On 13 oct 2004, Kevin Greiner wrote:

> Xavier Maillard <zedek@gnu-rox.org> writes:
> 
> > On 11 oct 2004, Xavier Maillard wrote:
> > 
> > > Hello,
> > > 
> > > New problem I have since 2 or 3 days now.
> > > 
> > > Sometimes for an unknown reason, I get this backtrace when
> > > trying to display an article:
> > > 
> > > ,----
> > > > Debugger entered--Lisp error: (error "Server closed
> > > > connection") signal(error ("Server closed connection"))
> > > > error("Server closed connection") apply(error "Server
> > > > closed connection") nntp-report("Server closed
> > > > connection") nntp-accept-process-output(nil) byte-code("
> > > > =ƒ% gnus-async-wait-for-article(15716)
> > > > gnus-async-request-fetched-article("nntp+gnu-rox.org:gnus.ding"
> > > > 15716 #<buffer *Article*>)
> > > > gnus-request-article-this-buffer(15716
> > > > "nntp+gnu-rox.org:gnus.ding") gnus-article-prepare(15716
> > > > nil) gnus-summary-display-article(15716 nil)
> > > > gnus-summary-select-article(nil nil pseudo)
> > > > gnus-summary-scroll-up(1)
> > > > call-interactively(gnus-summary-scroll-up)
> > > `----
> > > 
> > > It doesn't happen every time but in a random fashion. I can
> > > for example read 10 articles in a row and then, Gnus refuse
> > > to display one. Dunno what can cause this to happen since I
> > > have this behavior when plugged or not.
> > 
> > Any taker ?
> > 
> > I tried to (setq nntp-record... t) and switched to the
> > corresponding buffer but all I get is:
> > 
> > ,----
> > > 20041012T002908.472 gnu-rox.org *** CALLED nntp-report *** 
> > `----
> > 
> > This is kind of annoying since I don't really know what can
> > block this.
> > 
> > Any help *grantly* appreciated.
> > 
> > Hint: I just upgraded my local copy of Gnus from CVS and
> > still got this after having restarted Gnus.
> 
> The problem is that some nntp servers close idle connections to
> conserve their resources while others punish a client that
> leaves idle connections open by ignoring future requests. The
> nntp backend handles both of these problems by wrapping the
> nntp command logic in the nntp-with-open-group form.
> 
> Just guessing but I'd bet that the problem is that
> gnus-async-wait-for-article binds proc. The
> nntp-with-open-group recovers from a closed nntp connection by
> opening a new connection. The gnus-async-wait-for-article
> nevers sees that new connection because it binds proc before it
> enters its loop.

Oh yes I see.

> So, try replacing (nntp-accept-process-output proc) in
> gnus-async-wait-for-article with (nntp-accept-process-output
> (nntp-find-connection (current-buffer))).

Hmm, looking at the function I see: