[
https://issues.apache.org/jira/browse/HADOOP-6295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12762266#action_12762266
]
Hong Tang commented on HADOOP-6295:
-----------------------------------
One possible deficiency in the current native code compilation is that On
modern 64-bit capable intel processors, we would compile either 32-bit or
64-bit native library depending on the version of JVM being used by ant.
Typically, we invoke the build process twice, with 32-bit and 64-bit jvm
without cleaning the build directory, so that the release version can be used
without worrying about which version of jvm is used in the cluster.
Might it be better if we just build both 32-bit and 64-bit versions up front?
> Build process from tarball is flawed
> ------------------------------------
>
> Key: HADOOP-6295
> URL: https://issues.apache.org/jira/browse/HADOOP-6295
> Project: Hadoop Common
> Issue Type: Bug
> Components: native
> Reporter: Allen Wittenauer
>
> Currently, the natively compiled components have a requirement of building
> from within ant. To make matters worse, after talking this over with Lee, it
> appears that the ant build assumes to be executed from a svn-checkout and not
> from the tarballs that are generated. We need to
> a) start packaging up the svn repo trees as src tarballs
> b) separate out the compiled bits out from underneath ant completely
> c) be able to create real builds from the tarball
> As an admin, I shouldn't have to reach for svn to make Hadoop work optimally
> on non-Linux.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
