🚀 go-pugleaf

[I'm cross-posting this to comp.emacs in case anyone there can figure this
one out.  The problem is that GNEWS seems to die if one posts a longish
message using n-reply-program = nil, though smaller messages get out fine.
It signals a file error, "writing to process gnews, i/o error", and then
dies.]

> > >I was unable to post anything in GNEWS today with n-reply-program set
> > >to nil ...GNEWS hung with an error message that said "writing to
> > >process gnews, i/o error" or something like that.
> >
> >My n-reply-program is currently set to nil...
> >
> >I've gotten this message many times, always when posting a followup to
> >the empty article "$".  It seems to trash Gnews whenever it happens.
> 
> I've never had this problem posting from GNEWS.  Maybe it's only on the
> Apollos??

Possibly.  I don't remember getting this until last week, and after that I've
been getting it repeatedly, but only when I post a longish message.

Doing a backtrace, it looks like it comes when the message is sent to the
gnews process, after the "post" command, using send-string.  The entire
message is send-string'd to the process.  Perhaps there is a limit to the
size of the string you can send-string to a process?  The message I tried had
19110 characters.  This could conceivably be system-dependent.

If I turn on nntp-exec-log, it shows the correct entry: ("post " . "340 Enter
news, period on a line by itself to end.")  But I get the error anyway, and
it trashes the gnews process.

-----------------------------------------------------------------------------
Signalling: (file-error "writing to process" "i/O error" #<process gnews>)
  send-string(#<process gnews> "Newsgroups: yale.test\nFrom: ...\n")
  byte-code("..."] 16)
  nntp-exec-post()
  byte-code("..."] 15)
  n-reply-do()
  funcall(n-reply-do)
  byte-code("..."] 20)
  reply(1)
  byte-code("..."] 4)
  reply-send(1)
  byte-code("..."] 3)
  reply-return(1)
* call-interactively(reply-return)
-----------------------------------------------------------------------------

I have deleted the long function arguments for readability.  In
src/process.c, process-send-string (same as send-string) does seem to signal
a "writing to process: i/o error" under some condition, but I don't really
understand enough of the code to tell what's happening.

I tried send-string'ing a huge string (the same 19110-character string,
actually) to my shell process under Emacs, and Emacs hung for quite a while.
When I ^G'd out of it, however, the shell was alive (but hadn't received any
input).

-- Ashwin.

ARPA:    Ram-Ashwin@cs.yale.edu
UUCP:    {decvax,ucbvax,harvard,cmcl2,...}!yale!Ram-Ashwin
BITNET:  Ram@yalecs