Mpiexec release 0.78: mpich2/pmi update, bug fixes

[Pete Wyckoff]

Not too much happened since 0.77 was released last September.  Unless
you plan to use mpich2 or are seeing one of the bugs listed below, I
suggest you sit this release out.

mpich2/pmi update

    Recent releases of the fast-evolving mpich2 code with its PMI
    startup protocol have made a fair number of changes to the PMI
    protocol that broke compatibility with earlier mpich2 releases.
    
    The PMI protocol in this mpiexec release has been tested against
    mpich2-1.0.1 with the ch3 device and three of its eight different
    channels:  sock, shm, and ssm.  It may very well work with the other
    variants too, but has not been tested.  As each channel exercises a
    somewhat different set of PMI functions, getting good coverage is a
    challenge.

    This version of mpiexec will likely not start codes using older
    mpich2 versions, but that has yet to be tested.  In the future,
    however, as long as the developers are good about updating the
    version numbers along with protocol changes, we should be able
    to detect and adapt to older executables.  Kudos to the mpich2
    team for adding version numbers.

MAC OSX mpich/gm

    A few little bugs were fixed.  One that deserves attention is the
    use of mpich/gm on MAC OSX.  There is a magic environment variable
    that must be set to allow overloading of the malloc code by the
    mpich/gm library.  At least that is how Andrew Gallatin of Myricom
    explained it to me.  Anyway this should resolve that.

Little bugs

    The other little bugs resulted in the premature death of mpiexec and
    were hopefully only rarely encountered.  One was related to closing
    the wrong socket when running with "-nostdin".  The other happened
    as a race condition during application exit.


Full changelog and downloads at:  http://www.osc.edu/~pw/mpiexec/
Send bug reports, comments and suggestions to the mailing list.

		-- Pete