Releases: davidmoten/rtree
Releases · davidmoten/rtree
0.8-RC10
0.8-RC9
- change
Rectangle
to an interface so that can reduce memory used byPoint
instances (should save 24 bytes perPoint
instance)
0.8-RC8
0.8-RC7
- fix backpressure race condition and release volatile state (null it) before calling
onComplete
inOnSubscribeSearch
- avoid cast of
float
todouble
inSplitterQuadratic
0.8-RC6
- add final to static comparators in
SplitterRStar
- improve R*-tree insert throughput by up to 100% by optimizing
RStarSplitter
- decrease object allocations and casting to increase performance of insert for R-tree and R*-tree
0.8-RC5
- reduce parsing for flatbuffers searching, use curly brackets more
- presize
ArrayList
s to increase performance - upgrade to rxjava 1.1.2, remove patch in
OnSubscribeSearch
to force fast path for requestLong.MAX_VALUE - 1
as workaround for RxJava PR #3227
0.8-RC4
- use
Float.compare
andDouble.compare
inComparators
class to increase perf (insert/delete) - use
Float.compare
instead of boxing primitives and callingcompareTo
. Should help performance ofSplitterRStar
(insert/delete) - don't extract point from flatbuffers data structure more than required (improves performance) in
FlatBuffersHelper.toGeometry
- fix infinite loop when
rectangleGeographic
called with double parameters (introduced in 0.8) - switch order of
compare
parameters instead of negating result because of possbility of problems atInteger.MIN_VALUE
being negated - handle serialization of empty trees correctly
0.8-RC3
- add generic wildcards to
Serializers
api
0.8-RC1
Flatbuffers for serialization and low-memory RTree
- major refactor of internal data structures (many classes -> interfaces to support serialization use cases)
- instantiation of
Geometry
classes now should happen only throughGeometries.rectangle
,Geomtries.point
, etc. - Flatbuffers option for serializing (writing and reading an RTree to disk for instance)
- RTree built by deserializing using flatbuffers has 1/3 memory usage
- RTree built by deserializing using flatbuffers has reduced search performance (still O(log N) but 1/10 of throughput)
- fix redundant check of worst combinations in
QuadraticSplitter
as reported by Eric Perser (not in github issues), performance improved for usingQuadraticSplitter
by ~2%