This page is a snapshot from the LWG issues list, see the Library Active Issues List for more information and the meaning of Resolved status.
Section: 20.3.2.2.5 [util.smartptr.shared.mod] Status: Resolved Submitter: Jonathan Wakely Opened: 2008-09-15 Last modified: 2016-01-28
Priority: Not Prioritized
View all issues with Resolved status.
Discussion:
#include <memory> #include <cassert> struct A { }; struct B : A { }; int main() { std::shared_ptr<A> pa(new A); std::shared_ptr<B> pb(new B); std::swap<A>(pa, pb); // N.B. no argument deduction assert( pa.get() == pb.get() ); return 0; }
Is this behaviour correct (I believe it is) and if so, is it unavoidable, or not worth worrying about?
This calls the lvalue/rvalue swap overload for shared_ptr:
template<class T> void swap( shared_ptr<T> & a, shared_ptr<T> && b );
silently converting the second argument from shared_ptr<B> to shared_ptr<A> and binding the rvalue ref to the produced temporary.
This is not, in my opinion, a shared_ptr problem; it is a general issue with the rvalue swap overloads. Do we want to prevent this code from compiling? If so, how?
Perhaps we should limit rvalue args to swap to those types that would benefit from the "swap trick". Or, since we now have shrink_to_fit(), just eliminate the rvalue swap overloads altogether. The original motivation was:
vector<A> v = ...; ... swap(v, vector<A>(v));
[ Batavia (2009-05): ]
We agree with the proposed resolution. Move to
NAD EditorialResolved.
Proposed resolution:
Recommend NAD EditorialResolved, fixed by
N2844.