[Rd] Java on 64-bin Linux [was: cross tools (was Re: wine and build difference between R.2.4.0 and R 2.4.1 windows binaries?)]

[Hin-Tak Leung]

Simon Urbanek wrote:
> 
> On Jan 12, 2007, at 1:31 PM, Hin-Tak Leung wrote:
> 
>> Prof Brian Ripley wrote:
>>> On Fri, 12 Jan 2007, Simon Urbanek wrote:
>>>>
>>>> On Jan 12, 2007, at 7:31 AM, Hin-Tak Leung wrote:
>>>>
>>>>> I do use Java, just not in relation to R - it has been a while 
>>>>> since I played with SJava. Sun's JDK (32-bit) has been working 
>>>>> consistently.
>>>>> On FC5 x86_64 the default gcj-based JRE was a bit funny, but since 
>>>>> upgraded to FC6, I found the gcj-based JRE on x86_64 can run 
>>>>> haploview (http://www.broad.mit.edu/mpg/haploview/) reasonably 
>>>>> well. If you want 64-bit R working with a 64-bit JRE, the gcj-based 
>>>>> jre seems to be the only route.
>>>>
>>>> For the record, rJava and JRI work fine with Sun's Java 1.6 on 
>>>> 64-bit systems (fine being defined by the fact that JGR works :)). 
>>>> JGC-based JRE is still quite buggy on both 32 and 64-bit machines as 
>>>> far as we can tell, but for non-GUI task it seems to work. However, 
>>>> I don't think this is what Brian has in mind ...
>>> Sun's Java 1.6 does not run JGR or rJava on our FC5 Opteron boxes 
>>> unfortunately, which is behind my comment.  (It gives the same 
>>> problems as 1.5.0_x, which Simon also had)  An essentially identical 
>>> i686 installation (same set of FC5 RPMs) does run rJava etc.  We did 
>>> manage to get a Blackdown version of Java 1.4.2 to start rJava, but 
>>> it soon crashed.
>>
>> haploview is GUI and it does run alright-ish under the gcj-based jre 
>> under FC6. I know gcj is not as mature as blackdown or Sun's ; but 
>> given Sun's is 32-bit only, and gcj is shipped with FC6... and yes, 
>> gcj on FC5 was a bit funny.
>>
> 
> I'm running everything on Debian etch (=testing) and they are usually up 
> to date. GIJ/GCJ/GNU classpath have made astonishing progress, but they 
> are still not ready for the  prime-time. Swing & co. are just too 
> complex to have a bullet-proof implementation - even Sun took years to 
> come even close to an usable implementation ..

I don't intend to start a distribution war, but I found Debian testing
too *conservative* during my 1-week trial of Debian a couple of
years ago, and opted for Debian unstable almost right away. (and I try 
to avoid Debian/Debian hang-outs - too many egos and bickering and
just not getting anything done). Debian stable is a joke for a fossil.

As for "not ready for prime-time" and the other comments
- nothing is ready for prime time if nothing ever get tested/used.
I only disguish between usable and not-usable. FWIW, I work with
some people who describe R as not ready for prime-time as well :-P.

And you can also discard Wine as "windows is too complex to have
a bullet-proof implementation, even Microsoft took years to come even 
close to an usable implementation"...

Given that JNI with GCJ on FC6 does work with a reasonably complex
piece of software (gridengine.sunsource.net), and the GUI components 
does work with a fairly graphically-rich piece of software
(haploview), I'd say it is worth giving gcj is try. After all,
nothing is ready for prime-time if nothing ever get used/tested -
and it applies as much to GCJ as it is for R.

Just my 2-cents.

Hin-Tak