That fix is currently not good... removing std::is_base_of_v<X, C> && makes it ambiguous.

I am changing

template< typename C, typename X >
requires (std::is_base_of_v<X, C> && !std::is_same_v<C,X>)
auto is( X const& x ) -> bool {
     return dynamic_cast<C const*>(&x) != nullptr;
}

to

template< typename C, typename X >
    requires (
        ( std::is_base_of_v<X, C> || 
          ( std::is_polymorphic_v<C> && std::is_polymorphic_v<X>) 
        ) && !std::is_same_v<C,X>)
auto is( X const& x ) -> bool {
    return dynamic_cast<C const*>(&x) != nullptr;
}

That will make the code unambiguous and will solve the issue.

Thanks. Interim ack: I'm going to hold off on this and other changes to cpp2util.h for a short time, including most/all of the is and as fixes, because cpp2util.h is not yet covered by the current regression tests. (The current tests only cover regressions in .cpp lowering, and don't cover actually executing the resulting programs which is what exercises the code in cpp2util.h.)

I'm about to add support for actually executing the regression test programs (in at least one compiler, three would be ideal) and check in those results. Then they'll cover the cpp2util.h fixes too and we can work through those with some automated support for knowing we aren't getting regressions.

OK, that sounds good. I am also struggling with ensuring what is working or not. I am supporting myself with a compiler explorer to ensure if new ideas are working on gcc, clang, and MSVC.

Looks good! Thanks.