I think this is not a correct change: attributes go after the noexcept-specifier, not before.

However, there are two bugs here:

1. [] () constexpr [[using std:]] noexcept {}; should be ill-formed, but we don't prevent the constexpr decl-specifier from having attributes.
2. [] () constexpr [[]] {}; is ambiguous, because there are two different attribute-specifier-seqs that could consume the attributes.

The fix would be to stop reusing decl-specifier-seq here, but this doesn't seem editorial to me.

More bugs caused by the reuse of decl-specifier-seq:

• [dcl.type]/3 says "Except in a declaration of a constructor, destructor, or conversion function, at least one defining-type-specifier that is not a cv-qualifier shall appear in a complete type-specifier-seq or a complete decl-specifier-seq.", which renders all mutable or constexpr lambda-expressions ill-formed.
• Throughout [dcl.spec], wording assumes that a decl-specifier-seq has a corresponding init-declarator-list or member-declarator-list.

I think we simply should stop using decl-specifier_seq for lambdas. It was cute when constexpr was introduced, but the fall-out is scary plus we never actually refer to this as a decl-specifier-seq in the text that follows.

I have created a clean-up patch; see #2338. I hope to get CWG to review the issue in San Diego.

@jensmaurer: does https://wg21.link/p1102r2 fix the underlying issue?

Yes, the original issue is fixed with P1102R2. The decl-specifier-seq concerns will be addressed by #2338.