This came up in another thread today but I figure I'd throw a brief comment to the timeline. The concept of "grace periods" where you separate the logical and physical deletion of resources is something you see in RCU, EBR, QSBR, etc, but it's just as useful in single-threaded code where you only have pseudo-concurrency through subroutine calls. Like the age-old pattern of deferring deletion until the end of the game loop, or autorelease pools in Objective C which get drained by the run loop.
