Nitpicky change. I think it's reasonable that we don't have new test coverage. Since we're enforcing a better requirement than what the standard specifies, the problem cited in the issue isn't a problem for us (https://godbolt.org/z/dsoMseMnh).

EDIT: Oops - I apparently didn't commit the actual comment. For posterity (Nicole asked me in person) I asked for formatter<T[N], ...> and formatter<const T[N], ...> to derive from the same specialization of _Formatter_base since const on the first type parameter doesn't affect _Formatter_base.

No change requested. I guess we can test this with the (duplicate of) std::formattable concept added by P2286R8. But since this will be tested alongwith P2286R8, it may be not worthy to duplicate such test.