Is bbox=x,y,x,y a invalid/empty bbox or a point ?

[rouault]

The spec doesn't mention if a bbox degenerated to a single point (like bbox=2,49,2,49) should be considered as a geometry-with-point intersection or geometry-with-empty intersection. From a client side of view, it would be great to have some clarification. In the case where this would be legit for servers to interpret this a geometry-with-empty intersection, clients might replace their query with a small rectangle around the point (like box=1.99,48.99,2.01,49.01)

[rouault]

CC @aaime

[cportele]

Why would a degenerate bounding box (point or horizontal/vertical line) be considered "empty"? It specifies a non-empty set of positions. If the bbox is a point then everything that intersects that point should be matched.

[aaime]

The underlying library we use for intersections, JTS, considers the polygon generated from the empty bbox as non intersecting... probably because it's not a valid one, topologically speaking (the intersection process is a two step, first checking that envelopes intersect, and then doing a polygon vs polygon intersection that ensures the bbox actually intersects the geometry). I guess we could make it work so that if the bbox is a empty one, a point in polygon test is done instead.

[cportele]

Thanks @aaime, I wasn't aware of that behaviour in JTS.

The Simple Features for SQL spec does not discuss degenerate envelopes either, if I recall correctly. Does anyone know what the SQL implementations (PGIS, Oracle, etc.) do? I think PGIS would return true for something like SELECT ST_Intersects(ST_Envelope('POINT(2 49)'::geometry), 'LINESTRING ( 2 49, 3 49 )'::geometry)?

[rouault]

I think PGIS would return true for something like SELECT ST_Intersects(ST_Envelope('POINT(2 49)'::geometry), 'LINESTRING ( 2 49, 3 49 )'::geometry)?

Just tested it. Yes, it returns true

[jampukka]

My implementation seems to return features that intersect with point bbox (x,y,x,y) and also with a line (x1,y,x2,y) or (x,y1,x,y2). However for collections with point/linestring features it's unlikely for the specified point to hit any of the features due to inaccuracies caused by reprojection and floating point<->decimal conversions.

[cportele]

Web-meeting 2020-01-20: A degenerated bbox is a point and should be evaluated as a point.

We should check the other databases beside PostGIS to confirm that this is the interpretation of the DBs, too.

[aaime]

Worth having a CITE test for this?

[cportele]

Meeting 2020-03-03: Confirmed the conclusion in the 2020-01-20 meeting. Clarification text will be added (no change to the requirements).

(A test case in the CITE ETS could be useful.)