http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59775
--- Comment #8 from Jan Hubicka <hubicka at gcc dot gnu.org> --- OK, this testcase shows the get_binfo_at_offset bug I mentioned in the first virtual inheritance PR. We call get_binfo_at_offset for binfo of C looking for A at offset 64. This is correct query, but becase get_binfo_at_offsets walks fields and then it tries to match binfos, it actually looks up field representing A already in C and does not find A among direct bases of C. I suppose we will need to dive into the bases in get_binfo_at_offset same way as ipa-devirt does.