Bug in nntp-open-ssl-stream

Bug in nntp-open-ssl-stream

[Bill Perry]

There is a bug in nntp-open-ssl-stream in v5.8.3:

(defun nntp-open-ssl-stream (buffer)
  (let* ((ssl-program-arguments '("-connect" (concat host ":" service)))
	 (proc (open-ssl-stream "nntpd" buffer nntp-address
nntp-port-number)))
    (save-excursion
      (set-buffer buffer)
      (nntp-wait-for-string "^\r*20[01]")
      (beginning-of-line)
      (delete-region (point-min) (point))
      proc)))

This should NOT modify ssl-program-arguments, otherwise the latest version
of ssl.el (that uses OpenSSL by default) will fail to work, because the
's_client' parameter has been removed.

Also, the waiting for the string terminator should be removed, and we
should document that if people get garbage at the top of the buffer (the
-quiet switch I added ages ago to ssleay has been broken in openssl), they
should create a shell script like:

#!/bin/sh
exec 2> /dev/null
exec openssl "$@"

I was thinking of distributing an 'emacs.openssl' script or something like
that as part of ssl.el and making it use it by default.  What would be
ideal is if someone outside the united states would volunteer to
re-implement my SSL patches to Emacs from ages ago.

-Bill P.

Re: Bug in nntp-open-ssl-stream

[Lars Magne Ingebrigtsen]

Bill Perry <wmperry@aventail.com> writes:

> (defun nntp-open-ssl-stream (buffer)
>   (let* ((ssl-program-arguments '("-connect" (concat host ":" service)))
> 	 (proc (open-ssl-stream "nntpd" buffer nntp-address
> nntp-port-number)))
>     (save-excursion
>       (set-buffer buffer)
>       (nntp-wait-for-string "^\r*20[01]")
>       (beginning-of-line)
>       (delete-region (point-min) (point))
>       proc)))
> 
> This should NOT modify ssl-program-arguments, otherwise the latest version
> of ssl.el (that uses OpenSSL by default) will fail to work, because the
> 's_client' parameter has been removed.

Ok.  So the function shouldn't bind ssl-program-arguments at all?

> I was thinking of distributing an 'emacs.openssl' script or something like
> that as part of ssl.el and making it use it by default.  What would be
> ideal is if someone outside the united states would volunteer to
> re-implement my SSL patches to Emacs from ages ago.

Haven't the restrictions on crypty exports from the US largely been
lifted, or am I misinformed?

-- 
(domestic pets only, the antidote for overdose, milk.)
   larsi@gnus.org * Lars Magne Ingebrigtsen