[
https://issues.apache.org/jira/browse/HADOOP-11505?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15066842#comment-15066842
]
Colin Patrick McCabe commented on HADOOP-11505:
-----------------------------------------------
Thanks for working on this, everyone.
[~aw], if you want to run the native unit tests for hadoop-hdfs, you must build
the native part of hadoop-common first, so that you can have access to
{{libhadoop.so}}. That's why the old (pre-Yetus) build script had a hacky
piece of code that specifically compiled hadoop-common if anything native had
changed. I don't think the native components inside nativetask and yarn have
the same issue, since I don't think they actually make calls into libhadoop.
So theoretically we might be able to decouple those from the compilation of
hadoop-common. As [~cnauroth] commented, one way of doing this would be
splitting out stuff like the {{config.h}} generation code into a separate
module or file that could be invoked by all {{CMakeLists.txt}}.
As [~alanburlison] pointed out, the patch here doesn't add any additional
dependencies or change the existing ones. I think we should address the build
dependency issues in a separate JIRA.
{{MCollectorOutputHandler.cc}}: I'm confused by the {{Endium}} enum here.
First of all, shouldn't it be {{Endianness}}? "Endium" isn't a word.
Secondly, it looks hard-coded to {{LARGE_ENDIUM}}, but there is an "if"
statement. Is the intention to add additional flexibility in the future? If
we're not using this functionality right now, I would rather avoid "speculative
code" that we may or may not end up using.
> Various native parts use bswap incorrectly and unportably
> ---------------------------------------------------------
>
> Key: HADOOP-11505
> URL: https://issues.apache.org/jira/browse/HADOOP-11505
> Project: Hadoop Common
> Issue Type: Bug
> Affects Versions: 3.0.0
> Reporter: Colin Patrick McCabe
> Assignee: Alan Burlison
> Fix For: 3.0.0
>
> Attachments: HADOOP-11505.001.patch, HADOOP-11505.003.patch,
> HADOOP-11505.004.patch, HADOOP-11505.005.patch, HADOOP-11505.006.patch,
> HADOOP-11505.007.patch
>
>
> hadoop-mapreduce-client-nativetask fails to use x86 optimizations in some
> cases. Also, on some alternate, non-x86, non-ARM architectures the generated
> code is incorrect. Thanks to Steve Loughran and Edward Nevill for finding
> this.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
