This page is a snapshot from the LWG issues list, see the Library Active Issues List for more information and the meaning of New status.
pmr::memory_resource::do_allocate
needs clarificationSection: 20.4.2 [mem.res.class] Status: New Submitter: Jonathan Wakely Opened: 2021-11-12 Last modified: 2022-01-30
Priority: 3
View all other issues in [mem.res.class].
View all issues with New status.
Discussion:
20.4.2.3 [mem.res.private] says that pmr::memory_resource::do_allocate
returns "a pointer to allocated storage" and references 6.7.5.5.2 [basic.stc.dynamic.allocation].
But it's not really clear which parts of 6.7.5.5.2 [basic.stc.dynamic.allocation] define
"allocated storage". pmr::memory_resource::allocate
is not "an allocation function"
and not a "replaceable allocation function", so "the value returned by a replaceable allocation
function is a non-null pointer value" doesn't apply here, and neither does "different from any
previously returned value".
pmr::memory_resource::allocate
allowed to return a null pointer on success? Is it
allowed to return the same address twice, without an intervening deallocation? What about if you call
pmr::monotonic_buffer_resource::release()
, is that a deallocation?
When discussed on the reflector the consensus was that returning null should not be allowed,
it should throw an exception or return a valid dereferenceable pointer. The reference to
6.7.5.5.2 [basic.stc.dynamic.allocation] doesn't work to specify this though, so we should
restate the requirements without directly using the core wording for operator new
.
It was also suggested that returning the same value should not be allowed without an intervening
deallocation, but that "deallocation" should not only mean a call to deallocate on the resource,
but include things like pmr::monotonic_buffer_resource::release()
, and when a memory
resource's destructor returns memory to an upstream resource.
[2022-01-30; Reflector poll]
Set priority to 3 after reflector poll.
Proposed resolution: