On Monday, January 09, 2012 09:25:14 Steven Schveighoffer wrote: > Actually, not invalid for the current implementation. I don't know if > it's stated whether an AA specifically requires that elements do not > re-associate on a rehash.
Well, like I said, it depends on the current implementation. There are hash table implementations where rehashing would invalidate the pointer returned by in, and I don't believe that the spec specificies that D's AA guarantees that rehashing doesn't do that. So in the future, it _could_ be changed to an implementation which invalidates the pointers on rehash. As such, it doesn't really matter what the current implementation does. Relying on the current behavior is unsafe if it's not guaranteed by the spec. Now, if we want to change the spec to make such guarantees, that's fine, but I don't believe it makes them right now. Good to know what the current implementation is doing though. - Jonathan m Davis
