CGAL users discussion list

[Marc Glisse <>]

On Mon, 10 Mar 2014, Muhammad Abdelghaffar wrote:
> On Sun, Mar 9, 2014 at 1:10 AM, Marc Glisse 
> <>
>  wrote:
>       On Sat, 8 Mar 2014, Muhammad Abdelghaffar wrote:
>
>             I'm trying to construct the delaunay graph for a
>             point set in $d=6$ using
>             the CGAL dD Convex Hulls and Delaunay Triangulations
>             package , but I'm
>             getting a stack overflow error.
>
>
> Did you try increasing the stack size? In case it might be related to:
> http://stackoverflow.com/q/17404432/1918193
>
>
> Ok. I've tried to increase the stack reserve/commit size up to 1.3 GB, the
> overflow now happens at a later stage, but still happens.

1.3G is getting big indeed... On linux your program works with a stack of 
130M (didn't try anything between 8M (not enough) and 130M) and creates a 
2G output file. Did you compile as 64 bits?

> The Task Manager shows that the code is only using ~700 MB, shouldn't 
> this value be the one I've set ?   

I don't use that platform so I can't tell. Maybe it only shows the heap 
size? Maybe it refused to set the stack size to your value?

>       Also, could you show what the call stack looks like at the time
>       of the overflow?
>
> Please find the attachment.

That all it shows? It isn't being very helpful... Ideally it would show 
that we currently have visibility_search calling itself recursively with a 
depth of thousands, like it does on linux.

> I have another question, is this package surpasses Qhull's performance in
> terms of speed and memory ?

Delaunay_2 and Delaunay_3 are better. I don't have any experience with 
Delaunay_d, see if there are performance statements in the doc...

The num_of_unbounded_simplices issue is strange, the number becomes 
negative with just a few vertices. Good thing is, that number doesn't seem 
to be used for anything, so unless you have other issues, you can just 
ignore it.

--
Marc Glisse