Karamsetty Vishnu vardhan created CALCITE-6556:
--------------------------------------------------
Summary: Return Data Type of CASE Statement is Incorrect
Key: CALCITE-6556
URL: https://issues.apache.org/jira/browse/CALCITE-6556
Project: Calcite
Issue Type: Bug
Reporter: Karamsetty Vishnu vardhan
Assignee: Karamsetty Vishnu vardhan
I have encountered an issue where Trino is throwing a stale view error when it
shouldn't. My environment involves a table and a view created in hive. Here are
the DDL statements used:
Table I have created:
create table test.DECBIGINTCHECK ( ID DECIMAL(12,0), NAME varchar);
View I have created:
create view test.decview2 as select name,case when ID=1 then 123456789012 else
id end as newid from test.DECBIGINTCHECK;
As i dig deeper i found that trino is using linkedin-coral library to calculate
the data types of columns and this linkedin-coral is using apache calcite.
There the return type of "{*}case when ID=1 then 1234567890123 else id end{*}"
of statement is coming decimal(12,0) instead of decimal(19,0). This is
happening because the return type is calculated as the superset of then and
else clauses return type.
Here one clause return type is *BIGINT* and other one is {*}DECIMAL(12,0){*}.
Found the root cause is
SqlTypeFactoryImpl.leastRestrictiveSqlType function is not considering the edge
case of one type having precision and other not having precision. which is
pushing my use-case to
where the calcite returns the leastrestrictive based on the precision values
(Here precision of BIGINT is -1)
This can be fixed by adding another flow in the if-else clause where one of the
types doesn't have precision.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
