P2999 R3 Sender Algorithm Customization

[wg21bot]

P2999R0 Sender Algorithm Customization (Eric Niebler)

[inbal2l]

D2999R1 Sender Algorithm Customization (Eric Niebler)
Will be seen in Kona, Nov 2023 meeting

[inbal2l]

Library Evolution Telecon 2022-11-06

D2999R1: Sender Algorithm Customization

2023-11-06 Library Evolution Kona Minutes

Champion: Eric Niebler

Chair: Inbal Levi / Fabio Fracassi

Minute Taker: Ben Craig

Start: 2023-11-07

Summary

POLL: We want to add late customizations (as proposed by D2999R1)

Strongly Favor	Weakly Favor	Neutral	Weakly Against	Strongly Against
9	10	0	0	1

Attendance: 27 + 7 = 34

# of Authors: 1

Author Position: SF

Outcome: Strong consensus in favor

SA: Worried about the increasing complexity, and how it will impact user adoption. Would like to see investigation on removing early customization first

POLL: We should explore getting rid of early customization (paper required)

Strongly Favor	Weakly Favor	Neutral	Weakly Against	Strongly Against
5	9	7	2	0

Attendance: 27 + 7 = 34

# of Authors: 1

Author Position: N

Outcome: Consensus

WA: "If we just did this, we could make the whole thing simpler" around early customizations. We should just do that.

POLL: We generally approve on the customization mechanism presented in D2999R1

Strongly Favor	Weakly Favor	Neutral	Weakly Against	Strongly Against
6	12	1	1	1

Attendance: 27+7 = 34

# of Authors: 1

Author Position: SF

Outcome: Consensus

SA: Would like to see investigation in higher granularity than domains

The proposed mechanism for CPs was supported by LEWG.

Requested additions:

• Examples of usability of the sender_of concept (and - possibly - alternative design)
• The author will consider the following names, and come back with either reasoning for the existing one (in the paper) or an alternative one. Refer to the following name for the utility currently called “is_sender”/”is_receiver”:
  • is_sender, is_receiver
  • sender_tag, receiver_tag
  • enable_sender, enable_receiver

Next Steps

During the next meeting we will take the following steps:

• Review: What are the expected outcomes in different use cases (different pipelines, mixing different domains) - require more examples
• Whether or not we require the existence of an early CP (the room seems to prefer a single late CP, need to poll)
• Poll on late CP
• Poll on moving back from std:: namespace (LWG to come back with reasoning - take the following poll: POLL: Revert the recommendation by LWG: get_env, empty_env, and env_of_t are moved into the std:: namespace.)
• Poll / approve by the room the sender_of concept names

[inbal2l]

D2999R2: Sender Algorithm Customization
Scheduled for LEWG in Kona 2023 (Thursday)