Subject: Re: [boost] [geometry] robustness approaches
From: Simonson, Lucanus J (lucanus.j.simonson_at_[hidden])
Date: 2009-03-19 16:10:38
Paul A. Bristow wrote:
> I've been following this erudite discussion with interest and a modest
> degree of understanding.
> But one thing seems clear to me - that any quality implementation is
> to require at the very least 'big' ints and 'big' floats, and
> probably exact int and exact floats.
> It would seem that we need tried and tested Boost license
> implementations - preferably before starting work on a complex
> geometry problem. No solution can be considered for Boost if it uses
> any restrictive-licensed components.
GMP is licensed under the LGPL. That's not such a bad license. I don't know why we need to bypass it with a free implementation. Even so I don't depend on gmp header files in my geometry library. Instead I allow the "big" type to be overridden by specializing a meta-function that looks it up. I don't need to depend on GMP header files in my library code to use GMP with my library in an application. There is, incidently, a boost multi-precision library under development. Brandon tried it out about a year ago. It is a tall order to get performance up to the level of GMP, which uses inline assembly and clever C++ tricks to eliminate unnecesary copying of the heavy number types.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk