From bug-octave-request at bevo dot che dot wisc dot edu  Thu Feb 18 04:45:10 1999
Subject: Unidentified subject!
From: Rolf Fabian <fabian at TU-Cottbus dot De>
To: bug-octave at bevo dot che dot wisc dot edu
Date: Thu, 18 Feb 1999 04:45:08 -0600 (CST)

Hi,

incorrect transposition operator " ' "
instead of correct " .' " for complex input
to script files (already reported 990216 for
'polyfit') can also be found in scripts
'dot.m' and 'cross.m'.

EXAMPLE to reproduce the bugs:
:)x=1+[1,1,-1]*i; y=1+[1,1,1]*i;
##############################
# (A) preference for columns #
##############################
:)prefer_column_vectors = 1;
:)dot(x,y),dot(x.',y),dot(x.',y.'),dot(x,y.')
	ans = 4 - 2i
	ans = 2 + 4i
	ans = 4 - 2i
	ans = 2 + 4i
#REM:	this is obviously buggy
#	we get 2 different results depending only on
#	row/column orientation of input
:)cross(x,y),cross(x.',y),cross(x.',y.'),cross(x,y.')
ans =
  -2 - 2i   2 + 2i   0 + 0i	#BUG: WHY DO WE GET A ROW AT COLUMN-PREFERENCE?
ans =
  -2 + 2i			#BUG: CONJUGATE COMPLEX TO 1st ANSWER ABOVE
   2 - 2i
   0 + 0i
ans =
  -2 + 2i
   2 - 2i
   0 + 0i
ans =
  -2 + 2i
   2 - 2i
   0 + 0i
#################################
# (B) no preference for columns #
#################################
:)prefer_column_vectors = 0;
:)cross(x,y),cross(x.',y),cross(x.',y.'),cross(x,y.')
ans =
  -2 - 2i   2 + 2i   0 + 0i
ans =
  -2 - 2i   2 + 2i   0 + 0i
ans =
  -2 + 2i			#BUG: WHY DO WE GET A COLUMN AT ROW-PREFERENCE?
   2 - 2i			#BUG: CONJUGATE COMPLEX TO OTHER ANSWERS
   0 + 0i
ans =
  -2 - 2i   2 + 2i   0 + 0i

#######################################
#FIX -> see comments in scripts below #
#######################################
#application of these fixes result in unique correct answers to our example
:)dot(x,y),dot(x.',y),dot(x.',y.'),dot(x,y.')
ans = 2 + 4i
ans = 2 + 4i
ans = 2 + 4i
ans = 2 + 4i

:)prefer_column_vectors=0				#we expect rows!
:)cross(x,y),cross(x.',y),cross(x.',y.'),cross(x,y.')
ans =  -2 + 2i   2 - 2i   0 + 0i			#ROWS ok
ans =  -2 + 2i   2 - 2i   0 + 0i
ans =  -2 + 2i   2 - 2i   0 + 0i
ans =  -2 + 2i   2 - 2i   0 + 0i
:)prefer_column_vectors=1;				#we expect columns!
:)cross(x,y),cross(x.',y),cross(x.',y.'),cross(x,y.')
ans =
  -2 + 2i						#COLUMNS ok
   2 - 2i
   0 + 0i
ans =
  -2 + 2i
   2 - 2i
   0 + 0i
ans =
  -2 + 2i
   2 - 2i
   0 + 0i
ans =
  -2 + 2i
   2 - 2i
   0 + 0i

I suspect even more script to be candidates for the subtle bug inferred
by an incorrect usage of " ' " instead of ".' " -operator if
only a change in orientation is requested. I assume, authors tend
to check their procedures only at real-type input so that this type of
bugs are not detected even for a long time,
e.g. 'cross' dates back to 1994 !

Wouldn't it be a good idea to implement more complex-input-type tests
even for basic routines into the test-suite?

Furthermore, I suggest to give an easy-to-find hint in the manual,
that " ' " is *NOT* the appropriate operator for simple tranposition
operation only.

Bye

Rolf Fabian 990217
Atmospheric Chemistry and Air Quality Dep.
Technical University of Cottbus / FRG


##------------------------------------------
## usage:  dot (x, y)
##
## Computes the dot product of two vectors.

## Author: jwe
## Debug : fabian at tu-cottbus dot de 990217

function z = dot (x, y)
  
  if (nargin != 2)
    usage ("dot (x, y)");
  endif

  if (is_vector (x) && is_vector (y) && length (x) == length (y))
    [x_nr, x_nc] = size (x);
    [y_nr, y_nc] = size (y);
    if (x_nr == 1)
      if (y_nr == 1)
	#z = x * y';	#original line BUGGY for complex y, correct:
	 z = x*y.';
      else
	z = x * y;
      endif
    else
      if (y_nr == 1)
	z = y * x;
      else
	#z = y' * x;	#original line BUGGY for complex y, correct:
	 z = y.'* x;
      endif
    endif
  else
    error ("dot: both arguments must be vectors of the same length");
  endif

endfunction


##------------------------------------------
## usage:  cross (x, y)
##
## Computes the vector cross product of the two 3-dimensional vectors
## x and y.

## Author: KH <Kurt dot Hornik at ci dot tuwien dot ac dot at>
## Created: 15 October 1994
## Adapted-By: jwe
## Debug     : fabian at tu-cottbus dot de 990217

function z = cross (x, y)
  
  if (nargin != 2)
    usage ("cross (x, y)");
  endif

  if (length (x) == 3 && length (y) == 3)

    z = [x(2)*y(3) - x(3)*y(2); x(3)*y(1) - x(1)*y(3); x(1)*y(2) - x(2)*y(1)];

#**---- REMOVE WHOLE BLOCK OF ORIGINAL LINES, THEY LEAD TO THE 'ORIENTATION BUG'
#**  x_nr = rows (x);
#**  y_nr = rows (y);
#**
#**  if ((x_nr == y_nr && x_nr == 1)
#**		|| (x_nr != y_nr && ! prefer_column_vectors))
#**---- REMOVE WHOLE BLOCK OF ORIGINAL LINES

   if(!prefer_column_vectors))	#FIX fabian at tu-cottbus dot de (MUCH MORE SIMPLE!)
	#z = z';		#BUGGY original line for complex vectors x,y 
       	 z = z.';		#FIX fabian at tu-cottbus dot de
    endif

  else
    error ("cross: both x and y must be 3-dimensional vectors");
  endif

endfunction