This page is a snapshot from the LWG issues list, see the Library Active Issues List for more information and the meaning of C++20 status.

2843. Unclear behavior of std::pmr::memory_resource::do_allocate()

Section: 20.5.2.3 [mem.res.private] Status: C++20 Submitter: Jens Maurer Opened: 2016-12-13 Last modified: 2021-02-25

Priority: 3

View all other issues in [mem.res.private].

View all issues with C++20 status.

Discussion:

The specification of do_allocate() (20.5.2.3 [mem.res.private] p2+p3) says:

Returns: A derived class shall implement this function to return a pointer to allocated storage (3.7.4.2) with a size of at least bytes. The returned storage is aligned to the specified alignment, if such alignment is supported (3.11); otherwise it is aligned to max_align.

Throws: A derived class implementation shall throw an appropriate exception if it is unable to allocate memory with the requested size and alignment.

It is unclear whether a request for an unsupported alignment (e.g. larger than max_align) yields an exception or the returned storage is silently aligned to max_align.

This is editorial issue #966.

[2017-01-27 Telecon]

Priority 3; Marshall to ping Pablo for intent and provide wording.

[2017-02-12 Pablo responds and provides wording]

The original intent was:

• If the alignment is supported by the implementation (see below), then it must use that alignment or else throw.
• If the alignment is not supported it must use max_align else throw.

However, the description of do_allocate might have gone stale as the aligned-allocation proposal made its way into the standard.

The understanding I take from the definition of extended alignment in (the current text of) 3.11/3 [basic.align] and "assembling an argument list" in 5.3.4/14 [expr.new] is that it is intended that, when allocating space for an object with extended alignment in a well-formed program, the alignment will be honored and will not be truncated to max_align. I think this is a change from earlier drafts of the extended-alignment proposal, where silent truncation to max_align was permitted (I could be wrong). Anyway, it seems wrong to ever ignore the alignment parameter in do_allocate().

[2017-11 Albuquerque Wednesday issue processing]

Move to Ready.

[2018-3-17 Adopted in Jacksonville]

Proposed resolution:

Change the specification of do_allocate() (20.5.2.3 [mem.res.private] p2+p3) as follows:

Returns: A derived class shall implement this function to return a pointer to allocated storage (3.7.4.2) with a size of at least bytes, aligned to the specified alignment. The returned storage is aligned to the specified alignment, if such alignment is supported; otherwise it is aligned to max_align.

Throws: A derived class implementation shall throw an appropriate exception if it is unable to allocate memory with the requested size and alignment.