No reviewer.

build example executables.

No reviewer.

No reviewer.

library as well.

Reviewed by goodell.

Added a configure flag --enable-versioning, which is on by default.
This causes libtool to version shared libraries, using libtool
conventions.  Some platforms may use other versioning schemes, for
example PE MPI uses directory structure layout to version libraries.
In addition, PE must generate consistent, non-versioned "libmpi.so"
output from ldd on the final link of the application to allow for 2
way binary compatibility with other MPI implementations.

Reviewed by goodell.

success, instead of relying on an assignment statement.

No reviewer.

libraries.  This is required on BG/Q.

Reviewed by goodell.

pamid.

No reviewer.

No reviewer.

No reviewer.

No reviewer.

Reviewed by buntinas.

Using -all-static seems to throw a warning when the code uses a system
shared library function (such as gethostbyname).

Reviewed by goodell.

The new ABI string is 8:0:0 because we changed the prototype of
MPIX_Win_allocate_shared since rc1.  The rc2 release should have also
had this string, but I forgot to update the version.m4 file before
cutting the tarball.

No reviewer.

By converting the scratchpad structs into a union and then adding a pointer
type, the compiler becomes more inclined to align the field on a valid pointer
boundary.  This is a temporary, low-risk, low-ish-robustness fix for the problem
before the 1.5 release.  The proper fix is to stop aliasing the character arrays
with the structures and instead just place those structures in the VC via some
macro hook (think `MPID_REQUEST_DECL`).

Reviewed by buntinas@.

The information returned by MPI_Type_get_contents was incorrect for the
new MPI-3 routine, MPIX_Type_create_hindexed_block.  This commit fixes
the bug and adds a regression test for it ("hindexed_block_contents").

Reported by Lisandro Dalcin (thanks!):
http://lists.mcs.anl.gov/pipermail/mpich2-dev/2012-October/000978.html

Reviewed by buntinas@.

These tests themselves were a bit buggy.  They computed "goodval" values
for incorrect indices and had incorrect loop bounds.  They also were a
bit hard to read, which I think I've helped a bit.

Reviewed by buntinas@.

Added barriers to eliminate a data race in the lockopts test case.  The data
race was reported by Michael Blocksome.

Added a section for experimental features and included a statement about MPI-3
RMA.

break on this test.

No reviewer.

No reviewer.

suite to the test file.

No reviewer.

No reviewer.

No reviewer.

No reviewer.

[svn-r10273] Reset the attribute storage type to match the new value.  Without this, if the same attribute key is used from different languages to set the attribute (allowed since MPI2.2), incorrect handling of the value on attribute retrieval will occur