Re: [boost] boost.lockfree update

Subject: Re: [boost] boost.lockfree update
From: John Bytheway (jbytheway+boost_at_[hidden])
Date: 2010-07-25 05:49:40

On 25/07/10 09:50, Tim Blechmann wrote:
>> What exactly about PODs is it taking advantage of?  In C++0x will it
>> work for all standard layout types?
> 
> one needs to read the payload from the fifo node, before it is dequeued. 
> this may happen more than once otherwise a later dequeue may free the node. 
> it is nothing, c++0x can change, but a property of the michael/scott queue 
> [1]. the type essentially needs a trivial copy constructor.

If that's all it needs then I suggest you phrase the limitation in those
terms; it's a lot less restrictive than requiring a POD type (which
cannot, for example, have any user-defined constructors).

> [1] 
> http://www.cs.rochester.edu/research/synchronization/pseudocode/queues.html

If you implemented it exactly as it appears in this link then I guess it
really needs a trivial copy assignment operator, not a trivial copy
constructor, but of course the two tend to go hand-in-hand.  You should
probably require both.

John Bytheway