This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: PLEASE APPROVE: patch to fix PR 5879, a serious gcc 3.1regression
>>>>> "Anthony" == Anthony Green <green@redhat.com> writes:
> I agree with Adam here. The patch he is backing out appears to have
> introduced a C++ regression for sjlj-enabled targets. It looks like we
> lose a call to _Unwind_SjLj_Unregister at some point, and fc_static ends
> up pointing to some location beyond the top of the stack. It eventually
> gets overwritten and causes problems when we call the personality
> function.
> For libgcj, I've found that initializeClass in java/lang/natClas.cc is
> the first place that $sp ends up leaving fc_static behind.
Hmm. Can you (or Adam) be more specific about the miscompilation? After
reconfiguring with --enable-sjlj-exceptions I've tried compiling
natClass.cc with and without this patch, and the only difference I see is
label numbers.
A small testcase would be great.
Jason