit or baseline_benchmark.unit,
metric=current_benchmark.metric,
regressed=regressed,
improved=improved,
)
)
for name in sorted(set(current) - set(baseline)):
new_in_current.append(name)
return comparisons, missing_in_current, new_in_current
def _format_value(value: float, unit: str) -> str:
"""Format a benchmark value for Markdown output.
Args:
value (float): Numeric benchmark value.
unit (str): Display unit.
Returns:
str: Formatted value with optional unit suffix.
"""
suffix = f" {unit}" if unit else ""
return f"{value:.6g}{suffix}"
def _format_status(comparison: Comparison) -> str:
"""Format a comparison status for Markdown output."""
if comparison.regressed:
return ":red_circle: regressed"
if comparison.improved:
return ":green_circle: improved"
return "ok"
def write_summary(
path: Path,
comparisons: list[Comparison],
missing_in_current: list[str],
new_in_current: list[str],
threshold_percent: float,
higher_is_better: bool,
) -> None:
"""Write a Markdown benchmark comparison summary.
Args:
path (Path): Path where the summary should be written.
comparisons (list[Comparison]): Comparison rows for matching benchmarks.
missing_in_current (list[str]): Baseline benchmark names missing from the
current result.
new_in_current (list[str]): Current benchmark names not present in the
baseline result.
threshold_percent (float): Regression threshold in percent.
higher_is_better (bool): Whether higher benchmark values are considered
better.
"""
regressions = [comparison for comparison in comparisons if comparison.regressed]
improvements = [comparison for comparison in comparisons if comparison.improved]
direction = "higher is better" if higher_is_better else "lower is better"
sorted_comparisons = sorted(comparisons, key=lambda comparison: comparison.name)
lines = [
"<!-- bw-benchmark-comment -->",
"## Benchmark comparison",
"",
f"Threshold: {threshold_percent:g}% ({direction}).",
f"Result: {len(regressions)} regression(s), {len(improvements)} improvement(s) beyond threshold.",
]
lines.append("")
if regressions:
lines.extend(
[
f"{len(regressions)} benchmark(s) regressed beyond the configured threshold.",
"",
"| Benchmark | Baseline | Current | Change |",
"| --- | ---: | ---: | ---: |",
]
)
for comparison in regressions:
lines.append(
"| "
f"{comparison.name} | "
f"{_format_value(comparison.baseline, comparison.unit)} | "
f"{_format_value(comparison.current, comparison.unit)} | "
f"{comparison.delta_percent:+.2f}% |"
)
else:
lines.append("No benchmark regression exceeded the configured threshold.")
lines.append("")
if improvements:
lines.extend(
[
f"{len(improvements)} benchmark(s) improved beyond the configured threshold.",
"",
"| Benchmark | Baseline | Current | Change |",
"| --- | ---: | ---: | ---: |",
]
)
for comparison in improvements:
lines.append(
"| "
f"{comparison.name} | "
f"{_format_value(comparison.baseline, comparison.unit)} | "
f"{_format_value(comparison.current, comparison.unit)} | "
f"{comparison.delta_percent:+.2f}% |"
)
else:
lines.append("No benchmark improvement exceeded the configured threshold.")
if sorted_comparisons:
lines.extend(
[
"",
"<details>",
"<summary>All benchmark results</summary>",
"",
"| Benchmark | Baseline | Current | Change | Status |",
"| --- | ---: | ---: | ---: | --- |",
]
)
for comparison in sorted_comparisons:
lines.append(
"| "
f"{comparison.name} | "
f"{_format_value(comparison.baseline, comparison.unit)} | "
f"{_format_value(comparison.current, comparison.unit)} | "
f"{comparison.delta_percent:+.2f}% | "
f"{_format_status(comparison)} |"
)
lines.extend(["", "</details>"])
if missing_in_current:
lines.extend(["", "Missing benchmarks in the current run:"])
lines.extend(f"- `{name}`" for name in missing_in_current)
if new_in_current:
lines.extend(["", "New benchmarks in the current run:"])
lines.extend(f"- `{name}`" for name in new_in_current)
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text("\n".join(lines) + "\n", encoding="utf-8")
def main() -> int:
"""Run the benchmark comparison command line interface.
Returns:
int: ``1`` when a regression exceeds the threshold, otherwise ``0``.
"""
parser = argparse.ArgumentParser()
parser.add_argument("--baseline", required=True, type=Path)
parser.add_argument("--current", required=True, type=Path)
parser.add_argument("--summary", required=True, type=Path)
parser.add_argument("--threshold-percent", required=True, type=float)
parser.add_argument("--higher-is-better", action="store_true")
args = parser.parse_args()
baseline = extract_benchmarks(args.baseline)
current = extract_benchmarks(args.current)
comparisons, missing_in_current, new_in_current = compare_benchmarks(
baseline=baseline,
current=current,
threshold_percent=args.threshold_percent,
higher_is_better=args.higher_is_better,
)
write_summary(
path=args.summary,
comparisons=comparisons,
missing_in_current=missing_in_current,
new_in_current=new_in_current,
threshold_percent=args.threshold_percent,
higher_is_better=args.higher_is_better,
)
return 1 if any(comparison.regressed for comparison in comparisons) else 0
if __name__ == "__main__":
raise SystemExit(main())