Circular Buffer force_enque

[Tetragramm]

Circular Buffer force_enque

[cameron314]

A few notes:

• There's a race between no slots being available and overwriting the oldest element
• More generally, there's a race between the producer overwriting the oldest element and the consumer also trying to dequeue it
• This code is not exception safe (I believe neither is mine, to be fair)
• I don't quite understand how this would fit into the existing code (given there's no block manipulation)

[Tetragramm]

Hmm. I get the second race, (IE: the producer pauses, the consumer reads the entire queue including a possibly in mid modification oldest element).

I'm not sure what you mean by the first one though. Do you mean that an element might be overwritten even though a free slot has opened up since the semaphore was checked? Or is there a way for the consumer to be reading the element as it's being overwritten?

[cameron314]

The first race is a subset of the second, so it's not really important, but what I was referring to was:

1. slots_->tryWait() fails, so the producer enters the if
2. Meanwhile, the consumer consumes a single element (the oldest one, since this is a FIFO queue)
3. The producer now either destructs the next element (despite a slot being available) or double-destructs the same element that was just consumed

[Tetragramm]

I see where my mistake is now. I knew destructing two elements at once is possible, and just the cost of multi-threaded programming, but I thought double-destructing was impossible because size_t increment is atomic.
Unfortunately, that is true only for specific architectures, so it's not safe to rely on.

Thank you for your help and advice.

[cameron314]

size_t increment is not atomic but std::atomic<size_t> increment is. Not sure how this would help here though.