fornwall opened a new issue, #25281:
URL: https://github.com/apache/datafusion/issues/25281
### Describe the bug
`GROUP BY` (and `SELECT DISTINCT`) over a group key whose type contains a
zero-field struct (`DataType::Struct(Fields::empty())`) panics when the
aggregate emits.
A zero-field struct array has no child to derive its length from, so
arrow-rs refuses `StructArray::try_new` for it. The group-value emit path
rebuilds the key column with that constructor and unwraps the resulting error.
### To Reproduce
```console
$ datafusion-cli -c "select arrow_cast(null, 'Struct()') as s, count(*) from
(values (1),(2),(3)) group by 1;"
DataFusion CLI v55.1.0
thread 'tokio-rt-worker' panicked at
datafusion/physical-plan/src/aggregates/group_values/multi_group_by/row_backed.rs:225:14:
dictionary re-encode during emit: ArrowError(InvalidArgumentError("use
StructArray::try_new_with_length or StructArray::new_empty_fields to create a
struct array with no fields so that the length can be set correctly"), Some(""))
Error: Join Error
caused by
External error: task 63 panicked with message "dictionary re-encode during
emit: ..."
```
Same panic for `SELECT DISTINCT`:
```sql
select distinct arrow_cast(null, 'Struct()') as s from (values (1),(2));
```
and when the zero-field struct is nested inside the key rather than being
the whole key:
```sql
select named_struct('a', column1, 'z', arrow_cast(null, 'Struct()')) as s,
count(*)
from (values (1),(2)) group by 1;
```
### Expected behavior
_No response_
### Additional context
The query returns its groups. A zero-field struct is a valid Arrow type and
carries one bit per row (null or not), so it should group like any other key.
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
