https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118293
--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> --- It's possible that reusing the "trailing" capacity when pushing at the front of an empty deque will break some invariant which other members rely on (e.g we assume that popping the last element from a deque will always move the _M_start._M_cur pointer to _M_start._M_first, restoring the default-constructed state). In that case, this might be an ABI break, and then would have to be lumped in with PR 24693 as another nice thing we can't have.
