Re: [boost] Variant/Any and the never-empty guarantee

AMDG Mathias Gaunard wrote:

I see the problem now. If you move the object, then the pointer in the set doesn't point to a valid object anymore.

Would a move constructor that replaces the pointer in the set completely solve the problem?

Unfortunately, such a move constructor could throw. So we would be left right back where we started. Making a nothrow move constructor would require reimplementing std::set to allow elements to be modified in a way that changes their ordering without calling erase/insert. In Christ, Steven Watanabe