https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70940
Jonathan Wakely <redi at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|ASSIGNED |RESOLVED
Resolution|--- |FIXED
Target Milestone|--- |9.0
--- Comment #9 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Eric Fiselier from comment #0)
> Furthermore it seems that do_allocate returns ill-aligned pointers. It seems
> that do_allocate(s, a) returns the pointers from
> 'Allocator<char>.allocate(...)' directly even though they have no alignment
> guarantees.
Fixed by allocating a buffer with additional space. The returned pointer is
adjusted to the next alignment boundary, and a token is stored at the end of
the allocated buffer to allow removing the adjustment to retrieve the original
pointer for deallocation.
(In reply to Jonathan Wakely from comment #1)
> __null_memory_resource doesn't need to be a class template.
Fixed by using a local class.
> new_delete_resource() returns something that:
> - doesn't have the required is_equal() behaviour.
> - only uses new/delete if std::allocator uses __gnu_cxx::new_allocator.
Fixed by using new_allocator directly.
> And another one from Eric:
> Your new_delete_resource() also unnecessarily pads the allocation size
> before invoking ::operator new, but that isn't a real bug.
I'm not going to fix that for the TS. Without C++17's aligned new we can't rely
on calling new/delete directly (at least not for extended alignments). Going
via resource_adaptor<__gnu_cxx::new_allocator> gives us the alignment
guarantees, at the expense of some extra padding. I considered adding
specialization to resource_adaptor so that when using new_allocator (or
std::allocator when that uses new_allocator) it would avoid doing the padding
for fundamental alignments. Maybe at a later date.
For C++17's std::pmr::new_delete_resource() we can just use aligned new
directly and don't need to wrap new_allocator<char> and over-allocate (and if
the user uses -fno-aligned-new they don't get a new_delete_resource).
Closing as fixed on trunk.
