Re: [container] Proposal for ring queues

From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2020-05-22 14:08:12

On 2020-05-22 14:32, Mathias Gaunard wrote:
> On Fri, 22 May 2020, 11:24 Andrey Semashev via Boost, 
> <boost_at_[hidden] <mailto:boost_at_[hidden]>> wrote:
> 
>     circular_buffer is not a queue. The important difference is that it
>     overwrites older elements as you keep pushing new elements into it.
> 
> That's just the behaviour when you run out of capacity.
> Nothing prevents you from checking you're out of capacity and have 
> different behaviour instead.

Yes, that is a possible workaround, but definitely not the intended use 
of a circular buffer.

>     As I said, the main advantage is avoiding dynamic memory allocations
> 
> You cannot avoid memory allocations if your queue is unbounded.
> 
> If pre-allocation of some size is a concern you can just use an adequate 
> allocator.

One allocation might not be a problem, and small_ring_queue allows to 
avoid even that, if the optimal capacity is known at compile time.

However, std::deque continues to dynamically allocate chunks as you push 
and pop elements, even if the average number of enqueued elements stays 
the same. This is avoided by ring queues.