diff options
Diffstat (limited to 'Tools/importbench')
| -rw-r--r-- | Tools/importbench/README | 6 | ||||
| -rw-r--r-- | Tools/importbench/importbench.py | 252 |
2 files changed, 258 insertions, 0 deletions
diff --git a/Tools/importbench/README b/Tools/importbench/README new file mode 100644 index 0000000..81a5544 --- /dev/null +++ b/Tools/importbench/README @@ -0,0 +1,6 @@ +Importbench is a set of micro-benchmarks for various import scenarios. + +It should not be used as an overall benchmark of import performance, but rather +an easy way to measure impact of possible code changes. For a real-world +benchmark of import, use the normal_startup benchmark from +hg.python.org/benchmarks. diff --git a/Tools/importbench/importbench.py b/Tools/importbench/importbench.py new file mode 100644 index 0000000..714c0e4 --- /dev/null +++ b/Tools/importbench/importbench.py @@ -0,0 +1,252 @@ +"""Benchmark some basic import use-cases. + +The assumption is made that this benchmark is run in a fresh interpreter and +thus has no external changes made to import-related attributes in sys. + +""" +from test.test_importlib import util +from test.test_importlib.source import util as source_util +import decimal +import imp +import importlib +import importlib.machinery +import json +import os +import py_compile +import sys +import tabnanny +import timeit + + +def bench(name, cleanup=lambda: None, *, seconds=1, repeat=3): + """Bench the given statement as many times as necessary until total + executions take one second.""" + stmt = "__import__({!r})".format(name) + timer = timeit.Timer(stmt) + for x in range(repeat): + total_time = 0 + count = 0 + while total_time < seconds: + try: + total_time += timer.timeit(1) + finally: + cleanup() + count += 1 + else: + # One execution too far + if total_time > seconds: + count -= 1 + yield count // seconds + +def from_cache(seconds, repeat): + """sys.modules""" + name = '<benchmark import>' + module = imp.new_module(name) + module.__file__ = '<test>' + module.__package__ = '' + with util.uncache(name): + sys.modules[name] = module + for result in bench(name, repeat=repeat, seconds=seconds): + yield result + + +def builtin_mod(seconds, repeat): + """Built-in module""" + name = 'errno' + if name in sys.modules: + del sys.modules[name] + # Relying on built-in importer being implicit. + for result in bench(name, lambda: sys.modules.pop(name), repeat=repeat, + seconds=seconds): + yield result + + +def source_wo_bytecode(seconds, repeat): + """Source w/o bytecode: small""" + sys.dont_write_bytecode = True + try: + name = '__importlib_test_benchmark__' + # Clears out sys.modules and puts an entry at the front of sys.path. + with source_util.create_modules(name) as mapping: + assert not os.path.exists(imp.cache_from_source(mapping[name])) + sys.meta_path.append(importlib.machinery.PathFinder) + loader = (importlib.machinery.SourceFileLoader, + importlib.machinery.SOURCE_SUFFIXES, True) + sys.path_hooks.append(importlib.machinery.FileFinder.path_hook(loader)) + for result in bench(name, lambda: sys.modules.pop(name), repeat=repeat, + seconds=seconds): + yield result + finally: + sys.dont_write_bytecode = False + + +def _wo_bytecode(module): + name = module.__name__ + def benchmark_wo_bytecode(seconds, repeat): + """Source w/o bytecode: {}""" + bytecode_path = imp.cache_from_source(module.__file__) + if os.path.exists(bytecode_path): + os.unlink(bytecode_path) + sys.dont_write_bytecode = True + try: + for result in bench(name, lambda: sys.modules.pop(name), + repeat=repeat, seconds=seconds): + yield result + finally: + sys.dont_write_bytecode = False + + benchmark_wo_bytecode.__doc__ = benchmark_wo_bytecode.__doc__.format(name) + return benchmark_wo_bytecode + +tabnanny_wo_bytecode = _wo_bytecode(tabnanny) +decimal_wo_bytecode = _wo_bytecode(decimal) + + +def source_writing_bytecode(seconds, repeat): + """Source writing bytecode: small""" + assert not sys.dont_write_bytecode + name = '__importlib_test_benchmark__' + with source_util.create_modules(name) as mapping: + sys.meta_path.append(importlib.machinery.PathFinder) + loader = (importlib.machinery.SourceFileLoader, + importlib.machinery.SOURCE_SUFFIXES, True) + sys.path_hooks.append(importlib.machinery.FileFinder.path_hook(loader)) + def cleanup(): + sys.modules.pop(name) + os.unlink(imp.cache_from_source(mapping[name])) + for result in bench(name, cleanup, repeat=repeat, seconds=seconds): + assert not os.path.exists(imp.cache_from_source(mapping[name])) + yield result + + +def _writing_bytecode(module): + name = module.__name__ + def writing_bytecode_benchmark(seconds, repeat): + """Source writing bytecode: {}""" + assert not sys.dont_write_bytecode + def cleanup(): + sys.modules.pop(name) + os.unlink(imp.cache_from_source(module.__file__)) + for result in bench(name, cleanup, repeat=repeat, seconds=seconds): + yield result + + writing_bytecode_benchmark.__doc__ = ( + writing_bytecode_benchmark.__doc__.format(name)) + return writing_bytecode_benchmark + +tabnanny_writing_bytecode = _writing_bytecode(tabnanny) +decimal_writing_bytecode = _writing_bytecode(decimal) + + +def source_using_bytecode(seconds, repeat): + """Source w/ bytecode: small""" + name = '__importlib_test_benchmark__' + with source_util.create_modules(name) as mapping: + sys.meta_path.append(importlib.machinery.PathFinder) + loader = (importlib.machinery.SourceFileLoader, + importlib.machinery.SOURCE_SUFFIXES, True) + sys.path_hooks.append(importlib.machinery.FileFinder.path_hook(loader)) + py_compile.compile(mapping[name]) + assert os.path.exists(imp.cache_from_source(mapping[name])) + for result in bench(name, lambda: sys.modules.pop(name), repeat=repeat, + seconds=seconds): + yield result + + +def _using_bytecode(module): + name = module.__name__ + def using_bytecode_benchmark(seconds, repeat): + """Source w/ bytecode: {}""" + py_compile.compile(module.__file__) + for result in bench(name, lambda: sys.modules.pop(name), repeat=repeat, + seconds=seconds): + yield result + + using_bytecode_benchmark.__doc__ = ( + using_bytecode_benchmark.__doc__.format(name)) + return using_bytecode_benchmark + +tabnanny_using_bytecode = _using_bytecode(tabnanny) +decimal_using_bytecode = _using_bytecode(decimal) + + +def main(import_, options): + if options.source_file: + with options.source_file: + prev_results = json.load(options.source_file) + else: + prev_results = {} + __builtins__.__import__ = import_ + benchmarks = (from_cache, builtin_mod, + source_writing_bytecode, + source_wo_bytecode, source_using_bytecode, + tabnanny_writing_bytecode, + tabnanny_wo_bytecode, tabnanny_using_bytecode, + decimal_writing_bytecode, + decimal_wo_bytecode, decimal_using_bytecode, + ) + if options.benchmark: + for b in benchmarks: + if b.__doc__ == options.benchmark: + benchmarks = [b] + break + else: + print('Unknown benchmark: {!r}'.format(options.benchmark, + file=sys.stderr)) + sys.exit(1) + seconds = 1 + seconds_plural = 's' if seconds > 1 else '' + repeat = 3 + header = ('Measuring imports/second over {} second{}, best out of {}\n' + 'Entire benchmark run should take about {} seconds\n' + 'Using {!r} as __import__\n') + print(header.format(seconds, seconds_plural, repeat, + len(benchmarks) * seconds * repeat, __import__)) + new_results = {} + for benchmark in benchmarks: + print(benchmark.__doc__, "[", end=' ') + sys.stdout.flush() + results = [] + for result in benchmark(seconds=seconds, repeat=repeat): + results.append(result) + print(result, end=' ') + sys.stdout.flush() + assert not sys.dont_write_bytecode + print("]", "best is", format(max(results), ',d')) + new_results[benchmark.__doc__] = results + if prev_results: + print('\n\nComparing new vs. old\n') + for benchmark in benchmarks: + benchmark_name = benchmark.__doc__ + old_result = max(prev_results[benchmark_name]) + new_result = max(new_results[benchmark_name]) + result = '{:,d} vs. {:,d} ({:%})'.format(new_result, + old_result, + new_result/old_result) + print(benchmark_name, ':', result) + if options.dest_file: + with options.dest_file: + json.dump(new_results, options.dest_file, indent=2) + + +if __name__ == '__main__': + import argparse + + parser = argparse.ArgumentParser() + parser.add_argument('-b', '--builtin', dest='builtin', action='store_true', + default=False, help="use the built-in __import__") + parser.add_argument('-r', '--read', dest='source_file', + type=argparse.FileType('r'), + help='file to read benchmark data from to compare ' + 'against') + parser.add_argument('-w', '--write', dest='dest_file', + type=argparse.FileType('w'), + help='file to write benchmark data to') + parser.add_argument('--benchmark', dest='benchmark', + help='specific benchmark to run') + options = parser.parse_args() + import_ = __import__ + if not options.builtin: + import_ = importlib.__import__ + + main(import_, options) |