[rjb32]

Order points and take their min, max and average.

[rjb32]

Note

AI-generated summary.

Summary

Goal. Points compare with <, <=, > and >=, and take min, max and avg: MATCH (n:City) RETURN min(n.at), max(n.at), avg(n.at).

Approach. The comparison operators follow ORDER BY's total order: SRID, then x, y and z, across CRSs too. Neo4j instead answers null for < between two different points. min and max use that order everywhere: plain, per group, DISTINCT, and over tagged cells. avg is the coordinate-wise centroid of points of one CRS. It is held as a running mean (Point::addToMean), so the accumulator is always a valid Point; a running sum of latitudes would fail the [-90, 90] check. Points of two CRSs, or points mixed with numbers, raise.

Where to look.

• storage/metadata/Point.cpp:590 Point::addToMean
• query/ir/exec/NLExecutor.cpp:3491 addToAverageTotal: the count the running mean needs, threaded through the typed folds
• query/ir/exec/NLExecutor.cpp:4034 NLTaggedSum::addPoint: the same for tagged cells
• query/AST/decl/EvaluatedType.h:365 orderClassOf: points now have an order class

point({x:1,y:9}) < point({x:2,y:0})              true
point({x:9,y:9,crs:'wgs-84'}) < point({x:0,y:0}) true    4326 < 7203
point({x:1,y:2}) < 3                             null
avg([(0,0), (4,0), (2,6)])                       point({x: 2.0, y: 2.0, crs: 'cartesian'})

Evidence

• Before: < between points is null, and min, max and avg reject points.
  After:

  test_query_ir_point_order_and_aggregate   [  PASSED  ] 17 tests.
  ctest                                     100% tests passed, 0 tests failed out of 1169
  

Merge Danger

Door: two-way

Blast Radius: query

A geographic centroid is a plain mean of longitudes and latitudes, so points on both sides of the 180th meridian average to a longitude near 0. docs/points.md says so.