Okay as Martin was kind enough to point out; (d) was a mistake and did
not mean the same thing as the others... I created a little test
harness and tried each alternative from scratch; carefully ignoring the
first couple of runs so the JIT compiler does not confuse things. If you
want to see the raw data look at the end of this page:
-
http://docs.codehaus.org/display/GEOTDOC/Point+Set+Theory+and+the+DE-9IM+Matrix
The results are (for comparing 100 randomly generated little 2x2 squares
arranged onto a 10x10 grid):
a) 230 mills geometry.disjoint( geometry2 ) || geometry.touches( geometry2 )
b) 341 mills matrix.isDisjoint() || matrix.isTouches(2,2)
c) 441 mills geometry.relate( geometry2, "F********" )
My guess would be that the explicit relationship methods on the JTS
geometry objects are optimized and do not compute the full matrix; all
other options cost about the same as they are based on computing the
complete matrix and then asking questions about the result. I was hoping
the relate method would notice my pattern only cared about the
intersection of the interior ;-)
Thanks for everyones time,
Jody
_______________________________________________
jts-devel mailing list
jts-devel@lists.jump-project.org
http://lists.refractions.net/mailman/listinfo/jts-devel
