https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121506
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Why set_underlying_types? Those artificial RECORD_TYPEs aren't typedefs, and having TYPE_NAME being a TYPE_DECL with DECL_NAME then being IDENTIFIER_NODE is pretty standard thing, I think e.g. C++ FE does that all the time like that and relies on that, the C FE doesn't. But, the backends are creating the RECORD_TYPEs without knowing if it is for the C or C++ FEs, so IMHO c_type_tag should be able to cope with that. Because fixing the backends to make it C FE friendly could very well completely break it in C++.
