Re: gnus makes emacs lose response

[Bob Rogers <rogers-emacs@rgrjr.dyndns.org>]

   From: Richard Stallman <rms@gnu.org>
   Date: Wed, 23 Aug 2006 10:45:30 -0400

       > That means after reconnecting to the internet under Windows a new
       > process is started which has no problem communicating to the server
       > while on GNU/Linux the old one is reused which obviously cannot cope
       > with the new internet connection.

       . . .

       > I am not sure what the right course of action on GNU/Linux would be to
       > remedy the problem.  Should programs like openssl die when the
       > internet connection is being closed?  Or renegotiate a connection?

   I don't know whether openssl can keep working when a phone connection
   drops and is reestablished, but I have seen scp connections keep
   working when I removed and then reinserted a Wifi card.  These cases
   are not the same but they are analogous; if the openssl/phone case
   does not work now, it ought to work.

The problem is fundamental to dialups (and my apologies if this is old
hat, and by "it ought to work" you really meant "Gnus" and not
"openssl").

    For a wireless card or cable modem, the DHCP server remembers your
hardware address, which it uses to give you the IP address you had
before, so established TCP connections resume working as soon as the
interface comes back up.  When you redial, the computer attached to the
modem bank has no clue who are (and might even be a different computer
than before), so you usually get a different IP address, making TCP
connections established with the old IP address unusable.  Either way,
OpenSSL probably isn't even aware of all this commotion in the lower
layers of the networking stack.

   So you might consider this a bug in GNU/Linux that it doesn't close
open connections for old addresses.  However, it may not be possible to
decide that an old address is truly and forevermore gone, and so (I am
guessing) they figured that it is more robust to let higher-level
connection timeout mechanisms come into play.  (It may even be specified
that way, but I don't know the relevant RFCs to check.)  By that
interpretation, the observed difference in behavior is really a Windows
bug.  (And it wouldn't be the first time MS chose user convenience over
robustness, now, would it?  ;-)

       I see similar problems connection to one of the news servers at my ISP.
       I just interrupt Gnus, and make anoter refresh -- I guess Gnus could
       just as well do this automatically if there is no response from the
       server it used last time.

   I don't entirely understand that proposal . . .
   If openssl could resume its connection when you reconnect the phone line,
   would that make this proposal unnecessary?  Would that make this proposal
   undesirable or incorrect?

Would it be possible for Gnus to request a timeout when invoking
openssl?  I'm afraid I don't know enough about either to make a more
detailed suggestion, let alone whether this approach could work.  (And
http://www.openssl.org/docs/apps/s_client.html doesn't show any
timeout-related option, which is not hopeful.)

   If not, then restarting the connection after a suitable timeout is
probably the best available technology.

					-- Bob Rogers
					   [who sometimes pretends to be
					   a network engineer, often
					   with success.]
					   http://rgrjr.dyndns.org/