From Alain.Picard at memetrics.com  Wed Aug 31 04:11:01 2005
From: Alain.Picard at memetrics.com (Alain.Picard at memetrics.com)
Date: Wed, 31 Aug 2005 14:11:01 +1000
Subject: [Cl-store-devel] bug restoring long lists
Message-ID: <E1EAJwD-0003Vt-AG@imladris>


Dear Maintainers,

Under Lispworks and SBCL, trying to store
and restore a long list fails.  (SBCL can handle
a much, much longer list than LW, but still fails).

Here's what happens:

    XOS> (setq l (loop for i below 10000 collect i)
	       foo 'ok)
    OK
    XOS> (progn (store l "/tmp/list.1") 'ok)
    OK
    XOS> (progn (restore "/tmp/list.1") 'ok)
    ; Evaluation aborted
Raises:
Stack overflow (stack size 16000).
   [Condition of type CONDITIONS:STACK-OVERFLOW]

Restarts:
  0: [ABORT] Abort handling SLIME request.
  1: [ABORT] Quit process.

Backtrace:


    XOS> (let ((*check-for-circs* nil)) (progn (restore "/tmp/list.1") 'ok))
    ; Evaluation aborted
Raises same error.

The problem is non tail recursive manner in which it restores
lists as a nested series of conses.

I have a private patch, which just stores "proper lists" differently,
but I thought you'd want to know and perhaps implement something
better than my hack.

I notice you know have this *CHECK-FOR-CIRCS* variable (I was still
using an old version) --- perhaps when within a context of *CHECK-FOR-CIRCS*
of NIL you can dispatch to an entirely different (and stack-safe) algorithm?

Just a thought.

				--Alain Picard

p.s. please CC me; I'm not on this list (though maybe I should be...)

-- 
It would be difficult to construe        Larry Wall, in  article
this as a feature.			 <1995May29.062427.3640 at netlabs.com>


From sdr at jhb.ucs.co.za  Wed Aug 31 15:12:08 2005
From: sdr at jhb.ucs.co.za (Sean Ross)
Date: 31 Aug 2005 17:12:08 +0200
Subject: [Cl-store-devel] bug restoring long lists
References: <E1EAJwD-0003Vt-AG@imladris>
Message-ID: <87oe7e3zuf.fsf@sdr.ucs.co.za>

Alain.Picard at memetrics.com writes:

Hi Alain,
  thanks for the report, I've also run into this problem 
  a couple of times but wasn't quite too sure what the 
  right approach would be. 
> 
> I have a private patch, which just stores "proper lists" differently,
> but I thought you'd want to know and perhaps implement something
> better than my hack.
I'm not so sure :-)
Right now i'm trying a slightly long winded approach by using list-length 
to figure out what kind of list is passed in (proper, dotted or circular)
which looks something like this.

(defun proper-list-length (list)
  "Returns 2 values the first being the lenght of the list
if it is a proper-list or NIL of it is a circular list.
The second value can be checked to see if LIST is a dotted list."
  (ignore-errors (list-length list)))

(multiple-value-bind (length errorp)
      (proper-list-length list)
    (cond (errorp (dump-atomed-list list stream))
          (length (dump-proper-list list length stream))
          (t (dump-circular-list list stream))))

Each dumper function then takes a more respectable approach
to serializing the list.

Any comments?
What approach were you taking?


Regards,
  Sean.

-- 
"My doctor says that I have a malformed public-duty gland and a
 natural  deficiency in moral fibre," he muttered to himself, "and
 that I am therefore excused from saving Universes."
 - Life, the Universe, and Everything     Douglas Adams.

Confidentiality Notice: http://ucs.co.za/conf.html



From Alain.Picard at memetrics.com  Wed Aug 31 22:36:17 2005
From: Alain.Picard at memetrics.com (Alain.Picard at memetrics.com)
Date: Thu, 1 Sep 2005 08:36:17 +1000
Subject: [Cl-store-devel] bug restoring long lists
In-Reply-To: <87oe7e3zuf.fsf@sdr.ucs.co.za>
References: <E1EAJwD-0003Vt-AG@imladris>
	<87oe7e3zuf.fsf@sdr.ucs.co.za>
Message-ID: <17174.12513.182468.685653@memetrics.com>

Sean Ross writes:
 > Right now i'm trying a slightly long winded approach by using list-length 
 > to figure out what kind of list is passed in (proper, dotted or circular)
 > which looks something like this.
  [SNIP]

That's eerie.  That's almost _exactly_ what I did.  :-)

 > Any comments?

Well, it's the best I could come up with, so I'm obviously starting
to understand your code.  I can send it to you if you wish.

Lastly, I'd comment that.... I find cl-store quite useful.
We use CommonSQL, but it turns out that for a lot of odds and
ends, it's often just easier to store a "regular" lisp object
in a slot with cl-store than to worry about schemas, etc.

Cheers,
			--ap