lukeplatts opened a new issue, #1115:
URL: https://github.com/apache/sedona-db/issues/1115
## Describe the bug
Any query that combines **2 or more non-geometry columns** with an **`ST_*`
accessor function** (e.g. `ST_IsEmpty`, `ST_IsValid`, `ST_XMin`, `ST_NPoints`)
applied to a geometry column **sourced from a file scan** (e.g. `read_parquet`)
panics on every Arrow-materializing export path — `to_arrow_reader()`,
`to_arrow_table()`, `to_pandas()`, `to_parquet()`. `count()` and `show()` are
unaffected, since they don't go through Arrow export.
```
thread '<unnamed>' panicked at .../arrow-schema-57.3.0/src/schema.rs:375:10:
index out of bounds: the len is 2 but the index is 2
```
The index value scales with the number of extra columns (reproduced with 1,
2, and 3 — 1 never panics, 2+ always does), which points at a fixed-size
metadata array (likely CRS/extension-type metadata for the geometry column)
being indexed by column position without accounting for the other columns in
the schema.
This affects both the sync (`sedonadb`) and `sedona.db` Python entry points
and reproduces on a fresh `sedona.db.connect()` with no options set. It is not
related to column ordering, nullability, or the specific `ST_*` function —
every accessor function tried reproduces it identically.
## To Reproduce
```python
import pyarrow as pa
import pyarrow.parquet as pq
import shapely
import sedona.db
path = "/tmp/repro.parquet"
geom = pa.array(
[shapely.to_wkb(shapely.geometry.Point(1, 2)),
shapely.to_wkb(shapely.geometry.Point(3, 4))],
type=pa.binary(),
)
pq.write_table(pa.table({"geometry": geom}), path)
sd = sedona.db.connect()
sd.read_parquet(path).to_view("raw", overwrite=True)
# 2 extra non-geometry columns ('c1', 'c2') + one ST_* accessor on the
scanned geometry
sd.sql("""
CREATE OR REPLACE VIEW g1 AS
SELECT 'a' AS c1, 'b' AS c2, ST_SetSRID(ST_GeomFromWKB(geometry), 4326)
AS geometry
FROM raw
""")
sd.sql("CREATE OR REPLACE VIEW t AS SELECT *, ST_IsEmpty(geometry) AS _e
FROM g1")
for batch in sd.sql("SELECT * FROM t").to_arrow_reader():
pass # panics here
```
Reducing to a single extra column (`c1` only, dropping `c2`) does **not**
panic. Reducing to zero `ST_*` calls does not panic. Using a literal geometry
(`ST_Point(1, 2)`) instead of one read from a file scan does not panic.
## Expected behavior
The query executes and exports without panicking, as it did in `sedonadb`
0.3.0 (confirmed no reproduction there).
## Environment
- sedonadb: 0.4.0 (conda-forge, `py314h9a0e2ed_0` build, linux-64 and
osx-arm64 both reproduce)
- Python: 3.14
- OS: reproduced on both macOS (osx-arm64) and Linux (linux-64) conda-forge
builds
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
