Subject: Re: [boost] Formal Review: Boost.Polygon starts today August 24, 2009
From: Simonson, Lucanus J (lucanus.j.simonson_at_[hidden])
Date: 2009-09-01 12:02:13
Jeffrey Hellrung wrote:
> +1; I'm generally sympathetic to Barend's concerns, especially the
> (apparent?) lack of direct floating point support. A generic geometry
> or polygon library could be a component in some fracture simulation
> research software I plan to eventually write, but imprecision
> with converting to integer types, and hence the associated inexact
> constructions of intersection points, makes the proposed Boost.Polygon
> unuseable for that purpose, in my opinion.
We are currently using my library for that exact purpose of a component in fracture simulation research. Floating point does not provide more precision than integer, each has a certain number of bits of precision and both produce inexact constructions of intersection points.
> I would like to see at least a framework to be able to support
> point types and exact constructions. CGAL does this, I believe. I'm
> not familiar with their framework.
I can implement a transparent floating point to fixed point conversion wrapping the algorithm that is integer only. This should allow transparent use of the algorithm with floating point coordinates and no more loss of precision than floating point calculations themselves would produce.
> That said, I just want to say it looks like Luke's library is very
> well designed toward its intended goals. I'll try to give a more
> review saying as much.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk