[
https://issues.apache.org/jira/browse/CALCITE-2606?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16637167#comment-16637167
]
Julian Hyde commented on CALCITE-2606:
--------------------------------------
I'm not sure we should even allow AND and OR to have constantNull as an
argument. constantNull is a literal of type NULL, not boolean. We should
require that the argument has type BOOLEAN. This probably happens in
SqlToRelConverter but is not done automatically when tests call RexBuilder
automatically.
Same goes for nullInt and nullVarchar in RexProgramBuilderBase (used by
RexProgramTest).
One of those builder methods should throw, and force the caller to get their
arguments into shape.
> Return type inference for and/or may identify result type incorrectly
> ---------------------------------------------------------------------
>
> Key: CALCITE-2606
> URL: https://issues.apache.org/jira/browse/CALCITE-2606
> Project: Calcite
> Issue Type: Bug
> Reporter: Zoltan Haindrich
> Assignee: Zoltan Haindrich
> Priority: Major
>
> In case in an AND expression a null constant appears earlier than a nullable
> boolean; the return type is set to NULL instead of boolean.
> https://github.com/apache/calcite/blob/c39bfaa02a06ac91575076a6e74f29863923f5eb/core/src/main/java/org/apache/calcite/sql/type/ReturnTypes.java#L198
> The problem can be reproduced; however I'm not sure if it would happen
> naturally or not - I've discovered it during enhancing case simplification.
> {code}
> @Test
> public void testUsageOfConstantNull() {
> RexLiteral constantNull = rexBuilder.constantNull();
> RexNode node1 = or(constantNull, vBool());
> assertThat(node1.getType(), is(vBool().getType()));
> RexNode node2 = or(vBoolNotNull(), constantNull);
> assertThat(node2.getType(), is(vBool().getType()));
> RexNode node3 = or(constantNull, constantNull);
> assertThat(node3.getType(), is(constantNull.getType()));
> }
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
