CGAL users discussion list

[Marc Glisse <>]

On Thu, 26 Sep 2013, ax487 wrote:

> I am currently working with the
> Exact_predicates_inexact_constructions_kernel provided by CGAL 4.2. I am
> however unsure which of the following functions are actually geometric
> predicates and therefore evaluated exactly:

Any kernel functor with a discrete return type (like an enum) should be a 
predicate, that includes all those you are listing.

> - The output of the do_intersect function for a pair of objects each of
> which may be either a Ray_2, a Segment_2 or a Line_2.
> - The output of the orientation function: I would imagine that I can
> rely upon LEFT_TURN/RIGHT_TURN, but I believe that verifying whether
> three points are collinear is unlikely to ever work correctly using
> anything other than rational arithmetic.
> - The output of the bounded_side_2 for convex/simple/not simple
> polygons, specifically in the case of the result being ON_BOUNDARY

CGAL does use exact arithmetic (like rationals) if needed to evaluate 
predicates, and particularly for the ON_BOUNDARY case. It first tries 
faster methods, but if it can't prove the correct answer using those, it 
falls back to exact.

--
Marc Glisse