mirror of
https://github.com/catchorg/Catch2.git
synced 2024-11-22 13:26:10 +01:00
57 lines
1.6 KiB
Python
57 lines
1.6 KiB
Python
#!/usr/bin/env python3
|
|
|
|
import subprocess, os, sys
|
|
import xml.etree.ElementTree as ET
|
|
from collections import defaultdict
|
|
from statistics import median, stdev
|
|
from datetime import datetime
|
|
|
|
def get_commit_hash():
|
|
res = subprocess.run('git rev-parse HEAD'.split(), check=True, stdout=subprocess.PIPE, universal_newlines=True)
|
|
return res.stdout.strip()
|
|
|
|
if len(sys.argv) < 2:
|
|
print('Usage: {} benchmark-binary'.format(sys.argv[0]))
|
|
exit(1)
|
|
|
|
|
|
num_runs = 10
|
|
data = defaultdict(list)
|
|
|
|
|
|
def parse_file(file):
|
|
|
|
def recursive_search(node):
|
|
if node.tag == 'TestCase':
|
|
results = node.find('OverallResult')
|
|
time = results.get('durationInSeconds')
|
|
data[node.get('name')].append(float(time))
|
|
elif node.tag in ('Group', 'Catch'):
|
|
for child in node:
|
|
recursive_search(child)
|
|
|
|
tree = ET.parse(file)
|
|
recursive_search(tree.getroot())
|
|
|
|
def run_benchmarks(binary):
|
|
call = [binary] + '-d yes -r xml -o'.split()
|
|
for i in range(num_runs):
|
|
file = 'temp{}.xml'.format(i)
|
|
print('Run number {}'.format(i))
|
|
subprocess.run(call + [file])
|
|
parse_file(file)
|
|
# Remove file right after parsing, because benchmark output can be big
|
|
os.remove(file)
|
|
|
|
|
|
# Run benchmarks
|
|
run_benchmarks(sys.argv[1])
|
|
|
|
result_file = '{:%Y-%m-%dT%H-%M-%S}-{}.result'.format(datetime.now(), get_commit_hash())
|
|
|
|
|
|
print('Writing results to {}'.format(result_file))
|
|
with open(result_file, 'w') as file:
|
|
for k in sorted(data):
|
|
file.write('{}: median: {} (s), stddev: {} (s)\n'.format(k, median(data[k]), stdev(data[k])))
|