You can clone with
HTTPS or Subversion.
pycurl has a bug that causes integer overflow and
to return large negative values, which starts a really, really long chain of _handle_timeout() calls and tornado to stop responding.
See also http://curl.haxx.se/mail/curlpython-2013-04/0000.html .
Until pycurl is fixed, it would be a good idea to do:
if new_timeout >= 0:
Ignore negative return values from CurlMulti.timeout.
pycurl has a wraparound bug that can turn very large timeouts into