[
https://issues.apache.org/jira/browse/FLINK-4316?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15409384#comment-15409384
]
Stephan Ewen commented on FLINK-4316:
-------------------------------------
An alternative would be to copy the Hadoop {{Writable}} type (simple interface)
into {{link-core}}. We did that before, it works well and is a much easier
solution (no reflection work), but comes with two issues:
- We need to carry that class in the flink codebase
- There will be multiple versions of {{Writable}} and in theory, these could
lead to class cast exceptions.
> Make flink-core independent of Hadoop
> -------------------------------------
>
> Key: FLINK-4316
> URL: https://issues.apache.org/jira/browse/FLINK-4316
> Project: Flink
> Issue Type: Bug
> Components: Core
> Affects Versions: 1.1.0
> Reporter: Stephan Ewen
> Assignee: Stephan Ewen
> Fix For: 1.2.0
>
>
> We want to gradually reduce the hard and heavy mandatory dependencies in
> Hadoop. Hadoop will still be part of (most) flink downloads, but the API
> projects should not have a hard dependency on Hadoop.
> I suggest to start with {{flink-core}}, because it only depends on Hadoop for
> the {{Writable}} type, to support seamless operation of Hadoop types.
> I propose to move all {{WritableTypeInfo}}-related classes to the
> {{flink-hadoop-compatibility}} project and access them via reflection in the
> {{TypeExtractor}}.
> That way, {{Writable}} types will be out of the box supported if users have
> the {{flink-hadoop-compatibility}} project in the classpath.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
