Eclipse vs gij/libgcj, unit-at-a-time, -O2 and -O

Jeff Sturm jsturm@one-point.com
Sun Nov 30 04:25:00 GMT 2003


On 29 Nov 2003, Tom Tromey wrote:
> Jeff> So the eclipse code calls the single-argument Class.forName(),
> Jeff> which walks the stack frame to find a loader and fails,
> Jeff> presumably because _Jv_StartOfInterpreter and
> Jeff> _Jv_EndOfInterpreter no longer surround the
> Jeff> _Jv_InterpMethod::run method in generated code.
>
> This is a bogus approach and we knew it, we just were hoping to get
> away with it.

Should we enter into bugzilla then?  With a little effort we could
probably cobble together a test case too.

> I think a better approach is to register the ffi closure stubs
> instead.  Those are unique per method and they'll let us get rid of
> the linear search of interpreted stack frames.

I'm not really sure what you mean by this.  It is my understanding that
the closures are invisible to a stack trace.

BTW, Eclipse starts for me with the -fno-unit-at-a-time workaround for
interpret.cc and the rest of libgcj compiled with -O2, provided I disable
the verifier.  So things seem to be improving.

Jeff



More information about the Java mailing list