Re: *_SHARED_ALL & TYPE_SUBARRAY semantics

Bill Nitzberg (nitzberg@nas.nasa.gov)
Wed, 28 Aug 1996 13:03:38 -0700

In message <199608281750.LAA11141@hal300.cs.colorado.edu> John Wilson writes
> > 12. PROPOSAL: Remove *_SHARED_ALL or rename to *_SHARED_ORDERED [11.5.2]
> I would urge that *_SHARED_ALL be kept or renamed. I feel that these
> routines have a definite place and should not be removed.

I have been struggling with two problems with respect to
the *_SHARED_ALL routines: they are not orthogonal to the
other *_ALL routines, and I have not been able to justify
them in terms of a real application use.

The first problem can be eliminated by renaming, but the
justification problem is harder. The non-collective *_SHARED
routines support log files and work sharing, but I don't have
any examples of applications that need the *_SHARED_ALL (or
renamed *_SHARED_ORDERED) routines.

John, do you have an application use in mind? I would love to
include a real example in the text.

> > 14. PROPOSAL: TYPE_SUBARRAY semantics depend on language binding [11.6.3]
> Could you comment further upon this? There is no proposal in the draft
> at this point regarding this.

Oops. I should have marked this as a proposal in the draft as well
as in my email. The new draft text is marked by change bars in
section 11.6.3. The semantics of TYPE_SUBARRAY were not well defined
in the draft, and I simply added a paragraph specifying the current
operational behavior of the PMPIO prototype implementation of MPI I/O.
In any case, we should have a straw vote on it at the subcommittee
meeting.

- bill