LEWG-I in San Diego: Forward to LEWG for C++23.

P0211R3 Allocator-aware library wrappers for dynamic allocation (Thomas Köppe)

Prague 2020-02 LEWG Minutes

P0211R3 std::allocate_unique

Chair: Bryce Adelstein Lelbach

Champion: Thomas Koeppe

Minute Taker: Ben Craig

Start Review: 2020-02-13 15:52

Should allocator_new/allocator_delete use allocator_traits<>::construct/allocator_traits<>::destroy instead of placement new/destroy?

Do we want allocator_new to take a T and rebind the allocator?

Start Polling: 16:20

POLL: We must have support for arrays in allocate_unique/allocator_new/allocator_delete in the first version we ship.

Attendance:

# of Authors: 1

Author Position: N

POLL: Users should be able to name the types of the deleters returned by allocate_unique.

Attendance: 33

# of Authors: 1

Author Position: WF

That has weak consensus.

POLL: We must have allocate_unique_for_overwrite in the first version we ship.

Attendance: 33

# of Authors: 1

Author Position: N

That has weak consensus.

POLL: We must support cv-qualified types in the first version of allocate_unique/allocator_new/allocator_delete.

Attendance: 33

# of Authors: 1

Author Position: SF

That has consensus.

What allocator should allocate_unique use in the allocator_deleter:

• The original allocator (e.g. A).
• The rebound allocator used for allocation (e.g. A::rebind<T>).
• The original allocator rebound to std::byte (reduces the number of allocator types you generate).
  • This breaks fancy pointer.

End: 16:53

CONSENSUS: Bring a revision of P0211R3 (allocator-aware library wrappers for dynamic allocation), with the guidance below, to LEWG for further review.

• Array support is not required right now.
• allocator_delete should be nameable by users.
• Bikeshed allocator_delete on the LEWG mailing list.
• allocate_unique/allocator_new/allocator_delete must support cv-qualified types.
• Add allocate_unique_for_overwrite.

Given recent mailing list feedback, scheduling and backlog, this is very unlikely to be able to target c++23, removing the tag

This may need a new champion.

2022-11-11 10:00 to 11:30 UTC-10 Kona Library Evolution Meeting

P0211R4: allocate_unique

2022-11-11 10:00 to 11:30 UTC-10 Kona Library Evolution Minutes

Champion: Thomas Koeppe (in-person)

Chair: Bryce Adelstein Lelbach & Robert Leahy

Minute Taker: Steve Downey

Start: 2022-11-11 11:09 UTC-10

POLL: We should promise more committee time to pursuing P0211R4 (allocate_unique), knowing that our time is scarce and this will leave less time for other work.

Attendance: 16 (in-person) + 3 (remote)

# of Authors: 1

Author Position: WF

Outcome: Consensus in favor.

WA: I think the paper's great, I just want allocators to die in a fire.

POLL: P0211R4 (allocate_unique) should follow the Boost design.

Attendance: 16 (in-person) + 3 (remote)

# of Authors: 1

Author Position: WF

Outcome: Weak consensus in favor.

End: 11:34

Summary

Library Evolution briefly reviewed this paper, which proposes allocate_unique. The status quo design doesn't rebind allocators and as a consequences doesn't support arrays. We discussed the design of Boost's version of this facility as an alternative. No notable downside with the Boost approach was identified, and it supports more features that we felt were important, like arrays. So, we had consensus to adopt the Boost design.

Next Steps

Revise the paper, adopting the Boost design, and return to Library Evolution for further design review.