CGAL users discussion list

[Sylvain Pion <>]

Alejandro Aragon a écrit :
> Indeed that made the difference, using simple cartesian gave:
>
> $./a.out
> Time: 1.5705sec
> Measurement granularity: 1000000 of a second
> Time: 1.56849sec
> Measurement granularity: 1000000 of a second
>
> so I guess it's the same. Now, why would you use reference counting for 
> such a small type??? Let's say you use double precision Point_2 types, 
> you would need 64 bits to represent the x and y coordinates and then 
> half of that to keep track of the storage???? Isn't that something silly 
> to do? I would expect using reference counting in objects that require a 
> large amount of storage (the std::string class for example, even though 
> I believe that even that class is not using reference counting). If you 
> allocate 100 different points, 33% of your memory goes away in using 
> pointers!!!
>
> I guess that Cartesian uses objects that really need reference counting, 
> but I don't think that the Point_2 should be one of them. Then users of 
> the library who didn't bother to go over the entire documentation (like 
> me) think that Cartesian is their best candidate, when it is not.

Historical artefact, I owuld say (like many things in CGAL you may find 
strange).

Had you gone a bit at the documentation, you would not have used Cartesian
or Simple_cartesian at all, maybe.

Also, keep in mind that the primary goal of CGAL's kernels is to support
the higher level geometric algorithms, which have specific requirements
e.g. in terms of robustness.

We still don't know what you do with your points :-)

--
Sylvain Pion
INRIA Sophia-Antipolis
Geometrica Project-Team
CGAL, http://cgal.org/