| Index: sitescripts/subscriptions/bin/updateMalwareDomainsList.py | 
| =================================================================== | 
| --- a/sitescripts/subscriptions/bin/updateMalwareDomainsList.py | 
| +++ b/sitescripts/subscriptions/bin/updateMalwareDomainsList.py | 
| @@ -11,16 +11,17 @@ | 
| # GNU General Public License for more details. | 
| # | 
| # You should have received a copy of the GNU General Public License | 
| # along with Adblock Plus.  If not, see <http://www.gnu.org/licenses/>. | 
|  | 
| import os | 
| import subprocess | 
| import codecs | 
| +import contextlib | 
| import urllib2 | 
| import zipfile | 
| import tempfile | 
| import shutil | 
| import sys | 
| from StringIO import StringIO | 
| from sitescripts.utils import get_config | 
|  | 
| @@ -31,53 +32,63 @@ | 
| ! Last modified: %timestamp% | 
| ! Expires: 1d | 
| !''' | 
|  | 
| MALWAREDOMAINS_PATH = '/files/justdomains.zip' | 
|  | 
|  | 
| def try_mirror(mirror): | 
| +    url = mirror + MALWAREDOMAINS_PATH | 
| try: | 
| -        response = urllib2.urlopen(mirror + MALWAREDOMAINS_PATH) | 
| -        return response.read() | 
| -    except urllib2.HTTPError: | 
| -        return None | 
| +        with contextlib.closing(urllib2.urlopen(url)) as response: | 
| +            return None, response.read() | 
| +    except urllib2.HTTPError as exc: | 
| +        exc.close() | 
| +    except urllib2.URLError as exc: | 
| +        pass | 
| +    return 'Failed to fetch {}: {}'.format(url, exc), None | 
|  | 
|  | 
| -if __name__ == '__main__': | 
| +def main(): | 
| config = get_config() | 
| section = 'subscriptionDownloads' | 
| repository = config.get(section, 'malwaredomains_repository') | 
| mirrors = config.get(section, 'malwaredomains_mirrors').split() | 
|  | 
| tempdir = tempfile.mkdtemp(prefix='malwaredomains') | 
| try: | 
| subprocess.check_call(['hg', '-q', 'clone', '-U', repository, tempdir]) | 
| subprocess.check_call(['hg', '-q', 'up', '-R', tempdir, '-r', 'default']) | 
|  | 
| path = os.path.join(tempdir, 'malwaredomains_full.txt') | 
| file = codecs.open(path, 'wb', encoding='utf-8') | 
|  | 
| print >>file, FILTERLIST_HEADER | 
|  | 
| +        error_report = ['Unable to fetch malware domains list', 'Errors:'] | 
| for mirror in mirrors: | 
| -            data = try_mirror(mirror) | 
| +            error_message, data = try_mirror(mirror) | 
| if data is not None: | 
| break | 
| +            error_report.append(error_message) | 
| else: | 
| -            sys.exit('Unable to fetch malware domains list.') | 
| +            sys.exit('\n'.join(error_report)) | 
|  | 
| -        zip = zipfile.ZipFile(StringIO(data), 'r') | 
| -        info = zip.infolist()[0] | 
| -        for line in str(zip.read(info.filename)).splitlines(): | 
| +        zf = zipfile.ZipFile(StringIO(data), 'r') | 
| +        info = zf.infolist()[0] | 
| +        for line in str(zf.read(info.filename)).splitlines(): | 
| domain = line.strip() | 
| if not domain: | 
| continue | 
|  | 
| print >>file, '||%s^' % domain.decode('idna') | 
| file.close() | 
|  | 
| if subprocess.check_output(['hg', 'stat', '-R', tempdir]) != '': | 
| subprocess.check_call(['hg', '-q', 'commit', '-R', tempdir, '-A', '-u', 'hgbot', '-m', 'Updated malwaredomains.com data']) | 
| subprocess.check_call(['hg', '-q', 'push', '-R', tempdir]) | 
| finally: | 
| shutil.rmtree(tempdir, ignore_errors=True) | 
| + | 
| + | 
| +if __name__ == '__main__': | 
| +    main() | 
|  |