https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124121
--- Comment #5 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #4)
> Not sure what inplace_vector is for ...
std::vector that uses internal storage, not heap allocations.
So std::inplace_vector<T, N> stores between 0 and N objects of type T. It
contains
union {
T _M_elems[N];
};
and we do not start the lifetime of the entire array, because we want it to
contain zero objects initially. The start_lifetime_as_array call creates the
array but not its subobjects, so we just have an empty shell that we can then
construct individual T objects into as the inplace_vector grows.
Maybe we can do this instead:
constexpr void
_M_init()
{
// TODO: use new(_M_elems) _Tp[_Nm]() once PR121068 is fixed
if constexpr (is_trivial_v<_Tp>)
for (size_t __i = 0; __i < _Nm; ++__i)
_M_elems[__i] = _Tp();
else if !consteval
{
#if __glibcxx_start_lifetime_as
std::start_lifetime_as_array<_Tp>(data(), _Nm);
#endif
}
else
{
// only trivial types are supported at compile time
__builtin_unreachable();
}
}
This avoids the start_lifetime_as for trivial types.
