P3081 R2 Core safety profiles for C++26

[wg21bot]

P3081R0 Core safety Profiles: Specification, adoptability, and impact (Herb Sutter)

[jfbastien]

SG15 and SG23: please consider and forward to EWG.

[Bigcheese]

SG15 wants to pursue normatively requiring implementations to offer some specific “Fix” semantics as described in P3081R0, such as “when type_safety profile is enabled, a static_cast pointer downcast remains static_cast in the source code but is interpreted as-if the code had written dynamic_cast” — SG15 will separately describe how implementations could conform

SF	F	N	A	SA
3	5	2	0	0

SG15 wants to pursue normatively requiring implementations to offer some specific “Modernize” fixits (as a new form of diagnostic) as described in P3081R0, such as “when type_safety profile is enabled, for a static_cast pointer downcast offer a mechanical fixit to change the source code into a dynamic_cast” — SG15 will separately describe how implementations could conform

SF	F	N	A	SA
2	6	2	0	0

[rogerorr]

P3081R0 was seen in SG23 on Thurs in Wroclaw

POLL: Apply the SG23 syntax decisions (Weds) in Wrocław and other SG23 direction to P3081R0, and forward P3081 to EWG as part of an initial set of Profiles targeting C++26.

Favor	Neutral	Against
23	1	0

[jfbastien]

Seen in EWG on Friday:

Poll: P3081r0 Core safety Profiles, we are interested in pursuing profiles in the C++26 timeline, including aspects of the specification which potentially have language and library impact (i.e. despite the P1000 schedule).

SF	F	N	A	SA
25	19	4	2	1

Result: consensus in favor

EWG will perform telecons (@hanickadot) between now and the next meeting to discuss individual profile aspects.

[hanickadot]

EWG seen first section of paper's wording on 2024/12/11 telecon and took (non-binding, temperature in the room) following polls:

---

D3081R1: The profile attributes enforce and apply should be separated out from P3081 and done separately, leaving the initial version of profiles to just Implementation-Defined Translation-Unit level configuration.

SF	F	N	A	SA
6	10	10	1	1

Result: Consensus

---

D3081R1: The next revision of P3081 should explore, clarify, and justify the application of profile attributes on lexical vs declaration scoping.

SF	F	N	A	SA
11	10	2	0	0

Result: Consensus

---

D3081R1: EWG prefers that profile attributes include the profiles:: namespace.

SF	F	N	A	SA
1	3	12	7	3

Result: Not consensus

[hanickadot]

D3081R1 was seen again on an EWG teleconference. No poll was taken.

[jensmaurer]

Object lifetime safety was moved to P3465 "Pursue P1179 as a Lifetime TS" #2120 .

[wg21bot]

P3081R1 Core safety profiles for C++26 (Herb Sutter)

[hanickadot]

P3081R1 was seen again on an EWG teleconference (2025-01-15). No poll was taken.

[jfbastien]

Profiles were discussed in EWG in Hagenberg on Wednesday:

• P3589 C++ Profiles: The Framework
• P3611 Dealing with pointer errors: Separating static and dynamic checking
• P3081: Core safety profiles for C++26
• P3586R0 — The Plethora of Problems With Profiles

Polls aimed at increasing consensus

Poll: P3081r2 Core safety profiles: for dynamic checking, only focus on library hardening for the first version. Do not change static checking.

SF	F	N	A	SA
27	33	9	1	0

Consensus in favor.

Poll: Merge P3589r1 “profiles framework” into P3081r2 “Core safety profiles”.

SF	F	N	A	SA
21	34	14	0	0

Consensus in favor.

Poll: P3589r1: the first version of the profiles framework should offer a way to apply a profile in a scope, particularly to be useful for non-modules code and header-only libraries.

SF	F	N	A	SA
20	36	9	4	0

Consensus in favor.

Poll: P3081 “Core safety profiles”: separate out “5.1 pointer arithmetic (bounds.1)” from the proposal into its own paper.

SF	F	N	A	SA
7	17	31	1	1

No consensus.

Polls to determine ship vehicle

Poll: P3081 “Core safety profiles”: forward (with amendments voted above) to CWG for inclusion in C++26.

SF	F	N	A	SA
10	10	2	25	29

Consensus against.

Poll: Pursue a language safety white paper in the C++26 timeframe containing systematic treatment of core language Undefined Behavior in C++, covering Erroneous Behavior, Profiles, and Contracts. Appoint Herb and Gašper as editors.

SF	F	N	A	SA
32	31	6	4	4

Consensus in favor.

[jfbastien]

Profiles were discussed again in EWG on Friday:

Poll: Forward P3589r1 “profiles framework” (with amendment for scope support voted above) to CWG for inclusion in C++26.

SF	F	N	A	SA
18	16	4	14	20

Not consensus.

Poll: Forward P3589r1 “profiles framework” (with amendment for scope support voted above) and apply it to library hardening (P3471) for inclusion in C++26 (P3611).

SF	F	N	A	SA
17	17	2	13	25

Not consensus.

[wg21bot]

P3081R2 Core safety profiles for C++26 (Herb Sutter)