Proposed resolution (approved by CWG 2026-09-11):

Change in 13.8.1 [temp.res.general] bullet 6.6 as follows:

The program is ill-formed, no diagnostic required, if

• ...
• every there exists a template parameter pack P of a variadic template TT such that no valid specialization of, ignoring static_assert-declarations that fail, can be generated for TT a variadic template requires an empty template parameter pack when P is non-empty, or
• ...

(From submission #878.)

The phrasing in 13.8.1 [temp.res.general] bullet 6.6 is not parallel to similar phrasings in bullets 1-3:

The program is ill-formed, no diagnostic required, if

• ...
• every valid specialization of a variadic template requires an empty template parameter pack, or
• ...

For example, this bullet is vacuously true (but unintentionally so) if there is no valid specialization at all.

Also, a similar rule for empty structured binding packs is missing, although the example in 9.7 [dcl.struct.bind] paragraph 3 appears to indicate support for these. However, consider:

  int f(auto x) {
    auto [...y] = x;
    union U : decltype(y)... {}; // unions cannot have base classes
    return 0;
  }

  struct S {};
  int z = f(S());

This might be addressed by 13.7.4 [temp.variadic] paragraph 7, though, together with the existing rule about valid specializations for the function template f.