[postgis-users] Re: Difference throws an error

[strk at refractions.net]

On Tue, Nov 07, 2006 at 02:29:00PM -0800, Martin Davis wrote:
> The reason the error is thrown is that at least one of the geometries 
> has a very gnarly spike on one side of it.  The second geometry contains 
> the vertices:
> 
> [      0:28] POINT (5417154.92 5658376.37)
> [      0:29] POINT (5417155.18955743 5658376.89699992)
> [      0:30] POINT (5417154.919 5658376.371)
> [      0:31] POINT (5417155.814 5658378.111)
> 
> These form a spike/gore where the inside vertex 29 is almost coincident 
> with the edge 30-31.
> 
> This geometry combination causes an error in all previous versions of 
> JTS/GEOS as well.  Hopefully this problem can be addressed with some 
> further work on robustness.

The error is due to Geometry snapping making the first geometry invalid.
Only applying commonbits removal passes the 'areatest' op.
I disabled this in RC2 to try at being in sync with JTS, but might
be worth re-enabling in RC3. The cost would be an additional
run of the operation IFF it fails with both original and common-bits-removed
inputs.

What do you think ?

--strk;

PS: so GEOS *does* benefit JTS sometime uh ? :)