If you can somewhat solve the problem by better using the existing api, good. I am not 'stuck' on reusing str/repr*. If metavar is non-optional for non-iterable choices, the doc should say so in the entry for choices. (Does the test suite already have a testcase already for non-iterable choices + metavar?)

If you can solve it even better and remove that limitation by extending the 'default_metaver' idea from positional and optional args to choices ('choiced' args), even better.

I think the refactoring may still be needed, especially for #16418, but that is that issue.

* My main concern is that the attempt to provide helpful info to end users not hang or kill a program. IDLE used to sometimes quit while attempting to provide a nice calltip (#12510). It currently quits on Windows when attempting to warn users about bad custom configuration (#14576).