Use CoordinateSequence instead of Coordinate arrays

[dr-jts]

To improve the performance of using non-array based CoordinateSequenceimplementations the JTS core should be changed to using coordinate access based on CoordinateSequence accessors, instead of forcing the creation of Coordinate arrays.

Some places where this is needed:

• uses of PointLocation isInRingand locateInRing
• Centroid.add(Polygon)

See also:

• https://github.com/NetTopologySuite/NetTopologySuite/pull/193
• https://github.com/NetTopologySuite/NetTopologySuite/pull/71

[FObermaier]

All usages of org.locationtech.jts.geom.CoordinateArrays class (here) potentially benefit from a utility class that works on org.locationtech.jts.geom.CoordinateSequence instead of org.locationtech.jts.geom.Coordinate[]. This would have to be extended.

Not explicitly performance, but making the [WKT|WKB]Reader classes not work on org.locationtech.jts.geom.Coordinate[] can help to improve handling of measure ordinate values. I have an implementation using a CoordinateBuffer class at hand.

[dr-jts]

Good recommendations @FObermaier .

What is the nature of your CoordinateBuffer class?

[FObermaier]

What is the nature of your CoordinateBuffer class?

It encapsulates an ArrayList to store 3DM coordinate values. You can add ordinate values calling add2d, add3d, add2dm, add3dm functions. While adding the following is taken care of:

• Keep track of extent (x, y) and possibly the interval of z- and m- values.
• Check for "No data values" and transform them to and from Coordinate.NULL_ORDINATE
• Create a CoordinateSequence using any factory
• Get array of ordinate values (getXY(), getZ(), getM())

You can have a look at it here.