[Maxima-bugs] [ maxima-Bugs-1692651 ] MGRP gets sense of comparison backwards

[SourceForge.net <no...@so...>]

Bugs item #1692651, was opened at 2007-04-01 23:03
Message generated for change (Comment added) made by willisbl
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=104933&aid=1692651&group_id=4933

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Lisp Core
Group: None
Status: Open
Resolution: None
Priority: 7
Private: No
Submitted By: Robert Dodier (robert_dodier)
Assigned to: Barton Willis (willisbl)
Summary: MGRP gets sense of comparison backwards

Initial Comment:
The function MGRP (greater-than comparison) can get the sense of the comparison backwards. MGRP is called indirectly from MEVALP and maybe other places as well.

e.g.

?mgrp (x, 1); => x - 1 > 0  (OK)

?mgrp (1, x); => x - 1 > 0  (OOPS)

I'm pretty sure the cause of this is the use of special variables in src/compar.lisp. I think what's happening is that SIGN1 calls itself, clobbering some of the specials. I believe the right way to fix this is to rework the comparison code to use only lexical variables instead of specials. 

For the record:

(%i1) :lisp (trace sign1)
(SIGN1)
(%i1) ?mgrp (x, 1);
  1> (SIGN1 ((MPLUS SIMP) -1 $X))
  <1 (SIGN1 $PNZ)
(%o1)                       x - 1 > 0
(%i2) ?mgrp (1, x);
  1> (SIGN1 ((MPLUS SIMP) 1 ((MTIMES SIMP) -1 $X)))
    2> (SIGN1 -1)
    <2 (SIGN1 $NEG)
    2> (SIGN1 $X)
    <2 (SIGN1 $PNZ)
    2> (SIGN1 -1)
    <2 (SIGN1 $NEG)
    2> (SIGN1 ((MPLUS SIMP IRREDUCIBLE) -1 $X))
    <2 (SIGN1 $PNZ)
  <1 (SIGN1 $PNZ)
(%o2)                       x - 1 > 0




----------------------------------------------------------------------

>Comment By: Barton Willis (willisbl)
Date: 2007-04-13 06:11

Message:
Logged In: YES 
user_id=895922
Originator: NO

The calls to infsimp aren't needed -- the function sign1
does the infsimp. When 'memq' is restored, I'll patch
mgrp and mgqp.  




----------------------------------------------------------------------

Comment By: Robert Dodier (robert_dodier)
Date: 2007-04-07 12:40

Message:
Logged In: YES 
user_id=501686
Originator: YES

Barton, thanks for investigating. Please go ahead and commit the new code.
I don't understand exactly what is the purpose of MGRP (it seems like it is
one of several functions to make comparisons and I don't see how the they
all fit together) but I am more than willing to trust your judgement on
this one.

----------------------------------------------------------------------

Comment By: Barton Willis (willisbl)
Date: 2007-04-05 16:35

Message:
Logged In: YES 
user_id=895922
Originator: NO

I see that memq is no longer in Maxima. Replacing memq with a call to
member, and appending an infsimp (may or may not be a good idea, I don't
know), I built Maxima CVS using XP and sbcl 1.02. I passes the test suite.
The code:

(defun mgrp (a b)
  (setq a (infsimp (sub a b)))
  (let ((sgn (csign a)))
    (cond ((eq sgn '$pos) t)
	  ((eq sgn t) nil) ;; csign thinks a - b isn't real
	  ((member sgn '($neg $zero $nz) :test #'eq) nil)
	  (t `((mgreaterp) ,a 0)))))

(defun mgqp (a b)
  (setq a (infsimp (sub a b)))
  (let ((sgn (csign a)))
    (cond ((member sgn '($pos $zero $pz) :test #'eq) t)
	  ((eq sgn t) nil) ;; csign thinks a - b isn't real 
	  ((eq sgn '$neg) nil)
	  (t `((mgeqp) ,a 0)))))



----------------------------------------------------------------------

Comment By: Barton Willis (willisbl)
Date: 2007-04-05 06:16

Message:
Logged In: YES 
user_id=895922
Originator: NO

An attempt at a fix:

(defun mgrp (a b)
  (setq a (sub a b))
  (let ((sgn (csign a)))
    (cond ((eq sgn '$pos) t)
	  ((eq sgn t) nil) ;; csign thinks a - b isn't real
	  ((memq sgn '($neg $zero $nz)) nil)
	  (t `((mgreaterp) ,a 0)))))

(defun mgqp (a b)
  (setq a (sub a b))
  (let ((sgn (csign a)))
    (cond ((memq sgn '($pos $zero $pz)) t)
	  ((eq sgn t) nil) ;; csign thinks a - b isn't real 
	  ((eq sgn '$neg) nil)
	  (t `((mgeqp) ,a 0)))))

Notes:

(1) With these functions, the test suite gives an error break on rtest15;
    rtest16 #39 reports a problem about the sign of infinity; and rtestsum
#5
    reports that the result is correct, but that Maxima expected the
result to
    be wrong.

(2) The functions c-$pos and friends may have some logic in them that I've
overlooked.

(3) The subtraction in my functions can be a problem for inf, minf, und,
and ...

(4) It's possible that my functions don't properly set the values of the
'evens',
    'odds', and 'sign'. This scheme confuses me. I would like to expunge
the functions
    c-$pos and friends along with  'evens', 'odds', and 'sign'.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=104933&aid=1692651&group_id=4933