HTTPSConnection script fails, but only on some servers (long)

[Paul Winkler]

Thanks for the replies, Steve and Andreas!   I will check out pycurl,
thanks very much for the tip.

Meanwhile, I'm trying to prepare a bug report re. httplib and get as
much information as possible.

Something I neglected to mention: when the script hits the problematic
server, it always takes about 3 minutes to get the Bad Gateway
response. Don't know if that's indicative of anything.

I added a bunch of blather to httplib.py to see at what point things
are waiting, or if it was stuck in a loop or what. The result is pretty
clear:  we get as far as this point in SSLFile:

    def _read(self):
        buf = ''
        # put in a loop so that we retry on transient errors
        while True:
            try:
                buf = self._ssl.read(self._bufsize)

... at which point we simply wait for the server for three minutes,
Then a response finally comes back, no exceptions are caught or raised
within _read(), and finally _read()  returns buf. I can't easily trace
any deeper because self._ssl apparently comes from _ssl.so and I don't
fancy hacking at the C code.

Do these observations seem consistent with the hypothesis that
HTTPSConnection is failing to handle the HTTP 1.0 proxy?

I will also see what else I can find out from the admin. Maybe there's
more useful info in the logs somewhere.  Unfortunately IIRC our jboss
log is always clogged with a few zillion irrelevant messages ... that
should be fun.

-PW