-----Original Message----- From: boost-bounces@lists.boost.org [mailto:boost-bounces@lists.boost.org] On Behalf Of Phil Endecott Sent: Tuesday, November 17, 2009 5:39 PM To: boost@lists.boost.org Subject: Re: [boost] GGL Review
Paul A. Bristow wrote:
Phil Endecott wrote:
Personally, I find it better to use fixed point (e.g. for latitude/longitude) and relax knowing that I don't have to worry.
But isn't this assuming that your latitude/longitude (or whatever) is exact?
Both fixed and floating-point formats represent latitude and longitude values snapped to a grid. In the case of fixed point it's a grid of squares (on a flat earth, anyway); in the case of floating point it's a grid of rectangles whose sizes change depending on how far you are from the equator and prime meridian. Floating point is only "less exact" in the sense that we often don't worry about what the grid size is, but it is still there.
But the *actual position* is never known exactly, it always has some uncertainty, say + or - 1 grid unit. So, for example, if it appears to be in the wrong (next door or so) grid, you could decide that it doesn't matter to you. Is this why users seem to be happy with a potentially uncertain FP implementation? Is an interval based implementation another way to go - except that it will probably be too slow to be useful? (But I note that one author Sylvian Pion http://www.mpi-inf.mpg.de/~kettner/pub/nonrobust_cgta_06.pdf an interval enthusiast and Boost Interval author doesn't mention it, so perhaps not?). Paul --- Paul A. Bristow Prizet Farmhouse Kendal, UK LA8 8AB +44 1539 561830, mobile +44 7714330204 pbristow@hetp.u-net.com