On 20/10/2020 00.28, Hans Boehm via SG5 wrote:
> Agenda:
> P2066R3, aiming for a straw poll

I've uploaded the latest version:

https://wiki.edg.com/pub/Wg21summer2020/SG5/p2066r4.html

This should also be in the upcoming mid-October mailing.

Changes in R4 since R3

    Make throwing an exception from inside a transaction undefined behavior


Note: Any throwing of an exception is considered "bad" for starters,
even if it is caught again inside the atomic block.

Jens




> Discuss Mike Spear's paper proposing allocation support.
>
> Join with Google Meet
> meet.google.com/sbj-cvgh-vnd <http://meet.google.com/sbj-cvgh-vnd>
> Join by phone
> ‪(US) +1 208-925-0196‬ PIN: ‪255 542‬#
>
> On Tue, Sep 22, 2020 at 9:18 AM Hans Boehm <boehm_at_[hidden] <mailto:boehm_at_[hidden]>> wrote:
>
> Attendees:
>
> Hans Boehm, Nathan Burgers, Victor Luchangco, Jens Maurer, Michael Scott, Mike Spear
>
> Jens provided p2066r3, which is in the currently unpublished mailing.
>
> Mike Spear provided a draft of a paper about requiring allocation support.
>
> We discussed functions allowed in atomic blocks. We're basically happy with the list, but unsure that the current wording is sufficient.
>
> Some general discussion about addition of allocation, though we postponed discussion of Mike's paper.
>
> If we add malloc, we will want containers, uniqe_ptr, String, etc.
>
> Jens - Want to keep this non-challenging for implementers.
>
> Victor - Maybe suggest an approach for adding malloc, but don't require it.
>
> Jens - Wouldn't really be part of the specification.
>
> Hans - If we did add allocation, would implementation defined behavior for out-of-memory be OK?
>
> Consensus in favor. (But no consensus to require allocation. That remains to be discussed.)
>
> How important is out-of-memory handling?
>
> Jens and Hans suggested that it's commonly not critical, beyond error reporting. And this is a sore spot in other parts of the standard as well.
>
> Could we call std::terminate?
>
> We're still in transaction, which makes this tricky.
>
> Continued consensus to stick with implementation defined behavior for out-of-memory if we add allocation.
>
> Jens - How do we proceed?
>
> Consensus: Give ourselves a month to look over and think about 2066R3. Alert SG1 to the changes. Aim to vote it out of SG5 next time, and put it back on EWG's plate.
>
> Next meeting:
>
> Oct. 20, 8 a.m. PDT, 11 a.m EDT (I suspect Europe and the US are out-of-sync with respect to daylight time?)
>
> Agenda for next meeting:
> P2066R3, aiming for a straw poll
> Discuss Mike Spear's paper proposing allocation support.
>
>