Re: [boost] GGL Review

Subject: Re: [boost] GGL Review
From: Fernando Cacciola (fernando.cacciola_at_[hidden])
Date: 2009-11-18 23:21:20

Hi Steve,

[disclaimer: I'm just responding to current traffic totally out of thread and 
without knowing the current state of the GGL]

> Using floating-point filters the
> speed can be competitive with the naive algorithm, but without the
> crashes.
> 
There is in fact something VERY important to add here:

I know for a fact, since I have carefully looked into it, that Boost.Polygon is 
100% robust when used with integer coordinates provided the user properly sets 
the "extended integer type" needed by the filtering mechanism (which is already 
set when the coordinate type is "int" and sizeof(long) == 2*sizeof(int)).

IOW, Boost.Polygon uses a filter, just like CGAL.

This is why I specifically said that the library should be considered to be 
restricted to integer coordinates.

While this was not brought up during the formal review, it was in the preceding 
discussions.

I was aware of this and this *fundamental library property* was the most 
important reason why I accepted the library into boost.

If you google the developers list you should find my very lenghly explanations 
about why this is an absolute requirement, why GTL (at that time) was already 
meeting it, and why GGL was not.

I even sketched a general approach that could be followed to make GGL robust 
with floating point coordinates (essentially just explaining what a floating 
point filter is and how it works).

FWIW, Barend is very well aware of all this and has always been more than 
willing to do any required fixing. What happened is that I never had all the 
time needed to dedicatedly give him the neccesary guidance, so the issue is 
probably still unresolved.

Best

--
Fernando Cacciola
SciSoft Consulting, Founder
http://www.scisoft-consulting.com