[cpp-threads] pthreads (was: RE: C++ Connections proposal)

[Andrei Alexandrescu]

My opinion is that we can forge ahead and assume that capability. One 
thing about a good standard is that is knows where to break with the 
"bad ole days" and foresee future trends. Standardizing dead rats is the 
easiest.

Old architectures that can't deal with that will have to generate code 
to CAS it. Yes, it will be less efficient. If it's really important, 
programmers can design around the problem. That's why programmers for 
old architectures are paid big bucks :o)).


Andrei

Boehm, Hans wrote:
>>-----Original Message-----
>>From: Alexander Terekhov
>>
>>Are you absolutely positively sure that all hardware on which 
>>threaded C++ is supposed to run can provide required isolation for
>>
>>  char[2] a;
>>
>>  thread 1: a[0] = 1;
>>
>>  thread 2: a[1] = 2;
>>
>>without CAS/LL(LR)-SC (that would considerably slow things 
>>down)?
>>
> 
> No.  Which is why we need feedback.  But I know of none that can't.
> 
> I am positive that all hardware that can run Java 1.5/5 efficiently
> can do this.  I think the same holds for .NET.
> 
> Thus I'm also pretty sure that every future general purpose processor
> will be able to do this.
> 
> Hans
>