RE: Calling java code from C/C++ code.

["Boehm, Hans" <hans_boehm at hp dot com>]

It's essential that when the gc is loaded, the weak reference to data_start
be bound to the definition from the main executable.

The dlopen man page vaguely implies that this might happen if you link the
main program with "-rdynamic", which I couldn't find on the ld man page.

Probably the easiest workaround is to explicitly link against the GC library
instead of using dlopen to load it.  A GNU ld expert might be able to
suggest better solutions.

Hans

> -----Original Message-----
> From: Timothy Wall [mailto:twall@oculustech.com]
> Sent: Monday, March 05, 2001 1:49 PM
> To: Boehm, Hans
> Cc: Bryce McKinlay; java@gcc.gnu.org
> Subject: Re: Calling java code from C/C++ code.
> 
> 
> This is linux x86, RedHat 7.0, using gcc 3.0 branch from CVS, 
> and binutils
> from the distribution.  The java code is very simple:
> 
> public class test {
>     public static void execute() {
>         main(new String[0]);
>     }
>     public static void main(String[] args) {
>         System.out.println("Hello, world");
>         System.exit(0);
>     }
> }
> 
> I believe dlopen loads the gc, using LD_LIBRARY_PATH to find 
> it.  I have no
> idea how that affects the weak references.
> 
> Tim
> 
> "Boehm, Hans" wrote:
> 
> > > -----Original Message-----
> > > From: Timothy Wall [mailto:twall@oculustech.com]
> > > (gdb) p GC_dump()
> > > ***Static roots:
> > > From 0x0 to 0x804a1d8
> > >
> > ...
> > There's the problem.  The collector thinks that static data 
> for the main
> > program starts at location 0, and hence it's trying to scan 
> memory from 0x0
> > to 0x804a1d8.  Most of that isn't mapped.
> >
> > I assume your gcj installation works with a simple 100% 
> Java program?  This
> > is Linux/X86?  Which distribution?
> >
> > Does the garbage collector itself get loaded with a dlopen? 
>  Does that
> > affect processing of the weak references to __data_start 
> and data_start?
> >
> > Hans
>