OpenOffice and gcj at runtime

Thu Dec 9 16:31:00 GMT 2004

Caolan McNamara writes:
 > On Tue, 2004年12月07日 at 09:35 -0700, Tom Tromey wrote:
 > > >>>>> "Caolan" == Caolan McNamara <caolanm@redhat.com> writes:
 > > 
 > > Caolan> with (FC3) gcj/g++ 3.4.2 this example fails
 > > Caolan> with (rawhide) gcj/g++ 3.4.3 this example fails
 > > Caolan> with gcj/g++ from head (20041126) this example passes.
 > > 
 > > With the FC2 gcj, this does something at least:
 > > 
 > > opsy. type gcj
 > > gcj is /usr/bin/gcj
 > > opsy. rpm -q gcc-java
 > > gcc-java-3.3.3-7
 > > opsy. sh runme.gcj.sh
 > > registerInterfaceA is 959e668
 > > registerInterfaceB is 95b47e0
 > > A: registerInterface
 > > A: registerInterface
 > 
 > > Some versions of gcj did have a bug when calling an interface method
 > > via JNI. Is that what you're running into? Do you get a
 > > NullPointerException or something?
 > > 
 > 
 > That's what I want all right, and forcing the example to link directly
 > against libgcj.so.4 gives me the same results as you had, so libgcj.so.5
 > and .6 are the culprits, while libgcj.so.4 is behaving itself.
 > 
 > > We can fix this on some branch or another, especially since it seemed
 > > to work in 3.3 (regressions get special dispensation, usually).
 > > This example doesn't work in the BEA JDK :-)
 > 
 > Excellent that its a regression :-)
 > 
 > > Caolan> Mixing gcj from head and g++ 3.4.2 and the example fails, which is
 > > Caolan> more-or-less what I was doing with OOo (started with 3.4.2 and moved to
 > > Caolan> the head gcj because of the URL snafu) :-) I deserved that I guess, can
 > > Caolan> hardly complain about doing something odd like that.
 > > 
 > > Were there changes to libgcc? If not, I would actually expect this to
 > > work ok. It seems to me that you wouldn't hit C++ ABI problems as the
 > > JNI stuff is all inlined...
 > > 
 > > What goes wrong?
 > > 
 > > It would be nice to get this approach working, because most of the
 > > other java stuff we're looking at (for, say, FC4) will most likely be
 > > compiled with gcj4.
This is also broken with CVS HEAD.
I suggest this patch. The idea is pretty simple: if we have no method
index, and no code pointer, look the method up by name.
Andrew.
2004年12月09日 Andrew Haley <aph@redhat.com>
	* java/lang/reflect/natMethod.cc (_Jv_CallAnyMethodA): In the case
	of virtual dispatch, if we have no method index, and no code
	pointer, look a method up by name.

Index: natMethod.cc
===================================================================
RCS file: /cvs/gcc/gcc/libjava/java/lang/reflect/natMethod.cc,v
retrieving revision 1.36.6.1
diff -p -4 -c -r1.36.6.1 natMethod.cc
*** natMethod.cc	26 Feb 2004 15:21:11 -0000	1.36.6.1
--- natMethod.cc	9 Dec 2004 16:26:57 -0000
*************** _Jv_CallAnyMethodA (jobject obj,
*** 463,489 ****
 JvFail ("Unknown ffi_call return type");
 break;
 }

! void *ncode;

 // FIXME: If a vtable index is -1 at this point it is invalid, so we
 // have to use the ncode. 
 //
 // This can happen because methods in final classes don't have
 // vtable entries, but _Jv_isVirtualMethod() doesn't know that. We
 // could solve this problem by allocating a vtable index for methods
 // in final classes.
! if (is_virtual_call 
! && ! Modifier::isFinal (meth->accflags)
! && (_Jv_ushort)-1 != meth->index)
! {
 _Jv_VTable *vtable = *(_Jv_VTable **) obj;
! ncode = vtable->get_method (meth->index);
! }
! else
! {
! ncode = meth->ncode;
 }

 try
 {
--- 463,498 ----
 JvFail ("Unknown ffi_call return type");
 break;
 }

! void *ncode = meth->ncode;

 // FIXME: If a vtable index is -1 at this point it is invalid, so we
 // have to use the ncode. 
 //
 // This can happen because methods in final classes don't have
 // vtable entries, but _Jv_isVirtualMethod() doesn't know that. We
 // could solve this problem by allocating a vtable index for methods
 // in final classes.
! if (is_virtual_call)
! { 
 _Jv_VTable *vtable = *(_Jv_VTable **) obj;
! if ((_Jv_ushort)-1 == meth->index)
! 	{
! 	 if (ncode == NULL)
! 	 // We have no vtable index, and we have no code pointer.
! 	 // Look method up by name.
! 	 ncode = _Jv_LookupInterfaceMethod (vtable->clas, 
! 					 meth->name,
! 					 meth->signature);
! 	}
! else
! 	{ 
! 	 // We have an index. If METH is not final, use virtual
! 	 // dispatch.
! 	 if (! Modifier::isFinal (meth->accflags))
! 	 ncode = vtable->get_method (meth->index);
! 	}
 }

 try
 {