[rjb32]

Read points from GeoParquet.

[rjb32]

Note

AI-generated summary.

Summary

Goal. LOAD PARQUET reads GeoParquet point columns as points, with no clause: a column Parquet's GEOMETRY or GEOGRAPHY logical type marks, or one GeoParquet 1.x's geo file metadata names.

Approach. ParquetGeometry is built once per file. It finds the geometry columns and the CRS family of each: OGC:CRS84, EPSG:4326 or EPSG:4979 for wgs-84, an undefined CRS ("crs": null) for cartesian, and a PROJJSON CRS by its id. Any other CRS is rejected. It decodes each value as a WKB point, little- or big-endian, ISO or EWKB, 2D or 3D, with an empty point read as null. Lists of geometries become lists of points. A geometry that is no point, a point with a measure, and bytes that are no WKB are rejected.

Where to look.

• io/parquet/ParquetGeometry.cpp:157 findColumn: logical type first, then GeoParquet metadata
• io/parquet/ParquetGeometry.cpp:180 readPoint: the WKB decoding
• io/parquet/ParquetGeometry.cpp:277 crsOfString: the CRS mapping
• io/parquet/ParquetImportVisitor.cpp:199: where a geometry column becomes a Point property

OGC:CRS84 / EPSG:4326 / EPSG:4979 / default   -> wgs-84, or wgs-84-3d for Point Z
GeoParquet "crs": null                        -> cartesian, or cartesian-3d
anything else                                 -> rejected

Evidence

• Before: a WKB column loads as a String property of raw bytes.
  After: the test writes its node files with the Parquet C++ writer: native GEOMETRY and GEOGRAPHY logical types, and geo metadata.

  test_query_ir_load_parquet_point   [  PASSED  ] 16 tests.
  Parquet import suites              9/9 passed
  

Merge Danger

Door: two-way

Blast Radius: import

A GeoParquet file that used to load its geometry column as a string property now loads it as points, or fails if the column holds other geometries or a CRS no point is in.