P2214 A Plan for C++23 Ranges

[brycelelbach]

P2214R0 A Plan for C++23 Ranges (Barry Revzin, Conor Hoekstra, Tim Song)

[brycelelbach]

P2214R0: A Plan for C++23 Ranges

2020-10-27 Library Evolution Telecon Minutes

Chair: Bryce Adelstein Lelbach

Champion: Barry Revzin

Minute Taker: Ben Craig

Start: 2020-10-27 10:10 Pacific

Does this proposal have:

• Examples?
  • Yes (in slides).
• Implementation experience?
  • Yes (Ranges v3).
    • Note: Ranges v3 doesn't have adjacent, adjacent_transform, and filter_map.
• Usage experience?
  • Yes (Ranges v3).
• Deployment experience?
  • Yes (Ranges v3).
• Discussion of prior art?
  • Yes.
• Wording?
  • Only for some parts.

Naming for "two at a time" zip algorithms needs to be considered:

• Current: adjacent (we may want to use this for "n at a time").
• Alternatives: pairwise, zip_with_next.

Naming for "fold" needs to be considered.

Should we have a form of "fold" which does not take an initial value and requires a non-empty range?

How did you determine which features were most fundamental? Can you provide data that backs your categorization up?

Should we expose the ranges pipeline mechanism to users, so that they could write some of these things themselves?

Should we prioritize allowing users to define pipeable things that compose with ranges?

Two key questions:

• Do we want to prioritize some set of ranges work for C++23?
• Do we want to prioritize this set (P2214R0 Tier 1) of ranges work for C++23?

POLL: Prioritize exposing a mechanism that allows users to write range adapters that compose with standard range facilities over other ranges work.

Strongly Favor	Weakly Favor	Neutral	Weakly Against	Strongly Against
9	12	0	1	0

Attendance: 26

# of Authors: 3

Author Position: 3x WF

Outcome: That has consensus in favor.

WA: I think ranges::to is the most important thing.

POLL: Tier 1 of P2214R0 (A Plan for C++23 Ranges) is a reasonable set of ranges work to prioritize relative to the Tier 2 and Tier 3 ranges work.

Strongly Favor	Weakly Favor	Neutral	Weakly Against	Strongly Against
7	12	1	0	0

Attendance: 26

# of Authors: 3

Author Position: 3x SF

Outcome: That has strong consensus in favor.

POLL: Tier 1 of P2214R0 (A Plan for C++23 Ranges) should be treated as a priority for C++23 (P0592 bucket 1 item), prioritized over other improvements to existing features (P0592 bucket 2 items) and new features (P0592 bucket 3 items).

Strongly Favor	Weakly Favor	Neutral	Weakly Against	Strongly Against
3	2	3	4	2

Attendance: 26

# of Authors: 3

Author Position: SF, WF, N

Outcome: That has consensus against.

POLL: Tier 1 of P2214R0 (A Plan for C++23 Ranges) should be treated as improvements to an existing feature (P0592 bucket 2 item), prioritized over new features (P0592 bucket 3 items).

Strongly Favor	Weakly Favor	Neutral	Weakly Against	Strongly Against
8	8	0	1	0

Attendance: 26

# of Authors: 3

Author Position: 2x SF, WF

Outcome: That has consensus in favor.

WA: I think the most important feature is the ability for users to interoperate with what we already have; once this has that, I may feel differently.

We will classify Tier 1 of P2214R0 as a bucket 2 item.

We should consider reactivating the Ranges Study Group.

End: 11:38

Summary

We discussed P2214R0, which describes a variety of ranges work, inspired by Ranges V3, that we should pursue to build on top of C++20 ranges. The paper classifies the work into three tiers; the Tier 1 group is suggested as priorities for C++23. There was some technical discussion about specific features (such as chunk/slice and fold), but primarily we discussed the overall plan and what we should prioritize, relative to both other ranges work and to other Library Evolution work in general. It was mentioned that the proposed range adapters cannot be written by users in C++20. We had consensus that exposing a mechanism that allows users to write range adapters that compose with standard range facilities should be the highest priority ranges work. Once we have such a mechanism, users would be able to provide their own versions of all the proposed features. We discussed whether Tier 1 of P2214R0 should be proposed as an addition to the plenary approved set of C++23 priorities (P0592), however we did not have consensus for this. We did have consensus on the shape Tier 1 group of features in P2214R0, and agreed that they should be treated and prioritized as improvements to an existing feature (P0592 bucket 2 item). Finally, we briefly discussed the idea of reactivating the Ranges Study Group to drive this work, especially given that an informal group has been meeting regularly since the summer to pursue ranges enhancements.

Outcome

Revise P2214R0 (A Plan for C++23 Ranges), adding a mechanism for users to write range adapters as (at least) a Tier 1 item. Work on Tier 1 features will be considered as improvements to an existing feature (P0592 bucket 2 item).

[wg21bot]

P2214R1 A Plan for C++23 Ranges (Barry Revzin, Conor Hoekstra, Tim Song)