paul-heyse opened a new issue, #4846:
URL: https://github.com/apache/arrow-adbc/issues/4846
### What happened?
`adbc_ingest` into a PostgreSQL array column stores a NULL **element** of a
list as the element type's zero value. It raises no error and gives no warning.
The affected types are `float8[]`, `bigint[]`, `text[]`, `boolean[]` and a
domain over `double precision[]`.
Other NULL cases are stored correctly:
| Arrow value | Column type | Stored |
|---|---|---|
| `[1.5, None]` | `double precision[]` | `{1.5,0}` |
| `[1, None]` | `bigint[]` | `{1,0}` |
| `['a', None]` | `text[]` | `{a,""}` |
| `[True, None]` | `boolean[]` | `{t,f}` |
| `[1.5, None]` | domain over `double precision[]` | `{1.5,0}` |
| `None` (scalar) | `double precision`, `uuid` | `NULL` (correct) |
| `None` (whole list) | `double precision[]` | `NULL` (correct) |
| `[]` | `double precision[]` | `{}` (correct) |
Expected: `{1.5,NULL}`, `{1,NULL}`, `{a,NULL}`, `{t,NULL}`.
### Reproducer
```python
# adbc-driver-postgresql==1.12.0, adbc-driver-manager==1.12.0,
pyarrow==25.0.1, PostgreSQL 18.6
import adbc_driver_postgresql.dbapi as adbc
import pyarrow as pa
url = "postgresql:///scratch" # any empty database
with adbc.connect(url) as conn, conn.cursor() as cur:
cur.execute("CREATE TABLE t (v double precision[])")
cur.adbc_ingest("t", pa.table({"v": pa.array([[1.5, None]],
pa.list_(pa.float64()))}), mode="append")
conn.commit()
cur.execute("SELECT v::text FROM t")
print(cur.fetchall()) # [('{1.5,0}',)]; expected [('{1.5,NULL}',)]
```
### Probable cause
This is from reading the source on `main` at `63e765286` (2026-09-30); it is
not a tested patch. In `c/driver/postgresql/copy/writer.h`,
`PostgresCopyListFieldWriter::Write` puts `has_null_flags = 0` in the array
header. Its element loop then calls `child_->Write(...)` for every element
without testing the child's validity bitmap. Only the tuple writer and the
dictionary writer test validity. The binary array format expects a `-1` length
for a NULL element, and the flags word to say NULLs are present.
The list writer was last changed by #4320 (the fix for #4319, a sibling
silent-misalignment issue with sliced list arrays). The CHANGELOG for 1.12.0,
and for the 1.13.0 rc, lists no change for NULL elements. I found no existing
issue or PR for this.
### Environment
- `adbc-driver-postgresql` 1.12.0 and `adbc-driver-manager` 1.12.0 (PyPI
wheels)
- pyarrow 25.0.1
- CPython 3.14
- PostgreSQL 18.6 on Linux x86_64
--
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]
