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.

487 lines
16 KiB

#!/usr/bin/env python2.7
# Copyright (c) 2013 - present Facebook, Inc.
# All rights reserved.
#
# This source code is licensed under the BSD style license found in the
# LICENSE file in the root directory of this source tree. An additional grant
# of patent rights can be found in the PATENTS file in the same directory.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from __future__ import unicode_literals
import argparse
import csv
import io
import json
import logging
import multiprocessing
import os
import platform
import re
import shutil
import stat
import subprocess
import sys
import tempfile
import time
import traceback
import zipfile
from inferlib import analyze, config, issues, utils
ANALYSIS_SUMMARY_OUTPUT = 'analysis_summary.txt'
DEFAULT_BUCK_OUT = os.path.join(os.getcwd(), 'buck-out')
DEFAULT_BUCK_OUT_GEN = os.path.join(DEFAULT_BUCK_OUT, 'gen')
INFER_CSV_REPORT = os.path.join(config.BUCK_INFER_OUT,
config.CSV_REPORT_FILENAME)
INFER_JSON_REPORT = os.path.join(config.BUCK_INFER_OUT,
config.JSON_REPORT_FILENAME)
INFER_STATS = os.path.join(config.BUCK_INFER_OUT, config.STATS_FILENAME)
INFER_SCRIPT = """\
#!/usr/bin/env {0}
import subprocess
import sys
cmd = ['{0}'] + {1} + ['--', 'javac'] + sys.argv[1:]
subprocess.check_call(cmd)
"""
LOCAL_CONFIG = """\
[tools]
javac = %s
"""
def prepare_build(args):
"""Creates script that redirects javac calls to infer and a local buck
configuration that tells buck to use that script.
"""
infer_options = [
'--buck',
'--incremental',
'--analyzer', args.analyzer,
]
if args.debug:
infer_options.append('--debug')
if args.no_filtering:
infer_options.append('--no-filtering')
if args.debug_exceptions:
infer_options += ['--debug-exceptions', '--no-filtering']
# Create a temporary directory as a cache for jar files.
infer_cache_dir = os.path.join(args.infer_out, 'cache')
if not os.path.isdir(infer_cache_dir):
os.mkdir(infer_cache_dir)
infer_options.append('--infer_cache')
infer_options.append(infer_cache_dir)
temp_files = [infer_cache_dir]
try:
infer = [utils.get_cmd_in_bin_dir('infer')] + infer_options
except subprocess.CalledProcessError as e:
logging.error('Could not find infer')
raise e
# Disable the use of buckd as this scripts modifies .buckconfig.local
logging.info('Disabling buckd: export NO_BUCKD=1')
os.environ['NO_BUCKD'] = '1'
# make sure INFER_ANALYSIS is set when buck is called
logging.info('Setup Infer analysis mode for Buck: export INFER_ANALYSIS=1')
os.environ['INFER_ANALYSIS'] = '1'
# Create a script to be called by buck
infer_script = None
with tempfile.NamedTemporaryFile(delete=False,
prefix='infer_',
suffix='.py',
dir='.') as infer_script:
logging.info('Creating %s' % infer_script.name)
infer_script.file.write(
INFER_SCRIPT.format(sys.executable, infer).encode())
st = os.stat(infer_script.name)
os.chmod(infer_script.name, st.st_mode | stat.S_IEXEC)
temp_files += [infer_script.name]
return temp_files, infer_script.name
def get_normalized_targets(targets):
""" Use buck to convert a list of input targets/aliases
into a set of the (transitive) target deps for all inputs"""
# this expands the targets passed on the command line, then filters away
# targets that are not Java/Android. you need to change this if you
# care about something other than Java/Android
TARGET_TYPES = "kind('android_library|java_library', deps('%s'))"
BUCK_GET_JAVA_TARGETS = ['buck', 'query', TARGET_TYPES]
buck_cmd = BUCK_GET_JAVA_TARGETS + targets
try:
targets = filter(
lambda line: len(line) > 0,
subprocess.check_output(buck_cmd).decode().strip().split('\n'))
if len(targets) > 0 and args.verbose:
logging.debug('Targets to analyze:')
for target in targets:
logging.debug(target)
return targets
except subprocess.CalledProcessError as e:
logging.error('Error while expanding targets with {0}'.format(buck_cmd))
raise e
def init_stats(args, start_time):
"""Returns dictionary with target independent statistics.
"""
return {
'float': {},
'int': {
'cores': multiprocessing.cpu_count(),
'time': int(time.time()),
'start_time': int(round(start_time)),
},
'normal': {
'debug': str(args.debug),
'analyzer': args.analyzer,
'machine': platform.machine(),
'node': platform.node(),
'project': os.path.basename(os.getcwd()),
'revision': utils.vcs_revision(),
'branch': utils.vcs_branch(),
'system': platform.system(),
'infer_version': utils.infer_version(),
'infer_branch': utils.infer_branch(),
}
}
def store_performances_csv(infer_out, stats):
"""Stores the statistics about perfromances into a CSV file to be exported
to a database"""
perf_filename = os.path.join(infer_out, config.CSV_PERF_FILENAME)
with open(perf_filename, 'w') as csv_file_out:
csv_writer = csv.writer(csv_file_out)
keys = ['infer_version', 'project', 'revision', 'files', 'lines',
'cores', 'system', 'machine', 'node', 'total_time',
'capture_time', 'analysis_time', 'reporting_time', 'time']
int_stats = list(stats['int'].items())
normal_stats = list(stats['normal'].items())
flat_stats = dict(int_stats + normal_stats)
values = []
for key in keys:
values.append(flat_stats[key])
csv_writer.writerow(keys)
csv_writer.writerow(values)
csv_file_out.flush()
def get_harness_code():
all_harness_code = '\nGenerated harness code:\n'
for filename in os.listdir(DEFAULT_BUCK_OUT_GEN):
if 'InferGeneratedHarness' in filename:
all_harness_code += '\n' + filename + ':\n'
with open(os.path.join(DEFAULT_BUCK_OUT_GEN,
filename), 'r') as file_in:
all_harness_code += file_in.read()
return all_harness_code + '\n'
def get_basic_stats(stats):
files_analyzed = '{0} files ({1} lines) analyzed in {2}s\n\n'.format(
stats['int']['files'],
stats['int']['lines'],
stats['int']['total_time'],
)
phase_times = 'Capture time: {0}s\nAnalysis time: {1}s\n\n'.format(
stats['int']['capture_time'],
stats['int']['analysis_time'],
)
to_skip = {
'files',
'procedures',
'lines',
'cores',
'time',
'start_time',
'capture_time',
'analysis_time',
'reporting_time',
'total_time',
'makefile_generation_time'
}
bugs_found = 'Errors found:\n\n'
for key, value in sorted(stats['int'].items()):
if key not in to_skip:
bugs_found += ' {0:>8} {1}\n'.format(value, key)
basic_stats_message = files_analyzed + phase_times + bugs_found + '\n'
return basic_stats_message
def get_buck_stats():
trace_filename = os.path.join(
DEFAULT_BUCK_OUT,
'log',
'traces',
'build.trace'
)
ARGS = 'args'
SUCCESS_STATUS = 'success_type'
buck_stats = {}
try:
trace = utils.load_json_from_path(trace_filename)
for t in trace:
if SUCCESS_STATUS in t[ARGS]:
status = t[ARGS][SUCCESS_STATUS]
count = buck_stats.get(status, 0)
buck_stats[status] = count + 1
buck_stats_message = 'Buck build statistics:\n\n'
for key, value in sorted(buck_stats.items()):
buck_stats_message += ' {0:>8} {1}\n'.format(value, key)
return buck_stats_message
except IOError as e:
logging.error('Caught %s: %s' % (e.__class__.__name__, str(e)))
logging.error(traceback.format_exc())
return ''
class NotFoundInJar(Exception):
pass
def load_stats(opened_jar):
try:
return json.loads(opened_jar.read(INFER_STATS).decode())
except KeyError as e:
raise NotFoundInJar
def load_csv_report(opened_jar):
try:
sio = io.StringIO(opened_jar.read(INFER_CSV_REPORT).decode())
return list(utils.locale_csv_reader(sio))
except KeyError as e:
raise NotFoundInJar
def load_json_report(opened_jar):
try:
return json.loads(opened_jar.read(INFER_JSON_REPORT).decode())
except KeyError as e:
raise NotFoundInJar
def collect_results(args, start_time):
"""Walks through buck-gen, collects results for the different buck targets
and stores them in in args.infer_out/results.csv.
"""
buck_stats = get_buck_stats()
logging.info(buck_stats)
with open(os.path.join(args.infer_out, ANALYSIS_SUMMARY_OUTPUT), 'w') as f:
f.write(buck_stats)
all_csv_rows = set()
all_json_rows = set()
headers = []
stats = init_stats(args, start_time)
accumulation_whitelist = list(map(re.compile, [
'^cores$',
'^time$',
'^start_time$',
'.*_pc',
]))
expected_analyzer = stats['normal']['analyzer']
expected_version = stats['normal']['infer_version']
for root, _, files in os.walk(DEFAULT_BUCK_OUT_GEN):
for f in [f for f in files if f.endswith('.jar')]:
path = os.path.join(root, f)
try:
with zipfile.ZipFile(path) as jar:
# Accumulate integers and float values
target_stats = load_stats(jar)
found_analyzer = target_stats['normal']['analyzer']
found_version = target_stats['normal']['infer_version']
if (found_analyzer != expected_analyzer
or found_version != expected_version):
continue
else:
for type_k in ['int', 'float']:
items = target_stats.get(type_k, {}).items()
for key, value in items:
if not any(map(lambda r: r.match(key),
accumulation_whitelist)):
old_value = stats[type_k].get(key, 0)
stats[type_k][key] = old_value + value
csv_rows = load_csv_report(jar)
if len(csv_rows) > 0:
headers.append(csv_rows[0])
for row in csv_rows[1:]:
all_csv_rows.add(tuple(row))
json_rows = load_json_report(jar)
for row in json_rows:
all_json_rows.add(json.dumps(row))
# Override normals
stats['normal'].update(target_stats.get('normal', {}))
except NotFoundInJar:
pass
except zipfile.BadZipfile:
logging.warn('Bad zip file %s', path)
csv_report = os.path.join(args.infer_out, config.CSV_REPORT_FILENAME)
json_report = os.path.join(args.infer_out, config.JSON_REPORT_FILENAME)
bugs_out = os.path.join(args.infer_out, config.BUGS_FILENAME)
if len(headers) == 0:
with open(csv_report, 'w'):
pass
logging.info('No reports found')
return
elif len(headers) > 1:
if any(map(lambda x: x != headers[0], headers)):
raise Exception('Inconsistent reports found')
# Convert all float values to integer values
for key, value in stats.get('float', {}).items():
stats['int'][key] = int(round(value))
# Delete the float entries before exporting the results
del(stats['float'])
with open(csv_report, 'w') as report:
writer = csv.writer(report)
all_csv_rows = [list(row) for row in all_csv_rows]
writer.writerows([headers[0]] + all_csv_rows)
report.flush()
with open(json_report, 'w') as report:
json_string = '['
json_string += ','.join(all_json_rows)
json_string += ']'
report.write(json_string)
report.flush()
print('\n')
issues.print_and_save_errors(json_report, bugs_out)
stats['int']['total_time'] = int(round(utils.elapsed_time(start_time)))
store_performances_csv(args.infer_out, stats)
stats_filename = os.path.join(args.infer_out, config.STATS_FILENAME)
utils.dump_json_to_path(stats, stats_filename)
basic_stats = get_basic_stats(stats)
if args.print_harness:
harness_code = get_harness_code()
basic_stats += harness_code
logging.info(basic_stats)
with open(os.path.join(args.infer_out, ANALYSIS_SUMMARY_OUTPUT), 'a') as f:
f.write(basic_stats)
def cleanup(temp_files):
"""Removes the generated .buckconfig.local and the temporary infer script.
"""
for file in temp_files:
try:
logging.info('Removing %s' % file)
if os.path.isdir(file):
shutil.rmtree(file)
else:
os.unlink(file)
except IOError:
logging.error('Could not remove %s' % file)
if __name__ == '__main__':
parser = argparse.ArgumentParser(parents=[analyze.base_parser])
parser.add_argument('--verbose', action='store_true',
help='Print buck compilation steps')
parser.add_argument('--no-cache', action='store_true',
help='Do not use buck distributed cache')
parser.add_argument('--print-harness', action='store_true',
help='Print generated harness code (Android only)')
parser.add_argument('targets', nargs='*', metavar='target',
help='Build targets to analyze')
args = parser.parse_args()
utils.configure_logging(args.verbose)
timer = utils.Timer(logging.info)
temp_files = []
try:
start_time = time.time()
logging.info('Starting the analysis')
subprocess.check_call(
[utils.get_cmd_in_bin_dir('InferAnalyze'), '-version'])
if not os.path.isdir(args.infer_out):
os.mkdir(args.infer_out)
timer.start('Preparing build...')
temp_files2, infer_script = prepare_build(args)
temp_files += temp_files2
timer.stop('Build prepared')
# TODO(t3786463) Start buckd.
timer.start('Computing library targets')
normalized_targets = get_normalized_targets(args.targets)
timer.stop('%d targets computed', len(normalized_targets))
if len(normalized_targets) == 0:
logging.info('Nothing to analyze')
else:
timer.start('Running buck...')
buck_cmd = [
'buck', 'build', '--config', 'tools.javac=' + infer_script]
if args.no_cache:
buck_cmd += ['--no-cache']
if args.verbose:
buck_cmd += ['-v', '2']
compile_cmd = buck_cmd + normalized_targets
subprocess.check_call(compile_cmd)
timer.stop('Buck finished')
timer.start('Collecting results...')
collect_results(args, start_time)
timer.stop('Done')
except KeyboardInterrupt as e:
timer.stop('Exiting')
sys.exit(0)
except Exception as e:
timer.stop('Failed')
logging.error('Caught %s: %s' % (e.__class__.__name__, str(e)))
logging.error(traceback.format_exc())
sys.exit(1)
finally:
cleanup(temp_files)