[kaffe] ARM and kaffe

[Jim Pick]

Kevin D. Kissell wrote:
>>>Oh well, I am working through similar issue. To make Kaffe workiing
>>>somehow set into configuration
>>>'--with-engine=interp'. It'll solve your problem but makes your kaffe
>>>working terrible slowly. The real problem is somewhere into JIT3 module
>>>and it is a memory management issue. I've sent a few letters to the guy
>>>whom ported it but no answer what so ever.
>>
>>I tried already but no luck.
> 
> 
> I had kaffe JIT3 95% working with MIPS after hacking up the 1.0.7 
> source base back in 2002.  It failed 6-8 regression tests, but ran well
> enough to handle the embedded CaffeineMark benchmarks.  Then 1.1.0
> came out, which was a huge step backwards for non-x86 JIT3, and 
> after beating my head against it for a week or two, I gave up.  I was
> able to determine that incomplete internal data structures containing
> null pointers were being used, but couldn't figure out why, and whoever
> maintained JIT3 apparently couldn't care less about non-PC platforms.

I'll use this as an opportunity to drop a note to the list on what I
plan to be hacking on in the next few months.

If you read my Advogato diary, you might have read that after 6 years in
California, I'm quitting my job and I'm moving back home to Vancouver,
British Columbia, Canada.  I should be finished with the move in
mid-April.  I've been saving, so I'm planning to take at least 6 months
to just work on my own projects and do a little freelance  work on the
side to cover expenses.

The #1 project I want to do is build a full blown regression testing
setup for Kaffe.  It will be useable by other projects as well.  But
Kaffe is going to be the focus and the original guinea pig.

Really, I started working on this particular problem 5 years ago when I
was still at Transvirtual.  I was developing a testrig for testing Kaffe
for in-house use, but it wasn't very well developed by the time the
company shut down.  About 3 years ago, I also put some scripts together
for Kaffe, but there was too half-baked and I never deployed it publicly
because I just didn't have enough time to fully develop it.  (I think I
described some of my ideas to Dalibor at FOSDEM two years ago).  The
last couple of years, I've done a fair amount of similar work building
in-house systems for Digeo.  So it's sort of unfinished business for me...

The difference this time is that I've got time, money and server space -
so I think I'll be able to pull it off.

I've also got some ARM, MIPS, SuperH, and PowerPC hardware I've
collected over the years that I'd like to put online and dedicate to
Kaffe testing.

I'm going to set up a wiki in few weeks with all my ideas - it would be
a pretty long email if I list them all here.

To the end users, the system will look like an ordinary website where
they can log in, setup tests, and run them.  I'll be able to provision
Xen sessions on the fly for testing things like various x86 Linux
distributions, *BSD, Solaris, etc.  I'm aiming for a "folksonomy" model
of doing distributed testing.

There will be data collection and querying facilities -- I've been
playing around a lot with RDF, and I think I'm going to try to use it.
Think of it as a replacement for SQL - but with easier import/export and
 Web 2.0 "mashup" possibilities.  It will be really, really easy to
build web pages that query the RDF database, and generate reports on how
"healthy" Kaffe is for a particular target.

It's quite an ambitious plan -- but I'm going to start small.  The first
thing I'm going to try to nail is getting decent regression testing for
x86, then ARM and MIPS, because those are the platforms I want to use.  :-)

Oh, and everything will be released as free software too, in case you
had any doubts...

Cheers,

 - Jim