Github user DonnyZone commented on the issue:
https://github.com/apache/spark/pull/19568
This PR is similar to the initial commit when I try to fix SPARK-21441 in
https://github.com/apache/spark/pull/18656
(https://github.com/apache/spark/pull/18656/commits/92dc106aec59a0f2755d7621d2d03831250cccc0).
(1) The INPUT_ROW in SortMergeJoinExec's codegen context points to the
wrong row. In general, it works well. However, this behavior potentially
causes wrong result or even NPE in `bindReference`. I think it is necessary to
fix it.
(2) The `CollapseCodegenStages` rule before 2.1.1 has an issue which may
lead to code generation even the SortMergeJoinExec contains CodegenFallback
expressions, when it has an umbrella SparkPlan (e.g., ProjectExec).
Consequently, it triggers the above potential issue.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
