Anurag Mantripragada created SPARK-61506:
--------------------------------------------
Summary: Mark a CALL as bound only after BindProcedures binds it
Key: SPARK-61506
URL: https://issues.apache.org/jira/browse/SPARK-61506
Project: Spark
Issue Type: Bug
Components: SQL
Affects Versions: 5.0.0
Reporter: Anurag Mantripragada
CALL fails, or silently ignores arguments, when the procedure implements both
UnboundProcedure and BoundProcedure (for example a SimpleProcedure, or
Iceberg's procedures) and an argument is resolved later in the analyzer
iteration than BindProcedures.
Call.bound is derived from the procedure type, so such a procedure counts as
bound before BindProcedures has called bind(), rearranged the arguments and
filled in defaults. ProcedureArgumentCoercion and InvokeProcedures then see the
arguments as written. With p(ts TIMESTAMP, label STRING DEFAULT 'none')
implemented as a SimpleProcedure:
{code:sql}
CALL cat.ns.p(TIMESTAMP '2026-01-02 00:00:00' - INTERVAL '1' DAY)
-- IllegalArgumentException: requirement failed: number of args and params must
match after binding
CALL cat.ns.p(label => 'x', ts => TIMESTAMP '2026-01-02 00:00:00' - INTERVAL
'1' DAY)
-- IllegalArgumentException: requirement failed: args must be foldable
CALL cat.ns.p(TIMESTAMP '2026-01-02 00:00:00' - INTERVAL '1' DAY, 'x', 1)
-- runs and ignores the third argument instead of failing with WRONG_NUM_ARGS
{code}
The same happens for scalar subquery arguments and time zone dependent
expressions such as CAST('...' AS TIMESTAMP). If bind() returns a different
object, Spark also calls call() on the object that was never bound, and EXPLAIN
CALL binds it on every iteration until it reaches the iteration limit.
Affected since 4.0.0 for procedures implementing both interfaces, and since
4.2.0 for SimpleProcedure.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]