[geos-devel] intersect/intersection() problem under Win

[Christian Pschierer]

On Thu, 16 Oct 2003, Martin Davis wrote:

> This is definitely a robustness issue (related to the "numerical
> precision" you mention, but more complex).  JTS 1.3 exhibits the same
> problem.  I'm happy to say that the forthcoming JTS 1.4 contains some
> bug fixes which seem to solve this problem.  These will need to be
> rolled into GEOS sometime... (but first - FIXING LEAKS!)

This is true. When simply computing the intersection() for all polygons
and checking for areas > 0, the application runs 10% faster than with the
additional intersects/touches checks, but leaks about 70MB memory.


> Strange that Linux doesn't exhibit the same problem.  What hardware
> are you running on?  These kind of problems will work better on
> machines with higher precision floating point - could that be the
> difference?  

I don't think so. We are using the same Laptops under Linux and WinXP (see
below), so it must be a software/compiler problem.

Linux: Centrino 1.6GHz (SuSE8.2, gcc3.3) and Athlon 2GHz (SuSE8.1, gcc3.2)
Windows: Centrino 1.6GHz (WinXP, MS Visual Studio 6.0 with the standard MS
	 Compilers)
Irix: Onyx (R10000 CPU with R10010 FPU)

If I have some time I will try the Intel Compiler on Linux and see what
happens.


Christian