[
https://issues.apache.org/jira/browse/HADOOP-5700?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12714531#action_12714531
]
Tsz Wo (Nicholas), SZE commented on HADOOP-5700:
------------------------------------------------
{quote}
I'll prepare a patch to address the above things.
On a side note, what is the policy for dealing with updated patches after a
JIRA has already been committed? As far as I knew, discussion of a better/new
patch is supposed to go on in a new JIRA.
{quote}
Thanks, Todd. I think the right way is to revert the committed patch and work
on a new patch in this issue.
bq. I strongly disagree on the performance issue. I'd go into detail about why
this cannot possibly cause a performance problem, but clearly this is an
idealogical debate rather than a practical one for this case.
For practical reasons, we need numbers. Without measurement, this becomes a
debate. getPathComponents(..) is one of the central internal methods in
Namenode. All accesses to the namespace use it. Many of the accesses acquire
the namespace lock beforehand. Please try measuring the time required for lsr
on a directory with 10 millions descendants before and after the patch.
bq. ... a constant-time single-byte comparison.
When we check the code of String.startWith(String), it is not a single-byte
comparison. If the patch uses the return-value of getPathNames(..), I won't
have problems on it.
bq. I think of IAEs as stricly more informative than NPEs for the developer who
has committed the error.
I agree that IAEs is more informative than NPEs for the developers but it is
never worth to benifit developers by sacrificing the performance of production
systems.
> INode.getPathComponents throws NPE when given a non-absolute path
> -----------------------------------------------------------------
>
> Key: HADOOP-5700
> URL: https://issues.apache.org/jira/browse/HADOOP-5700
> Project: Hadoop Core
> Issue Type: Bug
> Components: dfs
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Priority: Minor
> Fix For: 0.21.0
>
> Attachments: HADOOP-5700.txt
>
>
> If you pass a path that doesn't start with '/' to INode.getPathComponents, it
> throws a NullPointerException. Instead it should throw
> IllegalArgumentException to make it clear that absolute paths are required in
> this code.
> The attached patch fixes this, clarifies, the javadoc, and adds a test case.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
