I think he's complaining that because the head_ and tail_ loads in push/pop are relaxed, rather than also being acquire, they can be reordered relative to the a

I think he's complaining that because the head_ and tail_ loads in push/pop are relaxed, rather than also being acquire, they can be reordered relative to the acquire tail_ and head_ loads respectively. I don't believe this impacts the correctness of the logic, but I could be missing something.