Follow-up to #1138 — test-only and docs cleanup, no production behavior change:
epsg_4326_behavior has its assert.Equal args reversed (actual passed where expected goes), so a failure prints the diff backwards.- The
wktToWKB test helper reuses the "parse WKT" error message on its wkb.Marshal branch — it should say "marshal WKB".
TestParquetGeoArrowExtensionMetadataRoundTrip uses unchecked type assertions that panic the goroutine on a mismatch instead of failing cleanly. Routing them through the comma-ok assertGeoArrowWKB pattern the PR already has would report regressions properly.- That same round-trip test only covers
srid:4326 variants; the default-CRS geometry/geography columns — the most common real-world case — never go through the Parquet metadata path. Worth adding them.
- Document the intentional contract at
geoArrowMetadataToIcebergType: a geoarrow.wkb field with no edges (the PyIceberg/arrow-rs default) reads back as GeometryType, so a foreign geography(…, planar) column lands as geometry. A short pinning test would make that explicit. While there, the "planar geography from other clients" half of the VisitGeography comment is misleading — planar isn't a valid geography algorithm — so it can be trimmed.
- The trickier
readOnlyCases entries (e.g. geometry_srid_0_with_crs_type, geometry_epsg_4326_incorrect_type) could use a one-line note on what each one pins.
- Minor:
TestWKTToWKB mostly exercises go-geom's own behavior rather than iceberg-go logic, and geomList/geomSRID are duplicate declarations in the schema subtest.
Mostly self-contained, low-risk changes — good first issue territory.
Follow-up to #1138 — test-only and docs cleanup, no production behavior change:
epsg_4326_behaviorhas itsassert.Equalargs reversed (actual passed where expected goes), so a failure prints the diff backwards.wktToWKBtest helper reuses the "parse WKT" error message on itswkb.Marshalbranch — it should say "marshal WKB".TestParquetGeoArrowExtensionMetadataRoundTripuses unchecked type assertions that panic the goroutine on a mismatch instead of failing cleanly. Routing them through the comma-okassertGeoArrowWKBpattern the PR already has would report regressions properly.srid:4326variants; the default-CRS geometry/geography columns — the most common real-world case — never go through the Parquet metadata path. Worth adding them.geoArrowMetadataToIcebergType: ageoarrow.wkbfield with noedges(the PyIceberg/arrow-rs default) reads back asGeometryType, so a foreigngeography(…, planar)column lands as geometry. A short pinning test would make that explicit. While there, the "planar geography from other clients" half of theVisitGeographycomment is misleading — planar isn't a valid geography algorithm — so it can be trimmed.readOnlyCasesentries (e.g.geometry_srid_0_with_crs_type,geometry_epsg_4326_incorrect_type) could use a one-line note on what each one pins.TestWKTToWKBmostly exercises go-geom's own behavior rather than iceberg-go logic, andgeomList/geomSRIDare duplicate declarations in theschemasubtest.Mostly self-contained, low-risk changes — good first issue territory.