clang allows for multiple declarations introduced by `using enum`

[Rush10233]

The following code is accepted by clang under -std=c++20:

struct S2 {
  enum B {
    A
  };
  using enum B ;
};

MSVC and GCC choose to reject it as there would be a name conflict:

<source>:5:16: error: 'using S2::B::A' conflicts with a previous declaration
    5 |   using enum B ;
      |                ^
<source>:3:5: note: previous declaration 'S2::B S2::A'
    3 |     A
      |     ^

While the using enum B statement here actually seems to be meaningless, I'm not quite sure whether this should be rejected or not.

https://godbolt.org/z/3e3aGxnWW

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: None (Rush10233)

The following code is accepted by clang under `-std=c++20`:

struct S2 {
  enum B {
    A
  };
  using enum B ;
};

MSVC and GCC choose to reject it as there would be a name conflict:

<source>:5:16: error: 'using S2::B::A' conflicts with a previous declaration
    5 |   using enum B ;
      |                ^
<source>:3:5: note: previous declaration 'S2::B S2::A'
    3 |     A
      |     ^

While the using enum B statement here actually seems to be meaningless, I'm not quite sure whether this should be rejected or not.

https://godbolt.org/z/3e3aGxnWW

[zygoloid]

Analysis:

• A using-enum-declaration is equivalent to a using-declaration for each enumerator.
• If a declaration is named by two using-declarators that inhabit the same class scope, the program is ill-formed. However, here we have a conflict between a using-declarator and a non-using declaration. There's no way to get into this situation other than for enumerators, because a using-declaration can otherwise only name a member of a base class.
• If a declaration named by a using-declaration that inhabits the target scope of another declaration B potentially conflicts with it ([basic.scope.scope]), and either is reachable from the other, the program is ill-formed unless B is name-independent and the using-declaration precedes B. However, the two declarations of the name declare the same entity, so they do not potentially conflict.

So, I think Clang is correct per the wording -- this code is valid. However, there appears to be an intent to prevent multiple / redundant declarations in class scopes in general, and this is the only way to introduce that situation with using declarations, so maybe we should ask CWG if they want this to be valid.

[zygoloid]

Sent a message to CWG asking for opinions.

[zygoloid]

This is now CWG2979, and so far it looks like the rules will be changed to make this ill-formed.