[
https://issues.apache.org/jira/browse/SPARK-61805?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated SPARK-61805:
-----------------------------------
Labels: pull-request-available (was: )
> Defer constant-folding failures in MERGE actions
> ------------------------------------------------
>
> Key: SPARK-61805
> URL: https://issues.apache.org/jira/browse/SPARK-61805
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 5.0.0
> Reporter: Szehon Ho
> Priority: Major
> Labels: pull-request-available
>
> With ANSI mode enabled, MERGE can fail during constant folding while
> evaluating an expression in a clause that is never reached.
> For a MERGE-capable target with schema (pk INT, salary BIGINT, dep STRING),
> target row (1, 100, 'hr'), and source row (1, 10, 'hr'):
> {code:sql}
> MERGE INTO target t
> USING source s
> ON t.pk = s.pk
> WHEN MATCHED AND s.salary < 0 THEN UPDATE SET salary = 1 DIV 0
> WHEN MATCHED THEN UPDATE SET salary = s.salary
> {code}
> Expected: update salary to 10.
> Actual: DIVIDE_BY_ZERO during optimization, although the first action is not
> selected.
> Conditions in later clauses and expressions in other match types can fail the
> same way. MergeRows evaluates only the applicable clause list and stops at
> the first matching clause, but ConstantFolding does not currently treat
> instruction children as conditional.
> The fix can reuse the existing conditional-expression error deferral for
> MergeRows.Instruction children. Successful expressions still fold; failures
> remain runtime expressions and still raise errors if reached.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]