On Tue, 2008-04-29 at 20:03 +0300, Peter Dimov wrote:
My current line of thinking is that they should not, as even Frank is having second thoughts about it. But I haven't made up my mind yet. There's not much time left though.
Trying to implement enable_shared_from_this with only the public interface of shared_ptr, weak_ptr, etc. (and of course sp_accept_owner and sp_deleter_wrapper) turns out to be a good test for completeness IMHO. I tried to do this and found one thing that I cannot do with the current implementation of weak_ptr: I cannot test if it's empty. It lacks the conversion to bool, similar to shared_ptr. Or is there any way to do this I missed? I attached a patch (tested, passing all regression tests) to turn the current trunk version into a more reference-like implementation. It shows that conversion for bool of a weak_ptr is a desirable feature (see init_weak_once), since only now I understand what _owner was really meant for: It worked around the lack of weak_ptr conversion to bool by manually tracking it's internal state. In case conversion to bool seems to dangerous to you, I'd suggest to add at least an explicit method to test for px != 0 (.empty()?), to allow implementations to avoid tracking internal state manually. Regards, Daniel