[
https://issues.apache.org/jira/browse/CALCITE-5884?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17750079#comment-17750079
]
Mihai Budiu commented on CALCITE-5884:
--------------------------------------
There is something unusual going on here.
First of all, I notice that the RexToLixTranslator does *not* look at the
function return type inference rules.
It has its own NullPolicy for some methods, which it uses to deduce when a
method returns null.
However, none of the NullPolicies that exist seems to match the behavior of
this method.
This method returns a nullable result iff any of ARG0 or ARG1 is nullable.
Doesn't the nullPolicy information duplicate some of the information in the
ReturnTypes? In this case the two are contradictory.
I appreciate any suggestions on how to fix this.
> Type Inference rule for ARRAY_TO_STRING is incorrect
> ----------------------------------------------------
>
> Key: CALCITE-5884
> URL: https://issues.apache.org/jira/browse/CALCITE-5884
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.35.0
> Reporter: Mihai Budiu
> Priority: Trivial
>
> This is the current definition of the function ARRAY_TO_STRING in
> SqlLibraryOperators:
> {code:java}
> /** The "ARRAY_TO_STRING(array, delimiter [, nullText ])" function. */
> @LibraryOperator(libraries = {BIG_QUERY})
> public static final SqlFunction ARRAY_TO_STRING =
> SqlBasicFunction.create(SqlKind.ARRAY_TO_STRING,
> ReturnTypes.VARCHAR_NULLABLE,
> OperandTypes.STRING_ARRAY_CHARACTER_OPTIONAL_CHARACTER);
> {code}
> So the result is nullable if any of the arguments is nullable. However, the
> nullability of the last argument does not influence the result nullabillity:
> a NULL value for the third optional argument will not cause a NULL value to
> be output.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
