[
https://issues.apache.org/jira/browse/HADOOP-10663?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14029703#comment-14029703
]
Chris Nauroth commented on HADOOP-10663:
----------------------------------------
bq. That code should work if {{mapOutput}} is a fully qualified {{Path}}, right?
Yes, that's correct. You mentioned that MAPREDUCE-5269 includes changes to
guarantee the {{Path}} instances are fully qualified. Thanks for pointing that
out. The problem is that MAPREDUCE-5269 isn't committed yet, and it's a larger
patch that will take more time to review, so I'm going to proceed with
committing MAPREDUCE-5912 for a short-term fix.
bq. Any heuristic that infers the FS from name can be defeated by a carefully
constructed future FS In other words scheme should be mandatory
In general, yes, I agree. I prefer using a fully-qualified {{Path}},
particularly around service or process boundaries where the {{Configuration}}
instances might be different. Even better would be a way to statically
type-check relative vs. fully-qualified, but this is a long-standing limitation
of the {{Path}} class and the underlying {{URI}} that it wraps.
However, we've also historically had the behavior that if the {{Path}} doesn't
have a scheme, then it gets mapped to the default {{FileSystem}} based on
configuration. Like you said, it's imperfect, but it's sometimes convenient,
and we need to maintain that behavior for backwards-compatibility. The patch
here would create a risk of bypassing that logic, which is an expected part of
the contract.
> Path.getFileSystem should identify Windows drive (c:) like files and return
> the localFS
> ---------------------------------------------------------------------------------------
>
> Key: HADOOP-10663
> URL: https://issues.apache.org/jira/browse/HADOOP-10663
> Project: Hadoop Common
> Issue Type: Bug
> Affects Versions: 3.0.0
> Reporter: Remus Rusanu
> Assignee: Remus Rusanu
> Attachments: HADOOP-10663.1.patch
>
>
> While investigating the problem with MAPREDUCE-5912 I considered that the
> fact that asking a Windows specific Path to get its file system should not
> return the default FS, but the localFS. Specifically, I believe that
> Path.initialize should identify such names as belonging to the 'file://'
> scheme (when null) and so create appropriate URIs.
> I do not have the know-how to evaluate the bigger impact of such a change.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
