UBarney commented on code in PR #16443:
URL: https://github.com/apache/datafusion/pull/16443#discussion_r2202340283
##########
datafusion/physical-plan/src/joins/nested_loop_join.rs:
##########
@@ -828,13 +833,127 @@ impl<T: BatchTransformer> NestedLoopJoinStream<T> {
handle_state!(self.process_probe_batch())
}
NestedLoopJoinStreamState::ExhaustedProbeSide => {
- handle_state!(self.process_unmatched_build_batch())
+ handle_state!(self.prepare_unmatched_output_indices())
+ }
+ NestedLoopJoinStreamState::OutputUnmatchedBuildRows(_) => {
+ handle_state!(self.build_unmatched_output())
}
NestedLoopJoinStreamState::Completed => Poll::Ready(None),
};
}
}
+ fn get_next_join_result(&mut self) -> Result<Option<RecordBatch>> {
+ let (left_indices, right_indices, start) =
+ self.join_result_status.as_mut().ok_or_else(|| {
+ datafusion_common::_internal_datafusion_err!(
+ "should have join_result_status"
+ )
+ })?;
+
+ let left_batch = self
+ .left_data
+ .as_ref()
+ .ok_or_else(|| {
+ datafusion_common::_internal_datafusion_err!("should have
left_batch")
+ })?
+ .batch();
+
+ let right_batch = match &self.state {
+ NestedLoopJoinStreamState::ProcessProbeBatch(record_batch) =>
record_batch,
+ NestedLoopJoinStreamState::OutputUnmatchedBuildRows(record_batch)
=> {
+ record_batch
+ }
+ _ => {
+ return internal_err!(
+ "state should be ProcessProbeBatch or OutputUnmatchBatch"
+ )
+ }
+ };
+
+ let current_start = *start;
+
+ if left_indices.is_empty() && right_indices.is_empty() &&
current_start == 0 {
Review Comment:
Now I understand why [this
test](https://github.com/apache/datafusion/blob/acf0bbe1cedf3fe0155de5f41a5b66046e262e04/datafusion/core/src/dataframe/mod.rs#L1250-L1263)
passed after I changed the return value from `None` to
`RecordBatch::new_empty`.
In this unit test, the join result is converted to a string and then
compared with the expected output. When converting to a string, it retrieves
the schema from the `record_batch` (as the passed `schema_opt` is `None`).
https://github.com/apache/arrow-rs/blob/7b219f98c25fcd318a0c207f51a41398d1b23724/arrow-cast/src/pretty.rs#L186
When executed in the CLI, there's no issue even if the Nested Loop Join
(NLJ) returns 0 record batches.
```
> select t1.value from range(1) t1 join range(1) t2 on t1.value + t2.value
>100;
+-------+
| value |
+-------+
+-------+
0 row(s) fetched.
```
From a compatibility standpoint, I think it's better to keep it consistent
with the previous behavior.
--
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]
