P2434 R2 Nondeterministic pointer provenance

[jfbastien]

P2434R0 Nondeterministic pointer provenance (S. Davis Herring)

[erichkeane]

EWG supports:

-users verifying that this model meets their needs, particularly for pointer zap -authors work with SG22 and WG14 to align C and C++ solutions -implementation experience be obtained -usage experience be obtained

then we want to see the paper again

SF	F	N	A	SA
10	8	1	0	0

Result: Consensus

[sebastian-wittmeier-bl-ext]

link to P2434R0 is broken

[jwakely]

The link looks fine, the paper just hasn't yet been published where the wg21 link bot can find it.

[jfbastien]

Related: #895 #553 #1084 #1011

[opensdh]

@NinaRanns: Should this issue get an update from the SG22 discussion?

[NinaRanns]

probably :) SG22 had a meeting to discuss the issue. WG14 felt the wording presented in P2434 doesn't solve the problem from WG14 perspective. The room didn't find a common way forward. Notes from the meeting can be found here

[wg21bot]

P2434R1 Nondeterministic pointer provenance (S. Davis Herring)

[hanickadot]

P2434r1 Nondeterministic pointer provenance: Expand the paper to specify the special angelic behavior on std::atomic<T*>.

SF	F	N	A	SA
2	7	8	3	0

Not consensus (too many neutrals).

P2434R1 Nondeterministic pointer provenance: forward to SG1/CWG/LEWG for inclusion in C++26.

SF	F	N	A	SA
7	11	2	0	0

Consensus.

[jensmaurer]

CWG 2024-06-25: Brief discussion on differences between WG14 (superimposed pointer values) and WG21 (angelic nondeterminism) approaches. Provenance of integers (and erasure thereof with arithmetic) mentioned.

[jensmaurer]

CWG 2024-06-27 in St. Louis: Reviewed; needs updates.

[ogiroux]

SG1 in St Louis

SG1 believes that the approach of P2434R1 can address the zap problem with some additional (simpler) changes e.g.:

1. Convenience / future-proofing in casts & atomics
2. Tightening invalid-pointer semantics SF F N A SA 3 7 1 0 0 Unanimous consent

[wg21bot]

P2434R2 Nondeterministic pointer provenance (S. Davis Herring)

[xmh0511]

There is still an issue for the wording in this paper

A pointer value P is valid in the context of an evaluation E if P is a null pointer value, or if it is a pointer to or past the end of an object O and E happens after the beginning and happens before the end of the duration of the region of storage for O. […]

Consider this example:

int main(){
   auto p = new int{0};
   *p  =10;  // E
}

There is no deallocation in this program, so the expression E happens before the end of the duration is unknown, in other words, Is the pointer valid when we leak the memory?

[jensmaurer]

CWG 2024-11-22 in Wroclaw: CWG found the following example from the paper a rather surprising consequence for optimizers, and asks EWG to affirm this consequence is expressly acknowledged:

void f(T* p) {
  T* t = new T();
  if (opaque_fn(t)) {
    // *t and *p can alias here
  }
}