https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68761
Jakub Jelinek <jakub at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |jakub at gcc dot gnu.org --- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Please report that into a separate PR, this one is graphite related, your is some C++17 C++ FE bug. What is going on is that cp_genericize_r first converts changes a by-reference RESULT_DECL into INDIRECT_REF of it, and then (dunno if because of tree sharing or some constexpr stuff) at some point that IL is passed again to cp_genericize_r again and again performs the same change, so we in the end have two nested INDIRECT_REFs and the types don't match any longer.