Should MacroCall remove visibility?

[ehuss]

Currently libsyntax rejects any visibility modifier on a macro invocation in ItemKind, ImplItemKind, and ForeignItemKind.

[petrochenkov]

That's correct behavior, I think.

The only similar case where macro call (foo!()) is not a whole item/statement is semicolon foo!(); and we already have issues with expanding macros with a semicolon in the token model (rust-lang/rust#61733).
Semicolon problems are reasonably solvable though, but what should be done with a pub attached to a macro invocation is not clear at all.

(Oh, also attributes #[attr] foo! { ... }, which also lead to confusing behavior incompatible with the token model if the attribute is inert.)