CGAL users discussion list

[Stefan Schirra <>]

Hi Monique and Stefan,

not sure whether my 2 cents might be helpful, but may be, they are.

I conclude from Monique's answer that the CGAL code uses CORE::Expr 
internally.

> > So I'm wondering: is it possible that previous calculations performed
> > with dt affect future calculations? I'm imagining something like if
> > some calculation needed a certain precision (or number field), the
> > consecutive calculations are performed with the same precision even
> > though they might not need it.

Basically, such expression-dag based number types like CORE::Expr or 
leda::real (or Real_algebraic which is still under development) use 
adaptive precision computation, i.e., somehow increase the local 
precision repeatedly until it suffices. By local I mean the precision 
used during a decision making (sign computation). There is no global 
overall precision.

There are two strategies when it comes to the next decision making. 
Either you keep all the correct information you already have or you 
round to a small local starting precision, thereby introducing some 
error but making initial computations faster again, and then later 
adjust the precision adaptively. To my knowledge, CORE keeps all the 
correct information it has in order to avoid redundant recomputation.

> > If this is the case, is there a way to "reset" the precision?

I don't know, but I believe there is nothing like that. Usually, keeping 
all the correct information is a fairly good strategy.

> > Now I'm experiencing the following behavior: I have a program (and a
> > set of input) that finishes "immediately" but if I add some code, it
> > takes on the order of minutes and more *even in the old part of the
> > code*. 

I do not quite understand what you mean by "add some code", but it is 
hard to believe that running time switches form seconds to minutes just 
because of keeping correct bits. I guess, there must be other causes, 
too, running out of memory for example. But without knowing what you are 
actually doing ...

Best regards

        Stefan