[
https://issues.apache.org/jira/browse/FLINK-1305?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14237840#comment-14237840
]
ASF GitHub Bot commented on FLINK-1305:
---------------------------------------
Github user rmetzger commented on the pull request:
https://github.com/apache/incubator-flink/pull/252#issuecomment-66112318
The reason why I added the dependency to `hadoop-common` is that I need it
to be able to instantiate the `NullWritable`.
Before that, we only needed the `Writable` interface which didn't require
any other classes from Hadoop. However the `NullWritable` is depending on a lot
of classes. I first tried it by copy-pasting classes from Hadoop, but after the
5th file or so I gave up. I don't know how many files it will be in the end.
But it quickly becomes dangerous shipping Hadoop code due to incompatible
versions.
Also, at the end of the day, people are going to have the Hadoop jars in
their classpath anyways, because flink-runtime is depending on it.
The only argument left is probably the collection based execution. I think
that one only requires flink-core and flink-java. But if somebody has an issue
with the hadoop dependency, they can exclude it.
> Flink's hadoop compatibility layer cannot handle NullWritables
> --------------------------------------------------------------
>
> Key: FLINK-1305
> URL: https://issues.apache.org/jira/browse/FLINK-1305
> Project: Flink
> Issue Type: Bug
> Components: Hadoop Compatibility
> Affects Versions: 0.7.0-incubating
> Reporter: Sebastian Schelter
> Assignee: Robert Metzger
> Priority: Critical
>
> NullWritable is a special object that is commonly used in Hadoop
> applications. NullWritable does not provide a public constructor, but only a
> singleton factory method. Therefore Flink fails when users to try to read
> NullWritables from Hadoop sequencefiles.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
