CGAL users discussion list

Le Lun 16 juillet 2007 14:27, Matthijs Sypkens Smit a écrit :
> Exact_predicates_inexact_constructions_kernel I encounter cases where:
> - coplanar_orientation() returns COLLINEAR, but the construction does not
> fail and returns a point,
> - coplanar_orientation() does not return COLLINEAR, but the circumcenter
> construction does end in an assertion failure.
>
> Are both these cases to be expected?

Yes they are: nothing is filtered inside the construction methods.

> In particular the last surprised me,
> since I was hoping to 'catch' the assertion failure, by testing in
> advance, but now it seems that this does not generally work.

What you test beforehand with coplanar_orientation is the "exact" property
of collinearity. The problem is that it has, for example, no relation with
some double becoming zero in the middle of an inexact construction.

> Is it just plain impossible to construct circumcenters without getting
> assertion failures using my choice of kernel, or are there ways around
> this (by testing beforehand)?

It depends on what you want exactly. For which purpose do you want to use
these circumcenters?

>   CGAL error: assertion violation!
>   Expr: !(i>s)
>   File: /user/CGAL-3.3/include/CGAL/Interval_nt.h
>   Line: 77
>   Explanation: Variable used before being initialized (or CGAL bug)
> but so far I cannot seem to find the cause of the problem. I do believe
> that all variables involved have actually been initialized.

This may be the bug that we corrected recently. What exactly did trigger
this assertion violation? Are you using the Regular Triangulation? If you
are, the attached diff-file may help you.
-- 
Camille

Attachment: diff-bug-regular
Description: Binary data