I realize my basic point is rather simple:

If you say MPI state is UNDEFINED after an error, you have part of the standard done. Simple and clear.

If you say MPI state is DEFINED after certain errors, the standard must define it.

The complexity of identifying and classifying errors and defining the state after each is mind boggling. The job of implementing and testing that is even more mind boggling.

BTW - I see no practical problem with mandating that an error code in class NOT_IMPLEMENTED be considered harmless.  I.E. "A subroutine that returns an error in error class NOT_IMPLEMENTED leaves the MPI library in the same state as if the call had not been made."  That is easy to define and implement.


Dick Treumann  -  MPI Team          
IBM Systems & Technology Group
Dept X2ZA / MS P963 -- 2455 South Road -- Poughkeepsie, NY 12601
Tele (845) 433-7846         Fax (845) 433-8363