On 02/27/09 14:13, Achilleas Margaritis wrote:
This seems a serious shortcoming. To justify this, I guess you'd have to somehow argue that this use-case doesn't happen very often or that if it does, the user can easily detect that it happens and take corrective active. Then, to justify this method vs. the use of weak-ptrs, you'd have to argue why it's easier for the user to detect "transferred to a newer ptr" than it is to detect where a cycle is created.
Indeed. I think the algorithm works for immutable ptrs only.
Aren't mutable ptrs necessary for creating cycles? Suppose the ptr graph is B->A->B. (IOW, object B has pointer, B.ptr which points to object A, and object A has pointer, A.ptr which points to B.) Now, either A or B must be created first. When the first one is created (suppose it's B) it can't be constructed already pointing to A because A hasn't been created yet. Consequently, B.ptr must be mutable in order to eventually point to A. Am I missing something?