New `GEOMETRY` and `GEOGRAPHY` columns in delta/Databricks are not supported

[danielsparing]

If the underlying delta table has a GEOMETRY or GEOGRAPHY type column, the delta_scan fails as below.

Steps to reproduce:

• You can use Databricks Free Edition, but need to have a Unity Catalog schema that points to an S3 bucket you can read from, such as an External Schema defined on an S3 bucket you have access to.
• Write any table from Databricks SQL with a GEO* data type column, such as:

-- Databricks SQL
create table duckgeo.e as
select
  st_point(2, 3) geo

• Make sure DuckDB has access to the delta lake table directory, either on S3 or just download the whole directory to a convenient location.
• In DuckDB 1.4, try:

-- DuckDB SQL
install spatial; load spatial;
from delta_scan('/path/to/delta_table');

The error you get:

IOException                               Traceback (most recent call last)
File <command-[...]>, line 1
----> 1 duckdb.sql("from delta_scan('/[...]/delta_table')")

IOException: IO Error: DeltaKernel UnsupportedError (35): Unsupported: Unknown ReaderFeatures: "geospatial". Supported ReaderFeatures: "columnMapping", "deletionVectors", "timestampNtz", "typeWidening", "typeWidening-preview", "vacuumProtocolCheck", "v2Checkpoint", "variantType", "variantType-preview", "variantShredding-preview"

You can read the table with read_parquet(), but then of course you don't have Delta Lake versioning support.

[danielsparing]

Attaching the example delta output (with userId, userName redacted in json).

example_delta_with_geospatial.zip

[danielsparing]

@Maxxen fyi

[djouallah]

@danielsparing can you add a vote here, this is a delta kernel rest limitation, not duckdb delta-io/delta-kernel-rs#1261