[
https://issues.apache.org/jira/browse/CALCITE-7691?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18102236#comment-18102236
]
Julian Hyde commented on CALCITE-7691:
--------------------------------------
I agree, dropping rows whose timestamps are NULL is the least surprising
semantics. (It matches most SQL aggregate functions.)
> The behavior of SESSION table functions is unspecified for NULL timestamps
> --------------------------------------------------------------------------
>
> Key: CALCITE-7691
> URL: https://issues.apache.org/jira/browse/CALCITE-7691
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.42.0
> Reporter: Mihai Budiu
> Priority: Minor
>
> The result produced by a SESSION table has non-null timestamps.
> So it is unclear what should happen to the input rows that have null
> timestamps.
> A reasonable solution seems to be to actually drop these rows when computing
> the session windows.
> The only other possible alternative I can imagine is putting all rows with
> NULL timestamps in the same session. In that case the type inference must be
> changed to produce nullable results.
> Ideally this choice would be guided by a spec or some reference
> implementation.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
