(j3.2006) further thoughts on Aleks' proposal

[Aleksandar Donev]

Dick Hendrickson wrote:

> 1)  Is there a problem to be solved.
> I think Bill's comment, that he's never seen a
> problem, is the hardest one to deal with.
But it is the easiest one to make :-) Sure, I agree with Bill that this 
is not, has not been, nor will it ever be the most pressing or even a 
very important issue of the day or even MPI+Fortran interop. But the 
issue exists (the MPI forum would not be writing whole sections in the 
MPI-2 standard on tricks to avoid compiler optimizations if someone did 
not actually have to use those tricks in some code) at least in some 
codes, under some conditions, and for some compilers.

The issue is certainly there from the stand point of the standard: We do 
not really properly account for asynchronous data transfer performed by 
external libraries. Even if compilers need to do nothing, we should do 
it in the standard, IMNSHO, regardless of the actual severity of the 
problem in practice, and most of all, regardless of the fraction of 
users affected by it. It is like the constitution of a country---someone 
has to worry about the minorities too. Always basing language design on 
numbers/polls is a bad idea, IMHO.

Best,
Aleks