CGAL users discussion list

[Marc Glisse <>]

On Thu, 18 Sep 2014, stfu wrote:

> we have an application where we need to determine the vertices of a convex
> hull in moderate dimensions (d<=50) and tried the Convex_hull_d code from
> CGAL.
>
> The following program unfortunately crashes with a segmentation fault, e.g.
> when trying to insert 1000 points in 10 dimensions after insertion of about
> 40 points:
[snip]
> By employing an exact kernel we should not experience robustness problems,
> right?
> Or is there a really stupid mistake in our above code?
>
> BTW for small dimensions (e.g. d=3) it seems more stable and does not crash
> as quickly. Tested on cgal 4.2 as included with Ubuntu 13.10 and 14.04 on an
> intel x86 system (64-bit).
>
> I hope we did not make a too obvious mistake here ...

I did not try your example, but past experience is that this code involves 
a recursion that becomes quite deep when the dimension is large. Could you 
try increasing significantly the size of the stack?

--
Marc Glisse