Hi,

I have some news regarding the rtree:

1. Segments are now supported by the rtree. They can be used as Indexables and stored directly in the spatial index.

2. nearest() predicate now uses the geometry::comparable_distance() which means that all combination of parameters supported by this algorithm are automatically supported by the rtree.
This means that geometries other than Point can be passed to the predicate. And that it should now work properly for non-cartesian coordinate systems.

The above 2 means that it should be possible to search e.g. for k Segments nearest to some other Segment, or to some Box, etc.

Synthetic benchmarking for random Segments shows that the rtree is slower than in the case when Boxes are stored however the results could be different for a real-life use-case. Without the Segment support the user would be be forced to store bounding Boxes of Segments. During a query some operations would be performed and then the user would be forced to perform them second time for the Segments coresponding of the returned Boxes. As usual benchmarking should be done for a specific use case.

Those are some synthetic results gathered for 1M random Values and rtree using linear balancing algorithm (http://github.com/boostorg/geometry/blob/develop/index/example/benchmark_experimental.cpp):

BOXES	SEGMENTS
0.436803 seconds	0.468003 seconds	pack 1000000
0.0624004 seconds	0.0780005 seconds	query(B)* 100000 found 100046
1.43521 seconds	1.49761 seconds	insert 1000000
0.748805 seconds	0.967206 seconds	query(B) 100000 found 100046
0.780005 seconds	0.998406 seconds	range queried(B) 100000 found 100046
1.01401 seconds	1.24801 seconds	qbegin(B) qend(B) 100000 found 100046
0.998406 seconds	1.20121 seconds	qbegin(B) qend() 100000 found 100046
1.04521 seconds	1.24801 seconds	range qbegin(B) qend(B)100000 found 100046
1.06081 seconds	1.27921 seconds	type-erased qbegin(B) qend() 100000 found 100046
1.07641 seconds	1.27921 seconds	range type-erased qbegin(B) qend() 100000 found 100046
0.405603 seconds	-	query(i && !w && !c) 100000 found 50025
0.717605 seconds	0.717605 seconds	query(nearest(P, 10)) 20000 found 200000
0.764405 seconds	0.842405 seconds	qbegin(nearest(P, 10)) qend(n) 20000 found 200000
0.748805 seconds	0.811205 seconds	qbegin(nearest(P, 10)) qend() 20000 found 200000
0.780005 seconds	0.858006 seconds	type-erased qbegin(nearest(P, 10)) qend() 20000 found 200000
0.358802 seconds	0.468003 seconds	remove 100000

* query(B) is a short form of query(intersects(box))

Regards,
Adam