https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104172
--- Comment #12 from Jason Merrill <jason at gcc dot gnu.org> --- I'm persuaded by your argument that we don't need to care about compatibility with the 8.1 handling of these types, which weren't yet really usable, so we can go with the simpler patch. But I'd like to hear from IBM.
