#3 sounds like a great approach to me. I agree with Kevin that if we keep the object husk approach that any use of a weak pointer that returns nil should drop any reference to a husk.
Spin locks are, unfortunately, illegal on iOS, which does not guarantee progress in the face of priority inversion.
John.
···
On Dec 12, 2015, at 7:04 PM, Chris Lattner <clattner@apple.com> wrote:
-Chris
On Dec 11, 2015, at 7:00 AM, Mike Ash via swift-dev <swift-dev@swift.org> wrote:
3. Borrow a bit from the weak pointer to implement a spinlock. This is really a special case of (2), with the activity count being capped at 1 and additional activity blocking. In fact, you could even do a hybrid approach by borrowing more bits. (I think it could safely steal up to 20 bits with current 64-bit architectures. This may not be wise. As long as targets are pointer-aligned you can safely steal 2/3 bits.)