This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: libjava test suite keeps getting stuck
>>>>> "Zack" == Zack Weinberg <zackw@Stanford.EDU> writes:
Zack> jvm.h was the first header I found which was included by all the
Zack> C++ files that needed to get hit by the pragma, and marked
Zack> "implementation internals"; there may be a better place.
If we want all native code, both ours and random CNI users, to be
affected by this pragma then cni.h is probably the best choice.
Or maybe java/lang/Object.h.
Zack> I don't know where to document the new pragma.
At least in the C++ manual, I'd imagine.
CNI is documented in libjava/doc/cni.sgml.
Documenting it there would also be nice.
Zack> How is it that 'throw new <object>', which is all over the C++
Zack> code in libjava, generates calls to _Jv_AllocObject not _Znwj?
Magic in g++.
Zack> libjava:
Zack> * Makefile.am (libgcj.la): Remove libsupc++convenience.la from
Zack> link line.
Zack> * Makefile.in: Regenerate (by hand).
Zack> * include/jvm.h: Insert #pragma GCC java_exceptions here.
Zack> * prims.cc (_Jv_ThisExecutable): Use _Jv_Malloc.
Zack> * java/lang/natClassLoader.cc (_Jv_RegisterInitiatingLoader):
Zack> Likewise.
These patches look fine to me.
I'd like to see this go in to the branch once the EH code is moved
over, too. This will solve the libstdc++/libsupc++ dependency
problem.
Tom