[cpp-threads] Yet another visibility question

[Paul E. McKenney]

On Fri, Jan 12, 2007 at 01:53:41PM -0800, Lawrence Crowl wrote:
> On 1/11/07, Herb Sutter <hsutter at microsoft.com> wrote:d
> >Even if we did the full lower non-SC layer, ... we need those [ref cnt, 
> >dcl,
> >mark] abstractions anyway because they're ... far easier to use correctly 
> >...
> 
> I agree with this statement, but we are currently lacking specific 
> proposals.
> 
> >And if we think we need to support those cases via abstractions anyway,
> >that removes that argument for providing the brittle individual knobs. Are
> >there any other arguments for providing them?
> 
> I doubt that we can identify, a priori, all of the necessary
> abstractions or that
> we can identify optimal implementations of those abstractions.  A standard
> means to describe and implement those abstractions will avoid committee
> participation in every concurrency advance and therefore result in a much
> shorter "time to market" for many good ideas.  (I am trusting a less formal
> process to also weed out bad ideas.)

I would argue that this committee's job is not to specify all the
abstractions that might be useful or necessary, but rather to provide
a memory-ordering definition that permits portable code for such
abstractions to be reasonably constructed.

							Thanx, Paul