You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
82 lines
2.3 KiB
82 lines
2.3 KiB
# Copyright 2018 The Chromium Authors. All rights reserved.
|
|
# Use of this source code is governed by a BSD-style license that can be
|
|
# found in the LICENSE file.
|
|
|
|
|
|
"""Writes a Perf-formated json file with stats about the given web file."""
|
|
|
|
|
|
from __future__ import print_function
|
|
import json
|
|
import os
|
|
import subprocess
|
|
import sys
|
|
|
|
|
|
def main():
|
|
input_file = sys.argv[1]
|
|
out_dir = sys.argv[2]
|
|
keystr = sys.argv[3]
|
|
propstr = sys.argv[4]
|
|
bloaty_path = sys.argv[5]
|
|
total_size_bytes_key = sys.argv[6]
|
|
magic_seperator = sys.argv[7]
|
|
|
|
results = {
|
|
'key': { },
|
|
'results': { }
|
|
}
|
|
|
|
print(magic_seperator)
|
|
print('If you see lots of func[19] and such, go check out the debug build')
|
|
print('Note that template instantiations are grouped together, '
|
|
'thus the elided types.')
|
|
print('If you notice an unsymbolized "duplicate" entry, it is simply how '
|
|
'many bytes the function name itself takes up')
|
|
print(subprocess.check_output([bloaty_path, input_file,
|
|
'-d', 'shortsymbols', '-n', '0']))
|
|
|
|
print(magic_seperator)
|
|
print('If you see lots of func[19] and such, go check out the debug build')
|
|
print(subprocess.check_output([bloaty_path, input_file,
|
|
'-d', 'fullsymbols', '-n', '0']))
|
|
|
|
props = propstr.split(' ')
|
|
for i in range(0, len(props), 2):
|
|
results[props[i]] = props[i+1]
|
|
|
|
keys = keystr.split(' ')
|
|
for i in range(0, len(keys), 2):
|
|
results['key'][keys[i]] = keys[i+1]
|
|
|
|
r = {
|
|
total_size_bytes_key: os.path.getsize(input_file)
|
|
}
|
|
|
|
# Make a copy to avoid destroying the hardlinked file.
|
|
# Swarming hardlinks in the builds from isolated cache.
|
|
temp_file = input_file + '_tmp'
|
|
subprocess.check_call(['cp', input_file, temp_file])
|
|
subprocess.check_call(['gzip', temp_file])
|
|
|
|
r['gzip_size_bytes'] = os.path.getsize(temp_file + '.gz')
|
|
|
|
name = os.path.basename(input_file)
|
|
|
|
results['results'][name] = {
|
|
# We need this top level layer 'config'/slice
|
|
# Other analysis methods (e.g. libskia) might have
|
|
# slices for data on the 'code' section, etc.
|
|
'default' : r,
|
|
}
|
|
|
|
print(magic_seperator)
|
|
print(json.dumps(results, indent=2))
|
|
|
|
with open(os.path.join(out_dir, name+'.json'), 'w') as output:
|
|
output.write(json.dumps(results, indent=2))
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|