> Consider if you are stopped in a method or friend function of a derived
> class.  In this case you do have access to the base -- so overload
> resolution should work.  But with this patch, it will not.
>

Gotcha. This makes sense. I have made the change.

> Maybe we could come up with some heuristic for when it would be
> acceptable and when it would not be.  And maybe overloading should be on
> the "respect access" list -- but I think for the time being we should
> still ignore it, until we have a good idea of what we want and the
> capability to implement it correctly.
>

If it becomes an issue in the future I think we can solve the problem by 
introducing a rank for private base conversions that is better than void 
conversions but worse than public conversions.

> Aside from this final issue the patch looks ok to me.

I will commit the patch as attached. I have added comments to the test 
cases to make debugging the .exp easier.