CWG2986 [basic.life] Creating object within a mutable member subobject of a const complete object

Full name of submitter (unless configured in github; will be published with the issue): Jiang An

Reference (section label): [basic.life]

Link to reflector thread (if any):

Issue description:

Per [basic.life] p12, the following program has undefined behavior.

int main() {
  struct S {
    alignas(int) mutable unsigned char buf[sizeof(int)];
  };
  const S s{};
  ::new ((void*)s.buf) int{42};
}

Disallowing creating objects within a mutable member looks like an oversight, since it seemingly defeats the purpose of mutable.

We should also determine whether the behavior is well-defined if a union type has both mutable and non-mutable non-static data members, and an object is created within the mutable member of such a const complete union object.

Implementations accept the following code (Godbolt link), although it includes UB (per current rules) in constant evaluation.

constexpr bool test1() {
  union U {
    int x;
    mutable float y;
  };
  constexpr U u{.x = 42};
  u.y = 42.0f;
  return true;
}
static_assert(test1());

Suggested resolution:

@jensmaurer

@jensmaurer, I think some clarification here will help (to reinforce that we do mean UB for the non-mutable case).

@hubert-reinterpretcast

@hubert-reinterpretcast , could you please add more detail which direction you want here?

The link for the first example talk about "transparently replaceable" and re-using pointers. However, there are no pointers in sight here, so I'm confused. Maybe the link is stale.

However, https://eel.is/c++draft/basic.life#12 has no carve-out for mutable subobjects, so this object creation is undefined behavior. Do you wish to have a carve-out for mutable subobjects here?

Changing the active member of a const union seems a situation the standard doesn't talk about at all; should we forbid that?

The link for the first example talk about "transparently replaceable" and re-using pointers. However, there are no pointers in sight here, so I'm confused. Maybe the link is stale.

Yeah. The paragraph used to be p10 but is now p12.

However, https://eel.is/c++draft/basic.life#12 has no carve-out for mutable subobjects, so this object creation is undefined behavior. Do you wish to have a carve-out for mutable subobjects here?

Yes.

Changing the active member of a const union seems a situation the standard doesn't talk about at all; should we forbid that?

That technically is object creation (https://wg21.link/class.union.general#5), so it is already forbidden.

CWG2986

Navigation Menu

Repository navigation

CWG2986 [basic.life] Creating object within a mutable member subobject of a const complete object #367

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

CWG2986 [basic.life] Creating object within a mutable member subobject of a const complete object #367

Description

Activity

hubert-reinterpretcast commented on Jan 20, 2025

jensmaurer commented on Jan 20, 2025

frederick-vs-ja commented on Jan 20, 2025

hubert-reinterpretcast commented on Jan 20, 2025

jensmaurer commented on Jan 20, 2025

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions