| 1 |
"""Benchmark some basic import use-cases. |
| 2 |
|
| 3 |
The assumption is made that this benchmark is run in a fresh interpreter and |
| 4 |
thus has no external changes made to import-related attributes in sys. |
| 5 |
|
| 6 |
""" |
| 7 |
from test.test_importlib import util |
| 8 |
import decimal |
| 9 |
from importlib.util import cache_from_source |
| 10 |
import importlib |
| 11 |
import importlib.machinery |
| 12 |
import json |
| 13 |
import os |
| 14 |
import py_compile |
| 15 |
import sys |
| 16 |
import tabnanny |
| 17 |
import timeit |
| 18 |
import types |
| 19 |
|
| 20 |
|
| 21 |
def bench(name, cleanup=lambda: None, *, seconds=1, repeat=3): |
| 22 |
"""Bench the given statement as many times as necessary until total |
| 23 |
executions take one second.""" |
| 24 |
stmt = "__import__({!r})".format(name) |
| 25 |
timer = timeit.Timer(stmt) |
| 26 |
for x in range(repeat): |
| 27 |
total_time = 0 |
| 28 |
count = 0 |
| 29 |
while total_time < seconds: |
| 30 |
try: |
| 31 |
total_time += timer.timeit(1) |
| 32 |
finally: |
| 33 |
cleanup() |
| 34 |
count += 1 |
| 35 |
else: |
| 36 |
# One execution too far |
| 37 |
if total_time > seconds: |
| 38 |
count -= 1 |
| 39 |
yield count // seconds |
| 40 |
|
| 41 |
def from_cache(seconds, repeat): |
| 42 |
"""sys.modules""" |
| 43 |
name = '<benchmark import>' |
| 44 |
module = types.ModuleType(name) |
| 45 |
module.__file__ = '<test>' |
| 46 |
module.__package__ = '' |
| 47 |
with util.uncache(name): |
| 48 |
sys.modules[name] = module |
| 49 |
yield from bench(name, repeat=repeat, seconds=seconds) |
| 50 |
|
| 51 |
|
| 52 |
def builtin_mod(seconds, repeat): |
| 53 |
"""Built-in module""" |
| 54 |
name = 'errno' |
| 55 |
if name in sys.modules: |
| 56 |
del sys.modules[name] |
| 57 |
# Relying on built-in importer being implicit. |
| 58 |
yield from bench(name, lambda: sys.modules.pop(name), repeat=repeat, |
| 59 |
seconds=seconds) |
| 60 |
|
| 61 |
|
| 62 |
def source_wo_bytecode(seconds, repeat): |
| 63 |
"""Source w/o bytecode: small""" |
| 64 |
sys.dont_write_bytecode = True |
| 65 |
try: |
| 66 |
name = '__importlib_test_benchmark__' |
| 67 |
# Clears out sys.modules and puts an entry at the front of sys.path. |
| 68 |
with util.create_modules(name) as mapping: |
| 69 |
assert not os.path.exists(cache_from_source(mapping[name])) |
| 70 |
sys.meta_path.append(importlib.machinery.PathFinder) |
| 71 |
loader = (importlib.machinery.SourceFileLoader, |
| 72 |
importlib.machinery.SOURCE_SUFFIXES) |
| 73 |
sys.path_hooks.append(importlib.machinery.FileFinder.path_hook(loader)) |
| 74 |
yield from bench(name, lambda: sys.modules.pop(name), repeat=repeat, |
| 75 |
seconds=seconds) |
| 76 |
finally: |
| 77 |
sys.dont_write_bytecode = False |
| 78 |
|
| 79 |
|
| 80 |
def _wo_bytecode(module): |
| 81 |
name = module.__name__ |
| 82 |
def benchmark_wo_bytecode(seconds, repeat): |
| 83 |
"""Source w/o bytecode: {}""" |
| 84 |
bytecode_path = cache_from_source(module.__file__) |
| 85 |
if os.path.exists(bytecode_path): |
| 86 |
os.unlink(bytecode_path) |
| 87 |
sys.dont_write_bytecode = True |
| 88 |
try: |
| 89 |
yield from bench(name, lambda: sys.modules.pop(name), |
| 90 |
repeat=repeat, seconds=seconds) |
| 91 |
finally: |
| 92 |
sys.dont_write_bytecode = False |
| 93 |
|
| 94 |
benchmark_wo_bytecode.__doc__ = benchmark_wo_bytecode.__doc__.format(name) |
| 95 |
return benchmark_wo_bytecode |
| 96 |
|
| 97 |
tabnanny_wo_bytecode = _wo_bytecode(tabnanny) |
| 98 |
decimal_wo_bytecode = _wo_bytecode(decimal) |
| 99 |
|
| 100 |
|
| 101 |
def source_writing_bytecode(seconds, repeat): |
| 102 |
"""Source writing bytecode: small""" |
| 103 |
assert not sys.dont_write_bytecode |
| 104 |
name = '__importlib_test_benchmark__' |
| 105 |
with util.create_modules(name) as mapping: |
| 106 |
sys.meta_path.append(importlib.machinery.PathFinder) |
| 107 |
loader = (importlib.machinery.SourceFileLoader, |
| 108 |
importlib.machinery.SOURCE_SUFFIXES) |
| 109 |
sys.path_hooks.append(importlib.machinery.FileFinder.path_hook(loader)) |
| 110 |
def cleanup(): |
| 111 |
sys.modules.pop(name) |
| 112 |
os.unlink(cache_from_source(mapping[name])) |
| 113 |
for result in bench(name, cleanup, repeat=repeat, seconds=seconds): |
| 114 |
assert not os.path.exists(cache_from_source(mapping[name])) |
| 115 |
yield result |
| 116 |
|
| 117 |
|
| 118 |
def _writing_bytecode(module): |
| 119 |
name = module.__name__ |
| 120 |
def writing_bytecode_benchmark(seconds, repeat): |
| 121 |
"""Source writing bytecode: {}""" |
| 122 |
assert not sys.dont_write_bytecode |
| 123 |
def cleanup(): |
| 124 |
sys.modules.pop(name) |
| 125 |
os.unlink(cache_from_source(module.__file__)) |
| 126 |
yield from bench(name, cleanup, repeat=repeat, seconds=seconds) |
| 127 |
|
| 128 |
writing_bytecode_benchmark.__doc__ = ( |
| 129 |
writing_bytecode_benchmark.__doc__.format(name)) |
| 130 |
return writing_bytecode_benchmark |
| 131 |
|
| 132 |
tabnanny_writing_bytecode = _writing_bytecode(tabnanny) |
| 133 |
decimal_writing_bytecode = _writing_bytecode(decimal) |
| 134 |
|
| 135 |
|
| 136 |
def source_using_bytecode(seconds, repeat): |
| 137 |
"""Source w/ bytecode: small""" |
| 138 |
name = '__importlib_test_benchmark__' |
| 139 |
with util.create_modules(name) as mapping: |
| 140 |
sys.meta_path.append(importlib.machinery.PathFinder) |
| 141 |
loader = (importlib.machinery.SourceFileLoader, |
| 142 |
importlib.machinery.SOURCE_SUFFIXES) |
| 143 |
sys.path_hooks.append(importlib.machinery.FileFinder.path_hook(loader)) |
| 144 |
py_compile.compile(mapping[name]) |
| 145 |
assert os.path.exists(cache_from_source(mapping[name])) |
| 146 |
yield from bench(name, lambda: sys.modules.pop(name), repeat=repeat, |
| 147 |
seconds=seconds) |
| 148 |
|
| 149 |
|
| 150 |
def _using_bytecode(module): |
| 151 |
name = module.__name__ |
| 152 |
def using_bytecode_benchmark(seconds, repeat): |
| 153 |
"""Source w/ bytecode: {}""" |
| 154 |
py_compile.compile(module.__file__) |
| 155 |
yield from bench(name, lambda: sys.modules.pop(name), repeat=repeat, |
| 156 |
seconds=seconds) |
| 157 |
|
| 158 |
using_bytecode_benchmark.__doc__ = ( |
| 159 |
using_bytecode_benchmark.__doc__.format(name)) |
| 160 |
return using_bytecode_benchmark |
| 161 |
|
| 162 |
tabnanny_using_bytecode = _using_bytecode(tabnanny) |
| 163 |
decimal_using_bytecode = _using_bytecode(decimal) |
| 164 |
|
| 165 |
|
| 166 |
def main(import_, options): |
| 167 |
if options.source_file: |
| 168 |
with open(options.source_file, 'r', encoding='utf-8') as source_file: |
| 169 |
prev_results = json.load(source_file) |
| 170 |
else: |
| 171 |
prev_results = {} |
| 172 |
__builtins__.__import__ = import_ |
| 173 |
benchmarks = (from_cache, builtin_mod, |
| 174 |
source_writing_bytecode, |
| 175 |
source_wo_bytecode, source_using_bytecode, |
| 176 |
tabnanny_writing_bytecode, |
| 177 |
tabnanny_wo_bytecode, tabnanny_using_bytecode, |
| 178 |
decimal_writing_bytecode, |
| 179 |
decimal_wo_bytecode, decimal_using_bytecode, |
| 180 |
) |
| 181 |
if options.benchmark: |
| 182 |
for b in benchmarks: |
| 183 |
if b.__doc__ == options.benchmark: |
| 184 |
benchmarks = [b] |
| 185 |
break |
| 186 |
else: |
| 187 |
print('Unknown benchmark: {!r}'.format(options.benchmark), |
| 188 |
file=sys.stderr) |
| 189 |
sys.exit(1) |
| 190 |
seconds = 1 |
| 191 |
seconds_plural = 's' if seconds > 1 else '' |
| 192 |
repeat = 3 |
| 193 |
header = ('Measuring imports/second over {} second{}, best out of {}\n' |
| 194 |
'Entire benchmark run should take about {} seconds\n' |
| 195 |
'Using {!r} as __import__\n') |
| 196 |
print(header.format(seconds, seconds_plural, repeat, |
| 197 |
len(benchmarks) * seconds * repeat, __import__)) |
| 198 |
new_results = {} |
| 199 |
for benchmark in benchmarks: |
| 200 |
print(benchmark.__doc__, "[", end=' ') |
| 201 |
sys.stdout.flush() |
| 202 |
results = [] |
| 203 |
for result in benchmark(seconds=seconds, repeat=repeat): |
| 204 |
results.append(result) |
| 205 |
print(result, end=' ') |
| 206 |
sys.stdout.flush() |
| 207 |
assert not sys.dont_write_bytecode |
| 208 |
print("]", "best is", format(max(results), ',d')) |
| 209 |
new_results[benchmark.__doc__] = results |
| 210 |
if prev_results: |
| 211 |
print('\n\nComparing new vs. old\n') |
| 212 |
for benchmark in benchmarks: |
| 213 |
benchmark_name = benchmark.__doc__ |
| 214 |
old_result = max(prev_results[benchmark_name]) |
| 215 |
new_result = max(new_results[benchmark_name]) |
| 216 |
result = '{:,d} vs. {:,d} ({:%})'.format(new_result, |
| 217 |
old_result, |
| 218 |
new_result/old_result) |
| 219 |
print(benchmark_name, ':', result) |
| 220 |
if options.dest_file: |
| 221 |
with open(options.dest_file, 'w', encoding='utf-8') as dest_file: |
| 222 |
json.dump(new_results, dest_file, indent=2) |
| 223 |
|
| 224 |
|
| 225 |
if __name__ == '__main__': |
| 226 |
import argparse |
| 227 |
|
| 228 |
parser = argparse.ArgumentParser() |
| 229 |
parser.add_argument('-b', '--builtin', dest='builtin', action='store_true', |
| 230 |
default=False, help="use the built-in __import__") |
| 231 |
parser.add_argument('-r', '--read', dest='source_file', |
| 232 |
help='file to read benchmark data from to compare ' |
| 233 |
'against') |
| 234 |
parser.add_argument('-w', '--write', dest='dest_file', |
| 235 |
help='file to write benchmark data to') |
| 236 |
parser.add_argument('--benchmark', dest='benchmark', |
| 237 |
help='specific benchmark to run') |
| 238 |
options = parser.parse_args() |
| 239 |
import_ = __import__ |
| 240 |
if not options.builtin: |
| 241 |
import_ = importlib.__import__ |
| 242 |
|
| 243 |
main(import_, options) |