Answer 6—No additional protections at all are required for a simple queue such as this. Since the head and tail index updates are implicitly atomic, the array of data items is already sufficiently protected. Both processes examine both indices to determine whether the queue contains data.

The sending process checks to see whether the head index plus one, modulo the length of the queue, equals the tail index. If so, the queue is full, and it must wait to add more data. Otherwise, it writes values to the data item indicated by the head index and then increments it. The new data item doesn’t become “visible” to the receiving process until the increment occurs, so it doesn’t matter if the data item is a complex structure that requires multiple writes to update.

The receiving process checks to see whether the two indices are equal. If so, the queue is empty, and it must wait before processing more data. Otherwise, it reads whatever information it needs from the data item indicated by the tail index and then increments it. The sending process won’t reuse this slot in the array until the increment occurs. So, once again, it doesn’t matter if the item requires multiple accesses to extract the data.

Contributor: David Tweed