Fixes a couple of problems:
1) The order of the tuples had two of the fields reversed (count and
   size).
2) The hosts were not being examined in the correct order, so some
   regular cases would be marked as irregular.

Reviewed by buntinas@.

This bug was introduced in r4666.

Reviewed by buntinas@.

This also adds a --with-autotools option in order to provide a mechanism
that works like the (now) old --with-autoconf behavior.

Reviewed by balaji@.

being correctly spread across the available partitions.

[svn-r4680] ugly fix in Newmad for now: getConninfo fails when BC size is MPID_NEM_MAX_NETMOD_STRING_LEN

throw out the help message instead.

[svn-r4674] Added support for exporting the debug versions of the nesting macros to help find nesting errors in ROMIO

environment variable. It's off by default (i.e., uses send).

Reviewed by thakur.

there?). Thanks to Dries for catching this!

and inaccurate statements in them.

Reviewed by buntinas.

[svn-r4667] Initial deposit of the determination of the fortran constant for MPI_IN_PLACE(that fixes req#363), and moved the determination other fortran constants (MPI_F_STATUS(ES)_IGNORE, MPI_STATUS_SIZE and fortran logical values in C) from configure to runtime, i.e. during MPI initialization.  Overall this modification has MPICH2 specific feature...

pointed out in the release notes.

[svn-r4661] Fixed comment in configure for MPIR_Pint and MPIR_Upint. Thanks to Joe @ IBM for pointing this out.

[svn-r4651] Handling the case where the MPICH2 rev num does not have REV or EXT numbers. Some more doc added.

independent file pointer is correct: next time through this routine, type
processing will start writing/reading at the correct location.

where we shift from Send to Ssend in Gatherv.

Reviewed by thakur.

[svn-r4638] Using a name with less duplicates - *more uniq* - for shm seg on windows. This should fix the failures seen with 32+ cores at MS - see ticket #623 for details

non-threaded version for persistent proxy launching. This should fix
ticket #620.

Reviewed by buntinas.

runtime check. This also avoids duplicating the bsend structure in the
source code as well as configure. Both read from a common header file
now.

Reviewed by buntinas.