[
https://issues.apache.org/jira/browse/SPARK-61341?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Eric Yang updated SPARK-61341:
------------------------------
Description:
When the pattern argument of `startswith`, endswith or contains is a column
rather than a string literal, V2ExpressionSQLBuilder renders the predicate as a
LIKE against a mangled copy of the column name instead of the column's value.
For example, startswith(email, email) becomes email LIKE 'mai%' ESCAPE '\'. A
DSv2 source that pushes these predicates down with V2ExpressionSQLBuilder
returns wrong results without any error. In a DSv2 JDBC catalog with a custom
JdbcDialect that builds SQL this way,
filter(col("EMAIL").startsWith(col("EMAIL"))) over rows MAIL@x and zzz returns
only MAIL@x. If the pattern column's name is a single character, calling
toString on the predicate throws StringIndexOutOfBoundsException.
This affects non-JDBC DSv2 sources and JdbcDialects that override
compileExpression with their own builder. Spark's built-in JDBC dialects are
fixed by SPARK-57500.
was:
When the pattern argument of `startswith`, endswith or contains is a column
rather than a string literal, V2ExpressionSQLBuilder renders the predicate as a
LIKE against a mangled copy of the column name instead of the column's value.
For example, startswith(email, email) becomes email LIKE 'mai%' ESCAPE '\'. A
DSv2 source that pushes these predicates down with V2ExpressionSQLBuilder
returns wrong results without any error. In a DSv2 JDBC catalog with a custom
JdbcDialect that builds SQL this way,
filter(col("EMAIL").startsWith(col("EMAIL"))) over rows MAIL@x and zzz returns
only MAIL@x. If the pattern column's name is a single character, calling
toString on the predicate throws StringIndexOutOfBoundsException.
This affects non-JDBC DSv2 sources and JdbcDialects that override
compileExpression with their own builder. Spark's built-in JDBC dialects are
fixed by SPARK-57500.
> Wrong LIKE pattern from startswith/endswith/contains with a column argument
> in DSv2 sources that use V2ExpressionSQLBuilder
> ---------------------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-61341
> URL: https://issues.apache.org/jira/browse/SPARK-61341
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 5.0.0
> Reporter: Eric Yang
> Priority: Major
>
> When the pattern argument of `startswith`, endswith or contains is a column
> rather than a string literal, V2ExpressionSQLBuilder renders the predicate as
> a LIKE against a mangled copy of the column name instead of the column's
> value.
>
> For example, startswith(email, email) becomes email LIKE 'mai%' ESCAPE '\'. A
> DSv2 source that pushes these predicates down with V2ExpressionSQLBuilder
> returns wrong results without any error. In a DSv2 JDBC catalog with a custom
> JdbcDialect that builds SQL this way,
> filter(col("EMAIL").startsWith(col("EMAIL"))) over rows MAIL@x and zzz
> returns only MAIL@x. If the pattern column's name is a single character,
> calling toString on the predicate throws StringIndexOutOfBoundsException.
>
> This affects non-JDBC DSv2 sources and JdbcDialects that override
> compileExpression with their own builder. Spark's built-in JDBC dialects are
> fixed by SPARK-57500.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]