PR tree-optimization/45605

[Richard Guenther]

On Tue, 21 Sep 2010, Jan Hubicka wrote:

> > 
> > > Bootstrapped/regtested x86_64-linux, OK?
> > 
> > Jan
> > 
> > what you're doing is very exciting and I'm very supportive of the changes. :-)
> > 
> > But unfortunately, this patch caused a regression in obj-c++/proto-lossage-3.mm 
> > on my i386 linux, where the compiler now segfaults (most likely nothing to with 
> > Objective-C++, I noticed just because I keep an eye on these ObjC/Obj-C++ tests).
> > 
> > It all works (the compiler can compile obj-c++/proto-lossage-3.mm fine) up until 
> > exactly the subversion version number before you applied this patch. ;-)
> > 
> > Could you have a look ? :-)
> > 
> > Thanks a lot!
> > 
> > PS: Here is a stack trace in case it helps --
> > 
> > (gdb) run ./proto-lossage-3.mm 
> > Starting program: /opt/gcc/trunk2/libexec/gcc/i686-pc-linux-gnu/4.6.0/cc1objplus ./proto-lossage-3.mm
> >  sel_eq objc_hash_ptr objc_hash_string objc_compare_ptrs objc_compare_strings main __objc_gnu_init _GLOBAL__I_main
> > Analyzing compilation unit
> > 
> > Program received signal SIGSEGV, Segmentation fault.
> > gimple_fold_obj_type_ref_known_binfo (token=0, known_binfo=0xb7d99070) at ../../trunk2/gcc/gimple-fold.c:1443
> > 1443      fndecl = TREE_VALUE (v);
> 
> This seems like an incomplette binfo comming form obj-C++ FE.  Martin, do you have some idea?

Likely PR45580 which I just checked in a patch for.

Richard.