[
https://issues.apache.org/jira/browse/CALCITE-5183?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17557100#comment-17557100
]
Steven Talbot commented on CALCITE-5183:
----------------------------------------
For the record, I did end up needing to use MULTIPLY (thanks!) because
TIMESTAMP_ADD (the workaround I was using) requires a TimeUnit rather than a
TimeUnitRange, which blows up in RelJson (which I am also using).
> Impossible condition in return type inference for SqlIntervalOperator?
> ----------------------------------------------------------------------
>
> Key: CALCITE-5183
> URL: https://issues.apache.org/jira/browse/CALCITE-5183
> Project: Calcite
> Issue Type: Bug
> Reporter: Steven Talbot
> Priority: Major
>
> I could be wrong here, but I don't see how to make this work, and I don't see
> any tests for it being used on the relBuilder side of the fence (where the
> "impossible" type inference is invoked for me).
>
> [https://github.com/apache/calcite/commit/03c76a7d2b896042ab417ddc36f1849f874ad3dd#diff-f6836cabfabc14be277e9f7406ab38996aaa45bac86969cecd64189c2fd7c745R57]
> requires the second argument to the Interval call to be literal wrapper
> around a SqlIntervalQualifier. In the context of RexCallBinding, I don't
> believe this is possible, since the argument has to be a RexLiteral, and
> RexLiteral exposes no way to directly wrap a value as a SqlIntervalQualifier
> like this. Again, this is where my reading of the code gets uncertain, but
> neither signature of RexBuilder.makeIntervalLiteral will give you what you
> want, and I don't see any way this could be possible.
>
> This issue means that there seems to be no way to use
> `SqlStdOperatorTable.INTERVAL` with RelBuilder.call.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
