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.

346 lines
9.2 KiB

#
# Copyright (c) 2013- Facebook. All rights reserved.
#
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from __future__ import unicode_literals
import csv
import fnmatch
import gzip
import json
import logging
import os
import subprocess
import sys
import tempfile
import time
BIN_DIRECTORY = os.path.dirname(os.path.realpath(__file__))
LIB_DIRECTORY = os.path.join(BIN_DIRECTORY, '..', 'lib', 'java')
TMP_DIRECTORY = tempfile.gettempdir()
MODELS_JAR = os.path.join(LIB_DIRECTORY, 'models.jar')
DEFAULT_INFER_OUT = os.path.join(os.getcwd(), 'infer-out')
CSV_PERF_FILENAME = 'performances.csv'
STATS_FILENAME = 'stats.json'
CSV_REPORT_FILENAME = 'report.csv'
JSON_REPORT_FILENAME = 'report.json'
BUGS_FILENAME = 'bugs.txt'
CSV_INDEX_KIND = 1
CSV_INDEX_TYPE = 2
CSV_INDEX_QUALIFIER = 3
CSV_INDEX_LINE = 5
CSV_INDEX_FILENAME = 8
CSV_INDEX_QUALIFIER_TAGS = 11
QUALIFIER_TAGS = 'qualifier_tags'
BUCKET_TAGS = 'bucket'
IOS_CAPTURE_ERRORS = 'errors'
IOS_BUILD_OUTPUT = 'build_output'
BUCK_INFER_OUT = 'infer'
FORMAT = '[%(levelname)s] %(message)s'
DEBUG_FORMAT = '[%(levelname)s:%(filename)s:%(lineno)03d] %(message)s'
# Monkey patching subprocess (I'm so sorry!).
if "check_output" not in dir(subprocess):
def f(*popenargs, **kwargs):
if 'stdout' in kwargs:
raise ValueError('stdout not supported')
process = subprocess.Popen(
stdout=subprocess.PIPE,
*popenargs,
**kwargs)
output, unused_err = process.communicate()
retcode = process.poll()
if retcode:
cmd = kwargs.get("args")
if cmd is None:
cmd = popenargs[0]
raise subprocess.CalledProcessError(retcode, cmd)
return output
subprocess.check_output = f
def configure_logging(debug, quiet=False):
"""Configures the default logger. This can be called only once and has to
be called before any logging is done.
"""
logging.TIMING = logging.ERROR + 5
logging.addLevelName(logging.TIMING, "TIMING")
def timing(msg, *args, **kwargs):
logging.log(logging.TIMING, msg, *args, **kwargs)
logging.timing = timing
if quiet:
logging.basicConfig(level=logging.TIMING, format=FORMAT)
elif not debug:
logging.basicConfig(level=logging.INFO, format=FORMAT)
else:
logging.basicConfig(level=logging.DEBUG, format=DEBUG_FORMAT)
def elapsed_time(start_time):
return time.time() - start_time
def error(msg):
print(msg, file=sys.stderr)
def get_cmd_in_bin_dir(binary_name):
# this relies on the fact that utils.py is located in infer/bin
return os.path.join(
os.path.dirname(os.path.realpath(__file__)),
binary_name)
def write_cmd_streams_to_file(logfile, cmd=None, out=None, err=None):
with open(logfile, 'w') as log_filedesc:
if cmd:
log_filedesc.write(' '.join(cmd) + '\n')
if err is not None:
errors = str(err)
log_filedesc.write('\nSTDERR:\n')
log_filedesc.write(errors)
if out is not None:
output = str(out)
log_filedesc.write('\n\nSTDOUT:\n')
log_filedesc.write(output)
def save_failed_command(
infer_out,
cmd,
message,
prefix='failed_',
out=None,
err=None):
cmd_filename = tempfile.mktemp(
'_' + message + ".txt",
prefix, infer_out
)
write_cmd_streams_to_file(cmd_filename, cmd=cmd, out=out, err=err)
logging.error('\n' + message + ' error saved in ' + cmd_filename)
def run_command(cmd, debug_mode, infer_out, message, env=os.environ):
if debug_mode:
print('\n{0}\n'.format(' '.join(cmd)))
try:
return subprocess.check_call(cmd, env=env)
except subprocess.CalledProcessError as e:
save_failed_command(infer_out, cmd, message)
raise e
def print_exit(s):
print(s)
exit(os.EX_OK)
def infer_version():
version = json.loads(subprocess.check_output([
get_cmd_in_bin_dir('InferAnalyze'),
'-version_json',
]).decode())
return version['commit']
def infer_branch():
version = json.loads(subprocess.check_output([
get_cmd_in_bin_dir('InferAnalyze'),
'-version_json',
]).decode())
return version['branch']
def infer_key(analyzer):
return os.pathsep.join([analyzer, infer_version()])
def vcs_branch(dir='.'):
cwd = os.getcwd()
try:
os.chdir(dir)
branch = subprocess.check_output([
'git',
'rev-parse',
'--abbrev-ref',
'HEAD',
]).decode().strip()
except subprocess.CalledProcessError:
try:
branch = subprocess.check_output([
'hg',
'id',
'-B',
]).decode().strip()
except subprocess.CalledProcessError:
branch = 'not-versioned'
finally:
os.chdir(cwd)
return branch
def vcs_revision(dir='.'):
cwd = os.getcwd()
try:
os.chdir(dir)
revision = subprocess.check_output([
'git',
'rev-parse',
'HEAD',
]).decode().strip()
except subprocess.CalledProcessError:
try:
revision = subprocess.check_output([
'hg',
'id',
'-i',
]).decode().strip()
except subprocess.CalledProcessError:
revision = 'not-versioned'
finally:
os.chdir(cwd)
return revision
class Timer:
"""Simple logging timer. Initialize with a printf like logging function."""
def __init__(self, logger=lambda x: None):
self._logger = logger
self._start = 0
def start(self, message=None, *args):
self._start = time.time()
if message:
self._logger(message, *args)
def stop(self, message=None, *args):
self._stop = time.time()
self._dt = self._stop - self._start
if message:
self._logger(message + ' (%.2fs)', *(args + (self._dt,)))
return self._dt
def interact():
"""Start interactive mode. Useful for debugging.
"""
import code
code.interact(local=locals())
def search_files(root_dir, extension):
# Input:
# - root directory where to start a recursive search of yjson files
# - file extension to search from the root
# Output:
# - list of absolute filepaths
files = []
if not os.path.isabs(root_dir):
root_dir = os.path.abspath(root_dir)
for dirpath, _, filenames in os.walk(root_dir):
for filename in fnmatch.filter(filenames, "*" + extension):
files.append(os.path.join(dirpath, filename))
return files
def uncompress_gzip_file(gzip_file, out_dir):
# This is python2.6 compliant, gzip.open doesn't support 'with' statement
# Input:
# - gzip file path
# - output directory where uncompress the file
# Output:
# - path of the uncompressed file
# NOTE: the file is permanently created, is responsibility of the
# caller to delete it
uncompressed_path = None
uncompressed_fd = None
compressed_fd = None
try:
# the uncompressed filename loses its final extension
# (for example abc.gz -> abc)
uncompressed_path = os.path.join(
out_dir,
os.path.splitext(gzip_file)[0],
)
uncompressed_fd = open(uncompressed_path, 'wb')
compressed_fd = gzip.open(gzip_file, 'rb')
uncompressed_fd.write(compressed_fd.read())
return uncompressed_path
except IOError as exc:
# delete the uncompressed file (if exists)
if uncompressed_path is not None and os.path.exists(uncompressed_path):
os.remove(uncompressed_path)
raise exc
finally:
if compressed_fd is not None:
compressed_fd.close()
if uncompressed_fd is not None:
uncompressed_fd.close()
def run_process(cmd, cwd=None, logfile=None):
# Input:
# - command to execute
# - current working directory to cd before running the cmd
# - logfile where to dump stdout/stderr
# Output:
# - exitcode of the executed process
p = subprocess.Popen(
cmd,
cwd=cwd,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
(out, err) = p.communicate()
if logfile:
write_cmd_streams_to_file(logfile, cmd=cmd, out=out, err=err)
return p.returncode
def invoke_function_with_callbacks(
func,
args,
on_terminate=None,
on_exception=None):
try:
res = func(*args)
if on_terminate:
on_terminate(res)
return res
except Exception as exc:
if on_exception:
return on_exception(exc)
raise
def create_json_report(out_dir):
csv_report_filename = os.path.join(out_dir, CSV_REPORT_FILENAME)
json_report_filename = os.path.join(out_dir, JSON_REPORT_FILENAME)
rows = []
with open(csv_report_filename, 'r') as file_in:
reader = csv.reader(file_in)
rows = [row for row in reader]
with open(json_report_filename, 'w') as file_out:
headers = rows[0]
issues = rows[1:]
json.dump([dict(zip(headers, row)) for row in issues], file_out)
# vim: set sw=4 ts=4 et: