Re: [OctDev] griddata / delaunay

[Jonathan S. <jjs...@sb...>]

David Bateman wrote:
> According to Jonathan Stickel <jjs...@sb...> (on 09/01/04):
> 
>>I have encountered a problem with griddata when using semi-regular input 
>>points.  The issue has to do with facet merging when running delaunay, 
>>via qhull, with defualt options.  To avoid this, delaunay should be run 
>>with the "QJ" option.  Perhaps this should be mentioned explicitly in 
>>the delaunay help.  For the worst case of a rectangular grid, a 
>>segmentation fault occcurs:
>>
>>octave:1> [x,y]=meshgrid(1:10,1:10);
>>octave:2> delaunay(x(:),y(:),"QJ")
>>(plot of triangles)
>>octave:3> delaunay(x(:),y(:))
>>panic: Segmentation fault -- stopping myself...
>>attempting to save variables to `octave-core'...
>>save to `octave-core' complete
>>Segmentation fault
>>
>>Attached is a patch for griddata which calls delaunay with the "QJ" option.
> 
> 
> Not sure if this is the right approach. What is the real reason for the
> seg fault. If it is in the underlying qhull code, it might be better to
> surround the calls to the qhull stuff with
> 
>       BEGIN_INTERRUPT_IMMEDIATELY_IN_FOREIGN_CODE_1;
> 
>       error ("delaunay: Internal qhull error. Please try the 'QJ' option.");
>       octave_throw_interrupt_exception ();
> 
>       BEGIN_INTERRUPT_IMMEDIATELY_IN_FOREIGN_CODE_2;
> 
>       <QHULL CODE HERE>
> 
>       END_INTERRUPT_IMMEDIATELY_IN_FOREIGN_CODE;
> 
> and inform the author of qhull of the problem within their code. At least
> this way octave won't crash for qhulls problems....
> 

I agree.  I do believe that qhull is the real cause of the example 
segfault, and octave functions should catch the problem like you 
suggest.  However, for other semi-regular grids, delaunay does not 
always segfault but will "drop points" when used without the QJ option. 
   I can provide an example if you like. I think this should be 
mentioned in the help for qhull based functions.

Please don't overlook my specific solution for griddata.  Griddata does 
not currently take arguments to pass to qhull.  This could be added, of 
course, but I see no reason not to always use the QJ option for delaunay 
within griddata.

Jonathan