CGAL users discussion list

[Marc Glisse <>]

On Sun, 29 Jan 2012, Bernhard Kornberger wrote:

> Thank you for the valuable hint regarding profiling. However,
> multithreading still doesn't work reliably. So I have one more
> question. Let's create segments like this:
>
> ***
> #define CGAL_HAS_THREADS
> typedef CGAL::Exact_predicates_exact_constructions_kernel K;
>
> K::Point_3 a(0,0,0);
> K::Point_3 b(1,0,0);
> K::Point_3 c(0,1,0);
>
> K::Segment_3 s0(a,b);
> K::Segment_3 s1(a,c);
> ***
>
> Are s0 and s1 independent so that they can be treated by different
> threads? Or does the reference counting of Gmpq some magic behind
> the scenes that links the segments to the same point?

Note that I haven't looked in details, this is just a quick guess.
What you get in a, b and c are lazy point constructors.
What you get in s0 and s1 are lazy segment constructors. They both keep a 
reference to 'a' lying around, which at some point they may even modify 
through update_exact. So I would say the objects are not independent.

Maybe something like Simple_cartesian<mpq_class>, possibly with some 
filtering on top for speed, would be happy with this code. Otherwise, you 
may have to rewrite a bit to make sure the objects really are independent.

--
Marc Glisse