The use of complete was vague. I meant complete in a different
sense. I changed it to:
MPI marks the request as complete (as if MPI_GR_MARK_COMPLETE was
called)
What I mean is that the complete_fn is now eligible to execute.
> Also, after a call to
> MPI_Cancel, the request either completes entirely or no part of it
> gets completed, but nothing in between. The statement in parenthesis on
> line 47 is a bit misleading.
I agree that this is desirable but am not sure we want to require
this. I was hoping the text following covered this. I wanted to say
that it is a good idea to clean up after oneself but that nothing in
MPI requires this for a GR. Is this ok?
Steve