https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85078
--- Comment #7 from Jan Hubicka <hubicka at gcc dot gnu.org> --- Created attachment 43878 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=43878&action=edit patch Hi, this patch makes the testcase to work by rebuilding type inheritance graph. It is one extra walk over the cgraph but should not be terribly expensive. Looking at BINFo cleaning I also notice that we do not walk BINFO_VIRTUALS (IMO) becuase we zero them anyway. Moreover we probably want to walk BASETYPE because it is where the BINFOs are taken by devirt machinery. It may make sense to implement checker that all BINFOs and other kind of global trees are seen by free lang data when they are streamed?
