For a standard, the issue is whether *all* implementations are Unix- like, and for MPI, there are definitely some that aren't Unix-like, and there are likely to be more (I don't want to see Unix running on each of the 10,000,000 nodes of my Exascale system). Handling this is what the advice to implementor is for - the Forum can remind implementors about things that may apply to some but not all implementations. We've also taken the position that many of these issues should be "quality of implementation" rather than mandating specific behavior. In this case, an advice to implementors should be all that is necessary to ensure that an implementation doesn't do something weird by mistake. Bill On Feb 21, 2010, at 8:57 AM, Marty Itzkowitz wrote:
Aren't most implementations of MPI Unix or Unix-like? Why can't this fact be used in writing a definition?
William Gropp Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign