[cpp-threads] seq_cst compare_exchange and store-load fencing

[Boehm, Hans]

> -----Original Message-----
> From: cpp-threads-bounces at decadentplace.org.uk 
> [mailto:cpp-threads-bounces at decadentplace.org.uk] On Behalf 
> Of Alexander Terekhov
> Sent: Tuesday, January 13, 2009 12:36 PM
> To: C++ threads standardisation
> Subject: Re: [cpp-threads] seq_cst compare_exchange and 
> store-load fencing
> 
> On Tue, Jan 13, 2009 at 6:27 AM, Hans Boehm <Hans.Boehm at hp.com> wrote:
> 
> [... putting the notes into the atomics section ...]
> 
> > It sounds like we need a library issue?
> 
> Putting the notes in 1.10 will work better (will receive 
> better attention). I suggest to connect 1.10's "sequentially 
> consistent"/"sequential consistency" terms with atomics 
> section's "seq_cst" beast via putting the notes into 1.10 in 
> addition to similar notes put into 29.X elaborating it with 
> more details (like the examples and the detailed explanations).
> 
I'd personally prefer to keep anything of interest only to users of explicit memory_order_... arguments out of 1.10, at least as much as possible.  I still hope and expect that this will be of interest to a relatively small number of low-level library implementors, who will have to read the atomics chapter in detail anyway.  Remaining users, who I would expect to use mostly locks, already have to plow through a lot of extra stuff, but I'd prefer to minimize that.

Hans