Clarification about the cons_virtual_derivation.cc tests

[Paolo Carlini]

Hi all, hi Benjamin,

today was having a look to the xfail-ed tests we have in the testsuite, in the hope of fixing some, and noticed this set, one for each bad_* class in clause 18.

As far as I can see, the reason those tests can only fail is that - contrary to the general advice from Andrew Koenig, explained in the Boost reference - the bad_* classes derive non-virtually from std::exception. If anything, that seems to me an issue with the specifications in the Standard, not with our implementation, which is straightforward and, I think, can't be changed to derive virtually instead.

Therefore, I'm a little puzzled. I'm not sure why we want to have fails here (then xfailed); I'm not sure whether we believe we can change something in our specific implementation in order to have passes, while remaining conforming. Or whether we want to pursue this issue as DR?!? Can you clarify?

Thanks,
Paolo.