https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71301
--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> --- For the record, the user-provided _Tuple_impl(_Tuple_impl&&) in the specialization that terminates the recursion was added by me when fixing Bug 56785. It didn't matter at the time, because we also had a user-provided move ctor in the non-terminal case, added in 2007 by r0-85143-gba60f6f9529729. So it was never trivially move constructible. Much later, r12-3022-g0187e0d7360f32 and r12-4356-g7481021364e75b changed the move ctor to be defaulted in the derived _Tuple_impl specializations, but for ABI reasons the move ctor in the ultimate base class could not be changed.
