java.net.NetworkInterface.getNetworkInterfaces does not work properly on AIX with IPv6

[Steve Poole]

On Mon, 2011-10-17 at 09:05 +0100, Alan Bateman wrote:
> Jonathan Lu wrote:
> > Hi net-devs,
> >
> > I've got a problem on AIX operating system when IPv6 is enabled,which 
> > can be easily reproduced by attached test case NetworkInterfaceTest.java.
> > On AIX paltform, the /proc file system behaves differently from Linux, 
> > so we cannot just read required lines from /proc/net/if_inet6 but to 
> > call ioctl to get needed information. A patch is available for 
> > OpenJDK8 code base, see attachment patch.diff.
> >
> > Is anybody interested in this topic?
> >
> > I'm quite aware of the fact that so far there's no publicly available 
> > building scripts for OpenJDK8 on AIX operating system, but before 
> > everything settles down, is anybody willing to review the patch? any 
> > ideas about how to integrate it?
> >
> > Best regards!
> >
> This patch adds AIX specific code and I guess the bigger question is 
> whether this is the start of a stream of patches to add an AIX port (or 
> portions of) to OpenJDK. If so then I think it needs wider discussion. 
> The Porters Group (cc'ed) sponsor several porting projects in OpenJDK 
> and should be able to advise. If the intention is to get an AIX port 
> into OpenJDK then a possible route is to propose an AIX porting project, 
> similar to the BSD and MacOX X porting projects.
> 
> -Alan.

Hi Alan -   We (as in IBM and Oracle)  and we (as in you and I ) have
had some conversations about how to go forward with getting AIX support
into OpenJDK.  I'm glad your raised the question on the mailing list.


At the crux of the discussion are two points 

1 - The scale of change 
2 - The need (or not) to port Hotspot to AIX.

I'll answer the second one first. We have no interest in porting hotspot
to AIX.  IBM already has a JVM and our intent is to make OpenJDK class
libraries etc work with it.  Creating a porting project for AIX would
thus be disingenuous as it would be signalling an activity that was not
really happening  . 

Now lets look at the scale of the changes.  

To get OpenJDK to build on AIX (sans Hotspot) requires  about 300 lines
of code change along the lines of extending #ifdefs to include AIX as
appropriate (BTW - since we really want to improve OpenJDK portability
generally we have and will try to recode these changes as being
capability based rather than platform based)  Beyond those simple
changes there is about an extra 3000 lines in new files.  These new
files cover building on AIX and providing support for AIX filesystems
etc as seen below


make/common/Defs-aix.gmk  
make/common/shared/Defs-aix.gmk  
make/common/shared/Compiler-xlc_r.gmk
src/solaris/native/sun/nio/fs/genAixConstants.c                        
src/solaris/native/sun/nio/fs/AixNativeDispatcher.c                    
src/solaris/classes/sun/nio/fs/AixDosFileAttributeView.java            
src/solaris/classes/sun/nio/fs/AixFileStore.java                       
src/solaris/classes/sun/nio/fs/AixFileSystem.java                      
src/solaris/classes/sun/nio/fs/AixFileSystemProvider.java              
src/solaris/classes/sun/nio/fs/AixNativeDispatcher.java                
src/solaris/classes/sun/nio/fs/AixUserDefinedFileAttributeView.java    
src/solaris/classes/sun/tools/attach/AixVirtualMachine.java            
src/solaris/classes/java/lang/UNIXProcess.java.aix                     
src/solaris/classes/sun/nio/ch/AixAsynchronousChannelProvider.java     
src/solaris/classes/sun/nio/ch/AixEvent.java                           
src/solaris/classes/sun/nio/ch/UnixEvent.java                          
src/solaris/classes/sun/nio/ch/UnixEventPort.java                      
src/solaris/native/sun/nio/ch/UnixEventPort.c                          
src/solaris/native/sun/tools/attach/AixVirtualMachine.c                



So really we don't see any need to have a porting project for AIX.

AIX is very close to both linux and solaris and IBM will maintain the
minimal AIX specific code as necessary.  Once we've recoded the majority
of the #ifdef style changes to be capability based there will be little
left to worry about.  In any case we wouldn't expect OpenJDK developers
who did not have access to an AIX machine to be concerned with breaking
AIX.  That's up to us to fix if required.   

Finally,  I've asked the IBM developers who are working in this space to
get these changes made visible on the mailing lists so you can all see
the gory details.   I'm sure there will be opportunities to reduce the
change sets down even further. 

Cheers

Steve