Home / Browse / Steel Bank Common Lisp / Mail / Archive

Steel Bank Common Lisp

Email Archive: sbcl-help (read-only)

[Sbcl-help] Re: [Sbcl-devel] Remove bogus handlers

From: Antonio Martinez <antonio.shotton@gm...> - 2005-12-28 16:26

(CC'd sbcl-help rather than sbcl-devel.)

On 12/28/05, Dmitri Hrapof <yavannadil@...> wrote:
> What is the right way to automatically remove bogus handlers?
> In my server I (ADD-FD-HANDLER <listening-socket> :input
> (PERFORM-WORK)), followed by (ADD-FD-HANDLER <accepted-socket> :input
> (PERFORM-WORK)) to leave slime REPL free. Everything is fine until
> client quits and condition is signalled. Of course I can choose
> REMOVE-THEM restart manually, but is it possible to do it automatically?
> Where do I (HANDLER-BIND)? I beg your pardon, but I'm not very good with
> condition system yet...

I surround (PERFORM-WORK) in a HANDLER-BIND form and use FIND-RESTART
in the handler body, followed by INVOKE-RESTART, something along these
lines:

              (handler-bind
                     ((error (lambda (condition)
                               (invoke-restart
                                (find-restart 'remove-them condition)))))
                   (sb-sys::serve-event period))
                 (when end-of-period-hook
                   (funcall end-of-period-hook)))

(You probably want to check that the 'remove-them symbol has been
imported into your package.)

However, SBCL currently signals an ERROR for bogus handlers, so you'd
probably want to do more condition processing to discriminate real
client-closed-connection conditions from, say, disk-full conditions.=20
(Perhaps depending on whether the restart exists or not?)

I submitted a patch a while back which defined a condition for this
situation, but, ah, it breaks the build because CLOS isn't up when
this file is loading, so it only works when patching a running system.
 I haven't had time to look into it, sigh.

Hope this helps,

--Tony

Thread View

Thread	Author	Date
[Sbcl-help] Re: [Sbcl-devel] Remove bogus handlers	Antonio Martinez <antonio.shotton@gm...>

[Sbcl-help] Re: [Sbcl-devel] Remove bogus handlers

From: Dmitri Hrapof <yavannadil@ya...> - 2005-12-30 11:21

Thank you for your answer!

Antonio Martinez wrote:

>I surround (PERFORM-WORK) in a HANDLER-BIND form and use FIND-RESTART
>in the handler body, followed by INVOKE-RESTART, something along these
>lines:
>
>              (handler-bind
>                     ((error (lambda (condition)
>                               (invoke-restart
>                                (find-restart 'remove-them condition)))))
>                   (sb-sys::serve-event period))
>                 (when end-of-period-hook
>                   (funcall end-of-period-hook)))
>  
>
I did the following
 (when *slime-friendly*
                (handler-bind
                  ((error (lambda (condition)
                      (invoke-restart (find-restart 'remove-them 
condition)))))
                 (sb-sys:add-fd-handler
                  (sb-bsd-sockets:socket-file-descriptor k)
                  :input
                   (lambda (fd)
                      (declare (ignore fd))
                       (loop while (op:work_pending *the-orb*)
                                     do (op:perform_work *the-orb*))))))

and it brought no relief.
Error was signalled, and backtrace was:

0: (SB-IMPL::HANDLER-DESCRIPTORS-ERROR)
1: (SB-IMPL::SUB-SERVE-EVENT NIL 0)
2: (SB-SYS:WAIT-UNTIL-FD-USABLE 0 :INPUT NIL)

(op:perform_work) doesn't call SB-SYS:SERVE-EVENT...
Is there any way to bind handlers for SB-IMPL::SUB-SERVE-EVENT?
Or around REPL, so to say? Kind of "global" HANDLER-BIND?

Thanks,
Dmitri

[Sbcl-help] Re: [Sbcl-devel] Remove bogus handlers

From: Antonio Martinez <antonio.shotton@gm...> - 2005-12-30 13:27

On 12/30/05, Dmitri Hrapof <yavannadil@...> wrote:
> I did the following
>  (when *slime-friendly*
>                 (handler-bind
>                   ((error (lambda (condition)
>                       (invoke-restart (find-restart 'remove-them
> condition)))))
>                  (sb-sys:add-fd-handler
>                   (sb-bsd-sockets:socket-file-descriptor k)
>                   :input
>                    (lambda (fd)
>                       (declare (ignore fd))
>                        (loop while (op:work_pending *the-orb*)
>                                      do (op:perform_work *the-orb*))))))
>
> and it brought no relief.
> Error was signalled, and backtrace was:
>
> 0: (SB-IMPL::HANDLER-DESCRIPTORS-ERROR)
> 1: (SB-IMPL::SUB-SERVE-EVENT NIL 0)
> 2: (SB-SYS:WAIT-UNTIL-FD-USABLE 0 :INPUT NIL)
>
> (op:perform_work) doesn't call SB-SYS:SERVE-EVENT...
> Is there any way to bind handlers for SB-IMPL::SUB-SERVE-EVENT?
> Or around REPL, so to say? Kind of "global" HANDLER-BIND?

Hi Dmitri,

Ah, right.  I don't know of any /portable/ way to define a global
HANDLER-BIND for a particular condition as you suggest.  In SBCL,
you'd have to hack src/code/toplevel.lisp.  To work around this,
instead of working in the REPL directly, I usually have a global entry
point to the server, and have the global HANDLER-BIND be one of the
first few forms, something like this:

(defun server ()
  (handler-bind (error ...)
     server-body ...))

SERVER-BODY manages the fd-handlers (ensuring no active fd-handler on
exit etc.) and calls SERVE-EVENT.  Would this approach be
unsatisfactory?  Do you need to work in the REPL?  (When working in
the REPL directly, I find the fd conditions more reminder than
irritant.)

Cheers,

--Tony

[Sbcl-help] Re: [Sbcl-devel] Remove bogus handlers

From: Dmitri Hrapof <yavannadil@ya...> - 2005-12-30 14:26

Antonio Martinez wrote:

>instead of working in the REPL directly, I usually have a global entry
>point to the server, and have the global HANDLER-BIND be one of the
>first few forms, something like this:
>
>(defun server ()
>  (handler-bind (error ...)
>     server-body ...))
>
>SERVER-BODY manages the fd-handlers (ensuring no active fd-handler on
>exit etc.) and calls SERVE-EVENT.  Would this approach be
>unsatisfactory?  Do you need to work in the REPL?  (When working in
>the REPL directly, I find the fd conditions more reminder than
>irritant.)
>
Well, the server has idiomatic entry point, OP:RUN, but it blocks the 
REPL, so development is literally interrupted with Control-C's...
Anyway, it seems I've solved my problem by redefining 
SB-IMPL::HANDLER-DESCRIPTORS-ERROR so it always "chooses" REMOVE-THEM 
restart:

(defun make-repl-happy ()
    (handler-bind
       ((error #'(lambda (c)
           (declare (ignore c))
           (invoke-restart (find-restart 'continue)))))
     (setf *repl-friendly* t
       (fdefinition 'sb-impl::handler-descriptors-error)
       #'(lambda ()
           (let ((bogus-handlers nil))
         (dolist (handler sb-impl::*descriptor-handlers*)
           (unless (or (sb-impl::handler-bogus handler)
                   (sb-unix:unix-fstat
                (sb-impl::handler-descriptor handler)))
             (setf (sb-impl::handler-bogus handler) t)
             (push handler bogus-handlers)))
         (setf sb-impl::*descriptor-handlers*
               (delete-if #'sb-impl::handler-bogus
                  sb-impl::*descriptor-handlers*)))))))

Good holidays,
Dmitri