Re: SableVM on ARM Zaurus
"Grzegorz B. Prokopski" <[email protected]>
| Newsgroups | gmane.comp.java.vm.sablevm.general |
|---|---|
| Organization | SableVM - LGPL'ed Free Java VM http://sablevm.org |
| Message-ID | <[email protected]> |
On Sun, 2005-09-01 at 15:02 +0100, Cortez The Killer wrote:
> > >
> > > sablevm: thread.c:635: _svmf_resuming_java: Assertion 'status == 3 ||
> > > status == 4' failed
> >
> > Does this happen on both ARM and x86? (it'd be good if we could
> > reproduce it on x86 hardware)
> >
> > Which example was that?
> >
> > It doesn't happen on a simple "HelloWorld" program, on ARM, I presume?
>
> It does not happen on "HelloWorld" kind of programs, no. I tried to
> run the gnu-classpath examples by executing:
>
> /usr/java/bin/sablevm
> -c/usr/java/share/sablevm-classpath/examples/examples.zip
> gnu.classpath.examples.awt.Demo
>
> Until now, this worked only once, just by repeatingly trying(?!).
> After pressing the "Cursors" button the program crashed with a
> segmentation fault though.
>
> Unfortunately, I'm unable to try this on x86, because ./configure
> complains about me having wrong versions of gtk+ on my RH9
It works fine on my x86 with SableVM 1.1.8 debs installed.
Can you try debug version of sablevm and call 'sablevm -v ...."?
This will give you a hyper-flooding output so you surely should
redirect it to some place with ~500MB of space. And then seeing
last ~500 lines of this file might help us.
The other option is to attach get a stack trace somehow. Either
by running sablevm under gdb until it segfaults and then getting
backtrace ('bt'), or asking it to drop core and then using gdb to
analyze it and get the backtrace.
I only guess the segfault might be happening ie. in GTK code that
does not handle some error gracefully, because of something that
never happens on standard GTK installations on desktop machines.
> > It suggests that either a bug in SableVM, or problems with atomic
> > compare&swap implementation we have for ARM, or ... sth other?
>
> So removing the --enable-debugging-features would remove the assertion
> feature completely?
Yes.
> Shouldn't assertion be a command line option, like
> it is with Sun's JVM?
Yes. We plan to provide a configure option that will automatically
build 3 versions of libsablevm:
1. Normal, fast (default)
2. Debug, assertions, slow as hell
3. JVMDI-enabled (to support debugging via JDWP/JDI)
The right flavour will be loaded by 'sablevm' starter depending on the
options passed. But this won't happen until March, I think.
Cheers,
Grzegorz B. Prokopski
--
Grzegorz B. Prokopski <[email protected]>
SableVM - Free, LGPL'ed Java VM http://sablevm.org
Why SableVM ?!? http://sablevm.org/wiki/Features
Debian GNU/Linux - the Free OS http://www.debian.org