Scoring out of bounds

Scoring out of bounds

[Norbert Koch]

Hi!

I just stumbled over the following error

Signaling: (range-error "floor" 992188263.1)
  floor(992188263.1)
  gnus-decay-score(1044408698)
  gnus-decay-scores(((touched nil) ("thread" ("<011120031152206042%mmanti@mac.com>" 1044408698 731520 s) ("<87r80s83r0.fsf@bird.agharta.de>" 1044408698 731520 s) .... (decay 731520)) 731520)
  gnus-score-load-file("/usr/users/nk/.xemacs/gnus/.scores/comp.lang.lisp.ADAPT")
  gnus-score-load-files(("/usr/users/nk/.xemacs/gnus/.scores/comp.lang.lisp.ADAPT" "/usr/users/nk/.xemacs/gnus/.scores/comp.lang.lisp.score" "/usr/users/nk/.xemacs/gnus/.scores/all.score"))
  gnus-score-headers(("/usr/users/nk/.xemacs/gnus/.scores/comp.lang.lisp.ADAPT" "/usr/users/nk/.xemacs/gnus/.scores/comp.lang.lisp.score" "/usr/users/nk/.xemacs/gnus/.scores/all.score") nil)
  gnus-possibly-score-headers()
  gnus-summary-read-group-1("comp.lang.lisp" nil t nil nil nil)
  gnus-summary-read-group("comp.lang.lisp" nil t nil nil nil nil)
  gnus-group-read-group(nil t)
  gnus-group-select-group(nil)
  gnus-topic-select-group(nil)
  call-interactively(gnus-topic-select-group)

Something's pushing scores up too high (my bad), but maybe it should
be bound in gnus-decay-score to prevent such errors?

norbert.

Re: Scoring out of bounds

[Norbert Koch]

Norbert Koch <viteno@xemacs.org> writes:

> Something's pushing scores up too high (my bad), but maybe it should
> be bound in gnus-decay-score to prevent such errors?

Here's a possible fix

Index: gnus-score.el
===================================================================
RCS file: /usr/local/cvsroot/gnus/lisp/gnus-score.el,v
retrieving revision 6.35
diff -u -u -r6.35 gnus-score.el
--- gnus-score.el	13 Aug 2003 16:34:46 -0000	6.35
+++ gnus-score.el	2 Nov 2003 12:14:46 -0000
@@ -2918,13 +2918,15 @@
 
 (defun gnus-decay-score (score)
   "Decay SCORE according to `gnus-score-decay-constant' and `gnus-score-decay-scale'."
-  (floor
-   (- score
-      (* (if (< score 0) -1 1)
-	 (min (abs score)
-	      (max gnus-score-decay-constant
-		   (* (abs score)
-		      gnus-score-decay-scale)))))))
+  (condition-case nil
+      (floor
+       (- score
+	  (* (if (< score 0) -1 1)
+	     (min (abs score)
+		  (max gnus-score-decay-constant
+		       (* (abs score)
+			  gnus-score-decay-scale))))))
+    (arith-error most-positive-fixnum)))
 
 (defun gnus-decay-scores (alist day)
   "Decay non-permanent scores in ALIST."

Re: Scoring out of bounds

[Katsumi Yamaoka]

>>>>> In <vz3cd7c6bu.fsf@redqueen.bytechase.cx>
>>>>>	Norbert Koch <viteno@xemacs.org> wrote:

>> Something's pushing scores up too high (my bad), but maybe it should
>> be bound in gnus-decay-score to prevent such errors?

> Here's a possible fix

> --- gnus-score.el	13 Aug 2003 16:34:46 -0000	6.35
> +++ gnus-score.el	2 Nov 2003 12:14:46 -0000
> @@ -2918,13 +2918,15 @@

>  (defun gnus-decay-score (score)
>    "Decay SCORE according to `gnus-score-decay-constant' and `gnus-score-decay-scale'."
> -  (floor
> -   (- score
> -      (* (if (< score 0) -1 1)
> -	 (min (abs score)
> -	      (max gnus-score-decay-constant
> -		   (* (abs score)
> -		      gnus-score-decay-scale)))))))
> +  (condition-case nil
> +      (floor
> +       (- score
> +	  (* (if (< score 0) -1 1)
> +	     (min (abs score)
> +		  (max gnus-score-decay-constant
> +		       (* (abs score)
> +			  gnus-score-decay-scale))))))
> +    (arith-error most-positive-fixnum)))

I've committed it with a slight modification[1].  Thanks.
BTW, isn't it an XEmacs bug?  Emacs's floor can handle the value
of an argument that it returns the maximum integer:

(floor (+ (lsh -1 -1) 0.999))
 => 134217727

[1] The most-positive-fixnum variable is not always available in
21.3 or older versions of Emacs since it is defined in cl.el.
-- 
Katsumi Yamaoka <yamaoka@jpl.org>

Re: Scoring out of bounds

[Norbert Koch]

Katsumi Yamaoka <yamaoka@jpl.org> writes:

> I've committed it with a slight modification[1].  Thanks.

Thanks for applying it.

> BTW, isn't it an XEmacs bug?  Emacs's floor can handle the value
> of an argument that it returns the maximum integer:
>
> (floor (+ (lsh -1 -1) 0.999))
>  => 134217727

Ah, I've always shifted in the other direction :-)

I'm not sure whether to call this behaviour a bug.  If an operation
returns an out of bound value, you've always got two possibilities:
flag an error or silently convert it to the maximum allowed error.

norbert.

Re: Scoring out of bounds

[Katsumi Yamaoka]

>>>>> In <vz65i0smwp.fsf@redqueen.bytechase.cx>
>>>>>	Norbert Koch <viteno@xemacs.org> wrote:

> Katsumi Yamaoka <yamaoka@jpl.org> writes:

>> (floor (+ (lsh -1 -1) 0.999))
>>  => 134217727

> Ah, I've always shifted in the other direction :-)

> I'm not sure whether to call this behaviour a bug.  If an operation
> returns an out of bound value, you've always got two possibilities:
> flag an error or silently convert it to the maximum allowed error.

I don't intend to complain to the XEmacs development team
because it is not an essential problem when the score value
grows unusually large. ;-)

I was simply surprised that XEmacs' floor (and other arithmetic
functions) cannot handle a floating point number over the half
of the maximum integer:

(floor (+ (/ most-positive-fixnum 2.0) 0.499))
 => 536870911
(floor (+ (/ most-positive-fixnum 2.0) 0.500))
 => ...
-- 
Katsumi Yamaoka <yamaoka@jpl.org>

Re: Scoring out of bounds

[Katsumi Yamaoka]

>>>>> In <b9yekwohj9a.fsf@jpl.org> Katsumi Yamaoka wrote:

>>>>>> In <vz3cd7c6bu.fsf@redqueen.bytechase.cx>
>>>>>>	Norbert Koch <viteno@xemacs.org> wrote:

>>> Something's pushing scores up too high (my bad), but maybe it should
>>> be bound in gnus-decay-score to prevent such errors?

>> Here's a possible fix

>>  (defun gnus-decay-score (score)
>>    "Decay SCORE according to `gnus-score-decay-constant' and `gnus-score-decay-scale'."

[...]

>> +  (condition-case nil
>> +      (floor
>> +       (- score
>> +	  (* (if (< score 0) -1 1)
>> +	     (min (abs score)
>> +		  (max gnus-score-decay-constant
>> +		       (* (abs score)
>> +			  gnus-score-decay-scale))))))
>> +    (arith-error most-positive-fixnum)))

> I've committed it with a slight modification[1].  Thanks.

I noticed that solution does not decay a score value eternally,
and replaced it in CVS with the following definition:

(defun gnus-decay-score (score)
  "Decay SCORE according to `gnus-score-decay-constant' and `gnus-score-decay-scale'."
  (let ((n (- score
	      (* (if (< score 0) -1 1)
		 (min (abs score)
		      (max gnus-score-decay-constant
			   (* (abs score)
			      gnus-score-decay-scale)))))))
    (if (and (featurep 'xemacs)
	     ;; XEmacs' floor can handle only the floating point
	     ;; number below the half of the maximum integer.
	     (> (abs n) (/ (lsh -1 -2))))
	(string-to-number
	 (car (split-string (number-to-string n) "\\.")))
      (floor n))))
-- 
Katsumi Yamaoka <yamaoka@jpl.org>

Re: Scoring out of bounds

[Norbert Koch]

Katsumi Yamaoka <yamaoka@jpl.org> writes:

> I noticed that solution does not decay a score value eternally,
> and replaced it in CVS with the following definition:

Yes, a good thing to think of :)

Thanks,
norbert.