[
https://issues.apache.org/jira/browse/FLINK-5884?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15997632#comment-15997632
]
ASF GitHub Bot commented on FLINK-5884:
---------------------------------------
Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/3808#discussion_r114808765
--- Diff:
flink-libraries/flink-table/src/main/scala/org/apache/flink/table/expressions/fieldExpression.scala
---
@@ -150,3 +133,30 @@ case class WindowReference(name: String) extends
Attribute {
}
}
}
+
+abstract class TimeAttribute(val expression: Expression)
+ extends UnaryExpression
+ with NamedExpression {
+
+ override private[flink] def child: Expression = expression
+
+ override private[flink] def name: String = expression match {
+ case UnresolvedFieldReference(name) => name
+ case _ => throw new ValidationException("Unresolved field reference
expected.")
+ }
+
+ override private[flink] def toAttribute: Attribute =
+ throw new UnsupportedOperationException("Time attribute can not be
used solely.")
+}
+
+case class RowtimeAttribute(expr: Expression) extends TimeAttribute(expr) {
--- End diff --
I think we should not allow `TimeAttribute` in batch environments.
In stream environments, there are different cases to distinguish:
- rowtime on new attribute (i.e., defining it as an additional attribute at
the end): Adds a new logical attribute to the schema
- rowtime on existing attribute: Replaces the existing (physical) attribute
with a new logical attribute. The reason is that data might contain a timestamp
attribute which should be replaced by the logical timestamp attribute. If we
would only allow to define rowtime on non-existing attributes, we would have
two timestamp attributes, one that can be used as a time indicator and one that
can't. I agree, this is not a super nice solution, but the best we could think
of. Please let us know if you have a better idea. Btw. I expect most queries to
consume data from TableSources. Those won't have this problem.
- proctime on new field (i.e., defining it as an additional attribute at
the end): adds a new logical attribute
- proctime on existing field: should be forbidden, IMO.
> Integrate time indicators for Table API & SQL
> ---------------------------------------------
>
> Key: FLINK-5884
> URL: https://issues.apache.org/jira/browse/FLINK-5884
> Project: Flink
> Issue Type: New Feature
> Components: Table API & SQL
> Reporter: Timo Walther
> Assignee: Timo Walther
> Priority: Blocker
> Fix For: 1.3.0
>
>
> We already discussed the need for a proper integration of time indicators
> (event-time or processing-time) for both the Table API & SQL on the ML:
> http://apache-flink-mailing-list-archive.1008284.n3.nabble.com/DISCUSS-Table-API-SQL-indicators-for-event-and-processing-time-tp15927.html
> This issue will track the progress. I will work on a design document how we
> can solve this issue.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
