Interoperability of RTTI information

[Martin von Loewis]

> To achieve this we can emit both the vtables, the virtual functions and
> the ABI entry points in an implementation's runtime and ensure
> that no object file will
        [...]
> 	use an A and B different from the rest of the executable
> all will be well.

I think this will be hard to assure. Users linking with -Bsymbolic and
integrating the C++ runtime will manage to break that (or, rather, the
ABI breaks -Bsymbolic ...)

> The only way to have multiple copies of the runtime is to statically link
> it into different libraries and then link those libraries together. Is
> this a situation which must be addressed, or can it be banned?

It might be a good idea to ban it. However, then I'd like to see an
exact specification of what is being banned... (e.g. "no executable
must include two different definition of the symbols ...") That would
essentially means that some symbols must be provided by the operating
system platform, rather than the C++ compiler implementation.

Regards,
Martin