Scott Meyers wrote:
Suppose I have an auto_ptr, shared_ptr, or scoped_ptr. If I want to set it to null, I do this (assuming the pointer is named p):
p.reset(); // okay
If I try this,
p = 0; // error
I get an error during compilation. In other words, reset works, but assignment of the null pointer does not.
But suppose p is an intrusive_ptr. Now
p.reset(); // error
fails, and
p = 0; // okay
succeeds. In other words, assignment works, but reset does not. I find this, er, unintuitive. Does anybody know why intrusive_ptr marches to its own nullifying drummer?
The reason for p = 0 working in intrusive_ptr's case is that intrusive_ptr has an assignment operator taking a raw pointer. The other smart pointer types do not. The lack of intrusive_ptr::reset is arguably a defect in intrusive_ptr's interface. My personal opinion is that moving between shared_ptr and intrusive_ptr doesn't have to be smooth because their semantics are different enough to warrant a careful code review after the switch. On the other hand, consistency does have its benefits.