Re: Qhull test changes

[Brad Barber]

At 03:09 PM 1/30/2012, Rik wrote:
> On 01/30/2012 06:39 AM, Ben
> Abbott wrote:
> >> Rik,
> >>
> >> I pushed a changeset to fix the tests.
> >>
> >>       
>
> http://hg.savannah.gnu.org/hgweb/octave/rev/2633baa831e2
> >>
> >> Ben
> >>
> I now get an error the other direction.  See below.
>
>   ***** testif HAVE_QHULL
>  cube = [0 0 0;1 0 0;1 1 0;0 1 0;0 0 1;1 0 1;1 1 1;0 1 1];
>  [h, v] = convhulln (cube, "Qt");
>  assert (size (h), [12 3]);
>  h = sortrows (sort (h, 2), [1:3]);
>  assert (h, [1 2 4; 1 2 6; 1 4 8; 1 5 6; 1 5 8; 2 3 4; 2 3 7; 2 6 7;
> 3 4 7;
> 4 7 8; 5 6 7; 5 7 8]);
>  assert (v, 1, 10*eps);
>  [h2, v2] = convhulln (cube); % Test defaut option =
> "Qt"
>  assert (size (h2), size (h))
>  h2 = sortrows (sort (h2, 2), [1:3]);
>  assert (h2, h);
>  assert (v2, v, 10*eps);
> !!!!! test failed
> assert (size (h),[12, 3]) expected
>    12    3
> but got
>    6   4
> values do not match
>
> Previously the tests were matched to Qhull <= 2010.  Now the
> tests are
> matched to Qhull >= 2011.  In either case, if Octave is linked
> against the
> other version then a test error results and users are going to think it
> is
> Octave's fault.  This is absolutely up-to-date Octave code from
> Mercurial
> with your changeset, but linked against the default Qhull in Kubuntu
> 10.04
> which is version 2009.
>
> A couple of quick ways to solve this.
>
> 1) Don't check the result.  Check only that h = convhulln (...,
> "Qt")
> matches h2 = convhulln (...) which shows that Octave is passing Qt as
> the
> default option.
>
> 2) Check the size of the result (12 x 3 or 6 x 4) and then compare
> against
> the appropriate array.
>
> 3) Embody the test in configure.ac so that Octave can identify which
> version of Qhull it is linking against.  Then you could use two
> testif
> macros, say HAVE_QHULL or HAVE_QHULL_NEW.
>
> Cheers,
> Rik

Rik,

This is a strange result.   As demonstrated below, Qhull 2003.1
recognized option Qt.  Early versions of qhull would have reported
an error for an unknown option if given Qt.  So I'm fairly confident
that Octave is not passing 'Qt' as the default option to Qhull
2003.1.  That would explain the '6   4' result.

ISPY-/c/backup/qhull/qhull-2003.1-win> rbox c | qhull Qt i
12
6 2 0
4 6 0
5 4 0
1 5 0
5 6 4
6 5 7
2 3 0
3 1 0
6 3 2
3 6 7
3 5 1
5 3 7 

---
Your previously reported result for 2009.1 doesn't agree with my copy of
2009.1.  

# Your result
convhulln (cube, 'Qt FO')
Options selected for Qhull 2009.1 2009/06/14:
  Qtriangulate  _pre-merge  _zero-centrum 
_max-width  1
  Error-roundoff 1.4e-15  _one-merge 9.7e-15  _near-inside
4.9e-14
  Visible-distance 2.8e-15  U-coplanar-distance 2.8e-15 
Width-outside 5.5e-15
  _wide-facet 1.7e-14
ans =

   4   1   2   3
   5   6   2   1
   8   5   1   4
   6   7   3   2
   7   8   4   3
   8   7   6   5

# Using qhull 2009.1
ISPY-/c/download/installers/qhull-2009.1/src> rbox c | qhull Qt FO
i
Options selected for Qhull 2009.1 2009/06/14:
  Qtriangulate  incidence  _pre-merge 
_zero-centrum  _max-width  1
  Error-roundoff 6.9e-016  _one-merge 4.9e-015 
_near-inside 2.4e-014
  Visible-distance 1.4e-015  U-coplanar-distance 1.4e-015
  Width-outside 2.8e-015  _wide-facet 8.3e-015
12
6 2 0
4 6 0
5 4 0
1 5 0
5 6 4
6 5 7
2 3 0
3 1 0
6 3 2
3 6 7
3 5 1
5 3 7

Something odd is going on.

        

        

        

        

        

--Brad