Semantics for 1d bbox or "bad" geojson geometries

[philvarner]

Question filed against OGC API here:

• Does a bbox have to be 2d or 3d? opengeospatial/ogcapi-features#786

Prompted by:

• POST search requests crashing due to bbox having the same pair of coords #375

Questions:

Bbox:

• Is a 1d bbox valid? In implementation, this could be handled by converting the bbox to either a Point or LineString geometry for query against the database.

Intersects geometry:

• What's considered an invalid geometry? Currently there's no language like "implementations may reject geometries that have semantic problems like self-crossing at their discretion with a 400 and an informative error message".
• Polygons that have an "obvious" self-crossing are invalid, but are two identical sequential points a self-crossing? These are rejected by Elasticsearch & OpenSearch for this reason, so it seems they should be.

[philvarner]

WRT to bboxes, Clemens said language was added to the corrigendum that states that degenerate bboxes are allowed, and should be treated with point or linestring semantics. This should be clarfied in the spec so implementers will remember to handle this case. E.g., in stac-server, bboxes are converted to Polygon for search, but a Polygon with multiple identical sequential points is rejected by ES/OpenSearch, so these have to be handles specially.

[philvarner]

After discussion at the STAC meetup 2023-Feb-13: add some language around the fact that implementions are allowed to reject what they consider semantically-invalid geometries. For example, OpenSearch rejects self-crossing polygons and polygons with sequential duplicate coordinates (which could be considered either self-crossing or not forming a line between them.)