https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108053
--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> --- (In reply to Jonathan Wakely from comment #1) > This applies to all the extended floating-point types too. For these, we're normalizing them to a standard FP type early, and then have no information allowing us to return them as a handle later. So we store e.g. std::float64_t in the format args, then visiting will give you double instead of handle.
