CGAL users discussion list

[Marc Glisse <>]

On Wed, 17 Feb 2016, Benjamin Kehlet wrote:

> I am struggling with memory leaks when doing some simple manipulations of
> the geomtric locations of the vertices of a Polyhedron_3 when using the
> EPECK kernel.
>
> Attached is the simplest example that reproduces the issue I could come up
> with. It is an implementation of a very simple smoothing scheme, which just
> replaces the location of every vertex with the centroid of its neighbors.
> The implementation does two passes: In the first pass the new locations are
> computed and collected in an std::vector<std::pair<Vertex_handle,
> Point_3>>. Then in the second pass the new locations are applied. This
> leaks heavily and the consumed memory (simply monitored with top) increases
> by each smoothing iteration.

What makes you say that it leaks? Epeck is lazy, it stores the tree of 
operations that led each object to its current value, with an approximate 
version of the object. It only drops that tree when something calls 
exact() on an object (usually because the approximate object was not 
precise enough to determine the result of a predicate). So it is normal 
that the size increases. You can disable that lazy behavior (look for 
macros in the header that defines Epeck), but the running time will 
increase since all constructions will be done eagerly with an exact NT.

> Another strange thing is that when running "valgrind --tool=memcheck", I
> get this error
> """
> terminate called after throwing an instance of 'CGAL::Assertion_exception'
>  what():  CGAL ERROR: assertion violation!
> Expr: -CGAL_IA_MUL(-1.1, 10.1) != CGAL_IA_MUL(1.1, 10.1)
> File:
> /home/benjamik/software/CGAL-4.8-beta1/local/include/CGAL/Interval_nt.h
> Line: 209
> Explanation: Wrong rounding: did you forget the  -frounding-math  option if
> you use GCC (or  -fp-model strict  for Intel)?
> """
> which doesn't show up when I just run the program (with a debug
> configuration of CGAL).

valgrind does not support floating point rounding modes, old issue. I wish 
this were a warning instead of a hard error though.

--
Marc Glisse