CGAL users discussion list

[Stefan Schirra <>]

On 02.03.2012 16:08, Marc Glisse wrote:
> On Fri, 2 Mar 2012, Zohar wrote:
>
> > 1. Why is the Simple_Cartesian better than the Cartesian?
> >
> > I found in the doc of simple_cart:
> >
> > "In contrast to Cartesian, no reference counting is used internally. This
> > eases debugging, but may slow down algorithms that copy objects
> > intensively."
>
> It is not better in general. When the number type is double, the heavy 
> reference
> counting machinery has only drawbacks (copying a pair of double takes no 
> time):
> dynamic allocation, thread-unsafe. When the number type is something heavy 
> (say
> mpq_class or some other bignum that is not already reference-counted) and you 
> do
> a lot a copying objects around, it can be a life-saver.

Sorry Marc, I do not fully agree.

Thread-safety is certainly an issue. However, even copying pairs of doubles 
takes some time. It is not clear whether it takes more time than copying a 
pointer and incrementing an int. Even if it would, your application must copy a 
lot, because of the indirection in the access operations (cache effects are an 
issue as well). So it depends. For more, probably somewhat outdated, information 
on this issue, see "S. Schirra; A Case Study on the Cost of Geometric Computing. 
In Algorithm Engineering and Experimentation (ALENEX99), pages 156-176, 1999. 
Springer, Berlin. Note: LNCS 1619.

For "heavy number types", you can usually use Simple_cartesian, because 
mpq_class and other bignum integers already use reference counting themselves! 
So you don't copy bignums anyway.


--Stefan