Any solution that encodes Unicode in a way that works for some characters but fails for others has the same problem that Unicode had in Python 3. Unfortunately we're stuck with such a solution (Latin-1) and for backwards compatibility reasons we can't change it. If we were to deprecate it, we should warn for *any* data given as a Unicode string, even if it's plain ASCII (even if it's an empty string :-).

But even if we don't deprecate it, we can still change the text of the error message (but not the type of the exception used) to be more clear.

Can we please start drafting a suitable error message here?