| #!/usr/bin/env python3 |
| # Copyright 2019 The Chromium Authors |
| # Use of this source code is governed by a BSD-style license that can be |
| # found in the LICENSE file. |
| |
| """A tool to extract size information for chrome. |
| |
| For a list of command-line options, call this script with '--help'. |
| |
| This script uses Python 2 due to dependence on tracing.value. |
| """ |
| |
| from __future__ import print_function |
| |
| import argparse |
| import errno |
| import glob |
| import json |
| import platform |
| import os |
| import re |
| import stat |
| import subprocess |
| import sys |
| import tempfile |
| |
| |
| SRC_DIR = os.path.abspath(os.path.join(os.path.dirname(__file__), '..', '..')) |
| |
| # Add Catapult to the path so we can import the chartjson-histogramset |
| # conversion. |
| sys.path.append(os.path.join(SRC_DIR, 'third_party', 'catapult', 'tracing')) |
| from tracing.value import convert_chart_json |
| |
| sys.path.insert(0, os.path.join(SRC_DIR, 'build', 'util')) |
| from lib.results import result_sink |
| from lib.results import result_types |
| |
| |
| class ResultsCollector: |
| |
| def __init__(self): |
| self.results = {} |
| |
| def add_result(self, name, identifier, value, units): |
| assert name not in self.results |
| self.results[name] = { |
| 'identifier': identifier, |
| 'value': int(value), |
| 'units': units |
| } |
| |
| # Legacy printing, previously used for parsing the text logs. |
| print('RESULT %s: %s= %s %s' % (name, identifier, value, units)) |
| |
| |
| def get_size(filename): |
| return os.stat(filename)[stat.ST_SIZE] |
| |
| |
| def get_linux_stripped_size(filename): |
| EU_STRIP_NAME = 'eu-strip' |
| # Assumes |filename| is in out/Release |
| # build/linux/bin/eu-strip' |
| src_dir = os.path.dirname(os.path.dirname(os.path.dirname(filename))) |
| eu_strip_path = os.path.join(src_dir, 'build', 'linux', 'bin', EU_STRIP_NAME) |
| if (platform.architecture()[0] == '64bit' |
| or not os.path.exists(eu_strip_path)): |
| eu_strip_path = EU_STRIP_NAME |
| |
| with tempfile.NamedTemporaryFile() as stripped_file: |
| strip_cmd = [eu_strip_path, '-o', stripped_file.name, filename] |
| result = 0 |
| result, _ = run_process(result, strip_cmd) |
| if result != 0: |
| return (result, 0) |
| return (result, get_size(stripped_file.name)) |
| |
| |
| def run_process(result, command): |
| p = subprocess.Popen(command, stdout=subprocess.PIPE) |
| stdout = p.communicate()[0].decode() |
| if p.returncode != 0: |
| print('ERROR from command "%s": %d' % (' '.join(command), p.returncode)) |
| if result == 0: |
| result = p.returncode |
| return result, stdout |
| |
| |
| def main_mac(output_directory, results_collector, size_path): |
| """Print appropriate size information about built Mac targets. |
| |
| Returns the first non-zero exit status of any command it executes, |
| or zero on success. |
| """ |
| result = 0 |
| # Work with either build type. |
| base_names = ('Chromium', 'Google Chrome') |
| for base_name in base_names: |
| app_bundle = base_name + '.app' |
| framework_name = base_name + ' Framework' |
| framework_bundle = framework_name + '.framework' |
| framework_dsym_bundle = framework_name + '.dSYM' |
| |
| chromium_app_dir = os.path.join(output_directory, app_bundle) |
| chromium_executable = os.path.join(chromium_app_dir, 'Contents', 'MacOS', |
| base_name) |
| |
| chromium_framework_dir = os.path.join(output_directory, framework_bundle) |
| chromium_framework_executable = os.path.join(chromium_framework_dir, |
| framework_name) |
| |
| chromium_framework_dsym_dir = os.path.join(output_directory, |
| framework_dsym_bundle) |
| chromium_framework_dsym = os.path.join(chromium_framework_dsym_dir, |
| 'Contents', 'Resources', 'DWARF', |
| framework_name) |
| if os.path.exists(chromium_executable): |
| print_dict = { |
| # Remove spaces in the names so any downstream processing is less |
| # likely to choke. |
| 'app_name': re.sub(r'\s', '', base_name), |
| 'app_bundle': re.sub(r'\s', '', app_bundle), |
| 'framework_name': re.sub(r'\s', '', framework_name), |
| 'framework_bundle': re.sub(r'\s', '', framework_bundle), |
| 'app_size': get_size(chromium_executable), |
| 'framework_size': get_size(chromium_framework_executable), |
| 'framework_dsym_name': re.sub(r'\s', '', framework_name) + 'Dsym', |
| 'framework_dsym_size': get_size(chromium_framework_dsym), |
| } |
| |
| # Collect the segment info out of the App |
| result, stdout = run_process(result, [size_path, chromium_executable]) |
| print_dict['app_text'], print_dict['app_data'], print_dict['app_objc'] = \ |
| re.search(r'(\d+)\s+(\d+)\s+(\d+)', stdout).groups() |
| |
| # Collect the segment info out of the Framework |
| result, stdout = run_process(result, |
| [size_path, chromium_framework_executable]) |
| print_dict['framework_text'], print_dict['framework_data'], \ |
| print_dict['framework_objc'] = \ |
| re.search(r'(\d+)\s+(\d+)\s+(\d+)', stdout).groups() |
| |
| # Collect the whole size of the App bundle on disk (include the framework) |
| whole_size = 0 |
| for root_dir, _, filenames in os.walk(chromium_app_dir, |
| followlinks=False): |
| for filename in filenames: |
| full_path = os.path.join(root_dir, filename) |
| if not os.path.islink(full_path): |
| whole_size += get_size(full_path) |
| print_dict['app_bundle_size'] = whole_size |
| |
| results_collector.add_result(print_dict['app_name'], |
| print_dict['app_name'], |
| print_dict['app_size'], 'bytes') |
| results_collector.add_result('%s-__TEXT' % print_dict['app_name'], |
| '__TEXT', print_dict['app_text'], 'bytes') |
| results_collector.add_result('%s-__DATA' % print_dict['app_name'], |
| '__DATA', print_dict['app_data'], 'bytes') |
| results_collector.add_result('%s-__OBJC' % print_dict['app_name'], |
| '__OBJC', print_dict['app_objc'], 'bytes') |
| results_collector.add_result(print_dict['framework_name'], |
| print_dict['framework_name'], |
| print_dict['framework_size'], 'bytes') |
| results_collector.add_result('%s-__TEXT' % print_dict['framework_name'], |
| '__TEXT', print_dict['framework_text'], |
| 'bytes') |
| results_collector.add_result('%s-__DATA' % print_dict['framework_name'], |
| '__DATA', print_dict['framework_data'], |
| 'bytes') |
| results_collector.add_result('%s-__OBJC' % print_dict['framework_name'], |
| '__OBJC', print_dict['framework_objc'], |
| 'bytes') |
| results_collector.add_result(print_dict['app_bundle'], |
| print_dict['app_bundle'], |
| print_dict['app_bundle_size'], 'bytes') |
| results_collector.add_result(print_dict['framework_dsym_name'], |
| print_dict['framework_dsym_name'], |
| print_dict['framework_dsym_size'], 'bytes') |
| |
| # Found a match, don't check the other base_names. |
| return result |
| # If no base_names matched, fail script. |
| return 66 |
| |
| |
| def check_linux_binary(binary_name, output_directory): |
| """Collect appropriate size information about the built Linux binary given. |
| |
| Returns a tuple (result, sizes). result is the first non-zero exit |
| status of any command it executes, or zero on success. sizes is a list |
| of tuples (name, identifier, totals_identifier, value, units). |
| The printed line looks like: |
| name: identifier= value units |
| When this same data is used for totals across all the binaries, then |
| totals_identifier is the identifier to use, or '' to just use identifier. |
| """ |
| binary_file = os.path.join(output_directory, binary_name) |
| |
| if not os.path.exists(binary_file): |
| # Don't print anything for missing files. |
| return 0, [] |
| |
| result = 0 |
| sizes = [] |
| |
| sizes.append((binary_name, binary_name, 'size', get_size(binary_file), |
| 'bytes')) |
| |
| result, stripped_size = get_linux_stripped_size(binary_file) |
| sizes.append((binary_name + '-stripped', 'stripped', 'stripped', |
| stripped_size, 'bytes')) |
| |
| result, stdout = run_process(result, ['size', binary_file]) |
| text, data, bss = re.search(r'(\d+)\s+(\d+)\s+(\d+)', stdout).groups() |
| sizes += [ |
| (binary_name + '-text', 'text', '', text, 'bytes'), |
| (binary_name + '-data', 'data', '', data, 'bytes'), |
| (binary_name + '-bss', 'bss', '', bss, 'bytes'), |
| ] |
| |
| # Determine if the binary has the DT_TEXTREL marker. |
| result, stdout = run_process(result, ['readelf', '-Wd', binary_file]) |
| if re.search(r'\bTEXTREL\b', stdout) is None: |
| # Nope, so the count is zero. |
| count = 0 |
| else: |
| # There are some, so count them. |
| result, stdout = run_process(result, ['eu-findtextrel', binary_file]) |
| count = stdout.count('\n') |
| sizes.append((binary_name + '-textrel', 'textrel', '', count, 'relocs')) |
| |
| return result, sizes |
| |
| |
| def main_linux(output_directory, results_collector, size_path): |
| """Print appropriate size information about built Linux targets. |
| |
| Returns the first non-zero exit status of any command it executes, |
| or zero on success. |
| """ |
| assert size_path is None |
| binaries = [ |
| 'chrome', |
| 'nacl_helper', |
| 'nacl_helper_bootstrap', |
| 'libffmpegsumo.so', |
| 'libgcflashplayer.so', |
| 'libppGoogleNaClPluginChrome.so', |
| ] |
| |
| result = 0 |
| |
| totals = {} |
| |
| for binary in binaries: |
| this_result, this_sizes = check_linux_binary(binary, output_directory) |
| if result == 0: |
| result = this_result |
| for name, identifier, totals_id, value, units in this_sizes: |
| results_collector.add_result(name, identifier, value, units) |
| totals_id = totals_id or identifier, units |
| totals[totals_id] = totals.get(totals_id, 0) + int(value) |
| |
| files = [ |
| 'nacl_irt_x86_64.nexe', |
| 'resources.pak', |
| ] |
| |
| for filename in files: |
| path = os.path.join(output_directory, filename) |
| try: |
| size = get_size(path) |
| except OSError as e: |
| if e.errno == errno.ENOENT: |
| continue # Don't print anything for missing files. |
| raise |
| results_collector.add_result(filename, filename, size, 'bytes') |
| totals['size', 'bytes'] += size |
| |
| # TODO(mcgrathr): This should all be refactored so the mac and win flavors |
| # also deliver data structures rather than printing, and the logic for |
| # the printing and the summing totals is shared across all three flavors. |
| for (identifier, units), value in sorted(totals.items()): |
| results_collector.add_result('totals-%s' % identifier, identifier, value, |
| units) |
| |
| return result |
| |
| |
| def check_android_binaries(binaries, |
| output_directory, |
| results_collector, |
| binaries_to_print=None): |
| """Common method for printing size information for Android targets. |
| |
| Prints size information for each element of binaries in the output |
| directory. If binaries_to_print is specified, the name of each binary from |
| binaries is replaced with corresponding element of binaries_to_print |
| in output. Returns the first non-zero exit status of any command it |
| executes, or zero on success. |
| """ |
| result = 0 |
| if not binaries_to_print: |
| binaries_to_print = binaries |
| |
| for (binary, binary_to_print) in zip(binaries, binaries_to_print): |
| this_result, this_sizes = check_linux_binary(binary, output_directory) |
| if result == 0: |
| result = this_result |
| for name, identifier, _, value, units in this_sizes: |
| name = name.replace('/', '_').replace(binary, binary_to_print) |
| identifier = identifier.replace(binary, binary_to_print) |
| results_collector.add_result(name, identifier, value, units) |
| |
| return result |
| |
| |
| def main_android(output_directory, results_collector, size_path): |
| """Print appropriate size information about built Android targets. |
| |
| Returns the first non-zero exit status of any command it executes, |
| or zero on success. |
| """ |
| assert size_path is None |
| binaries = [ |
| 'chrome_public_apk/libs/armeabi-v7a/libchrome.so', |
| 'lib/libchrome.so', |
| 'libchrome.so', |
| ] |
| |
| return check_android_binaries(binaries, output_directory, results_collector) |
| |
| |
| def main_android_cronet(output_directory, results_collector, size_path): |
| """Print appropriate size information about Android Cronet targets. |
| |
| Returns the first non-zero exit status of any command it executes, |
| or zero on success. |
| """ |
| assert size_path is None |
| # Use version in binary file name, but not in printed output. |
| binaries_with_paths = glob.glob( |
| os.path.join(output_directory, 'libcronet.*.so')) |
| num_binaries = len(binaries_with_paths) |
| assert num_binaries == 1, "Got %d binaries: %s" % ( |
| num_binaries, ', '.join(binaries_with_paths)) |
| binaries = [os.path.basename(binaries_with_paths[0])] |
| binaries_to_print = ['libcronet.so'] |
| |
| return check_android_binaries(binaries, output_directory, results_collector, |
| binaries_to_print) |
| |
| |
| def main_win(output_directory, results_collector, size_path): |
| """Print appropriate size information about built Windows targets. |
| |
| Returns the first non-zero exit status of any command it executes, |
| or zero on success. |
| """ |
| assert size_path is None |
| files = [ |
| 'chrome.dll', |
| 'chrome.dll.pdb', |
| 'chrome.exe', |
| 'chrome_child.dll', |
| 'chrome_child.dll.pdb', |
| 'chrome_elf.dll', |
| 'chrome_proxy.exe', |
| 'chrome_watcher.dll', |
| 'elevated_tracing_service.exe', |
| 'elevation_service.exe', |
| 'libEGL.dll', |
| 'libGLESv2.dll', |
| 'mini_installer.exe', |
| 'notification_helper.exe', |
| 'resources.pak', |
| 'setup.exe', |
| 'WidevineCdm\\_platform_specific\\win_arm64\\widevinecdm.dll', |
| 'WidevineCdm\\_platform_specific\\win_x64\\widevinecdm.dll', |
| 'WidevineCdm\\_platform_specific\\win_x86\\widevinecdm.dll', |
| ] |
| |
| for f in files: |
| p = os.path.join(output_directory, f) |
| if os.path.isfile(p): |
| results_collector.add_result(f, f, get_size(p), 'bytes') |
| |
| return 0 |
| |
| |
| def format_for_histograms_conversion(data): |
| # We need to do two things to the provided data to make it compatible with the |
| # conversion script: |
| # 1. Add a top-level "benchmark_name" key. |
| # 2. Pull out the "identifier" value to be the story name. |
| formatted_data = {} |
| for metric, metric_data in data.items(): |
| story = metric_data['identifier'] |
| formatted_data[metric] = {story: metric_data.copy()} |
| del formatted_data[metric][story]['identifier'] |
| return {'benchmark_name': 'sizes', 'charts': formatted_data} |
| |
| |
| def main(): |
| if sys.platform in ('win32', 'cygwin'): |
| default_platform = 'win' |
| elif sys.platform.startswith('darwin'): |
| default_platform = 'mac' |
| elif sys.platform.startswith('linux'): |
| default_platform = 'linux' |
| else: |
| default_platform = None |
| |
| main_map = { |
| 'android': main_android, |
| 'android-cronet': main_android_cronet, |
| 'linux': main_linux, |
| 'mac': main_mac, |
| 'win': main_win, |
| } |
| platforms = sorted(main_map.keys()) |
| |
| parser = argparse.ArgumentParser() |
| parser.add_argument( |
| '--output-directory', |
| type=os.path.realpath, |
| help='Chromium output directory, e.g. /path/to/src/out/Debug') |
| parser.add_argument( |
| '--platform', |
| default=default_platform, |
| help='specify platform (%s) [default: %%(default)s]' % |
| ', '.join(platforms)) |
| parser.add_argument('--size-path', default=None, help='Path to size binary') |
| |
| # Accepted to conform to the isolated script interface, but ignored. |
| parser.add_argument('--isolated-script-test-filter', help=argparse.SUPPRESS) |
| parser.add_argument( |
| '--isolated-script-test-perf-output', help=argparse.SUPPRESS) |
| parser.add_argument('--isolated-script-test-repeat', help=argparse.SUPPRESS) |
| parser.add_argument('--isolated-script-test-launcher-retry-limit', |
| help=argparse.SUPPRESS) |
| |
| parser.add_argument( |
| '--isolated-script-test-output', |
| type=os.path.realpath, |
| help='File to which simplified JSON results will be written.') |
| |
| args = parser.parse_args() |
| |
| real_main = main_map.get(args.platform) |
| if not real_main: |
| if args.platform is None: |
| sys.stderr.write('Unsupported sys.platform %s.\n' % repr(sys.platform)) |
| else: |
| sys.stderr.write('Unknown platform %s.\n' % repr(args.platform)) |
| msg = 'Use the --platform= option to specify a supported platform:\n' |
| sys.stderr.write(msg + ' ' + ' '.join(platforms) + '\n') |
| return 2 |
| |
| isolated_script_output = { |
| 'valid': False, |
| 'failures': [], |
| 'version': 'simplified' |
| } |
| test_name = 'sizes' |
| |
| results_directory = None |
| if args.isolated_script_test_output: |
| results_directory = os.path.join( |
| os.path.dirname(args.isolated_script_test_output), test_name) |
| if not os.path.exists(results_directory): |
| os.makedirs(results_directory) |
| |
| results_collector = ResultsCollector() |
| result_sink_client = result_sink.TryInitClient() |
| try: |
| rc = real_main(args.output_directory, results_collector, args.size_path) |
| isolated_script_output = { |
| 'valid': True, |
| 'failures': [test_name] if rc else [], |
| 'version': 'simplified', |
| } |
| finally: |
| if results_directory: |
| results_path = os.path.join(results_directory, 'test_results.json') |
| with open(results_path, 'w') as output_file: |
| json.dump(isolated_script_output, output_file) |
| |
| histogram_path = os.path.join(results_directory, 'perf_results.json') |
| # We need to add a bit more data to the results and rearrange some things, |
| # otherwise the conversion fails due to the provided data being malformed. |
| updated_results = format_for_histograms_conversion( |
| results_collector.results) |
| with open(histogram_path, 'w') as f: |
| json.dump(updated_results, f) |
| histogram_result = convert_chart_json.ConvertChartJson(histogram_path) |
| if histogram_result.returncode != 0: |
| sys.stderr.write( |
| 'chartjson conversion failed: %s\n' % histogram_result.stdout) |
| rc = rc or histogram_result.returncode |
| else: |
| with open(histogram_path, 'wb') as f: |
| f.write(histogram_result.stdout) |
| if result_sink_client: |
| status = result_types.PASS |
| if not isolated_script_output['valid']: |
| status = result_types.UNKNOWN |
| elif isolated_script_output['failures']: |
| status = result_types.FAIL |
| result_sink_client.Post(test_name, status, None, None, None) |
| |
| return rc |
| |
| |
| if '__main__' == __name__: |
| sys.exit(main()) |