[
https://issues.apache.org/jira/browse/FLINK-1664?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14394233#comment-14394233
]
ASF GitHub Bot commented on FLINK-1664:
---------------------------------------
Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/541#discussion_r27721821
--- Diff:
flink-java/src/main/java/org/apache/flink/api/java/operators/DataSink.java ---
@@ -208,6 +214,28 @@ public DataSink(DataSet<T> data, OutputFormat<T>
format, TypeInformation<T> type
return this;
}
+ private void isValidSortKeyType(int field) {
--- End diff --
I agree, there is some duplicated code for key handling, not only for sort
keys.
I am planning to rework the expression key handling in the API to make it
more consistent as of FLINK-1666 soon.
> Forbid sorting on POJOs
> -----------------------
>
> Key: FLINK-1664
> URL: https://issues.apache.org/jira/browse/FLINK-1664
> Project: Flink
> Issue Type: Bug
> Components: JobManager
> Affects Versions: 0.8.0, 0.9
> Reporter: Fabian Hueske
> Assignee: Fabian Hueske
>
> Flink's groupSort, partitionSort, and outputSort operators allow to sort
> partitions or groups of a DataSet.
> If the sort is defined on a POJO field, the sort order is not well defined.
> Internally, the POJO is recursively decomposed into atomic fields (primitives
> or generic types) and sorted by sorting these atomic fields. Thereby, the
> order of these atomic fields is not well defined (I believe it is
> lexicographic order of the POJO's member names).
> IMO, the best approach is to forbid sorting on POJO types for now. Instead,
> it is always possible to select the nested fields of the POJO that should be
> used for sorting. Later we can relax this restriction.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
