java patch: initialize obstack before use

[Anthony Green]

Mark wrote:
BTW, I was just at ALS2000, and I was asked about GCJ's support for
dynamically loading Java bytecodes.  I know this has been thought
about, but I promised to pass along the feature request.

It's already there!  The system class loader creates vtables on the fly 
with pointers to dynamically generated closures.  The closures invoke the 
interpreter with the appropriate bytecode so you can freely mix compiled 
and interpreted classes.

The interpreter currently only works on IA-32 and IA-64 systems.  All 
that's needed to support other architectures is to implement the target 
dependent closure interface in libffi.  Contributions welcome! :-)

AG