[slime-devel] Extra Swank threads after slime-disconnect

[Alan Caulkins]

Hello list,

To start, I'm new to the list, and I'd like to congratulate everyone on
creating such a wonderful piece of software. I've been continually
impressed by SLIME ever since I started keeping up with it a few months
ago. However, I think I've found a bug in Swank's spawn connection-style.
I've been connecting to an already-running sbcl (on Linux) as per the
instructions found on on Bill Clemenson's weblog, linked from CLiki's
SLIME-HOWTO page.

fatman at hubris:~$ sbcl --load slime.lisp
This is SBCL 0.8.13, an implementation of ANSI Common Lisp.
More information about SBCL is available at <http://www.sbcl.org/>.
.
<snip>
.
;; Swank started at port: 4005.
*

My slime.lisp is simple:
(load "lib/emacs/slime/swank-loader.lisp")
(swank::create-swank-server 4005 :spawn #'swank::simple-announce-function t)

Before connecting to Swank for the first time, there are 2 sbcl threads
running:

fatman at hubris:~$ ps -e | grep sbcl
 9831 pts/0    00:00:04 sbcl
 9832 pts/0    00:00:00 sbcl

After running (slime-connect "localhost" 4005), we also see the new
reader, control, and repl threads:

fatman at hubris:~$ ps -e | grep sbcl
 9831 pts/0    00:00:04 sbcl
 9832 pts/0    00:00:00 sbcl
 9844 pts/0    00:00:00 sbcl
 9845 pts/0    00:00:00 sbcl
 9846 pts/0    00:00:00 sbcl

CL-USER> (in-package swank)
#<PACKAGE "SWANK">
SWANK> (mapcar (lambda (c) (list (connection.control-thread c)
				 (connection.reader-thread c)
				 (connection.repl-thread c)))
	       *connections*)
((9844 9845 9846))


But, after running (slime-disconnect), only the reader thread has
terminated:

fatman at hubris:~$ ps -e | grep sbcl
 9831 pts/0    00:00:04 sbcl
 9832 pts/0    00:00:00 sbcl
 9844 pts/0    00:00:00 sbcl
 9846 pts/0    00:00:00 sbcl

With each successive connection and disconnection, two more threads are
added. I suspect that's not the intended behavior. :)

First of all, does this happen to anyone else? I'm using the laterst
release of SBCL and the 1.0 Alpha release of SLIME, but some of my system
libraries are pathetically out of date. I wouldn't be surprised to learn
that my own system is the problem.

Secondly, assuming this is a bug, I have come up with a fix. I simply
defined a function for the :cleanup slot in the connection struct. My fix
is decidedly simple-minded - I just kill the other two threads. It seems
to work, but I have no idea if those threads were left alive for some
reason, or if there is a better way to get them to terminate. Also, I'm
still very new to the Lisp game, and although I've tried to honor your
indirection and abstractions, I have no idea if I succeeded.

-Alan

PS- I tried to make the following diff from the current CVS instead of
1.0alpha, but I got these kind of errors on the sbcl console when the
connection was closed:

debugger invoked on a UNDEFINED-FUNCTION in thread 10396:
  The function :DEBUG is undefined.


			Linux: The ultimate video game.


fatman at hubris:~/lib/emacs$ diff slime.alan/swank.lisp slime-1.0alpha/swank.lisp
488,491d487
< (defun cleanup-connection-threads (connection)
<   (kill-thread (connection.control-thread connection))
<   (kill-thread (connection.repl-thread connection)))
<
599,600c595
<                      :serve-requests #'spawn-threads-for-connection
<                        :cleanup #'cleanup-connection-threads))
---
>                      :serve-requests #'spawn-threads-for-connection))