[dcl.type] p3 already has a footnote talking about the "implicit int rule". Let's remove the footnote.

@zygoloid, I don't understand the point on the note. I believe there are template-declaration s where we can, in fact, use a nodeclspec-function-declaration (admittedly, those are definitions, but declaration is not limited to non-defining declarations, I believe). And the note is correct in saying that a nodeclspec-function-declaration cannot appear outside of the listed contexts, although it is true that not all listed contexts will allow a nodeclspec-function-declaration in all cases. But I don't read the text of the note as attempting to say the latter.

@zygoloid: I get it now: The grammar for nodeclspec-function-declaration only allows a non-defining declaration.

@zygoloid: It seems to me that

struct S {
  template<class T> S(T);
};

recurses through member-declaration / template-declaration / declaration / nodeclspec-function-declaration and that's where the note applies. [temp] p2 says the declaration must declare a function, which it does, if [dcl.fct] is ever supposed to be applicable to constructors ("returning T"?)

Ah, yes, OK then. So the note is subtle but correct.