Aleksandar Trajković created SPARK-62090:
--------------------------------------------
Summary: Use SQL type names for unmapped integral targets of
pushed JDBC casts
Key: SPARK-62090
URL: https://issues.apache.org/jira/browse/SPARK-62090
Project: Spark
Issue Type: Bug
Components: SQL
Affects Versions: 5.0.0
Reporter: Aleksandar Trajković
JDBCSQLBuilder.visitCast spells a cast's target with the dialect's getJDBCType
and falls back to Spark's type name, not the common JDBC mapping that writes
use. Dialects that don't map LongType or ShortType (generic, Postgres, SQL
Server, DB2) therefore push casts such as CAST(int_col AS long), which these
databases reject, so queries like WHERE CAST(int_col AS BIGINT) > 0 or WHERE
int_col = bigint_col fail when pushed down.
Proposal: spell unmapped integral targets with SQL type names for exact
integral up-casts (MySQL: SIGNED), and keep string-to-integral casts in Spark,
since databases convert strings differently. SPARK-58372 already does this for
casts from fractional types.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]