Re: Qhull test changes

[Rik]

On 01/30/2012 02:54 PM, Ben Abbott wrote:
> On Jan 30, 2012, at 3:09 PM, 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
> I'm not certain, but doesn't "Qt" imply that the convex hull should be made 
> up of triangles ? (perhaps I should study the qhull docs a bit ?)
>
> In any event, I favored the more recent qhull because it matches Matlab's 
> result.
>
Yes, the output should be triangulated when we pass the 'Qt' option and the
new post-2011 Qhull behavior is mathematically correct.  The problem is
that Qhull is not returning triangulated output for versions less than 2011
and users will blame Octave when they see a failing test in the test
report.  I am proposing that Octave work around the different Qhull
versions so we don't generate a lot of spurious bug reports.

On the other hand, if we want we could leave the test in and also put in
some comments that specifically say, "If you see this test failing, then
you must upgrade your Qhull installation."  This might do a bit towards
pushing users and distributions to upgrade to a new Qhull.

--Rik