Steel Bank Common Lisp

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

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